TR-4423-99-C
                      June 1999
       IR Open-Path Monitoring
        Guidance Document
                Third Edition
  O)
  C
        1000
2000
3000
                           -i
              Wavenumber (cm )
            ManTech Environmental Technology, Inc.
                  P.O. Box 12313
              Research Triangle Park, NC 27709

               A ManTech International Company
4000

-------
Submitted to:
                                                           TR-4423-99-03
                                                               June 1999
               FT-IR Open-Path  Monitoring

                     Guidance  Document

                              Third Edition

                                   by

                  George M. Russwurm and Jeffrey W. Childers
                    ManTech Environmental Technology, Inc.
                  Research Triangle Park, North Carolina 27709
                            William A. McClenny
                         Work Assignment Manager
               Human Exposure and Atmospheric Sciences Division
                     National Exposure Research Laboratory
                  Research Triangle Park, North Carolina 27711
                           Contract 68-D5-0049
Reviewed and Approved by:
                                       	
 eorge M. Russwurm, Principal Investigator    E. Hunter Daughtrey,^., Area Supervisor
                    ManTech Environmental Technology, Inc.
                             P.O. Box 12313
                      Research Triangle Park, NC 27709

                         A ManTech International Company

-------
                                                                    TR-4423-99-03
                                   Foreword

      This report  presents the results  of work performed'by  ManTech Environmental
Technology, Inc., under Contract 68-D5-0049 for the Atmospheric Methods and Monitoring
Branch,  National Exposure Research Laboratory, U.S. Environmental Protection Agency,
Research Triangle Park, NC. This  report  has  been  reviewed  by ManTech Environmental
Technology, Inc., and  approved for publication.  Mention  of trade names or commercial
products does not constitute endorsement or recommendation for use.
                                       in

-------
                                                                      TR-4423-99-03
                           Preface to the Third Edition

       The  Fourier  transform  infrared  remote sensing  technique  for  measuring  gas
 concentrations in the atmosphere has undergone a vigorous growth and development period
 over the last 10 years. There seems to be an expanding awareness of the capabilities of this
 technique and therefore a continuing demand for a guidance document that is useful to the
 people entering the field  for the first time. It is our hope that this document  will fulfill that
 need. The intent of this document is to provide information about the FT-IR technique that will
 assist the user in understanding how the system functions.

       While there is some difficulty in producing a document that addresses all the questions
 operators  may have about the FT-IR remote sensing technique, we have tried to include as
 much pertinent information  as is available at this time. Some of the topics included here are
 more rigorous than would at first be deemed necessary.  But both authors  have come to
 understand that the operator should have an in-depth understanding of the instrumentation in
 order to make appropriate choices about the data acquisition and processing.

       The data that has been used to compile the information in this document was acquired
 over a several year period and with several different instruments. We have tried to present
 guidance that would be common to all instruments, but in some instances that is not possible.
 The judgements included in  this document are based on our study of spectra  acquired with
 high (0.1 25 cm'1)  resolution  and with low (1.0 cm'1) resolution instruments and with both the
 monostatic and the bistatic systems. During the data acquisition phase of this project we also
 acquired the ancillary data of relative humidity, temperature, and atmospheric pressure. This
 provided us with much of the information necessary to understand the effects of water vapor
 on the data and the manufacture of a background spectrum and a water vapor reference.

       The document contains 1 2  chapters that we  believe address  the  most important
 aspects of atmospheric monitoring  with the FT-IR remote  sensing  technique. As we have
 defined this technique, we mean the use of an open-air path up to 1 km long.  Chapter 1 2 is
 a bibliography that contains more than 330 citations of papers and presentations that describe
the technique. While this is not an exhaustive compilation, it shows that there  is a wealth of
 information about the use and efficacy of the technique.

      The authors wish to emphasize that this document is meant to be a primer for the new
 users of the FT-IR remote sensing technique and to give them some guidance in the overall
 operation of the instrument. It is not meant to be a standard operating  procedure. For that,
                                         v

-------
                                                                    TR-4423-99-03
there is an EPA-approved method (compendium method TO-1 6) and also two ASTM methods
that are available. These are cited  in the text and in the bibliography.

GMR
JWC
June 1999
                                       VI

-------
         i^jijjj                                                     TR-4423-99-03
 TF/*U-
 ICtrrt: '
      : 'L- = =
                                     Contents

Foreword  	iii
Preface to the Third Edition	'. . . .  v
Figures	xi
Tables  	  xv
Acknowledgement	  xvii

   1  Introduction	  1-1
      1.1    Overview of Document	  1-1
      1.2    References	  1-2

   2  The Fourier Transform Spectrometer  . .	  2-1
      2.1    Introduction and Overview 	  2-1
      2.2    The Michelson Interferometer	  2-3
             2.2.1  Interference  	  2-3
             2.2.2  Resolution	  2-7
             2.2.3  Throughput	  2-8
             2.2.4  The Detector	  2-8
             2.2.5  The IR Source	  2-10
      2.3    Transfer Optics, Telescopes, and Beam-Return Optics	  2-10
             2.3.1  Bistatic System	  2-13
             2.3.2  Monostatic System	  2-14
      2.4    The Electronics  	  2-15
      2.5    The Computer	  2-16
      2.6    The Data Output	  2-17
             2.6.1  Beer's Law  	  2-17
             2.6.2  The Interferogram  .,	  2-19
                   2.6.2.1 Truncation  	  2-19
                   2.6.2.2 Phase Shift	  2-19
             2.6.3  The Transform	:	  2-20
             2.6.4  The Single-Beam Spectrum	  2-20
             2.6.5  Data Analysis	  2-21
                   2.6.5.1 Generation of the Absorption Spectrum	  2-21
                   2.6.5.2 Generation of the Reference Spectrum  	  2-21
                   2.6.5.3 Analytical Methods	  2-22
                          2.6.5.3.1 Comparison Technique 	  2-23
                          2.6.5.3.2 Scaled Subtraction Technique  	   2-23
                          2.6.5.3.3 Multicomponent Analysis Techniques ....   2-24
      2.7    References  	   2-24

   3  Initial Instrument Operation	   3-1
      3.1     Introduction and  Overview	   3-1
      3.2    The Single-Beam Spectrum	   3-2
             3.2.1  Wave Number Shift	   3-4
             3.2.2  Change in Resolution	   3-4
      3.3    Distance to  Saturation  	   3-5
                                        VII

-------
                                                                    TR-4423-99-03
    3.4   Return Intensity as a Function of Distance  	  3-5
    3.5   Determination of the Stray Light Signal  	  3-6
    3.6   Determination of the Random Noise of the System  	  3-7
    3.7-   Return Intensity as a Function of Water Vapor  	  3-10
    3.8   References  	  3-10

4  Background Spectra	  4-1
    4.1    Introduction and Overview  	  4-1
    4.2   Synthetic Background Spectra  	  4-3
    4.3   Upwind Background Spectra	  4-3
    4.4   Short-Path Background Spectra	  4-4
    4.5   Averaged Background Spectra	  4-6
    4.6   Why Use a Background  	  4-7
    4.7   General Advice About Background Spectra	  4-8

5  Water Vapor Spectra	  5-1
    5.1    Introduction and Overview  	  5-1
    5.2   Water Vapor Spectra Considerations  	  5-2
    5.3   General Process for the Production  of a Water Vapor Spectrum  	  5-2
          5.3.1  Selection of Spectra	  5-3
          5.3.2 Creation of Synthetic Background .	  5-3
          5.3.3 Creation of the Absorption Spectrum	  5-3
          5.3.4 Subtraction  of the Target Gas	  5-4
    5.4   Calculated Water Spectra	  5-4
    5.5    Methane and Ozone Examples	  5-5

6  Siting	  6-1
    6.1    Introduction  and Overview  	  6-1
    6.2    Selecting the Path	  6-3
          6.2.1  The Longest Path	  6-5
          6.2.2 Shortest Path Requirements  	  6-5
          6.2.3 Short Path Versus Long Path	  6-6
          6.2.4 Prevailing Winds  	  6-8
          6.2.5  Slant Path Versus Horizontal Path 	  6-8
    6.3    Changing the Path	  6-8
    6.4    Ancillary Measurements	  6-9
    6.5    A Specific Case	  6-9
    6.6    References	  6-12

7   Resolution Considerations  in Long-Path, Open-Path FT-IR Spectrometry	  7-1
    7.1    Introduction  and Overview   	  7-1
    7.2    Definition of Resolution  	  7-3
    7.3    Trading Rules in FT-IR  Spectrometry  	  7-4
    7.4    Example Spectra of CO2 and Water  Vapor 	  7-6
          7.4.1  Resolution Effects  	  7-8
                7.4.1.1 Laboratory Measurements 	  7-8
                7.4.1.2 Long-Path Measurements	  7-10
          7.4.2  Zero-Filling Effects	' .  7-11

                                     viii

-------
                                                                    TR-4423-99-03
          7.4.3  Apodization Effects  	  7-12
    7.5    Effect of Resolution on Quantitative Analyses	  7-14
          7.5.1  Studies from the Literature	  7-15
          7.5.2  Case Study:  The Effect of Resolution and Related
                 Parameters on the CIS Analysis of Multicomponent
                 Mixtures	  7-17
                 7.5.2.1 Mixtures of CO and 13CO	  7-18
                 7.5.2.2 Mixtures of Acetone, Methylene Chloride,
                        and  Ethanol  	  7-19
                        7.5.2.2.1  Effect of the Number of Data Points
                                  on the  CIS Analysis	  7-20
                        7.5.2.2.2 Effect of S/N Ratio on the CIS Analysis .  .  7-22
                 7.5.2.3 Mixtures of Methylene Chloride and Nitrous Oxide  .  .  7-22
                 7.5.2.4 Conclusions and Recommendations Based
                        on Case Study  	  7-23
    7.6    General Conclusions and Recommendations	  7-24
    7.7    Guidance for Selecting Resolution and Related Parameters	  7-25
    7.8    References 	  7-28

8   Nonlinear Response Caused by Apodization Functions and Its
    Effect on FTIR Data	  8-1
    8.1    Introduction and Overview 	  8-1
    8.2    Procedure and Theoretical Basis  	  8-4
    8.3    Results of Calculations	  8-7
    8.4    Analysis	  8-13
    8.5    Discussion  	  8-17
    8.6    Conclusions and Recommendations	  8-19
    8.7    References 	  8-20

9   The Technique of Classical  Least Squares  	  9-1
    9.1    Introduction and Overview 	  9-1
    9.2    Least Squares Analysis  for One Gas  	  9-1
    9.3    Matrices	  9-5
          9.3.1  Matrix Types	  9-5
          9.3.2 Some Matrix Properties  	  9-6
          9.3.3 Multiplication of  Matrices	  9-7
          9.3.4 The Identity Matrix	  9-8
          9.3.5 The Transpose of a Matrix 	  9-9
          9.3.6 The Determinant of a Matrix	  9-9
          9.3.7 Cofactors  of Matrices	  9-11
          9.3.8 The Inverse of a  Matrix  	  9-11
   9.4    Matrices and Algebraic Equations  	  9-13
   9.5    Least Squares and Matrices	  9-14
   9.6    Expansion to Many Gases	  9-18
   9.7   Least Squares Errors  	  9-20
   9.8    References  	  9-21
                                     IX

-------
                                                                  TR-4423-99-03
 10 Quality Assurance and Quality Control	  10-1
    10.1   Introduction and Overview  	  10-1
    10.2  Project Plan Categories	  10-2
          10.2.1   Category Definitions	  10-3
          10.2.2   Category I Points to be Addressed	  10-3
                10.2.2.1   Project Description	  10-4
                10.2.2.2  Project Organization and Responsibilities  	  10-4
                10.2.2.3  QA Objectives	  10-5
                10.2.2.4  Site Selection and Sampling Procedures	  10-5
                10.2.2.5  Sample Custody  	  10-6
                10.2.2.6  Calibration Procedures and Frequency  	  10-6
                10.2.2.7  Analytical Procedures	  10-7
                10.2.2.8  Data Reduction, Validation, and Reporting  	  10-7
                10.2.2.9  Internal Quality Control Checks	  10-8
                10.2.2.10 Performance and System Audits	  10-8
                10.2.2.11  Preventive Maintenance	  10-8
                10.2.2.12 Calculation of Data Quality Indicators  	  10-9
                10.2.2.13 Corrective Action	  10-9
                10.2.2.14 Quality Control Reports to Management	  10-9
                10.2.2.15 References	  10-9
                10.2.2.16 Other Items	  10-9
    10.3   Case Study:  QA Data Collected Over Two and One-Half Months
          at a Semipermanent Field Site	10-10
    10.4   Recommendations of Tests to Be Included in a QA Program
          for FT-IR Long-Path Monitors	10-14
          10.4.1   Noise Measurements  	10-15
          10.4.2  Stability of Instrument  	10-15
          10.4.3  Accuracy and Precision	10-16
          10.4.4  Completeness and Representativeness of Data  	10-18
          10.4.5  Comparability of the Data	10-18
          10.4.6  Ancillary Measurements  	10-19
          10.4.7  Documentation  	10-19
    10.5  References 	10-19

11 Glossary of Terms for FT-IR Open-Path Remote Sensing   	   11-1
   11.1   Introduction and Overview  	   11-1
   11.2   Terms	   11-1
   11.3   References 	   11-7

12 Bibliography  	   12-1
   1 2.1   Introduction and Overview  	   1 2-1
   12.2   Publications	   12-1

-------
                                                                     TR-4423-99-03
                                     Figures


Number                                                                      Page

  2-1   A Schematic of the Simplest Form of a Michelson Interferometer  	  2-4
  2-2   Schematic of Interference Created by Division of Amplitude  	  2-5
  2-3   Center Burst Increasing as the Wave Number Range Expands	  2-6
  2-4   Interferograms for a Range of 3500 cm"1  	  2-6
  2-5   Interferogram of Two Cosine Waves as a  Function of AT1	  2-7
  2-6   The Bistatic Configuration	  2-11
  2-7   The Monostatic Configuration  	  2-12
  2-8   Data Reduction Flow Chart  	  2-22

  3-1   Single-Beam  Spectrum Along a 414-m Path   	  3-2
  3-2   Single-Beam  Spectrum Recorded at a 20-m Total Path Length
       Indicating Nonlinear Operation	  3-3
  3-3   Region Between 1000 and  1025 cm"1	  3-4
  3-4   Subtraction of  Spectra for the Determination of Line Shifts and
       Resolution Changes  	  3-5
  3-5   Effect of Stray Light	  3-7
  3-6   The RMS Baseline Noise Measured Between 980 and 1020 cm'1,
       2480 and 2520 cm'1, and 4380 and 4420 cm'1	  3-9

 4-1   Synthetic 70 Spectrum	  4-3
 4-2   A Possible Configuration for I0 Spectrum Acquisition	•	  4-4
 4-3   Procedure for Acquiring a Short-Path Background Spectrum 	  4-5

 5-1   The Portion of a Single-Beam Spectrum over Which Methane  Absorbs	  5-5
 5-2   Methane Region with Synthetic Background Spectrum Superimposed  	  5-5
 5-3   Methane Reference Spectrum and the Calculated Absorption Spectrum ....  5-6
 5-4   Water Vapor  Spectrum Made for the Methane Absorption Region 	  5-6
 5-5   Atmospheric  Ozone Absorption Spectrum  and Ozone
       Reference Spectrum	;	  5-6
 5-6   Ozone Measured at Research Triangle Park During June	  5-7

 6-1   Sulfur Hexafluoride Reference Spectrum	  6-7
 6-2   Aerial Photograph of a Superfund Site	  6-11

 7-1   Single-Beam IR Spectra of CO2 Measured at 0.25-, 0.50-, 1.0-, and
       2.0-cm"1 Resolution with No Apodization and No Additional Zero  Filling  ....  7-7
 7-2   Single-Beam IR Spectra of Water Vapor Measured at 0.25-, 0.50-, 1.0-,
       and 2.0-cm"1  Resolution with No Apodization and No Additional
       Zero Filling	  7-7
 7-3   Single-Beam IR Spectra of Water Vapor Measured at 2-, 1-, and 0.5-cm'1
       Resolution over a 1 50-m Path  	  7-7
                                       XI

-------
                                                                    TR-4423-99-03
 7-4  IR Spectra of Water  .......................................  7-11
 7-5  Absorption Spectra of C02 Measured at 0.25-cm'1 Resolution with a
      Zero-Filling Factor of  1,  0.5-cm"1 Resolution with No Zero-Filling, and
      0.5-cm"1 with a Zero-Filling Factor of 2  ..........................  7-1 1
 7-6  Absorption Spectra of Water Vapor Measured at 0.25-cm"1 Resolution
      with a Zero-Filling Factor of 1, 0.5-cm"1 Resolution with a Zero-Filling
      Factor of 2, and 1-cm"1  Resolution with a Zero-Filling Factor of 4 ........  7-12
 7-7  Absorption Spectra of CO Measured at a Nominal 0.125-cm"1 Resolution
      with No, Triangular, Happ-Genzel, and Norton-Beer-Medium
      Apodization Functions ......................................  7-13
 7-8  Absorption Spectra of Water Vapor Measured at 0.5-cm"1 Resolution
      with a Zero-Filling Factor of 2 and with No, Triangular, Happ-Genzel,
      and Norton-Beer-Medium Apodization Functions ....................  7-13
 7-9  Reference 0.25-cm'1 Spectra of 13CO and CO and Spectra of Synthetic
      Mixtures of 150 ppm  CO and 100 ppm 13CO  Measured at 0.25-, 0.5-,
      1 .0-, and 2.0-cm"1 Resolution .............. . ..................  7-19
 7-10  Concentration Calculated from CLS Analysis vs. Known
      Concentration for 13CO/CO Mixtures Measured at 2-cm"1  Resolution ......  7-19
 7-1 1  Reference 0.25-cm"1 Spectra of Acetone, Methylene Chloride, and
      Ethanol and Spectra of Synthetic Mixtures of 100 ppm Acetone,
      100 ppm Methylene Chloride, and 500 ppm Ethanol  Measured at
      1 .0-, 2.0-, and 4.0-cm"1  Resolution .............................  7-20
 7-12  Spectra of Synthetic Mixtures of 100 ppm Acetone, 100 ppm
      Methylene Chloride, and 500 ppm Ethanol Measured at 1-cm"1
      Resolution with 0, 1,  5,  10, and 25% Noise Added  .................  7-22
 7-13  Reference 0.25-cm"1 Spectra of N2O and Methylene Chloride and
      Spectra of Synthetic Mixtures of 50 ppm N20 and 100 ppm Methylene
      Chloride Measured at  0.25-, 0.5-, and 1 .0-cm"1 Resolution  ............  7-23
 7-14  Concentration Calculated from CLS Analysis vs. Known Concentration for
      N2O/Methylene Chloride Mixtures Measured at 0.25-cm"1  Resolution .....  7-23

 8-1   A Portion of a Water Spectrum Using Boxcar Apodization and
      Triangular Apodization .......................................  8-2
 8-2   Schematic of Actual and Assumed FT-IR Responses   .................  8-3
 8-3   Measured Concentration of Methane vs. the Experimental Response
      of the FT-IR  ..............................................  8-7
 8-4   Methane Absorbance  at  2927 cm"1  .............................  8-9
 8-5   Ammonia Absorbance at 967 cm"1 ..............................  8-9
 8-6   Water Absorbance at  1014.5 cm"1 .............................  8-10
 8-7   Methane Absorbance  vs. CL at 2927 cm"1  .......................  8-10
8-8   Ammonia Absorbance at 2927 cm"1  ............................  8-11
8-9   Water Absorbance vs. P at 1 01 4.5 cm"1 .........................  8-11
8-10  Water Absorbance for 0.5 torr at 1014.5 cm"1  ....................  8-12
8-1 1  Water Absorbance for 35 torr at 1014.5 cm"1 .....................  8-12
8-12  Match of Water Absorbance at 1014.5 cm"1 ......................  8-14
8-1 3  Analysis Results for Methane from 291 5 to 2929 cm"1  ..............  8-16

                                      xii

-------
                                                                 TR-4423-99-03
8-14 Analysis Results for Methane from 2900 to 3000 cm"1  ..............  8-16
8-1 5 Methane Analysis Allowing the Water Reference Concentration to
     Vary  ..................................................  8-17
8-16 Plot of Regression Slopes vs. Temperature .......................  8-18
8-17 Difference After Regression Coefficients Have Been Applied  ........ '. . .  8-19

9-1  Least Squares Fit of a Data Set  ................................  9-3

10-1 Return Signal Magnitude of the FT-IR Monitor Measured Daily at
     0700 and  1 200  .......................................... 10-11
10-2 The RMS Baseline Noise Measured Between 980 and 1020 cm'1,
     2480 and  2520 cm'1, and 4380 and 4420 cm'1 .................... 10-11
10-3 Repeatability of the Position of the Water Vapor Singlet at 1014.2 cm"1
     Measured on November 10, 1993, December 22, 1993, and
     January 4, 1 994  ......................................... 10-12
10-4 Measurement of Ambient Methane Concentration and Single Beam
     Intensity at 987 cm'1 on November 17 and 18, 1993  ................ 10-13
10-5 Peak Area  of 2998. 8-cm'1 Absorption Band of CH4 and the 1014.2-cm"1
     Absorption Band of Water Vapor Measured on November 17-18, 1993  ... 10-14
                                    XIII

-------
              2f                                                     TR-4423-99-03


                                     Tables

Number                                                                      Page

  6-1   Estimated Method Detection Limits for Selected Gases	   6-4
  6-2   Minimum Usable Path Lengths	   6-7

  7-1   Resolution Test Data	   7-6
  7-2   Optimal Wave Number Region and Minimum Resolution  	  7-16
  7-3   Effect of the Number of Data Points on the CLS Analysis	  7-21
  7-4   The Effect of Zero Filling on the CLS Analysis	  7-21
  7-5   Effect of Noise on the CLS Analysis	  7-22

  8-1   Maximum Values Over Which Response Can be Considered Linear and
       Associate Errors	  8-13

  9-1   Partial Listing of Spectral Absorbance Data	   9-2
                                       xv

-------
         75:.v/5:?/=                                                     TR-4423-99-03
                                Acknowledgement

       There have been many helpful comments and  reviews of this document that the
authors have received from many people since the first edition was published. In particular,
we would like to thank the following people. Dr. William  Herget (now deceased) reviewed the
original document and the contents  of Chapter 8 in this version. Dr. Steven Levine of the
University of Michigan provided much support and an excellent review of an earlier version.
Dr. Peter Griffiths of the University of Idaho provided an  excellent and comprehensive review
of the  second edition of this document.

       Professor Konradin Weber of the  Fachhochschule in  Dusseldorf,  Germany, and a
doctoral candidate, Mr. Alexander Ropertz, provided us  with the experimental verification of
the nonlinear response discussed in Chapter 8. Professor Weber was also the first person to
provide us with the information that the detection limits when measured from the actual data
do not seem to change with path length.

       Dr. William Phillips of SpectraSoft Technology in Tullahoma, Tennessee, provided us
with much of the mathematical development and some of the software that allowed us to
perform the calculations in Chapter 8.

       The authors wish to thank two companies for  their support in this endeavor. The
MIDAC Corporation in Irvine,  California,  provided  us  with a bistatic system capable  of
acquiring data at 0.5-cm"1 resolution. Kayser-Threde Gmbh in Munich, Germany, provided us
with a  high-resolution instrument which gave us a great deal  of insight  about the need for
higher  resolution spectra.

       We  must also express  our gratitude to Ms. Janet  Parsons  for editing this entire
document again. We are convinced that Ms. Parsons has made this document a more readable
document.

       Finally, we wish to thank Dr. William McClenny for his support in the preparation of this
work.
                                        XVII

-------
                                                                       TR-4423-99-03
                                     Chapter 1
                                    Introduction
       The Michelson interferometer has had
 a remarkable history in that new uses for the
 deyice  have  been  found for more  than
 100 years.   One use of the  interferometer
 that has experienced rapid growth since the
 mid-1960s is as the main optical component
 of   Fourier   transform   infrared    (FT-IR)
 spectrometers.  Although there  have been
 several  applications  of FT-IR  spectrometers
 to unique and difficult problems, the majority
 of FT-IR systems have been  used to make
 qualitative  measurements under controlled
 conditions  in  the  laboratory.  More than
 20 years ago, some efforts were made to
 use the  instrument for making quantitative
 measurements  of  atmospheric   gaseous
 pollutants over extended open paths  (Hanst
 1970; Herget and Brasher 1979).  Although
these efforts were largely successful, they
were  overlooked  by the great majority of
people engaged in environmental monitoring.
During the 1 980s there was steady but slow
progress in development of the technique. In
the late  1980s, a revival of  the technique
occurred, initiated in part during a meeting of
the Chemical Manufacturer's  Association in
Houston (Russwurm and  McClenny  1990;
Levine et al. 1991; McClenny et al.  1991),
and  today  there  is a  large amount of
developmental activity taking place. (See the
bibliography in Chapter  10.)

      This   document   describes   the
components  of  FT-IR   monitors  and is
 intended to provide  guidance for the FT-IR
 operator in field monitoring applications. It is
 a point of reference for further development
 and evaluation of FT-IR open-path monitors
 as field instruments.

 1.1    Overview of Document

       A brief discussion of the FT-IR open-
 path  monitor  and its function is given  in
 Chapter  2,  along  with a  more in-depth
 description of the various components of the
 sensor.  Chapter 3 includes  the preliminary
 procedures   for  setting  up  the  FT-IR
 instrumentation for monitoring. Chapter 4 is
 a discussion  of background spectra,  and
 Chapter  5 is  a  discussion of  water vapor
 spectra.   Chapter  6  presents  guidance on
 how to set up the monitoring instruments
 within the physical  constraints  of a site.
 Chapter 7  presents experimental  data that
 illustrate  the effect of resolution and related
 parameters on the spectral data. Chapter 8
 contains  a  discussion of  the effects  of the
 apodization function on the FTIR data using
 classical  least squares  analysis. It also
 discusses some of the effects due to ambient
temperature.   Chapter  9  describes  the
 classical  least  squares  analysis  technique
 itself.  It starts with a description of a linear
 regression  for  a  dependent   and  one
 independent variable  and proceeds to  the
multiple  regression   case  using   matrix
notation.  Chapter 10 contains quality control
                                        1-1

-------
                                                                     TR-4423-99-03
and   quality   assurance   guidelines,
incorporating portions of an approved quality
assurance plan,  and includes selected  QA
data we collected over a recent  one-year
period.  Chapter 11 is a glossary of terms,
and Chapter 1 2 is a general bibliography of
work that addresses FT-IR  monitoring and
the principles of FT-IR spectrometry.


       Each chapter begins with a summary
highlighting  the  primary  contents of the
chapter.  This is followed by an introduction
and overview of the chapter.


1.2    References


Hanst, P.L.  1970.  Infrared Spectroscopy
and Infrared Lasers in Air Pollution Research
and Monitoring.  Appl. Spectrosc.  24:161-
174.

Herget, W.F.,  and  J.D.  Brasher.   1979.
Remote Measurement of Gaseous Pollutant
Concentrations   Using  a  Mobile   Fourier
Transform  Interferometer  System.
Opt. 18(20):3404-3420.
Appl.
Levine, S., H. Xiao, W. Herget, R. Spear, and
T. Pritchett.  1 991. Remote Sensing (ROSE)
FTIR.  In Proceedings of  the  1991  U.S.
EPA/A&WMA International Symposium on
the Measurement of Toxic and Related Air
Pollutants,   Air   &  Waste  Management
Association, Pittsburgh, PA, pp. 707-711.

McClenny,   W.A.,   G.M.   Russwurm,
M.W.  Holdren,  A.J.  Pollack,  J.D.  Pleil,
J.L.  Varns,  J.D.  Mulik,  K.D.  Oliver,  R.E.
Berkley,  D.D. Williams,  K.J.  Krost,  and
W.T. McLeod. 1991.  Superfund Innovative
Technology Evaluation. The Delaware SITE
Study,  1989. U.S. Environmental Protection
Agency, Research Triangle Park, NC.

Russwurm,  G.M., and   W.A.  McClenny.
1990.  A Comparison of FTIR Open Path
Ambient Data with Method TO-14 Canister
Data.  In Proceedings of the 1990 U.S.
EPA/A&WMA International Symposium on
the Measurement of  Toxic and Related Air
Pollutants,   Air  &   Waste   Management
Association,  Pittsburgh, PA, pp. 248-253.
                                       1-2

-------
           T.r.;s.:r =
                                                                        TR-4423-99-03

                                      Chapter 2
                       The Fourier Transform Spectrometer
                                      SUMMARY

         The major topics discussed in this chapter are the following.

            •  The basic principles of FT-IR spectrometers
               •   Resolution and throughput
               •   Detectors and sources
               •   Electronics and computer requirements

            •  The fundamental aspects of the interferogram, the Fourier
               transform, and single-beam spectra

            •  The optics used in long-path, open-path FT-IR monitors
               •   Transfer optics, telescopes, and beam return optics
               •   Monostatic and bistatic configurations

            •  Beer's law and data analysis procedures
 2.1    Introduction and Overview

        This    chapter    describes   the
 components of a complete FT-IR monitoring
 system, which include  the  following: the
 FT-IR spectrometer, the  transmitting  and
 receiving  optics,   the  electronics,   the
 computer,   and   the  data   output.   The
. discussions in this chapter are based on the
 general configurations of instruments that are
 commercially available  at  the  time of this
 writing.   There   are   currently  other
 manufacturers with instruments in the design
 or developmental stages.

       It is not necessary to have a thorough
 understanding  of  the  underlying  physics
 describing  how   an  FT-IR   spectrometer
functions to obtain reliable data with a long-
path, open-path FT-IR monitoring system.
However, familiarity with the basic principles
of FT-IR spectrometry is required if proper
operational choices are to be made under
varying  field conditions. And, the better the
operator understands the functions of the
instrument, the more likely it is that reliable
data will be produced. This chapter includes
a description of long-path, open-path FT-IR
monitors and an in-depth discussion of the
various components of FT-IR spectrometers.
The  integral  components  of  an  FT-IR
monitoring   system,  which   include  the
interferometer, detector, IR source, transfer
and  beam-return optics,  electronics,  and
computers, are described. The fundamental
processes  of FT-IR spectrometry, including
                                         2-1

-------
 MAIWM
 TECH>

                            TR-4423-99-03
 the interference phenomenon, generation of
 the   interferogram,   optical   throughput,
 resolution, and the-Fourier  transform, are
 explained. A brief discussion of Beer's law
 and its application to the data analysis  is
 provided. In addition to providing quantitative
 results, the relationships that are explained
 by Beer's law are important when estimating
 detection  limits  and determining  optimum
 path lengths.

       The heart of an FT-IR system is the
 interferometer. Most, but not all, commercial
 instruments    use   the   Michelson
 interferometer. A detailed description of the
 Michelson interferometer   is  provided  in
 Section 2.2. The trace of the output of the
 interferometer   is   referred   to   as  an
 interferogram. The interferogram is the actual
 data produced by an FT-IR spectrometer and
 contains  all  of  the  information about the
 spectrum.   However,  the   information
 contained  in the interferogram  is  not in  a
 form  that is readily recognizable  to most
 spectroscopists. To change the data into  a
 form that  is  more  easily  interpretable, the
 raw  data  are converted into a spectrum (a
 plot  of  intensity versus  wave number) by
 performing  a  Fourier  transform   on the
 interferogram. A computer system with the
 appropriate software packages is  used to
 apply   this  and   all  other  necessary
 mathematical functions to the data. Although
the execution of these calculations is virtually
 invisible   to   the  operator,   a   basic
 understanding of the principles  involved  is
 necessary  to  ensure  that  the optimum
 parameters are used to collect and process
 the FT-IR data.

       All quantitative data analysis in long-
 path, open-path FT-IR spectrometry is based
 on Beer's law. Beer's  law states that for a
 constant   path  length,  the   IR   energy
 traversing  an  absorbing medium diminishes
 exponentially    with   concentration.
 Mathematically, this is written as
where /0(v)  is the intensity of the  incident
beam,  a(v)  is   the   optical   absorption
coefficient of the absorbing material (e.g.,
target gas) as a function of wave number (v),
C is the concentration of the target gas, and
L is the path length.

       Two  primary  configurations, mono-
static and bistatic, are used to transmit the
IR  beam  along the  path,  as described  in
Section 2.3. The monostatic system has
both the IR  source and the detector at one
end of the path and a retroreflector at the
other.  The retroreflector returns the  beam
either along  or collinear to the original path,
which doubles the effective path length and
thus the measured absorbance of the target
gas. The bistatic system has the detector at
one end of the path and the source at the
other.  This  configuration  minimizes the
optical  components  that  are required for
open-path  monitoring.  However,  in the
bistatic system, the IR beam is limited to a
single pass  along  the path. Both types  of
configurations  are currently  in   use  for
environmental monitoring.
                                         2-2

-------
                                                                       TR-4423-99-03
 2.2   The Michelson Interferometer

       The primary optical.component in an
 FT-IR   instrument   is   a   Michelson
 interferometer. It is  not generally necessary
 to have a fundamental understanding of how
 the interferometer functions to obtain reliable
 data  with  an FT-IR instrument. However,
 familiarity with some of the  aspects of the
 interferometer   is   required   if   proper
 operational choices  are  to be made under
 varying field  conditions.  To that purpose, a
 brief  discussion of the  optics of the FT-IR
 instrument is included in this subsection. The
 following  major  topics  are   discussed:
 interference   (Section   2.2.1),   resolution
 (Section 2.2.2), throughput (Section 2.2.3),
 the detector  (Section 2.2.4),  and  the  IR
 source (Section 2.2.5).

       A variety of devices have been used
 over the last 200 years to study interference
 phenomena. These devices are conveniently
 classified by  the  amount  of four  primary
 attributes that they exhibit: monochrom-
 atism, fringe localization, fringe production by
 division  of wave  front  or by  division  of
 amplitude, and double or multiple beams. The
 interferometric device that today bears  his
 name  was   first   introduced  by  A.  A.
 Michelson in  1881  (Michelson  1881). It is
the   most   famous  of   a   group    of
 interferometers that  produce interference
fringes  by the division  of  amplitude. 'Four
years    after   Michelson   introduced  the
interferometer, it was shown that the Fourier
transform of  the  interferogram  was  the
original spectrum or intensity as a function of
wavelength. The  Michelson  interferometer
 has been used to define  and measure  the
 standard  meter,  to measure the  angular
 separation of binary stars,  and to provide the
 experimental  data  for one  of  the  four
 cornerstones  of  relativity  theory.  During
 recent times, the  Michelson interferometer
 has been  used successfully to measure  the
 concentrations of  various  chemicals  that
 absorb  energy in  the  IR portion  of  the
 electromagnetic spectrum. (See Chapter  10,
 Bibliography.) It is currently being developed
 as   an   instrument  to   make   similar
 measurements over extended open  paths,
 and  it  is   in   this  context  that   the
 interferometer is discussed here.

       A schematic of the simplest form of a
 Michelson   interferometer  is  shown   in
 Figure 2-1. It consists of a  beam splitter and
 two mirrors,  one  of them movable.  The
 figure also shows an arrangement  for  the
 light source  and the detector. For the most
 accurate use, the two mirrors must be kept
 perpendicular to one another. One of the two
 mirrors moves along the optic axis. During
this motion  the  perpendicularity  cannot
change. This requirement  can represent a
stringent  limitation  for  the  mechanisms
involved with the  motion. The light incident
on  the beam splitter should be collimated,
because uncollimated light  gives rise to poor
resolution.

2.2.1 Interference

      This   section   is   presented   for
completeness and  because there seems to
be   some  confusion  as  to   how  the
interferogram   arises.   It   is  somewhat
                                        2-3

-------
                                                                       TR-4423-99-03
                Moving
                Mirror
 •a
 01
 X
                                 Detector
               Infrared
               Source
   Figure 2-1. A Schematic of the Simplest
   Form of a Michelson Interferometer.
 mathematically rigorous and can be omitted
 without  jeopardizing  the  ability  of  the
 operator to obtain reliable FT-IR data.

       Interference is the underlying physical
 phenomenon that allows a Fourier transform
 instrument to obtain spectrometric data. The
 interference   phenomenon   cannot  be
 physically explained by the simple addition of
 the intensities of two or more optical beams.
 The amplitudes of the individual interfering
 beams  must, be added according to  the
 principle of  superposition,  and  the  total
 intensity must be calculated from that result.
 Interference  phenomena  are   linear   in
 amplitude.   The   principle   of   linear
 superposition,  which  is operating  here,
 follows directly from Maxwell's  equations
 and the fact that these  equations are  linear
 differential equations. To arrive at the  basic
 equation that describes  how the Michelson
 works,   consider  the   arrangement  of
 Figure  2-2.  A monochromatic electromag-
 netic plane wave is incident on  a device at A
that  divides   its   amplitude   into  two
 components. After the division, the individual
 beams  traverse a medium  along  different
 paths  and are somehow recombined  at  a
 point P in space. On arrival at point P, the
 two beams,  which need not be collinear,
 have the following amplitudes.

             ^  _  ^ gi(u/-27M7yA.)
       The two A0 terms are the amplitudes
of the individual beams, the co is the angular
frequency of the radiation, n is  the index of
refraction of the medium,  and  the two T
terms are the physical path lengths that each
beam has traversed. The product nT is called
the optical path length that the beam has
traveled. At point P, where the two beams
are recombined, the total amplitude is the
sum of these  terms. The intensity is then
given by the product  of this  sum and its
complex conjugate.  Thus the  intensity at
point P is given by Equation 2-1 .
                                    (2-1)
The first two terms are the intensities of the
original two beams, and the last two terms
are called the interference terms. When the
amplitudes AQ and A0' are equal, they can be
combined.
                                        2-4

-------
                                                                       TR-4423-99-03
        Source
             Plane
             Wave
Amplitude
 Divider
       Figure 2-2. Schematic of Interference Created by Division of Amplitude.
       Path 1  has physical length 7,, and Path 2 has physical length 72.
       By using the relation 2cosx = e" + e*, the
intensity at point P is given by Equation 2-2.
               cos
                                    (2-2)
       Here 70 is the intensity of either beam.
Thus as the  difference of the path length,
T2 - 7,, changes, the intensity at point P can
vary from 0 to 4/0. The fact that/can be 4/0
does not violate the  conservation of energy
law. There is no  physical requirement that
the intensity  at every point in space be 2/0.
The requirement is that the interference term
averaged  over  space must  be zero  (Rossi
1957).

      When  a plane monochromatic wave is
incident  on  the  beam  splitter of  the
Michelson  interferometer,   the   amplitude
ideally is evenly divided along each leg. At
any  position of . the  moving mirror, the
detector output is.proportional to the integral
of the intensities over wavelength, and this
                      recording is  called the  interferogram. From
                      Equation  2-2, it  is seen that at zero path
                      (T2 - r, = 0) difference,  the cosine term  is 1
                      for   all   wavelengths.   Thus  for   all
                      wavelengths, the intensity  is 4/0, and  the
                      output  of the detector is large compared to
                      any other mirror  position.  This is quite
                      noticeable in  the  interferogram  and is
                      commonly called  the center  burst.

                            This center burst  does  not  appear
                      when   the   radiation   is   monochromatic.
                      Figure 2-3 shows  how the center burst builds
                      as the  wave number range  is expanded to
                      include   more   wavelengths.   The
                      interferograms in  this figure were calculated
                      from Equation 2-1 in the following way. All
                      the wave  numbers have the same intensity
                      and  add  incoherently.  The wave number
                      value was stepped in increments.of 0.1 cm"1.
                      The retardation (actually,  the term T2 - 7,)
                      was taken in increments of the wavelength
                      of a He-Ne laser. At each  position  of  the
                      mirror  the proper phase for  each  wave
                                        2-5

-------
                                                                      TR-4423-99-03
 B
D
                     T
Figure 2-3. Center Burst Increasing as the
Wave Number Range Expands.

number was used to calculate the intensity,
and then  the  intensities were added. The
interferograms were actually calculated for
6000 incremental movements of  the mirror;
however, only a portion of the data is shown
for clarity. The interferograms in  Figure 2-3
are for (A) a 2-cm"1range, (B)a 50-cm"1 range,
(C)a 500-cm'1 range,  and (D) a  3500-cm"1
range.  The  two. interferograms  shown in
Figure 2-4 are for a range of 3500 cm'1, but
curve B has  a  1500.K  blackbody radiation
curve superimposed on it, and it appears quite
similar to the interferogram actually recorded
by the FT-IR spectrometers.

      Equation 2-2 shows that as the mirror
moves,  the  path  difference   causes  a
modulation   of  the  intensity  at   each
wavelength. The modulation can  be used to
advantage in open-path FT-IR monitors. For
example,  if  the  IR   beam traverses  the
interferometer before  it is  sent  along the
open path, any background radiation entering
Figure 2-4. Interferograms for a Range of
3500 cm'1. Interferogram B has a 1500  K
blackbody radiation spectrum
superimposed on it.
                                        2-6

-------
•••jrr «•?../».•'« =
                                                                       TR-4423-99-03
 the system  from the  surroundings is not
 modulated and will not be processed by the
 electronics.  However, a  portion  of this
 unmodulated light will still be incident on the
 detector  and in extreme  situations  could
 cause the detector to become  saturated.
 Therefore, it is prudent to avoid setting the
 instrument up along a path that  includes
 bright (hot) IR sources.

 2.2.2 Resolution

       The  resolution  of  an  instrument
 determines   how   close  two  absorption
 features can be and still be separated enough
 for analysis. There are several criteria for this
 instrument parameter,  but  the one  most
 often  used  for  the  FT-IR  instrument  is
 described below. Equation 2-2 shows that all
 wavelengths  have  a maximum  and are  in
 phase  with  one   another  at  zero,  path
 difference. The most common definition of
 resolution for the FT-IR spectrometer states
 that two absorbing features centered at
 wavelengths A, and A2 will be resolved if the
 mirror moves  at least to the point where
these two wavelengths are again in phase.
To determine when this occurs, the following
example may be  considered. If  only  two
spectral features situated at A, and A2 make
up the spectrum, then the interferogram is
made up of two spectra, each described by
Equation 2-2. The result of adding these two
spectra is shown in Figure 2-5 and  is given
by Equation 2-3.
                                    Figure 2-5. Interferogram of Two Cosine
                                    Waves vs. AT. The wavelengths differ by
                                    10 cm"1. The minimum occurs when the
                                    two  waves are  180° out of phase.
                                          The  second cosine  term produces a
                                    high-frequency signal that is modulated by a
                                    low-frequency signal described  by the first
                                    cosine  term.  It is the first term that is of
                                    interest when determining  the resolution of
                                    the system. The signal is a maximum when
                                    the  argument of this cosine term  is 2Mi,
                                    where N =  0,  1, 2, ... . Thus, setting n the
                                    index of refraction equal to 1, the first time
                                    that the two wavelengths  are in phase after
                                    the center burst is when N  - 1,  so that

                                               7tA7T.l/A2- I/A,) = 2n

                                    This implies  that

                                                A7 = 2/(l/A2-l/A,)
                                        2-7

-------
                                                                       TR-4423-99-03
 However, the term in the denominator is the
 difference  in the  wave  numbers  of  the
 absorption peaks,.so that AT" = 2/Av. Thus, if
 the operator  desires a resolution of 0.5 cm"1,
 the  optical path  difference must be 4  cm.
 Because the  beam traverses the path in the
 interferometer twice, the actual motion of the
 mirror must be only 2 cm. It should  be noted
 here that this is an idealized result.  The fact
 that the interferogram  is first truncated and
 then apodized changes this result somewhat
 (Marshall and Verdun 1990; Beer 1992).

       The  question  of  what  resolution
 should be used for a specific data collection
 task is not addressed in this section. It is
 discussed in  more depth in Chapter 7 and
 Chapter 8. The  answers to the  resolution
 questions are specific  to  the gases to  be
 monitored and the effects of water vapor in
 their regions  of absorbance. At the present
 time,  each  monitoring  situation  must  be
 considered separately.

 2.2.3  Throughput

       The throughput of an optical system is
 defined as the product  of the area of an
 aperture A and the solid angle Q of the light
 beam  at  that aperture. This  quantity  is
theoretically   a  constant  throughout the
system, so that  once it is defined for an
aperture it is  known for all apertures.  For
small angles, the solid angle of the beam can
be shown to  be  equivalent to the  product
ir62, where 6  is the half angle of the field of
view of the instrument. It can be shown that
the throughput is related to the f# of the
system by recognizing  that 9 = 1/(2/#), so
 that the throughput is equal to An[M(4f#)]2.
 With FT-IR instruments, the selection of the
 system 1# is generally a compromise. An
 important consideration is the solid angle of
 the beam as it traverses the interferometer.
 A  portion  of  the  beam  traversing  the
 interferometer at  a  large angle will travel
 over   a    longer  path   through   the
 interferometer, and a beam traversing at a
 smaller angle will travel over a shorter path.
 This angular dispersion tends to degrade the
 resolution of the instrument, because energy
 at the  same wavelength  appears  to  the
 interferometer as though it covers a range of
 wavelengths.  Smaller  f#s are   at  first
 attractive  because  they  indicate  that  a
 smaller aperture  can  be  used.  However,
 small  f#s imply  large  solid  angles and
 therefore   a   loss   in   resolution.  The
 manufacturers  of  these instruments have
 taken this into  account in  the  instrument
 design,  but nevertheless the aperture size is
 fixed,  and once a specific instrument is
 purchased, there  is  little,  if anything,  the
 operator can do to change the throughput.

 2.2.4 The Detector

      The   detector   in   most   FT-IR
 instruments used for monitoring atmospheric
 pollutant gases  is  a  semiconductor device
 made of mercury, cadmium, and  telluride,
 commonly called an MCT detector. There are
three modes of operation for this device, as
 a photovoltaic device,  as  a photoelectro-
magnetic device, and as a photoconductive
device.  The MCT photoconductive  detector
 is the  one most often used in the  FT-IR
instrument. This device converts a  beam of
                                        2-8

-------
MAIWi
           ::
                                                                      TR-4423-99-03
 photons to an electrical current that can be
 measured. In addition to the spectral region
 that the detector responds to, the two most
 important parameters of the detector are the
 noise  equivalent  power  (NEP)   and   the
 sensitivity of  the detector  in  terms of a
 quantity called D* (pronounced "Dee Star").
 For  an MCT photoconductive  device,  the
 spectral response ranges from 2 to 20 fjm, or
 from 500 to 5000 cm'1. The NEP is given in
 terms of W/(Hz)'/1. For the detectors used in
 FT-IR instruments, this parameter has a value
 of about  5  x  10~12 .  The user should be
 aware  that this  parameter  represents a
 measure of the inherent noise in the detector
 and that small numbers are better than large
 numbers. The  D*  is   a  measure of   the
 sensitivity of the detector and has  units of
 cm(Hz)'/1/W. This number is actually defined
 as the ratio of the square root of the detector
 area to the  NEP, or D*  = JA/NEP. For  the
 MCT photoconductive detectors, this number
 is  about 5 x  1010 at  10 /urn. Here, larger
 numbers are better. However, there is little
that the operator can do about the magnitude
of these parameters once the instrument is
purchased. But  if the  detector has to  be
replaced,  some  acceptance criteria for the
NEP  and the D* should  be specified.

      There are, in  general, two types of
MCT detectors  available, wide  band  and
narrow band. Each has somewhat  different
characteristics.  Wide band MCT detectors
cut off  at around 500  cm"1, whereas  the
narrow  band MCT  detectors cut off  near
600  cm"1. For long-path measurements the
region  below 722  cm'1  is nearly  opaque
because . of absorption  by  CO2   in  the
                                          atmosphere, so a wide band detector does
                                          not offer any real advantages. Also, the D*
                                          for the narrow band detector is 5 to 10 times
                                          higher than that for the wide band detector.
                                          There is  also an indium antinomide (In-Sb)
                                          detector that can be used in the higher wave
                                          number  region   with   advantage.   One
                                          particular application for using this type of
                                          detector is the measurement of HF.

                                                An important  requirement  for  the
                                          detectors in FT-IR monitoring systems is that
                                          they must  be cooled to  operate properly.
                                          Liquid nitrogen temperatures  (77  K)  allow
                                          optimum  operation  of  these  detectors.
                                          Currently, two techniques  are used to cool
                                          the detector.  The  first  is  to place  the
                                          detector in a Dewar that uses liquid nitrogen
                                          as a refrigerant.  For this mode, a supply of
                                          liquid  nitrogen must be available for use in
                                          the field, and the  operator must  fill  the
                                          Dewar periodically.  This has  not  been a
                                          major  problem in the  past, as the  liquid
                                          nitrogen requirement is only a few pints per
                                          day of operation. The second technique to
                                          achieve  cold  temperatures   is  with  a
                                          cryogenic cooler,  such as a Stirling engine, a
                                          Joule-Thompson  cooler, or a closed-cycle
                                          helium refrigeration system.  In the  Stirling
                                          engine, the heat is exchanged through a wall
                                          from the enclosure to cool the gas. Currently,
                                          the major problems with this cooling device
                                          is the mean time between failures is too
                                          short,  and these coolers seem to add noise
                                          to the  spectra. One-half year of continuous
                                          operation is  about the maximum that can be
                                          expected.  If  unattended  operation  is  a
                                          necessity, the Stirling engine  is one  choice.
                                          The Joule-Thompson  cooler  forces  dry
                                        2-9

-------
                                                                        TR-4423-99-03
 nitrogen through  an  orifice, after  which it
 expands   and  cools.  This  device  uses
 nitrogen,  which has to be of high purity, at
 the  rate of about one  300 size  cylinder in
 40  h  of  operation.  Other  options for
 unattended operation  are the use of larger
 detector Dewars  with  longer hold times or
 devices that automatically refill the detector
 Dewar.

 2.2.5  The IR Source

       All IR sources that are available today
 for use with the  FT-IR monitor are heated
 elements  that are open to the  atmosphere.
 They  are   resistive  devices that  radiate
 approximately as black body radiators. These
 devices operate at a color temperature from
 1 200 to 1 500 K. The Wien displacement law
 states that the product of  the wavelength of
 maximum power output and the temperature
 of the source, Amaxr,  is a  constant equal to
 0.2987 cm K"1. This indicates that there is an
 inverse relation  between  peak wavelength
 and  source temperature.  Planck's radiation
 law shows that there is more energy at all
 wavelengths for hotter sources.  The  ideal
 would be  a source that is at about 3000 K,
 so it would have  a peak  at about 1.1 //m.
The  materials that are  necessary to make
 such a source have not been available until
 now.

       Perhaps   the    most   detrimental
characteristic of the available sources is that
they are large compared to the focal length
of the collimating  optics.  From geometrical
optics, it is  clearly seen that the  beam can
never be.  better, collimated  than  the  angle
 that the source subtends at the collimating
 optics (lens  or  mirror). Thus,  all  available
 FT-IR spectrometers have beam divergences
 that are too  large for the  rest of the optics.
 This means that retroreflectors or receiving
 optics are overfilled, and much of the initially
 available  energy  is  lost. Ultimately,  this
 divergence restricts the path length that can
 be used to advantage. Perhaps as  further
 developments occur, a small hot source will
 be  developed  that  will  minimize   this
 difficulty.

 2.3    Transfer Optics, Telescopes, and
       Beam-Return Optics

       There  are two primary geometrical
 configurations available for transmitting the
 IR beam along the path.  One is a bistatic
 system (Figure 2-6); the other a monostatic
 system (Figure 2-7). The monostatic system
 has both the  IR  source and the detector at
 the same  end of  the path,  whereas  the
 bistatic system has the detector at one  end
 of the  path and  the source  at the  other. In
 the bistatic system, the optical path length is
 equal to the physical path length, whereas in
 the monostatic  configuration,  the  optical
 path length is twice the physical path length.
 In this document, we always refer  to  the
 optical path length.  The reflecting optics for
 a  bistatic  FT-IR  monitoring  system  are
 relatively   straightforward  (see  Section
 2.3.1), whereas the optics in a monostatic
 system may include one or more telescopes,
 an additional beam splitter, and return-beam
optics. The possible configurations for a
monostatic   system   are  described   in
Section 2.3.2.
                                        2-10

-------
MAP/:
                                                                         TR-4423-99-03
  A)
Detector

Receiving
Optics

» i
i i
IR Path \
-^--v 	 « 	 \
\ * «
* * *

Transmitting
Optics


Interferometer


"*---.
-^"•""" IR
Sou re
  B)
    Detector
                                 t
                           Absorbing Medium

                                  X
                                                                      .-*
                                                                        IR
                                                                      Source
                                    ....'••••"
                 Figure 2-6. The Bistatic Configuration.
                                   2-11

-------
                                                                    TR-4423-99-03

   A)


/
• 1
f f

\


Translating
Retroref lector
*"•"*"»

/ Transmitting IR Pattf
« »

*• \ W.
7 Return IR Path i--**





Transmitting
Optics



Receiving
Optics





Interferomet*



*-&


Detector


;t








^
..-*
-*•*" IR
Source






                Absorbing Medium
   B)
Retroreflector   *



....
4
/
*
*
.—
.'•••""••
4
* **•»*«.*...
** *
IR Path

« .

'-.t
«
•
— A1*-.-

«
*,



..^^
«
t
•
^^"
4




rransmitting/Recetvtng
Optics





^
\

• i
Additional
Beam Splitter
/•
r
Interferometer





-^"- .
**
.-•*
•^•-*" IR
Source

                                               Detector
             Figure 2-7. The Monostatic Configuration.
                                2-12

-------
                                                                        TR-4423-99-03
       There are several types of telescopes
 that could be used to transmit and collect the
 IR  beam.  Those in current  use are the
 Cassegrain  and the Newtonian telescopes.
 They are optically  equivalent, and the only
 difference is the placement of the diagonal
 mirror that  removes  the  beam from the
 telescope. In the Cassegrain, the beam exits
 from the end of the  telescope, and  for the
 Newtonian  it exits from the side. Optically,
 the number of reflections is the same, and
 otherthings being equal, the reflection losses
 are   also  essentially  the   same.  The
 geometrical  configuration  for  the  whole
 instrument is slightly  different for these two
 designs. In one, the overall package enlarges
 vertically, whereas in  the  other  it grows
 longitudinally. These  are minor  points as far
 as instrument operation is concerned.

       In  most  cases, the beam is expanded
 before  it is sent along the path,  and in
 principle there  are essentially no size limits.
 The  physical quantity that is of interest to
 the operator is how much of the IR beam is
 absorbed by the gases of interest. This  is, in
 general, very small (about 1 part in a  1000).
 This fact is not changed by expanding the
 beam. In general,  beam expansion  allows
 more energy to  be transmitted  along  the
 path, thereby increasing the overall signal-to-
 noise ratio  (S/N).  There  are,  of  course,
 practical limitations to the size of the optics
that can be  accommodated.
 must reduce the size of the beam so that it
 can pass through the system to the detector
 without vignetting and also to set the solid
 angle  of the  beam so that the resolution
 remains acceptable.

 2.3.1  Bistatic System

       The bistatic configuration minimizes
 the optical components that are required for
 open-path monitoring. At  the source end of
 the  path there  must be  some  method for
 collimating the beam. This can easily be done
 with a mirror shaped as a  parabola or one of
 the other conic sections. At the receiving end
 of the path, a collector, similar in design to
 the collimator, may be used to transfer the
 beam to the interferometer and the detector.
 In commercially available instruments,  the
 diameter of the collector generally is  the
 same as that of the transmitter,  although
 there is no optical  necessity for this choice.

       There are two configurations that can
 be   used   for   bistatic   systems.   One
 configuration    places   the   IR   source,
 interferometer, and transmitting optics at one
 end of the path and the receiving optics  and
 detector at the other end (Figure 2-6A). The
 advantage of this configuration is that the IR
 beam is modulated along the path, which
 enables   the   unmodulated   background
 radiation to be  rejected  by the system's
 electronics.
      As a rule, the optics in either system
are reflecting  optics rather than refracting
optics to avoid transmission losses. Once the
IR energy  has been  collected,  the  optics
      The other configuration places the IR
source and transmitting optics at one end of
the   path  and   the   receiving  optics,
interferometer, and detector at the other end
                                        2-13

-------
 TECH£jUltm
                                                                       TR-4423-99-03
 of the path (Figure 2-6B). This is the more
 common configuration of bistatic systems in
 current use.  The -main drawback to. this
 configuration  is that the IR source is not
 modulated before it is transmitted along the
 path. Therefore, the system has no way to
 distinguish between the active IR source and
 the background IR radiation.

       Still another version is being tested at
 the present time. In this bistatic system the
 light   source  is  modulated  by   some
 mechanical  means.    This   configuration
 requires   a   feedback   loop   to    the
 interferometer (via radio link or optical cable)
 so that proper phasing can be obtained.

       Because the IR radiation makes only a
 single  pass through the optics, less of the
 radiation is lost in bistatic systems. Also, this
 makes  the  systems  more  amenable  to
 passive measurements, such  as  emission
 measurements  or absorption measurements
 with a natural  background hot source.  For
 various reasons, this single pass through the
 absorbing  medium is  not seen as  a severe
 drawback.

       Ambient monitoring in confined areas
 or rough terrain may  pose certain logistics
 problems with a bistatic system. One is that
this mode of operation  requires two power
 sources, one  at  each end  of  the path.
Another is  that  there is only  one  pass
through the absorbers. The absorbance for
most gases of interest is very small, and for
short  paths, as  those encountered  with
plumes, a  single pass through  the  gas may
be insufficient.
 2.3.2 Monostatic System

       There are currently two techniques in
 use for returning the beam along the optical
 path when the monostatic mode is used. One
 is to set up an arrangement of mirrors that
 translates the  beam  slightly for its return
 path, and another is to place a retroreflector
 array at the end  of  the path. These two
 configurations are optically equivalent.

       In the first configuration,  an optical
 system is placed at the end of the path that
                 /
 translates the IR beam slightly so that it does
 not  fold back on itself (Figure 2-7A). The
 receiving end then  has a second telescope
 slightly removed from the transmitter with
 the  detector  at  the  primary  focus. This
 technique circumvents a possible objection to
 the second monostatic configuration.

       The  second  configuration  for the
 monostatic monitoring mode uses the same
 telescope  for  the transmitting  and  the
 receiving optics and  uses a retroreflector
 array at the end of the beam (Figure 2-7B). A
 retroreflector is an  arrangement of mirrors
 that reflects the beam so that the  incident
 and  reflected directions of propagation are
 collinear but opposite to one another. It is
 made of three  reflecting surfaces that are
 mutually orthogonal, such as the floor and
 two adjacent walls of a room.  It is often
 assumed that  after  reflection  the beam
 returns  along the same path  over which it
 was transmitted. That is true only in a gross
 sense because if the beam is small compared
to the  reflecting surfaces, a  measurable
translation takes place, and there  is always
                                        2-14

-------
                                                                       TR-4423-99-03
 an inversion and a reversion of the beam. In
 order to transmit and receive with the same
 optics,  a beam splitter must be placed in the
 optical   path.   An   objection   to  this
 configuration  is  that  the  IR  energy must
 traverse this beam splitter twice, once on the
 transmitting end  and once on the receiving
 end.  The  most  effective  beam   splitter
 transmits 50% of the light and rejects the
 other  50%.   Thus,,  in  two  passes,  the
 transmission  is only 25%  of the original
 beam. Because this  loss affects the S/N, it
 may  be a  significant  drawback  of this
 configuration of the monostatic mode.

      Because   of   the  design  of   the
 orthogonal  mirror retroreflector,  it  is quite
 insensitive to  small motions such as those
 caused  by a wind. Retroreflector  arrays are
 also  very   easy   to  align   with   the
 transmitting-receiving telescope,  and a few
 degrees  of  misalignment  will  pose  no
 problem to the operator. This does not seem
 to  be the  case  with  the  spherical mirror
 configuration.  Although this arrangement is
 also easy to align, it seems somewhat more
 sensitive to a small error.

 2.4   The Electronics

      Ah   in-depth    discussion   of  the
 electronics of the FT-IR system is beyond the
scope of this document. However, some
points that are of interest to the operator are
covered. The interferogram is in the form of
intensity versus  position  of  the  moving
mirror. As the  mirror moves,  the detector
measures a varying intensity, and this signal
is first amplified and then sent to an analog-
 to-digital  (A/D) converter for digitization.
 There are two important features that pertain
 to this digitization  process.

       The first concerns the dynamic range
 of  the  signal, which  can actually  be  too
 large. There must be enough resolution in the
 A/D converter so that the least significant bit
 can always be reserved  for recording  the
 noise in the system. If  this not the case, the
 spectrum derived by performing the  Fourier
 transform  on  the  interferogram will   be
 distorted. For this reason, most commercially
 available instruments today use a  1 6-bit A/D
 converter. Higher range converters exist but
 the  trade-off  comes in the noise and  the
 speed of the device. So a 20-bit converter
 may not offer much of an advantage over a
 16-bit converter. This means that if the noise
 is recorded  on the  least significant  bit,  the
 highest  signal  that can be recorded  is  a
 factor of 215  above the noise. This is  not
 really as high as it seems. For example, if the
 noise is about  1 /nV, the largest signal that
 can be recorded is  about 0.5 V.

      The   second feature concerns  the
 amount of  data that has to be recorded in
 digital form so that the original  waveform
 can  be reproduced. There is some relation
 between the rate at which a signal  varies and
the  number  of sampling pulses that  are
needed to reproduce it exactly. The sampling
theorem from modern communication theory,
sometimes   called   the  Nyquist  theorem,
states that at least  2/m  equally  spaced
samples  are   needed  each   second   to
reproduce the  waveform without distortion.
Here  fm   is   the   maximum  frequency
                                        2-15

-------
                                                                       TR-4423-99-03
 component that is contained in the original
 waveform.

       As  the  mirror  of  the  Michelson
 interferometer  moves, each  wavelength is
 modulated at a frequency that is related to
 the velocity of the mirror and the particular
 wavelength. Since the light must traverse the
 distance from the beam splitter to the mirror
 twice (down and back) in any one cycle, 2
 times the actual velocity can be used for the
 determination  of the  frequency range to
 which the electronics must respond. If  the
 mirror moves at a speed of 1 cm/s  and  the
 wavelength range is from 2.5 /xm to 20 /*m,
 then the frequency range is from 8000 Hz to
 1000 Hz.  Thus,  according  to the  Nyquist
 theorem, the digital sampling rate must be at
 least  16000 equally  spaced  samples   per
 second.

      There is also a need for measuring the
 position of the mirror and for signaling to the
 electronics when to record data. This is done
 by using an He-Ne laser. The laser beam is
 sent  through   the   interferometer   and
 modulated in the same  way as all other
 wavelengths are. The amplifier output is
 capacitively coupled  to  the  rest  of  the
 electronics so that the He-Ne interferogram
 is  an  ac signal with negative and  positive
 parts. The zero crossings of this ac signal are
 sensed,  and the instrument  records a data
point  at the zero crossings. Electronically,
zero  crossings  are  easier  to  detect than
maxima   or  minima because  of the  sign
change.
 2.5   The Computer

       The final  requirement  for an  FT-IR
 system is a computer.  This  discussion is
 meant to be a discussion  of  the minimum
 requirements only.

       The data storage requirements depend
 on the resolution used, and  the  capacity
 must be fairly  large if the  interferogram is
 stored. This requires about 100 kilobytes for
 each  interferogram  recorded  at   1-cm'1
 resolution, and some means for archiving the
 data  must  be  available.  Most software
 packages that  are currently  available are
 written for Windows'95, so a computer with
 a Pentium processor is required. For ordinary
 field  work  at  monitoring  sites such  as
 Superfund sites or waste  sites, the ideal
 system seems to be either a single computer
 with the ability to operate in both  foreground
 and   background   modes  or   two
 computers—one to control the instrument and
 to record  the data  and the other, a much
 more powerful machine, to be  used for data
 analysis.

       In addition to the computer hardware,
 a  software  package  is  required that  will
 control the FT-IR  system and  record either
the interferogram (preferably) or  the single-
 beam spectrum  (Section 2.7.4) produced by
the  Fourier  transform  performed on  the
interferogram. There is currently one generic
software package that can be configured for
any  of the  commercially  available  FT-IR
systems.   Other   FT-IR  systems   use
proprietary software for data collection. In
addition to data  collection, the software
                                       2-16

-------

                                                                        TR-4423-99-03
 should also  provide  some  means  for data
 analysis.   This   is   discussed  further   in
 Section 2.6.5.

 2.6   The Data Output

       This section contains a discussion of
 the interferogram generated  by the FT-IR
 system, the Fourier transform that is applied
 to  the interferogram, and the single-beam
 spectrum  into which the interferogram  is
 transformed.  The data reduction  process,
 starting with  the interferogram  and ending
 with  the  unknown  gas  concentration,  is
 described. A discussion of Beer's law of how
 energy  diminishes   as   it  traverses   an
 absorbing medium is presented first.

 2.6.1  Beer's  Law

       The fundamental physical law that  is
 quoted as the basis for all FT-IR data analysis
 is Beer's law (Beer 1852). It states  that, for
 a constant path length, the intensity of the
 incident light energy traversing an absorbing
 medium   diminishes   exponentially   with
 concentration. Mathematically, this is written
as
       v) = I0(v)e
                    -a(v)CL
(2-4)
where 70(v)  is the intensity of the incident
spectrum,   a   is  the  optical  absorption
coefficient of the gas and is a function of the
wave number v, C is the concentration of the
gas, andZ, is the path length. There are many
possible  sets  of units for  these quantities
that are  variously used by the workers in
 FT-IR  open-path  monitoring. Whatever the
 set chosen, it must be noted that the product
 aCL must be a unitless quantity. Thus, if the
 absorption coefficient has units of (cm-atm)'1,
 the  concentration must  be in atmospheres
 and the path length must be in  centimeters.
 One primary difficulty that confronts the user
 of FT-IR open-path monitors is determining
 the quantity/0. This  is discussed in detail in
 Chapter 4, Background Spectra.

       The mathematical functional form of
 Beer's   law   explains    many  physical
 phenomena.  These  include   atmospheric
 pressure as a function of  altitude, thermal
 expansion of metal rods, radioactive decay,
 and the electrical  discharge of capacitors, to
 name  but a  few.  Although- there is no
 physical  basis for doing  so, in the field of
 optical spectroscopy, this functional form is
 often stated by using logarithm to the base
 10. The available analysis software also uses
 logarithms  to the  base  10. To understand
 how the change is to be made, consider the
 following.

       The fundamental formula is Y = a?.
The problem is to  solve this equation for the
quantity X, which is the power that a must
 be raised to  obtain  Y. To  solve  this,  the
concept of logarithms is  introduced so that
\oga(Y)  =X\oga(a). This  is read  as  "the
logarithm of /to  the base  a equals Xtimes
the  logarithm of  a  to  the base a."  By
definition, \oga(a) = 1 so that

               X=\oga(Y}
                                        2-17

-------
                                                                       TR-4423-99-03
 Logarithms can be determined by using any
 number  for the  base,  but  only  two are
 commonly  used in the  physical sciences.
 They are the  base 10 and the base e. The
 number e is defined as the limit of (1  + MN)N
 as N goes to  infinity. The number e occurs
 naturally in mathematics, and particularly it
 occurs naturally in equations like Beer' law.
 The question is then how must Beer's law be
 written to account for the change  to base
 10.

       The logarithmic form of the equation
 Y = c? can be written in any other base b
 such that

             \ogb(Y) = X\ogb(a)

 Solving this for X gives

            X = \ogb(X)/\ogb(aJ

 Substituting this X into the original equation
 gives

          \ogb(Y) = \oga(Y)\ogb(a)

 In terms of the bases e and 10, this is written
 as

          Iog10/r; = \OQ,0(e)\ogem

       Historically,  natural  logarithms are
written   by  using  the  prefix  In,   while
logarithms to the base 10 are written with
the  conventional  log  as  a  designator.
Logarithms to all other bases also use the
convention log as nomenclature, but then the
actual base is  specified as a subscript.
       If the power of 1 0 is used, Beer's law
 is written as follows.
 For convenience, the exponential term in this
 expression is defined as the absorbance. The
 mathematics given here is transparent to the
 operator and of little significance throughout
 the remainder of this document. However, it
 should be noted that when absorbances are
 given as numerical  values, the logarithm to
 the base 10 has been used.

       Although most FT-IR workers cite and
 discuss Beer's law, it is not directly used.
 The  absorption coefficient is  generally not
 known. One implication of Beer's law that is
 used is the concept of reciprocity. That is,  if
 the concentration diminishes by a factor of 2
 but the  path length  increases by a factor of
 2, the  measurement  will  yield the same
 results.  This is  not always true,  and  it is
 generally accepted that if the  quantity aCL
 becomes larger than about 0.1 , the concept
 of reciprocity is no longer valid.

      Beer's law has been restated so that
 it  includes  many  applications,   and,  as
 restated, the law has assumed several other
 names,  as the  Lambert-Beer law or the
 Bouguer-Lambert-Beer law, but these other
 names are not correctly used. Beer wrote the
 law for a purpose other than the way  it is
 used today. When Beer published his original
 work   in  1852,   he   was   conducting
 experiments  designed  to  measure  the
 absorption  of  various  materials  that were
then being  used in the field of photometry.
                                        2-18

-------
                                                                        TR-4423-99-03
 His  entire endeavor was directed  toward
 investigating the effects of the thickness of
 a material. He therefore did not write the law
 in terms of concentration, nor is there any
 evidence that he considered the effects of a
 changing  concentration.  The law  as used
 today, with concentration as an explicit term
 in the exponent, seems to have first been
 published by B. Walter almost 40 years after
 Beer's original work (Walter 1889).

 2.6.2  The Interferogram

       The  primary data  produced  by  an
 FT-IR instrument is the interferogram, and it
 is the piece of data that should be recorded.
 However,  the  mechanics  of  the  FT-IR
 instrument itself can alter the appearance of
 the interferogram, and this influence must be
 accounted for during data analysis. Two of
 these effects, truncation and phase shift, are
 discussed below.

 2.6.2.1 Truncation

       As discussed earlier, the interferogram
 is the intensity measured by the  detector as
 a function of the  position of the moving
 mirror. It contains all the  information about
 the  spectrum  that is  familiar to  most
 operators. In actual operation, the mirror in
 the interferometer moves, at most, a few
 centimeters and stops and then returns to its
 original position.  This   finite  movement
 truncates  the interferogram at each  end. It
 can be shown that this truncation  actually
 limits the sharpness of the absorbing features
that are of interest to the experimentalist.
This is analogous to creating a square wave
 from a series of sine and cosine functions.
 There it is  seen that the higher frequency
 components sharpen the edges of the square
 wave. The  situation with the FT-IR data is
 identical. The information at the ends of the
 interferogram is really information about the
 high-frequency   components.  A   simple
 truncation (stopping the mirror motion after
 a  certain distance) behaves mathematically
 as though the interferogram were multiplied
 by what is called a boxcar function. That is,
 the interferogram is multiplied by a  function
 that is 1 in a region from mirror position 1 to
 mirror position 2 and is  zero elsewhere. The
 effect of this multiplication is to broaden the
 spectral line features. Truncation also causes
 a phenomenon called ringing in the wings of
 the spectral features. That is, the truncation
 of the interferogram adds oscillations into the
 wings  of  the  spectral  features.  These
 unwanted  features can  be  removed  by
 applying an  apodization  function  to  the
 interferogram  prior to the Fourier transform.
 There are several apodization functions (see
 Filler or Norton and Beer) that can be applied
 to the data, but an in-depth  description is
 beyond the scope of this chapter. They are
 addressed briefly in Chapters  7  and 8. The
 point is that the operator should be aware of
 this effect and  that some choices  can  be
 made during data analysis that will affect the
 shape and intensity of the spectral features.

 2.6.2.2 Phase Shift

       A second instrumental  effect on the
 data that occurs is a shifting in the relative
 phase of the wavelength, which is caused by
the optics and the electronics. There  are two
                                        2-19

-------

                                                                       TR-4423-99-03
 components for  each of the wavelengths
 whose intensities are added to make up the
 interferogram. They are the magnitude of the
 intensity and the relative phase. The optics
 and the electronics cause slight phase shifts
 as the signals are processed, and these are
 frequency dependent. The shifts are normally
 accounted for when the Fourier transform is
 done,  but  no  record of  them  is  saved.
 Therefore,   if  the   interferogram  is   not
 recorded, it cannot be retrieved  by simply
 performing  an inverse transform  on   the
 spectrum itself. It is primarily for this reason
 that the interferogram should be saved, even
 though it is somewhat more costly in  disk
 space.

 2.6.3 The Transform

      The transform  is  performed on the
 interferogram by  machine and therefore is
 done numerically. There are many algorithms
 to accomplish this, and  which of these is
 used  in  the software provided  with  the
 commercially available instruments  is  not
 known.   The mathematical basis for  the
 transform is described  below.

      The complex motion of items such as
 vibrating  strings  or  drumheads  or other
 periodically varying quantities can always be
 described by a sum of sine and cosine terms
 known as a Fourier series. The frequencies in
these terms are called the  fundamental
frequencies at which the item or quantity can
vibrate. The actual motion is then a linear
combination   of   these   fundamental
frequencies. When this summing of terms is
done, it is said that a harmonic analysis has
 been performed  on  the  original  vibratory
 motion. A study of such analyses shows that
 there-are related pairs of variables such as
 time   and  frequency  or  position   and
 momentum.   In   an  analogous   manner,
 functions can be analyzed, but here the more
 general Fourier transform must be used. The
 Fourier series  is used to describe a periodic
 function as an infinite sum of sine and cosine
 terms  whose  frequencies  are  multiples of
 some fundamental. The transform allows the
 analysis  of nonperiodic  functions  as  an
 integral (also a summation) over a continuous
 range of frequencies. In one of its forms that
 relates  time   and frequency,  the Fourier
 transform /^co), a function  of frequency, is
 related  to  G(t), a function of  time,  as is
 shown  in  Equation  2-5.  In  the  present
 situation,  the  function   G(t)   is   the
 interferogram produced by  the system, and
 F(co) is called the single-beam spectrum. The
 t in the term G(t) is a dummy variable, but in
 this case it  is really the  position of  the
 moving mirror from the center burst position.
       -T**J
    1   I  —./ \  -«fl*
= -,—J(7(f)e
                            dt
                                (2-5)
2.6.4  The Single-Beam Spectrum

       Some of the literature in this field
refers  to the single-beam spectrum as the
inverse  transform  of  the  interferogram
because the instrument takes the transform
of  the  incoming  signal  to start   with.
Mathematically, this merely changes the sign
in the exponent of Equation 2-5,  which
implies a phase shift in the sine  and cosine
terms. The term single-beam spectrum is a
                                       2-20

-------
 MAIMn
                            TR-4423-99-03
 historical holdover  from  the  time  when
 spectroscopists   used   a  double-beam
 instrument and determined the transmission
 directly from a ratiometer in the electronics
 of the  instrument. The  present-day  FT-IR
 systems do not use a double-beam system,
 but some of the terminology remains.

       The single-beam spectrum contains all
 the information about the absorbing species
 of interest. But most workers do not use this
 spectrum for  any direct data  analysis. In
 some systems, it is this  spectrum that has
 been stored on disk. At 1 -cm"1 resolution this
 spectrum takes   about  35  kilobytes  of
 memory,  which  is  much  less  than the
 interferogram   (100   kilobytes).  However,
 storage capacity of the computers available -
 today makes this  a non-issue.

 2.6.5  Data  Analysis
 the  interferogram. Although in some  cases
 the  interferogram  is analyzed  directly  to
 determine  the  concentration  of the target
 gas, more commonly, the interferogram is
 automatically converted  into a single-beam
 spectrum through the numeric process that is
 called a fast Fourier transform. A single-beam
 spectrum is generated and recorded for each
 sampling period.  We call this  spectrum the
 analytical,  or field, spectrum. A background
 spectrum is generated by one of the methods
 described in Chapter 4. Then a transmission
 spectrum is  obtained by  dividing the field
 spectrum by the background. The absorption
 spectrum is obtained by taking the negative
 logarithm of the transmission spectrum. The
 absorption  spectrum is used  for all further
 data analysis.

 2.6.5.2    Generation of the Reference
           Spectrum
       The data analysis includes generating
an  absorption  spectrum  from  the  raw
interferogram data, developing or obtaining
the appropriate reference spectra, and then
applying  the  chosen  analytical method to
determine the concentration  of  the  target
gases. The  analytical  methods  and  the
procedures for  generating  an  absorption
spectrum   from  the  interferogram   and
reference  spectra of the target compounds
are discussed below.

2.6.5.1  Generation of the Absorption
         Spectrum

       As shown  in Figure  2-8,  the data
analysis generally starts with the recording of
       A  reference  spectrum  is  usually
generated by using a high concentration of
gas  in a  relatively short cell. The  cell is
usually at least 1 m long, although multipass
cells with longer path lengths are also used.
A  pure sample  of gas  mixed with an inert
gas,  such  as   nitrogen,   is  used.  The
concentration of  gas used  to generate  the
reference spectrum should  yield a range of
absorbance values that match as closely as
possible  those  expected  to  be  found  in
atmospheric measurements. The system can
use a flowing stream  of  gas, but the total
pressure   should  be  around  1  atm.  The
process of producing a reference spectrum is
then the same as outlined above.
                                       2-21

-------

                                                                         TR-4423-99-03
                    Interferogram
                                                                   Single-Beam
                                                                    Spectrum
     0          1
Path Difference (cm)
                                                       1000   2000   3000   4000
                                                            Wave Number (crrr1)
                             Analytical Spectrum
                             Generated from Each
                             Sampling Period
                             Background Spectrum
                             Generated as Described
                             in Chapter 4
                     Transmission Spectrum =
                                Analytical Spectrum
                               Background Spectrum
                                          I
                        Absorption Spectrum = -In [transmission]
                        The absorption spectrum is used for all further
                        data analysis.
                         Figure 2-8. Data Reduction Flow Chart.
       The production of reference spectra is
an exacting undertaking and requires great
attention to the experimental details. It is not
likely that most users of the FT-IR technique
will  prepare  their own  reference  spectra.
Reference  spectra are  currently available
commercially.  The  National  Institute for
Standards  and  Technology  (NIST)   has
undertaken the task of producing  reference
spectra that are available at a minimal cost.
                               The investigators can at present  (1999) be
                               contacted  at  301-975-3108  or  on  the
                               Internet at http://gases.nist.gov.

                               2.6.5.3 Analytical Methods

                                     After the  reference  spectra  of the
                               target gases are  obtained, the appropriate
                               wave  number  region  for analysis must be
                               selected. The selection  should be based on
                                         2-22

-------
                                                                       TR-4423-99-03
 an examination of reference spectra and the
 type of  analytical  method  chosen.  Two
 issues  must  be  addressed  to make this
 selection. Ideally, the gas should have a high
 absorption coefficient in the selected region,
 and the region should be free of absorption
 bands from interfering species. If interfering
 species are present they must be  identified
 and accounted for in the analysis methods.

       Once an appropriate  wave number
 region is selected, data analysis can proceed.
 The concentration of the unknown gas can
 be determined  in three general ways,  as
 described  below:  the comparison method,
 scaled  subtraction,  and  multicomponent
 analysis  techniques.  Each method uses  a
 reference  spectrum  of   the   gas   being
 investigated.

 2. 6. 5. 3. 1 Comparison Technique

      One method  of  determining the
 concentration  is  to measure the absorbance
 at a particular wave number and compare  it
 with  the  absorbance  of  the  reference
 spectrum at the same wave number. Then,  if
 reciprocity holds (as implied by Beer's law),
the concentration is obtained as follows. The
 absorbance (A) is the product of a, the optical
absorption coefficient, C, the concentration
of the gas and L, the path  length. Thus A =
aCL and the unknown concentration can be
found from the following expression.
CunkLunk
                                 (2-6)
 Solving for the unknown concentration gives
 the following.


     ^unk ~ ^ref^ref^unk ' ^unk^ref   (2-7)


 This concentration has the same units as the
 units of the reference concentration, which
 is prepared as described in Section 2.6.5.2.

 2.6.5.3.2 Scaled Subtraction Technique

       The scaled subtraction  technique  is
 similar  in  principle   to  the   comparison
 technique. This  technique  is  particularly
 useful if there  are spectral features due to
 interfering species that overlap with those of
 the  target  compound. However, for scaled
 subtraction to be  successful,  either  the
 target compound or the interfering species
 should have at least one  unique absorption
 band. High-resolution data can be used to an
 advantage  with this technique.

       The scaled subtraction can be done as
 follows. Most software packages allow two
 spectra to be subtracted interactively. In this
 case the  reference  spectrum  should  be
•subtracted from the analytical spectrum until
 the  absorption maximum of the band of
 interest is zero.  Once the  subtraction is
 completed  the software  reports a  scaling
 factor. This factor can be multiplied by  the
 concentration used to generate the reference
 spectrum to obtain the concentration of  the
 target gas in the analytical spectrum. There
 is some operator skill involved in subtracting
 spectra   interactively;   therefore,   some
 practice   in    using   this   technique   is
                                        2-23

-------
                                                                      TR-4423-99-03
 recommended before the actual field spectra
 are measured.

 2,6.5.3.3  Multicomponen t A na lysis
           Techniques

       Multicomponent  analysis techniques
 can  be used to advantage when there are
 several target compounds to be analyzed for
 and  there  are several  interfering  species
 present. This is the case often encountered
 in open-path FT-IR monitoring, so some type
 of multicomponent  analysis  technique  is
 generally the preferred method of analysis.
 There are several techniques that are used to
 perform  multicomponent  analyses  of  IR
 spectra. Multicomponent analysis techniques
 encompass a discipline unto themselves, and
 a  complete  discussion  of   the   various
 techniques is beyond  the  scope  of this
 document. Chapter 9 includes a discussion of
 classical least squares  analysis. But this is
 intended to be  only an introduction. The
 reader is referred to an excellent review by
 Haaland   (1990).  The  most   common
 multicomponent analysis  method  used in
 open-path  FT-IR monitoring  is based on  a
 classical least squares (CLS) fitting algorithm.
 This is discussed below.

      The  CLS technique  performs  a linear
 regression by using the unknown  and  the
reference  spectra over  a wave  number
region. The slope calculated in the regression
is then used as  a multiplier of the reference
concentration to obtain  the unknown. The
ratio  of the  path lengths must  also be
accounted for. Thus if the  slope is found to
be 1  and  the ratio of the reference path
 length  to  the path  length  used  for the
 measurement is  1/10,  then the unknown
 concentration  is  1/10  of  the  reference
 concentration.

       The  process  of using  the  linear
 regression  is more  suitable than either the
 comparison  technique   or   the   scaled
 subtraction technique because the shape of
 one spectrum is compared with the shape of
 the reference spectrum. If  the  correlation
 coefficient is also  calculated,  it  gives  a
 measure of this comparison. There  is one
 significant  problem  with this technique that
 has   generally   been   overlooked.   The
 procedures are generally written by assuming
 a  linear response  of  the  instrument  to
 changes in concentration. The response is
 actually  slightly  nonlinear,   and  this
 contributes to the overall error in the data.
 This topic is discussed in depth in Chapter 8.

 2.7    References

 Beer,  A. 1852. Ann. Physik. 86:78

 Beer,  R. 1992. Remote Sensing  by Fourier
 Transform Spectrometry, John Wiley & Sons,
 New York.

 Filler,   A.S.    1964.   Apodization  and
 Interpolation  in   Fourier  Transform
 Spectroscopy, J. Opt.  Soc. Am. 54:  762

 Haaland, D.M. 1 990. Multivariate Calibration
 Methods  Applied   to  Quantitative  FT-IR
Analyses.   Practical   Fourier   Transform
Infrared  Spectroscopy-lndustrial  and
Laboratory  Chemical Analysis, J.R.  Ferrar
and K. Krishnan, Eds.,  Academic Press, San
Diego, CA,  pp. 396-468.
                                       2-24

-------
                                                                   TR-4423-99-03
Marshall, A.G., and  F.R.  Verdun.  1990.   Michelson, A.A. 1881. The Relative Motion
Fourier  Transforms in NMR, Optical,  and   of the Earth and the Luminiferous Ether. Am.
Mass Spectrometry, Elsevier, Amsterdam.     J. Sci. 22: 120-129.

Norton,  R.H.  and  Beer,  R. 1976,  New   Rossi,  B.  1957.  Optics. Addison-Wesley
Apodizing   Functions   for   Fourier   Publishing Company, Inc., Reading, MA.
Spectrometry. J. Opt.  Soc. Am. 66: 259.
                                         Walter, B.  1889. Ann. Physik (Wiedemann)
                                         36: 502, 518.
                                     2-25

-------
                                                                        TR-4423-99-03
                                     Chapter 3
                            Initial Instrument Operation
                                      SUMMARY

          This chapter offers guidance and recommendations with respect to the initial
    tests that should be performed on the FT-IR system to verify that the instrument is
    set up to operate properly.  Specific areas that are addressed include the following.

       •  The characteristic features of the single-beam spectrum

       •  The distance at which the detector is saturated and operating in a nonlinear
          fashion

       •  The return signal  intensity as a function of distance

       •  The uniformity of the IR beam intensity

       •  The contribution of stray light to the total return signal

       •  The determination of the system noise

       •  The effect of water vapor concentration on the  return signal intensity
3.1    Introduction and Overview

       The   assumption   made   for   the
discussion  in  this chapter  is  that  the
manufacturer has  set up the FT-IR and  it is
running  according to  his  specifications.
Initially, the setup  procedure for  each field
study should be the same, although  certain
procedural  differences  are  dictated  by
specific data requirements.

       Before putting  the instrument  into
continuous monitoring  service, the operator
should  conduct   some  initial  tests   and
determine the following.
  •  The distance at which the detector is
     saturated and operating in a nonlinear
     fashion (Section 3.4)
  •  The  return  signal  as  a  function  of
     distance (Section 3.5)
  •  The stray light  inside  the instrument
     (Section 3.6)
  •  The uniformity of the  beam intensity
     (Section 3.7)

The  operator should become quite  familiar
with the single-beam  intensity  profile and
with the gross features of the  single-beam
spectrum.  Beyond that, the operator must
start several control charts that will provide
long-term   information  about  the   return
                                         3-1

-------
                                                                      TR-4423-99-03
 intensity and the noise levels. Some ancillary
 items such as water vapor concentration,
 ambient temperature, -and ambient pressure
 should also be recorded.

       This chapter includes a discussion of
 each of these items and also addresses the
 items that  must  be  recorded  so  that
 adequate information concerning the long-
 term stability of the FT-IR can be obtained.
 In addition, some data are presented  that
 have been recorded by FT-IR instruments in
 the past.

       The  tests outlined  in this chapter
 should be performed before data are recorded
 with  the FT-IR monitor.  A failure to do so
 could result in the acquisition of  erroneous
 data   that   could    lead   to   inaccurate
 concentration measurements. Many of the
 tests involving the initial instrument setup are
 similar to  those  proposed for use in the
 routine  quality   assurance   procedures
 presented in Chapter 10 of this document.

 3.2     The Single-Beam Spectrum

     Figure,  3-1   shows  a  single-beam
 spectrum taken with 1-cm"1 resolution.  The
 total scan time was  5 min and the total path
 length was 414 m.  The vapor pressure of
 water during this measurement was 1 2 torr.
There are several features in the  spectrum
that should be noted. First, the regions 1415
to 1815  cm'1 and 3547 to  3900  cm'1 are
where the  infrared  energy  in the  beam  is
totally absorbed  by  water  vapor.    The
operator  will  notice that, for a given  path
length, the width of the region for complete
   400   1000      2OOO      3000
               Wave Number (cm-')
                                  4000
 Figure 3-1. Single-Beam Spectrum Along a
 414-m Path. S indicates stray light.

 absorption varies as the  amount water vapor
 in the atmosphere  changes from one  day to
 the  next.  Also the  center  of the  region
 should become somewhat transparent as the
 path is made shorter.

     The strong absorption in the 2234- to
 2389-cm"1  region is due to carbon dioxide,
 and  the  atmosphere  in this wave number
 region  should remain opaque at all  times,
 even when the instrument is used to monitor
 over short paths. The opaque regions should
 be flat, and they represent the baseline of
 the spectrum.  Any deviation from zero in
 these regions indicates that  something is
 wrong   with  the  instrument   operation.
 However, Figure 3-1 shows these regions to
 be elevated.  This is due to stray light in the
 instrument, and  these regions  are marked
 "S" in the figure.

     The  operator  should pay particular
 attention to  the spectrum  in  the region
 around 600-700 cm"1. The spectrum  below
this wave number region should be flat and
                                        3-2

-------
                                                                       TR-4423-99-03
 at the baseline.  If  the  spectrum has an
 elevated baseline below the detector cutoff,
 in this-example the-650-cm'1-~region, the
 instrument may be operating in a nonlinear
 manner. If this is the case, the operator will
 see  what seems like  a  dip appear as the
 retroreflector or source is brought closer to
 the  FT-IR.   An  example of this is given  in
 Figure  3-2  for  a  single-beam  spectrum
 recorded at  a 20-m path  length.
 c
 a
 _
 01
 c
                2000
               Wave Number (cm'1)
                                 4000
Figure 3-2. Single-Beam Spectrum Recorded
at a  20-m  Total Path Length  Indicating
Nonlinear Operation.
     When  the path  is  sufficiently long
(200 m and 10 torr H20) or the water vapor
concentration  is large, an absorption  band
should be noticeable at 2720 cm'1. This peak
is the Q-branch of  deuterated water (HDO),
and it is also possible to observe the P and
the R branches.

     The spectral region around 3000 cm"1 is
also strongly  absorbed  by water vapor,
although it is  not  opaque. The  absorption
features .of methane are in this region.  This
 is  also  the  region  of  the  C-H stretching
 frequencies.  The atmosphere from 3500 to
 3900 cm"1 is opaque, again because of water
 vapor.   There  is still some sensitivity and
 therefore an elevated signal return above
 4200  cm"1,  and this is the region  where
 hydrogen fluoride is absorbing.

     The  return   beam   intensity    at
 approximately 987,  2500, and 4400 cm"1
 should be recorded so as to form a basic set
 of  data  about  the  instrument's operation.
 Along with this, the operator should record
 the path length.  The total  return  signal  is
 dependent on  the  path  length  and  the
 amount of water vapor  in the atmosphere.
 When  using  the single-beam  spectrum to
 gauge    how   well   the    instrument   is
 functioning, the operator should try to select
 regions  that  are not greatly impacted  by
 water vapor.

     Figure 3-3  shows the region between
 1000 and 1025 cm'1 enlarged and plotted  in
 absorbance. The operator should notice that
 there are water vapor lines at 1010,  1014,
 and 1017 cm"1.  These lines will be in every
 spectrum as long as the product of the water
 vapor concentration and the path  length  is
 large enough.  The lines at 1010 and 1017
 are actually doublets and cannot be resolved
 at 1-cm"1 resolution. The line at 1014.2 cm"1
 is a singlet and can be used as  a check for
 wave  number  shifts and  resolution.    A
 procedure for doing each of these is given in
the subsections below.

    Both wave number shifts and resolution
changes   indicate  that  something  has
                                        3-3

-------

                                                                      TR-4423-99-03
             1010            1020
               Wave Number (cm'1)
 Figure  3-3.  Region  Between 1000  and
 1025 cm'1. The line  at 1014.2 cm'1 can be
 used as a check for wave number shifts and
 resolution.
 changed in the instrument geometry, and if
 these occur they should be  discussed with
 the  instrument  manufacturer.  A  subtle,
 apparent wave number shift can be observed
 if the atmospheric absorption line used for a
 shift determination is an unresolved doublet.
 In this  case, if one  line  becomes more
 intense   with  respect  to  the other,  the
 envelope peak will appear to be shifted.

 3.2.1   Wave Number Shift

      To determine whether wave number
 shifts have  occurred, the operator  should
 have an absorption spectrum that contains
the water vapor line at 1014.2 cm"1 and one
for which there is no  shift  present.  The
HITRAN database for water vapor is used in
this document as a guide  (University of South
Florida 1993), and it positions this water
vapor line at 1014.2 cm"1.
        For any particular instrument, the line
 assignment   may   be  slightly  different
 (±0.2 cm"1)  because of the instrument's
 geometry, but it should not shift in time.
 However that may be, it is the responsibility
 of the operator to determine precisely where
 the water line is and whether shifts  occur
 with time. The operator may also choose to
 determine wave number shifts by using the
 HDD lines in the  2720-cm'1 region.   This
 measurement is somewhat more sensitive to
 shifts in the higher wave  number (shorter
 wavelength)  region.

      In principle,  any absorption or single
 beam spectrum can be used as  a guide to
 determine  wave number shifts.  There  are
 two methods available for determining shifts.
 The first is simply to compare the positions
 of the  peaks of  the two  spectra  on the
 computer monitor.  The second is to subtract
 the second spectrum from the first and study
 the result. The second spectrum should be
 normalized  to  the  first   by   a  simple
 multiplication before the subtraction is done,
 or the subtraction can be done interactively.
 After subtraction, wave number  shifts will
 result in a curve that appears to be the first
 derivative of the line  shape.  The  wave
 number where zero amplitude occurs will be
 shifted from the original peak wave number
 by an amount  that is proportional  to the
 wave  number shift.    This  is   shown
 schematically in Figure 3r4A.

3.2.2 Change in Resolution

      The other possible change that can
occur in  the  spectra as time passes is  a
                                       3-4

-------
                                                                       TR-4423-99-03
                             Suited
            OngiruJ
Figure 3-4.  Subtraction of Spectra for the
Determination  of  (A)  Line   Shifts  and
(B) Resolution Changes.
resolution  change  in the  instrument. If  a
change in  the  resolution  has occurred  but
there is no peak shift, the result will appear
to have the shape of an  'M' or a  'W,
depending  on  which  spectrum  has  been
recorded with the largest amount of water
vapor.   This  is shown  schematically  in
Figure 3-4B.  If there are no changes in the
line,  then the result of subtraction will be
random noise.

3.3   Distance to Saturation

      One of the early pieces of information
to obtain with an FT-IR monitor is the path
length at which the detector is saturated. As
 discussed  in  Section  3.2, this is easily
 noticed by a negative dip in the single-beam
-return in the 650-750-cm'1 region below the
 detector cutoff. As the retroreflector or the
 light source is brought closer to the detector,
 this  dip will  appear.    This  response  is
 opposite to the response that  is due to an
 absorption  feature in the spectrum.   This
 distance is important because  it  represents
 the minimum path length over which  it  is
 possible to operate without making changes
 to the instrument.  In the monostatic case, it
 is  possible to rotate the retroreflector to
 lower the return intensity.  If necessary, it is
 possible to lower the intensity of the FT-IR
 instruments by simply using a fine wire mesh
 screen to cover  the aperture.  A plastic
 screen should not be used because  plastics
 have absorption features in the infrared.

      To measure this distance, simply move
 the light source  or the retroreflector away
 from . the transmitting  telescope  until  the
 negative dip just  disappears from  the single-
 beam spectrum.   This distance  should be
 recorded as the minimum working distance
 available   without  making   instrument
 changes.

 3.4  Return Intensity as a Function of
      Distance

      Some attempt is made to  collimate the
 infrared beam before it is transmitted along
 the  path.   It  is,  however, impossible to
 completely  collimate the  beam because of
 the size of the light source. Therefore, most
 beams  either are  always  diverging as they
 traverse the path  or become  diverging at
                                         3-5

-------
                                                                        TR-4423-99-03
 some point along the path.  Once the beam
 is  bigger  than the  retroreflector  or  the
 receiving telescope, the return signal should
 diminish as the square of the distance.  That
 is,  beyond  some  distance the  signal  is
 reduced by  a  factor of 4  when the  path
 length is doubled. The reason that this return
 signal   versus   distance  determination  is
 necessary  is twofold. The  first  is that the
 commercially available instruments that use
 the monostatic configuration both have  a
 stray  light  signal  when  the telescope  is
 blocked. The return  beam  should never be
 allowed to  approach this signal. The second
 is that  at some distance  the system noise
 will become an  appreciable part of the signal,
 and -this represents  the  maximum  usable
 distance.

      To determine the  return signal as  a
 function of distance, the operator must start
 with the retroreflector or the light source at
 the minimum working distance as determined
 above.  Then, the operator should move the
 light source  or the retroreflector back by
 some distance  and record the signal.  This
 process should  be  continued until the signal
 level reaches the noise level or just levels off.
 It should be noted that the leveling-off effect
 can also be  caused by the return signal's
 reaching the stray  light level.  These data
 should then be plotted and  used  for quality
 assurance/quality control purposes.

 3.5  Determination of the Stray Light Signal

     The stray light in the instrument can be
measured without  regard to  the distance to
the light source  or the  retroreflector. It is not
 expected that instruments using the bistatic
 configuration will have any measurable stray
-light,  but  a  one-time check is appropriate.
 To measure the stray light, the operator must
 block the receiving telescope while the signal
 is being recorded.  It is important to use an
 appropriate blocking  material to do this.  No
 surface that can reflect any of the infrared
 energy back to the instrument can be used,
 nor any material that is  transparent.  The
 best blocking material is a piece  of black
 cloth such as is used  in the construction of a
 photographic film changing bag. For systems
 that  transmit   the   beam  through  the
 interferometer before transmitting it along
 the path,  the beam  can  simply  be slewed
 away from the retroreflector.  This return
 signal should be recorded and then plotted on
 the graph  as return signal versus distance,
 discussed  above.  A record of the stray light
 spectrum should be made and compared to
 the single-beam  spectrum  recorded  at the
 selected working distance.

      Stray light inside the instrument can
 also be caused  by  strong  sources of IR
 energy that  are in  the field  of view of the
 instrument.  For example,  it is possible to
 have the sun in the instrument's field of view
 during sunrise and sunset. This will probably
 give rise to an unwanted signal that actually
 comes from reflections inside the instrument.

      The stray light actually causes an error
 in   the   determination   of   the   gas
 concentrations and must be subtracted from
 the  data spectra  before processing. Thus it
 has to be recorded  at  every monitoring
 session and periodically  in the  case of  a
                                         3-6

-------
                                                                       TR-4423-99-03
permanent installation.  The effect  of stray
light on photometric accuracy is illustrated by
the absorption feature-shown in Figure 3-5.
    1.0-
    0.5-
            1.0
                 Wave Number
   1.0-
   0.5-
            1.1
                                    B
                 Wave Number
Figure 3-5. Effect of Stray Light. A. Spectral
feature  without  stray  light.  B.  Spectral
feature with stray light.
      The absorption line in Figure 3-5A has
a transmission of 0.5 at the peak. The base-
line   (100%   transmission)   has   unity
amplitude.  Now  suppose that  stray  light
exists  in the instrument in  the  amount  of
10%  of the original return signal.   This
means that the baseline goes to an amplitude
of 1.1, but the absorption feature goes to an
amplitude of 0.6, as shown  in Figure 3-5B.
Thus the new transmission of the absorption
is 0.6/1.1, and this is not equal to 0.5 but to
0.5454, or is in error by about 9.2%.
      Mathematically,  this is just verifying
the fact that  A/B * (A  + C)/(B +  C).  It
might be questioned whether the effect of
this stray light is offset by the effect of the
air  in the  interferometer  enclosure, which
most likely contains the pollutant gas also.

      However, the transmission due to the
gas in the  cell is increased from that along
the path in the  ratio of the path lengths.
Thus, for our data taken at Research Triangle
Park, NC, along  a  414-m  path, the ratio of
the  path  lengths  is at  most  1/414, and
therefore   the  additional  absorption  is
negligible.  A  more in-depth analysis of the
problems introduced  by stray light indicates
that actual line shape distortions may take
place.  It is also quite likely that the simple
subtraction of  stray light as suggested above
will  not  remove  all  the error incurred.
Therefore,  all  efforts should be made to at
least  minimize the amount of  stray  light
reaching the detector.

3.6   Determination of the Random Noise of
     the System

     The random noise of the system is
determined  from  an  absorption  spectrum
made from two  single-beam spectra taken
sequentially. These spectra are to be taken
under the same operating  conditions as will
be used for the acquisition of data spectra.
That is,  the same acquisition time and path
length   should  be  used  for   the   noise
determination  as  will   be used for  data
acquisition. There should be no time allowed
to elapse between the acquisition of the two
                                         3-7

-------
MA/mf^j
    f:'i:J:.-^.ii
                                                                     TR-4423-99-03
 spectra. This determination will be somewhat
 dependent on the water vapor concentration
 in the  atmosphere, so  the  water  vapor
 concentration should also be determined.

      The use of the word "noise" suggests
 a random signal that is primarily produced by
 the system electronics. When measurements
 are taken in the  open  air, this  may  not
 exactly be the  case.  If the period of data
 acquisition for  the  two  spectra  is long
 (because of a  large  number of scans,  for
 example),  then  atmospheric  effects may
 contribute to the  measurement.  For that
 reason, the  wave  number regions that  are
 used  for  the   determination   should  be
 carefully  chosen.  Three  regions   are
 suggested below, but the low wave number
 region may not  be  suitable in all situations
 because of the presence of gases other than
 water vapor.
                                         the three regions  958-1008 cm'1,  2480-
                                         2530 cm'1, and 4380-4430 cm'1.

                                               There  are  several  ways  that  are
                                         described in statistics texts to determine the
                                         noise.  We will specify the use of the root
                                         mean  square   (RMS)  deviation  as   the
                                         appropriate measure of the noise.  The first
                                         step is to perform a linear regression over the
                                         wave number region and determine the slope
                                         and the intercept of the line.  At each wave
                                         number,  the next step  is then to calculate
                                         the difference  between the calculated line
                                         and the actual  ordinate  value.  The squares
                                         of these  differences  are  then  used to
                                         calculate the RMS deviation as is described
                                         below.

                                              To   calculate  the   slope  and  the
                                         intercept  from  a linear  regression, use the
                                         following expressions.
      The actual wave number range over
which the  noise should be calculated will
vary with the resolution used.   Statistically,
it can be shown that about 98 data points is
an optimum number (Mark and Workman,
1991).   For 1-cm"1  resolution, this means
that  data should be  taken over a 50-cm"1
region.
                                              NUMERA TOR =7V~£ XtYf - £ x£ Y,
                                            DENOMINATORS^, (*,)2-£*,£*,
                                            SLOPE=NUMERA TORIDENOMINA TOR
     Once  the  two  spectra  have  been
acquired, an absorption spectrum should be
made by using one of the two spectra as a
reference or background spectrum. Which
one  is  used for the  background is  not
important.  The noise is then determined  in
                                             INTERCEPT=—(£ Y-SLOPED X)
                                              In this case the Xt values are the wave
                                         numbers  and   the   Yt  values  are  the
                                       3-8

-------
                                                                    TR-4423-99-03
absorbance  values at the particular wave
numbers.

      The differences are then found as £>, =
Yi  - yt ,  where now  the  Yt  values  are
calculated  from the  line  by  using  the
expression Yt =  slope*X( - intercept.

      Once this is  accomplished the  RMS
deviation is  determined with the following
expression.
     RMSDEVIATION=
                          D;
                         N-2
                              1/2
      Mathematical representations  of the
 RMS deviation vary in what the denominator
 is.  Quite often the denominator  is written
 simply as N. The  term  N-2  is used here
 because  the slope and the  intercept are
 calculated from the data. This reduces the
 degrees of freedom by 2  and hence the N-2
 (see, e.g., Mark and Workman 1991).

      Some results of these measurements
 are shown in Figure 3-6.  The data shown  in
 the figure  were  taken over  the  region
 980-1020 cm"1 in order to include the water
 vapor peaks at 1014 cm"1.  To reduce the
 effect of water vapor to a minimum it  is
 possible to create two spectra  by using an
 artificial  background,  subtract the  water
vapor of one from the other, and then make
the noise determination.
             0.005
             0.004 -
          0)
          tn  0.003 -
          O
             0.002 -
             0.001 -
                                  Date of Measurement
         Figure 3-6. The  RMS  Baseline Noise Measured  Between  980 and
         1020 cm"1 (•), 2480 and 2520 cm"1 (•), and 4380 and 4420 cm"1 (A).
                                      3-9

-------
                                                                      TR-4423-99-03
3.7   Return Intensity as a Function of
      Water Vapor

      The return-beam intensity is a function
of the absorption due to water vapor in the
atmosphere. It  is therefore a function not
only of  the  path  length but  also the water
vapor concentration in the atmosphere.  It
must be clear to the operator that relative
humidity is not important in this case and has
no  relevance to  the FT-IR data,  and it  is
actually the  water vapor partial pressure  in
torr that must be used.  However, over  a
period  of  one   day,   the   water  vapor
concentration may not change very much, so
acquiring a set of data over a range of water
vapor concentrations will take some time.
Some thought must also  be given to the
problem  of  measuring the  water  vapor
concentration when the temperature is below
freezing.  The  Smithsonian psychrometric
tables give data  to an ambient temperature
of 5 °F,  but it is not  clear that a simple sling
psychrometer should be used to make the
wet and dry  bulb measurements.

      The measurements for water vapor can
be made at  any  place  along the path. The
operator should  note, however, that some
investigators feel that the concentration of
water vapor-along.the path actually changes.
We have made some measurements with a
sling psychrometer and  have  not  seen  any
appreciable changes along the path. We have
seen rather subtle changes in the absorption
due  to  water  vapor   from  the  spectra
themselves. This is most easily noticed if the
subtraction technique is  used. For example,
we acquired a set of data taken at  1-min
intervals.  After the spectra were converted
to absorption spectra, the first spectrum was
subtracted from  the others in  order. The
residual water in the spectra indicated that
minor  but  noticeable  changes take  place
minute by minute.

3.8  References

University of South Florida.  1993.   USF
HITRAN-PC.  University  of South Florida,
Tampa, FL

Mark, H., and J. Workman. 1991. Statistics
in  Spectroscopy.   Academic Press, New
York.
                                       3-10

-------
                                                                      TR-4423-99-03
                                     Chapter 4
                               Background Spectra
                                     SUMMARY

          The topics and specific points of emphasis discussed in this chapter include
    the following.

          •  The generation of the transmission spectrum and the absorption spectrum

          •  The need for a background spectrum and the difficulties in obtaining one
             that is adequate

          •  The methods for generating a suitable background spectrum
             •   Synthetic backgrounds
             •   Upwind backgrounds
             •   Short-path backgrounds
             •   Averaged backgrounds

          •  Specific field guidance for measuring the background spectrum
4.1    Introduction and Overview

       In current  use, long-path, open-path
FT-IR data are obtained from single-beam
measurements.    That  is,  there   is  no
reference,  or background,  spectrum taken
simultaneously with the sample spectrum to
null  the  spectral  features  due   to  the
characteristics of the source,  beam  splitter,
detector,  and  interfering  species  in  the
atmosphere.  To  remove these background
spectral  features,  the   single-beam  field
spectrum   is   divided  by  a  single-beam
background spectrum, or I0 spectrum. This
operation (illustrated in Figure 2-8) generates
a  transmission spectrum.    According  to
Beer's  law, the absorption spectrum is then
calculated by taking the negative logarithm
of the transmission spectrum.   It  is  the
absorption  spectrum that is used for data
analysis.

       Ideally, the background spectrum is
collected  under  the  same  experimental
conditions as those for the sample spectrum,
but without the target gas or gases present.
However,  in the  field it  is not  possible to
obtain the /0 spectrum directly because the
target gas cannot be easily removed from the
atmosphere.    This  chapter  presents a
discussion of the  problems associated with
obtaining the 70 spectrum and the methods
that are used  to  generate  a background
spectrum.

       There are currently four methods for
obtaining 70: synthetic, upwind,  short-path,
and averaged background spectra. Synthetic
                                        4-1

-------
                                                                       TR-4423-99-03
 background spectra can  be generated  by
 selecting  data points  along  a  single-beam
 spectrum and  then calculating the high-order
 polynomial  function   that  best  fits  the
 selected points or by repeatedly subtracting
 the  spectral  features due  to  interfering
 species from a single-beam spectrum.  Both
 of these methods can introduce distortions or
 spurious  features in   the  actual  intensity
 profile.  Therefore, much care must be taken
 when  generating  synthetic  background
 spectra.

       For short-term monitoring efforts, the
 FT-IR  path  is   generally  chosen  to  be
 perpendicular to the wind field. If the area of
 the  emission  source  is relatively small, a
 background spectrum can be acquired along
 a  path  that  is  upwind from  the  source.
 However, it is difficult  to make this type of
 measurement  frequently if the FT-IR system
 has to be moved from one side of the source
 area to  another.  Also,  errors  may  be
 introduced in the measurements if the water
 vapor   in  the  atmosphere   changes
 significantly between  the  times that the
 background spectrum and the sample spectra
 are acquired.

      Another option  for  obtaining the /0
spectrum  is to  bring   the retroreflector or
external  IR  source  close  to  the receiving
telescope.   This  effectively eliminates the
absorption caused by the target gases and
records a true  instrument background.  One
problem with this method is that the detector
can be saturated at short paths because too
much IR radiation is incident on the detector
element..
       When the experimental conditions are
 fairly constant over a measurement period, it
 is possible-to average several backgrounds
 that have been taken  over this time.  This
 average 70 an then be used for the entire data
 set for that period.  However,  most of the
 time, the experimental conditions  are not
 constant enough to perform this type of long-
 term   averaging.      For  example,   the
 concentrations of water  vapor,  C02, and
 gases emanating  from other sources are
 constantly changing.

       The   change   in   water    vapor
 concentration  must   be   considered  the
 biggest potential  source of  error  in  the
 background  measurement. This is because
 changes  in water  vapor concentration can
 change the curvature of the baseline  in the
 field  spectra.  When  that  happens,  the
 background  spectrum and the field spectra
 do not properly match, and errors  occur. An
 accurate record of the partial  pressure  of
 water vapor  should be  maintained.  These
 data should be taken continuously.

       Acquisition   of  the  70   spectrum
 represents one of the more difficult tasks in
 FT-IR  long-path,   open-path  monitoring.
 Currently, there is not a universal method for
 obtaining   a  satisfactory   background
 spectrum. The method chosen to obtain /0
 must  be determined on a  site-by-site basis.
 This chapter includes a rationale for the use
 of an  70 spectrum,  as well as advice on the
 appropriate  techniques for generating  it.
These techniques are each discussed below.
                                        4-2

-------

                                                                        TR-4423-99-03
 4.2   Synthetic Background Spectra

       The software that is supplied with the
 commercially available instruments  has a
 routine that  allows a synthetic spectrum to
 be generated.  This is best accomplished by
 selecting data  points along some  original
 single-beam  spectra  and then calculating a
 high-order polynomial function that best fits
 the selected  points. Generally, however, the
 individual  data  points are connected .with
 straight line segments, and in most instances
 this is satisfactory. Thus, once a single-beam
 spectrum  is  produced, it  can be used to
 generate a synthetic I0. Synthetic 70 spectra
 can be made that cover only selected wave
 number regions, or  they  can be  made to
 cover the entire wave  number region that
 the  FT-IR  uses.   An  example   synthetic
 spectrum is shown in Figure 4-1.
                 Original Spectrum
                   1000
              Wave Number (cm )
Figure 4-1. Synthetic I0 Spectrum. The
peak at 1110 cm"1 has intentionally been
left in as a fiducial point.
       Some  care  must  be  used  when
synthetic 70 spectra  are  generated  so  that
distortions   are  not  introduced  into   the
intensity, function.  For  this  reason, when
 data points are selected, they should never
 be selected at the peaks or even within an
 absorbing  feature.  The final curve that is
 produced must be a smooth function without
 artificial dips and peaks and must follow the
 baseline of the single-beam spectrum from
 which it is made.

       It  is   essentially   impossible   to
 construct a synthetic spectrum in the wave
 number  regions where  water vapor  and
 carbon  dioxide   absorb   strongly.     The
 individual lines are overlapping so that  it is
 very difficult to judge where the background
 curve should  be set, and in  much of the
 region there  is  almost  no  energy  being
 returned to the detector. Even at the shortest
 path length possible, there are still portions
 of   the  spectrum  that  are  completely
 absorbed by the water vapor and the carbon
 dioxide.  Because of this strong absorption,
 these regions of the spectrum are not used in
 the data analysis.

 4.3    Upwind Background Spectra

       For short-term monitoring efforts, the
 path  is  generally  set out  so  as  to  be
 perpendicular  to the wind field.  If need be,
 the operator can change the  orientation of
 the path  so that this geometry is maintained.
 If the area  of  the  source is relatively small
 and its upwind side is accessible, an upwind
/0  spectrum  can  be  acquired.  A  usable
 background spectrum  can  also be acquired
 by taking data  along the side of source. (See
 Figure 4-2.)   As long as the  wind is  not
 blowing  across   the  source  area   and
transporting the emissions across the path
                                        4-3

-------
 re/*U.:7s~?sTi± -^
 I tun- 'i-j -^j.-i ^=
                                                                        TR-4423-99-03
 used  for  70,  these  spectra  should  be   transport the entire system from one place to
 satisfactory.                                another.
   Retroreflector
     Path for
  I0 Measurements
                   Prevailing
                   Wind
      J_D
      FT-IR                     Retroreflector
          Y— Primary Data Path -*j
Figure 4-2. A Possible Configuration for 70
Spectrum Acquisition.
       Another technique  for acquiring  an
upwind background spectrum is to wait until
the wind shifts so that the path is along an
upwind side of the source. This works well
for isolated sources,  but  if there  are other
places emitting chemicals, then this method
can lead to errors in identifying compounds
and in quantifying just what is.coming from
the source under  study.

       There are some advantages to taking
true upwind  background spectra this way.
First,  it is likely that sources are not isolated
and the  chemical species of  interest  are
emanating  from several places in  the area.
The  compounds  entering  the area  being
investigated are thus included in the upwind
background spectrum.  If the  configuration
can be set  up so the  side of the source area
can be used, a second retroreflector or IR
source can be used, and the 70 spectrum can
be taken, frequently  and without having  to
       There are also some disadvantages to
this procedure. It is difficult to take upwind
and downwind  spectra  frequently if the
system has to be  moved from one side of the
source area to another. Generally, this type
of spectrum is taken once at the beginning of
the  monitoring period  and once at the end
(each day).

4.4   Short-Path Background Spectra

       Another   possible   technique   for
obtaining the 70  spectrum is to bring the
retroreflector or external IR source  close to
the  receiving  telescope.  This  effectively
eliminates  the absorption caused  by  the
compounds of interest and records a true
background.  This background is called the
short-path  70.  If  two retroreflectors  are
available, this task is fairly easy to perform.
The FT-IR monitor can be pointed first to one
retroreflector and then  the other quite easily
with some  regularity.

       Figure 4-3  presents a procedure for
acquiring a short-path background spectrum.
The recommendation for  the  frequency of
repeating this procedure is that a background
should be considered valid for no longer than
one  day or until there  is a major change in
the  operational parameters.   Although  a
short-path  background spectrum may  be
valid for an extended  period, it should  be
revalidated on a   daily basis.   Necessary
checks for a short-path background spectrum
include the following.
                                         4-4

-------
                                                                             TR-4423-99-03
               ACQUIRING A SHORT-PATH BACKGROUND SPECTRUM

 1.  Calculate the required path length for each gas that is being measured.
    a.  Calculate the absorption coefficient of the gas in the wave number region that is being used
        for analysis.  From the reference spectrum, obtain the product of the concentration and the
        path length used to generate the reference data. Then measure the absorbance of the peak that
        is being used for analysis.  The absorption coefficient is given by the following.

                                      a=A/CL

    b.  Estimate the maximum anticipated concentration of the target gas. Express this in the same
        units as the concentration of the reference has been expressed (generally, ppm).
    c.  Set the absorbance to the number that has been experimentally determined from the RMS
        noise as an estimate of the lowest possible detection limit for the FT-IR system, for example,
        10"*, and use the expression

                                     L = A/aC

        to find the path length to be used for the background spectrum.
    d.  A sample calculation with the toluene peak at 1031.6 cm"1 is given below.
               a=A/CL
               Read A at 1031.6 as 0.0203.
               The CL product = 496 ppm-m

                             a = 0.0203/496 = 4.1 x 10'5

           Estimate the maximum concentration to be 0.050 ppm.
           Calculate L:

                         L =A/aC= 10-4/(0.000041 x 0.050)

           or
                                     L =  48.8m

2.   Repeat Step 1 for each target gas and set up the  retroreflector or light source at the  minimum
    calculated distance.

3.   Take a spectrum at the same resolution as  will be used to take data. So that the noise in the data
    spectrum will  be the predominant source of noise, take the background spectrum for at least 4
    times the number of scans used for the data spectrum. (This judgement should be based on the
    time involved with taking a spectrum.)

4.   Use this  spectrum for the  background spectrum.
      Figure 4-3.  Procedure for Acquiring a Short-Path Background Spectrum.

                                          4-5

-------

                                                                       TR-4423-99-03
  •  Comparison of  the  curvature  of the
     baseline in the  short-path spectrum
     with the curvature of the baseline in the
     field spectra
  •  Inspection   of   the  spectrum   for
     wavelength   shifts   and   resolution
     changes, as discussed in Section 3.2

       One difficulty  with this procedure is
deciding  on  an appropriate  distance for
placing  the  retroreflector or  external  IR
source.  One difficulty is that the detector
can easily be  saturated when the path length
is too short. However,  if the  reflector is
placed far enough  away  to overcome this
problem, then the absorption  may become
large enough to be detrimental.   Thus, this
distance  must  be  determined  for  each
instrument at least once.  When the detector
is saturated,  the signal seems to  drop below
zero at the  low-wave-number end  of  the
single-beam spectrum, and the response of
the detector  is nonlinear.  The retroreflector
or external IR source must be  far enough
away so that this dip in the signal strength
disappears. One way to accomplish this is to
reduce the light intensity by rotating  the
retroreflector or  using wire mesh screens to
attenuate the signal.  (See Section 3.3.)

      A second difficulty in   monostatic
systems is that the retroreflector will subtend
different angles  when it is  at  different
distances.  The retroreflector may be the
actual optical field  stop of the instrument,
and  changing   the   distance  can  cause
distortions  in the  spectrum.    When the
distance  is   increased,  the  retroreflector
subtends, smaller angles, and the instrument
 uses different cones of light.  This problem
 can be overcome by placing a field stop in
 the  instrument that uses a smaller field of
 view than the smallest anticipated from the
 retroreflector. However, that is a job for the
 manufacturer because  stops  cannot  be
 placed just anywhere  in the  optical  train
 without causing other problems

 4.5    Averaged Background Spectra

       When the experimental conditions are
 fairly constant over  a measurement period, it
 is possible to average  several background
 spectra that have been taken over this time.
 This  average 70  can then be used for the
 entire data set for that  period. Because all
 the individual spectra making up the average
 should have the same noise and there should
 be  no other  errors, the final error in this
 average background should  be smaller by a
 factor of  //V.   Here, A/ is  the number  of
 spectra in the average.

       However,  most   of  the  time the
 experimental conditions are not constant
 enough to perform the averaging. The water
vapor concentration is changing most of the
time, and so is the concentration of carbon
dioxide. If other sources are in the area, the
concentrations of the gases emanating from
them are not likely to be constant.  If any of
these gases are also  being  monitored, the
use  of an average 70  will  not  give  true
absorption spectra for the entire monitoring
period.

       Currently, no published data seem  to
be  available  that  have been  taken  over
                                        4-6

-------
                                                                      TR-4423-99-03
 periods  of  more   than   a   few  weeks.
 Therefore it is not known whether 70 spectra
 can be taken over-extended periods, sorted
 according to the similarity of conditions, and
 then averaged.  If this could be done, it might
 be possible to acquire a set of 70 spectra that
 are  universal.   The experimenter could use
 these   as  he  uses  the   library  of  pure
 compound spectra.  Thus, with 10 torr  of
 water  vapor and a path length of 200 m,
 background VII might be used, etc.

       Acquisition  of  the  70  spectrum
 represents one of the more  difficult tasks
 associated with using an FT-IR spectrometer.
 Most of the data that have been published
 until   now  has  come  from  short-term
 monitoring programs.  The 70 spectra that
 have been used have been  taken at various
 times during the days of operation.  These
 spectra have then been  used for the next
 series of atmospheric spectra to be analyzed
 until the next background is  taken. Although
 shifts in the wind direction have been used to
 determine the need for a new background,
 changes in water vapor concentration have
 not.  Little information has been published on
 the stability of the instrument itself or on the
 effect  of  taking  a  new background  after
 some instability develops.  There  is some
 evidence that has been derived from bench-
 top  laboratory  work that  indicates the  I0
 should   be taken  as  often  as  every other
 atmospheric spectrum.  There is also some
 evidence that a single 70 can be used over a
 long  period  of time  with  no  detrimental
 effects. Neither of these observations has
 been corroborated by any in-depth study of
the background spectra.
       The  work  performed  at  Research
 Triangle Park, NC,  indicates that a synthetic
 background (70)-spectrum can-be used for
 extended periods with some care.  After one
 year of operation, the 70 spectrum had to be
 changed four times.   Three  times  were
 caused by  instrument component changes,
 and only one time  did  the  atmospheric
 conditions require a change in 70. That time
 the water vapor concentration changed from
 23 torr to 9 torr in a 1 5-min period.

       Thus, much work is yet to be  done in
 establishing the  appropriate  manner and
 frequency for acquiring the 70 data.

 4.6    Why  Use a Background

       The primary data that is produced by
 an FT-IR is the interferogram.  It contains all
 the information that is required to obtain the
 concentrations  of  the  gases that  the
 experimenter   wants.   However,   the
 information in  the interferogram is in a form
 that is somewhat cumbersome  to  most
 people that  are familiar with spectroscopy.
 In addition,  the  primary  physical law  that
 governs the analysis of the  data is Beer's
 law, and this is defined in terms of the  more
 conventional spectra that are familiar to most
 people.  Thus,  with  few exceptions, the
 interferogram is converted to a single-beam
 spectrum via a Fourier transform and  divided
 by a background, or 70, spectrum to get a
transmission   spectrum.  This   is   then
 converted to  an absorption  spectrum  by
finding  the  negative   logarithm  of  the
transmission spectrum.  At this point, the
 absorption spectrum is compared with the
                                        4-7

-------
      -
      : 'L-j -.-i •
                                                                       TR-4423-99-03
 absorption spectrum (after it has undergone
 the same mathematical processing) of a pure
 gas to obtain the concentration.

       All this  mathematical processing is
 performed by computer and therefore is done
 numerically with whatever algorithms were
 available at the time the  analysis  software
 was written. A relevant question is whether
 all the  processing is necessary, or if  it is
 being done merely because it  has always
 been done that way.

       It is likely that most workers would
 essentially   refuse  to   work   with   the
 interferogram directly because spectra have
 historically been used.  However, this does
 not seem  to  be true of  the single-beam
 spectrum.  The  single-beam spectrum  also
 contains  all the  information  about   the
 concentrations,  and  it looks like a normal
 spectrum.  The water vapor and the carbon
 dioxide   absorption   peaks   are   readily
 discernable, and any  absorption due to other
 gases should also be discernable.  Thus, for
 identification purposes, using the single-beam
 spectrum should not  be a problem.

      The  problem  seems to arise when
quantitation is  required.    If the reference
spectrum were also  available in terms of a
single-beam spectrum, a direct comparison
could  be made between the data spectrum
and the reference spectrum, but only if the
intensity levels of each were known  on some
absolute basis.  Beer's law gives no hint of
how  the data are to  be  analyzed in  the
absence of an /0 spectrum. It is true that the
single-beam spectrum is recorded with some
 intensity level for the ordinate. But unless it
 is put on an absolute basis, the single-beam
 spectrum alone  is not a sufficient piece of
 information to  determine the transmission
 through the atmosphere.

       Currently, there does not seem to be
 a  satisfactory way to  use the  single-beam
 spectrum alone for the final analysis.

 4.7    General Advice About Background
       Spectra

       All of the currently used methods for
 generating  a background  spectrum   are
 fraught with difficulties.  No one method is
 generally accepted  as the best method for
 acquiring a background.  It is crucial for the
 operator  to be aware of the  importance of
 this spectrum and of two criteria for a valid
 background.

   1. The background cannot contain any
      absorption features due to the target
      gas  or gases.

   2. The background  spectrum must be
      valid for the time period over which it
      is  used.  .

Although there seems to be agreement that
the first  point above  is a requirement, no
such consensus  has been reached for the
second. The time periods over which single
backgrounds  are  used by various  workers
vary  from  a  day  to  several   months.
However,  one  point has become  clear.
Whenever  any  optical  component   (light
source, mirror, window, etc.) is changed in
                                        4-8

-------
        ?*?*'. ?• i -'• ' =
        •.*•=•• ••!=•: =
        s ~ ss 7ai =••=
                            TR-4423-99-03
the instrument, a new background must be
acquired.

     There are few guidelines as  to what
represents a  valid background spectrum for
the production of accurate data.  One point
is   that  the  curvature  of  the   baseline
(maximum intensity) must be quite close to
the curvature of the baseline of  the  field
spectra.

     There are two primary choices for the
background that is to be used with a specific
data set. If the absorbing  peaks are narrow,
as they are for methane or hydrogen fluoride,
it  is  possible to  construct  a   synthetic
background for the analysis. But for broad
absorbing features like those exhibited  by
acetone, this is difficult. With broad features,
even small changes in the curvature of the
baseline can produce large errors. In general,
the operator is advised to use a synthetic
background whenever possible.  Taking a
short-path background should be considered
only when  the  absorption feature  of  the
target gas is very broad.  A final reason to
consider the use of a synthetic background is
that it is essentially the only background that
allows actual atmospheric concentrations to
be determined.
                                        4-9

-------
              . : r
               •
                            TR-4423-99-03
 TCFU- ~:~ss~sj. =••==
 ICifft: ••.-.' -.-.i * i ^=
                                     Chapter 5
                               Water Vapor Spectra
                                     SUMMARY

          Specific topics that are addressed in this chapter are the following.

          •   Selection of spectra that can be used for generating a water vapor
              reference spectrum

          •   Creation of the water vapor reference spectrum itself

          •   Subtraction from the water vapor spectrum all absorption features of
              C02, N20, CH4, CO, and the so-called pollutant gases

          •   Example water vapor spectra for methane and ozone, selected because
              they present different problems to the operator
5.1    Introduction and Overview

       Water  vapor  absorption  lines  are
present  in   all  regions   of   the   mid-IR
wavelength   region,  as  was   shown   in
Figure  3-1.    The  water vapor  spectrum
interferes with the spectrum of almost every
volatile organic compound in the atmosphere.
Because of this, the absorption features of
water vapor have to be accounted for during
the analysis of field  spectra.

       Some  amount of the  water vapor
absorption is  accounted for if there is water
vapor   absorption   in   the   background
spectrum,  as  described  in  the previous
chapter.    However,  when   a  synthetic
background is used,  all the water will still be
in  the  field  spectrum,  and some residual
amount  will   be   there   when    other
backgrounds  are used.  It is  possible to
account for the water vapor by considering it
as  an interfering  species  in the  analysis
package.     The  software  commercially
available for performing a classical  least
squares  analysis  allows the operator to
choose  interfering  gas  species  that are
present  in  the  wave  number  region of
interest.  To do this, however, a water vapor
spectrum must be available.  Although  there
are   water  vapor   spectra   available
commercially, they are not suitable for use in
this  application  because  of  line  shape
differences  and other  small  instrumental
effects, as well as  insufficient path length.

      This chapter explains how to use field
spectra  on site to produce a spectrum of
pure water that can subsequently be used in
the analysis of field spectra. Examples of
water vapor spectra that can be used for the
analysis  of  methane   and  ozone   are
                                        5-1

-------
                                                                      TR-4423-99-03
 discussed,  as   these   gases  represent,
 perhaps,  the  extreme challenges  that the
 operator will encounter.

 5.2   Water Vapor Spectra Considerations

       Any   single-beam   spectrum   that
 exhibits a sufficient amount of water vapor
 absorption in  the wave number region of
 interest can be used for the production of a
 water vapor reference spectrum.  Spectra
 taken at short path lengths or during very dry
 periods  may  not  be  satisfactory.    At
 Research Triangle Park, NC, we have seen
 the water vapor partial pressure change from
 a low of less than 1  torr in the winter to a
 high of 28 torr during the summer.  Changes
 in the water  vapor concentration of  this
 magnitude,  along  with  any   instrument
 changes, may require that a new water vapor
 spectrum be produced. It is the responsibility
 of the operator to determine when the water
 vapor spectrum  has  to be remade, and no
 hard  and fast rules on the frequency for
 creating  a  new spectrum are  presently
 available.  If the error bars  of the analysis
 increase from one data set to another, a first
 step in determining the cause is to  compare
 the water vapor reference spectrum with the
 water vapor in the field spectra.

      The  primary   concern   for   the
 production of a water vapor spectrum is that
the final result must not contain  any of the
target gas. If the water vapor spectrum does
 contain even a small amount of a target gas,
the analysis will  be in error by that  amount.
The ease with which the absorption features
of the target gas can be removed from the
water vapor reference spectrum is dependent
on  many instrumental factors,  and  the
process can be quite time-consuming. The
removal of the target gas absorption is done
by  subtraction  and   in  some  instances
requires great attention to detail.

5.3   General Process for the Production of
      a Water Vapor Spectrum

      The general procedure that  is to  be
followed to produce a water vapor reference
spectrum is given below.

   1.  Select two single-beam spectra that
      will be combined into a water vapor
      absorption spectrum.

   2.  Use one of the  spectra to  create  a
      synthetic spectrum that is to be used
      as the background.

   3.  Create the absorption spectrum that
      is  to  be used  as the water  vapor
      reference spectrum.

   4.  Subtract any absorption features that
      are known to be present from the
      target gas.

   5.  Analyze a number of the field spectra
      for the  target gas,  using the water
      vapor  reference  spectrum  as  an
      interfering species.

   6.  Determine if  any of the target gas
      absorption remains in the water vapor
      spectrum.

   7.  Repeat Steps 4-6 until you  are sure
      that there is no absorbance due to the
      target gases remaining.
                                        5-2

-------
                                                                       TR-4423-99-03
 A word of caution  is necessary  here.  If
 analysis is to be done for more than one gas,
 the synthetic background in Step 2 should be
 created for all the gases of interest at the
 same time. Otherwise, when the absorption
 spectrum is created,  some or all of the water
 vapor  absorbance will  disappear.   If this
 occurs the process has to be started over.

 5.3.1  Selection of Spectra

       The  spectra  selected in Step 1  that
 are to be  used to produce a water vapor
 reference spectrum  should be taken at the
 same resolution as  the  field  spectra.  The
 water vapor spectrum should have a signal-
 to-noise ratio that is the same as  or better
 than the field spectra. If possible, the water
 vapor concentration  for these two spectra
 should be representative  of the water vapor
 concentration  in the field  spectra  to be
 analyzed.  Curvatures and any  other special
 features  of the  baselines of these spectra
 should be  the same as  those of  the field
 spectra baselines  over  the wave  number
 regions of interest. If both of these spectra
 are closely spaced  in  time,  they  should
 contain approximately the same amount of
the target  gas. The  operator should select
 spectra at times when  a minimum  of the
target  gas  is expected to  be  present.
However, the operator should be aware that
the target gas will be  present in both spectra
and should consider the ramifications of that
fact.
 5.3.2  Generation of Synthetic Background

       Either of the two spectra selected
 above (Step 1} can  be used to  create  a
 synthetic background that is then used to
 create an absorption spectrum. The synthetic
 background must be created over the same
 wave number region as will be used for the
 final analysis. The wave number region can
 be larger than that used for analysis,  but  it
 cannot  be  smaller.  If  the  two spectra
 selected above are closely spaced in  time
 they will probably both contain approximately
 the same amount of the target gas. This will
 certainly be  the  case  with  methane  and
 nitrous oxide regardless of the time selected,
 but any set of spectra for ozone may have
 widely  varying  absorption due  to ozone.
 Again, which one of the spectra is used for
 the  synthetic  background  generation  is
 arbitrary.   Once the synthetic  background
 has been prepared, it should be stored  with
 an appropriate name.

 5.3.3  Generation of the Absorption
       Spectrum

       Use the remaining spectrum  (after
 Step 2) as the data spectrum and the newly
 created synthetic background to create an
 absorption spectrum (Step 3).  This is done
 exactly as  all the  other absorption spectra
 are created. (See Figure 2-8.)  At this point
 it is not  likely  that  the baseline  of  the
 absorption spectrum is at zero absorbance.
To make the baseline  zero,  measure  the
 height of the baseline above zero and simply
subtract  that amount from the  spectrum.
This  completes the process of creating an
                                        5-3

-------
                                                                      TR-4423-99-03
 absorption spectrum, and it should be saved
 with an appropriate name.

 5.3.4  Subtraction of the Target Gas

       This step (Step 4) is really an iterative
 process.   The  newly created water vapor
 spectrum must  be  used  in the analysis of
 other spectra. The results of these analyses
 must be examined in an attempt to  determine
 whether any of the target gas remains.  If
 some remains,  it must be  subtracted from
 the water vapor spectrum.   The process
 must then be repeated until the operator is
 sure that no target gas absorbance remains.

       To analyze the water vapor spectrum,
 the  operator  must  process  several  data
 spectra and  use the newly created  water
 vapor  spectrum as  an interfering species
 spectrum.  The  field  spectra  to  be  used
 should be chosen so that the target gas is a
 minimum.  If  the analysis shows the target
 gas to go through zero  and actually become
 negative, then the water vapor spectrum still
 contains an absorbance due  to the target
 gas. The negative  concentration should be
 subtracted from the water vapor  reference
 spectrum. Some gas concentrations will not
 go  to  zero at  any time but  will reach  a
 minimum.  This minimum can be set to zero
 in some instances, but  it should at least be
 set  to the minimum that the gas is  known to
 achieve from  other sources.

      The actual subtraction can be done in
two ways.   The first  way is  to use  the
reference  spectrum of the target gas  and
create  a   spectrum that  represents  the
 amount of the target gas to be subtracted.
 This is  done by  multiplying the reference
- spectrum  by an  appropriate  factor.   This
 factor  can be  calculated  by using  the
 concentration path length product used in the
 reference  gas and the path length that was
 used for data acquisition.

       The  second  way  is  to   do  the
 subtraction interactively with the software.
 In this way, the operator can see the results
 of the  subtraction directly  and has a little
 more   control  of  the  process.     Both
 procedures require some practice,  and the
 operator must be aware that his first attempt
 may not be satisfactory.

 5.4 Calculated Water Spectra

       Some time  ago investigators at the Air
 Force Geophysics  Laboratory in Cambridge,
 Massachusetts, compiled a high-resolution
 transmission molecular  absorption database
 for  the   primary  atmospheric  gaseous
 constituents.  This database is known  as
 HITRAN and contains the data for calculating
 a   water  spectrum.  Software has  been
 developed   that   allows  for  the  actual
 generation  of  the  spectrum, and  that
 software  is called  FASCODE. FASCODE
 generates  a high-resolution  spectrum  using
 Lorentzian   line  shapes. These calculated
 spectra can then be mathematically shaped
 to fit the spectra  produced by a particular
 FT-IR instrument by using the mathematics
 described in Chapter 8.

       While the  software  is  not readily
 available, this procedure has been tested for
                                        5-4

-------
                                                                      TR-4423-99-03
several different resolutions and for several
different  FTIR . instruments. It has  been
determined that this -procedure  works well,
and all of the observable absorption features
are quite accurately reproduced.

      The above-referenced mathematical
procedure will probably be  used by NIST to
generate lower resolution reference spectra
from the high-resolution spectra that they are
acquiring.

      If this procedure can be used,  it is
recommended as the  method of choice for
the production of a water vapor  reference. It
eliminates all of the difficulties  described in
the previous sections about having to remove
the .absorptions of  the other gases  in  the
atmosphere. It has a flat baseline, and once
the procedure is determined, a new reference
can be generated  in a  few minutes' time.

5.5   Methane and Ozone  Examples

      Figure 5-1  shows the  portion of a
single-beam spectrum  over which methane
absorbs.  The methane concentrations at
Research   Triangle   Park   are  generally
measured at about 2.5 ppm. We have seen
methane concentrations as high as about
6  ppm  in  this  area. The spectrum  in
Figure 5-1 actually contains water vapor and
methane, although the methane is not very
noticeable. Figure 5-2 has superimposed on
it the synthetic background  that will be used
to manufacture an absorption spectrum. The
synthetic background has been raised slightly
above the single beam spectrum for clarity.
 •e
                 W«v» Number (cm-')
Figure 5-1. The Portion of a Single-Beam
Spectrum Over Which Methane Absorbs.
           2900
                        2950
                                     3000
               Wave Number (cnr1)
Figure 5-2. Methane Region with Synthetic
Background Spectrum Superimposed.
      Figure  5-3 shows  the  absorption
spectrum that has been made from the two
spectra shown in Figure 5-2. Also shown in
this  figure   is  the  methane   reference
spectrum. The task is now to subtract out
the methane that is present in this absorption
spectrum. There are two absorption peaks in
                                        5-5

-------
 TECHmOHm
                                                                       TR-4423-99-03
 o
 o
 c
 n
 a
 o
 in
 a
      Methane Reference Spectrum
Absorption Spectrum
          2900           2950
               Wave Number (cm-1)
                                   3000
 Figure 5-3. Methane Reference Spectrum
 and the Calculated Absorption Spectrum.
the methane spectrum that are virtually not
impacted by  water  vapor.  They are  at
2916.7  and 2926.8  cm'1.  The  methane
concentration  can be  calculated from  the
peak  height  of  these  two peaks.  The
concentration  path length product of  the
reference spectrum is 81 ppm-m.  The field
spectrum was  taken at a path length  of
414 m, so the reference spectrum represents
a concentration of 81/414 =  196ppb. The
measured peak height of the .2926.8-cm"1
line is 0.00771  absorbance  units for  the
reference spectrum and 0.1069 absorbance
units  for  the field spectrum.  Thus if  the
reference spectrum is multiplied by the factor
0.1069/0.00771  =  13.9 it will represent the
same  amount of methane as is in the field
spectrum. That is to say that the absorbance
in  the  data  spectrum   is  indicative   of
196 ppb  x  13.9 =  2.7 ppm of methane.
After  multiplication,  the  reference can be
subtracted from the field spectrum, and the
methane should be removed from the water
vapor spectrum.
       Figure 5-4 shows the spectrum that
has the methane removed and is now usable
as a water vapor spectrum in the region of
the methane absorbance. The operator must
note, however, that many gases other than
methane absorb in this region and, if they are
still in the spectrum, they will cause errors in
the analysis.
                                                  _ Water Vapor over Mithane Region
                                                  2900         2950
                                                      Wave Number (crrr1)
                                                                          3000
                                      Figure 5-4. Water Vapor Spectrum Made for
                                      the Methane Absorption Region.
                                            Removal of ozone from the  water
                                      vapor spectrum  is  much  more  difficult,
                                      primarily because the absorption feature  is
                                      broad.   Figure  5-5 shows  an atmospheric
                                      .Q
                                      <
                                            Field Spectrum
                                            Showing Ozone Absorbance
                                               1000
                                                            1050
                                                                         1100
                                                    Wave Number (crrr1)
                                            Figure 5-5. Atmospheric Ozone Absorption
                                            Spectrum and Ozone Reference Spectrum.
                                        5-6

-------

                                                                       TR-4423-99-03
 ozone absorption  spectrum and the ozone
 reference spectrum. The slight elevation of
 the  absorption spectrum in the vicinity of
 1050 cm"1 is indicative of an absorbance due
 to  ozone  with  a  concentration of about
 100 ppb.  A significant problem with ozone
 occurs  when the  synthetic background is
 made. Here the absorption spectrum is so
 broad that at least one point  in the ozone
 spectrum region  must  be  chosen  for  the
 baseline.  It is easiest to think that the point
 in the center of the spectrum where  the
 absorbance dips  almost  to  the  baseline
 should  be   used.  However,   the  ozone
 absorbance does not go to zero at that wave
 number.  So, a very  careful  estimate for
 where that point should be placed needs to
 be made. A difficulty arises that is associated
 with the MCT detector, for which there is a
 curvature  of  the   baseline in this wave
 number region. This means that the baseline
 cannot  be  made  by  connecting  a  line
 between two points along the curve.

      Once the synthetic  background has
 been  made   and   the  operator  has  an
 absorption spectrum, the ozone still has to be
 subtracted from  it.  There is almost  no
 possibility that the ozone concentration will
 actually go to zero at any location but it
 should go through  a  minimum.  For areas
 such as Research Triangle  Park, where the
 atmospheric ozone  is  produced locally  and
 not transported into the area from elsewhere,
that  minimum occurs  at about  6:00 in the
 morning.

      A  plot of  several   days of  ozone
concentration  taken at Research  Triangle
 Park during the month of June is shown in
 Figure 5-6. The negative values indicate that
 there is a significant amount of ozone in the
 water vapor spectrum. The question is just
 c
 o
 O
 0.08 -

 0.06 -

 0.04 •

 0.02 -

 0.00 -

-0.02 -

-O.04 -

-0.06 -
             —I—
             20
              —r—
              40
—I—
 60
—1—
 80
—I—
 100
—I—
 120
                 Elapsed Time (hrs)
 Figure 5-6. Ozone Measured at Research
 Triangle  Park During June.
how much to subtract from the spectrum,
because the ozone concentration does not go
to zero.    In this  area,  we  are  fortunate
because  there  are  other  instruments that
make measurements of ozone,  and they can
be used to determine the ozone minimum.
Ozone is  a  criteria  pollutant  and  is also
monitored by the individual states.   These
data  are  generally  available as   hourly
averages and may be useful to the operator
who is trying  to subtract the correct amount
of ozone from  the  water vapor  reference
spectrum.
      The operator  must  be aware that
gases other than methane and ozone must be
subtracted from  the water  vapor reference
spectrum,  even  though  they are all  not
covered here. Certainly the gases C02, N20,
                                        5-7

-------
          itizjjj                                                      TR-4423-99-03
 TCFU- "slssJs^ '•==
 itUft: 'it.! -.?• > • ^
and CO must be subtracted from the water
vapor reference.   When data are taken  at
industrial  sites,  any gas that  is  to be
monitored or used as an interfering species
must  be subtracted from the water vapor
reference spectrum.
                                        5-8

-------
                                                                        TR-4423-99-03

                                     Chapter 6
                                        Siting
                                      SUMMARY

           The topics and specific points of emphasis discussed in this chapter include
    the following

           •  Siting considerations for long-term and short-term monitoring efforts

           •  Factors to consider when selecting the path

              •  Short path versus long path
              •  Prevailing winds
              •  Slant path versus horizontal path

           •  What conditions warrant changing the path

           •  What ancillary measurements are required

           •  Calculation of the minimum path length required to detect specific
              concentrations of selected target gases

           •  An example of a specific monitoring site
6.1    Introduction and Overview

       There are  two  kinds  of  monitoring
programs  for  which  siting  needs  to  be
discussed. One is a long-term effort with the
instrument placed in a more or less permanent
position. The second is a short-term program
designed to take data  at  a site for a period
from a few days to a  few weeks.   Each of
these  situations,  while   similar,  requires
somewhat different thinking to actually site
the instrument.   The short-term  program is
more flexible in that the  path configuration
can be based on the meteorological conditions
at the time of the monitoring program. Long-
term monitoring programs must be designed
to allow for changes in the direction of the
path as dictated by changing meteorological
conditions, or  useful  data  might be  lost.
Siting considerations for both situations are
described   in  this  chapter.    Criteria  for
selecting the path, changing the path,  and
choosing  the  ancillary measurements  to
make at a monitoring site are also discussed.

       There   is  little  information in  the
literature   pertaining  to   the   long-term
monitoring program.  Even for the short-term
program,   the  parameters    that   were
considered in selecting the actual direction
and length of the path have been discussed
in  only  a cursory  manner.    A typical
                                         6-1

-------
                                                                       TR-4423-99-03
 statement is "The path was set up based on
 a knowledge of the prevailing winds."  But
 what the wind field actually, was during the
 observation period or where the  prevailing
 wind  data were  taken  is  almost never
 presented.    There  is  a  similar   lack of
 information concerning the selection of the
 path length and the partial pressure of water
 vapor  in the atmosphere.   For that matter,
 there is almost no discussion in the literature
 about how the  length of the path is selected.

       There are some sophisticated methods
 being studied by various groups that use  real-
 time meteorological data for making decisions
 about  the  path.    It  seems  that these.
 techniques  are  more  suited for permanent
 monitoring installations and not for short-term
 programs such as monitoring at small waste
 sites.    Under  any   circumstances these
 methods have  not been adequately tested,
 and  an  evaluation  of  these  methods is
 considered to be beyond the scope of  this
 document at the present time.

       Other ongoing relevant work is being
 documented by  the  U.S.   Environmental
 Protection Agency, which has prepared a set
 of changes to Part 58 of Chapter 1 of Title 40
 of the Code of Federal Regulations (40CFR58)
 that  define  the  appropriate  ambient   air
 monitoring criteria for  open-path (long-path)
 monitors  (U.S.  Environmental  Protection
 Agency,  1994).  These  amendments have
 been finalized   and  approved,  and  they
 specifically  address  the monitoring of  the
 gases  called  the  criteria  pollutants.  The
 amendments are  significant  in that they
 describe just how the path is to be chosen in
terms  of obstructions,  height  above  the
 ground, and changes in path height.  They
 also describe the appropriate positioning of
 the path in relation to buildings, stacks, and
 roadways.

       Several  factors must  be  considered
 when  selecting the path.   These factors
 include (1) instrumental parameters, such as
 the signal-to-noise ratio (S/N) of the system
 and the divergence of the IR beam; (2)  the
 characteristics  of the target gases, such as
 concentrations  and absorption coefficients;
 (3)  the  presence  and concentrations   of
 interfering species, such as water vapor and
 C02; (4)  meteorological data, such as wind
 direction  and  speed;  and  (5)  physical
 constraints, such as the area of the emission
 source, the extent of the plume,  and the
 availability of suitable sites to accommodate
 the instrument and peripherals.

       Example calculations using Beer's law
 are given in this chapter to illustrate the
 minimum path length required to measure a
 specific concentration of  a target gas.  For
 example, the minimum path length  required
 to measure ammonia at a concentration of
 10 ppb  would  be  approximately  21  m,
 assuming  a  minimum detection  limit  of
 3  x  10"4  absorbance units,  no  interfering
 species,   and   a   uniform   concentration
throughout  the  path.  In  contrast,  the
minimum path  length required to measure
 10 ppb of chlorobenzene  would be 230  m.
 In general, the  length of  the  path must  be
chosen to be the  minimum length that will
allow the measurement to be made with a
meaningful statistical accuracy.

       As an aid to the operator, Table 6-1,
listing minimum detection  limits for various
                                        6-2

-------
                                                                       TR-4423-99-03
 gases,  is included. The limits in Table  6-1
 were   calculated   by  using  a  minimum
 detectable absorbance of 10"3...The units used.
 in this table are ppm«m.

       An example using a specific Superfund
 site is given in this chapter. Procedures  are
 described for selecting a usable path for a
 short-term intensive study.

 6.2   Selecting the Path

       To select the length and the position of
 the  path, the investigator must have  some
 understanding of the  ramifications of  these
 choices.  The immediate  questions concern
 (1) the effect that the path has on the data
 that is produced and  (2) the procedure that
 the operator follows for selecting a path.

       Preliminary answers to those questions
 are found by referring  to Beer's law.  But the
 complete answers  are more  complicated.
 They include  scattering  and  absorption  by
 aerosols,  the  effects  of  water  vapor and
 carbon  dioxide  on the  S/N,  and  spectral
 interferences.   When  measuring plumes of
 finite extent, a path longer than the width of
the plume is actually detrimental.

       The amendments to 40CFR58 describe
the following considerations for selecting the
path.

   •  At least 80%  of the path must  be
      between  3  and  15  m above the
      ground.
    •  At least 90% of the path must be at
       least  1  m  vertically or horizontally
       away .from, walls, etc.

    •  If the path has to be near a building,
       then it must be on the windward side
       of the building.

    •  Buildings or  other obstructions  may
       possibly  scavenge   the  gases  of
       interest.  At least 90% of  the  path
       must have unrestricted airflow and be
       located  away from  obstructions  so
       that it is removed by at least twice
       the height that the obstacle protrudes
       above the path.

    •  At least 90% of the path must be at
       least 20 m  from the  drip lines  of
       trees.

    •  When monitoring is  done for  ozone,
       90% of the  path must be at least
       10m from a road that carries fewer
       than 10,000 cars a day. This criterion
       changes to 250 m for heavily traveled
       roads (> 110,000 cars per day).

       There are other  changes to 40  CFR
Part 58  that are applicable to the use  of
FT-IR open-path monitors, but they are for
concerns other than siting.  The interested
reader should obtain a copy of 40 CFR Part
58  from the  Office of  Federal  Register,
National  Archives  and Records  Adminis-
tration, Washington, DC.  It is also available
in most public libraries.
                                        6-3

-------
•\i:s:'::s-.'= =
mm
                                                                  TR-4423-99-03
   TABLE 6-1. Estimated Method Detection Limits (MDLs) for Selected Gases1
Compound
acetaldehyde
acetonitrile
acrolein
acrylic acid
acrylonitrile
ammonia
benzene
bis-(2-chloroethyl)ether
bromomethane
1,3-butadiene
2-butanone
carbon dioxide
. carbon disulfide
carbon monoxide
carbon tetrachloride
carbonyl sulfide
chlorobenzene
chloroethane
chloroform
chloromethane
m-dichlorobenzene
o-dichlorobenzene
dichlorodifluoromethane
1,1-dichloroethane
1,2-dichloroethane
1,1-dichloroethene
1,2-dichloroethene
dichloromethane
1,1-dimethylhydrazine
ethylbenzene
ethylene oxide
formaldehyde
hexane
hydrogen chloride
hydrogen fluoride
hydrogen sulfide
isooctane
methane
methanol
methylmethacrylate
nitric oxide
nitrobenzene
nitrogen dioxide
nitrous oxide
ozone
phosgene
phosphine
propionaldehyde
propylene oxide
styrene
sulfur dioxide
sulfur hexafluoride
tetrachloroethene
toluene
1,1,1-trichloroethane
1,1,2-trichloroethane
Class"
caa
caa
pp.caa
caa
pp,caa
PP
pp.caa
pp.caa
pp.caa
caa
pp.caa
ag
pp.caa
cp
pp.caa
caa
pp.caa
pp.caa
pp.caa
pp.caa
PP
PP
PP
pp.caa
pp.caa
PP
pp.caa
pp.caa
caa
pp.caa
pp.caa
caa
caa
caa
caa
caa
caa
ag
caa
caa

pp.caa
cp
ag
cp
caa
caa
caa
caa
caa
cp
tracer
pp.caa
pp.caa
pp.caa
pp.caa
v b
'max
(cm-1)
1761
1463
1730
1726
954
967
673
1138
1306
908
1745
2361
1541
2173
795
2070
740
1288
772
732
1581
749
1161
705
731
869
864
750
2775
2975
3066
1745
2964
2945
4038
1293
2961
3017
1033
1169
1894
1553
1629
2213
1054
849
2326
1762
3001
695
1377
947
915
728
725
742
MDLb
(ppb-m)
2063
8403
1297
639
3398
620
266
2157
11547
1445
1483
637
191
4583
178
240
1341
6744
359
6652
1266
1428
294
2049
1983
1241
5024
1174
1962
2031
987
1248
1023
3164
578
535003
554
1597
1249
1199
4388
852
540
932
2533
318
7699
2305
2838
1720
372
42
708
1632
533
1615
v c
max
(cm-1)
2729
1042
958
1439
971
931
3047
767
2983
1014
1175
668
1527
2112
773
2051
1483
677
1219
1459
784
1462
921
1060
1237
793

1276
909
697
872
2802
1467
2822
3877


1305
2982
1748
1843
1355
1599
1300
1040
1832
992
2992
837
909

615
781
3018
1088
941
MDLC
(ppb-m)
6674
46095
4509 '
1326
4548
718
4449
4372
12455
5719
3224
608
266
5417
1027
330
3980
6871
1927
9517
1305
5142
303
3053
6803
1814

4113
3774
2277
3327
2581
7710
3620
761


2998
5933
1341
6816
1049
742
3946
3971
667
12468
4107
4549
2908

420
2654
3583
1183
7933
                                  6-4

-------
                                                                         TR-4423-99-03
 refi/.-rs-ssT^i -^
 /CUfts •i.-J -.=i J ; ^=
           TABLE 6-1. Estimated Method Detection Limits (MDLs) for Selected Gases1
Compound
trichloroethene
trichlorofluoromethane
vinyl acetate
vinyl chloride
vinylidene chloride
m-xylenc
o-xylene
p-xylene
Class3
pp.caa
PP
caa
pp.caa
caa
pp.caa
pp.caa
pp.caa
v b
max
(cm'1)
849
846
1225
942
868
768
741
795
MDLb
(ppb-m)
1173
178
688
2824
1669
1601
1070
1765
v c
max
(cm-1)
944
1084
1790
1620
1086
690
2949
2936
MDLC
(ppb-m)
1578
634
1327
3643
2501
3825
5797
3340
      MDLs were estimated by using value's of the absorptivity calculated from 1-crrr1 reference spectra with
 triangular apodization from a commercially available spectral library and a minimum detectable absorbance of
 1 * irj-3.
  "Pollutant classification: priority pollutant (pp); criteria pollutant (cp); hazardous air pollutant from the 1990 Clean
 Air Act Amendment (caa); atmospheric gas (ag).
  "Peak position and MDL for the most intense absorption band.
  cPeak position and MDL for the second most intense absorption band in a different spectral region.
6.2.1  The Longest Path

       It  is   possible  to  determine   the
maximum usable path in several ways.  One
is to use the noise equivalent power of the
detector as the minimum signal that can be
recorded. Another is to use a minimum S/N
that the operator is willing to accept.  Then
if  either the  noise equivalent  power (NEP)
signal  or the S/N  is  known at  one distance
and the energy falls off as the inverse square
of the  distance,  the maximum possible path
can be calculated. Any attempt to actually
do this calculation results in the conclusion
that the maximum path is  essentially infinite.

       In a practical sense the absorption due
to water vapor will limit the usable path long
before the theoretical limit calculated above
can be reached.  The  water vapor line at
1014.2 cm"1 has an absorbance of 0.01 at a
total path length of about 30  m when  the
water vapor partial pressure is 10 torr. If an
absorbance of 1  is considered the maximum
allowable for this line, then the  maximum
usable total path is about 3 km.

       The  companion  to this  document
(Russwurm  1997)  describes a  method for
determining the minimum detection limit for
open-path FTIR systems. It has recently been
shown by investigators in Germany (Dovard
et al.  1 997) that this detection limit does not
vary  with  path  length as  is predicted by
Beer's law.  This is probably caused  by the
variability in the atmospheric  constituents
themselves. At any rate, the work by Dovard
et al.  implies that a maximum usable path is
more  on the order of 400-500 m.

6.2.2  Shortest Path Requirements

   The shortest  path for various gases can
be calculated from the absorbance  measured
in the reference spectra, a knowledge of the
minimum  measurable  absorbance, and  the
assumption that reciprocity holds.   To make
this  calculation, the  operator  must  have
chosen the wave number region that will be
                                         6-5

-------
                                                                     TR-4423-99-03
used   for   analysis  and   obtained   the
absorbance  of the gas from the reference
spectrum  over  that region.  The.operator
must also choose a minimum concentration
that is  to be measured. Then,  by using the
minimum   detectable   absorbance,   the
minimum path can be calculated as follows.

   1.   Measure  the  absorbance  at   the
       appropriate  wave  number for  the
       target  gas  from   the  reference
       spectrum.  Record the concentration
       path  length product  at  which  this
       spectrum was taken.

   2.   Calculate the absorption coefficient a
       for this gas  by  using the following
       formula.
              CL=Ar/CrLr
       where A  is  the absorbance and CL is
       the   concentration-path   length
       product.  The subscript r refers to the
       reference spectrum.

   3.   Assume  a   minimum  concentration
      that will  be measured, and set  the
       minimum detectable  absorbance at
       3 times the RMS baseline noise as
      measured  under  normal  operating
      conditions,  for example, 3 x  10~4.

   4.  Calculate the minimum  usable path
      (Lm) from
             Lm = Am/aCm
      where Am is the minimum absorbance
      (3 x  10~4)  and  Cm is the minimum
      concentration assumed in Step 3, and
      a  is   the   absorption   coefficient
      calculated in Step 2.
       The results of the above calculations
 for  four  different  gases  are  given   in
 Table 6-2.       .          -  -

 6.2.3  Short Path Versus Long Path

       As shown in the previous section, the
 selection  of  the  path length  begins  by
 calculating the minimum usable length from
 Beer's law.  If  a  retroreflector is used, the
 physical path can be half the optical path
 determined above.   This  is advantageous
 when  plumes  of  finite  size  are  being
 measured  because the path length may be
 chosen close to the  physical extent of the
 plume.

       The  length of  the  path  must  be
 chosen to be the minimum length that will
 allow the measurement to be made with a
 meaningful  statistical  accuracy.   For the
 calculations   above,   this  distance  was
 determined by using a minimum absorbance
 of 3  x  10'4, or about 3  times the  best
 detection  limit  that  is achievable  at the
 present time.  For homogeneously distributed
 gases, the  path can be made longer  with
 some advantage.   But for plumes of finite
 extent, making the  path  longer  than the
 plume  is wide  would  be a  detriment and
 should not  be done.   This  is because the
 measurement actually determines the  path
 average concentration, and if a portion of the
 path  has  zero  concentration,  there  is  a
dilution effect. Another reason for choosing
 a path that is as short as possible is that the
effects of  spectral  interferences will  be
minimized.
                                       6-6

-------
                                                                    TR-4423-99-03

               Table 6-2.  Minimum Usable Path LengthsH
Wave
Number CrLr
Gas
p-dichlorobenzene
chlorobenzene
toluene
benzene
(cm'1}
822
1025
1031
1038
Ar (ppm-m) a
0.085
0.0227
0.0203
0.0027
500
170
496
27
1.7 x 10'4
1.34 x 10"4
4.09 x 10"5
1.0 x 10"4
cm
-
(ppb) (m)
10
10
10 '
10
176
223
734
300
* A, = absorbance of the reference spectrum, C,L, = concentration-path length product at
which the reference spectrum was taken, a = absorption coefficient, Cm = minimum
concentration, Lm = minimum usable path length.
       A different example can be described
as follows.  There are times when a release
of a tracer gas such as SF6 is desirable. The
question is how much must be in the path if
it is to be detected.   From Figure 6-1, it is
seen that the absorbance of SF6 at 947 cm"1
is  1.56.   The concentration path length is
66  ppm  • m.  Thus for a working detection
limit of 3  x  10"4 absorbance units and a path
length of 50  m, the  minimum  average
concentration in the path must be C = A/aL.
The absorption coefficient a is obtained from
a = A/CL =  1.56/66. Thus C= (3 x 10'4 x
66)7(1.56 x  50)  =  0.25 ppb.  It is clearly
seen in this example that because of its large
absorption  coefficient,  not much  SF6  is
required for detection.         '.

      There is no distance with any of the
available  instruments  that  will reduce the
absorption due to water vapor and carbon
dioxide below the detection limits.  In fact, in
the  wave  number  region  of  strongest
absorbance  for these gases the atmosphere
is genera.lly totaljy opaque.  That is, there is
                                        so much  light  being absorbed that none
                                        returns to the detector.  These regions are
                                        not usable for data analysis with the FT-IR
                                        systems.
                                          1.71S
                                          0.156
                                          0.000
                                             007  816  924  933  942  950  959  968  976 985
                                                     Wave Number (cm-')
                                        Figure 6-1.  Sulfur Hexafluoride Reference
                                        Spectrum. (Used with permission of P.L.
                                        Hanst)
                                              For long-term  monitoring  programs
                                       with  permanent  installations, the only real
                                       option  is to  place  retroreflectors or  light
                                       sources  (depending  on  the   instrument
                                       configuration)  at  various  distances  and
                                       switch  from one  to the other  periodically or
                                    6-7

-------

                                                                       TR-4423-99-03
 on  some   predetermined   schedule.    A
 scanning  system  is  available with  some
 versions of open-path FT--IR-monitors that
 facilitates this.  Currently, almost no work
 has been done to define  various lengths for
 various conditions. Thus, this chore must be
 individually  repeated  for each  monitoring
 program.

 6.2.4  Prevailing Winds

       When  using   the  FT-IR  long-path
 technique, the operator depends on the wind
 to  deliver  the gases  being  emitted  by a
 source to the infrared beam.  Knowledge of
 the  prevailing  winds  is  important  when
 setting up the path for long-term monitoring
 programs, but may be much less  important
 for short-term programs.  Most operators of
 open-path  monitors have been  concerned
 with short-term programs and know that the
 wind almost never comes from  where the
 prevailing wind rose predicts. The short-term
 program usually demands that the operator
 be prepared to change  the path configuration
 when the wind changes. For either the long-
term or the short-term program,  the  ideal
situation  is  to  have  more  than   one
retroreflector or light source.  This allows the
path direction  and length to be changed as
the  requirements of  the program dictate
without having to transport  the instrument
itself.

      When emission rates  need to  be
calculated from data  taken  with  an FT-IR
instrument,  the wind  direction and speed
must be known. The direction of the path
with respect  to the  wind  must also  be
 known.   A knowledge  of  the  historical
 prevailing winds is of little use for this task.
 When emission rates are required, the wind
 field at the path must be measured directly.

 6.2.5  Slant Path Versus Horizontal Path

       Path orientation is important because
 the   wind   is  the   primary   mode   of
 transportation of the gases being monitored.
 Wind  speed  and  direction  can  change
 dramatically   over   small   regions  when
 measured close to the ground.  This is true
 not only because of the changing terrain but
 also because the motion of the air (a wind)
 must at least approach zero at the surface.
 There   is   some   indication   that   the
 concentration contours of gases become very
 complex  with  altitude,  at  least  in  part
 because of turbulence. There are no data in
 the FT-IR literature that describe the variation
 of concentration with altitude.  Because of
 these uncertainties, a comparison of the use
 of a slant path and a horizontal path cannot
 be made.

 6.3    Changing the Path

       The beginning of this chapter included
 a discussion of some of the ramifications of
 path selection. The question here is when
 should the  path  length  or direction  be
 changed?  Obviously, if the plume from a
 point source or an  area source  is being
 monitored and the wind changes direction,
the path  should be changed.  Changing the
 path, however, should be done in accordance
 with some  plan.   Items that need to be
 covered in  the plan include the conditions
                                        6-8

-------
 TCfU r~s75s~si ~.^=
 ItUtl- -i-J -.-i * : ^=
                                                                      TR-4423-99-03
 that make a change necessary, as described
 above.   They  should  also  consider  the
 ramifications  of  the  change,   both -the
 advantages and the disadvantages.   For
 example,  if the concentrations  of  gases
 crossing a fence line are being  monitored,
 there is iittle point in changing the direction
 of the path.

       Change  in  the  length  of the  path
 should be considered only for purposes of
 taking a  background  spectrum  or  when
 spectral interferences from compounds like
 water vapor become  so strong that  the
 absorption  due  to the target compounds is
 overwhelmed.     Whether  any  accurate
 monitoring can be done under that condition
 has not been studied. Certainly, it makes no
 sense to reduce the path  length to the point
 where the  target compounds cannot be
 monitored.

 6.4   Ancillary Measurements

      There are several reasons  why some
 ancillary measurements must be made when
taking data with an FT-IR open-path sensor.
 One is the requirement to take data that can
 be  used  for quality  control  and quality
 assurance purposes.  (See Chapter 10 for a
discussion  of  quality control  and quality
assurance procedures.) Another is that many
programs will require ancillary data such as
wind  speed and direction.  Also, for the
foreseeable  future,  the  amount   of water
vapor in the atmosphere should be  monitored
because too many  unanswered  questions
about water vapor exist.  By far, water vapor
represents   the   strongest   spectral
 interference,  and  unless  it is measured
 separately,  problems  may  arise when the
 data are analyzed.  It should be noted that a
 measurement  of relative humidity is not
 satisfactory for this work. The actual partial
 pressure of water vapor must be found, and
 if relative humidity is  measured,  then the
 temperature must  also  be  measured.   The
 ambient pressure should also be recorded.
 At any one monitoring location operators can
 expect  to experience  a small change in
 ambient atmospheric  pressure.   In some
 cases, the data may have to be corrected for
 these changes.  However,  when  acquiring
 data  in places  of high  altitude,  such as
 Denver, CO, a substantial change in pressure
 can  be expected when  compared  to  sea
 level. The operator must determine whether
 his experiment demands that these changes
 be accounted for in the data.

       Guidance  for selecting and setting up
 the instruments  for making  meteorological
 measurements can be found  in a government
 handbook  (U.S.  Environmental Protection
 Agency 1989).  Although  this document
 does  not  directly  address  the long-path
 measurements,  it  does   present  useful
 information    about   meteorological
 instrumentation and measurements.  For the
 long-path situation, the only measurements
that should be obtained are probably those at
 or along the path itself.

 6.5   A Specific Case

      The  literature   offers   very   little
information about procedures for selecting
the path at any given site. The task is again
                                       6-9

-------
                                                                       TR-4423-99-03
 divided into two parts:  (1) selecting a path
 or set of paths for long-term monitoring at a
 fixed installation and (2) selecting a usable
 path  for short-term intensive studies.  For
 this document, we will discuss the latter
 case  only  and do  so by presenting  a real
 case.

       Figure 6-2 is an aerial photograph of
 a  Superfund  site  undergoing remediation.
 The active region of the site is at the middle
 left of the photograph. Two large repository
 pits can be seen, one in the top middle of the
 photograph and the other in the top right.
 The former pit has been  filled and capped
 with dirt while the  latter is open and  in the
 process of being lined. To gauge the size of
 the site, note that the vehicles immediately
 to  the  left of the FT-IR monitor  are D8
 bulldozers.

       The site lies  between two ridges (not
 shown), and the prevailing winds blow along
 the valley through the site from upper left to
 lower right. The first pit rises sharply from
 the active area for about 50 ft, and then the
 terrain falls off about 20 ft to the second pit.
 The road in front of the FT-IR monitor rises
 sharply  in  front of the  second pit.   The
 surrounding terrain  is forest,  with the trees
 rising about 40 ft above  the ground  level.
 Permission  had been obtained  to  make
 measurements  with the FT-IR monitor with
the  proviso   that  there  would  be  no
 interference with  the ongoing remediation.
The FT-IR operators were not given access to
the active  area,   which  was  defined as
starting at the buildings in the foreground and
extending to the forested region to the left of
 and behind the capped pit.  The remediation
 operation entailed digging soil from the active
 area,-  repacking  it-in  metal  drums,  and
 moving it to the lined pit areas.  Fluids that
 were encountered in the active area or in old
 drums were  brought  to  the settling tanks
 seen in the middle of the photograph.  The
 predominant  chemical in the site was the
 herbicide Dicamba.  Dicamba has a very low
 vapor  pressure, but two by-products were
 thought  to   be   present.    They  were
 benzonitrile and benzaldehyde, and the goal
 of  the FT-IR  study  was  specifically  to
 measure these two compounds.  Funding had
 been allocated for one week of field work.
 Although the  proposed  amendments to 40
 CFR Part 58 were not available at the time of
 this study,  they  were  almost  exactly
 followed.

       Benzonitrile  has  a  single   usable
 absorption  band  at  757  cm"1.     The
 absorbance of this band is 0.0346 when the
 concentration-path   length   product   is
 186 ppm-m.   Repeating  the   calculation
 described  above   for   the   absorption
 coefficient  gives  a  =  0.0346/186   =
 1.9 x 10~4.  It  was thought that benzonitrile
 would have a concentration of about 10 ppb.
 At  the  time   of  this   study,   the  FT-IR
 instrument had a minimum detection limit  of
 about  1  x  10"3  absorbance units.    The
 minimum  usable  path   is  calculated   as
follows.

 Lm  =   1 x 10'3/{1.9 x 10'4 * 10 x 10'3)
    =   525 m
                                (Eq. 6-1)
                                        6-10

-------
Figure 6-2. Aerial Photograph of a Superfund Site Undergoing Remediation.
M
CO

CD
CD

6
CO

-------

                                                                      TR-4423-99-03
        The problem  is  somewhat  more
 cqmplicated  than  this because  there is a
 weak  interfering  carbon  dioxide  peak  at
 757 cm'1.   The task was then to site the
 instrument so that the retroreflector could be
 placed 250 m  away while adhering to the
 constraints  imposed by  the remediation
 process. To obtain electrical power without
 the use of a generator, the only logical place
 to put the FT-IR monitor  was at the  main
 entrance to the site,  as it is shown in the
 Figure 6-2 photograph. From there, only two
 possible  paths  of  250 m were  available.
 They are shown in  the photograph at the RR
 positions.  The path that extends from the
 FT-IR monitor to the right RR position in the
 photograph was selected as the primary path
 because  this would  encompass  the entire
 plume coming from the active area according
 to the prevailing winds. The secondary path,
 extending from  the  FT-IR monitor to the
 bottom of the capped repository  pit (at the
 left RR position), was not really satisfactory
 because it  rose too high above the ground
 level and  went directly  over the settling
 tanks.  A path that  is too high  above the
 active area would  leave the possibility that
 the plume might go under the beam. If the
 beam were to go directly  over the settling
 tanks, flux calculations would be  virtually
 impossible.

        As an aside, during the week-long
field program the remediation process  was
 halted because of a problem with the lining
 of the second pit. Also, during this time, the
 wind never blew more than 0.5 mph, and its
 direction was almost always from the bottom
 to the top of  the photograph, contrary to
 expectation.
6.6
References
Douard,  M., J. Zentzius-Reitz,  T.  Lamp,
A.  Ropertz, and K. Weber.  1997. Quality
Assurance Procedures and Measurements for
Open  Path  FTIR   Spectroscopy  Europto
Series. Proceedings of the Envirosense '97
Meeting  in   Munich,   Germany.   SPIE
3107:114-125.

Russwurm, G.M. 1 997. Compendium Method
TO-16.   Long-Path  Open-Path  Fourier
Transform   Infrared   Monitoring   of
Atmospheric Gases.  In  Compendium of
Methods for the  Determination of  Toxic
Organic Compounds in Ambient Air, Second
Edition,   EPA/625/R-96/01 Ob,   U.S.
Environmental Protection Agency, Research
Triangle Park, NC.

U.S.  Environmental  Protection  Agency.
1989.  Quality Assurance Handbook for Air
Pollution Measurement Systems, Vol.  IV -
Meteorological   Measurements.      U.S.
Environmental Protection Agency, Research
Triangle Park, NC.

U.S.  Environmental  Protection  Agency.
1994. Ambient air quality surveillance siting
criteria for  open path  analyzers  (proposed
rule).  Fed. Reg. 59(1 59):42541-42552.
                                       6-12

-------
                                                                       TR-4423-99-03
      - 'if -.- - •
                                     Chapter 7
   Resolution Considerations in FT-IR Long-Path, Open-Path  Spectrometry
                                     SUMMARY

          The topics and specific points of emphasis discussed in this chapter include
    the following.

          •  The definition of resolution in FT-IR spectrometry

          •  The trading rules between resolution, the signal-to-noise (S/N) ratio, and
             measurement time

          •  Example spectra of atmospheric constituents  and selected  VOCs  that
             illustrate the following effects.

             •   Effect of resolution on peak shape and intensity

             •   Effect of apodization and zero filling on peak shape and intensity

          •  A discussion  of the effect of resolution on quantitative analysis

          •  A case study illustrating the effects of resolution, zero filling, and baseline
             noise on the CLS analysis of multicomponent mixtures
7.1    Introduction and Overview

       One important issue regarding the use
of long-path, open-path  FT-IR systems for
monitoring hazardous air pollutants  is the
appropriate  spectral  resolution  to be used
during data acquisition.  A resolution  should
be chosen to maximize the ability to resolve
spectral overlap while maintaining a balance
between the S/N, analysis time, and data
storage requirements. Several factors must
be considered when determining the optimum
resolution for measuring the  IR  spectra of
atmospheric constituents along an open long
path.  These factors include (1) the ability to
distinguish between the spectral features of
target  analytes  and those  of  ambient
interfering species in the atmosphere, such
as water vapor and CO2; (2) the trade-offs
between resolution, IR  peak  absorbance,
and S/N; and  (3) practical considerations,
such as measurement time, computational
time to process the interferogram, and the
size  of  the  interferogram  file for data
storage.   The  use  of  an   inadequate
instrumental resolution can distort the true
absorption spectrum, affect the quantitative
relationship   between  absorbance  and
concentration, and diminish the ability to
resolve spectral overlap. Resolutions ranging
                                        7-1

-------

                                                                       TR-4423-99-03
 from 0.25 to 2 cm"1 have been suggested for
 use in FT-IR monitoring, but there currently is
 no  consensus  as  to what  resolution  is
 generally applicable. The nonlinear response
 caused by the apodization function discussed
 in Chapter 8  strongly indicates that higher
 resolution improves the accuracy of the data.
 practical terms this means that  the  scan
 time  will  be approximately twice as  long,
 the interferogram file will be approximately
 twice the size for data storage, and the time
 required to process the interferogram will be
 longer   for   the   higher   resolution
 measurement.
       This   chapter   describes   the
fundamental aspects of resolution in FT-IR
spectrometry  and illustrates the effects  of
resolution   and   related   instrumental
parameters on the measured spectrum. The
trading rules  that determine  the balance
between resolution and S/N are discussed.
Test spectra were obtained in the laboratory
and  along an open path to illustrate the
effects of  resolution, apodization, and zero
filling on the  IR spectra of C02 and water
vapor, common atmospheric species that can
interfere with  analytical measurements, and
selected gases and VOCs. Studies from the
literature   that   address    resolution
requirements in  long-path FT-IR monitoring
are discussed. A case study illustrating the
effects of resolution, zero filling, and baseline
noise on the CLS analysis of multicomponent
mixtures is also presented.

       In FT-IR spectrometry, the minimum
separation  in  wave numbers  (cm"1)  of two
spectral features that can be just resolved is
inversely related to the maximum optical path
difference in centimeters of the  two mirrors
employed in the Michelson interferometer.  If
the desired resolution is increased by a factor
of 2,  for  example,  from  1-  to 0.5-cm"1
resolution,   the   moving   mirror  in  the
interferometer must travel twice as far.  In
       The  instrumental  resolution  also
affects the S/N.  In general, if the  size of
the  aperture, or Jacquinot  stop,  in  the
interferometer is held constant, the baseline
noise  in  an   FT-IR  spectrum  is  directly
proportional  to  the  resolution  of  the
interferometer for measurements made in
equal times.   For example,  changing  the
resolution from 1- to 0.5-cm'1 increases the
noise  level  by  a  factor  of 2  for  equal
measurement times.  Therefore, to obtain
the  same baseline noise  level for  the
0.5-cm"1 spectrum as was measured for the
1-cm"1  spectrum, the  measurement time
would have to be quadrupled (because  the
S/N is proportional to the square root of the
measurement time).

       As described in Chapter 8, resolution
also  affects  the  peak absorbance  of  the
bands being measured.  For narrow spectral
features,  the  peak absorbance will only
approximately  double  on   halving   the
resolution. In this case, the  S/N is nearly
the  same for the spectra acquired at  the
higher  and  lower  resolution   settings,
provided the  measurement time is equal.
For broad spectral  features  whose peak
absorbance does not change  appreciably as
a function of resolution, the lower resolution
                                        7-2

-------

                                                                       TR-4423-99-03
 measurement  is preferable.   For  strongly
 absorbing bands, whether they are  broad or
 narrow, calibration curves  of  absorbance
 values measured at different resolutions and
 plotted  versus  concentration  must  be
 developed  to   ascertain   the   optimum
 resolution to  be used.  In light of the  work
 outlined in Chapter  8 it seems to be better to
 use the  highest resolution at which the
 instrument can operate.

       In  general,  the  minimum  limit  of
 detection  (LOD)   should  be  found  for
 measurements made  at the lowest  possible
 resolution  that  adequately  resolves  the
 spectral features of the analyte from those of
 interfering species.  The use of an inadequate
 resolution  can distort the  true  absorption
 spectrum, affect the quantitative relationship
 between absorbance and concentration, and
 diminish  the   ability  to  resolve  spectral
 overlap.  Conversely,  the use of a higher
 resolution than  is required  can result in  a
 poorer S/N and  an  unnecessary increase in
 measurement  time, processing  time, and
 data storage requirements.

       There is currently no consensus as to
 what resolution  and related  parameters are
 generally applicable in long-path,  open-path
 FT-IR monitoring. Most likely, the optimum
 resolution will need to be determined  on  a
 case-by-case  basis,   depending  on  the
 spectral   characteristics  of  the   target
 compounds and their concentration, the path
 length,  and  the presence  of  interfering
 species. In field measurements, a qualified
judgement must by made taking into account
these factors  in addition to the practical
 considerations  discussed  above  and  in
 Chapter 8.

 7.2    Definition of Resolution

       An understanding of the resolution
 requirements in FT-IR long-path, open-path
 monitoring requires an understanding of the
 basic principles  involved in generating  an
 interferogram and the operations performed
 on the interferogram  prior to converting it to
 a spectrum. The following  discussion is an
 attempt to describe these basic principles in
 a way that will be of general use to analysts
 in FT-IR monitoring.   For a more rigorous
 treatment of the  fundamentals  in  FT-IR
 spectrometry, the  reader is referred to the
 definitive  text by  Griffiths  and de Haseth
 (1 986) or several other excellent references
 (Horlick  1968;  Bell 1972;  Herres and
 Gronholz 1984)  and Section  2.4.2  of this
 document.

       As shown  in Section  2.4.2,  the
 minimum  separation in  wave numbers  of
 two  spectral features that can be resolved
 is inversely related to the maximum optical
 path difference, in  centimeters, of the two
 interferometer  mirrors  employed  in the
 Michelson interferometer.  The closer the
 separation of the two spectral features, the
greater the optical  path difference must  be
 before the spectral features can be resolved.

       In terms  of the measured spectrum,
resolution can be defined as the minimum
separation that  two  spectral  features can
have  and still be  distinguished from one
another. A commonly used  requirement for
                                         7-3

-------
                                                                       TR-4423-99-03
 two  spectral  features  to be  considered
 resolved  is  the  Raleigh  criterion.   This
 criterion states that  two  bands that  have
 identical intensity, band  shape, and  peak
 width are resolved when the minimum of one
 band falls  on  the  maximum of the other.
 When this  is  the  case,  there  is a dip
 corresponding to approximately 20% of the
 absorption   maxima   between   the  two
 overlapping spectral features.  It should be
 noted that this criterion is valid only  for a
 sine2 instrument line shape, such as  that
 found in a dispersive spectrometer or  an
 FT-IR instrument using triangular apodization.

       The actual spectral resolution in the
 frequency domain that can be obtained by an
 interferometer   is  also  affected   by   the
 truncation  of  the  interferogram  and   the
 application of various apodization functions.
 The apodization functions  can increase the
 bandwidth  and also change the  line shape.
 Apodization   is   discussed   further   in
 Section 7.3.3 and in Chapter 8.

 7.3    Trading Rules in FT-IR Spectrometry

       The   quantitative   relationships
 between   the   S/N,   resolution,   and
 measurement time in FT-IR spectrometry are
 referred to as "trading rules". The factors
that affect the S/N and dictate the trading
 rules  are expressed in Equation 7-1, which
gives the S/N of a spectrum measured  with
a rapid-scanning  Michelson interferometer.
 (The derivation of Equation 7-1 is  given by
Griffiths and de Haseth [1986].)
         N
 where t/v(T) =
          9

         Av
          t =
        £>»  =
        AD =
                      1/2
                  (Eq. 7-1)

the spectral energy density
at wave number v from a
blackbody source at a
temperature T
the optical throughput of
the spectrometric system
is the resolution of the
interferometer
is the measurement time in
seconds
the efficiency of the
interferometer
the specific detectivity,  a
measure of the sensitivity
of the detector
the area of the detector
element
       As shown in Equation 7-1, the S/N of
a spectrum is proportional to the square root
of  the   measurement  time   (tK).   For
measurements made with a rapid  scanning
interferometer operating at a constant mirror
velocity at  a  given  resolution,  as would
most likely be the case in FT-IR monitoring
applications, the  S/N  increases  with  the
square root of the number of  scans being
averaged.

       The relationship  between  the S/N
and resolution  is not as straightforward as
implied in  Equation  7-1.   If  the physical
parameters  of the spectrometric system,
                                        7-4

-------
      .rfsissiii*^
                                                                       TR-4423-99-03
 such  as  the  measurement time,  optical
 throughput,   and   the   interferometer
 efficiency, are assumed to be constant for
 measurements made  at  both high and low
 resolution, the S/N will be halved on doubling
 the    maximum   retardation    of  the
 interferometer (Amax) or halving the resolution
 (Av/2). Because the S/N is proportional to the
 square root of the measurement time, the
 measurement time required to maintain the
 original  baseline   noise  level   must  be
 increased by a factor  of 4 each time Amax is
 doubled, or Av is halved, for measurements
 made at a constant optical throughput.

       The  optical throughput  does  not
 necessarily remain constant throughout the
 range  of resolutions that could be used to
 measure atmospheric gases. In low-resolution
 measurements, a large optical throughput is
 allowed  for  the  interferometer,  and  the
throughput is  limited by the area  of the
detector element or the detector  foreoptics.
 Most  commercial  low-resolution  FT-IR
spectrometers  operate   with a  constant
throughput for all resolution settings.

       Instruments capable of high-resolution
measurements  are equipped with adjustable
or interchangeable aperture (Jacquinot) stops
installed in the source optics that reduce the
solid angle of the beam passing through the
interferometer.  Spectra collected at high
resolutions are generally measured with  a
variable throughput, which decreases as the
spectral resolution increases.

       In high-resolution measurements made
under  variable  throughput  conditions,  the
 throughput is  halved  as Amax is doubled.
 This results in an additional decrease in the
 S/N by one-half, which requires increasing
 the number of co-averaged scans by another
 factor of 4 to obtain the original S/N.  Thus,
 for  high-resolution FT-IR  spectrometers
 operating   under   variable   throughput
 conditions, the  total measurement time  is
 increased  by a  factor of 16 when  Amax  is
.. doubled, if the S/N ratio is to stay the same.

      ' The above discussions  apply only to
 the effect of resolution on the baseline noise
 level. Resolution may also affect the  peak
 absorbance of the  bands being  measured.
 For a narrow spectral feature whose full
 width at half height (FWHH) is  much less
 than the instrumental  resolution, the  peak
 absorbance will only approximately double
 on doubling Amax. Assuming this band was
 measured  under  constant-throughput
 conditions, its S/N  would be the same for
 measurements taken at the higher and lower
 resolution   settings,   provided   the
 measurement times are  equal.  However,
 the degree of overlap by nearby spectral
 features  will   be   reduced   when   the
 measurement is taken at a higher resolution.
 Therefore,  in this case  the higher resolution
 measurement is preferred.

       For  weak,  broad  spectral features
 whose peak absorbance does not change as
 a function of resolution, the lower resolution
 measurement is preferable when the optical
throughput   is  constant.  For  strongly
 absorbing bands, whether they are broad or
 narrow,  calibration curves of absorbance
values measured at different resolutions and
                                        7-5

-------
            ..  .
       : •.>•=••••• -.••= =
                                                                       TR-4423-99-03
 plotted   versus   concentration  must  be
 developed   to   ascertain   the  optimum
 resolution to be used.

       In general, the minimum  LOD should
 be found for measurements made at the
 lowest possible  resolution that adequately
• resolves the spectral features of the analyte
 from those of interfering species. Increasing
 the resolution beyond this point degrades the
 S/N.   In  FT-IR   monitoring,  the  optimum
 resolution will be determined by the  band
 widths of the absorption lines in the spectra
 of the target compounds, the presence of
 interfering  species, and  the  S/N of  the
 system.  The  optimum  resolution will most
 likely vary with respect to specific analytes
 and measurement conditions.

 7.4   Example  Spectra of CO2 and Water
       Vapor

       Water vapor  and  CO2 have  IR
 absorption bands with theoretical bandwidths
as narrow as 0.1 cm"1, according to the USF
HITRAN-PC database (University  of South
Florida 1993).  To fully characterize the IR
spectra of these compounds, which have
absorption bands  that may  overlap with
those  of  target  compounds,  an  FT-IR
spectrometer  capable of  high-resolution
measurements must be employed.

       A   series  of  experiments   was
conducted   on  a   benchtop  FT-IR
spectrometer in  the laboratory and with a
transportable FT-IR monitor in the field to
illustrate the effects of resolution on the IR
spectra of C02  and  water vapor.  Three
separate  series   of  experiments   were
performed.  In the first set of experiments,
single-beam sample and background spectra
were   collected  at  various  instrumental
resolution  settings  with  the  benchtop
spectrometer purged  with nitrogen.  These
spectra were used to generate the data
given below in Section 7.4.1.1 (Table 7-1).
                           Table 7-1.  Resolution Test Data

Resolution
(cm'1)
0.25
0.50
1.0
2.0
4.0
8.0
16.0
Fourier
Transform
Points
131072
65536
32768
16384
8192
4096
2048

File Size
(bytes)
300268
168268
100267
67803
35034
18650
10266

Scan Time
(s)*
194
112
69
49
28
18
13
Process
Time
(s)
249
119
60
32
16
10
8
RMS Noise
2150-2100 cm'1
(10-3 Abs)**
1.5115
0.9007
0.4504
0.2347
0.1102
0.0590
0.0225
   Time to collect 100 scans.
   * RMS noise for 1-min measurement times.
                                         7-6

-------
                                                                       TR-4423-99-03
 In  the  second   set  of  experiments,   a
 background interferogram was collected  at
 0.125 cm"1 with the spectrometer purged,
 and a sample interferogram  was  collected
 with the sample  compartment open to the
 laboratory  air.  These interferograms were
 used to generate lower resolution spectra by
 reducing the number of data points used for
 the Fourier transform.  These spectra are
 shown in Figures 7-1 and 7-2. In  the third
 set of experiments, single-beam spectra were
 collected along an open  path of 1 50 m  at
 0.5-,  1.0-,  and 2.0-cm'1  resolution with  a
 transportable FT-IR monitor.  These spectra
 are shown  in Figure 7-3.
               Wive Number (cur')
Figure 7-2. Single-Beam IR Spectra of Water
Vapor  Measured   at  (A)   0.25-cm1,
(B) 0.50-cm"1, (C) 1.0-cm'1, and (D) 2.0-cm'1
Resolution with No Apodization and  No
Additional Zero Filling.
               Wov« Number (err1)

Figure  7-1. Single-Beam IR Spectra of C02
Measured at  (A) 0.25-cm'1, (B) 0.50-cm1,
(C)   1.0-cm"1, and (D) 2.0-cm"1 Resolution
with No Apodization and No Additional Zero
Filling.
       All laboratory spectra were collected
on a benchtop FT-IR spectrometer, which has
a nominal instrumental resolution selectable
to 0.125 cm'1.  The data system uses two
68000   data   processors   and   contains
2 megabytes of RAM.  The long-path spectra
                                            of water vapor were obtained on a portable
                                            FT-IR   spectrometer   with  resolutions
                                            selectable  to  a nominal  0.5 cm"1.   Data
                                            acquisition and manipulations were carried
                                            out by using a commercial software package
                                            on  a  486/33-MHz personal computer with
                                            8 megabytes of RAM.
           1020          1040
               Wave Number (cm-')
                                   1060
Figure 7-3. Single-Beam IR Spectra of Water
Vapor Measured at  (A) 2-cm"1, (B)  1-cm'1,
and (C)  0.5-cm'1 Resolution over a 150-m
Path.
                                        7-7

-------
                                                                       TR-4423-99-03
 7.4.1  Resolution Effects

       The effects of resolution  on the  IR
 spectra of C02 and  water vapor obtained
 from laboratory measurements and long-path
 measurements are addressed in this section.
 The laboratory measurements illustrate the
 relationship between  resolution  and  scan
 time, data processing  time, data  storage
 requirements,  spectral definition,  and.S/N.
 The long-path measurements are used  to
 characterize the water vapor spectrum in a
 region in which it interferes with the analysis
 of an example target compound, toluene.

 7.4.1.1 Laboratory Measurements

       Single-beam spectra of  the p- and
 r-branch of CO2 were recorded at resolutions
 ranging from 0.25 to  2 cm"1 as  shown  in
 Figure 7-1. No apodization or additional zero
 filling was applied to the interferograms prior
 to performing  the Fourier transform.  When
 plotted   on  the  scale  from   2400   to
 2280 cm"1, the spectral features of the C02
 bands appear  to be defined equally well  at
 0.25- and 0.5-cm"1 resolution, although slight
 differences were  observed  in  the FWHH
 measurements. In the spectrum measured  at
 1-cm"1 resolution, there is a more noticeable
degradation of the rotational  fine  structure.
This structure is  completely lost in the
spectrum measured at 2-cm'1 resolution, and
the r-branch appears as a broad continuum.
The  absorption  bands  that  make  up the
rotational   fine  structure  of  CO2  have
bandwidths  of  approximately  0.2  cm"1,
according  to the USF HITRAN-PC database
(University of  South Florida 1993).  Thus,
 these  bands are not fully resolved, even at
 0.25-cm'1  resolution,  and a  resolution  of
 0.125 cm"1 is required to fully characterize
 these bands.

       Similar  results  were  observed  in
 spectra  of  water   vapor  measured  at
 resolutions of 0.25, 0.5, 1, and 2 cm"1 with
 no apodization  or  additional zero  filling
 (Figure 7-2). The  single-beam spectrum  of
 water vapor between 3720 and 3620 cm"1
 exhibits several  isolated, sharp features  as
 well as overlapping features that are nearly
 baseline resolved in the spectrum measured
 at  0.25-cm"1 resolution.   The  spectrum
 measured at 0.5-cm"1 resolution exhibits a
 slight degradation  in  spectral definition  as
 compared   to  the  0.25-cm'1  spectrum,
 although the general characteristics of the
 bands  are   retained.   The  overall  band
 structure  is  still present in  the  spectrum
 measured at 1-cm"1 resolution; however, the
 distinction  between  some of  the  closely
 spaced,  weaker  bands is  lost.   In the
 spectrum  measured  at  2-cm"1  resolution,
there  is  no longer  any  evidence  of the
 spectral definition  exhibited in the previous
spectra,  and the  -bands are significantly
 broader.

      The effect of resolution on scan time,
data   processing   time,   data   storage
requirements, and  baseline noise levels was
also determined. For these tests, single-beam
sample   and  background   spectra  were
measured independently  at each resolution
setting, while holding  the Jacquinot  stop
constant.   The  number of scans for each
spectrum was 100.  The RMS noise levels
                                        7-8

-------
                                                                       TR-4423-99-03
 were calculated from spectra with a 1-min
 measurement time.  The results from  this
 test are presented in Table 7-1.

       As shown  in  Table  7-1, the  time
 required  to  collect  100  scans increases
 significantly  upon  acquiring  data at higher
 resolutions.   This  is  because  the moving
 mirror in the interferometer must  travel  a
 greater  distance as  Amax  is increased for
 higher resolution scans.  On average,  the
 total  scan  time for  100  scans  for  this
 particular instrument increased by a factor of
 1.6 each time Amax was doubled, indicating
 the poor duty cycle efficiency.

       The time required  to process  the
 interferogram also increases as Amax (and as
 the number of data points collected for. the
 interferogram) increases. In these examples,
 the processing  time  is that  required to
 perform  the  Fourier transform  and related
 operations on both  the background   and
 sample interferograms and to calculate  the
 absorption spectrum.  The  time  required to
 process  the  interferogram  increased  on
 average   1.8  times  each  time  Amax  was
 doubled.  It should be noted that these data
 were  processed  on  a relatively old data
 system using a 68000 processor chip. With
 newer  and  faster   computers   the  time
 required  to perform the Fourier transform is
 not as much  of a factor. For example, on a
 486/33MHz machine with  8 megabytes of
 RAM,  the time  required to perform   the
 Fourier  transform  and  plot  a single-beam
 spectrum is 1.65, 3.23, and  7.74 s for 2-,
 1-,  and  0.5-cm"1-resolution interferograms,
respectively. However, if time resolution is an
 important  parameter  in  a specific  FT-IR
 monitoring application, then the combination
 of scan time and processing time should be
 considered, or the interferograms should be
 stored for post-run processing.

       The amount of disk space required for
 data  storage increases almost by a factor
 of 2 each time the resolution is increased by
 a factor  of 2.  For 3.5-in floppy disks with
 1.4  megabytes of storage capacity,  this
 means that, for example, 14 interferograms
 collected at 1-cm"1 resolution could be stored
 on  one disk, whereas only eight  0.5-cm"1
 interferograms could be stored on one disk at
 a time. The newer data acquisition software
 packages  and  data  stations   make  more
 efficient  use of disk space.  For example,
 21   interferograms  collected   at   1-cm"1
 resolution on a newer system could be stored
 on one 3.5-in. floppy disk.  If large amounts
 of data are expected to be collected, such as
 might be  the case in routine FT-IR monitoring
 studies, data storage requirements could be
 an important consideration.

      The  RMS  noise measured between
 2200 and  2100  cm'1  increases  as Amax
 increases. The data in Table 7-1 were taken
from  absorption  spectra  created   from
background  and  sample spectra collected
over a 1-min scan time  with  a constant
aperture  at each  resolution setting.  These
data follow closely the  twofold increases in
baseline  noise  expected  each time  the
resolution is increased by a factor of 2.  It
should be noted that only the baseline noise
level  was  measured  in this  experiment.
Resolution  may  also  affect   the   peak
                                        7-9

-------
iwu\/m
                                                                     TR-4423-99-03
 absorbance of  the  bands being  measured.
 For example, for a weak spectral feature
 whose  FWHH  .is  much  less  than  the
 instrumental resolution, the peak absorbance
 will approximately double on increasing the
 resolution by a factor of  2.  Therefore, the
 S/N would be the same for measurements
 taken  at  the  higher and  lower  resolution
 settings, provided the  measurement  times
 are equal.   For broad  bands,  the  peak
 absorbance will not  be  affected by changes
 in resolution,  and  the  lower  resolution
 measurement would be preferred.

 7.4.1.2 Long-Path Measurements

       The   laboratory    measurements
 described above illustrate some of the trade-
 offs  encountered between resolution and
 other  experimental parameters.  However,
 the path length used in those  studies was
 insufficient to  detect many of the water
 vapor bands  that interfere' with bands  of
 target   pollutants,   such   as   toluene.
 Russwurm   (1992)   has  addressed  the
 limitations that  the presence of overlapping
 water vapor bands impose on the ability to
 detect and quantify toluene  by  long-path
 FT-IR spectrometry.  With a CIS analysis of
the toluene band at 1031 cm"1, the detection
 limit for toluene in the presence of 10.5 torr
of water   vapor was  estimated  to  be
approximately 1 ppm. In this spectral region
over a path length of 420 m, the absorbance
due to water vapor was found  to be strong
compared  to that of toluene.  These data
clearly indicate that to optimize the detection
limits  of FT-IR monitors for difficult target
                                         compounds,  such as  toluene, the water
                                         vapor spectrum must be well characterized.

                                                Single-beam spectra measured at 2-,
                                         1-, and 0.5-cm"1 over a 1 50-m path length
                                         are shown in Figure 7-3. The interferograms
                                         were processed with triangular apodization
                                         and no additional zero filling.  The spectra are
                                         plotted over the wave number region used to
                                         quantify toluene.  As expected, the FWHHs
                                         of the water vapor absorption bands between
                                         1000 and  1060  cm"1  are narrower in  the
                                         0.5-cm"1 resolution spectrum as compared to
                                         the  1- and 2-crrf1 resolution spectra.   In
                                         addition, spectral features are resolved in the
                                         0.5-cm"1 spectrum that appear as a single
                                         band.in the other two spectra. For example,
                                         the  band  at  1010  cm"1  in  the  1-cm"1
                                         spectrum is resolved into a doublet at 1010
                                         and  1010.7 cm"1 in the 0.5-cm"1 spectrum.
                                         Also, bands appearing at 1028.3 and 1029.5
                                         cm'1 are much better resolved in the 0.5-cm'1
                                         spectrum.  In fact, they are  not resolved at
                                         all in the 2-cm"1 spectrum.

                                               To  completely  resolve all  of  the
                                         overlapping bands in the spectrum of water
                                         vapor over this wave  number region, the
                                         spectrum must be recorded  at 0.125-cm'1
                                         resolution.   The  theoretical  spectrum  of
                                         water  vapor   from  the  USF  HITRAN-PC
                                         database is shown in Figure 7-4A.

                                               A 0.125-cm'1 resolution spectrum  of
                                         water vapor recorded on the ROSE system
                                         (Herget  1992) is shown in Figure 7-4B.  In
                                         these spectra  the band at  1018 cm"1 can be
                                         resolved  into  two  components,  and  the
                                         bands  at   1010  and  1010.7  cm"1  are
                                      7-10

-------
                                                                        TR-4423-99-03
 |
     A.
   1006 1007 1008 1009 1010 1011 1012 1013 1014 1015 1016 1017 10161019 1020
                W«vo Number (cm-1)
Figure 7-4. IR Spectra of Water.  (A) 10 torr
of water vapor over a 300-m path, from the
USF  HITRAN-PC  database (University of
South Florida  1993).   (B)  Water  vapor
recorded  at 0.125-cm"1 resolution of the
ROSE system (reproduced with permission
from  W.F. Herget).
completely baseline resolved.  Whether or
not measuring ambient spectra at 0.1 25-cm"1
resolution    improves   the   results   of
quantitative  analyses for  difficult  target
compounds,  such as toluene,  has not yet
been fully investigated.

7.4.2  Zero-Filling Effects

       When  the  interferogram  contains
frequencies that do not coincide  with the
frequency  sample  points,  the  spectrum
resembles  a  "picket  fence"  (Herres  and
Gronholz 1 984). An example of this effect is
shown in Figure 7-5 in the spectrum of C02
measured on  a  benchtop FT-IR instrument.
In  this example,  the  spectrum  of   CO2
measured at 0.25-cm"1  resolution  with no
apodization  and  a  zero-filling  factor  of  1
(Figure 7-5A) exhibits excellent peak shape.
However,  in  the  spectrum measured at
0.5  cm"1  with  no  additional   zero  filling
(Figure 7-5B), the peaks of several absorption
bands  are squared  off.  This effect can be
overcome by adding zeros to the end of the
interferogram before the Fourier transform is
performed.  This operation is referred to as
zero filling.  Zero filling  increases the number
of points per wave number in the spectrum,
and, in effect,  interpolates the spectrum.
Normally, some multiple (e.g., 2, 4, etc.) of
the original number of data points is added to
the  interferogram.   This  improves   the
photometric  accuracy of the FT-IR spectrum
 Figure 7-5. Absorption Spectra of CO2
 Measured at (A) 0.25 cm"1 with a Zero-
 Filling Factor of 1, (B) 0.5 cm"1 with No
 Zero Filling, and (C) 0.5 cm"1 with a Zero-
 Filling Factor of 2.
                                        7-11

-------

                                                                         TR-4423-99-03
 and  increases  the digital  resolution.   As
 shown  in  Figure  7-5C,  zero  filling  the
 interferogram  measured  at 0.5 cm"1 by an
 additional factor of 2 eliminated the picket
 fence effect.

        It should be  noted  that  zero  filling
 improves only the digital  resolution, and not
 the resolution  of  the FT-IR spectrum.  An
 example of this is  illustrated in Figure 7-6 for
 spectra of water  vapor measured at 0.25-,
 0.5-, and 1-cm'1 resolution. The spectrum
 measured at 0.25-cm"1 resolution was zero
 filled by a factor of 1, the 0.5-cm"1 spectrum
 was  zero filled by a  factor of 2, and .the
 1-cm"1  spectrum by  a factor  of 4.  In this
 case, each of the interferograms contained
 the same number  of data points after being
 zero filled. Even with additional zero filling,
 the 0.5-cm"1 spectrum does not  match the
 spectral definition of  the spectrum obtained
 at 0.25-cm"1 resolution. (Compare Spectra A
                Wave Number (cnr<)

Figure 7-6.  Absorption Spectra of Water
Vapor Measured at (A) 0.25-cm 1 Resolution
with a Zero-Filling Factor of 1, (B) 0.5-cm°
Resolution with a  Zero-Filling Factor of 2,
and (C) 1-cm'1 Resolution with a Zero-Filling
Factor of 4.
 and  B in  Figure 7-6.)  The loss  of  spectral
 features is more dramatic  in the zero-filled
 1-cm"1 spectrum.  For example, shoulders at
 3905 and 3884 cm"1 that are detectable in
 the 0.25-  and 0.5-cm"1  spectra were  not
 observed in the 1-cm'1 spectrum.  Also, side
 lobes appear in the 1 -cm"1 spectrum that was
 zero  filled  by a factor  of  4 (Figure  7-6,
 Spectrum  C).   These  side lobes are  also
 present,  but  are  not as  severe,  in  the
 0.5-cm'1 spectrum zero filled by a factor of 2
 (Spectrum B in Figure 7-6).

       The  picket  fence  effect  is  less
 extreme if the spectral components are broad
 enough to be spread over several sampling
 positions.  As  a rule of thumb, the original
 interferogram size should  be doubled by zero
 filling by an additional factor of 2. When a
 CIS  analysis   of   the   spectral  data   is
 performed,  in general it has been found that
 one order of zero filling (which is 2 times the
 original number of  data points used in the
 Fourier transform) yields a factor of 2 lower
 error than that with no additional zero filling.
 An example of  this  is  given in Section
 7.5.2.2.1.  It should be noted that zero filling
 does  increase  the  file size and  the time
 required for data processing.

 7.4.3  Apodization Effects

       As shown in  Chapter 2, the Fourier
transform integral has infinite limits  for the
optical path difference.  Thus, to measure
the true  spectrum   of  the  source,  the
interferometer must  scan  infinite  distances.
However, because the mirror can  move only
a finite distance, the  exact reconstruction of
                                        7-12

-------
 "rCt*U."s':s:~^i r-=
 TcCH,!i=J3.=i.*S==5
                                                                       TR-4423-99-03
the  spectrum  is  impossible.   The finite
movement  of   the   interferometer  mirror
truncates,  or cuts-off,  the  interferogram.
This, in effect, multiplies the  interferogram
by  a  boxcar  truncation function.    This
function may cause the appearance  of side
lobes on both   sides  of  the  absorption
band.   The   corrective   procedure   for
eliminating  these  side  lobes   is  called
apodization. (For an in-depth  discussion of
the effects  of apodization, see Chapter 8).
Apodization is performed by multiplying the
measured interferogram by a  mathematical
function.   Typical  apodization   functions
include triangular, Happ-Genzel, and Norton-
Beer functions.  An example of the effect of
these apodization functions  on the FT-IR
spectrum of CO is shown in Figure 7-7.

      Apodization  affects  the   effective
spectral  resolution,  the  apparent  peak
absorbance,  and the  noise  of  any. FT-IR
spectrum.   The apparent absorbance  of
narrow  bands will be  most affected  by the
               n«      JIM
               Wave Number (cnr1)
Figure 7-7. Absorption Spectra of CO
Measured at a Nominal 0.125-cm 1
Resolution with (A) No, (B) Triangular,
(C) Happ-Genzel, and  (D) Norton-Beer-
Medium Apodization Functions.
choice of apodization function.  In general,
the bands in a spectrum computed with no
apodization will be more intense than bands
in  the   spectrum  of  the  same  sample
computed from the same interferogram after
applying  an apodization function.

      .Apodization also degrades resolution.
An example  of  this  is  illustrated  by the
spectra of water vapor in Figure 7-8.  In this
case, subtle differences are observed,  for
example,  at  3948  and  3947  cm'1  and
3924.4 cm"1, in the spectrum generated with
no apodization (Figure 7-8A) and the spectra
generated  by using  the three  types  of
apodization functions.

      In general, to obtain the optimum S/N
for  spectra   of  small  molecules  with
resolvable fine  structure, the use  of  no
apodization is  preferable if side lobes from
neighboring intense lines do  not present an
interference.  If  side lobes are present and
                 Wive Number (cnr>)
 Figure 7-8. Absorption Spectra of Water
 Vapor Measured at 0.5-cm'1 Resolution
 with an Additional Zero-Filling Factor of 2
 and with (A) No, (B) Triangular, (C) Happ-
 Genzel, and (D) Norton-Beer-Medium
 Apodization Functions.
                                        7-13

-------
                                                                       TR-4423-99-03
 interfere   with   either  qualitative   or
 quantitative analyses, apodization becomes
 necessary.  However,  when  classical least
 squares is performed, the field spectra must
 be processed by using the same apodization
 function that was used  for the reference
 spectra.  For broad absorption bands,  the
 measured absorbance is about the  same in
 apodized  and unapodized spectra. This  can
 pose some problem  in  the  analysis using
 classical least squares.  Gases that exhibit
 broadband features  in  the  P  and the R
 branches, but also have a sharp Q branch like
 that  of  the gas toluene,  will exhibit a
 different   response    in  these  branches
 because of the apodization function.  Overall,
 the  greatest  noise  suppression  will  be
 obtained  with  the  strongest   apodization
 function, but the spectral resolution and band
 intensities will  be  greatest  for  weaker
 apodization   functions   (Griffiths   and
 de Haseth 1986).  The optimum  apodization
 function has yet to be determined for general
 use in long-path FT-IR monitoring. In  general,
 there is a reciprocal  relation between  the
 suppression  of  the  side  lobes and   the
 broadening of the absorption feature for any
 of the apodization functions.  That is, as  the
 side  lobes became smaller relative to  the
 peak, the feature becomes broader compared
to the unapodized feature.

      Triangular    and    Happ-Genzel
apodization functions are commonly used in
OP/FT-IR monitoring, although Griffiths et al.
 (1995) have indicated that  a Norton-Beer
medium function  actually gives a better
representation of the true absorbance. In all
cases, however, the same parameters should
 be used to collect the field spectra that were
 used to record the reference spectra. The
 choice of apodization function may be limited
 by this  requirement.  If spectra  from  a
 commercial or user-generated library  are to
 be the reference spectra for  quantitative
 analysis, then the parameters that were used
 to generate those reference spectra should
 be  used   to collect  the  field   spectra.
 Otherwise,   errors  in  the  concentration
 measurement will occur.

 7.5    Effect of Resolution on  Quantitative
       Analyses

       The   determination    of   analyte
 concentrations   by   FT-IR   spectrometry
 depends on the linear relationship between IR
 absorbance and  concentration as described
 by Beer's law. The discussion in Chapter 8
 shows that this seems never to  be true if an
 apodization  function is used, particularly at
 low resolution.   If the FWHH of the band is
 narrower than the instrumental resolution,
the  measured  spectrum   is   actually  a
convolution of the instrument line shape and
true band shape.  As a result, the measured
absorbance will be only approximately linear
with concentration. The higher the resolution
for the spectral  features of the  IR  band
chosen  for  quantification, the  better the
approximation. The apodization function also
has  an  effect on linearity.   This  section
describes  studies  from  the  literature that
have addressed the effects of resolution and
related parameters on quantitative analyses.
                                        7-14

-------

                                                                       TR-4423-99-03
 7.5.1  Studies from the Literature

       Strang et ai.  (1989) designed and
 evaluated  an FT-IR system for monitoring
 toxic   emissions   from   semiconductor
 manufacturing processes. This system was
 used to analyze  part-per-billion levels  of
 organic vapors and metal hydrides  such  as
 arsine, phosphine, and diborane in simulated
 workplace  environments. The optimal wave
 number, region  for quantification  and the
 effects of resolution and spectral overlap  on
 the  accuracy of quantitative  results were
 studied. Spectral measurements were taken
 at resolutions ranging from 0.5  to 8 cm"1  to
 determine  the optimum balance between
 (1) analysis  time, (2) data  storage  space,
 (3) S/N, (4) accuracy of quantitative analyses
 using a CIS  program, and (5) the ability  to
 differentiate  compounds with  overlapping
 spectra. It should be noted  that these data
 were acquired over a 20.25-m path in a
 multipass cell.  Therefore, findings from this
 study may not be applicable to long,  open-
 path measurements. Discussion of this study
 is included in this text because of the  interest
 in   workplace  monitoring.     Also,  the
 methodology used to determine the optimum
 resolution for the short-path measurements
 can   also  be  applied  to  longer  path
 measurements.

       The authors specified the following
four  issues that must  be resolved for a CLS
analysis  at   a  given  resolution  to  be
acceptable.
    1.  Whether the CLS  result  varies  by
       more than 50%  of the theoretical
       value
    2.  Whether  false   positives   or  false
       negatives develop  as  a  result  of
       degraded resolution
    3.  Whether the  amount of error in  the
       measurement will  cause potentially
       toxic  concentrations  of the target
       analyte   in  air  to   be  measured
       incorrectly
    4.  Whether the detection limit obtained
       with the CLS program changes as a
       function of resolution

Using these criteria, the authors determined
the  minimum allowable  resolution for  the
target   compounds.  The   results   are
summarized below in Table 7-2.

       In this study, the higher resolutions
required  for the  metal  hydrides arsine,
diborane,  and  phosphine  were a  result  of
spectral overlap with other target analytes
and with interfering atmospheric compounds.
For example, phosphine overlaps  with C02
and  arsine  overlaps with water vapor.  The
authors  also determined  the effects   of
decreased resolution on the accuracy of the
quantitative results. In the case of diborane,
only the 0.5-cm'1 resolution measurements
exhibited   a  linear   relationship   for   all
concentrations.  Measurements taken at  2-
and 4-cm"1  resolution deviated from linearity
as the concentration decreased.

      The effect of resolution on the ability
to quantify  overlapping compounds by the
CLS analysis was investigated  by  using
mixtures of Freons 11, 1 3B1, and 22. These
                                        7-15

-------

                                                                      TR-4423-99-03
           Table 7-2.  Optimal Wave Number Region and Minimum Resolution1
Compound
Acetone
Arsine
Diborane
o-Dichlorobenzene
2-Ethoxyethanol
Freon 1 1
Freon 13B1
Freon 22
Nitrogen trifluoride
Phosphine
Sulfur hexafluoride
Wave Number
Region
(cm'1)
1287-1167
2132-2106
2522-2515
1060-1002
1192-984
876-813
1137-1031
1193-1063
960-833
2440-2390
965-915
Minimum
Resolution
(cm'1)
8
2
0.5
8
8
8
8
8
8
4
8
 'From Strang et a/. (1989).  These data are applicable to a 20.25-m path.
compounds  have relatively  broad spectral
features  that  overlap.  The mixtures were
analyzed  at  0.5-,  2-,   4-,  and   8-cnrv1
resolution. Each of the individual compounds
could  be quantified  accurately at  each
resolution   in   a    1:1:1   mixture  at
concentrations of 10,  1, and 0.1  ppm.

       Strang and Levine  (1989)  have also
determined the LODs for  the same  target
compounds  in  the previous study   as  a
function of resolution.  For most compounds,
there was very little difference in the LODs
estimated at resolutions of  0.5, 2,  4, and
8  cm'1.     However,  for  diborane  and
phosphine   the   LOD  was  difficult  or
impossible to measure at 8 cm"1  resolution.
In the case of diborane, the peak selected for
quantification had a FWHH of 7 cm"1.  At 8-
cm'1  resolution there are only two data  points
every 8 cm"1, so a peak 7 cm'1 in width is not
defined well enough to be quantified by the
CLS program.   For phosphine, the  peak
shape  is  severely  degraded  at  8   cm'1.
Although the CLS program could quantify the
peak,  the  LOD  was  significantly higher
(0.7 ppm-v/v) for spectra measured at 8 cm'1
resolution as compared to those measured at
0.5 cm'1 resolution (0.07 ppm-v/v).   This
example   also   illustrates  one   of   the
advantages of the CLS program over  single
peak   absorbance   measurements  in
quantitative  analysis.   The  single   peak
absorbance measurement is difficult to make
for  broad bands, whereas the CLS  program
uses multiple data  points over the  entire
spectral range of the broad band.

       Spellicy  et  al.   (1991)  addressed
several  issues  regarding  spectroscopic
remote sensing with respect to the Clean Air
Act. One issue that was addressed  was the
                                       7-16

-------
                                                                      TR-4423-99-03
 optimum resolution for remote sensing FT-IR
 applications.     The   authors   presented
 theoretical   calculations   describing  the
 relationship   between  absorbance   and
 concentration for a single Lorentzian line with
 a  half-width  of  0.1   cm"1  measured   at
 resolutions from  0.01  to 0.1 cm"1.  Under
 these conditions, linearity was observed only
 at a highest resolution case and at the lowest
 concentrations.  The deviation from linearity
 most likely would  be observed  in  small
 molecules such as HCI, CO, C02, and H20,
 which have sharp  spectral features.  For
 larger  compounds,   such   as   heavy
 hydrocarbons that exhibit broader IR bands,
 the linear relationship  between  absorbance
 and  concentration   is  more likely to  be
 followed.

      More recently, Marshall et al. (1994)
 conducted a laboratory study to determine
the   effect   of   resolution   on   the
 multicomponent analysis of VOCs with a CLS
 program.  When they analyzed for  target
VOCs, such as acetone, chloroform, toluene,
methanol,  1,1,1 -trichloroethane, methyl ethyl
 ketone,  carbon tetrachloride, and the xylene
isomers, over a short path, resolutions lower
than 4  cm"1 had  an  adverse effect on the
multicomponent analysis. Resolutions of 1 to
2 cm"1 were found to be adequate for these
target compounds when the CLS  program
was used.

      Griffiths  et al.  (1993) reported the
advantages and disadvantages of using low-
resolution  measurements in long-path FT-IR
monitoring.  Among the advantages  cited
were the.smaller size and greater portability
 of the instrument, an improved S/N, and a
 lower cost.  The disadvantages included a
 greater  difficulty in visualizing the IR bands
 of  the  target compounds  and  potential
 deviations from Beer's law. A test  case of
 measuring the xylene isomers at resolutions
 of 2, 4, 8,  and 16 cm"1 was presented.  By
 using a  partial least squares (PLS) program,
 good quantitative results were  obtained at
 the relatively low resolution measurements.
 These results also  indicated  that the  PLS
 program might  be better than  the  CLS
 program for distinguishing  and  quantifying
 target compounds with overlapping features.

      Bittner et al. (1 994) have reported on
 high-resolution FT-IR measurements of VOCs
 at a variety of monitoring sites. By recording
 spectra  at 0.125-cm"1 resolution, detection
 limits for  benzene of  0.5  ppm-m were
 achieved at path lengths between  60  and
 100 m  at  a  fuel storage  area.   The high-
 resolution measurements allowed the narrow
 benzene band  at 674  cm"1  to be separated
 from the strong C02 absorption bands in that
 spectral region.

 7.5.2 Case Study: The Effect of Resolution
      and  Related Parameters on the CLS
      Analysis of Multicomponent Mixtures

      A study using  laboratory-generated
spectral   mixtures  that have  overlapping
features was conducted to investigate  the
effect of instrumental resolution  and  related
parameters on the CLS  analysis  results
(Childers and  Thompson 1994).  The study
was  designed to  simulate  conditions that
might   be    encountered   in   long-path
                                       7-17

-------
         t:sr::si'=i
                                                                       TR-4423-99-03
 measurements.    The  results from  three
 separate cases are discussed.

    1.  Analytes  with  narrow  bands  that
       overlap, such as  those for CO and
       13C-labeled CO
    2.  Analytes  with   broad   bands  that
       overlap, such as  those for acetone,
       methylene chloride, and ethanol
    3.  Analytes  with  narrow and   broad
       bands that overlap, such as those for
       nitrous oxide and methylene chloride.

 The effect of the number of data points and
 the noise level on the  CIS analysis is also
 illustrated. Because the mixtures analyzed in
 this study  were  created  from  a   linear
 combination of reference spectra,  the effect
 of  resolution  on  the  relationship  between
 concentration  and   absorbance  is  not
 addressed.

       The  spectra  were  collected   on  a
 research-grade, benchtop FT-IR spectrometer
 equipped with an MCT  detector.  A gas cell,
 50  mm long and 32 mm in diameter, was
 used  to  obtain  reference spectra of  CO,
 13CO, acetone, methylene chloride, ethanol,
 and nitrous  oxide at room temperature and
 atmospheric pressure. The reference spectra
 were acquired at resolution settings of 0.1 25
 and 1  cm'1. The original interferograms were
 processed by using the appropriate number
 of data points to yield spectra with nominal
 resolutions ranging from 0.25 to 8 cm"1.  No
 additional zero  filling  was  used on the
0.25-cm"1  spectra because  of   memory
limitations in the data system. The 0.5-cm'1
spectra were  zero filled by  an  additional
 factor of  2, and the 1.0-cm"1 spectra were
 zero filled by an additional factor of 4. As a
 result,  the  0.25-,  0.5-,  and  1.0-cm'1-
 resolution spectra had the same number of
 data points, that is 1 31,072. The 2,- 4-, and
 8-cnY1  spectra   were   generated   from
 interferograms that contained 16,384, 81 92,
 and 4096 data points,  respectively.   No
 additional zero filling was performed on these
 interferograms.   A  triangular  apodization
 function was applied to each interferogram
 prior   to   performing   the   fast   Fourier
 transform.

       The concentrations of the individual
 analytes in the gas cell were determined by
 comparing the maximum absorbance values
 of the 0.5-cm'1 spectra to those of reference
 spectra in a commercial spectral library.  The
 absorbance values of the reference spectra
 were   then  normalized   to  values   cor-
 responding to a concentration'of 100 ppm.
 These spectra were added mathematically to
 produce  synthetic  mixtures  with  varying
 concentrations of each analyte.  Synthetic
 noise corresponding to an amplitude of 1, 5,
 10, and 25% of the most intense peak  in
 each spectrum was added to the mixtures.
 Mixtures with 10% noise  added  were then
 analyzed by using a CLS algorithm.

 7.5.2.1   Mixtures of CO and 13CO

       Synthetic mixtures  of  CO and 13CO
 were generated by adding reference spectra
of CO  corresponding  to concentrations of
 150, 300, 450, and 600 ppm to reference
spectra of 100  ppm 13CO. The 0.25-cm'1
                                       7-18

-------
                                                                       TR-4423-99-03
 reference spectra  and the spectra of the
 synthetic mixtures corresponding to 1 50 ppm
 of CO and 100 ppm of 13CO recorded at
 0.25, 0.5,  1.0, and 2.0 cm"1  resolution are
 shown in Figure 7-9.  CO and 13CO  have
 several bands that overlap or nearly overlap
 in  the  spectral region  between  2118  and
 2137 cm'1. For example, overlapping bands
 at  2123.6 and 2124.2 cm"1 and  at 2131.0
 and 21 31.6 cm"1 are nearly baseline resolved
 in the 0.25 cm"1 spectrum. These bands can
     2120        2125       2130       2135
              Wave Number (cm'1)
 Figure 7-9. Reference 0.25-cm"1 Spectra of
 (A)  13CO  and  (B)  CO  and  Spectra  of
 Synthetic Mixtures of 150 ppm CO and
 100  ppm  13CO  Measured  at  (C)  0.25-,
 (D)  0.5-,  (E)  1.0-.  and  (F)  2.0-cm1
 Resolution.

 still  be  distinguished at  a  resolution of
 0.5 cm"1, but appear as only one band in the
 1.0-  and  2.0-cm"1 spectra.  Even  though
 these bands  are  not  resolved at 1.0-  and
 2.0-cm"1   resolution,   the   CLS  analysis
 accurately determined the concentration of
 CO in the mixture when it was analyzed for
 both CO  and 13CO. Plots of the calculated
concentration   versus   the   known
concentration  of  CO  were  linear over  the
range of 0 to 600 ppm of CO in the presence
 of 100 ppm 13CO. When the mixtures were
 analyzed for only CO, a positive bias and an
 increase in the magnitude of the errors in the
 measurements were observed. Although the
 bias  was  relatively  constant,  the error
 increased as the resolution  decreased from
 0.25 to 1.0 cm"1.  In the case of the 2-cm"1
 measurements, CO could not be detected at
 1 50 ppm in the mixture if 13CO was excluded
 from the analysis. (See Figure 7-10.)
                                               TOO
          100   MO   MO   400   MO   tOO  700
             Known Concentration (ppm)
 Figure 7-10. Concentration Calculated from
 CLS Analysis vs. Known Concentration for
 13CO/CO  Mixtures  Measured  at  2-cm"1
 Resolution.  The (•)  represents  a  value
 obtained during analysis for both 13CO and
 CO, and the (+) represents a value obtained
 during analysis for CO only.
7.5.2.2   Mixtures of Acetone, Methylene
          Chloride, and Ethanol

       Synthetic  mixtures   of   acetone,
methylene  chloride,  and  ethanol  were
generated by adding reference spectra  of
ethanol corresponding to concentrations  of
125, 250, 375, and 500 ppm to reference
spectra of  100 ppm each of acetone and
methylene chloride. The 0.25-cm"1 reference
                                       7-19

-------
MAi&m
TECHm^
                                                                       TR-4423-99-03
 spectra  and the  spectra of the  synthetic
 mixtures corresponding  to  500  ppm  of
 ethanol  and  100 ppm each of  acetone and
 methylene chloride recorded at 1.0-,  2.0-,
 and  4.0-cm"1   resolution are  shown  in
 Figure 7-11. In this case, the  spectrum of
 each analyte could be adequately measured
 at 1.0-cm"1  resolution. At 4-cm"1 resolution,
 the Q-branches of these compounds were no
 longer  detected.  However,  this   did  not
 diminish the ability to quantify ethanol in
 these mixtures with the CLS algorithm. Plots
 of the calculated  concentration versus the
 known concentration of ethanol were linear
 over the entire  range  from 0 to  500  ppm
 ethanol  in the  presence of 100  ppm  of
 acetone  and -100 ppm of methylene chloride
 for each resolution setting.  Although  the
 concentration of ethanol could be determined
 in the  low-resolution measurements,  the
 errors  in the CLS analysis increased  with
 decreasing  spectral  resolution.  The high-
 0>
 o
 c
 a
 a
             Wave Number (cm'1)
Figure  7-11. Reference  0.25-cm'1 Spectra
of (A) Acetone, (B) Methylene Chloride, and
(C)  Ethanol  and  Spectra  of Synthetic
Mixtures of  100 ppm Acetone, 100 ppm
Methylene Chloride, and 500 ppm Ethanol
Measured at   (D)  1.0-,  (E)  2.0-,  and
(F) 4.0-cm 1 Resolution.
                                           resolution spectra contain more data points
                                           per wave number than do the low-resolution
                                           measurements.   To   determine   if  this
                                           contributed to the increase in the magnitude
                                           of the errors, the effect of the number  of
                                           data  points   on  the  CLS   analysis  was
                                           investigated. The effect  of S/N on CLS was
                                           also investigated.

                                           7.5.2.2.1 Effect of the Number of Data
                                                    Points on the CLS Analysis

                                                 The average error  in the CLS analysis
                                           for  ethanol  in   the  acetone,  methylene
                                           chloride, ethanol synthetic mixtures for each
                                           resolution setting is shown in Table 7-3.

                                                 For  the 0.25-,  0.5-,  and 1.0-cm"1
                                           measurements, in which additional zero filling
                                           was used to keep the number of data points
                                           the same, the average error  was relatively
                                           constant at approximately 9%. However, for
                                          the 2.0-, 4.0-, and 8.0-cm"1 measurements,
                                           in which no additional zero filling was used,
                                          the average error increased with a decrease
                                           in the number of  data points.   (Note that the
                                          0.25-, 0.5-,  and  1.0-cm'1  spectra  were
                                          generated from  an original  interferogram
                                          collected at 0.125-cm"1 resolution, whereas
                                          the 2.0-, 4.0-, and 8.0-cm"1  spectra were
                                          generated from  an original  interferogram
                                          collected at 1.0-cm"1 resolution.)

                                                To show that the increase  in  error  is
                                          related to the number of data points  per
                                          wave number, and is not necessarily a direct
                                          result of  degrading the spectral resolution,
                                          spectral  mixtures  of  acetone, methylene
                                       7-20

-------
        rs.-=».f./s.fa =
 TFFH ^"^Va; r=
 I tUtii i=v -. -^ « r ^=
                                                                       TR-4423-99-03
           Table 7-3. Effect of The Number of Data Points on the CLS Analysis
Resolution
(cm'1)
0.25
0.5
1.0
2.0
4.0 ...
8.0
Additional Zero
Filling
None
2x
4x
None
None.
None
Data
Points
131072
131072
131072
16384
8192
4096
Average
Error
9.13
8.93
9.55
24.54
32.73
42.27
chloride, and  ethanol obtained at  1-cm"1
resolution  were  processed  by using  no
additional zero filling, an additional zero filling
factor  of 2,  and an additional  zero filling
factor  of 4. This resulted in interferograms
having 32,768, 65,536,  and  131,072 data
points, respectively. These results were also
compared  to  those  obtained  for spectra
measured at 2 cm'1, which were generated
from interferograms containing 16,384 data
points.
      As  can be  seen in Table 7-4, the
accuracy  of  the measurements  was not
affected by the  number of data points per
spectral element. However, the magnitude
of the error in the measurements was related
to the number of interferogram data  points
used to  generate the spectra.  On average,
the error in the CLS analysis decreased by a
factor of 1.4 each time the number of data
points used to process the interferogram was
doubled.
               Table 7-4.  The Effect of Zero Filling on the CLS Analysis

Cone.
(ppm)
0
125
250
375
500
1-cm'1
Resolution
4 x Zero Fill
Below MDL
121.40 (8.49)
248.33 (9.35)
379.85 (9.58)
500.81 (10.29)
1-cm-1
Resolution
2 x Zero Fill
Below MDL
119.65 (11.89)
248.87 (12.87)
368.40 (13.83)
509.94 (14.73)
1-cm-1
Resolution
No Zero Fill
Below MDL
124.53 (17.24)
252.07 (18.15)
380.78 (17.94)
492.42 (20.83)
2-cm-1
Resolution
No Zero Fill
Below MDL
119.85 (22.47)
252.18 (22.53)
373.87 (24.27)
509.24 (28.90)
                                       7-21

-------
      .
      • -L-J -,-j. • /
                                                                       TR-4423-99-03
 7.5.2.2.2  Effect  of  S/N
           Analysis
on  the  CLS
                Table 7-5. Effect of Noise on the CLS
                Analysis
       Synthetic noise was added to 1-cm'1
 spectral  mixtures   containing  100  ppm
 acetone, 100 ppm methylene chloride, and 0
 to 500 ppm ethanol  at levels corresponding
 to  1,  5,  10,  and 25%.of the maximum
 absorbance value in each spectrum.

       Spectral mixtures containing  100 ppm
 acetone, 100  ppm methylene  chloride, and
 500 ppm ethanol at each  noise  level are
 shown in Figure 7-1 2. In these  mixtures, the
 average error in the CLS analysis was found
 to be directly proportional to the percentage
 of  noise  added to  the spectrum.  (See
 Table 7-5.)
            1200         1250         1300
              Wave Number (cm'1)

Figure 7-12. Spectra of Synthetic Mixtures of
100 ppm Acetone, 100 ppm Methylene
Chloride, and 500 ppm Ethanol Measured at
1-cm"1 Resolution with (A)  0, (B) 1, (C) 5,
(D) 10, and (E) 25% Noise Added.
                                                  % Noise
                                        Average Error
0
1
5
10
25
0.07
0.96
4.78
9.55
23.89
                7.5.2.3    Mixtures of Methylene Chloride
                          and Nitrous Oxide

                      Synthetic  mixtures  of  methylene
                chloride and nitrous oxide were generated by
                adding  reference spectra of nitrous  oxide
                corresponding to concentrations of 1 2.5, 25,
                37.5,  50, 75, and  100 ppm to reference
                spectra of 100 ppm methylene chloride. The
                0.25-cm"1 reference spectra and the spectra
                of the synthetic mixtures corresponding to
                50 ppm of  nitrous oxide with both 0 and
                100 ppm of methylene chloride recorded at
                0.25-, 0.5-,  and  1.0-cm"1  resolution  are
                shown in Figure 7-13.

                     The spectrum of nitrous oxide exhibits
                sharp bands that are resolved at 0.25 cm'1,
                but  are  not  as  well resolved  at  0.5-cm'1
                resolution.  These bands become a broad
                continuum in the 1.0-cm"1 spectrum. At first
                glance, one  would expect the CLS analysis
                to perform better for the 0.25-cm"1 spectra,
                in which the N20 bands  are fully resolved.
                However, this is  not the case in these
                mixtures. When analyses are performed for
                N20 over the entire  band envelope  from
                1231 to  1329 cm'1,  N20 is not detected at
                concentrations less  than 75  ppm in the
                0.25-cm"1  spectra.    (See  Figure 7-14.)
                                       7-22

-------
  MAIWiin
  TECH
                            TR-4423-99-03
  0)
  g
  ra
  o
  (0
 <
          1250               1300
               Wave Number (cm"1)
 Figure 7-13. Reference 0.25-cm"1 Spectra of
 (A)  N2O and  (B)  Methylene Chloride and
 Spectra  of Synthetic Mixtures of 50 ppm
 N2O  and  100 ppm  Methylene  Chloride
 Measured  at  (C)  0.25-,   (D)  0.5-,  and
 (E) 1.0-cm1 Resolution.
   120
          20    40    «0    BO   100
           Known Concentration (pom)
                                     120
Figure 7-14. Concentration Calculated from
CIS Analysis vs. Known Concentration for
N2O/ Methylene Chloride Mixtures Measured
at 0.25-cm"1 Resolution. The (•) represents
a value obtained during  analysis  over  the
methylene  chloride  region,  and  the  (4)
represents a value obtained during analysis
over the N2O region.
 However, when the mixture was analyzed for
 N2O  by using the methylene chloride region
 from  1243  to   1292  cm"1,   N2O  was
 accurately quantified with a high precision.
 Similar results were obtained for the 0.5-cm'1
 resolution spectra.  In contrast, the mixtures
 recorded at 1.0-cm"1 resolution  could  be
 analyzed successfully over both regions.  At
 2-cm"1   resolution,  the   CLS   analysis
 performed  best   over  the  N20   region.
 Apparently, in multicomponent mixtures, the
 CLS  algorithm does not perform well in
 regions  where one or  more  components
 exhibit only baseline noise.

      This  effect seems to be amplified in
 higher   resolution   measurements   of
 compounds with  sharp spectral  features,
 such as N20. These results  indicate that for
 spectra with sharp features  the CLS should
 be  performed  over a  narrow range  that
 contains absorbing features.

 7.5.2.4  Conclusions and Recommendations
         Based on Case Study

      The following conclusions regarding
 resolution requirements  in long-path  FT-IR
 monitoring can be drawn from this simulated
 study using  well-characterized spectral data
 sets.

      In spectra  with  overlapping  sharp
features, the CLS algorithm  can accurately
quantify  target analytes, even when the
bands  used for   analysis  are  not  fully
resolved. However, a failure  to identify all of
the overlapping components in a mixture can
result in a bias and an increase in the error in
                                        7-23

-------
 TECHi
                                                                        TR-4423-99-03
 the  CLS analysis.  Thus, the real  value  in
 performing higher resolution measurements
 might be to facilitate identifying the species
 present to be included in the CLS  analysis
 set.

       In   the   case  of   spectra   with
 overlapping broad features,  such as those
 found in acetone, methylene chloride,  and
 ethanol, the accuracy of the  CLS analysis is
 not   affected  by the resolution  setting.
 However, the magnitude of the errors in the
 CLS analysis is related to the number of data
 points per  wave number  in  the  analyte
 spectrum.  Therefore, the errors in  the CLS
 analysis   will  increase  with  decreasing
 resolution unless  additional  zero filling, or
 some other means, is used  to increase the
 number  of  data  points in   the  spectrum.
 However,  the   use  of  zero   filling   or
 interpolation to indiscriminately increase the
 number of data points in the  spectrum is not
 recommended, because interpolated data
 points   do   not   contain   independent
 information.  In these mixtures, the  errors in
 the CLS analyses were also found to  increase
 proportionally with increases in  the  noise
 level.

      The results from the CLS  analysis of
 spectral mixtures with overlapping broad and
 sharp  bands,  as  was  the  case  with
 methylene chloride and nitrous oxide, were
 not as straightforward to interpret.  When
analyzing for nitrous oxide in spectral regions
where methylene chloride did  not exhibit any
absorption   bands,   the  CLS   algorithm
performed better at  lower resolutions.   In
regions where the two compounds exhibited
 overlapping  spectral features,  comparable
 results were  obtained  for  measurements
 taken at 0.25-, 0.5-, and 1.0-cm"1 resolution.

       In  summary,  resolution requirements
 will vary for different target compounds and
 sampling  conditions. In field measurements
-these requirements  will  depend  on  several
 factors, such as path length, concentration
 of the target compounds, and the presence
 of   interfering  species.     Although  the
 simulated  studies  described here  do not
 provide a definitive  answer regarding the
 resolution  question, similar studies  using
 target  analytes and possible  interfering
 species should  be  performed prior to field
 studies to  establish  guidelines  for  data
 acquisition and analysis.

 7.6   General Conclusions and
       Recommendations

       As stated in the introduction of this
 chapter, there  is currently no consensus as
 to what resolution is generally applicable in
 FT-IR  long-path, open-path  monitoring.  A
 spectral resolution of 0.125 cm'1  is required
 to   fully   characterize  the   spectra  of
 atmospheric C02 and water vapor.  Spectra
 taken  along a 1 50-m path show that there
 are significant differences in the water vapor
 spectra measured at nominal resolutions of
 2,  1,  and  0.5 cm'1. The effect of these
 differences   on   the  computer-assisted
 quantitative analyses for target  pollutants
 has  not, however, been  fully examined for
 long-path  FT-IR measurements.  In previous
 studies, Strang et al. (1989) have  shown
 that for several organic vapors a resolution of
                                        7-24

-------

                                                                       TR-4423-99-03
 8  cm"1  is  sufficient to obtain quantitative
 results over a short path if a CIS program is
 used. In contrast, Spellicy et al. (1 991) have
 presented  theoretical results  that  suggest
 that the FT-IR spectra  of  small molecules
 with  very  fine spectral features will obey
 Beer's law  at only high resolution (0.01 cm'1)
 and at very-low concentrations.  Recently,
 Marshall et al. (1994)  and Griffiths et al.
 (1995)  have  indicated  that  1- to  4-cnrT1
 resolution,  or possibly a lower resolution, is
 adequate for measuring certain VOCs using
 CIS  and  PLS multicomponent  analysis
 programs.

      Clearly,  there is  much fundamental
 research that must be done to "resolve" the
 resolution question.  Experiments similar to
those done by  Strang et al. (1989), Strang
 and Levine (1989), Marshall  et al.  (1993),
 and  Griffiths   et  al.  (1993) should  be
conducted  over a  long, open path  for the
hazardous air pollutants stipulated under the
Clean  Air   Act  Amendments  of   1990.
Measurements  of these compounds should
be   taken  at   different   resolutions,
concentrations,   and   path   lengths   to
determine   the  optimum   experimental
conditions  for  obtaining the  best S/N  and
detection limits.  Most likely,  the optimum
resolution  will  be different for the  various
compounds.  However, at least minimum
resolution requirements could be determined.

      Although the  question  of  what
resolution should be used in FT-IR long-path,
open-path   monitoring   has  not  been
answered,  the  reader  should   have   an
appreciation  for  the  factors related  to
 resolution that affect spectral measurements.
 Instrument  manufacturers  and  software
 vendors   have  made   great  strides   in
 simplifying the  use of FT-IR  instruments.
 Most FT-IR software  is menu driven and
 some instruments can be  operated  at the
 push  of   a   button.  Although   these
 developments facilitate  the collection  of
 FT-IR data, they  also allow  data  to  be
 collected  without  a  knowledge  of  the
 principles behind the measurement. Analysts
 working  in this field must be aware of the
 effects of different instrumental parameters
 on the measured spectrum.  Grasselli et al.
 (1982) have published criteria for presenting
 spectra  from computerized IR  instruments,
 with an emphasis on FT-IR measurements.
 The  authors  established recommendations
 and  guidelines for  reporting experimental
 conditions,  instrumental parameters,  and
 other pertinent information  describing the
 acquisition of FT-IR spectra. These guidelines
 should be  followed  when  reporting FT-IR
 data.

 7.7   Guidance for Selecting Resolution and
      Related Parameters

      In  this section,  general criteria  and
guidelines are  suggested for choosing the
optimum  resolution  for acquiring  spectral
data. The choice of resolution  and related
parameters, such as  apodization  and zero
filling, to be used for data collection will be
determined by several factors.   As  stated
before, there is no consensus as to what the
optimum   parameters   should   be.  The
parameters need to be optimized for the
specific  experiments  planned, taking into
                                       7-25

-------
 TECH'- :L=J -.-1 n ^^
                                                                        TR-4423-99-03
 consideration the goals of the monitoring
 study.  The following guidelines should be
 taken  into  account  when choosing  the
 optimum instrumental parameters.

 7.     Consider  the   bane/widths  of  the
 absorption  features  used  to  analyze for
 specific target compounds.  If the absorption
 bands of the target compounds are relatively
 broad, there may be no need to acquire high-
 resolution spectra. When this is the case, no
 additional information will be gained, and the
 measurements will have a  poorer S/N  and
 will   require   longer  data   collection,
 computational times, and larger data storage
 space.    The   analyst  must   be aware,
 however,   that  the   spectral  features  of
 atmospheric constituents such as CO2, H2O,
 and CH4 can be completely  resolved only at
 a resolution of 0.125 cm"1.  Because  these
 compounds are in every long-path spectrum
 and often overlap with the target analyte,
 access to  high-resolution  data  may  be
 required to develop  the  analysis  method.
 There   is   some  thought  that  the  real
 advantage of high-resolution spectral data is
 the ability to visualize the spectral features
 and to identify  interfering  species.   This
 information  can then be used in developing
 the analysis method.

 2.   Determine if interfering species  are
present. If the comparison method  or scaled
 subtraction is used for quantitative  analysis,
the  resolution  should  be  sufficient  to
separate spectral features  of  the  target
compounds from those of interfering species.
 For example,  in  the  case  of toluene the
 absorption   band  used   for  analysis   at
 1031 cm"1 is relatively broad.  At first, this
 would   indicate  that   a   low-resolution
 measurement would be sufficient. However,
 this  band  overlaps  with  bands  due  to
 atmospheric   water   vapor   and   CO2.
 Therefore, a higher resolution measurement
 is required  to  separate the toluene  from
 those of interfering species.

 3.  Acquire reference spectra of the target
 compounds.      If  the   specific  target
 compounds are known prior to beginning the
 monitoring study, reference spectra of the
 compounds of interest should be recorded at
 various   resolutions.   This   can   be
 accomplished  by  collecting  a  reference
 spectrum at the highest resolution setting on
 the instrument,  and then processing the data
 by  using the  required  number of Fourier
 transform  data  points  for  each  desired
 resolution.  When this method is used, only
 one -spectrum  has to  be 'collected.   By
 comparing the spectra processed at different
 resolutions,  the operator can determine the
 lowest  resolution measurement that  still
 resolves the spectral features  of interest.
This resolution  setting should be used  as a
starting point for future measurements.  If
this  is  not  possible,  the operator should
consult  reference libraries to help determine
the optimum resolution setting  required to
characterize the target analyte.

4. Develop calibration curves of the target
compounds.  If an inadequate resolution is
used, the relationship between  absorbance
and concentration will not be linear.   This
relationship   is   also   affected  by   the
apodization  function.   Calibration  curves
                                        7-26

-------
 MAIWM
 TECH:
                        TR-4423-99-03
 covering  the  concentration  range of  the
 target compounds expected in the ambient
 measurements  must  be  developed   at
 different  resolutions  and with the use  of
 different apodization functions to determine
 the optimum settings.  If the  compound of
 interest  does  not respond   linearly  with
 respect to concentration, a correction curve
 will need to be fitted to the data and. used in
 the quantitative analysis package.

 5. Use the same parameters to collect field
 spectra as were used to record the reference
 spectra.  If spectra from a commercial  or
 user-generated library  are to be the reference
 spectra for  quantitative  analysis, then the
 parameters that were used to generate those
 reference  spectra should be used to collect
 the field spectra.  Otherwise, errors in the
 measurement will occur.

 6.    Determine  that the  instrument  is
producing data at the specified resolution.
 The following  factors  should be considered
 here:  (a) that  the operator has selected the
 proper  parameters,  and  (b)  that  the
 instrument is operating to the manufacturer's
 specifications  and that the manufacturer's
 specifications  are  a true  indication of the
 capabilities of the instrument.

   a.   Most software packages allow the
   resolution to  be selected from  a menu.
   The software then  automatically sets the
   proper parameters  to collect data at the
   selected  resolution.  Therefore, there is
   very little opportunity for operator error.
   In  older versions of software that are not
   menu driven, but instead require entering
 line commands, many of the parameters
 affecting resolution, such as the number
 of  data points  used  for the  Fourier
 transform,  must be entered manually.  In
 this case, the operator  must  know, and
 enter  correctly,  all   of  the  proper
 parameters, and there is a greater chance
 of error.

 b.  If the instrument is not producing data
 of  the  selected resolution,  it  is also
 possible   that  the   instrument   is
 malfunctioning or that the manufacturer
 overstated   the   capabilities  of   the
 instrument. The following procedure can
 be  used to  determine if the instrument is
 producing   data   at   the   specified
 resolution.   There   is  a   cluster  of
 absorption  bands  between  1008  and
 1020 cm"1  due to water vapor that can
 be  used to verify  the resolution of the
 FT-IR  monitor. There   is -a  doublet
 centered at 1010.5 cm"1, a single band at
 1014.2 cm"1, and a  pair of bands at
 1017.5 and 1018 cm'1.  The singlet at
 1014.2 cm"1 has a theoretical  bandwidth
 of approximately 0.3 cm"1 (USF HITRAN-
 PC  [University of South Florida 1993]).
 This band is a  good  reference band for
 determining   the   actual   resolution
 measured by a medium- or low-resolution
 spectrometer.  For example, if this  band
 is measured at an  instrument setting of
 0.5cm"1, the FWHH should be 0.5cm"1.
 If measured with an instrument capable
 of  achieving  a higher  resolution, for
 example 0.25 cm"1, the FWHH should be
the  theoretical  value of 0.3 cm"1.  The
doublet centered at 1010.5 cm"1 is just
                                       7-27

-------
                                                                      TR-4423-99-03
    resolved at 0.5-cm"1 resolution, but is not
    resolved  at  1-cm~1  resolution.   The
    theoretical bandwidth  of each of these
    two absorption  bands is approximately
    0.1 cm'1 (USF HITRAN-PC), which makes
    them   good   reference  peaks  for
    instruments capable of measuring at a
    resolution of 0.125 cm"1.  The two peaks
    at 1017.5 and 1018 cm'1 are resolved at
    0.125 cm'1, but not at lower resolutions.
    If the instrument  is not performing  to
    specifications, there  is  most   likely
    an    alignment    problem   with  the
    interferometer or source optics.  Unless
    the operator is trained to perform this
    alignment,  a  representative  from the
    manufacturer  must   service  the
    instrument.

    The bands centered at 1014.2 cm"1 are a
    good test of resolution, but they  are not
    as  sensitive  to  misalignment  in the
    interferometer.  Other  bands that may
    also be used are the 2169-cm"1 band  of
    CO and  the  HDD  doublet  centered  at
    approximately 2720 cm"1.  These bands
    at shorter  wavelength  (higher   wave
    number)   are  more   sensitive  to
    interferometer misalignment and can also
    be used to determine the stability of the
    interferometer.

      These are  general guidelines to be
used when choosing instrumental parameters
to collect data.  In reality, the user's  choice
of parameters that can be actually used may
be limited by either the specifications of the
spectrometer or  by  the  software.  For
example,  one  software package supplied
 with  FT-IR  long-path,-open-path  systems
 allows only triangular apodization  with  no
 additional zero  filling  for  processing  the
 interferogram   with   the   menu-driven
 commands.   These parameters cannot  be
 changed unless the user has the capability of
 editing the software code.  As the resolution
 requirements of  long-path, open-path FT-IR
 monitors  become  better  defined,   the
 manufacturers  will  most  likely   produce
 instruments  and  software  to meet  those
 needs.

 7.8    References

 Bell,   R.J.   1972.   Introductory  Fourier
 Transform Spectroscopy.  Academic Press,
 New York.

 Bittner, H.,  T.  Eisenmann, H.  Mosebach,
 M.   Erhard,  and   M.   Resch.   1994.
 Measurements  of  Diffuse  Emissions  of
 Volatile  Organic  Compounds   by   High
 Resolution  FTIR  Remote  Sensing.   SP-89
 Optical   Sensing   for    Environmental
 Monitoring,  Air   &  Waste  Management
 Association, Pittsburgh, PA, pp. 443-454.

 Childers,  J.W.,  and  E.L.  Thompson,  Jr.
 1994.  Resolution Requirements in Long-Path
 FT-IR Spectrometry, SP-89 Optical  Sensing
 for Environmental Monitoring,  Air & Waste
 Management Association, Pittsburgh, PA, pp.
 38-46.

 Grasselli,   J.G.,   P.R.   Griffiths,   and
 R.W. Hannah. 1982. Criteria for Presentation
 of  Spectra  from  Computerized    IR
 Instruments. Appl. Spectrosc. 36:87-91.

 Griffiths,  P.R., and J.A. de Haseth.  1986.
Fourier Transform  Infrared Spectrometry.
John Wiley and Sons, New York.
                                       7-28

-------
 TFrU;TI;lSSJrir=
 I f,\ffl: i'j -.-j-j	
                                                                      TR-4423-99-03
 Griffiths, P.P., D. Qin, R.L. Richardson, and
 C. Zhu. 1993. Atmospheric Monitoring with
 FT-IR   Spectrometry:   Strengths   and
 Weaknesses  of  Measurement  at  Low
 Resolution.  Presented at the 44th Pittsburgh
 Conference  and  Exposition  on Analytical
 Chemistry   and  .Applied  Spectroscopy,
 Atlanta, GA, March 7-12, paper no.  1142.

 Griffiths, P.R., Richardson, R.L., Qin, D., and
 Zhu,  C.  1995.  Open-Path   Atmospheric
 Monitoring   with  a  Low-resolution FT-IR
 Spectrometer.   Proceedings   of   Optical
 Sensing  for Environmental  and  Process
 Monitoring, O.A. Simpson,  Ed., VIP-37, Air &
 Waste Management Association, Pittsburgh,
 PA, pp. 274-284.

 Herres,  W.,  and  J.  Gronholz.     1984.
 Understanding FT-IR data Processing Part 1:
 Data Acquisition and Fourier Transformation.
 Computer Applications  in  the Laboratory
 4:216-220.

 Horlick, G.   1968.  Introduction to  Fourier
 Transform Spectroscopy.  Appl. Spectrosc.
 22:617-626.

 Marshall,    T.M.,   C.T.   Chaffin,   V.S.
 Makepeace,  R.M. Hoffman, R.M. Hammaker,
W.G. Fateley, et al.  1994.  Investigation of
the Effects of Resolution on the Performance
of Classical  Least-Squares (CLS) Spectral
 Interpretation  Programs When  Applied  to
Volatile Organic  Compounds  (VOCs)  of
 Interest in  Remote Sensing Using Open-Air
 Long-Path Fourier Transform Infrared (FT-IR)
 Spectrometry. J. Mo/. Structure 324:19.

 Russwurm, G.M.  1992.  Quality Assurance,
 Water Vapor, and Analysis of  FTIR Data,
 Presented at the Air & Waste Management
 Association Annual Meeting, Kansas  City,
 MO, June.

 Spellicy, R.L.,  W.L.  Crow,  J.A.  Draves,
 W.F.  Buchholtz,  and W.F.  Herget.  1991.
 Spectroscopic Remote Sensing:  Addressing
 Requirements  of  the   Clean   Air  Act.
 Spectroscopy 6:24-34.

 Strang, C.R., and S.P. Levine.  1989.  The
 Limits of Detection for  the Monitoring  of
 Semiconductor Manufacturing Gas and Vapor
 Emissions  by Fourier Transform  Infrared
 (FTIR) Spectroscopy.  Am. Ind. Hyg. Assoc.
J.  50:78-84.

 Strang, C.R., S.P. Levine, and W.F. Herget.
 1989.   A  Preliminary  Evaluation  of the
 Fourier  Transform    Infrared    (FTIR)
 Spectrometer as  a Quantitative Air Monitor
for Semiconductor Manufacturing  Process
 Emissions.    Am. Ind.   Hyg.  Assoc.  J.
 50:70 77.

University of South Florida.  1993.   USF
HITRAN-PC,  University  of  South  Florida,
Tampa, FL.
                                       7-29

-------
                                                                      TR-4423-99-03
                                       Chapter 8
          Nonlinear Response Caused by Apodization Functions and
                             Its Effect on FT-IR Data
                                     SUMMARY

        This chapter discusses the effects of apodization and temperature on the accuracy
  of the FT-IR data.  It places special emphasis on the following.
        •  Why apodization causes  the  FT-IR response to be nonlinear  and  how this
           response is affected by resolution.
        •  The FT-IR response to absorption of water, methane, and ammonia as functions
           of concentration, temperature, and resolution.
        •  The errors  incurred by using  classical least squares and assuming a linear
           response.

  This chapter uses a set of spectra calculated by using the HITRAN database over a 207-m
  path at various concentrations and temperatures.
8.1    Introduction and Overview

       The technique  of using  an  FT-IR
system over a long, open atmospheric path
to  monitor  atmospheric  pollutants   has
undergone a vigorous development over the
past several  years. However, an in-depth
analysis of the error associated with the data
has never been reported. One source of error
in an FT-IR measurement is the application of
an apodization function to the interferogram,
causing a nonlinear response  to changes in
gas concentration.

       In an FT-IR instrument, the moving
mirror  of the interferometer must travel  a
certain distance in order to achieve a specific
resolution.  At the end  of  the  travel,  the
mirror  returns to its  original position  and
repeats the  movement.  The abrupt cutting
off of the interferogram at the end of the
mirror travel is called truncation. When the
Fourier   transform   of  the   truncated
interferogram  is  performed,   an  exact
spectrum is  not  reproduced, and there are
spurious oscillations  in  the  reproduced
spectrum. These  oscillations  distort  the
shape of the spectral features,  both in the
bandwidth  and  in  the   amplitude  of  the
various  spectral  peaks.  A  graph of  an
unapodized  (boxcar)  water  spectrum  is
shown  in   Figure  8-1 a.  The  graph  in
Figure 8-1 b is the same portion of the water
spectrum, but it has been processed  with
triangular apodization. Both spectra are at
0.25-cm'1 resolution and  are calculated for a
                                        8-1

-------
                                                                         TR-4423-99-03
           UJ
           o
           m
           a
           o
           tn
           m
                   A. boxcar apodization
                   B. triangular apodization
                     1010
                                1012        1014         1016

                                     WAVE NUMBER (CM-1)
                                                               1018
                                                                          1020
       Figure 8-1. A Portion of a Water Spectrum Using (A) Boxcar Apodization and
       (B) Triangular Apodization.
path length of 207 m and a partial pressure
of water of 1 5 torr. The spurious oscillations
are clearly visible in the upper curve.
changes in absorbance. This is separate from
the detector's  nonlinearity, which has not
been considered in this discussion.
       To  reduce the magnitude of these
oscillations,  various  apodization  functions
can  be applied to the interferogram before
the   transform   is  performed.   Many
mathematical  forms  of  the  apodization
functions have been investigated (Happ and
Genzel  1961; Filler 1964; Norton and Beer
1976;  Kauppinen  et al.  1981) and  are
available for use by the operator. In general,
these apodization functions create a broader
line width than is  available in the unapodized
spectrum and are also partly responsible for
a nonlinear response of the instrument to
       The FT-IR instruments commercially
available for remote sensing generally do not
provide  a  large selection of  apodization
functions for  the operator, and  triangular
apodization is commonly  used. This means
that the interferogram is  multiplied  by a
triangular mathematical  function  with the
peak at the so-called center burst before the
Fourier transform is performed."

       The  analysis  of  the  spectra  for
concentration is generally done by using the
method of classical least squares and  using
                                         8-2

-------
MAfi/:
jECHmimm
                                                                       TR-4423-99-03
reference spectra of known concentration-
path  length products, and these  spectra
have  also been processed with triangular
apodization.  Included  in  these  reference
spectra is a water vapor reference that is
generally  manufactured  by the  operator
(Russwurm  1996)  from  a field spectrum
taken during an actual monitoring program.
Generally,  there is  only  one  reference
spectrum per gas, and it has been acquired
for a single   concentration-path  length
product at a specific temperature, although
the temperature information is normally not
supplied with the spectrum.

       The procedure is  usually set up so
that  the  least  squares  calculation  is
performed over a  range  of  wave numbers
that encompasses a major absorption peak of
the target gas and generally accounts for all
known spectral interferences. The ensuing
mathematical  analysis process assumes a
linear  relation  between  absorbance  and
concentration,  as described  by Beer's  law.
Since the apodization function produces a
nonlinear   instrument   response   with
concentration  but  the mathematical data
process,   usually classical  least  squares,
assumes  linearity,  an  error  in  the data
occurs.

       How  this manifests itself is  shown
schematically in Figure 8-2. The linear curve
in Figure 8-2 represents the assumed change
in absorbance with concentration-path length
product while the quadratic curve represents
the actual  response expected  in  the field
                                                   Actual response calculated using triangular apodization
                                            o
                                            ca
                                            < 0.2
                                               0     200    400    600    800    1000
                                                     concentration path length product (PPMM)
                                           Figure 8-2. Schematic of Actual and
                                           Assumed FT-IR Responses.
                                           spectra. While Figure 8-2 is intended to be
                                           indicative  of the  instrument response, the
                                           actual curve is the response for methane at
                                           2927 cm"1. The point where the two curves
                                           cross  is  the  concentration-path  length
                                           product of the reference  spectrum.  From
                                           Figure   8-2  it   is   seen  that  if   the
                                           concentration-path  length  product  of the
                                           field  spectrum  is less than  that  of the
                                           reference,  the  actual  concentration  is
                                           overestimated, and the  actual concentration
                                           is underestimated when the concentration-
                                           path  length product of the field spectrum is
                                           greater than that of the  reference. Evidently,
                                           triangular apodization creates a broader final
                                           absorption  feature  (Marshall  and Verdun
                                           1990)  than most  of the  other  commonly
                                           used  functions   and  therefore  a   more
                                           pronounced  nonlinear response.  Triangular
                                           apodization  has  been used   exclusively
                                           throughout the remainder  of this chapter
                                           because use of this apodization function is
                                           considered a worst case.
                                        8-3

-------
       This chapter examines the magnitude
 of the error in several ways—by examining
 the error for a single line in a pure gas, then
 considering  the  effects of  the  spectral
 interference created  by water  vapor,  and
 finally examining the error created when a
 range of  wavelengths  is  used  with  the
 classical least-squares technique.

 8.2    Procedure and Theoretical Basis

       The primary data used in this chapter
 is a  set  of spectra  calculated from  the
 HITRAN database.1 The three gases used are
 ammonia, methane, and water vapor. These
 three   gases   were   chosen   because
 absorbance due to ammonia is  not heavily
 impacted by water, and the absorbance due
 to methane  is  strongly affected by water.
 Water is the primary interfering species, and
 it perhaps presents the largest problem in the
 atmospheric spectroscopy of the mid-infrared
 region. Indeed, the spectrum obtained from
 the open  atmosphere is primarily a water
 vapor and carbon dioxide spectrum, and the
 gases of interest (the target gases) represent
 only small  perturbations to  that spectrum.
 The HITRAN database was used to calculate
 Lorentzian  absorption  lines  over a 207-m
 path at  a total  pressure of one atmosphere
 and   with    varying  concentrations   and
temperatures. Following that calculation, the
spectra  were reprocessed with an algorithm
that  allows an apodization  function to be
                                                                        TR-4423-99-03
 applied  to  the  spectra  and  also allows
 changing the  resolution  to  some  other
 desired  resolution. To  match  the wanted
 resolution  and  to  apply  the  apodization
 function,  the   following   mathematical
 procedure has been used.

    •  Calculate an unapodized high-
       resolution spectrum 7 from  HITRAN.
    •  Calculate the inverse Fourier
       transform  of the  spectrum T.
    •  Multiply this inverse transform by
       the apodization function.
    •  Calculate the Fourier transform of
       the product spectrum  from the step
       above.

       The mathematical justification for this
 procedure is straightforward and is described
 as follows. The spectrum actually measured
 by an FT-IR instrument, Tr (w), is given by the
 convolution  of the true spectrum  7"(o))  and
 the instrument line function 5(o)) as
                                    (8-1
However, the instrument line function for an
FT-IR instrument is the Fourier transform of
the apodization function /4(6), where 6 is the
optical path difference in the two legs of the
interferometer.  That is,  5(co) = F[/4(6)], or
X(8) = F -'[5(0))].  Therefore, Equation (8-1)
can be rewritten as
                                   (8-2)
       'The AFGL HITRAN molecular absorption
parameters database. See, for example, Rothman
etal. (1987).
                                        8-4

-------

                                                                       TR-4423-99-03
       Application of the inverse transform
followed by the  forward  transform  to  the
right-hand side of Equation  (8-2) yields
                                   (8-3)
       The convolution theorem for Fourier
transforms states that the Fourier transform
(or  inverse)   of  the  convolution  of two
functions is  equal  to  the  product of their
Fourier transforms (or inverses). Applying this
theorem to Equation  (8-3)  gives  the final
expression that



and all the spectra that were calculated from
HITRAN and  FASCODE and  used for this
effort were processed  in this manner.

       It is convenient to discuss  here two
other  points  about  the  linearity  of the
response of the FT-IR instrument to changes
in concentration of the atmospheric gases.
These  points  consider  what concentration
levels  must   be  exceeded to  make the
response nonlinear  and what resolution will
make the response nonlinear. To  evaluate
these points,  note that the intensity 7(co) that
is  measured  by   a   Fourier  transform
spectrometer  is   proportional   to  the
convolution of the incident intensity and the
Fourier  transform  of  the  instrument line
function  S(u>). But  the function 5(co) is  a
function of the resolution,  which in turn is
dependent upon the maximum optical path
difference between the two mirrors of the
interferometer. The measured intensity can
be written as
        /m(o>) = D(a>)/7(co+8)S(8,7?)d8
In keeping with Beer's law, the measured
background  intensity is given as

       7mo(co) = D(G))//0(&>+6)S(6,/?)d8
where -D(co) is the response of the instrument
optics  and   the  electronics. The  actual
transmission 7"(a)) through  the   absorbing
medium can be written (from Beer's law) as
                                                             = /0(o))3Tto)
                                            and for a single  gas  component 7*(co)  =
                                            exp(-ca(o))). Here cis the concentration-path
                                            length product (with units of ppm-m) and ), can then be written as
                                           If  the  background  intensity is a  slowly
                                           varying function over the  instrument  line
                                           function 5, then it can be taken  outside the
                                           integral sign, and the measured transmission,
                                           Tr (co), can be rewritten as
                                           Because the  instrument line function has a
                                           unity normalized integral (^S(b,R)db = 1), the
                                           measured  transmission  T, can simply  be
                                           written as
                                        8-5

-------
        r±J=r.?. ia.Is =
        •.-• -.: •••=•••= =
                            TR-4423-99-03
                                    (8-5)
       The  measured  absorbance  can be
obtained by taking the negative logarithm of
Equation (8-5). Substituting for Tfrom Beer's
law, the expression



is obtained. From Equation (8-6), the above
two points can be examined. If the exponent
ca(o)) is small, only the first two terms of the
expanded exponential term can be retained,
and the expression can be written as
    Am((*>) = -log{/[l-ca(cj + 5)]S(5,/?)rf§}

Then, making use of the unity integral  of S
and the fact that logd - x) - - x for small x,
the measured absorbance can be written as
        Xm(w) = c/a(o)
and   this   expression   is   linear   with
concentration.  The  assumption  that  is
important here is that the product ca must
be small, at least compared to the quadratic
term in the expansion of the exponential.

      Another limiting case is that  of high
spectral resolution;  that  is, the instrument
line function should be small compared to the
half  width of the gas absorption features.
With  that assumption,  the  transmission
function   is  a  slowly   varying  term  in
comparison with the instrument line function
and  can  be  brought outside the integral.
 Then, again using the unity normal integral of
 5, the absorbance A can simply be written as
      Am(u>) = -log[exp(-ca(o)))] =  ca(co)
 which is again linear with concentration.

       The predominant absorption features
 in spectra  taken over an open  path  come
 from  water vapor in  the  atmosphere, and
 these  features  have  a FWHH  of  about
 0.1-0.2 cm"1.  Therefore,  the  assumption
 above  implies that the instrument resolution
 should  be  about 0.05 cm"1. However, this
 requirement is not satisfied at present by any
 of the commercially available instruments for
 OP/FT-IR monitoring. Therefore, the response
 of the  instruments that are available should
 be  expected  to be nonlinear, at least for
 water.

       It is  not easy to verify experimentally
 the nonlinear response of an FT-IR instrument
 since the concentrations normally present in
 the open  air  do  not  encompass  a  large
 enough range. The nonlinear response has
 been verified  experimentally for methane by
 Ropertz (1997),  and a copy of a  portion  of
 his data is shown in Figure 8-3. Additionally,
 reference  spectra  made  from  known
 concentrations of  pure gases are available
 commercially. One ammonia reference that is
 commonly  used has  a  concentration-path
 length  product of 550 ppm-m.  When the
 absorption  of that  spectrum is compared to
the absorption of the calculated spectrum  at
the same concentration-path length product,
the agreement is within about 7%.
                                        8-6

-------

                                                                        TR-4423-99-03
                        PolynomiKfte Rttgntsion 2.0r4xng d«r
                        MvQargibniiM bd dner opUcnen Weg-
                        Iflng* von 96.88m
                PdynomJscfia Regression 2.0nJnung d«f
                MeOeigebnlsM t>d «ln«r optiscfwn W«j-
                Urga von 20.7Vm
                              aooo     loooo
                              ZuttindMctu* CM, / ppfn*m
Figure 8-3. Measured Concentration of
Methane vs. the Experimental Response of
the FT-IR (reprinted with permission of
A. Ropertz).

8.3    Results of Calculations

       The effect of the nonlinear response
has been studied for three gases: methane,
ammonia,   and  water  vapor.  All  the
absorbances were calculated for a range of
concentration-path   length   products   at
specific  wavelengths and  then  fit to  a
polynomial of order 2, since this seemed to
be the best fit. In one case (an ammonia line
at 1046  cm"1), the quadratic fit was  not
really satisfactory.  The absorbance  at that
ammonia  line  was  best fit with  a linear
approximation over a portion of the data and
a   quadratic   approximation   over  the
remainder. This is a clear indication that the
functions describing  the  absorbance  at
various wave  numbers  are  different. The
absorbance was also calculated as a function
of temperature and for the four resolutions
0.25, 0.5,1.0,  and 4.0 cm'1.

       The  coefficients  for   the   various
polynomials  were  calculated  by  using
 nonlinear regression. It was then possible to
 determine the maximum value of either the
 concentration-path  length  product or  the
 temperature  that  can be  used before  the
 response must be considered nonlinear. To
 do that, the  ratio  of the quadratic  to linear
 terms of the  polynomial fit should be small,
 and this ratio was arbitrarily set equal to 0.1.
 This  is  equivalent  to  requiring  that  the
 concentration-path  length  product or  the
 temperature  be less than the term 0.1c, lcq,
 where the c, and the cq are the linear and the
 quadratic coefficients, respectively.

      The   methane   absorbance    was
 measured   at  various   points   in    the
 2900-3000-cm'1 region because that is  the
 region most commonly used for the analysis.
 Within  this  region  the  absorbance   was
 calculated for the concentration-path length
 product range from 300 to 2100 ppm-m. This
 corresponds to a path length of 200 m and a
 range of concentrations from 1.5 to  10 ppm.
 This range was selected because it covers
 the  range from  slightly   below accepted
 ambient  levels to  slightly above the  levels
 that could be expected  at  coal mines,  land
 fills, or hog farms.

      The   ammonia   absorbance   was
 measured  at   various   points   in   the
 850-1050-cm'1  region  and   covering  the
 range from essentially 0 to 1100 ppm-m. For
 a 200-m path length, this range covers the
concentration from  the  ambient levels to
 5  ppm,  or that concentration  seen at hog
farms.
                                         8-7

-------

                                                                      TR-4423-99-03
       The water  spectra were calculated
 over the wavelength ranges of methane and
 ammonia since the absorbance due to water
 is the predominant interfering absorbance in
 the mid-infrared atmospheric spectrum. The
 water  vapor  absorbance was calculated in
 terms  of the  partial pressure of water from
 0.5  to 35 torr,  or from slightly  below to
 slightly  above that range  seen  here at
 Research Triangle Park (RTP), North Carolina,
 over a year.  The various graphs  for water
 that follow  are plotted with  the  partial
 pressure in torr along  the  abscissa.  The
 water  absorbances were  calculated for a
 path length of 207  m,  which  is  the path
 length  used when we take measurements at
 RTP.

       All the absorbances were calculated
 over the temperature range from  250 K to
 310 K in 5-K increments. This range was
 chosen because  it  covers the yearly range
 normally  seen  at  RTP.  This  range  is
 appropriate also in  that it encompasses the
 temperature   range   over   which   the
 commercially  available instruments can work.
 Finally, all the spectra were  calculated for
the following  four resolutions:  0.25,  0.5,
 1.0, and 4.0 cm'1.

       Figures 8-4, 8-5,  and  8-6 show the
overall   absorbance  due to  methane  at
2927 cm'1, ammonia at 967 cm"1, and water
at 1014.5 cm"1, respectively. The plots for
methane and ammonia show the  absorbance
as a  function of  the concentration-path
length  product and the temperature. The
 water absorbance is shown as a function of
 the   partial   pressure   of   water   and
 temperature,  with  the absorbance  being
 calculated for a path length of 207 m.

       Figures   8-7   and   8-8   show,
 respectively,  the absorbance for methane
 and    ammonia   as   a   function   of
 concentration-path  length  product  at  a
 constant temperature  of 295 K.  Figure 8-9
 shows the absorbance due to water  as  a
 function of the partial pressure of water, also
 at a  constant temperature of 295 K. These
 graphs  also   contain  the  second-order
 polynomial that best describes the curvature
 of the absorbance.

       As  stated above, the dependence of
 the absorbance  on  temperature  was  also
 calculated for these three gases. The change
 in absorbance as a function of temperature
 for methane and ammonia is not very strong,
 as is depicted in  Figures  8-4 and 8-5,  but
 that is not the case for water. The change in
 absorbance with temperature for water is
 shown in  more  detail  in  Figures  8-10 and
 8-11. Figure 8-10 shows the dependence of
 water absorbance on temperature at a partial
 pressure of 0.5 torr,  and Figure 8-11  shows
this dependence  at a  partial pressure  of
 35 torr.
                                       8-8

-------
                                                                        TR-4423-99-03
 A. METHANE AT 1/4 CM"1 RESOLUTION
                                                  B. METHANE AT 1/2 CM"' RESOLUTION
C. METHANE AT 1.0 CM"' RESOLUTION
                                                  D. METHANE AT 4.0 CM "RESOLUTION
        Figure 8-4. Methane Absorbance at 2927 cm'1.
A. AMMONIA AT 1/4 CM"' RESOLUTION
                                                  8. AMMONIA AT 1/2 CM"' RESOLUTION
C. AMMONIA AT 1.0 CM"' RESOLUTION
                                                     0. AMMONIA AT 4.0 CM'1
        Figure 8-5. Ammonia Absorbance at 967 cm   .
                                 8-9

-------
                                                                                    TR-4423-99-03
    A. WATER AT 1/4 CM "' RESOLUTION
                                                            8. WATER AT 1/2 CM*1 RESOLUTION
    C. WATER AT 1.0 CM'1 RESOLUTION
                                                            D. WATER AT 4.0 CM"' RESOLUTION
          Figure 8-6.Water Absorbance at  1014.5 cm1.
        A. 114 CW1 RESOLUTION
ADS n -2.7eS«10^ » 2.401I10~*(CL) - 1.0323«»10*(CL)*
   400     100    1200    1600   2000    2400
            CL (PPM-M)
                                                              e. in CM 'RESOLUTION
                          r'(CL|'
                                               0.35

                                            S  0.25

                                            I  «•"

                                               0.05

                                              -O.OS
    400    100    1200    1000    2000   2400
           CL (PPM*M)
         C. 1 CM-' RESOLUTION
 ABS • 1.892110-" + 1.931Iia^(CL) • 1.258X1Q*(CL)2
    400     100     1200    1100   2000    2400
            CL (PPWM)
        D. 4 CM*1 RESOLUTION
ABS» 3.737x10^ * 6.21 x10~ft(CL)-
                                            fj 0.07
                                            I
   400     100    1200    1000    2000    2400
           CL  (PPM-M)
      Figure 8-7.  Methane Absorbance vs. CL at 2927 cm"
                                       8-10

-------
wmm
                                                                                      TR-4423-99-03
              A. 1M CM'1 RESOLUTION
             i1ffD» 1.73t4HOaCL -S.i72623i1ff'|CL|1
            200    400    100    100   1000   1100
                 CL (PPMM)
                                                                             B, iaCM 'RESOLUTION
                                                                            I193* 1.H73i10':lCL.3.713J44X10''(CL)a
                                                                           100    400    100    100   1000    1200
                                                                                 CL (PPMM)
             C. 1.0 CM'1 RESOLUTION
      ABS s 1.4M4i«3 * 1.077x10°CL • 3.1S021x1'"'
_»-"-*
.- ,•"•''" : :
^••*'** •
><"'": • " i 	
S IS IS JS
PARTIAL PRESSURE fTORR)
0.22
| 0.11
a
0 O.lQ
0.04
' .



0.10
0
a
or
o
! 0.01

.

.«•"
.*"*
..-**
»""**
S i 1* 21 31
PARTIAL PRESSURE FTORR)
o. 4.0 CM* 'RESOLUTION
ABS a 0.117X10T4* 1.4l07i10'~V * I.SJilO^P1
• ' "- XI-'-

„***
^r*' ';. . . 	 	 ; . .
• —•**".. .'..:. 	 • -
5 15 15 35
PARTIAL PRESSURE fTORR)
          Figure 8-9. Water Absorbance vs. fat  1014.5 cm'1.
                                         8-11

-------
                                                                                                 TR-4423-99-03
                  A.1HCM •'RESOLUTION
                Mi10  J.2*43110 "*T*6J172i10 -
    240         210
                   TEMPERATURE (K)
                                   300         320
        6.112 CM •' RESOLUTION
ABS.1.7H49110 J-1.«U10 •*T»4.C04T««10 '7(
                                                        g 0*3
                                                            240         260
                                                                                 no         MO
                                                                           TEMPERATURE (K)
                 C. I CM1 RESOLUTION
         ABS-1.02741110 J-8J9itO <5T*224S7S4t10  -'ft)3
                        230        300
                  TEMPERATURE [K]
       D.
-------
8.4    Analysis

       Once  the  absorbance  function  is
determined,  it is instructive  to find the
maximum value  in concentration for which
the absorbance  can be  considered  linear.
This  can be  done  by requiring that the
quadratic term in the polynomial representing
the absorbance be small in comparison with
the linear term. That is, if the absorbance is
represented by the polynomial ABS  = a0 +
a^X + a2X2, then for the absorbance to be
considered linear, the ratio a2X2/a{X must be
small. The  value  of X can  be found  by
requiring that  this ratio be  less than some
value  k, which is equivalent to requiring that
the independent variable X must  be less than
the quantity atk/a2. These values have been
calculated  for  a k  value  of  0.1  and  a
temperature of 295  K and are shown for
methane  and ammonia  in column 4  of
Table  8-1.

       Next, the error is estimated. The error
made  when a linear response  is assumed
depends  on   how    far  removed   the
                                                                     TR-4423-99-03
                                          concentration-path  length  product  of the
                                          field spectrum (the  actual measurement)  is
                                          from the concentration-path length product
                                          of the reference spectrum. (See Figure 8-2.)
                                          Column 5 of Table 8-1 shows the predicted
                                          error at the calculated maximum values Xmca
                                          when references at  81  and 550  pprrvm are
                                          used for methane and ammonia, respectively.
                                          These  reference  values  come  from two
                                          commercial sources  that are commonly used
                                          by the  instrument  users.  The errors  have
                                          been calculated  in the following way. The
                                          linear  response  is  simply based  on  the
                                          assumption  that  the  absorbance   goes
                                          through the 0 and the reference absorbance.
                                          at the  concentration-path  length  product
                                          listed. Thus the linear response is given by
                                          the expression ABSL  = (ABSREF )(CLj/CLR£/r
                                          where the (CL)REF is  the concentration-path
                                          length  product of the reference  spectrum.
                                          The absorbances for both the linear response
                                          and the quadratic response can be calculated
                                          for the same  CL and the error found  from
                                          %E = (ABSQ-ABSL)/ABSQ. The results of these
                                          calculations are given in Table 8-1.
Table 8-1. Maximum Values Over Which Response Can Be Considered Linear and Associated Errors
Gas v (cm"1)
Methane 2927



Ammonia 967



Resolution (cm"1)
0.25
0.5
1.0
4.0
0.25
0.5
1.0
4.0
Xmax (ppm-m)
2367
1799
1232
861
307
330
342
305
% Error at Xm(Lr

-9.3, Ref. at 81 ppm-rn
-10.2
-11.9
-17.4
+ 8.9, Ref. at 550 ppm
+ 7.6
+ 6.9
+ 9.2



•m



                                      8-13

-------
                                                                       TR-4423-99-03
       Similar calculations can be carried out
 for water, but in that case it is found that the
 absorbance is not linear for resolutions poorer
 than 0.5 cm "1 throughout the entire range of
 partial pressures for which the absorbances
 were calculated. This is also true for  the
 water  absorbance  as  a   function   of
 temperature. The water absorbance was also
 examined  at   1012.4  cm"1,  which   is
 considered to be at the baseline  between
 peaks. As the partial pressure of water rises
 and the absorbance increases, the  wings of
 the individual peaks merge  to  become a
 continuum,  which then forms the spectral
 baseline. This is particularly troublesome for
 the  least-squares  analysis   because  the
 absorbance  at the  baseline  changes in a
 nonlinear way over the entire range  of partial
 pressures for which the calculations were
 done.

       While  the actual  concentration   of
 water is of little interest  as an atmospheric
 pollutant  and water is  seldom  a  target
 compound,   it   must  almost  always  be
 included in the least-squares analysis  as an
 interfering species.  One  question that this
 work set out  to answer is how to overcome
 the difficulty  with matching the water vapor
 reference  to  the actual  water in the field
 spectrum. The concentration of water can
 and does change rapidly and dramatically in
 the atmosphere. Therefore, it was instructive
to look  at the case  where the water vapor
 reference was obtained from a spectrum that
 was taken when the partial pressure was
 10  torr but the actual partial pressure of
water  for the spectra being analyzed  was
20 torr.

       Figure 8-12  is a  plot of the water
absorbance  in the 1014.5-cm'1 region. The
solid curve is a water spectrum calculated for
20 torr of water at 207 m. The dotted curve
is a   10 torr spectrum  that  has  been
normalized  to  the  20  torr  spectrum  at
1014.5 cm"1. Note that the only place where
the curves seem  to match is at that peak,
and there is a particularly large discrepancy
in the baseline. But this kind of normalization
is not  exactly what classical least squares
does.  The mathematics  of classical least
squares adjusts the curves, one to the other,
so that the sum of the differences between
them squared is a minimum. It does that by
calculating a slope as the single  coefficient
by which to  multiply the entire curve (in this
case the 10 torr  spectrum) and by adding
some constant value to the result. However,
because  of  the nonlinear response of the
instrument and the fact that the polynomials
describing  the absorbance  seem  to  be
    1006   1008  1010  1012  1014  1016  1018  1020  1022
                WAVE NUMBER (CM"1)
 Figure 8-12. Match of Water Absorbance
 at 1014.5 cm"1.
                                        8-14

-------
                                                                       TR-4423-99-03
 different for different wave numbers, no such
 process can really match the curves.

       All the above analysis describes what
 error would, be  made  if the analysis  for
 concentration were done at a single  wave
 number using  a single reference. However,
 the  classical least squares analysis uses a
 range of wave numbers for the calculation. In
 the  remainder of this Analysis section, the
 classical  least-squares technique is applied to
 this  data  to  determine  the  actual  error
 involved  with  this procedure in light of the
 nonlinearities discussed above.

       To explore the effect of a reference
 spectrum whose concentration-path length
 product  is not  at the  concentration-path
 length product  of  the field  spectrum,  a
 methane spectrum calculated for 6 ppm of
 methane at 295 K was used as a "field"
 spectrum. The spectra  used as  references
 were all  at 295  K, but the concentrations
 were allowed  to  vary from  1.5 ppm  to
 10 ppm.  The  results for these calculations
 are  shown in  Figures  8-13  and  8-14.
 Figure  8-13 is  a  plot of the analyzed
 concentrations  of methane over the wave
 number range 2915-2929 cm"1. This region
 was  chosen for this analysis because there
 are two methane lines that are not strongly
 impacted by the interfering species water. In
 Figure  8-14, the analysis region  has  been
 expanded to cover the region from 2900 to
 3000 cm'1. These plots clearly indicate that
the best accuracy is at the higher resolution,
 and  also that the smaller analysis region
seems to also have better  accuracy.  This
must  indicate   that   the   least-squares
technique is less efficient at matching the
two spectra when the  analysis region  is
large.

       To explore the effect of water as an
interfering  species, a  "field" spectrum of
methane and water was calculated from
•HITRAN and then analyzed by classical least
squares.   The   field   spectrum   used  a
concentration of 6 ppm for methane and a
partial  pressure of water of 1 5 torr, all  at a
temperature of 295 K. The partial pressure of
the spectra used as water references  was
allowed to  range  from  10  to  20 torr at
295 K. Figure 8-15  shows the results from
classical least-squares analysis for methane
for  the   two   wave  number   regions
2915-2929 cm'1 and 2900-3000 cm'1. The
data points for the  two regions  have been
shifted slightly  for clarity.  These results
indicate that the methane concentration is
not strongly  impacted  by the presence of
water.  There does seem to  be  a bias  of a
few percent  for each  region from  the  true
value, and the magnitude of the error bars is
larger for the larger region. It is not clear why
the small but definite biases in these graphs
are present. A possible reason is  that in the
presence of  water, because of  its high
concentration-path  length  product,  the
resulting absorption spectrum is not  simply a
linear combination of the absorbances from
methane and water.
                                       8-15

-------
                                                           TR-4423-99-03
                    FIELD SPECTRUM IS FIXED AT 6 PPM
O
Z
O
O
Q
UJ
   6.8
   6.4
6.0
<  5.6
   5.2
   0          2         4         6         8        10

                REFERENCE SPECTRUM CONC (PPM)

  Figure 8-13. Analysis Results for Methane from 2915 to 2929 cm'1.
                                                                12
             FIELD SPECTRUM CONCENTRATION FIXED AT 6 PPM
   7.8
2
Q.
£L_

O
z
O
O
Q
OJ
<

<
6.6
   5.4
   4.2
  0         2         4         6         8         10

                REFERENCE SPECTRUM CONC (PPM)

  Figure 8-14. Analysis Results for Methane from 2900 to 3000 cm"1.
                                                                12
                               8-16

-------
           ..
 TECHm^mm
                                                                      TR-4423-99-03
           Field spectrum Is 6 ppm CH« plusl 5 torr water
 i.»
 s-
 o
     J
T
        PARTIAL PRESSURE OF WATER REFERENCE (TORR)

Figure 8-15. Methane Analysis Allowing the
Water Reference Concentration to Vary.
       Some further study has shown that as
the resolution gets poorer, the errors for the
above calculations also get worse (data not
shown).  The   reason for  water  as  an
interfering species to not strongly impact the
methane   analysis,   although   somewhat
surprising,   is  clear.  The  mathematical
procedure of classical least squares adjusts
the  spectra so  that for  the  individual
components most of the variability of the
absorbance  in   the  field  spectrum  is
accounted  for. It  makes that adjustment by
minimizing  the  sum  of the squares of the
residuals. But, in  the case of water—just as
with methane, the calculated concentration
is  not the  correct value  unless the  water
reference has the same partial pressure and
was taken  at the same temperature as the
water  in the field spectrum. Nonetheless,
because the residuals have  been minimized,
classical least squares accounts for most of
the variability in the absorbance of the  water
and also minimizes the effect of water on the
analysis of  methane.
 8.5    Discussion

       The  calculated spectra have  been
 analyzed to determine the magnitude of the
 error when a linear function is assumed for
 the response of the FT-IR instrument. From
 Figure  8-2 it is  seen that for a measured
 absorbance   the   concentration   is
 overestimated when the concentration-path
 length  product of the field spectrum  is less
 than  that  of   the   reference  spectrum.
 Conversely,   the   concentration   is
 underestimated when the concentration-path
 length product of the field spectrum is high in
 comparison with the reference. It  is also
 shown here that the  magnitude of the error
 is larger with poorer resolution.

       At present there seems to be  no
 simple  correction that can be applied to the
 calculated concentration-path length product
 to account  for  the  error.  In the  cases
 presented  here,  the  spectra  could  be
 calculated from the HITRAN database, but
 the number of gases  available in HITRAN is
 quite limited to the predominant atmospheric
 species.  What must  be made available  is
 either a set of references for each gas that
 encompasses   a   large   range   of
 concentration-path length products or a set
 of high-resolution spectra  that  can  be
 modified through the mathematics  shown
 here  to account for the nonlinear response.
Then the analysis should proceed by using a
polynomial fit to  the absorbance  versus
concentration.
                                       8-17

-------
                                                                        TR-4423-99-03
       Even  with  a  reasonable estimate of
 the  maximum  concentration-path  length
 product  that  can  be  used  with  some
 assurance  that the  errors  will not be very
 large,   the  implied  maximum   allowable
 concentrations in  terms  of parts  per million
 over the 207-m  path are fairly  small. For
 example,   Table   8-1  shows  that  at  a
 resolution of 4 cm'1 the maximum allowable
 concentration would not really encompass
 the   methane  concentration   variability
 normally seen in the atmosphere. While the
 higher resolutions imply higher  maximum
 values, measurements at coal mines might
 have  high  unexpected errors that are not
 reported at all by the classical least-squares
 technique.  The criterion used to arrive at the
 values in this paper was that the ratio of the
 quadratic term to  the linear term  should be
 less than 0.1. It  is  felt  that this is  not  a
 stringent criterion,  and  that implies that
 higher reported values of methane, even at
 normal atmospheric concentrations, probably
 have some unexpected and unreported error
 associated with them. The magnitude  of
 such  errors  is probably  unknown  to  the
 experimenter.

       The results for ammonia are similar to
those  of  methane   except that  at  the
967-cm"1 peak there does not seem to be as
much effect due to changing resolution.

       Water presents an entirely different
problem to the analysis of FT-IR  data. The
absolute value of the concentration of water
is  rarely,  if  ever,  calculated,   and the
 important concern seems to be whether the
 absorbance  due  to  water  is  adequately
 accounted for. This evaluation indicates that
 analyzing  for  an  actual   value  for  the
 concentration of water may be difficult but
 that this  may not make much difference
 since the variability of the absorbance due to
 water  seems to  be accounted for  almost
 entirely.  Figures  8-15  and  8-16 tend  to
 corroborate these assumptions. Figure  8-16
 plots the value of the  linear regression slope
 when  the  dependent variable is  a water
 spectrum at a partial pressure of 1 5 torr and
 a temperature of  275 K. The  independent
 variables were a set of water spectra whose
 partial  pressures were all 1 5 torr but whose
temperatures varied from 255 to 295 K. The
 expected value of all  the slopes was 1.00.
This curve  implies that  the temperature
difference plays  a major role in the error of
the analysis. If, for example, the calculations
are  repeated for  a  set  of  independent
variables, all at the same temperature but at
changing  partial  pressure,  the regression
slopes  are  all within  a  few percent of the
expected values.
          Partial pressure Is fixed at 15 torr for all spectra
       Field spectrum at 275 K , analyze from 1008 cm  "' to 1019 cm''
       255  260  265  270  275  280  285  290  295
          TEMPERATURE OF REFERENCE SPECTRA (K)
 Figure 8-16. Plot of Regression Slopes Vs.
 Temperature.
                                        8-18

-------
                                                                       TR-4423-99-03
       Figure 8-17 is  a  plot  of  the
 difference of two  spectra. One is  the
 original field spectrum where the partial
 pressure  of  water  is 15 torr at 275 K.
 The reference spectrum was at 1 5  torr
 but at a temperature of 255 K. To obtain
 the plot in  Figure  8-17, the difference
 was found  between the field spectrum
 and the reference spectrum to which the
 regression  results  had been   applied.
 These results show that the difference is
 less than 10%  of  the  original,  thereby
 accounting  for  at  least most  of  the
 absorbance  due to water. Figure 8-17
 also shows  a number  of other  curious
 results. The two spectra were aligned at
 the 1014.5-cm"1 line, but the difference
 clearly   shows   that   the   reference
 spectrum  was somewhat  broader than
 the field spectrum  because of the "W"
 shape of the curve. The line at 1017 cm"1
 clearly shows that these  two  spectra
 were not properly aligned, although they
 were at 1014.5 cm'1. Atthe 1010.7-crrV1
 line, the  difference  indicates  both a
 misalignment and  a  difference  in  the
 widths of the two lines. The discussion of
this chapter does not speculate on why
                Field spectrum at 27S K
        reference spectrum at 255 K regression results applied
 o
 til
 t. 0.001
              1010   1012  1014
                WAVE NUMBER (CM'1)
 Figure 8-17. Difference  After Regression
 Coefficients Have Been Applied.
 that happens, but most workers in the field of
 open-path  FT-IR measurements have  seen this
 occur on many occasions.

 8.6    Conclusions and Recommendations

       The work that has been carried out here
 has been restricted to molecules that have fairly
 narrow absorption  features,  and broadband
 features have yet to  be studied.  It is likely that
 whenever the instrument line function (or slit
 function)  is  broad  in  comparison  with  the
 absorption  feature, the results presented here
 will be noticeable. Because the response of the
 FT-IR instrument is inherently nonlinear but the
 most  commonly  used  analysis  technique
 assumes it to be linear, errors of an  unknown
 magnitude  can occur in the data. These errors
 occur  for  two primary  reasons:  First,  the
 reference spectra that  are  commonly used by
 workers with the FT-IR instrument have not been
taken at the same concentration-path length
 product or  at the same temperature as existed
during  the  field  spectra  acquisition phase.
Second, the differences in the temperatures at
which the reference and the field spectra  were
acquired seem to be more important than was at
first thought. This is clearly true for water.

       From this work, the conclusions that can
 be drawn are as follows.

    •  Errors of unknown magnitude  occur in
       the  FT-IR data whenever  the reference
       spectrum  of  the target gas does not
       have  the  same   concentration-path
       length  product  as  the field spectrum.
       There is probably at least a 10%  error
                                       8-19

-------
                                                                       TR-4423-99-03
       whenever the reference and the
       field   spectrum   differ   in
       concentration-path length product
       by a factor of 2.
    •  Differences in temperature cause
       some further error.
    •  In  all  the  cases  studied  here,
       higher  resolution indicated lower
       predicted errors.
    •  Water,   when   used   as   an
       interfering  species,   does  not
       strongly impact the analysis, even
       when    the    difference    in
       concentration-path length product
       is large. It is anticipated that this
       is true for all interfering species.
    •  The analysis for water directly  is
       limited  by the same difficulty as
       analysis for any other target gas.
    •  It  is  essentially  impossible  to
       determine the absolute  error  in
       concentration calculated for any
       one  field  spectrum.  This   is
       because  the  magnitude  of  the
       error depends on the magnitude of
       the    difference   in   the
       concentration-path length product
       of  the  reference  and  the  field
       spectra.

       Although  this may  seem to be  a
formidable problem for the  analysis of
FT-IR data when the response is assumed
to be  linear,  there are some  ways to
overcome it. The recommendations are
as follows.
                                     •  The combination of  high resolution and
                                        boxcar   apodization   (or   unapodized
                                        spectra) seems to create the most linear
                                        response. Because it  is likely that at
                                        present  few  reference  spectra  using
                                        boxcar apodization exist, they must be
                                        manufactured.
                                     •  A new mathematical analysis approach
                                        that accounts for the nonlinear response
                                        directly during  the analysis  should be
                                        developed.
                                     •  Future databases like the NIST database
                                        should   incorporate   high-resolution
                                        spectra,  along  with the  capability to
                                        create apodized, lower resolution spectra
                                        that account for the inherent nonlinearity
                                        created by the instrument line function.

                                  8.7    References

                                  Filler, A.S. 1964. Apodization  and  Interpolation
                                  in Fourier-Transform  Spectroscopy. J. Opt. Soc.
                                  Am. 54:762-767.

                                  Happ, H.,  and  L.   Genzel.  1961. Interfernz-
                                  Modulation Mit Monochromatischen Millimeter-
                                  wellen. Infrared Phys. 1:39-48.

                                  Norton, R.H., and R. Beer. 1 976. New Apodizing
                                  Functions for Fourier spectrometry. J. Opt.  Soc.
                                  Am. 66:259-264.

                                  Kauppinen, J.K.,  D.  J.  Moffatt, D.G. Cameron,
                                  and H.H. Mantsch. 1981. Noise in  Fourier Self-
                                  Deconvolution. Appl. Opt. 20:1866-1879.
The   use  of  high  resolution
(0.25  cm'1)  makes  the  errors
smaller and manageable.
                                        Marshall, A.G. and FiR. Verdun. 1990. Fourier
                                        Transforms  in  NMR,  Optical,   and  Mass
                                        Spectrometry, Elsevier, Amsterdam.
                                       8-20

-------

                                                                  TR-4423-99-03
Ropertz,  A. 1997.  Kalibrierung Eines
FT-IR    La ng weg a bsorpt i ons-
spektrometeres  in Verbindung Mit Einer
Einstelbaren   I nf ra rot-m u Iti-
reflexionsgaszelle  Und Validerung  Der
Ergebnisse Wahrend Einer Messkampagne
Bei Einer Raffinerie.  Diplomatarbeit im
Fachbereich   Maschienen   und
Vefahrenstechnik an der Fachhochschule
Dusseldorf,  Matrikel-Nr  240415,
Dusseldorf.
Russwurm,  G.M. 1996. Long-Path Open-Path
Fourier Transform Infrared Method Monitoring of
Atmospheric Gases. Compendium  of Methods
for   the  Determination  of  Toxic  Organic
Compounds  in  Ambient  Air—Compendium
Method   TO-16,   EPA/625/R-96/010b,
U.S. Environmental Protection Agency, Research
Triangle Park, NC.
Rothman,   L.S.,   R.R.   Gamache,
A. Goldman, R.A. Toth, H.M. Pickett,
R.L. Poynter, J.M. Flaud, C. Camy-Peyret,
A. Barbe, N. Hussen, C.P. Rinsland, and
M.A.H.  Smith.   1987.  The   HITRAN
Database:   1986  Edition.  Appl.  Opt.
26:4058-4096.
                                     8-21

-------
                                                                       TR-4423-99-03
                                     Chapter 9
                   The Technique of Classical Least Squares
                                     SUMMARY

     This chapter can be considered a tutorial on the use of classical least squares
     analysis as applied to the FT-IR data. The following specific points are addressed.
        •  Least squares analysis for one dependent and one independent variable or
           simply regression analysis
        •  Development of the least squares techniques in matrix terms
        •  Using more than one reference gas
        •  Expansion of the technique to include the quadratic terms
        •  Discussion of the errors as calculated by this technique
9.1    Introduction and Overview

       At  the  time  of  this writing  the
preferred technique for data analysis of the
FT-IR spectra is the method of least squares.
In order for this technique to be used, a set
of reference spectra must be available to the
operator.   These spectra .are acquired  by
using pure samples of gas  under controlled
conditions of pressure and temperature. They
are generally acquired with the gas in  an
enclosed cell  in  a  laboratory.   Once the
absolute pressure of the target gas in the cell
is  adjusted, the  cell  is backfilled to a total
pressure of one  atmosphere by using  an
nonabsorbing gas such as nitrogen, and the
spectrum is acquired. In order to accurately
analyze a so-called field spectrum there has
to be one reference  spectrum for each gas
whose absorbance is contained in the field
spectrum.
       This chapter describes in some detail
the process of the analysis itself, and this
description  is  by  necessity   somewhat
mathematical. Section 9.2 of this chapter
considers the case where there is one gas in
the field  spectrum and only one gas is used
as a reference. The result of this is to derive
the general expressions for the  linear least-
squares fit, or what is generally referred  to
as linear regression. Section 9.3 introduces
matrix terminology and describes how the
analysis  is generalized to  several   gases.
Section 9.4 describes those changes in the
matrix terminology that are necessary  to
include a quadratic term in the equations so
that a nonlinear response can be accounted
for.

9.2    Least Squares Analysis for One Gas

       The least squares technique has been
developed  for  many  years and  is  amply
                                        9-1

-------
                                                                       TR-4423-99-03
described in many texts (Sokolnikoff and
Redheffer 1966; Draper and Smith 1966).
As applied here, it consists of trying to fit the
absorbance  of  a reference spectrum to the
absorbance  measured in a field spectrum in
some so-called  best way. The assumption is
that at each wave number of  the  spectra
there is  a linear relation between the  field
spectrum absorbance and the absorbance of
the  reference  spectrum. This  assumption
arises  naturally from  Beer's   law,  which
implies that there is a linear relation between
the absorbance  and the concentration-path
length product from which the spectrum was
obtained. Mathematically, this linear relation
is  written as A = aCL, where CL is the
concentration-path length product and the
constant  of   proportionality   is  a,  the
absorption coefficient of the gas. This implies
that for a fixed path length L the absorbance
is  a linear function of the concentration C.
Thus at each wave number there is  a linear
relation between the absorbance of the field
spectrum and   the  absorbance  of  the
reference spectrum. If the absorbance of the
field  spectrum  is  considered  to  be  the
dependent variable Y and the absorbance of
the reference the independent variable  X,
then at each wave number a relation of the
form Y = mX + b should exist. The problem
then  becomes  a  mathematical  one  of
determining  the slope m and the intercept b.
As  an  illustration,  consider the situation
where there is  a single  gas in the  field
spectrum, and therefore  only  one  gas is
necessary in  the reference set.    If  the
instrument  is  operating at  a resolution  of
0.5 cm'1, then there is a data point at every
0.25  cm"1. A partial tabular listing  of the
absorbance  data looks like  that given  in
Table 9-1 below.

       The data in this table shows that the
baseline for both the field spectrum and the
reference spectrum is at an absorbance (abs)
of 0.001 and that an absorbance peak exists
at  1000.75  cm'1.  If  a  graph of  these
absorbances is plotted with abs (field) along
               Table 9-1.  Partial Listing of Spectral Absorbance (abs) Data
       Wave Number                 abs (field)                 abs (reference)
1000.00
1000.25
1000.50
1000.75
1001.00
1001.50
1001.75
0.001
0.050
0.085
0.100
0.090
0.055
0.030
0.001
0.010
0.017
0.020
0.018
0.011
0.006
                                        9-2

-------
 TECHmlmm
                                                                       TR-4423-99-03
the Y-axis (dependent variable) and the abs
(reference)  along  the  X-axis  (independent
variable), then the linear  relation  becomes
apparent.   In the  absence of noise in the
spectra, the intercept should be 0 according
to Beer's law, but because of noise it is not,
and  the  individual  data points  do not lie
precisely on a straight line.

       The question then  becomes what is
the "best" straight line that can be fit to the
data  points.  The   classical  least  squares
technique defines what is meant by the term
best, in that it requires that the sum of the
squares of the residuals be a minimum. Once
the slope and the intercept are determined,
then  a Y value  can be calculated  for any
point X from  Y = rnX + b.  The X values are
chosen to  coincide with  the  actual data
points along  the   X-axis,  and  then  the
difference between the calculated Yand the
actual 7data value is found.  This difference
is  the residual. The individual residuals are
squared and then summed, and the best fit is
found when this sum is a minimum.

       This procedure gives rise to the term
least squares. The minimization process  is
completely transparent to  the user, and he
does not have to make  small adjustments to
the  line  and  recalculate  the sum of the
residuals squared  and  then  select the one
slope and intercept that gives a minimum
result. The process does that automatically
because the equations used already make the
sum a minimum.

       The  situation is shown schematically
in  Figure 9-1. The original data points are
                         residual  {
                  0.02
               ABSORBANCE (REFERENCE)
 Figure 9-1. Least Squares Fit of a Data Set.
 The  squares represent actual data but do
 not fall on the  regression line.
shown as squares in the figure along with
the best fit line.  The  residual at one data
point (0.04,0.045) is also depicted.

       As a demonstration of how the usual
least squares  equations for this  simplified
case arise, consider the following. Suppose
that a field spectrum has been acquired and
that the  analysis region covers a range  of
discrete data points (total N) along the wave
number axis.  The reference spectrum has
the same number N oi data points along the
wave number  axis. A plot similar to that in
Figure  9-1  is made, and the problem is  to
determine the slope m and the intercept b for
the best fit straight line between the two
sets of spectral data. Let the equation of the
best fit line be given as y = mX+b and let the
original data points in the field spectrum be
at  (Xj , Yj. At  each  data point (X-t ,  Yj,
calculate a new y from the equation such
thaty, = mXj + b.  Then form the residuals at
each point such that  /?,  =  Y,  -  yf  =  Yt -
(mXj + b). The individual residuals are then
squared and summed over all the data points.
This leads to the expression
                                        9-3

-------

                                                                       TR-4423-99-03
 where the sum is taken over all the data
 points from 1  to N.  In order to minimize S,
 the derivative of  S with respect to both m
 and b must be found and each set equal to 0.
 The derivatives are each given below.
                                   (9-2)
 dm
and
                                            This expression for b can be substituted into
                                            Equation 9-3, which can then be solved for m
                                            to get

                                             (9-7)
         It should be noted that in Equation 9-6 the
         terms
  c&
                                   (9~3)
                                           and
Equations  9-2  and  9-3  are  a  set  of
simultaneous equations that can  be solved
for m the slope and b the intercept. These
expressions can be rewritten as
  m
and
(9-4)
                                  (9-5)
The term  ££ is equal to Nb,  where N is the
total number of data points, so Equation 9-4
can be solved for b as
                         N
                                           are just the average values of the X's and the
                                           Fs, so Equation 9-5 becomes
                                                                              (9-8)
               Equations 9-6,  9-7, and 9-8 are the
         usual  equations that are used for  doing  a
         linear  regression between two data sets. As
         applied to the spectral data from the FT-IR,
         the  X's   in  these  equations   are   the
         absorbances at the various wave numbers for
         the reference spectrum,  and the Fs are the
         absorbances of the field  spectrum.

               As  a numerical  example, the data in
         Table  9-1  can be used. There are a total of
         seven  data points, so N = 7. There are really
                                        9-4

-------
                                                                      TR-4423-99-03
only  four  quantities   that  have  to  be
calculated. They are the sum of the X's, the
sum of the Fs, the sum of Byproducts and
the sum of the X's squared. From the data in
Table 9-1, these quantities are: £X = 0.083,
£y= 0.411, £*y = 0.006351, and OT2  =
0.001271.  Substituting  these  values  into
Equation 9-7, the slope is obtained as
      m =
      0.006351-(0083)7(0411)
                           = 5.151
The  intercept  can be calculated next from
Equation 9-8.  From  the  above  data the
intercept is
                 N
      0.411  (5.15l)(0.083)
                        = -0.002367
      When the  results of this calculation
are applied to the  original data  to determine
the atmospheric concentration of the gas,
the concentration  product of the reference is
multiplied by the  slope and then divided by
the actual path length used to acquire the
field  spectrum.   So  if   the  reference
concentration-path   length   product  is
20  ppm-m and the actual path length used
was  200  m,   then  the   atmospheric
concentration is 20*5.15/200  =  515 ppb.
 9.3    Matrices

       The  technique  outlined  above  is
 limited to a single gas in the field spectrum
 and  therefore  uses  a  single  reference.
 Continuing on to the more realistic case of
 several gases and several interfering species
 is a tedious  if not impossible  task  in the
 method described  above, and  some  other
 method of describing the process has to be
 found. Fortunately, there is such a method,
 and it involves the use of  matrices. Thus a
 short tutorial  session on the description and
 manipulation  of matrices is presented next.
 Many excellent texts  (Hohn 1964; Belman
 1970)  cover  the mathematics of matrices,
 and  the  interested reader should consult
 them for more in-depth descriptions  of the
 use of  matrices.

 9.3.1  Matrix Types

       A  matrix  is  simply  an  array  of
 numbers.  The array can have i rows and j
 columns.  If the number of rows / equals the
 number of columnsy, the matrix is said to be
 square. The convention used in this tutorial
 will be to designate a matrix by an italicized
 bold letter. A square  matrix A  that  has  3
 rows and 3 columns is shown below
            A =
                       a
                        n
where  the  elements  of  the  matrix  are
designated by the subscripts ij. The element
a,, is the  element  in the first row and the
first  column,  while the element a32  is the
                                        9-5

-------
MAIWl
           ..
         ;=•••;:=••= =
                                                                      TR-4423-99-03
 element in  the  third row  and  the  second
 column. The numerical elements of a matrix
 will always be  enclosed  by  the  square
 brackets as shown.  A matrix can have only
 1  row as A = [<2,, a,2 an • • • a{j ] or it can
 have only 1 column as
                     a
                      11
                     a
                      21
                     a
                      31
                     a,-
 In these cases the matrices are called either
 a row vector or a column vector. A general
 matrix of / rows andy columns can be written
 as A = [a,y] or just simply  as A. There are
 other matrices that are necessary for  this
 work, and they  will be introduced at  the
 appropriate time.

      Matrices     arise    naturally   in
 mathematics from the study of the solutions
 to  simultaneous   equations,  and  they
 represent  a concise way to manipulate  the
 coefficients of the equations. For a  pair of
 simultaneous equations
           aQX
           b0X
                       =  c
there are a number of matrices that can be
written  to  describe  these  equations. The
coefficient  matrix is shown below as the
[2x2] matrix
             c =
                   a0  a,
                                          and the augmented matrix shown below as
                                          the [2x3] matrix
                                                    A =
                                          9.3.2  Some Matrix Properties

                                                Matrices    have   a   number   of
                                          mathematical properties that allow them to
                                          be used  for obtaining  the  solutions of
                                          algebraic equations. They can be added and
                                          subtracted from one another so that if A =  [
                                          a{j ] and B = [by ], then A ± B = [a&. ± bv ]. A
                                          matrix A  can be multiplied by a scalar k,
                                          (simply a number) such that if A = [a:j ], then
                                          kA  =  [ka^].   These   properties   are
                                          demonstrated with actual numbers below.  If
                                                       A =
                   2   1
                   5   12
                                           and
                                                         B =
then
                                                             4  9
                                                             3  8
                                                              6   16'
                                                              8   20
                                         and
                                                    A-B =
                    -2   -2
                    2    4
                                       9-6

-------

                                                                      TR-4423-99-03
Also, for multiplication by a scalar, say k =
3, then
                     ~6   21"
                      15   36_

Division  of  a  matrix by a  scalar is also
possible.
                                                C =
                                                        #22^2!   021^12 + ^22^22
       Multiplication  of  one  matrix  by
another is allowed, but it is not possible to
divide one matrix by another. Multiplication
of one  matrix  by  another  follows certain
mathematical rules, and these are described
in the next section.

9.3.3  Multiplication of Matrices

      The    description   of    matrix
multiplication begins  with  a discussion of
how  to  multiply  two  square   matrices
together. Suppose that two matrices, A and
B, are each [2x2] matrices and they are to be
multiplied as C — AB. Then if
             A =
                  a
                    n
                        a
                         l2
and
             B =
                   n    n
                  b2l   b22
The first thing to note here is that each of
the elements of the product matrix is made
up of a sum of products of the original matrix
elements. Each of the product elements is
found by multiplying a row element of A with
a  corresponding  column  element  of  the
matrix B and then  summing the individual
terms. So, the elements of row  1  of the
matrix A are multiplied by the elements of
column 1 of the matrix B, and  then the
individual   terms are summed to  get the
element (1,1) of the product matrix C. The
element  (1,2) of the product matrix C is
found by using the elements from row 1  of
the matrix A and the elements from column 2
of the  matrix B. Thus the subscripts of the
product  matrix  elements   provide   the
instructions of which row and which column
to use from  the  original  matrices.  The
numerical examples above for the matrices A
and B can  be used to  demonstrate this
multiplication.  Thus
then the product C = AB is given by
                                        9-7

-------
                                                                       TR-4423-99-03
                                            then the result is a [2x2] matrix C such that


"2 7"
5 12
"8+21 18+56"
20+36 45+96
"4 9"
3 8




"29 74"
56 141

i
"4 9"
3 8
"2 7"
5 12
       However,  an important  point about
matrix  multiplication  is  that  it  is   not
commutative. That is, the product AB does
not equal the  product BA.  This is clearly
shown in the example below (again using the
same matrices  as above).
     Cf = BA =
      8+45   28+108]  f53  136'
      6+40   21+96   =  46  117
       Matrices need not be square in order
to multiply them together. For example, a
matrix with 3 rows and 2  columns can be
multiplied by a matrix that has 2 rows and 3
columns.  An  example is  shown  below.
Suppose that
         A =
an   al2   al3
a2l   a22   a23
and
             B =
    \\    \2

  b2\   b22
                   '31
        '32
                              \ A I + G\2b2\

                              2\b\ 1 + a22b2\
                                                             3\  °2\b\2
 If the product BA is found, then the result is
 a  [3x3]   matrix.   Thus  the   order   of
 multiplication is again important,  and one
 generally speaks of either pre-multiplying one
 matrix by  another  or  post-multiplying one
 matrix by another.  For the product AB, one
 says that A pre-multiplies B, while for the
 product BA the matrix A post-multiplies the
 matrix B.   In general a  matrix of order (i,f)
 (one that has / rows and/ columns) can pre-
 multiply a matrix of order (&,/) only if j = k.
 This gives  rise to a matrix of order (/,/) or a
 matrix with / rows and / columns.

       To solve the matrix equations that are
 used for the least squares analysis of FT-IR
 data, several further definitions are required.
 They  include  the  identity  matrix,  the
 transpose of a matrix, cofactors of matrices
 or submatrices, and the inverse of a matrix.
 Also  required   is  a  discussion  of  the
 determinant of a matrix.  Rather than provide
 a  purely  mathematical  definition  of the
 determinant,  several  examples  are  used,
 which  provide the  information  about  them
that is  required for the present discussion.

9.3.4  The Identity Matrix

       The identity matrix/is simply a matrix
that has all ones along the principal diagonal
of  the matrix  and  zeroes   in  all   other
                                        9-8

-------
                                                                       TR-4423-99-03
 elements.  Thus a [3x3] identity matrix is
 written as
                   1  0   0
                   0  1   0
                   0  0   1
The principal  diagonal  of  a  matrix  is the
diagonal going from the upper left element to
the lower right element. The usefulness of
this matrix is explained below.

9.3.5  The Transpose of a Matrix

       The transpose of a matrix A is written
as  AT and  is  that   matrix  formed  by
interchanging the rows  and the columns of
the matrix A. So if the  matrix A is a [3x3]
matrix written as
          A =

-------
 TECHmdl^m
                                                                      TR-4423-99-03
       The determinant of a [2x2] matrix is
 found by multiplying the elements  of the
 principal diagonal together and subtracting
 from that product the result of multiplying
 the elements of the secondary diagonal. The
 secondary diagonal   of  a  matrix  is  that
 diagonal running from the upper right to the
 lower left of the matrix.  Thus if the [2x2]
 matrix is given by
             A =
                  #n   #12
                   42\   "22
the determinant is given by the expression
det(A)  =  aua22  ~  3i2a2i-  A  numerical
example of this is given below. If the matrix
A is
              A =
1   4
8   3
then  the  determinant  of the matrix A  is
detW) = (7K3) -  (4)(8) = -11.

       In order to find the determinant of a
larger, say [3x3], square matrix, the problem
must always be reduced to one of finding the
appropriate [2x2] matrices by using the rules
to be  discussed. To obtain the proper [2x2]
matrices from the original [3x3] matrix, the
elements of the top row  are each used in
turn.  The row and the column that element
is  in  is then  eliminated  from the original
matrix.  The [2x2] matrix that  remains is the
one that  has to  be used. To complicate
matters a little, the [2x2] must be multiplied
by  the matrix element   whose  row and
column were  eliminated.  So, if the  [3x3]
matrix is given by
                                 A =
                flf
                 31
                                                  #
                                                   33
                        then the determinant is given by
                        det(/l)  -
       It is important to note the change in
 sign from a plus to a minus in the second
 term in the determinant. The first term of the
 determinant is formed by starting with the
 element [1,1], then crossing out the first row
 and the first  column and then  finding the
 determinant  of the [2x2]  cofactor  of the
 element [1,1].  The second term  is found by
 starting with  the  negative of the element
 [1 ,2], then crossing out the first row and the
 second column and using the determinant of
 the cofactor of the element [1,2]. The  third
 term is found by starting with the element
 [1 ,3] and then crossing out the first row and
 the third column to get the determinant of
 the  [2x2] cofactor  of  the element [1,3].
 Whether the  sign  in front of any particular
 term is .positive or negative is  found by
 actually using the  expression (-1)l+; where
 the / and the j are the element subscripts.
 Thus for the first term the element subscripts
 are 1 , 1 and  1  + 1  =2; therefore (-1 )2 = 1 ,
 For the second term, however, the / and the
j subscripts total  3  and (-1)3 = -1.  The
 subscripts for the third term again yield a
 + 1  multiplier.  A  numerical  example of
                                       9-10

-------

                                                                       TR-4423-99-03
 calculating the determinant for a [3x3] matrix
 is given below.  Let the matrix A be given by
 the original matrix. Thus if the matrix A is
 given by
                   3   9   1
                   675
                   12  2   8
       The  determinant  for  this matrix is
found following the rules above as

 det(/l)=3[(7)(8)-(5)(2)]-
       = 3(56- 10]- 9[48- 60]+ 1[12- 84]
       = 138+108-72
       -174

       In general the determinant is found by
using all the elements along the top row as
described above and then continuing down to
the  [2x2]   matrix  level.  This  procedure
becomes very cumbersome very quickly, and
for a [4x4] matrix the determinant has four
terms  that  each multiply  a [3x3] matrix.
Each of these [3x3] matrices have a three-
term determinant, so that for a  [4x4] matrix
there are 1 2 terms to calculate to obtain the
determinant.

9.3.7  Cofactors of Matrices

       A cofactor of the [i,j] element of a
matrix  is the determinant of the matrix that
remains after the fth row and/th column are
removed from the original matrix. Just as for
the determinant,  each cofactor has to be
multiplied by the term (-"\Y+J. There can be
as many cofactors as  there  are elements in
          A =
 and the row and the column that contain the
 element a,, is eliminated from the matrix, the
 cofactor that remains is the determinant of
 the [2x2] matrix Z)n, given by
                      a22   a23
                      a32   033
 Or,  the  cofactor  of  the element  [1,1] is
 det(Du)  = a22a^ - a23a}2.

       By  using the  same  procedure,  the
 cofactor  of  the  element a22 is given  by
 det(.D22) and is seen to be
 det(£>22) = det
                a,
a,
                      13
                a.
                 31
a
                      33.
where again  the  -1  multiplier is a  plus
because (-1)2+2 =  +1. These cofactors are
important for the discussion of the inverse of
a matrix that follows below.

9.3.8  The Inverse of a Matrix

       Division of one matrix by another is
not allowed, but there is  a process that is
similar to a division, and it is called inversion.
The inverse of  a matrix A is simply written as
A"1.  The principal property of the inverse is
                                        9-11

-------

                                                                       TR-4423-99-03
 that the result of multiplying a matrix by its
 inverse yields the  identity matrix.  That is,
 given the matrix A and its inverse A'1
                  =04-')04) =
                    det(X)
 Thus for the [2x2] matrix
       There   are   two   very  important
 properties that a matrix must have in order
 for an inverse even to exist.  First, the matrix
 must be square; it must have as many rows
 as it has columns. Second,  the determinant
 of the matrix cannot be equal to zero.  In a
 very  loose sense, this  merely implies that
 dividing by zero is not allowed.  Nonsquare
 matrices  must  first  be made into a square
 matrix before the inverse can be found. This
 can be accomplished by simply multiplying a
 matrix by its transform (explained below).

       Since this chapter is meant to be  a
 simplified discussion of matrices and  how
 they are applied to the classical least squares
 analysis of FT-IR data, a definition, without
 proof, and a numerical example of finding the
 inverse will be sufficient.

       In  order  to  find  the inverse of  a
 matrix, one first has  to find the determinant
 of the matrix as defined  above. Then  for
 each element of the  matrix,  the  cofactor of
that element of the original matrix has to be
found. The transform of the matrix formed by
the cofactor matrix must then be found by
interchanging the rows and the columns. If
the cofactor matrix is given by C, then the
transform is given by CT, and the inverse A"1
of the [2x2] matrix A is then given by
             A =
 an   al2
 a2l   a22
the cofactor matrix is seen to be
                  a.
            c =
 '22    "21
-an    an
The transform of this matrix is given by
           CT =
                   a22    -an
                  -a
                    2l
and  the inverse of the matrix A  is finally
found as
                  1
               det(/4)
      a22    -al2
      -a2l   au
       For  a  numerical  example,  let the
matrix A be
             A =
  4   9
 -2   -5
The cofactor matrix is given by
                   -5   2
                   -9   4
                                        9-12

-------
                                                                       TR-4423-99-03
 and the transform of the cofactor matrix is
             CT =
                   -5  -9
                    2    4
       The determinant of the matrix A is
 easily found to be -2. Thus the inverse of the
 matrix A is given by
"-5
~^2
2
.-2
-9"
_2
4
^2.



2.5
_ i


4.5"
-2

      /*-' =
From here it is an easy matter to show that
this last  matrix is indeed the inverse of A
simply by multiplying A and A"1 together to
get the identity matrix.  Thus

AA'l =

"4 9"

-2 -5
"10-9 18-18"
-5+5 -9+10
"2.5 4.5"

-1 -2


"1 0"
0 1
       At this point all the tools  that  are
required  to  continue  processing the  FT-IR
spectra data are available in matrix form.

9.4    Matrices and Algebraic Equations

       The utility of matrices can easily be
demonstrated by considering the solution for
a  set  of simultaneous equations.  As  an
illustration,  consider  the  following set of
equations
              cX+dY = k{
 To solve these equations for X and Y, the
.first equation can be solved  for Xto get
                                                         X
                                                              kn-bY
                                                                 a
                                           This  can be  substituted into  the second
                                           equation to get
                                                      ckn-cbY
                                                               •+dY= k,
                                                          a
                                            which when solved for /yields
                                                             ak, - ckn
                                                                io
                                                              ad-cb
                                            Then  by using  the expression for Y and
                                            solving for X, it is seen that
                                                        X =
                                                              ad- cb
                                                  To solve these equations by  using
                                           matrices, the equations are first written in
                                           matrix form so that
'a b
c d_
' X
Y
                                           This expression is seen  to  be a  column
                                           matrix that is multiplied  by the coefficient
                                           matrix,  and that product  is equal to the
                                           column matrix of the constant terms. It can
                                        9-13

-------
          .'«.;r./ = .:r =
                                                                        TR-4423-99-03
 be rewritten in purely matrix format as CV =
 K, where the  C represents the coefficient
 matrix, the  V represents the variable column
 matrix, and the K  represents the column
 matrix of the constants. To solve this for the
 variable matrix, the  coefficient matrix must
 be removed from the left-hand side  of the
 equation.  This  is easily  accomplished  by
 multiplying both sides of the equation  by the
 inverse of the  coefficient matrix  C"1  to get
 C'CY  =  C[K.  Then, after  noting that  a
 matrix  times its inverse is just the identity
 matrix /, the final expression for the variable
 matrix is obtained as V = C{K. Thus in order
 to solve this equation for the X and the Y, the
 inverse of  the coefficient matrix must  be
 found.  In reviewing  the rules for finding the
 inverse of a matrix given in Section  9.3.8, it
 is seen  that  in order to find the inverse of the
 coefficient  matrix,  one  has to  obtain  its
 determinant and then form the transpose of
 the cofactor matrix.  The determinant  of the
 coefficient  matrix is just the term  ad - be.
 The cofactor matrix is found to be
                 d   -C
                 -b   a
The transpose of this is

                 d   -b
                 -c   a
and the inverse of the coefficient matrix is
just   this   transpose   divided   by  the
determinant, or
          C"1 =
                   1
                ad -be
d  -b
-c  a
 Now the final matrix expression becomes
'X
Y
1
ad -be
' d -b
-c a
"*0~
A
 By multiplying  out the matrix terms on the
 right-hand side of  this equation, the final
 result is obtained as
' X'
Y
1
ad -be
dk0 -bk{
-ck0 + akl
The X  term is  the  upper element  of the
matrix divided by the determinant, and the Y
term  is  the lower element  of  the  matrix
divided    by  the   determinant.   These
expressions are  seen to  be identical to the
expressions derived  above by the algebraic
substitution method.

9.5   Least Squares and Matrices

      The next  step in this discussion is to
show that the matrix technique allows the
same equations  for the regression analysis,
which was initially presented, to be derived.
Note that in that discussion a linear relation
was assumed between the absorbance of the
field spectrum and the absorbance  of the
reference spectrum so that at each particular
data point (or wave number) an expression of
the form Y = mX  + b could be used.  In terms
                                        9-14

-------
                                                                       TR-4423-99-03
 of  the  spectra,   the   Y represents  the
 absorbance  of  the  field  spectrum  as  a
 function of  wave  number. Therefore,  the
 matrix that represents this absorbance  has
 one element for each data point over  the
 range of the analysis. It is represented by a
 column matrix given by
                      a
                       N
The FT-IR usually acquires data at a rate so
that there are two data points per resolution
unit. Thus if the instrument is operating at a
resolution of  1  cm"1 there is one data point
every  half wave  number. It  is  best,  for
statistical reasons, to use about  80 data
points  in the  analysis, so the matrix above
would  have 80 elements and N =  80, and
the wave number  range for  the  analysis
would  cover  40  wave  numbers at  1-cm'1
resolution.

       In the case of the FT-IR, the variables
that  are  to be calculated are actually the
slope m and the intercept b.  Therefore, the
variable matrix  Vis given as
       The  matrix that  is  used  for  the
 absorbance of the reference spectrum is  a
 two-column matrix that will also have Nrows
 to  match  the  absorbance of  the  field
 spectrum matrix. This matrix is represented
 by AR and is given by
                     m
       This matrix has one column that is all
 1 's because  of  the intercept term b that
 arises  from the  linear fit.  The aR's are the
 absorbances  of  the reference spectrum at
 the individual data points.  They must  be
 chosen to be  at the same wave numbers as
 the data in the  field  spectrum, and  there
 must be an equal number of data points Win
 each   matrix.   The  fundamental  matrix
 equation that has to be solved is then AF =
AKV +  e. The e is a matrix that describes the
 errors in the data. Without the error term all
the data would fall on a straight line, and the
 least  squares process has  essentially  no
 meaning. The e matrix is the matrix that
 represents the residuals, and that can  be
solved for and then squared  to get the sum
of the  residuals  squared. This gives rise to
the expressions below.
                                       9-15

-------

                                                                     TR-4423-99-03
   STS = (AF-ARV}T(AF-ARV)
        = ATFAF-ATFARV-(ARV)T AF-
         (ARV)TARV

       Although it will not be  carried  out
 here, when this expression is differentiated
 with respect to V and set equal to zero,  and
 after making use of the fact that
            (AB)T  =  BTAT
 twice the primary matrix equation that must
 be solved is obtained. That is,
The terms in the variable matrix  V, namely
the intercept b and the slope m, are the terms
that have to be  found and  are the  best
estimates of the slope and the intercept. In
order to solve for the b and the m,  the matrix
AR must be removed from the right-hand side
of the equation. The way to do that is, just
as before, to pre-multiply both sides by the
inverse  of  the  matrix  AR. But there  is a
problem here because only square matrices
have inverses, and the matrix AR is decidedly
not square. It has  at least 80 rows but only
2 columns. So the first step is to make the
matrix  AR square   by multiplying it by its
transpose
At this point the fundamental matrix equation
that needs to be solved is
                                                                             (9-9)
                                           This is, in reality, a fairly simple equation and
                                           comprises only two spectra, the measured
                                           absorbance spectrum AF and  the reference
                                           spectrum  AR.  The   following discussion
                                           describes how these two matrices  must be
                                           manipulated to derive the expressions used
                                           for normal linear regression  calculations.

                                                 The  matrix   that   represents  the
                                           reference spectrum AR requires the most
                                           manipulation, and although  that chore may
                                           seem formidable, it  is seen that  the task
                                           quickly becomes fairly simple. The first step
                                           is to multiply the AR matrix by its transpose in
                                           order to get a square matrix. The transpose
                                           of AR is given by
                                                  1    1
•  •  •     1
•  •  •  a
                                                                      R(N-\)
                                          and it has to multiply the AR matrix in the
                                          following way
                                                                          1   °R\
                                                                          1   QB->
                                                  11...    1     1
                                                  J«l  aRl  *  *  «  *«AM)  aKN
              t   •
              •   •
              •   •
                                                Now, utilizing the rules set out above
                                          about the multiplication of matrices,  it is
                                          seen that this product will result  in a [2x2]
                                          matrix.  Before writing down the result, it is
                                       9-16

-------
       . tss : frTsJ -
       i 'L-J -.Ai • r
                                                                       TR-4423-99-03
 instructive to first contemplate the individual
 elements in the product matrix.

   •  For the element [1,1 ] it is seen that the
      first  row  elements  of  the transpose
      matrix must multiply the  first column
      elements of the original matrix and then
      be  summed.  But these elements are
      just all 1 's and when summed yield the
      number N,  the total  number  of  data
      points.
   •  To  obtain  the element  [1,2] the first
      row elements of the transpose matrix
      must  multiply  the   second  column
      elements of the original.  But this just
      turns into the sum of the absorbances
      of the individual data points from  the
      reference  spectrum.
   •  The element [2,1 ] of the product matrix
      is  again the  sum  of  the individual
      absorbances because  it is found by
      multiplying the elements of row 2 of
      the  transpose  by  the  elements of
      column 1 of the original  (all 1 's).
   • The  fourth  and final element  of the
      product matrix (element [2,2]) is found
      by multiplying the elements from row 2
     of the transpose by the elements of
     column 2 of the original  and summing.
     These individual terms are  seen to be
     the squares of the absorbances at the
     individual data points, so the element
     [2,2]  becomes  the  sum  of  these
     squares.

The result of multiplying the matrix AR by its
transpose becomes
                          I an
 Next, the inverse of this matrix  has  to be
 found, and this is done in three steps. The
 first is to find the cofactor matrix, the second
 it to take the transpose of that, and the third
 is to divide  the  individual elements by the
 determinant  of  the  original  matrix.  The
 cofactor matrix is
                    IK)3
                                  N
Because the elements [1,2]  and [2,1]  are
identical, the transpose of this matrix is  the
same as the original.  The determinant of  the
original  matrix is found to be
The inverse that is required here is just the
cofactor   transpose   divided   by   the
determinant, so that
                                  (9-10)
      The next step is to multiply the matrix
that represents the absorbance in the field
spectrum by the transpose of the reference
spectrum matrix, that is, to find the product
                                       9-17

-------

                                                                       TR-4423-99-03
 This product is
far =
       1    1   •  t  .    '1
am  aR2  *  •  »  flfl(AM)
                                    "F\

                                    aF2
                                    t

                                    *

                                    *
       Here  again,  a  little  inspection  is
helpful. The result of this product will be a
column matrix with only two rows. The first
element is made by multiplying the first row
of the transformed reference matrix by the
elements  of the field spectrum matrix.  But
that is  just  the sum of the absorbances at
the individual data points. The second row is
obtained by multiplying the second row of
the  transformed  reference  matrix by the
elements of the field spectrum matrix. This is
again a sum,  but now the terms  to  be
summed  are  the  products of  the two
absorbances such that each individual term
has the form
       It should  be  clear now  why there is
the requirement that the number of data
points  in  the  field  spectrum  must  be the
same as the number of  data  points  in the
reference  spectrum. If they were not, then
this last matrix multiplication  could not be
done.  At  any rate, this product is given by
                                    The final step in obtaining the variable matrix
                                    is then multiplying the two matrices from
                                    Equations  9-10  and   9-11   together.
                                    Remember here  though that  the order  of
                                    multiplication is important. When this  final
                                    product is performed,  the result is
                                            v = -
                                                                            y
                                                                            .LI aRiaFi\
                                    or
                                    v' =
                                   When these results for the intercept b (the
                                   element  [1,1] of V) and the  slope m (the
                                   element  [2,1] of V) are compared to the
                                   results derived from an algebraic standpoint,
                                   it is seen that the expressions are the same.

                                   9.6    Expansion to Many Gases

                                          Once the expressions for the  slope
                                   and the intercept are obtained as above, it is
                                   fairly simple to see  how  to  expand this
                                   solution  to  analyze   for   several  gases
                                   simultaneously or to analyze for one gas  in
                                   the presence of  several interfering species.
                                   The fundamental matrix equation  stays the
                                   same. That is, the expression
                                           always  stays the same,  but  the  individual
                                  (9-11)   matrices  V and  AR  have to  change to
                                           incorporate more information.  In most cases
                                           the problem is  formulated a  bit differently
                                       9-18

-------

                                                                      TR-4423-99-03
 than was used above, in that the intercepts
 are actually ignored. When several gases are
 to be  analyzed  for, the  problem is usually
 stated   by  considering  the   final  field
 absorption  spectrum   to  be   a  linear
 combination   of  the  individual  reference
 absorbance  spectra.  The  mathematical
 statement is that
and  the  problem now  becomes one  of
determining the individual mf terms.  These
nij's  are  the coefficients for  the  individual
references.  In  the mathematical  equation,
they determine the amount of each  of the
individual references to use to achieve the
final result of matching the field spectrum.

       Thus  the  expanded  AR  reference
matrix changes from a two-column matrix,
with the  first column being 1 's, to a  matrix
with N columns, but there are no 1 's. That
is,
          a
            n
a
                 22

Here there are j gases  and N absorbance
values  for  each gas. Typically, there  are
perhaps six or seven gases that are analyzed
                           for simultaneously,  but there can be many
                           more, and there are cases where as many as
                           30 have been targeted.  The final maximum
                           number of  gases  that  can  be  used  is
                           determined by how large a matrix is allowed
                           by the computer and the algorithm being
                           used. The final  variable  matrix V is then a
                           one-column matrix withy rows that represent
                           the m's as given above.  That is
                                                                m,
                                                mj
                                 It is now easy to see how to expand
                           the least squares technique even further to
                           account for  a  quadratic fit  such  as is
                           indicated by the development in Chapter 8.
                           In that case the AR matrix needs to contain
                           terms  that  are
                           absorbances, or
                  the  'squares   of  the
                                                             a
                                                               Ri
and for each addition column in the reference
matrix, there  has to be one  additional m in
the  variable   matrix  that  needs  to  be
determined.

      Calculating anything but the simple
case that is presented above  is not really
possible, nor would  it  be  overly instructive.
However, it is of interest that the absorbance
due to  water actually contributes  to the
continuum, and that makes up the baseline
as seen in the single beam spectrum.  The
                                       9-19

-------
                                                                       TR-4423-99-03
 water concentration can and does change in
 time rather dramatically, and therefore there
 are  changes in the baseline.   If the water
 absorbance is measured from the baseline to
 a peak,  an error will occur because of the
 continuum, which is  already at a lower
 transmission than 100% by a few percent.
 The work in Chapter 8 indicates that the
 absorbance in the water continuum actually
 rises as the square of the partial pressure of
 water. This seems to indicate  that a square
 term ought to  be included in the reference
 matrix for the  water. At the  present  time
 none of  these considerations  have been
 tested with spectra to determine whether the
 analysis  results are better  than  the current
 methods. Also, only two software packages
 that include that  term (the Nicolet  Omnic
 software  [Madison, Wl]  and  the  MIDAC
 AutoQuant software [Irvine, CA]) seem to be
 available at the time of this writing.

 9.7    Least Squares Errors

       Once  the   coefficients  for   the
 individual gaseous components  have been
 determined from the least squares technique,
 the uncertainties in these  coefficients must
 be  examined  if  the  uncertainties  in   the
 concentrations  are to  be understood.  The
 overall derivation of the expressions for the
 errors  and a thorough examination  of  the
 errors can be quite involved, and an in-depth
 discussion is felt to be beyond the scope of
this work.  The interested reader is directed
to two excellent references  (Draper  and
 Smith 1966; Bevington 1969) for an in-depth
 analysis of these questions.
       However, it is important to note here
 that  the errors  calculated  by  statistical
 techniques are associated with only the least
 squares   technique,  and   they  are  not
 indicative of the overall error  in the FT-IR
 data.  The expressions for determining the
 errors are  derived  from a  study  of  the
 residuals or the  analysis of variance. These
 errors are related to the amount  of  the
 variance in the original field spectrum that is
 explained by the variance in the sum of the
 individual reference spectra, and these errors
 have more to do with how well the sum of
 the reference spectra match the original field
 spectrum rather  than any absolute error.

       While these errors are important to
 the overall discussion of the FT-IR data, they
 do  not comprise the entire  picture  of  the
 errors. The errors generally reported by the
 least squares technique do  not take into
 account  the  actual  magnitude of the real
 errors in the data  because they  do  not
 account for several major contributors to the
 error. These unreported errors are errors like
the nonlinear response error produced by the
 apodization  function or  the temperature
 effect on the reference spectra, as described
in Chapter 8 of this document. They also do
not  include  any  errors in   the  reference
spectra themselves.

       Any study of these unreported errors
shows that  they can be quite  large when
compared to the errors calculated from  the
least squares technique.  Because of this, it
is felt that no claim for accuracy better than
about  ±30%  can be made  for  the  FT-IR
                                        9-20

-------
                                                                    TR-4423-99-03
data,   regardless   of   what   the   error
calculations of the least squares technique
produce.

9.8   References

Belman, R. Introduction  to Matrix Analysis
2nd ed. McGraw Hill Book Company,  1970.

Bevington,  P.R. Data Reduction and Error
Analysis for the Physical Sciences. McGraw
Hill Book Company, 1969.
Draper,   N.R.   and  H.  Smith.  Applied
Regression Analysis. John Wiley & Sons Inc,
1966

Hohn, F. E. Elementary Matrix Algebra 2nd ed.
The Macmillan Company, New York, 1964.

Sokolnikoff, I.S., and R.M. Redheffer. 1966.
Mathematics  of   Physics   and  Modern
Engineering.  McGraw-Hill Book  Company,
New York.
                                      9-21

-------
                                                                      TR-4423-99-03

                                    Chapter 10
                     Quality Assurance and Quality Control
                                     SUMMARY

          The topics and specific points of emphasis discussed in this chapter include
    the following.

          •   The need for a quality assurance (QA) project plan
          •   QA project plan categories and relevant EPA documents
          •   A general format for a 1 6-point QA project plan
          •   A  discussion  of specific  quality  assurance  and quality control
              (QA/QC) issues related to FT-IR long-path, open-path monitoring
          •   Portions of an approved QA project plan
          •   A  case study  presenting  QA  data collected over a two and one-half
              month period
          •   Recommendations of procedures to be included in a QA program
10.1   Introduction and Overview

       For open-path FT-IR spectrometry to
become   an   accepted   method   for
environmental monitoring,  QA  procedures
must be established. While QA issues have
been   addressed  (Kricks  et  al.  1992;
Russwurm  1992a,b; Weber  et  al. 1992;
Kagann et  al. 1994), there is currently no
consensus   regarding   the  proper   QA
procedures required  to  validate  open-path
FT-IR  data.   In fact,  there  is  very  little
information in the literature that addresses
the quality of the data generated by FT-IR
long-path, open-path systems. For example,
when  error bars are given, they are often
merely stated, and no discussion of how they
were derived is supplied. As the FT-IR long-
path technique begins to be used for routine
monitoring,  this  approach  will   not   be
satisfactory, and a more extensive QA plan
must be developed.

      The development of and adherence to
a QA  project plan requires  the operator to
consider exactly how the data generated by
an  FT-IR  long-path,  open-path  monitoring
program   will   be  obtained,  processed,
interpreted, and used.  When implemented
properly, the QA plan will alert the operator
if the instrument is not  functioning properly
or is generating erroneous data, and  it  will
contain recommendations for the corrective
action to be taken. The  various levels  of QA
                                       10-1

-------
AM/KW-"-'^
TECHZm
                                                                      TR-4423-99-03
 project plan  designs  apply to  programs
 ranging  from  research  and development
 programs to routine monitoring programs that
 must produce legally defensible data.  This
 chapter covers some of the points that will
 have to be addressed for any QA program.

       The  general  points  that  must be
 addressed for any QA program are given in
 this  chapter.  These points are drawn from
 specific   documents   that   address  QA
 requirements for data obtained for the U.S.
 Environmental Protection Agency.  Points of
 emphasis discussed  in  these  documents
 include the  following:   project description,
 organization,   and   responsibilities;   QA
 objectives;  site  selection  and  sampling
 procedures;  sample  and   data  custody;
 calibration   procedures   and   frequency;
 analytical    procedures;   data   reduction,
 validation,  and  reporting;  internal  quality
 control  checks;  performance and  systems
 audits; preventive maintenance; calculation
 of data quality indicators; corrective action;
 quality control reports to management; and
 references.

      These, and other, items are addressed
with respect to FT-IR long-path,  open-path
monitoring  in this chapter.   In  addition,
portions of the QA project plan for a 1989
Superfund Innovative Technology Evaluation
study in  Delaware  are  presented as  an
example for use in FT-IR long-path, open-path
monitoring.  Also, a case study involving the
acquisition of QA data over a two and one-
half  month  period is presented in Section
10.3, and recommendations for procedures
                                          to be included in a QA program are given in
                                          Section 10.4.

                                          10.2  Project Plan Categories

                                                 The U.S.  Environmental  Protection
                                          Agency  has  published  a  document that
                                          defines four  different  categories  of  QA
                                          project plans, as described in Section 10.2.1.
                                          The program with the least requirements is a
                                          research and development program (Category
                                          IV), whereas a program  that produces data
                                          by  routinely   monitoring  the  atmosphere
                                          (Category I) has the most. When considering
                                          programs that are specifically designed to
                                          obtain data  for  the U.S.  Environmental
                                          Protection Agency, it is  convenient to refer
                                          to two documents that describe QA  project
                                          plans:  (1) Preparing Perfect Project Plans: A
                                          Pocket Guide for the Preparation of Quality
                                          Assurance Project Plans  and (2) Preparation
                                          Aids for the Development of RREL Quality
                                          Assurance   Plans   (U.S.   Environmental
                                          Protection Agency 1989, 1991). Actually,
                                          there are four parts to the second document,
                                          one for each of the four categories, and each
                                          has its own document number.

                                                Some of the features  of the  items
                                          that need to be addressed  in the Category I
                                          project plans are covered in Sections  10.2.1
                                          and  10.2.2.   It  is intended  that  these
                                          sections  be a  paraphrasing  of the  two
                                          documents listed above.  People who have to
                                          deal with these issues must obtain a copy of
                                          the documents and follow the guidance given
                                          there.   To give  some  specificity to  the
                                          various points of  a QA project plan, the  QA
                                          project  plan  used for  a  recent program
                                       10-2

-------
                                                                       TR-4423-99-03
conducted by EPA  is used as an example.
The  program entailed taking data with an
FT-IR  open-path  monitor  in  an industrial
complex and comparing that  data to data
obtained by a canister technique according to
method TO-14 (Russwurm  and McClenny
1 990).  Funding was allocated for 10 days of
measurements with the FT-IR instrument in
the field.
10.2.1
Category Definitions
   •  Category I projects are those that are
      designed so that their results can be
      used   directly,  without  additional
      support  for   compliance  or  other
      litigation. As such, they must be  able
      to  withstand  legal  challenge  and
      therefore have the most rigorous and
      detailed requirements. These projects
      are critical to the  goals  of the U.S.
      Environmental Protection Agency.

   •  Category II projects are those whose
      data  complement  other  projects.
      When combined with the output from
      other projects,  these  data  can  be
      used for rule making or policy making.

   •  Category III   projects  are   those
      producing  data  that   allow   the
      evaluation  and  selection  of  basic
      options for use in feasibility studies.

   •  Category IV projects are those that
      are associated  with  research  and
      development projects. The results are
      used to assess the basic or underlying
      assumptions or suppositions of other
       work. Because of the nature of these
       projects,  they  have  the  minimum
       number  of  items that need  to be
       addressed in a QA program.

       These categories are intended to be
fairly general  and broad, and any project
must fit into one of the four. The number of
items that must be addressed for each of the
categories   is   16,   13,  12,  and  6  for
Categories I through IV, respectively.  The
project  described above  for  acquisition of
data for comparison purposes and referred to
as an example in Sections 10.2.2.1 through
10.2.2.16 was considered to be  a research
program,   and  therefore   it  fell   into
Category IV.

10.2.2    Category I  Points to Be Addressed

      The 16 items that must be addressed
for the QA project  plan in this category are
listed below.

    1. Project description
    2. Project organization and
      responsibilities
    3. QA objectives
   4. Site selection  and sampling
      procedures
    5. Sample  custody
   6. Calibration procedures and
      frequency
   7. Analytical procedures
   8. Data reduction, validation, and
      reporting
   9. Internal  quality control checks
   10. Performance and systems audits
   11. Preventive maintenance
   12. Calculation of  data quality  indicators
                                       10-3

-------
                                                                       TR-4423-99-03
   13.  Corrective action
   14.  Quality control reports to
       management
   1 5.  References
   16.  Other items

These  16 items are discussed briefly below.

10.2.2.1 Project Description

       The  most important  feature of the
project description is  that a person who  is
unfamiliar with the project, but is familiar
with  the  technology,  must  be  able  to
understand this section.

       For the example project, the following
items were included in this section.

   •  An   in-depth   discussion   of  the
      comparison program.  The primary
      aspect of this was to relate in detail
      how the data  would be taken.  The
      FT-IR  instrument  is  a   long-path
      monitor, and the canister technique  is
      a point monitor.  Thus, it was decided
      to transport the canister along the
      path  while  the FT-IR  monitor  was
      acquiring data.  Topics that had to be
      included in the  QA project plan were
      the number of traversals along the
      path  and  the  number  of  scans the
      FT-IR spectrometer would make.

   •  A  brief description  of  the  FT-IR
      technique and the canister technique.
      The techniques were described, and
      appropriate  documents   for  each
      technique were referenced.
 10.2.2.2 Project Organization and
          Responsibilities

       This   section  must  describe   the
 relationships  among  all  of  the   people
 connected with the project, including the QA
 manager, and give their  responsibilities.  It
 should  be noted  that,  somewhere  in  the
 organization conducting the program, there
 should be an autonomous QA representative.
 Most  organizations  have   this  function
 removed from the  technical staff and under
 the jurisdiction of an administrative manager.
 The primary function of this person is to act
 as final arbiter for any disputes about the QA
 aspects  of  the program.    Therefore, he
 should  not  be  administratively connected
 with the technical  program.

       For the project being outlined as an
 example, the management and administrative
 staff   consisted   of  U.S.   Environmental
 Protection Agency staff and staff from two
 contractors. The following descriptions were
 therefore included  in  this  section.

    •   All the principal people involved and
       their duties

    •   The relationship of one person to the
       other  concerning   decision-making
       responsibilities

    •   The lines of communication among
       the project personnel

All lines of communication among the various
project   personnel,  including   any   sub-
contractors,  must be described.    If  no
                                       10-4

-------

                                                                      TR-4423-99-03
 subcontractors are used, it is sufficient to
 simply state that fact.

 10.2.2.3 QA Objectives

       This  section of the QA project plan
 must cover items such as detection limits,
 precision, accuracy,  completeness of  the
 data, representativeness  of the data, and
 comparability of the data. There must be a
 discussion of the impact of  not meeting
 these objectives  and  how these  indicators
 will affect the legal defensibility of the data.

       Inadequately addressing these items is
 probably  the most frequent cause of the
 rejection of a QA project plan.

       Three   items—detection   limits,
 precision, and accuracy—must be numerically
 defined as QA objectives. Completeness of
 the data defines what percentage of the total
 number  of  possible  data  points that are
 available  under the sampling schedule are
 expected   to    be   captured.
 Representativeness of the data implies how
 well  the  acquired  data  account for the
 variability   of   the    real    situation.
 Representativeness of the  data  concerns
 itself with sampling.  Comparability of the
 data indicates how well  the data can  be
 compared  to  that   taken  with   other
 instrumentation.
 specific definitions were  required.  At the
 present time there are no generally accepted
 definitions  for these  quantities for  FT-IR
 open-path monitors.

       For the completeness of the data, the
 following procedure was  used.   A  10-h
 working day was assumed, and 2  h  were
 allocated for setup and warm-up time of the
 instrument.    The data-taking  period was
 assumed to be  0.5  h for each spectrum.
 Therefore, it was anticipated that 16 spectra
 a day would be taken.  During the study, it
 was noted  that  the  concentration of the
 gases  being monitored  was changing at a
 much    higher   rate   than   could   be
 accommodated by the  0.5 h allocated to
 each   data  spectrum.    Therefore,  the
 sampling schedule was altered to account for
 the change. It should be noted that this  is
 certainly an appropriate response to changing
 conditions.  The  QA  plan  is a guide for
 subsequent  operations and  not  something
 that is unchangeable.  All changes must be
 recorded, and  the rationale  for the  change
 must be  presented.

       A large  section of the plan concerned
 itself   with  the   representativeness  and
 comparability of the data.  The reason that
the canisters had to be transported along the
 path   was   precisely  to   satisfy   the
 requirements of these two items.
      A portion of the example field program
(Russwurm  and   McClenny   1990)  was
designed to determine how to best make
measurements of the detection  limits, the
accuracy,  and the precision.  Therefore, no
10.2.2.4  Site Selection and Sampling
          Procedures

       In addition to a physical description of
the site,  the  rationale  for  selecting  any
                                       10-5

-------

                                                                       TR-4423-99-03
individual site  must  be presented.   This
discussion must address how the site will
allow the data objectives to be met.

       The sampling portion of this section
must address the scientific and regulatory (if
any) objectives that the sampling protocol
allows.   It must  also  address  how any
calibration samples are to  be obtained and
delivered to the system.  In the case of the
FT-IR system, if a gas sample cell is  to be
used, then its preparation and use must be
addressed in this section.

       Site    selection    and    sampling
procedures played an important role in the
example study.  Several  months before the
field portion of the example study, a search
for  suitable  sites was started. The general
area of the study had been selected, but the
individual sites and FT-IR paths needed to be
chosen.   Therefore,  this  section  of the
example QA plan  contained  the  following
descriptions for  the site selection process
and sampling procedures.

   •  Background  information   on   site
      selection. During a number of trips to
      the general area, various land owners
      were contacted,  and  several  sites
      were selected as usable for the study.

   •  Predominant requirements for the site.
      Proximity   to  the  source,   an
      unobstructed path  length  of  up to
      300 m, a  path that could be used to
      transport the canister while the FT-IR
      monitor was taking data, and a site
      that was safe for the operators.
    •  Procedure for using a QA gas cell. A
       description of using a short cell filled
       with a high concentration  of gas for
       calibration purposes  was  described
       here.  The use of this cell  was also
       intended for determining the precision
       of the instrument.

 10.2.2.5 Sample Custody

       This  section must present complete
 sample custody procedures and  personnel
 responsibilities in handling samples. Because
 the  FT-IR  data  are  stored  on   disk,  all
 procedures for ensuring the integrity of  the
 data on the disk and the legal defensibility of
 those procedures must be addressed.

       For the example program, no sample
 custody procedures were required because it
 was a  research program.

 70.2.2.6 Calibration Procedures and
          Frequency

       The FT-IR  open-path  monitor  is not
 calibrated in the classical sense. That is, a
 sample  of  known  concentration  is  not
 presented   to  the    instrument   for
 measurement.  During  FT-IR sampling, the
 absorbance values in all the spectra obtained
 at  various wave  numbers for the specific
 gases   are   always  compared   to  the
 absorbance values of the reference spectra.
These reference spectra are made  with pure
 samples of the gas.  Production of  reference
spectra  is  a  formidable task,  and few
laboratories   are  equipped  for  such  an
undertaking.  Because of this, only a limited
                                       10-6

-------
                                                                        TR-4423-99-03
 number of spectral libraries that can be used
 for quantitative analysis exist.

       Although   measurements   of  the
 precision  can be  made  by using a  short cell
 filled with a  pure sample  of the gas,  this is
 not routinely done at the present time.  There
 is currently no agreed-upon procedure for the
 use  of such a cell.

       The project being used as an example
 did not require that the system be calibrated.
 However,  a  written procedure for using a
 short cell to make  precision measurements
 was part of the QA plan.

 7 0.2.2.7  Analytical Procedures

       For the most  part,   unless  other
 techniques are to  be  used  along  with the
 FT-IR system, there is nothing to address  in
 this  section.  The analytical procedures that
 are   to   be   used   to   determine   the
 concentrations of any short-cell gas mixtures
 would have to be addressed here.

       In the QA  project plan being used as
 an example, the procedures for preparing the
 short cell  and the gas mixtures were given
 here.  The cells are generally filled with a
 pure gas  sample and  then  backfilled with
 nitrogen so that the total pressure  is 1 atm.
All the apparatus used for this procedure was
described  in this section.

       This   section  also   contained   a
description of the procedure to be used for
the analysis of the  canister  samples.   This
was a brief description, but it referenced the
 TO-14  procedure  manual.  The entire QA
 procedure for that portion of the effort did
 not have to be presented in the plan, but it
 had  to  be referenced.  If  one had not been
 available, then  it would have  had to  be
 written and given in this section.

 10.2.2.8  Data Reduction, Validation, and
          Reporting

       The data reduction procedures must
 be discussed in  this  section. This includes
 the  least-squares regression  analysis,  or
 other multicomponent analysis method, if it
 is to  be used.   All statistical  methodology
 that  is used as an aid to  data interpretation
 must be described here.   This  section can
 also   include  sample  calculations.    All
 procedures to be used for flagging the data
 and removing outliers from the data set must
 be stated in this section.  The flow of data
 and  the procedures that will  be used  to
 transfer the data from where it is generated
 to the end user must also  be described.

      The QA project plan  being used for
 discussion purposes included an analysis of
the  spectral data with   a  classical least
 squares algorithm.  The  actual  process  of
 recording the interferogram,  performing the
Fourier  transform,  and  the least-squares
analysis was not discussed  in detail.  The
published   papers  that   describe   these
techniques were referenced.  The process of
selecting the wave number range for analysis
was   discussed,  and  the  wave number
regions  that were used were listed  in this
section of the QA project plan.
                                        10-7

-------
                                                                      TR-4423-99-03
       The   TO-14   procedure   for  data
 reduction also was referenced, but the actual
 details were  not presented in the plan.

       The major portion of this section dealt
 with the comparison of the data which was
 to  be  done  by  regression techniques.  At
 various stages in the program, two canister
 samples were taken simultaneously, and the
 comparison procedure for these  data was
 discussed.

       Validation of the data  was  described
 as  the  process of  reviewing the  study
 logbooks to  make sure that no  unwanted
 contamination of the samples had  occurred.
 For example, one sample was invalidated
 because an  automobile  had followed the
 person transporting the canister  along the
 path.

 10.2.2.9 Internal Quality Control Checks

       This section is designed to determine
 what internal QC checks are  to be made.  It
 must  also cover  why  these checks  are
 necessary and how they will help to achieve
 the data quality objectives. For example, this
 section  would describe the  use of control
 charts that show the peak-to-peak readings
 of the single-beam spectrum at various path
 lengths and with varying amounts of water
 vapor.

      At the time  of the example study,
there were no requirements for internal QC
 measurements written into  the  plan.   A
 portion of the study  was designed as an
 attempt to  develop procedures for making
 these measurements.

 10.2.2.10 Performance and System Audits

       System  audits  are generally  done
 before  any  data  are  taken.   They are
 designed  to answer  questions  about the
 proposed procedures and sampling protocols
 to be used in the program.  The performance
 audits are designed to determine whether the
 instrument is operating as it was described in
 the other sections of the study's QA project
 plan. Both of these audits are generally done
 by people who  are not associated  with the
 daily operation of the instrument.

      There were no provisions for either
 system  or performance audits for the study
 being used as an example.

      Although it is easy to  envision the
 system  audit for the FT-IR instrument, the
 performance audit  of the instrument would
 be much more  difficult to do at this time.
 This is usually done by using the instrument
 to measure  a known concentration of gas
 and  determining the  response.  There has
 been no  systematic  development  of  a
 procedure for  doing  this for the  FT-IR
 systems  to date.    Currently,   not  all
 commercially  available  systems   have
 provisions for putting  a short cell in the
 beam.

 10.2.2.11 Preventive Maintenance

      This section requires a description of
the preventive maintenance procedures and
                                       10-8

-------
                                                                       TR-4423-99-03
 the   schedule   on  which  they   will   be
 performed.  The time involved in performing
 preventive maintenance will affect the total
 amount of data that can be taken,  and this
 must be accounted for in the description of
 the completeness requirements.

       In  the example QA project plan,  no
 preventive maintenance schedule was given
 because the study was a short-term intensive
 program.  Over the 10-day field program,  no
 maintenance of the instrument was required.

 10.2.2.12 Calculation of Data Quality
           Indicators

       This section must contain  detailed
 procedures  that   are  to  be  used  for
 determining the data quality. It must include
 all the statistical routines that are to be used.
 Items that must be addressed are precision,
 accuracy, outliers, etc.

       Currently,  no  generally  accepted
 definitions for  precision  and accuracy are
 available for the FT-IR open-path technique.

 10.2.2.73 Correc tive A c tion

       The corrective action portion of the
 QA project plan is a set of contingency plans
that  try  to address  "what if"  questions.
These  corrective action  plans  serve  as a
check for the general tendency to want  to
perform quick fixes of equipment to get  as
much data as possible. This is certainly the
case with  short-term field programs,  but it is
generally better  to take the instrument off
line and repair  or adjust it  properly.  The
 corrective action plans should describe how
 this is  to  be done and  specifically what
 criteria  will  be used to make the judgement
 as to when to discontinue data collection and
 shut down  the instrument for repair.  This
 section  was  not  a  requirement for the
 example QA project plan.

 10.2.2.14  Quality Control Reports to
          Management

       This  section must state what reports
 will be transmitted to  whom and when they
 will  be  transmitted.   There should be  a
 description  of the contents of each  report,
 and  all  the QA/QC  data  that  must be
 included in each report  should be stated  in
 this section  of the QA project plan.

      The QA project plan for the example
 study  did   not  address  this requirement
 because there was no consistent QA  data
 generated in the program.

 10.2.2.15 References

      When  references   related  to  the
 present  program are available, they should be
 included in  the  QA  project plan.    For
 example, as  mentioned in Sections 10.2.2.1
 and 10.2.2.7, appropriate documents were
 referenced for the canister technique used  in
the example QA project  plan.

 10.2.2.16 Other Items

      The   QA   project   plan  contains
information that the principal investigator will
need  at various  points of  the  program.
                                        10-9

-------
                                                                      TR-4423-99-03
 Because of this, it is somewhat personalized,
 and this section can include any other items
 that are considered important to the program.

       There were no other items included in
 the QA project plan for the example study.

 10.3  Case Study:  QA Data Collected
       Over Two and One-Half Months at a
       Semipermanent Field Site

       The   study   described  here  was
 designed to evaluate the stability  of a long-
 path  FT-IR  system  and to determine the
 precision and accuracy of the concentration
 measurements (Thompson et al. 1994). The
 following criteria  were  used  to assess the
 stability of the instrument:  electronic noise,
 the magnitude of the. return signal, the RMS
 baseline  noise,  and the  repeatability of the
 position and full width at half height (FWHH)
 of  selected  absorption  bands.   Ambient
 concentrations  of CH4,  N20,  and  CO  were
 measured to  test the use of these data for
 determining  the precision and accuracy of
 the FT-IR open-path monitor. Measurements
 were made  daily  over two  and  one-half
 months,  from  November  1993  to  mid-
 January 1994.

       Spectral data were acquired by using
 a monostatic  FT-IR monitor. Each  spectrum
 consisted of 64 co-averaged scans recorded
 at a nominal  1-cm"1  resolution.  Triangular
 apodization was used.   The collection  of
 each   spectrum   required  approximately
 5  minutes. A spectrum was  taken  every
 15  minutes.  Single-beam  spectra  were
typically, acquired  over a  7-  to  8-h  time
 period.  Absorption spectra were created by
 ratioing  the  single-beam  spectra  to  a
 synthetic background  spectrum generated
 from a  2048-scan  single-beam  spectrum
 recorded  over  the  414-m  path.    This
 background spectrum was recorded at the
 beginning of the experiment and was used
 throughout  the study.   The  data  were
 analyzed by using a CLS software package
 and reference  spectra from a  commercial
 library.

      The site is located  near I-40, one of
 the  main traffic arteries for the  Research
 Triangle Park, NC, area. The instrument was
 kept in  a climate-controlled shed, which is
 heated during the winter months.  The total
 path length was 414 m, and it extended over
 an  open,  grassy field and a small parking
 area with very limited traffic. The beam path
 rose from about 1.8 m to 1 2.8 m above the
 ground  as  it was directed from the FT-IR
 spectrometer to  the retroreflector array,
 which was mounted on a tower.

      The instrumental electronic noise was
 measured each morning before the detector
 was cooled with liquid nitrogen.  This signal
typically ranged  between 600 and  620
 counts with the instrument in  the  single-
 beam mode. Shortly  after  the detector was
cooled, the instrument was aligned and the
maximum return signal was recorded.  The
return signal was recorded again (without
realignment) around  noon to  check  the
stability of the  signal.   On clear  days the
single-beam  return  signal  ranged  from
 10,500 to 13,500 counts. (See Figure 10-1.)
Certain atmospheric conditions  caused the
                                       10-10

-------

                                                                      TR-4423-99-03
  14,000 -

|^

3 10,000 -
 to
    •-000'
  Z 4,000-
  oc
   2,000 -

     0
                          4 ,
                Data of Measurement
 Figure 10-1. Return Signal Magnitude of the
 FT-IR Monitor Measured Daily at 0700 (A)
 and 1200 (•).  (The data points have been
 connected by lines for the convenience of
 the  reader and not to indicate  continuous
 data.)
                                            I
                                            M
                                                          Date of Measurement
                                                                           .
                                                                          yf yf-
                                         Figure  10-2.  The  RMS  Baseline  Noise
                                         Measured Between 980 and 1020 cm 1 (•),
                                         2480 and 2520 cm'1 (•), and 4380 and
                                         4420 cm"1 (A). (The data points have been
                                         connected by  lines for the convenience of
                                         the reader and not to indicate continuous
                                         data.)
 return signal to vary from day to day.  For
 example,  the return signal   dropped   by
 20-30%  during fog. On some  mornings,
 when the humidity was close to or below the
 dew point, condensation or ice formed on the
 retroreflector, resulting  in  a  lower return
 signal in the  early morning measurement. As
 the condensation evaporated,  an increase in
 return  signal counts was measured.   To
 remedy the problem of condensation, a heat
 lamp was mounted on the tower and directed
 at the retroreflector.  After the heat lamp
 was installed on December 10, the noon and
 early morning return signals were nearly the
 same.  The  use of the heat  lamp did not
 cause  an  increase in noise or detected  IR
signal.

       The RMS  baseline noise measured
over 26 days is illustrated in Figure 10-2.
The  baseline  noise was determined  by
collecting  two back-to-back,  64-scan, co-
                                         added spectra.  One spectrum  was ratioed
                                         against the other to obtain  an absorption
                                         spectrum.  The RMS noise (in  absorbance
                                         units) was  calculated over three spectral
                                         regions:  980-1020,   2480-2520,   and
                                         4380-4420 cm"1. During operations when
                                         condensation   did  not   form  on   the
                                         retroreflector, the baseline noise was  on
                                         average approximately 2  x  10'4  for the 980-
                                         1020-cm"1   region,  2.5 x 10"4  for  the
                                         2480-2520-cm"1 region,  and 9 x 10"4 for
                                         the   4380-4420-cm"1   region.   In   these
                                         measurements, the 980-1020-cm'1 region
                                         included water  vapor  bands.  For a true
                                         measurement of instrument performance that
                                         is not influenced  by  temporal  changes in
                                         water   vapor   concentration,   it   is
                                         recommended that the region from  968 to
                                         1008 cm"1  be used.   During  measurement
                                         periods when  condensation formed  on the
                                         retroreflector,  the  baseline  noise  for these
                                      10-11

-------
                                                                      TR-4423-99-03
 regions increased to 9.7 x 10'4,5.5 x  10"4,
 and 2.9 x 10"3, respectively.

       The  wave  number  stability  of the
 instrument was determined by monitoring the
 peak position and the  FWHH of the water
 vapor singlet at 1014.2 cm'1. Band positions
 typical of data collected at the beginning, in
 the middle, and near the end of the study are
 depicted  in Figure  10-3.   No shift in the
 frequency was  observed  during this  time
 period.  Also, no shifts were observed in the
 1-cm"1 spectra  collected under a variety of
 weather conditions, including rain, freezing
 rain,  sleet,  snow, and  low  (single-digit)
 temperatures.  To determine if  the water
 vapor singlet at 1014.2 cm"1 broadened,  a
 spectrum collected at the beginning of the
 experiment was subtracted from a spectrum
 in the middle and end of the experiment. No
 broadening was evident during the middle of
 the experiment; however, a slight broadening
 for  some of the spectra at the end of the
 o
 I
 o
 M
 a
           1010       1015       1020
             Wavenumbers (cm '1)
Figure 10-3. Repeatability of the Position of
the Water  Vapor Singlet at 1014.2  cm'1
Measured on  (A) November 10,  1993, (B)
December 22, 1993,  and  (C) January  4,
1994.
 experiment was observed.  The FWHH of the
 water vapor singlet in spectra  taken during
 different atmospheric conditions was also
 examined.  When a clear day spectrum was
 subtracted from any of these spectra,  no
 broadening was evident. It should be noted
 that   short   wavelengths  (higher  wave
 numbers) will be more sensitive to spectral
 shifts and changes in resolution.  The HDO
 doublet centered at 2720 cm'1, the CO band
 at 2169 cm"1, or other water vapor bands in
 the higher wave number region can also  be
 used  to test  for  shifts  and  changes  in
 resolution.

       The feasibility of using ambient gases
 for accuracy and  precision measurements
 was    also   investigated.   Ambient
 concentrations of N20, CH4, and CO were
 measured  on a  daily basis.  Each morning
 between 071 5 and 0930 the concentrations
 of  these  gases increased, then  steadily
 decreased during the remainder of the day.
 However,  concentrations of N2O and CH4
 remained constant, approximately 250 ppb
 and 1.7 ppm, respectively.  To determine
 whether the increases  in  concentration
 during the first 3 h  of operation were due  to
 an instrument effect or to  the  proximity  of
the site near a  major  highway, data were
 collected continuously for 36 h.

       Data    were    collected    from
 November  17  at  0730  until  1730  on
 November 18.  The CH4  concentration data
exhibited scatter during an early morning fog
episode and  decreased steadily during the
day. (See Figure 10-4A.) A step in the CH4
                                      10-12

-------

                                                                       TR-4423-99-03
 concentration  measurement was observed
 when the liquid nitrogen in the detector was
 depleted   at   approximately   2345   on
 November 1 7. The CH4 concentration value
 was 1.70 ppm just before the liquid nitrogen
 was depleted, increased to 1.9 ppm after
 liquid nitrogen was refurbished, and remained
 10% higher compared to the previous levels.
 The  concentration data for CO showed a
 similar, stepped increase.
 c
 o
 O
    07:12  12:12   17:19  22:15   03:30  08:30  13:30  11:30
       09:42  14:40  19:45  01:00  06:00   11:00  IfcOO
                    Time
 Figure 10-4. Measurement of (A) Ambient
 Methane Concentration and (B) Single-Beam
 Intensity at 987 cm'1 on November 17 and
 18, 1993.
       The CH4 concentration data exhibited
irregular behavior during a 6-h period shortly
after the detector Dewar was refilled with
liquid  nitrogen.   To  determine   if  the
instrument was operating properly during this
time period,  the single-beam intensity at
987 cm"1  was  measured  from  archived
spectra.   The  single-beam  spectra  had  a
lower intensity during the fog episode, then
leveled off until  the detector  ran  out of
coolant.   (See  Figure  10-4B.)   After  this
sudden  .drop,  the  single-beam  intensity
 returned to its original reading and remained
 relatively constant throughout the remainder
 of the experiment.  This  indicates that the
 instrument was working properly during the
 episode of high  measured CH4 levels.

        One other observation during this time
 period  concerned  the effect of water vapor
 concentration on CIS analysis for CH4.  On
 November 17 a cold front moved through the
 area in the late evening, and the water vapor
 pressure dropped rapidly. Because the water
 vapor spectrum is used  as an interfering
 species in the CLS concentration analysis for
 CH4, the  sudden  change in  water  vapor
 pressure could have had an effect on the CH4
 concentration measurements.   The relative
 concentration of water vapor along the path
 was determined by measuring the peak area
 of  the  absorption  band  at  1014.2  cm'1.
 Likewise, the relative concentration of CH4
 was determined  by measuring the peak area
 of the absorption band at 2998.8 cm'1. This
 peak was chosen because the water vapor
 bands do not interfere with it.   However, it
 was later brought  to our attention that this
 CH4 band actually overlapped nearly exactly
 with a water vapor band (W.F.  Herget, ETC,
 personal communication). This might explain
 the  dip in the peak  area  measured  at
 2998.8 cm"1 during the time that the water
 vapor concentration was decreasing rapidly.
 Therefore, this band is not a good choice for
 this type of data analysis;  the CH4 bands at
 2916.8 and 2927  cm"1 do not  overlap with
 water vapor and should be used instead. As
 shown  in Figure 10-5,  the  relative water
 vapor concentration decreased rapidly when
the  front moved  through the area. The
                                       10-13

-------

                                                                       TR-4423-99-03
 relative CH4 concentration increased during
 this period. Similar trends were  observed
 between plots of the CH4 peak area and CH4
 concentrations  determined  by  the  CLS
 concentration  analysis. This indicates that
 the change in water vapor concentration did
 not greatly affect the CLS  analysis for CH4,
 and   the  fluctuations   in   the   CH4
 concentrations were real.
 a
 < 0.10
  (Q

  I,
 I.
          A
     07:12  12:12  17:13  22:15  03:30  00:30  13:30   18:30
        09:42  14:40   19:49  01:00   09:00  11.-00  16:00
                     Time
 Figure  10-5. Peak Area  of 2998.8-cm1
 Absorption  Band  of  CH4  (A)  and  the
 1014.2-cm'1  Absorption  Band of Water
 Vapor (B) Measured  on November  17 and
 18, 1993.
       The  data collected  in  this  study
indicate that for this particular FT-IR monitor
the  return  signal and  baseline  noise  are
repeatable and are instrumentally stable over
extended   periods,   but  are   subject  to
variations due to weather conditions.  The
peak positions and the FWHHs of the water
vapor singlet at 1014.2 cm"1 were repeatable
from day to day  and  were not affected  by
rain,  freezing  rain,  snow,  or  single-digit
temperatures.   The  variability   in  the
concentrations  of  CH4  limits  its  use  in
instrument stability studies and for accuracy
and precision determinations.  A step in the
concentration measurements associated with
the depletion and refurbishment of detector
coolant  is  not  yet  understood.    Future
experiments  with  a  QA  cell  and surrogate
standards are planned to further investigate
this effect.

10.4   Recommendations for Tests to be
       Included in a QA Program for FT-IR
       Long-Path Monitors

       We  are  currently  evaluating  and
developing  procedures   to determine  the
quality of data taken with FT-IR monitors.
The following is an outline of criteria that we
have used in a preliminary QA program to
verify the performance of an FT-IR long-path
system.  Development of  a QA plan for FT-IR
monitoring   should   include,   but   not
necessarily  be  limited to,  these types  of
measurements. These tests are designed to
determine that the instrument is operating
properly and producing good data.  Some of
these issues were discussed in Chapter 3 for
the   initial   verification   of   instrument
performance, but can be used for routine QA
procedures  as   well.    Other  criteria  for
development of  a  QA plan, such as  siting
criteria or data chain-of-custody, should be
addressed  as   warranted,  but  are   not
discussed here.    These procedures  were
developed for a  research and development
program,  but factors  relevant  to  routine
monitoring programs were  also  taken into
consideration.   Bear  in mind  that  two
separate  issues must be  addressed in a  QA
plan.  One is whether or not the  instrument
is working properly. The other issue is if the
                                       10-14

-------

                                                                       TR-4423-99-03
 method  used  for  quantitative  analysis is
 producing the correct results.
 analysis of a target gas to give an estimate
 of detection limits.
 10.4.1    Noise Measurements

       Measurements of two types of noise
 can  be   routinely  taken,  instrumental
 electronic noise and random baseline noise.
 Electronic  noise  is  recorded  before  the
 detector Dewar is filled with liquid nitrogen.
 This  small  signal  is  indicative  of  the
 electronic  noise of the  system  with no
 detector signal.  This should remain relatively
 constant and typically contributes less than
 0.25%  of the total  return signal.  If  some
 electrical  component   of  the  system  is
 producing  spurious  noise, it  will become
 apparent from this measurement.

       Random baseline noise is measured by
 recording back-to-back spectra  after  the
 detector has been filled  with liquid nitrogen.
 One spectrum  is then ratioed to  the  other
 and the absorption spectrum is calculated.
 The result is a  spectrum of the random
 system  noise.   The RMS or peak-to-peak
 noise in  absorbance  units  can  then be
 calculated  from these  spectra.   These
 spectra  should  be  acquired  by using  the
 instrumental parameters to be used during
the analytical measurements. The baseline
 noise measurements should be taken  in a
spectral region that is devoid of absorption
due to  water vapor or other atmospheric
gases.   If not,  changes  in water  vapor
concentration  over the  measurement  time
will influence  the magnitude of the noise
calculations.  Noise measurements can also
be taken over the spectral  region chosen for
       For the baseline noise measurement it
 is best to record these two  spectra back to
 back, as passage of time between the two
 spectra, might  also  include  changes  in
 atmospheric conditions or concentrations of
 species in the path. Spectra taken at longer
 time intervals during the study can be ratioed
 in this manner to determine baseline stability
 or systematic noise.

 10.4.2    Stability of Instrument

       Several aspects related to the stability
 of the instruments can be measured.  One is
 the repeatability of the noise measurements
 described above. These noise measurements
 should  be  taken daily and  recorded  on a
 control  chart  to alert the operator of any
 gross changes or trends in the deterioration
 of the baseline noise.

      Another measurement to  be taken
 daily, or several times during the day, is the
 return intensity. This can  be measured either
 as the single-beam intensity at a selected
 wave number region or the magnitude of the
 interferogram. At this time, the position of
the zero peak difference of the interferogram
 should also be recorded, and the single-beam
spectrum should be examined for evidence of
system   nonlinearity   (see   Chapter   3,
Section  3.4).   The  single-beam  intensity
should  be  measured  in  different  wave
number  regions to determine if the source
characteristics  have  changed   or   the
interferometer alignment  has been altered.
                                       10-15

-------
        - "-'=*' •" •' -'. '= =
            •   •
        -.•'•:• •=••==
         f=7?± ~^=
          _-.i •
                                                                       TR-4423-99-03
 The high wave number  (short  wavelength)
 portion  of  the  spectrum  will  be  most
 sensitive to interferometer misalignment and
 will show a decrease in intensity relative to
 the other wave number  regions if changes
 have occurred.

       In addition to being dependent on the
 instrument's   performance,   the   return
 intensity  also  depends  on   atmospheric
 conditions.     For  example,   fog  has  a
 deleterious effect on the return signal. Thus,
 the atmospheric  conditions  must be noted
 when the measurement  is taken.  As with
 the noise measurements, the return intensity
 should also be plotted daily on a control
 chart.  A decrease  in return intensity could
 be  related to a drop in the source intensity,
 misalignment   of  the   external  optics,
 misalignment  of the  interferometer optics,
 deterioration of the system optics, or a loss
 in the detector Dewar hold time.

       The positions of selected absorption
 bands should also be recorded.  In long-path
 measurements the  water vapor  singlet  at
 1014.2 cm"1 works well  for this purpose.
 The FWHH  of  this  band   can  also  be
 measured to determine the repeatability  of
the  resolution  of   the  system.     The
subtraction   technique  described   by
 Russwurm  (1992b) can be used  to detect
small frequency shifts or subtle changes  in
the instrument resolution. If the resolution of
the system  is  deteriorating, and  the band
becomes broader than a spectrum recorded
previously, the subtraction result will appear
as an "M".   Shifts in frequency will produce
a derivative  shape in the subtraction result.
 Note that the FWHH measurement and the
 spectral subtractions should be done on an
 absorption spectrum, and not on  a single-
 beam spectrum or transmission spectrum.

       As mentioned previously, bands  in
 higher wave number regions actually  may
 work better  for this test,  as they will be
 more sensitive to spectral shifts and changes
 in resolution.  In addition to the water vapor
 bands centered  at 1014.2 cm'1, the  HDO
 bands centered at 2920  cm"1 can be used for
 this test.

       All of the above measurements should
 be  recorded  on  at least a daily basis  and
 compared to existing data to  establish  that
 the instrument is performing properly.

 10.4.3    Accuracy and Precision

       The determination of  accuracy  and
 precision are not  as straightforward as the
 tests used to determine  if the instrument is
 performing properly.  The concentrations of
 ambient gases, such as  CH4 or N20, can be
 used to a certain extent  for these purposes.
 This approach has the advantage that certain
 gases  are always present  in  open-path
 spectra and no changes  have to be  made to
the instrumental configuration to measure
these gases.  If the ambient concentrations
of these gases are assumed to be constant,
precision measurements  can be made.  For
example,  we   have   measured    the
concentration  of  N2O continuously over  a
five-day period in the late spring to be within
 ±3.5% of the mean value.   On the other
hand, as discussed in Section 10.3, we have
                                       10-16

-------

                                                                      TR-4423-99-03
 seen CH4 concentrations increase by a factor
 of 2.5 in a short time during measurements
 taken in the late fall.  Therefore, care must
 be  taken to  account for possible  emission
 sources  if ambient gases are  used for  QA
 purposes.  Ambient gases can also be used
 to test for frequency shifts or changes in
 resolution as described  above.  The  use of
 ambient  gases for QA purposes can be used
 to   estimate   the   precision   of   the
 measurement,  but this  approach does  not
 really address the accuracy of the method.

       The  alternative  approach to using
 ambient  gases  for QA  data is to insert a
 short  gas  cell  that   contains  surrogate
 standards into  the beam.  This  has  the
 disadvantage of an attenuated  IR beam due
 to the transmitting and  reflecting properties
 of the windows used in the cell. Thus,  the
 performance of the instrument  is somewhat
 degraded.   However,  this approach does
 have the advantage  of having  a known
 quantity  of target gas in  the path. Assuming
 this  quantity is accurately  known  and it
 remains  constant, accuracy and precision
 measurements can be made with a short cell
 technique.  To date, however, no universally
 applicable technique using  a short cell has
 been developed.

      As discussed in Chapter 3, stray light
 in   the   monostatic    configuration  and
 background   blackbody   radiation  in  the
 bistatic   configuration  can   affect  the
quantitative results. The signal due to stray
 light  or  background  radiation  must   be
 subtracted from the sample spectrum prior to
 quantitative analysis.  The stray light in the
 spectrometer of a monostatic instrument can
 be measured by  blocking the return beam
 with some type of opaque and nonreflective
 material.  In our experience the signal due to
 stray  light  has   been  relatively  constant,
 provided no components of the system  are
 changed.     The  background  blackbody
 radiation  of a   bistatic  system  can  be
 measured with  the  IR  source off.   This
 response will vary for different sites and can
 also change throughout the day. Therefore,
 this signal  must be recorded  on a more
 frequent basis. It is not obvious at this point,
 however, that simple subtraction adequately
 compensates for the effects of stray light or
 background  radiation linearly over a range of
 absorption values. Thus, the accuracy of the
 system might be affected.

       In  addition  to   determining  the
 accuracy and precision of the instrument, the
 accuracy and precision of the method used
 for  quantitative  analysis  must  also   be
 determined.   In most cases, an automated
 software package, such as one that uses the
 CLS algorithm, is used  to determine  the
 concentrations of the target gases.  These
 procedures  can  be checked  manually  by
 comparing the sample spectra to spectra of
 reference gases with a known concentration.
 Interactive subtraction procedures that yield
 a scaling factor for the reference spectrum
 can  be used to  check the  concentration
 measured  by the  CLS software.  In addition,
the reference spectra can be  scaled to  the
desired concentration and then added to the
sample  spectrum.  When the composite
spectrum  is then analyzed, the measured
concentration should reflect the amount of
                                       10-17

-------
                                                                       TR-4423-99-03
 reference gas added.  Care must be taken in
 choosing the spectral regions used to analyze
 for each target compound.  For example, the
 optimum region for analysis does not always
 encompass the entire  absorption envelope.
 Possible interfering  species  must also be
 accounted for in the analysis  method.

       The  operator should also be  aware
 that any time a concentration spike appears
 that cannot  be immediately attributed to a
 known  source, the  actual  spectral  data
 should be examined to verify the presence of
 the   compound   in   question   and  its
 concentration.  This can be  done by first
 subtracting   the   appropriate   absorption
 spectra of any interfering species from the
 sample   spectrum.    The  signature  and
 absorbance values of the resultant spectrum
 should then  be compared to the reference
 spectrum for proper features and intensities.
 Also, any time that the concentrations of the
 target compounds seem to correlate  with
 changes in  water  vapor  concentration, the
 spectra of the target compounds should be
 examined to  verify that the changes in
 concentration are real.  If the concentrations
 of  the  target compounds  exhibit  either
 positive or negative inflections with respect
to changes in water vapor concentration, the
 analysis method should be altered to alleviate
the problem.

      Ultimately,   if   the  instrument is
operating properly and  a suitable analysis
method is  developed,  the  accuracy of the
FT-IR technique will be determined by the
accuracy of the reference spectra. To  date,
 no way  of  validating  or  certifying  these
 reference spectra exists.

 10.4.4   Completeness and
          Representativeness of Data

       These  requirements  will vary  with
 specific monitoring applications. Care must
 be taken to ensure  that data points  are
 acquired frequently enough  to  account for
 the   variability  of   the   target   gas
 concentration.  Failure to do this will make it
 difficult to discern between  real changes in
 the target gas concentration and  possible
 variability in the FT-IR measurements.

 10.4.5    Comparability of the Data

       If possible, the FT-IR  data should be
 initially compared to an established method.
 This  can be  difficult because the  FT-IR
 produces a  path-averaged  concentration,
 whereas most established methods use some
 type  of  point  monitor.  As discussed  in
 Section 1 0.2.2, some of the FT-IR data have
 been compared to the  canister method. In a
 current  study,  we are  comparing  ozone
 measurements  recorded  with   the   FT-IR
 monitor  to  those  taken   with   personal
 sampling  devices and to hourly averages
from the  state ozone monitoring program.
Although not exact, these comparisons can
give  the  operator  an  idea if  the   FT-IR
measurements are within generally accepted
values.   If not, corrective action should be
taken.
                                       10-18

-------
                                                                     TR-4423-99-03
 10.4.6   Ancillary Measurements


       The type of ancillary measurements
 required will vary, depending on the type of
 study  being conducted.  For any long-path,
 open-path   measurements,  the  ambient
 temperature,  water  vapor  concentration,
 ambient pressure,  and wind velocity should
 be  recorded.  The operator should also  be
 aware of the effect of changes in altitude  on
 ambient  pressure.   If  the  instrument  is
 housed in  an  enclosed  environment, the
 temperature of that environment should also
 be recorded. We have also found it useful to
 record   the  temperature   inside  the
 spectrometer  itself,  especially  in   cold
 weather situations.


 10.4.7   Documentation


       As with any analytical methodology,
 a log of  instrument usage, downtime, and
 repairs, as  well as notes regarding unusual
 observations, should  be maintained.  These
 notes can prove invaluable for analyzing data
 that appear to  be abnormal. Records should
 be kept that are appropriate for the type  of
 study  being  conducted.    For example,
 requirements for a research and development
 project may be different from those required
 for legally defensible data.

 10.5   References


 Kagann,  R.H., J.G.  Jolly,  D.S.  Shoop,
 M.R. Hankins, and J.M. Jackson.   1994.
Validation  of  Open-Path   FTIR Data   at
Treatment, Storage, and Disposal Facilities.
SP-89  Optical Sensing  for  Environmental
 Monitoring,  Air  &  Waste  Management
 Association, Pittsburgh, PA, pp. 437-442.

 Kricks,  R.J.,  Scotto, R.L.,  Pritchett, T.H.,
 Russwurm,   G.M.,   Kagann,  R.H.,  and
 Mickunas, D.B.  1992.  Quality Assurance
 Issues Concerning the  Operation of Open-
 Path  FTIR Spectrometers.  Proceedings  of
 Optical Remote Sensing.   Applications  to
 Environmental  and  Industrial   Safety
 Problems, SP-81, Air & Waste Management
 Association, Pittsburgh, PA, pp. 224-231.

 Russwurm,   G.M.     1992a.     Quality
 Assurance, Water Vapor, and the Analysis of
 FTIR Data. Proceedings of the  85th Annual
 Meeting and Exhibition of the Air &  Waste
 Management Association.   Air  &  Waste
 Management Association, Pittsburgh, PA, pp.
 92-73.03.

 Russwurm, G.M. 1992b.  Quality Assurance
 and  the  Effects of Spectral  Shifts  and
 Interfering  Species  in  FT-IR  Analysis.
 Proceedings  of  Optical  Remote Sensing.
 Applications to Environmental and Industrial
 Safety  Problems,  SP-81,  Air &   Waste
 Management  Association.  Pittsburgh, PA,
 pp. 105-111.

 Russwurm, G.M. and W.A. McClenny  1 990.
 A  Comparison of FTIR Open Path Ambient
 Data  with Method  TO-14  Canister Data.
 Proceedings of the 1990 U.S. EPA/A&WMA
 International   Symposium  on   the
 Measurement  of Toxic  and  Related Air
 Pollutants.    Air  &  Waste Management
 Association, Pittsburgh, PA, pp. 248-253.

 Thompson, E.L, Jr., J.W. Childers, and G.M.
 Russwurm. 1994.  Development of Quality
 Assurance Procedures in  Open-Path FT-IR
 Monitoring. Proceedings  of the 1994 U.S.
EPA/A&WMA International  Symposium on
Measurement of Toxic   and Related Air
Pollutants.    Air & Waste Management
Association, Pittsburgh,  PA, pp. 529-534.
                                      10-19

-------
                                                                     TR-4423-99-03
U.S.  Environmental  Protection  Agency.
1989.  Preparing Perfect Project Plans:  A
Pocket Guide for the Preparation of Quality
Assurance   Project  Plans.    EPA/600/9-
89/087.   Available  from Guy  F. Simes,
Quality   Assurance   Manager,   U.S.
Environmental   Protection  Agency,  Risk
Reduction Engineering Laboratory, Cincinnati;
OH 45268.

U.S.  Environmental   Protection  Agency.
1991. Preparation Aids for the Development
of RREL Quality Assurance Plans (Category I
Project   Plans).      EPA/600/8-91/003.
Available  from  Guy  F.  Simes,  Quality
Assurance  Manager,  U.S.  Environmental
Protection   Agency,   Risk   Reduction
Engineering  Laboratory,  Cincinnati,  Ohio
45268.

Weber, K., H.J. van de Wiel, A.C.F. Junker,
and C. de LaRiva, C.  1992.  Definition and
Determination of Performance Characteristics
of Air Quality Measuring Methods  as Given
by  the   International   Organization   for
Standardization  (ISO)  -  Applicability  to
Optical Remote  Sensing.  Proceedings of
Optical Remote  Sensing.  Applications to
Environmental  and   Industrial   Safety
Problems, SP-81, Air & Waste Management
Association,  Pittsburgh, PA, pp. 30-42.
                                      10-20

-------
                                                                       TR-4423-99-03
                                    Chapter 11
                               Glossary of Terms for
                        FT-IR Open-Path Remote Sensing
 11.1  Introduction and Overview

       This chapter contains a glossary of
 terms for remote sensing, with an emphasis
 on those terms relevant to FT-IR long-path,
 open-path   monitoring.    When   possible,
 definitions of terms have been drawn from
 authoritative texts  or  manuscripts  in  the
 fields  of  remote   sensing,  air  pollution
 monitoring,  spectroscopy,  optics,   and
 analytical chemistry. In some cases, general
 definitions   have   been  augmented   or
 streamlined to be more specific to long-path,
 open-path  monitoring  applications.  These
 definitions   were   intended   to    remain
 scientifically rigorous and still be  generally
 applicable  to a  variety of  FT-IR open-path
 remote-sensing issues.

 11.2 Terms

 Absorbance: The negative logarithm of  the
 transmission.  A = -ln(///0), where / is  the
 transmitted intensity of the light and 70 is  the
 incident intensity. Generally, the logarithm to
 the base 10 is used, although the quantity /
 really  diminishes  exponentially   with   A
 (Pfeiffer and  Liebhafsky 1951).  If  the term
 "fractional transmission" is used for the ratio
///0,  then  the   implication  is  that   the
 instrument's  slit  function (see "Instrument
function") is accounted for (Penner 1959).
 Active  system:   A system that radiates
 energy  to the surrounding environment (for
 example, a radio transmitter).

 Apodization: A mathematical transformation
 carried  out  on  data  received  from  an
 interferometer  to  alter the  instrument's
 response function.  There are various types
 of  transformation;  the  most common are
 boxcar, triangular,  Happ-Genzel,  and Beer-
 Norton functions.

 Average  concentration:    For  FT-IR  or
 differential absorption spectroscopy systems,
 this quantity is the result of dividing the
 integrated concentration (the quantity that is
 measured) by the path length used for the
 measurement.   It  has  units of  parts per
 million,  parts per  billion, micrograms per
 cubic meter, etc. (McClenny and Russwurm
 1978).

 Background  spectrum:  1.  With all other
 conditions being equal, that spectrum taken
 in the absence of the particular  absorbing
 species  of interest.  2.  Strictly,  that radiant
 intensity incident on the front plane  of the
 absorbing medium.  3. A spectrum obtained
from  the  ambient  black  body  radiation
entering the  system. This background must
be considered in FT-IR systems, in  which the
 IR  beam  is  not  modulated before  it is
transmitted  along  the  path.    For  FT-IR
systems that do not use a separate source of
                                        11-1

-------
i!!!si:SS
 TECH&/mm

 infrared energy, the background is the source
 of infrared energy.

 Band pass  filter:  A  filtering  device that
 allows the transmission of only a specific
 band of  energies.  It can be wide band or
 narrow band.

 Bandwidth: The  width of a spectral feature
 as recorded by a spectroscopic instrument.
 This width is listed as the full  width at the
 half  maximum of the feature or as the half
 width at the half maximum of the spectral
 feature.  This is also referred to as the line
 width (Lengyel 1971).

 Beer's law:   Beer's law states  that  the
 intensity of a monochromatic  plane  wave
 incident on an absorbing medium of constant
 thickness diminishes exponentially with  the
 number of absorbers in the beam.  Strictly
 speaking,  Beer's  law  holds   only  if  the
 following conditions are met:   (1)  perfectly
 monochromatic radiation,  (2) no scattering,
 (3)   a  beam  that   is  strictly  collimated,
 (4)   negligible pressure-broadening effects
 (Pfeiffer  and  Liebhafsky 1951;  Lothian
 1963).  For an excellent  discussion of the
 derivation of Beer's law, see Penner (1959).

 Bistatic system:   A system in which the
 receiver   is   some  distance   from  the
transmitter. This  term is actually taken from
the field  of radar technology.   For remote
 sensing, this implies that the light source and
the detector are  separated and are at the
ends of the monitoring path.
                                                            TR-4423-99-03
                                 Broad band system: Any system that admits
                                 a broad  range  of  energies into its signal-
                                 processing section. Alternatively, a system
                                 that has  a flat response to a large  range of
                                 energies.

                                 Closed path: The optical path over which
                                 the beam travels in a sensor that is entirely
                                 enclosed. This is the case when White cells
                                 or Harriot cells are  used with the system.

                                 Cooler:  A device into which the  detector is
                                 placed  for  maintaining   it  at   a   low
                                 temperature in  an  IR  system.  At a  low
                                 temperature, the detector provides the high
                                 sensitivity that is required for the IR system.
                                 The two primary types of coolers are a liquid
                                 nitrogen  Dewar and a  closed-cycle Stirling
                                 cycle refrigerator.

                                 COSPEC:    An acronym  for  "correlation
                                 spectrometer."  This is a misnomer for the
                                 type of instrument implied.  It is really an
                                 instrument  that uses  a diffraction grating
                                 either in  the active, bistatic or the passive
                                 mode.   This instrument  is more correctly
                                 included  in the  class of instruments using
                                 differential   absorption   spectroscopy
                                 techniques.  It is used primarily to measure
                                 sulfur dioxide and nitrogen oxide.

                                 DIAL:    An  acronym   for  "differential
                                 absorption  lidar."   This system uses two
                                 pulsed frequencies from the same laser or
                                 from  different  lasers  to  measure  the
                                 concentration of gas over  a path. The two
                                 laser lines are at different positions within
                                 the absorption feature. The difference of the
                                 amount of light  backscattered at these two
                             11-2

-------
                                                                      TR-4423-99-03
 wavelengths is the quantity  used for  the
 measurement.

 DOAS:  An acronym for differential  optical
 absorption  spectroscopy.     A  technique
 whereby, in principle, any known difference
 in absorbance  is used  to determine  the
 concentration  of  a gas.   Generally,  the
 absorption difference is taken between  the
 spectral  line center and the wing.

 Electromagnetic spectrum:  The total of all
 possible   frequencies  of  electromagnetic
 radiation.  Different  sources may emit over
 different  frequency   regions.      All
 electromagnetic waves  travel at the same
 speed  in  free space (Halliday and Resnick
 1974).

 Fingerprint region:   The   region  of  the
 absorption spectrum of a molecule that
 essentially  allows   its   unequivocal
 identification.  This region covers the wave
 number  range  from 650   to  1300 cm'1
 (Willard et al. 1974).

 Flux:  The number or mass of particles or
 molecules that pass through a given unit area
 of surface per unit of time (Calvert 1990).

 Fourier transform: A mathematical transform
that  allows an aperiodic  function  to  be
expressed  as  an  integral  sum  over  a
continuous range of frequencies (Champeney
 1973).    The  Fourier  transform of the
interferogram produced  by the Michelson
interferometer in an FT-IR is the intensity as
a function of frequency.
 FT-IR: An abbreviation for "Fourier transform
 infrared." A spectroscopic instrument using
 the infrared portion of the  electromagnetic
 spectrum.  The  working component of this
 system is a Michelson interferometer.  To
 obtain the absorption spectrum as a function
 of frequency, a  Fourier transform of the
 output of  the  interferometer  must  be
 performed. For a brief overview of the FT-IR,
 see  the  publication  by  Nicolet (Nicolet
 Analytical Instruments 1986).   For  an in-
 depth  description of the FT-IR, see Griffiths
 and deHaseth  (1986).

 GASPEC:    An  acronym  for  "gas  filter
 correlation spectrometer."  The earliest of
 these  devices was  described  by Luft in
 1943, and they have been used  in various
 configurations ever  since.   The primary
 feature of these devices is a pair of gas cells.
 One   cell  contains  a  carefully  selected
 quantity of the target gas and  the other a
 gas that is spectroscopically inactive.  The
 difference in spectral transmittance of the
 two cells is an  indicator of the concentration
 of the target gas in the atmosphere (Ward
 and Zwick 1975).

 Infrared spectrum:   That  portion  of the
 electromagnetic  spectrum that  spans the
 region from  about   10  cm"1  to  about
 12,500 cm'1.  It is divided (Willard  et al.
 1974)  into (1)  the near-infrared region (from
 12,500 to 4000 cm'1), (2) the mid-infrared
region  (from 4000 to  650 cm'1), and (3) the
far-infrared region (from 650 to 10 cm'1).

Instrument   function:   The   function
superimposed  on the absorption  line shape
                                       11-3

-------
 TECHmuI^m
                                                                       TR-4423-99-03
 by  the  instrument.   This is sometimes
 referred to as the slit function,  a term taken
 from  instruments that use  slits  to  obtain
 resolution.

 Intensity: The radiant power per unit solid
 angle.  When the term "spectral intensity" is
 used, the units are  watts per steradian per
 nanometer. In  most spectroscopic literature,
 the term "intensity" is used to describe the
 power in a collimated beam of light in terms
 of power per unit area per unit wavelength.
 However, in  the  general  literature,  this
 definition  is more often used  for the term
 "irradiance," or "normal irradiance" (Calvert
 1990; Stone 1963).

 Interference:    The  physical   effects  of
 superimposing  two or more light waves.  The
 principle of superposition states that the total
• amplitude of the electromagnetic disturbance
 at a point is the vector sum of the  individual
 electromagnetic components incident there.
 For a two-component system of  collinear
 beams  of  the   same   amplitude,   the
 mathematical  description of the  result of
 addition is given by  I(p) =  2/0(1 + cos/4/),
 where 70 is the  intensity of either beam, and
 A   is  the  phase  difference  of  the  two
 components. The cosine term  is called the
 "interference term"  (Halliday and Resnick
 1974;  Stone  1963).   See also  "Spectral
 Interference."

 Interferogram:   The  effects of  interference
 that  are  detected  and  recorded  by  an
 interferometer;  the output of an  FT-IR  and
 the primary data that is collected and stored
 (Stone  1963; Griffiths and deHaseth 1986).
 Interferometer:   Any  of several  kinds  of
 instruments  used to produce  interference
 effects.  The Michelson interferometer used
 in FT-IR instruments is the most famous of a
 class  of  interferometers  that  produce
 interference by the division of an amplitude
 (Tolansky 1962).

 Irradiance: Radiant power per unit projected
 area of a specified surface. This has units of
 watts  per square centimeter.  The term
 "spectral irradiance"  is used to describe the
 irradiance as a function  of wavelength.  It
 has units of watts per square centimeter per
 nanometer (Calvert 1990).

 Laser:    An  acronym  for the  term  "light
 amplification   by  stimulated  emission  of
 radiation".  A source of  light that  is highly
 coherent,  both   spatially  and  temporally
 (Lengyel  1971).

 LIDAR:   An  acronym  for  the term  "light
 detection and ranging"  (Calvert 1990).  A
 technique for (1)  detecting the presence  of
 gases  and   aerosols  by  measuring  the
 backscattered portion of a laser beam and
 (2) determining the range of these gases and
 aerosols by electronically gating the detected
 signal and performing a calculation based on
the speed of light (about 30 cm/ns).

 Light: Strictly, light is defined as that portion
of the electromagnetic spectrum that causes
the sensation  of vision. It extends from about
 25,000 cm'1 to about 14,300 cm'1 (Halliday
and Resnick 1 974).
                                        11-4

-------
                                                                      TR-4423-99-03
 Light scattering:  The  redirection of light
 waves due to interaction with molecules or
 aerosols.  If the size of the body causing the
 scattering  is   small  compared  to  the
 wavelength of the incident  radiation, the
 scattering is termed "Raleigh scattering."  If
 the size is large compared to the wavelength,
 the scattering is termed "Mie scattering."
 Scattering by molecules is generally Raleigh
 scattering, and scattering by aerosols is Mie
 scattering.   As light  travels through  a
 medium, two physical processes diminish the
 intensity in the forward direction - scattering
 and  absorption.   The sum of these two
 effects  is called extinction  (van  de Hulst
 1981).

 Long-path monitoring:  This is a monitoring
 technique that uses an extended, open path.
 LIDAR systems  can  make  measurements
 over  a  path  length  of  a few kilometers.
 DOAS systems can make measurements of
 ozone over a path of up to 2 km. The FT-IR
 systems customarily  use paths with length
 less than 1  km.

 Minimum  detection limit:   The  minimum
 concentration of a compound that can be
 detected by  an  instrument with  a  given
 statistical probability.  Usually the detection
 limit  is  given as  3  times  the  standard
 deviation of the noise in the system. In this
case,  the minimum concentration can be
detected with a probability of 99.7% (Calvert
 1990; Long and Winefordner 1983).

Monitoring path:  The actual path in space
over which the  pollutant concentration is
measured and averaged.
 Monitoring path length:  The length  of the
 monitoring  path  in  the atmosphere over
 which the average pollutant concentration
 mea-surement   is   determined   (U.S.
 Environmental Protection Agency 1994).

 Monostatic  system:  A system  with the
 source and the receiver at the same end of
 the path.  For FT-IR systems and for  DOAS
 systems, the beam is generally returned by a
 retroreflector.   For LIDAR  systems,  the
 backscattered portion of the laser beam is
 measured directly.

 Open-ended system: A system in which the
 remote  sensor uses light  reflected  from
 targets of opportunity (walls, trees, etc.) or
 skylight as a  source.

 Open-path analyzer: An automated analytical
 instrument that  is  used for a method  of
 measuring the average atmospheric pollutant
 concentration in  situ along  one or more
 monitoring paths  that are 5 m or more in
 length  (U.S.  Environmental   Protection
 Agency 1994).

 Open-path monitoring: Remote sensing over
 a  path  that  is  completely  open to  the
 atmosphere.  Thus, the concentration of a
 particular  gas in the beam  path can be
 changed by winds and diffusion.  The open
 path is the most frequently used in remote
 sensing.

 Optical remote sensing: A generic term used
to describe   any  of  a  number of optical
 measurement techniques that measure some
quantity or constituent of the atmosphere.
                                       11-5

-------

                                                                      TR-4423-99-03
 These  techniques  include  DIAL,  DOAS,
 FT-IR, GASPEC, LIDAR, etc. One thing that
 is common to  all the techniques  is that no
 sample must be collected.

 Parts per million meters:  The unit term for
 the  quantity  that  is  measured  by many
 remote sensors. It is the unit  associated
 with   the   quantity   path-integrated
 concentration.  It is a possible unit of choice
 for  reporting  data from remote  sensors
 because it is independent of the path length.

 Passive system: Any system that does not
 radiate energy to its surroundings.

 Path-averaged concentration:  The result of
 dividing the path-integrated concentration by
 the   path   length.      Path-averaged
 concentration gives  the average value of the
 concentration along the path (McClenny and
 Russwurm'1978).

 Path-integrated concentration:  The quantity
 that is measured by a remote sensor over a
 long path. It has units of concentration times
 length.

 Plume:  The gaseous  and aerosol effluents
 emitted from a chimney or other source and
 the volume of space  they occupy.  The shape
 of  a  plume  and  the  concentration  of
 pollutants within it are  very  sensitive  to
 meteorological conditions (Calvert  1990).

 Point analyzer:  An automated  analytical
 method  that  measures   pollutant
concentration  in an  ambient air  sample
extracted from the atmosphere at a specific
 inlet  probe  point (proposed changes  [U.S.
 Environmental Protection Agency 1994]).

 Probe: The actual inlet where an air sample
 is extracted from the atmosphere for delivery
 to a  sampler or point  analyzer (proposed
 changes  [U.S.  Environmental  Protection
 Agency 1994]).

 Radiometry:  The  measurement of various
 quantities such as  intensity associated with
 radiant  energy  (Calvert 1990).  This  is in
 contrast to  the term photometry, which
 assumes  the spectral  sensitivity  of  the
 human eye as the detector (Walsh 1965).

 Real-time system:   Any monitoring system
 that acquires and records data at a rate that
 is  comparable  to  the  rate at which the
 concentration is changing.

 Reference   spectra:     Spectra   of   the
 absorbance versus wave  number for a pure
 sample of a set of gases. The  spectra are
 obtained  under  controlled  conditions  of
 pressure and temperature and  with known
 concentrations.   For most instruments, the
 pure  sample is pressure-broadened  with
 nitrogen  so  that  the   spectra   are
 representative of atmospherically broadened
 lines.  These spectra are  used for obtaining
the unknown concentrations of gases in
 ambient air samples.

 Resolution:  The minimum  separation  that
two spectral  features can have  and still, in
some  manner, be  distinguished from  one
another.  A commonly used requirement for
two spectral features to be considered just
                                       11-6

-------
MAm
                                                                       TR-4423-99-03
 resolved is the Raleigh criterion.  This states
 that two features are just resolved when the
 maximum intensity of one falls  at the first
 minimum of the other (Jenkins and White
 1950; Tolansky  1962).  This definition of
 resolution and the Raleigh criterion are also
 valid for the FT-IR, although there is another
 definition in common use for this technique.
 This  definition  states  that  the minimum
 separation in wave numbers of two spectral
 features that   can   be  resolved   is   the
 reciprocal  of the maximum  optical  path
 difference   (in   centimeters)  of the  two
 interferometer  mirrors  employed (Griffiths
 and  deHaseth 1986; Nicolet 1986).

 Retroreflector:     The  CIE   (Commission
 Internationale   de   I'Eclairage)   defines
 retroreflection  as  "radiation  returned  in
 directions close to the direction from which
 it came, this property being maintained  over
 wide variations  of  the  direction  of   the
 incident radiation."  Retroreflector devices
 come in a variety of forms and  have many
 uses.   The  one  commonly  described by
 workers in remote sensing uses total internal
 reflection from three  mutually perpendicular
 surfaces.    This  kind  of retroreflector  is
 usually  called  a  corner cube  or prismatic
 retroreflector (Rennilson 1980).
                                           several lasers that also are used as sources
                                           for DIAL and LIDAR instruments.

                                           Spectral intensity:  See "Intensity."

                                           Spectral interference: When the absorbance
                                           features from two or more gases cover the
                                           same wave number  regions, the  gases are
                                           said to exhibit spectral interference.  Water
                                           vapor   produces  the  strongest  spectral
                                           interference   for  infrared   spectroscopic
                                           instruments that take atmospheric data.

                                           Synthetic background:  A spectrum that is
                                           made from a field spectrum by choosing
                                           points along the baseline  and connecting
                                           them with  a high-order polynomial or short,
                                           straight lines.  The synthetic background is
                                           then used to find the absorbance spectrum.

                                           Truncation: The act of stopping  a process
                                           before   it   is  complete.      In   FT-IR
                                           spectrometers, the theoretically infinite scale
                                           of the interferogram is truncated by the finite
                                           movement  of the interferometer mirror.

                                           Unistatic system:  A system that has the
                                           source and the receiver at the same place;
                                           now  more  commonly  referred  to as  a
                                           monostatic system.
Slit function: See "Instrument function."

Source:   The  device  that  supplies the
electromagnetic energy  for  the  various
instruments  used to measure atmospheric
gases. These generally are a Nernst glower
or globar for the infrared region or a xenon
arc lamp for  the ultraviolet region.  There are
                                          Wave number:  The number  of waves per
                                          centimeter. This term has units of reciprocal
                                          centimeters (cm'1).
                                          11.3  References

                                          Calvert,  J.G.
                                          Atmospheric
1990.     Glossary  of
Chemistry   Terms
                                        11-7

-------
                                                                     TR-4423-99-03
 (recommendations 1990). PureAppl. Chem.
 62 (111:2167-2219.

 Champeney, D.C. 1973. Fourier Transforms
 and Their Physical Applications.  Academic
 Press, London.

 Griffiths, P.R.,  and J.A. deHaseth.  1986.
 Fourier Transform  Infrared Spectrometry.
 John Wiley and Sons, New York.

 Halliday,  D.,   and  R.  Resnick.    1974.
 Fundamentals of Physics.  Wiley and Sons,
 New York.

 Jenkins, F.A.,   and  H.E.  White.   1950.
 Fundamentals of Optics. McGraw-Hill, New
 York.

 Long,  G.L.,  and J.D. Winefordner.  1983.
 Limit of Detection:  A Closer Look at the
 IUPAC Definition. Anal. Chem. 55(7): 712A-
 724A.

 Lothian, G.F.  1963.  Beer's  Law and Its Use
 in Analysis.  Analyst 88:678.

 Lengyel, B.A. 1971.  Lasers, 2nd Ed. Wiley-
 Interscience, New York.

 Luft,  K.F.  1943. Uber Eine Neue Methode
 der Registrierenden Gas Analyze Mit Hilfe der
 Absorption Ultraroter Stahlen Ohne Spektrale
 Zerlegung. J. Tech. Phys. 24:97.

 McClenny,  W.A.,  and  G.M. Russwurm.
 1978. Laser-Based Long Path  Monitoring of
 Ambient Gases - Analysis of Two Systems.
A tmos. Environ.  1 2:1443.
 Nicolet Analytical Instruments.  1986. FT-IR
 Theory.   Nicolet  Analytical  Instruments,
 Madison, Wl.

 Penner, S.S.  1959.  Quantitative Molecular
 Spectroscopy and Gas Emissivities. Addison-
 Wesley, Reading, MA.

 Pfeiffer, H.G., and  H.A.  Liebhafsky.  1951.
 The Origins of Beer's Law.  J. Chem. Educ.
 28:123-125.
 Rennilson,  J.J.
 Measurements:
 19:1234.
 1980.    Retroreflection
A  Review.   Appl.  Opt.
Stone, J.M.  1963.  Radiation and Optics.
McGraw-Hill, New York.

Tolansky, S.  1962.  An  Introduction  to
Interferometry.  John Wiley and Sons, New
York.

U.S.  Environmental  Protection   Agency.
1994. Ambient air quality surveillance siting
criteria for open path analyzers  (proposed
rule). Fed. Reg. 59(1 59):42541-42552.

van de Hulst, H.C.   1981.  Light Scattering
by Small Particles.  Dover Publications, New
York.

Walsh, J.W.T.  1965.  Photometry.  Dover
Publications, New York.

Ward, T.V.,  and  H.H. Zwick.  1975.  Gas
Cell  Correlation  Spectrometer:   GASPEC.
Appl. Opt. 14:2896.

Willard,  H.H., L.L.  Merritt, and J.A. Dean.
1974.  Instrumental Methods of Analysis,
5th Ed.  D. Van Nostrand, Princeton, NJ.
                                      11-8

-------
                                                                       TR-4423-99-03
                                    Chapter 12
                                    Bibliography
 12.1   Introduction and Overview

       This chapter contains a bibliography
 of journal  articles, books, and  conference
 proceedings  that  address FT-IR long-path
 monitoring, as well as general references that
 present discussions of the basic principles of
 FT-IR spectrometry. This is  not a reference
 section for literature cited in the text of this
 guidance   document,   but  is  a  general
 bibliography. References in this document are
 listed at the end of the chapter in which they.
 are cited.

       This bibliography  is by  no  means
 exhaustive, but is  intended to give a broad
 overview  of the available  literature  in the
 FT-IR long-path discipline. During the initial
 phase of this overview, a computer-assisted
 on-line literature search was conducted for
 citations listed in Chemical Abstracts through
 September 29, 1992. A subsequent  on-line
 search was conducted  for citations listed in
 Chemical Abstracts from 1 992 through July
 22, 1 994. These literature searches focused
 on,  but  were not  limited  to,  current
 publications regarding FT-IR long-path, open-
 path  monitoring,   in  keeping  with  the
emphasis  of  this document. For this latest
edition, a literature search of the documents
available through  NTIS  in Springfield,  VA,
was   conducted   through  1998.   Other
citations were  gleaned  from the reference
sections of articles on file.  Several early
citations  are  also given  to  provide  an
 important, historical perspective  into long-
 path IR monitoring in environmental analysis.
 Only articles that are on file in our laboratory
 are   listed   in   the   bibliography.  This
 bibliography will continue to be updated as
 revisions to this document are made.

       Several of the more recent citations
 are from conference proceedings, which may
 limit  their  general  availability.  They are
 included to give the reader access to current
 research in the field that has in many cases
 not  yet   appeared  in  the peer-reviewed
 literature. Also, inclusion of the conference
 proceedings  provides  the   reader  with
 pertinent information regarding conferences
 and meetings that  typically address FT-IR
 long-path  or remote-sensing issues.

 12.2   Publications

 Amoto, I.  1988.  Remote Sensing: A Distant
 View  of   Chemistry.   Anal.   Chem.
 60(23): 1339A-1344A.

 Anderson,  R.J.,  and  P.R.  Griffiths.  1975.
 Errors  in   Absorbance   Measurements  in
 Infrared  Fourier  Transform  Spectrometry
 Because  of  Limited Instrument Resolution.
Anal.  Chem. 47(14):2339-2347.

Andreas, E.L., J.R. Gosz, and C.N. Dahm.
 1992.  Can  Long-Path  FTIR Spectroscopy
Yield  Gas  Flux  Measurements Through  a
Variance   Technique?  Atmos.   Environ.
26A(2):225-233.

Bangalore, A.S., G.W. Small,  R.J. Combs,
R.B. Knapp, R.T. Kroutil, C.A.  Traynor, and
                                        12-1

-------
 TECH!
                                                                    TR-4423-99-03
 J.D.  Ko.  1997.  Automated  Detection of
 Trichloroethylene  by  Fourier  Transform
 Infrared  Remote Sensing  Measurements.
 Anal. Chem. 26:118-129.

 Barnes,  H.W.,   Jr.,  W.F.   Herget,  and
 R. Rollins. 1974. Remote Sensing  of Sulfur
 Dioxide  in  Power  Plant  Plumes  Using
 Ultraviolet Absorption and Infrared  Emission
 Spectroscopy. Analytical Methods Applied to
 Air Pollution Measurements  (R.K.  Stevens
 and W.F. Herget, Eds.), Ann Arbor  Science,
 Ann Arbor, Ml, pp. 245-266.

 Barnett,  R.W.,  and  B.T.  Smith.  1994.
 Modeling   Background  Temperatures  in
 Passive Infrared  Remote  Sensing.  SP-89
 Optical   Sensing   for   Environmental
 Monitoring,  Air  &  Waste   Management
 Association, Pittsburgh, PA, pp. 293-304.

 Batterman, S.A., C. Peng, and P.  Milne.
 1994. Sequential  Extractive  Sampling of
 Indoor  Air   Contaminants  Using  FT-IR
 Spectroscopy.  SP-89 Optical Sensing for
 Environmental Monitoring,  Air  & Waste
 Management  Association,  Pittsburgh,  PA,
 pp. 242-254.

 Beer,  A. 1852. Ann. Physik 86:78.

 Beer,  R. 1992.  Remote Sensing by Fourier
 Transform Spectrometry,  John Wiley  and
 Sons, New York.

 Beil,  A., R. Daum, R. Harig, and G. Matz.
 1998. Remote  Sensing  of  Atmospheric
 Pollution by Passive FTIR Spectrometry. Proc.
 SP/£3493:32-43.

 Bell,   R.J.   1972.   Introductory   Fourier
 Transform Spectroscopy,  Academic Press,
 New York.

 Bennett, C.L.  1994. FTIR Measurements of
Thermal   Infrared   Sky   Radiance   and
Transmission. Department of Energy Report
 UCRL-117864.
 Bhattacharyya,  R., and  L.A. Todd.  1995.
 Two   Dimensional   Mapping   of  Air
 Contaminant   Movement   Using   a
 Tomographic System. Proceedings of the
 SP/E Specialty Conference Optical Sensing
 for Environmental and Process Monitoring,
 McClean,   VA,   International  Society for
 Optical Engineering, Bellingham, WA.

 Biermann, H.W., E.G. Tuazon, A.M. Winer,
 T.J. Wellington,  and  J.N.  Pitts, Jr.  1988.
 Simultaneous Absolute  Measurements  of
 Gaseous Nitrogen Species in Urban Ambient
 Air  by  Long   Pathlength  Infrared and
 Ultraviolet-Visible  Spectroscopy.  Atmos.
 Environ. 22(8): 1 545-1 554.

 Bishop,  G.A., J.R. Starkey,  A. Ihlenfeldt,
 W.J. Williams, and  D.H. Stedman. 1989. IR
 Long-Path Photometry: A Remote Sensing
 Tool for Automobile Emissions. Anal. Chem.
 61(10):671A-677A.

 Bishop, G.A., and D.H. Stedman. 1990. On-
 Road   Carbon   Monoxide   Emission
 Measurement Comparisons  for the  1988-
 1989 Colorado Oxy-Fuels Program. Environ.
 Sci. Technol. 24(6):843-847.

 Bishop, G.A., D.H. Stedman, and T. Jessop.
 1992.   Infrared   Emission   and  Remote
 Sensing.  J. Air Waste  Manage. Assoc.
 42(5):695-697.

 Bittner,  H.,  T.  Eisenmann,  H.  Mosebach,
 M.   Erhard,  and  M.   Resch.   1994.
 Measurements  of  Diffuse  Emissions of
Volatile   Organic  Compounds  by   High
 Resolution  FTIR  Remote  Sensing.  SP-89
 Optical   Sensing   for   Environmental
Monitoring,  Air  & Waste  Management
Association, Pittsburgh, PA, pp. 443-454.

Blumenstock,T.,  H. Fischer, A.  Friedle,  F.
Hase,   and  P.   Thomas.   1997.  Column
Amounts of CIONO2 HCI, HN03 and HFfrom
Ground  Based FTIR Measurements  Made
                                      12-2

-------
                                                                     TR-4423-99-03
 near Kiruna, Sweden in Late Winter 1994.
 J. Atmos. Chem. 26 (3):311-321.

 Brandon,   R.W.,   S.D.   Garbis,   and
 R.H.   Kagann.   1992.   Quantitative  Gas
 Standards for the Calibration of Open  Path
 Optical Sensors.  SP-81  Optical  Remote
 Sensing. Applications to Environmental and
 Industrial Safety Problems,  Air  &  Waste
 Management Association, Pittsburgh,  PA,
 pp. 434-445.

 Briz,  S.,  A.J.  De  Castro,  J.  Melendez,
 J. Meneses,  J.M.  Aranda, and  F.  Lopez.
 1997.  Proc. SP/E3106:159-170.

 Brown, C.W., P.P. Lynch, R.J. Obremski, and
 D.S. Lavery. 1982. Matrix Representations
 and   Criteria   for   Selecting   Analytical
 Wavelengths    for    Multicomponent
 Spectroscopic   Analysis.  Anal.   Chem.
 54(9):1472-1479.

 Byrd,  L.A.B. 1994.  Ambient  Air Monitoring
 Siting  Criteria  for  Open Path  Analyzers
 Measuring  Nitrogen Dioxide,  Ozone,  and
 Sulfur  Dioxide.  SP-89  Optical Sensing for
 Environmental  Monitoring,  Air  &  Waste
 Management  Association, Pittsburgh,  PA,
 pp. 349-357.

 Calvert, J.G. 1990. Glossary of Atmospheric
 Chemistry Terms (recommendations 1990).
 PureAppl. Chem. 62(11):21 67-221 9.

 Cantu, A., G. Pophal, S. Hall, and C.T. Laush
 1998.  Unique Application of an  Extractive
 FTIR Ambient Air Monitoring System for the
 Simultaneous Detection of Multiple ppb Level
 VOC's.  Appl.   Phys.  B.  Lasers   Opt.   67
 (4):493-496.

 Carlson, R.C., A.F. Hayden, and W.B. Telfair.
 1 988. Remote Observation of Effluents from
Small   Building  Smokestacks  Using FT-IR
Spectroscopy.   Appl.   Opt.   27(23):4952-
4959.
 Carter,  R.E., Jr., D.D. Lane,  G.A.  Marotz,
 M.J.  Thomas, and  J.L. Hudson. 1992. A
 Method of  Interconversion Between Point
 and  Path-Averaged  Ambient   Air  VOC
 Concentrations,  Using Wind  Data. SP-8J
 Optical Remote Sensing. Applications  to
 Environmental   and   Industrial   Safety
 Problems,   Air  &   Waste   Management
 Association, Pittsburgh, PA, pp. 529-540.

 Carter,  R.E., Jr., D.D. Lane,  G.A.  Marotz,
 C.T.  Chaffin, T.L.  Marshall,  M.  Tucker,
 M.R.   Witkowski,  R.M.   Hammaker,
 W.G.   Fateley,   M.J.   Thomas,   and
 J.L. Hudson. 1993.  A Method of  Predicting
 Point and  Path-Averaged  Ambient Air VOC
 Concentrations, Using Meteorological Data.
 J. Air Waste Manage. Assoc. 43:480-488.

 Carter, R.E., Jr., M.J. Thomas, G.A.  Marotz,
 D.D.  Lane,  and   J.L.   Hudson.   1992.
 Compound  Detection  and  Concentration
 Estimation by Open-Path  Fourier Transform
 Infrared Spectrometry and Canisters Under
 Controlled  Field  Conditions. Environ.  Sci.
 Technol. 26(111:2175-2181.

 Carter,  R.O.,   III,   N.E.  Lindsay,  and
 D.  Beduhn.  1990. A  Solution to Baseline
 Uncertainty  Due  to   MCT   Detector
 Nonlinearity  in  FT-IR.  Appl.  Spectrosc.
 44(71:1147-1151.

 Chaffin, T., T.L.  Marshall, J.M.  Poholarz,
 M.D.  Tucker, M.  Hammaker, and W.G.
 Fately. 1993. What Height for Remote FTIR
 Sensing of  Volatile Organic Compounds
 Emitted  near Ground Level. Proceedings of
 the 86th Annual Meeting and Exhibition, Air &
Waste Management Association, Pittsburgh,
PA.

Chaffin,   C.T.,   Jr.,   W.G.   Fateley,
M.D. Tucker, and R.M. Hammaker.  1 994. An
Alternative Sampling Technique for Fourier
Transform Infrared (FT-IR) Remote Sensing of
Fugitive  Emissions  from   Industrial Sites.
SP-89  Optical Sensing  for Environmental
                                       12-3

-------

                                                                    TR-4423-99-03
 Monitoring,  Air  &  Waste  Management
 Association, Pittsburgh, PA, pp. 814-825.

 Chaffin,  C.T., T.L.  Marshall,  R.J.  Combs,
 R.B. Knapp, R.T. Kroutil, W.G. Fately, and
 R.M.  Hammaker.  1995.  Passive  Fourier
 Transform  (FTIR)  Monitoring  of  S02  in
 Plumes: A  Comparison of Remote  Passive
 Spectra  of  an  Actual  Emission Spectra
 Collected with a Heatable Cell.  Proc. SPIE
 2365:303-313.

 Champeney, D.C. 1973.  Fourier Transforms
 and Their Physical Applications.  Academic
 Press, London.

 Chakraborty, O.K. 1995. Examination of the
 Long Path Open Air FTIR Technique for Air in
 the State of  Kentucky.  Proc. SPIE 2365:
 347-358.

 Chaney,   L.W.   1983.   The   Remote
 Measurement of Traffic  Generated  Carbon
 Monoxide.  J.  Air Pollut.  Control  Assoc.
 33(3):220-222.

 Chang,   S.-Y.,   and  T.-L.  Tso.   1994.
 Measurement of the Taiwan Ambient Trace
 Gas Concentration by Kilometer-Path Length
 Fourier-Transform  Infrared  Spectroscopy.
Anal. Sci. 10(1): 193-201.

 Chang,  S.-Y.,  and T.-L. Tso.  1994.  Field
 Applications of Long  Infrared Absorption in
Taiwan.   SP-89   Optical   Sensing  for
Environmental  Monitoring,  Air  &  Waste
 Management  Association,  Pittsburgh, PA,
 pp. 157-181.

Childers,  J.W.    1993.   Resolution
Considerations in Long-Path, Open-Path FT-IR
Spectrometry.   Paper   93-RA-121.05,
Proceedings of the 86th Annual Meeting and
Exhibition,  Air  &   Waste  Management
Association, Pittsburgh, PA.

Childers, J.W., and E.L. Thompson, Jr. 1 994.
Resolution Requirements  in Long-Path FT-IR
 Spectrometry. SP-89 Optical  Sensing  for
 Environmental Monitoring,  Air & Waste
 Management Association, Pittsburgh,  PA,
 pp. 38-46.

 Childers,  J.W.,  G.M.   Russwurm,   and
 E.L. Thompson,  Jr.  1996.  Instrumental
 Parameters and Their Effect on Open-Path
 FT-IR Data. Proceedings of the 89th Annual
 Meeting  & Exhibition of  the Air & Waste
 Management Association.  Paper 96-MP5.07,
 Air  &  Waste  Management  Association.
 Pittsburgh, PA.

 Childers,  J.W.,  G.M.   Russwurm,   and
 E.L. Thompson.  1997.  QA/QC Issues in
 OP/FTIR  Monitoring.  Proceedings  of  the
 1997 Annual Meeting AWMA, Air & Waste
 Management Association, Pittsburgh, PA.

 Childers L.O. 1996.  USEPA QA Auditor Is
 Scheduled for a Visit. What Can I Expect?
 Proceedings  of   the   1995  Specialty
 Conference In San Francisco,  Air & Waste
 Management Association, Pittsburgh, PA.

 Chu, P.M., G.C.  Rhoderick, W.J. Lafferty,
 F.  R.  Guenther,and  S.J. Wetzel. 1997.
 Quantitative Infrared Data Base of Hazardous
 Air Pollutants.  Proceedings  of the 89th
Annual Meeting of A WMA in Nashville, Tenn.
 Air & Waste  Management  Association,
 Pittsburgh, PA.

 Chughtai, A.R., and D.M. Smith. 1 991. Long
Optical Path  Cell for Photochemical Kinetics
in  Heterogeneous   Systems  of   Low
Concentration. Appl. Spectrosc. 45(7):1204-
 1207.

Clark,  J.M.  1994.   Mercury  Cadmium
Telluride Cryocooled Detector  Performance
Parameters for FTIR  Spectroscopy. SP-89
Optical   Sensing   for   Environmental
Monitoring,  Air  &  Waste  Management
Association,  Pittsburgh, PA, pp. 591-606.
                                      12-4

-------

                                                                     TR-4423-99-03
 Collins,  J.D.,   and  L.A.  Todd.  1992.
 Evaluation   of   Infrared   Optical  Remote
 Sensing Equipment in an Exposure Chamber.
 SP-81 Optical Remote Sensing. Applications
 to  Environmental  and  Industrial  Safety
 Problems,   Air   &  Waste  Management
 Association, Pittsburgh, PA, pp. 351-355.

 Cronin, J.T.  1992.  Stack-Gas  Monitoring
 Using  FT-IR Spectroscopy.  Spectroscopy
 7(5):33-39.

 Dando,    N.R.,   T.O.   Montgomery,
 L.A.  Schneider,  and  J.E.  Gibb.  1994.
 Applications of Open-Path FTIR Spectroscopy
 for Characterizing Fugitive Emissions at Metal
 Production  Plants. SP-89 Optical Sensing for
 Environmental  Monitoring, Air  &  Waste
 Management Association,  Pittsburgh,  PA,
 pp. 123-133.

 Demirgian,  J.C., and M.D. Erickson. 1990.
 The   Potential   of   Continuous   Emission
 Monitoring  of Hazardous Waste Incinerators
 Using   Fourier   Transform   Infrared
 Spectroscopy. Waste Manage. 10:227.

 Demirgian,    J.C.,   C.L.   Hammer,  and
 R.T. Kroutil. 1992. The Potential of Passive-
 Remote    Fourier  Transform   Infrared
 Spectroscopy to Detect Organic  Emissions
 Under  the  Clean Air Act. SP-81  Optical
 Remote    Sensing.   Applications    to
 Environmental   and  Industrial  Safety
 Problems,  Air   &   Waste   Management
 Association, Pittsburgh, PA, pp. 464-476.

 Demirgian,  J.C., C. Hammer, E. Hwang, and
 Z. Mao. 1994. Advances in Passive-Remote
 and  Extractive  Fourier Transform Infrared
 Systems.   SP-89  Optical   Sensing   for
Environmental Monitoring,  Air  &  Waste
 Management Association, Pittsburgh,  PA,
pp. 780-790.

 Douard,  M.,  J.  Zentzius-Reitz,  T.  Lamp.
A. Ropertz, and K.  Weber. 1997.  Quality
Assurance Procedures and Measurements for
 Open Path FTIR Spectroscopy. Proc.  SPIE
 3107:114-125.

 Dowling, J.A. 1994. A Review of the Naval
 Research Laboratory Program of Long-Path
 Air Measurements  Using Lasers and FT-IR.
 SP-89 Optical  Sensing  for Environmental
 Monitoring,  Air  &  Waste  Management
 Association, Pittsburgh, PA, pp. 145-156.

 Draves, J.A., J.P.  LaCosse, D.M. Hull, and
 R.L. Spellicy. 1992. A Comparison of Open
 Path Fourier Transform Infrared Spectrometry
 with Conventional  Ambient Air Monitoring
 Methods.  SP-81 Optical Remote  Sensing.
 Applications to Environmental and Industrial
 Safety Problems, Air & Waste Management
 Association, Pittsburgh, PA, pp. 252-265.

 Dresscher,   A.C.,  M.G.  Yost,  D.Y.Park,
 S.P. Levine, A.J. Gadgil, M.L.Fischer, and
 W.W.  Nazaroff.  1995.  Measurement  of
 Tracer Gas Distributions Using an Open Path
 Coupled   with   Computed   Topography.
 Proceedings of Specialty SPIE Conference in
 McClean,   Va.  International   Society  for
 Optical Engineering, Bellingham, WA.

 Dubois, A.E. , J.W.  Engle, P.L. McKane, and
 S.H. Perry. 1996.  Open Path FTIR Quality
 Assurance   Data:   Demonstration   of
 Technology   Reliability.   Proceedings   of
AWMA   Specialty  Conference  in   San
 Francisco.  Air   &  Waste   Management
 Association, Pittsburgh, PA.

 Edney, E.O., J.W.  Spence, and P.L. Hanst.
 1979.  Synthesis and  Thermal  Stability  of
 Peroxy Alkyl Nitrates. J. Air Pollut.  Control
Assoc. 29(7):741-743.

 Egert, S.,  D. Peri  and Y. Danziger.  1996.
 Efficient Monitoring of Toxic Gases over an
 Industrial  Zone  Using  Remote Sensors.
Proceedings  of   SPIE/AWMA  Specialty
 Conference in San  Francisco.  Air & Waste
Management Association, Pittsburgh, PA.
                                       12-5

-------
                                                                     TR-4423-99-03
 Eisenmann, T.,  H. Mosebach,  H. Bittner,
 M. Resch, R. Haus, and K. Schafer. 1993.
 Selected  Applications  of  Remote  Sensing
 Measurements  with  the  Double Pendulum
 Interferometer  in  Germany  with  Special
 Consideration of Quality Assurance Aspects.
 Paper  93-FA-165.01,  Proceedings of the
 86th Annual Meeting and Exhibition,  Air &
 Waste Management Association, Pittsburgh,
 PA.

 Eisenmann, T., H. Mosebach, and H. Bittner.
 1994.   Remote  Sensing  FTIR-System  for
 Emission Monitoring and Ambient Air Control
 of  Atmospheric  Trace   Gases  and  Air
 Pollutants.    Erdoel   Erdgas   Kohle
 110(1):28-34.

 Fischer,   H.  1992.  Remote  Sensing  of
 Atmospheric   Trace  Constituents   Using
 Fourier   Transform   Spectrometry.   Ber.
 Bunsen-Ges. Phys. Chem. 96(3):306-314.

 Flanagen, J., R. Shores and S. Thorneloe,
 1996.  Uncertainity Estimate for Open  Path
 Remote  Sensing  of   Fugitive  Emissions.
 EPA/600/A-96/067,   U.S.   Environmental
 Protection Agency.

 Franzblau,  A.,  S.P. Levine, L.A.  Burgess,
 A.S. Qu,  R.M.  Schreck,  and J.B.  D'Arcy.
 1992.  The Use  of a Transportable Fourier
 Transform  Infrared (FTIR) Spectrometer for
 the Direct Measurement of Solvents in Breath
 and Ambient Air - I: Methanol. Am. Ind. Hyg.
Assoc. J.  53(4):221-7.

 Garbis, S.D., and R.W. Brandon.  1994. An
 Assessment of Dynamic Equilibrium Methods
 of Generating  FTIR  Reference  Spectra.,
 SP-89  Optical  Sensing for Environmental
 Monitoring,  Air  &  Waste  Management
 Association, Pittsburgh, PA, pp. 281-292.

.Gardner, D.G., W.J. Phillips, D.J. Gonnion,
 L.T.  Lay,  and R.  Dishakjian.  1994.  High
Temperature  Reference  Spectra.   SP-89
 Optical   Sensing   for    Environmental
 Monitoring,  Air  &  Waste  Management
 Association, Pittsburgh, PA, pp. 266-280.

 Gay,  B.W.,  P.L.  Hanst, J.J.  Bufalini, and
 R.C. Noonan. 1976. Atmospheric Oxidation
 of  Chlorinated  Ethylenes.  Environ.  Sci.
 Techno/. 10(1):58-67.

 Gay, B.W., R.C. Noonan, J.J. Bufalini, and
 P.L. Hanst. 1976. Photochemical Synthesis
 of  Peroxyacyl  Nitrates  in  Gas  Phase  via
 Chlorine-Aldehyde  Reaction. Environ.  Sci.
 Techno!. 10(1):82-85.

 Gay, B.W., Jr., R.C. Noonan, P.L. Hanst, and
 J.J. Bufalini.  1975.  Photolysis of  Alkyl
 Nitrites   and   Benzyl   Nitrite   at  Low
 Concentrations—An Infrared Study. Removal
 of  Trace Contaminants from the Air, ACS
 Symp. Ser. 17:132-151.

 Giese-Bogdan, S., M. Simonds-Malachowski,
 and S.P. Levine.  1994. Evaluation of the
 Role of  Fixed  Beam Open Path  Fourier
 Transform  Infrared   Spectroscopy  in Air
 Monitoring Strategies. SP-89 Optical Sensing
 for Environmental Monitoring, Air & Waste
 Management Association, Pittsburgh, PA, pp.
 263-265.

 Gosz, J.R.,  C.L.  Dahm,  and  P.G.  Risser.
 1988.  Long-Path  FTIR  Measurement  of
 Atmospheric  Trace  Gas  Concentrations.
 Ecology 69(5)^ 326-1 330.

 Gosz, J.R.,  D.I. Moore,  C.N. Dahm,  and
 S. Hofstadler. 1990. Field Testing Long-Path
 Fourier    Transform  Infrared   (FTIR)
 Spectroscopy   for   Measurements   of
 Atmospheric Gas Concentrations. Remote
 Sens. Environ. 32:103-110.

 Grant,   W.B.,   Kagann,   R.H.,  and
W.A.  McClenny.  1992.   Optical  Remote
 Measurement of Toxic Gases. J. Air Waste
Manage. Assoc. 42(1 ):1 8-30.
                                       12-6

-------
                                                                     TR-4423-99-03
 Grant, W.B., and  R.T.  Menzies.  1983. A
 Survey  of  Laser  and  Selected  Optical
 Systems  for   Remote   Measurement  of
 Pollutant Gas Concentrations. J. Air Pollut.
 ControlAssoc. 33(3): 187-1 94.

 Grasselli,   J.G.,   P.R.   Griffiths,  and
 R.W. Hannah. 1982. Criteria for Presentation
 of   Spectra   from   Computerized   IR
 Instruments. Appl.  Spectrosc. 36(2):87-91.

 Gravel,  D.,  A.  Rilling,  V.  Karfik,  and
 M.   Schmaeh.   1997.  Fourier Transform
 Infrared Spectrometry—A Mature Analytical
 Method   for   Industrial-Level   Emission
 Monitoring.  Proceedings  of  the  AWMA
 Annual Meeting  in Toronto,  Canada, Air &
 Waste Management Association, Pittsburgh,
 PA.

 Green, M., Seiber, J.N., and Biermann, H.W.
 1991.  In  Situ  Measurement  of  Methyl
 Bromide in Indoor Air Using Long-Path FTIR
 Spectroscopy. Measurement of Atmospheric
 Gases, Proc. SPIE 1433:270-274.

 Green, M., H.W.  Biermann, and J.N.  Seiber.
 1992.  Long-Path Fourier Transform Infrared
 Spectroscopy for Post-Fumigation Indoor Air
 Measurements. Analusis 20(8):455-460.

 Green, M., J.N. Seiber, and H.W. Biermann.
 1 993. In-Situ Measurements of Volatile Toxic
 Organics in Indoor  Air  Using Long-Path
 Fourier Transform  Infrared  Spectroscopy.
 Proc. SPIE 171 6:157-64.

 Griffiths,  P.R., and J.A. de Haseth. 1986.
Fourier  Transform  Infrared  Spectrometry,
John Wiley and Sons, New York.

Griffiths,  P.R.,  A.R.H.  Cole,  P.L.  Hanst,
W.J. Lafferty, R.J. Obremski, J.H. Shaw, and
R.N. Jones. 1 983. Specifications for Infrared
Reference Spectra of Molecules in the Vapor
Phase.  Appl. Spectrosc. 37(5):458-463.
 Griffiths, P.R., R.L. Richardson, D. Qin, and
 C.  Zhu.  1995.  Open-Path  Atmospheric
 Monitoring with  a  Low-Resolution  FT-IR
 Spectrometer.   Proceedings   of  Optical
 Sensing  for  Environmental  and  Process
 Monitoring (O.A. Simpson,  Ed.).  VIP-37,
 Air  &  Waste  Management Association,
 Pittsburgh, PA, pp. 274-284.

 Griggs,  M., and  C.B. Ludwig. 1978.  Legal
 Aspects   of  Remote  Sensing   and  Air
 Enforcement. J.  Air Pollut. Control Assoc.
 28(21:119-122.

 Grim, L.B., Th.C.  Gruber, Jr.,  and J. Ditillo.
 1996, Algorithm  development for passive
 FTIR. Proceedings of the 1995 AWMA/SPIE
 Specialty Con f in San Francisco. Air & Waste
 Management Association, Pittsburgh,  PA.

 Grosjean,  D., E.G. Tuazon, and  E.  Fujita.
 1990. Ambient  Formic Acid in Southern
 California   Air:   A  Comparison   of   Two
 Methods,   Fourier   Transform   Infrared
 Spectroscopy  and   Alkaline  Trap-Liquid
 Chromatography with UV Detection. Environ.
 Sci. Techno/. 24(1 ):144-146.

 Haaland, D.M. 1 990. Multivariate Calibration
 Methods  Applied  to  Quantitative  FT-IR
 Analyses.  Practical   Fourier   Transform
Infrared  Spectroscopy—Industrial   and
Laboratory Chemical Analysis, J.R.  Ferrar
 and K. Krishnan, Eds., Academic Press, San
 Diego, CA, pp. 396-468.

 Haaland, D.M.,  and R.G.  Easterling. 1980.
Improved  Sensitivity   of   Infrared
Spectroscopy by  the Application of  Least
Squares   Methods.   Appl.   Spectrosc.
34(5):539-548.

Haaland, D.M., and  R.G.  Easterling. 1982.
Application of New Least-Squares Methods
for  the  Quantitative Infrared Analysis of
Multicomponent Samples. Appl. Spectrosc.
36(6):665-673.
                                       12-7

-------
 TECH&imm
                                                                     TR-4423-99-03
 Haaland,   D.M.,  R.G.   Easterling,   and
 D.A.  Vopicka.  1985. Multivariate Least-
 Squares Methods Applied to the Quantitative
 Spectral   Analysis   of   Multicomponent
 Samples. Appl. Spectrosc. 39(1):73-84.

 Hall,  M.J., D.  Lucas,  and C.P.  Koshland.
 1991. Measuring Chlorinated Hydrocarbons
 in Combustion by Use of  Fourier Transform
 Infrared Spectroscopy. Environ. Sci. Techno!.
 25(2):260-267.

 Halliday,   D.   and  R.   Resnick.   1974.
 Fundamental  of Physics.  John Wiley  and
 Sons, New York.

 Hammaker, R.M., W.G. Fateley, C.T. Chaffin,
 T.L. Marshall, M.D. Tucker, V.D. Makepeace,
 and  J.M.   Poholarz.  1993. FT-IR  Remote
 Sensing of Industrial Atmosphere for Spatial
 Characterization. Appl. Spectrosc. 47:1471-
 1475.

 Hanst, P.L. 1 970. Infrared Spectroscopy and
 Infrared Lasers in Air Pollution Research and
 Monitoring. Appl. Spectrosc. 24(2): 1 61-1 74.

 Hanst,   P.L.   1971.   Mechanism   of
 Peroxyacetyl Nitrate Formation. J. Air Pollut.
 ControlAssoc. 21(5):269-271.

 Hanst, P.L. 1 971. Spectroscopic Methods for
 Air Pollution  Measurement.  Advances in
Environmental Science and Technology, Vol.
 2  (J.N.  Pitts, Jr., and  R.L. Metcalf, Eds.),
Wiley-lnterscience, New York, pp. 91-213.

 Hanst, P.L. 1976.  Optical  Measurement of
Atmospheric  Pollutants: Accomplishments
and  Problems.  Opt.   Quantum  Electron.
8:87-93.

Hanst, P.L. 1 978. Air Pollution Measurement
by Fourier  Transform  Spectroscopy. Appl.
Opt.  17(91:1360-1366.
 Hanst, P.L.  1978. Noxious Trace Gases in
 the  Air.  Part  I.  Photochemical  Smog.
 Chemistry 51(1):8-15.

 Hanst, P.L.  1978. Noxious Trace Gases in
 the  Air. Part II.  Halogenated  Pollutants.
 C/7em/sr/y51(2):6-12.

 Hanst,  P.L.  1979.  Pollution:  Trace  Gas
 Analysis.    Fourier  Transform  Infrared
 Spectroscopy.  Applications  to  Chemical
 Systems, Vol. 2 (J.R. Ferraro and L.J. Basile,
 Eds.),   Academic   Press,   New   York,
 pp. 79-110.

 Hanst, P.L.  1986. IR-Spectroscopy  of the
 Atmosphere.  Fresenius  Z.  Anal.  Chem.
 324(6):579-588(209-218).

 Hanst, P.L. 1989. Infrared Analysis of Engine
 Exhausts:  Methyl  Nitrite  Formation from
 Methanol Fuel. Spectroscopy 4(9):33-38.

 Hanst, P.L.  1991. Earliest  Spectroscopic
 Studies of  Polluted  Air:  A Tribute  to  Dr.
 Edgar R. Stephens. Proceedings of the 1991
 U.S. EPA/A&WMA International Symposium
 on the Measurement of Toxic and Related Air
 Pollutants,   Air   &  Waste  Management
 Association, Pittsburgh, PA, pp.  37-38.

 Hanst, P.L.  1991.  Analyzing Air for  PPB
 Concentrations   of   Trace  Gases,  Using
 Spectral  Subtraction. Proceedings of the
 1991  U.S.   EPA/A&WMA  International
 Symposium  on  the Measurement of Toxic
and  Related Air Pollutants, Air &  Waste
 Management Association, Pittsburgh, PA,
pp. 83-96.

 Hanst,  P.L.  1992.  Infrared  Spectra  for
 Quantitative  Analysis  of  Gases.  SP-81
 Optical Remote  Sensing.  Applications to
Environmental   and   Industrial   Safety
Problems,   Air  &  Waste   Management
Association, Pittsburgh, PA, pp. 490-510.
                                       12-8

-------
                                                                     TR-4423-99-03
 Hanst,  P.L.,  and  B.W.  Gay,  Jr.  1977.
 Photochemical   Reactions   Among
 Formaldehyde, Chlorine, and Nitrogen Dioxide
 in Air. Environ. Sci.  Techno/. 11(1 2): 1105-
• 1109.

 Hanst,  P.L.,  and  B.W.  Gay,  Jr.  1983.
 Atmospheric  Oxidation  of Hydrocarbons:
 Formation   of   Hydroperoxides   and
 Peroxyacids. Atmos. Environ. 17(11):2259-
 2265.

 Hanst,  P.L.,  and   J.A.   Morreal.   1968.
 Detection and Measurement of Air Pollutants
 by Absorptions of Infrared  Radiation. J. Air
 Pollut. Control Assoc. 18(111:754-759.

 Hanst, P.L., A.S.  Lefohn, and B.W. Gay, Jr.
 1 973. Detection of Atmospheric Pollutants at
 Parts-per-Billion   Levels   by   Infrared
 Spectroscopy. Appl.  Spectrosc. 27(3):188-
 198.

 Hanst,   P.L.,  L.L.   Spiller, D.M.  Watts,
 J.W. Spence,  and M.F. Miller. 1975. Infrared
 Measurement  of Fluorocarbons,   Carbon
 Tetrachloride, Carbonyl  Sulfide, and Other
 Atmospheric  Trace  Gases. J. Air Pollut.
 Control Assoc. 25(1 2): 1 220-1 226.

 Hanst,  P.L.,  J.A. Cooney, E. Hesstvedt,
 P.L.  Kelly,  J.J.  Kennedy, J.E. Lovelock,
 C.K.N.   Patel,  and   G.   Wang.  1976.
 Stratospheric  Chemistry and  Measurement
 Techniques. Opt.  Quantum  Electron. 8:187-
 191.

 Hanst, P.L.,  J.W. Spence, and M. Miller.
 1977. Atmospheric Chemistry of A/-Nitroso
 Dimethylamine.   Environ.   Sci.  Techno/.
 11(4):403-405.

 Hanst, P.L.,   N.W.  Wong,  and  J.  Bragin.
 1982. A Long-Path Infra-Red Study of Los
 Angeles  Smog.    Atmos.   Environ.
 16(5):969-981.
 Harris, D.B.,  and E. L.  Thompson.  1998.
 Evaluation  of  Ammonia  Emissions  from
 Swine  Operations   in   North  Carolina.
 EPA/600/A-98/142.  U.S.   Environmental
 Protection Agency.

 Hashmoney, R.A., Y. Mamane, Y. Benayahu,
 and A. Cohen. 1 996. Estimation of Emission
 Rates  from   Non-homogeneous  Fugitive
 Sources Using Open Path FTIR and Inversion
 Techniques.  Proceedings   of the   1995
 SPIE/AWMA  Specialty Conference  in  San
 Francisco.  Air  &  Waste   Management
 Association, Pittsburgh, PA.

 Haus, R., and K. Schaefer. 1992. Modeling
 of Smoke Stack Emissions for FTIR Remote
 Sensing. Proc. SPIE 1575:328-330.

 Haus, R., K. Schafer, D. Wehner, H. Bittner,
 and H. Mosebach. 1992. Remote Sensing of
 Air Pollution  by Mobile  Fourier-Transform
 Spectroscopy: Modeling and First Results of
 Measurements.  SP-81  Optical  Remote
 Sensing. Applications to Environmental and
 Industrial Safety  Problems,  Air  &  Waste
 Management  Association,  Pittsburgh,  PA,
 pp. 67-75.

 Haus,   R.,   K.  Schafer,   J.   Heland,
 H. Mosebach,  and T. Eisenmann. 1 993. FTIS
 in Environmental Research:  Mobile Remote
 Sensing  of  Air   Pollution.  Proc.   SPIE
 2089:318-319.

 Haus,  R.,    K.   Schafer,   W.  Bautzer,
 H. Mosebach, H. Bittner, and T. Eisenmann.
 1994.   Mobile  FTIS-Monitoring  of  Air
 Pollution. Appl. Opt.  33(24):5682-5689.

 Haus, R., K. Schaefer, J. Hughes, J. Heland,
 and W. Bautzer. 1 995. FTIR Remote Sensing
 of Smoke Stack and Test Flare Emissions.
Proc. SPIE 2506:45-54.

 Heland, J., K.  Schaefer and R. Haus.  1995.
FTIR Emission Spectroscopy and Modeling of
                                       12-9

-------
        '•=•••••?:= =
                                                                     TR-4423-99-03
 Radiative Transfer Layered Plume: Analysis
 of Aircraft Engine Exhausts. Proc. SPIE 236.

 Heland, J., and K. Schaefer. 1997. Analysis
 of Aircraft Exhausts with Fourier Transform
 Infrared Emission Spectroscopy. J. Eng. Appl.
 Sci. 36(21 ):4922-4931.

 Herget, W.F.  1979.  Air  Pollution: Ground-
 Based Sensing of Source  Emissions.  Fourier
 Transform   Infrared   Spectroscopy.
 Applications  to Chemical Systems,  Vol. 2
 (J.R. Ferraro and L.J. Basile, Eds.), Academic
 Press, New York, pp. 111-127.

 Herget,   W.F.   1981.   Measurement  of
 Gaseous Pollutants Using a Mobile  Fourier
 Transform Infrared (FTIR) System. Proc. SPIE
 289:449-456.

 Herget, W.F.  1 982. Analysis of Gaseous Air
 Pollutants Using a Mobile  FTIR System. Am.
 Lab. 14(121:72-78.

 Herget, W.F. 1982. Remote and Cross-Stack
 Measurement of Stack Gas Concentrations
 Using  a  Mobile  FT-IR System.  Appl.  Opt.
 21(41:635-641.

 Herget,  W.F.,  and  J.D.  Brasher.  1979.
 Remote Measurement of  Gaseous Pollutant
 Concentrations  Using  a Mobile   Fourier
 Transform Interferometer System. Appl. Opt.
 18(201:3404-3420.

 Herget, W.F.,  and  J.D.  Brasher.  1980.
 Remote  Fourier  Transform   Infrared  Air
 Pollution Studies.  Opt. Eng. 19(41:508-514.

 Herget, W.F.,  and W.D. Conner.  1977.
 Instrumental Sensing of Stationary Source
 Emissions.   Environ.   Sci.   Techno/.
 11(101:962-967.

Herget, W.F., and  S.P. Levine.  1986. Fourier
Transform Infrared (FTIR)  Spectroscopy for
 Monitoring Semiconductor   Process  Gas
Emissions. Appl. Ind. Hyg. 1(21:110-112.
 Herget, W.F., and  J.W. Muirhead. 1970.
 Infrared Spectral Absorption Coefficients for
 Water Vapor. J.  Opt. Soc. Am. 60(2): 180-
 183.

 Herget, W.F., J.A. Jahnke, D.E. Burch, and
 D.A.  Gryvnak.  1976.   Infrared  Gas-Filter
 Correlation   Instrument   for  In   Situ
 Measurement   of   Gaseous   Pollutant
 Concentrations.  Appl.   Opt.   15(5): 1222-
 1228.

 Herres,  W.,  and  J.  Gronholz.   1984.
 Understanding FT-IR Data Processing Part 1:
 Data Acquisition and Fourier Transformation.
 Computer Applications  in  the Laboratory
 4:216-220.

 Hilton, M., A.H.  Lettington, and I.M. Mills.
 1994.  Passive   Remote  Detection   of
 Atmospheric   Pollutants   Using    Fourier
 Transform  Infrared  Spectroscopy.  SP-89
 Optical   Sensing   for  Environmental
 Monitoring,  Air   &  Waste   Management
 Association, Pittsburgh, PA, pp. 134-144.

 Hirshfeld,  T.  1979. Quantitative FT-IR:  A
 Detailed  Look at the  Problems Involved.
 Fourier  Transform Infrared Spectroscopy.
Applications to Chemical Systems,  Vol.  2
 (J.R. Ferraro and L.J. Basile, Eds.), Academic
 Press, New York, pp. 193-242.

 Hodgeson,  J.A.,  W.A.  McClenny,  and
 P.L. Hanst. 1 973. Air Pollution Monitoring by
 Advanced   Spectroscopic   Techniques.
Science 182:248-258.

 Hommrich,   D.N.,   R.L.   Kump,   and
 R.H. Kagann. 1992. An Evaluation of Remote
Sensing  FTIR vs.  Point Source   Sample
Collection  and  Analysis.  SP-81   Optical
Remote  Sensing.   Applications   to
Environmental   and  Industrial   Safety
Problems,  Air  &  Waste   Management
Association, Pittsburgh, PA, pp. 207-211.
                                      12-10

-------
                                                                     TR-4423-99-03
 Hook, S.J., and  A.B.  Kahle. 1996.  Micro
 Fourier Transform Interferometer (muFTIR)—A
 New Field Spectrometer for Acquisition  of
 Infrared Data  of Natural Surfaces. Remote
 Sens. Environ. 56(31:172-181.

 Horlick, G.  1968.  Introduction to Fourier
 Transform Spectroscopy. Appl.  Spectrosc.
 22:617-626.

 Hudson,   J.L.,  J.  Arello,  M.J.  Thomas,
 H.E. Kimball, T.T. Holloway, B.J. Fairless,
 M.L. Spartz, M.R. Witkowski, T.L.  Marshall,
 C.P. Chaffin, R.M. Hammaker, W.G. Fateley,
 and  D.F. Gurka. 1991. Remote  Sensing  of
 Toxic Air Pollutants at a High Risk  Point
 Source Using Open-Path  Fourier Transform
 Infrared Spectroscopy: A Case Study.  Paper
 91-57.1,  Proceedings of the 84th Annual
 Meeting   and  Exhibition,  Air  & Waste
 Management Association, Pittsburgh, PA.

 Hudson,   J.L.,  M.J.  Thomas,  J. Arello,
 J.R.  Helvig, B.J.  Fairless, R.E. Carter, Jr.,
 D.D.  Lane,  and  G.A.  Marotz.  1992.  An
 Overview   Assessment  of    the
 Intercomparability   and  Performance   of
 Multiple Open-Path FTIR Systems as Applied
 to the Measurement of Air  Toxics.  SP-81
 Optical  Remote Sensing. Applications  to
 Environmental  and  Industrial   Safety
 Problems,  Air  &  Waste   Management
 Association, Pittsburgh, PA, pp. 112-122.

 Hudson,  J.L.,  M.J.  Thomas,  J. Arello,
 J.R.  Helvig, T.T.  Holloway,  B.J.  Fairless,
 R.E. Carter, Jr., D.D. Lane, and G.A. Marotz.
 1992.   The  Kansas  Open-Path   Fourier
 Transform   Infrared   Spectrometer
 Intercomparison Study: A Project Overview.
 Paper 92-73.08, Proceedings of the  85th
Annual Meeting and Exhibition, Air  & Waste
 Management Association, Pittsburgh, PA.

 Hull, D., R. Brewer, W. Dieringer, J. Lange,
 J. Pophal,  and R.  Spellicy.  1996.  Full
 Perimeter  FTIR Fence  Line Monitor.  Proc.
AWMA/SPIE Specialty  Conference in  San
 Francisco.   Air  &  Waste   Management
 Association, Pittsburgh, PA.

 Hunt, R.N. 1992. Continuous Monitoring of
 Atmospheric Pollutants by Remote Sensing
 FTIR.  SP-81   Optical  Remote  Sensing.
 Applications to Environmental and Industrial
 Safety Problems, Air & Waste Management
 Association, Pittsburgh, PA, pp. 446-455.

 Hunt, R.N. 1994.  Installation of Industrial
 Bistatic  Long-Path  FTIR  Systems.  Paper
 807-813,  SP-89   Optical   Sensing   for
 Environmental  Monitoring,  Air   &   Waste
 Management Association, Pittsburgh, PA.

 Hunt,  R.N.,  and  P.A.   Fuchs.   1995.
 Applications in Continuous  Monitoring  of
 Atmospheric Sensing. Proc. SPIE2365.

 Jenkins,  F.A.,  and  H.E.  White.   1950.
 Fundamentals of Optics.  McGraw-Hill, New
 York.

 Kagann,  R.H.,  and  O.A. Simpson.  1990.
 Open Path FTIR Measurements of Gaseous
 Emissions from  a  Chemical  Plant  Waste
 Water  Treatment  Basin.  Paper  90-86-7,
 Proceedings of the 83rd Annual Meeting and
 Exhibition,  Air  &  Waste   Management
 Association,  Pittsburgh, PA.

 Kagann, R.H., and O.A. Simpson. 1 992. FTIR
 Remote Sensor Data from the EPA Region VII
 FTIR Intercomparison Study.  SP-81  Optical
 Remote   Sensing.  Applications   to
 Environmental  and  Industrial   Safety
 Problems,  Air   &  Waste   Management
 Association,  Pittsburgh, PA, pp. 145-156.

 Kagann, R.H., R. DeSimone, O.A. Simpson,
 and  W.F.  Herget.  1990.  Remote  FTIR
 Measurement   of   Chemical  Emissions.
Proceedings of the 1990 U.S. EPA/A&WMA
International   Symposium   on    the
Measurement of  Toxic  and Related  Air
Pollutants,  Air  &  Waste   Management
Association, Pittsburgh, PA, pp. 244-247.
                                      12-11

-------
 TCt*U .TTs'ss Tsj. ~^=
 ICCH-L:.".:J.*:==
                                                                     TR-4423-99-03
 Kagann,   R.H.,    O.A.   Simpson,
 G.M. Russwurm, and W.A. McClenny. 1 991.
 New Developments in the FTIR  Remote
 Sensor. Paper 91-57.7, Proceedings of the
 84th Annual Meeting and Exhibition,  Air &
 Waste Management Association, Pittsburgh,
 PA.

 Kagann,  R.H.,  and O.A. Simpson.  1992.
 Comparison of  Open Path  FTIR Data with
 Point Samplers  in the  EPA Region VII FTIR
 Intercomparison Study.  Paper 92-73.09,
 Proceedings of the 85th Annual Meeting and
 Exhibition,  Air  &  Waste  Management
 Association, Pittsburgh, PA.

 Kagann,   R.H.,  J.G. Jolley,  D.S.  Shoop,
 M.R. Hankins,  and J.M. Jackson.  1994.
 Validation of  Open-Path  FTIR   Data   at
 Treatment, Storage, and Disposal Facilities.
 SP-89  Optical  Sensing for Environmental
 Monitoring,   Air  &  Waste  Management
 Association, Pittsburgh, PA, pp. 437-442.

 Kagann,  R.H.,  W.T. Walter, C.D. Wang,
 O.K. Kotter, and G.J. McManus. 1995. Field
 Test  on   Signal Processing Methods  for
 Passive FTIR Remote Sensing Measurements.
 Proc. SPIE 2366.

 Kagan,  R.H.  1997.  Instrumentation  for
 Optical Remote  Sensing. Procedings of the
 1997 Annual AWMA  Meeting in  Toronto.
 Air  & Waste  Management  Association,
 Pittsburgh, PA.

 Kert, J. 1992. Remote Sensing of On-Road
 Vehicle Emissions by an FTIR. SP-81. Optical
Remote    Sensing.   Applications    to
Environmental and Industrial Safety Problems,
Air  &  Waste  Management  Association,
Pittsburgh, PA, pp.  325-330.

Kirchgessner,   D.A.,  S.D.  Piccot,   and
A. Chadha.  1993. Estimation  of Methane
Emissions  from  a Surface Coal Mine Using
Open-Path FTIR  Spectroscopy and Modeling
Techniques. Chemosphere 26:23-44.
 Koehler,  F.W.,  and  G.W.  Small.  1998.
 Calibration  Transfer Results for Automated
 Detection  of  Acetone  and  SF6  by FTIR
 Remote Sensing Measurements. AIP Conf.
 Proc. 430:231-234.

 Kricks, R.J., T.R. Minnich, D.E. Pescatore,
 P.J.  Solinski,  D.  Mickunas,   L.   Kaelin,
 O.   Simpson,   M.J.    Czerniawski,  and
 T.H. Pritchett. 1991. Perimeter Monitoring at
 Lipari  Landfill  Using   Open-Path  FTIR
 Spectroscopy:  An  Overview  of  Lessons
 Learned. Paper 91-57.11, Proceedings of the
 84th Annual Meeting  and Exhibition, Air &
 Waste Management Association, Pittsburgh,
 PA.

 Kricks, R.J.,  R.L.  Scotto,  T.H. Pritchett,
 G.M. Russwurm, R.H.  Kagann, and D.B.
 Mickunas.  1992. Quality Assurance Issues
 Concerning the Operation of Open-Path FTIR
 Spectrometers.  SP-81   Optical  Remote
 Sensing. Applications  to Environmental and
 Industrial Safety  Problems,  Air &  Waste
 Management  Association, Pittsburgh,  PA,
 pp. 224-231.

 Kricks, R.J., S.H Perry, and D.E. Pescatore.
 1994.   Continued  Investigation   of   an
 Automated  I0  Generation  Technique  for
 Open-Path  FTIR Application. SP-89 Optical
 Sensing for Environmental Monitoring, Air &
 Waste Management Association, Pittsburgh,
 PA, pp. 47-55.

 Kricks, R.J., and R.L. Scotto,  1997. Update
 of QA/QC Issues for Field Data. Proceedings
 of the 1997 AWMA  Meeting  in  Toronto.
 Air   & Waste  Management  Association,
 Pittsburgh,  PA.

 Kroutil, R.T., J.T. Ditillo, and G.W.  Small.
 1990.  Signal  Processing  Techniques  for
 Remote Infrared Chemical Sensing. Comput.-
Enhanced Anal. Spectrosc., Vol. 2 (H.L.C.
 Meuzelaar,  Ed.), pp. 71-111.
                                      12-12

-------
                                                                     TR-4423-99-03
 Kump,  R.L., and  D.N.  Hommrich.  1992.
 Fenceline and Ambient Air Monitoring  with
 Open Path  FTIR  Optical Remote Sensing.
 SP-81 Optical Remote Sensing. Applications
 to  Environmental  and  Industrial  Safety
 Problems,   Air  &  Waste   Management
 Association, Pittsburgh, PA, pp. 269-272.

 Lacey,    R.F.   1989.   Elimination   of
 Interferences  in  Spectral   Data.  Appl.
 Spectrosc. 43(7): 11 35-11 39.

 LaCosse,   J.P.,   W.F.   Herget,   and
 R.L.  Spellicy.   1994.  Measurements of
 Ambienf  CS2 Concentrations  Using  FTIR
 Spectroscopy. SP-89 Optical  Sensing for
 Environmental Monitoring,  Air  &  Waste
 Management Association, Pittsburgh,  PA,
 pp. 466-491.

 Lamp,  T.,   M.   Radmacher,  K.  Weber,
 A. Gaertner, R. Nitz, and G. Broeker. 1997.
 Calibration   of   an  Open    Path  .FTIR
 Spectrometer  for  Methane  Ethylene  and
 Carbon Monoxide. Proceedings  of the 1997
AWMA  Meeting in Toronto. Air &  Waste
 Management Association, Pittsburgh, PA.

 Lane, D.D.,  R.E. Carter, Jr., G.A. Marotz,
 M.J. Thomas, and  J.L. Hudson. 1992.  The
 Design of and Results  from  a  Study to
 Intercompare Open-Path FTIRs and Canisters
under Controlled Field  Conditions.  Paper
92-73.06, Proceedings of the 85th Annual
Meeting  and Exhibition,  Air  &   Waste
Management Association, Pittsburgh, PA.

Lawson, D.R., P.J. Groblicki, D.H. Stedman,
G.A.  Bishop,  and  P.L.  Guenther.  1990.
Emissions from In-Use Motor Vehicles in Los
Angeles:  A Pilot  Study of Remote Sensing
and   the   Inspection  and   Maintenance
Program.  J.  Air  Waste  Manage.  Assoc.
40(81:1096-1105.

Lengyel, B.A. 1971. Lasers, 2nd Ed.,  Wiley-
Interscience, New York.
 Leo,  M.R.,  T.R.  Minnich,   R.L.  Scotto,
 R. Lute, D.B. Mickunas, and  T.H. Pritchett.
 1994. Estimation of  Hydrocarbon Emission
 Rates Using Open-Path FTIR Spectroscopy
 During a Controlled Evaporation Study. SP-89
 Optical   Sensing   for  Environmental
 Monitoring,   Air  &  Waste   Management
 Association, Pittsburgh, PA, pp. 639-658.

 Levine,    S.P.   1992.    Metrological
 Recommendations   for  Air-Monitoring
 Instruments - A Guidance Document. Am.
 Ind. Hyg. Assoc. J. 53(101:673-676.

 Levine, S.P., H-K. Xiao, T. Pritchett, and R.D.
 Turpin.  1990.  Fourier  Transform Infrared
 (FTIR) Spectroscopy for Monitoring Polar and
 Labile   Airborne   Gases   and   Vapors.
 Proceedings of the 1991 U.S. EPA/A&WMA
 International  Symposium   on   the
 Measurement  of Toxic and Related Air
 Pollutants,   Air  &   Waste   Management
 Association, Pittsburgh, PA, pp. 176-182.

 Levine, S., H. Xiao, W. Herget, R. Spear, and
 T. Pritchett. 1991. Remote Sensing (ROSE)
 FTIR.   Proceedings   of  the   1991  U.S.
 EPA/A&WMA  International Symposium  on
 the Measurement of  Toxic and Related Air
 Pollutants,   Air  &   Waste   Management
 Association, Pittsburgh, PA, pp. 707-711.

 Levine, S.P., and G.M.  Russwurm.  1994.
 Fourier Transform Infrared  Optical Remote
 Sensing for Monitoring Airborne Gas and
 Vapor Contaminants  in  the  Field.  Trends
Anal. Chem. 1 3(7):258-262.

 Liebhafsky, H.A.,  and H.G. Pfeiffer.  1953.
 Beer's Law in Analytical Chemistry. J. Chem.
Educ.  30(9):450-452.

 Long, G.L. and J.D. Winefordner. 1 983. Limit
of Detection: A Closer Look  at the IUPAC
 Definition. Anal. Chem. 55(7):71 2A-724A.

Lord, H.C. 1974. Absorption  Spectroscopy
Applied  to  Stationary Source  Emissions
                                      12-13

-------
                                                                     TR-4423-99-03
 Monitoring. Analytical Me thods Applied to A ir
 Pollution Measurements (R.K. Stevens  and
 W.F. Herget, Eds.), Ann Arbor Science, Ann
 Arbor, Ml, pp. 233-243.

 Lothian, G.F. 1963. Beer's Law  and Its Use
 in Analysis. Analyst 88:678.

 Low, M.J.D., and F.K. Clancy. 1967. Remote
 Sensing and Characterization of Stack Gases
 by  Infrared  Spectroscopy.  Environ.  Sci.
 Techno/. 1(1):73-74.

 Lubin, D.,  and S.A. Simpson.  1994. The
 Longwave  Emission  Signature  of Urban
 Pollution:  Radiometric  FTIR  Measurement.
 Geophys. Res. Lett. 21(1):37-40.

 Luft, K.F. 1 943. Uber Eine Neue Methode der
 Registrierenden Gas  Analyze  Mit  Hiife  der
 Absorption Ultraroter Stahlen Ohne Spektrale
 Zerlegung. J. Tech. Phys. 24:97.

 Lupo, M.J., C.E. Magnuson, R.L. Shriver, F.K.
 Hamilton,  R.L.  Scotto, T.R. Minnich,  D.E.
 Pescatore, R.H. Kagann, O.A. Simpson, S.E.
 McLaren,  and  D.H.  Stedman.  1991.  Air
 Pathway Monitoring for a Land  Ban No-
 Migration    Petition.   Paper   91-57.4,
 Proceedings of the 84th Annual Meeting and
 Exhibition,   Air  &  Waste   Management
 Association, Pittsburgh, PA.

 Lupo,  M.J.,  C.E.   Magnuson,  Z. Wang,
 G.B. Evans,  Jr.,  R.L. Shiver, T.R.  Minnich,
 R.L. Scotto, D.E. Pescatore, and R.J. Kricks.
 1991. The  Suitability  of Open  Path  Air
 Monitoring  for RCRA  Application:  A  Case
 Study. Part I.  Project Overview. Proceedings
of the 1991 U.S. EPA/A&WMA International
Symposium on  the Measurement  of Toxic
and Related  Air Pollutants,  Air  & Waste
Management  Association, Pittsburgh,  PA,
pp. 685-690.

Lute, R., R.J. Kricks, D.E. Pescatore, and P.J.
Solinski. 1992. Effects of Atmospheric Water
Vapor  on  Open-Path  FTIR  Spectroscopy
 Data.   SP-81  Optical  Remote  Sensing.
 Applications to Environmental and Industrial
 Safety Problems, Air & Waste Management
 Association, Pittsburgh, PA, pp. 425-433.

 Mellqvist, J., D.W.  Arlander, B. Galle,  and
 B.   Berqvist.   1995.  Measurements   of
 Industrial Fugitive  Emissions by  the FTIR
 Tracer Method. IVL Report B 1 214, Swedish
 Environmental Research Institute, Goteborg,
 Sweden.

 Marotz, G.A., R.E. Carter, Jr., and D.D. Lane.
 1992. Sampling Design and Protocols Used
 in a Recent Study  of Long-Path  FTIR  and
 Canister Compound Detection and Estimation
 Under Controlled Field  Conditions.  SP-81
 Optical Remote Sensing. Applications  to
 Environmental  and  Industrial   Safety
 Problems,  Air  &   Waste   Management
 Association, Pittsburgh, PA, pp.  123-132.

 Marshall,  A.G.  and F.R. Verdun.  1990.
 Fourier  Transforms  in  NMR,  Optical, and
 Mass Spectrometry, Elsevier, Amsterdam.

 Marshall, T.L., C.T. Chaffin, R.M. Hammaker,
 and W.G. Fateley. 1994. An Introduction to
 Open-Path  FT-IR Atmospheric  Monitoring.
 Environ. Sci. Techno/. 28:224A-232A.

 Marshall,    T.L.,   C.T.    Chaffin,
 V.D.   Makepeace,   R.M.   Hoffman,
 R.M. Hammaker, W.G. Fateley, P. Saarinen,
 and J. Kauppinen. 1994. Investigation of the
 Effects of Resolution on the Performance of
 Classical  Least-Squares   (CLS)  Spectral
 Interpretation  Programs  When  Applied  to
 Volatile Organic  Compounds   (VOCs)   of
 Interest in Remote  Sensing Using Open-Air
 Long-Path Fourier Transform Infrared  (FT-IR)
 Spectrometry. J. Mo/. Struct. 324:19-28.

 Martino,  P.A.  1992.  A  Comparison   of
 Remote  Sensing   and   Point   Sampling
 Measurements  of  Aromatic  Hydrocarbon
 Emissions. SP-81 Optical Remote Sensing.
Applications to Environmental and Industrial
                                      12-14

-------

                                                                    TR-4423-99-03
 Safety Problems, Air & Waste Management
 Association, Pittsburgh, PA, pp. 212-223.

 McCauley,   C.L,   M.J.   Czerniawski,
 R.H.  Kagann,  and  O.A.  Simpson.  1992.
 Development of a  User-Friendly Intelligent
 Spectroscopic  Computer  System:  A  Case
 Study.  SP-81  Optical  Remote  Sensing.
 Applications to Environmental and Industrial
 Safety Problems, Air & Waste Management
 Association, Pittsburgh, PA, pp. 456-463.

 McClenny,   W.A.,   and  G.M.  Russwurm.
 1978. Laser-Based, Long Path Monitoring of
 Ambient Gases—Analysis of Two Systems.
 Atmos.  Environ. 12:1443-1453.

 McClenny,   W.A.,   W.F.   Herget,   and
 R.K. Stevens. 1974. Comparative Review of
 Open-Path  Spectroscopic  Absorption
 Methods  for   Ambient   Air  Pollutants.
 Analytical Methods Applied to Air Pollution
 Measurements   (R.K.   Stevens   and
 W.F. Herget, Eds.), Ann Arbor Science, Ann
 Arbor, Ml, pp. 107-120.

 McClenny,  W.A.,  Baumgardner, R.E.,  Jr.,
 Baity,  F.W., Jr.,  and  R.A. Gray.  1974.
 Methodology for Comparison of Open-Path
 Monitors with Point  Monitors. J. Air Pollut.
 ControlAssoc.  24(11):1044-1046.

 McClenny,  W.A.,   G.M.   Russwurm,
 M.W.  Holdren,  A.J.  Pollack,  J.D.   Pleil,
 J.L.   Varns,  J.D.  Mulik,  K.D.   Oliver,
 R.E.  Berkley, D.D. Williams, K.J. Krost, and
 W.T.  McLeod.  1991. Superfund Innovative
 Technology Evaluation.  The Delaware  SITE
 Study,  1989.   EPA  600/3-91/071.  U.S.
 Environmental Protection Agency, Research
Triangle Park, NC.

 McLaren, S.E., J.W. Hannigan, D.L.  New,
and   D.H.   Stedman.   1991.   Direct
Measurement of Freeway Emissions  Using
Long-Path Spectroscopy. Proceedings of the
 1991   U.S.  EPA/A&WMA   International
Symposium on the Measurement of Toxic
and Related  Air Pollutants,  Air  & Waste
Management  Association,  Pittsburgh,  PA,
pp. 741-746.

McLaren,   S.E.,   D.H.    Stedman,
P.O. Greenlaw,  R.J. Bath, and R.D. Spear.
1992. Comparison of an Open Path UV and
FTIR Spectrophotometer.  Paper  92-73.10,
Proceedings of the 85th Annual Meeting and
Exhibition,  Air   &   Waste  Management
Association, Pittsburgh, PA.

McNesby,  K.L,  C.F. Chabalowski,   and
R.A. Fifer. 1991.  Approximation  to  True
Peak  Absorbance  from  Observed   Peak
Absorbance for Gas Phase Fourier Transform
Spectroscopy. Appl. Opt. 30(4):378-379.

Michelson, A.A. 1881. The Relative Motion
of the Earth and the Luminiferous Ether. Am.
J. Sci. 22(31:120-129.

Michelson,   A.A.   1882.    Interference
Phenomena in a New Form of Refractometer.
Phil. Mag. 13:236-242.

Michelson,   A.A.   1891.   Visibility  of
Interference-Fringes  in the  Focus  of a
Telescope. Phil.  Mag.  31:256-259.

Minnich,   T.R.,   R.L.   Scotto,   and
T.H.  Pritchett.   1990.  Remote  Optical
Sensing of VOCs: Application to Superfund
Activities. Proceedings of the  1990  U.S.
EPA/A&WMA  International Symposium on
the  Measurement of Toxic and Related Air
Pollutants,  Air   &   Waste   Management
Association, Pittsburgh, PA, pp. 928-936.

Minnich,  T.R.,  M.R.  Leo,   R.L.  Scotto,
M.J.  Czerniawski,   R.H.   Kagann,   and
O.A. Simpson. 1991. A Software Package
for Assessing  Downwind Air Quality Impact
in Real  Time Based   on  Open-Path   FTIR
Measurement  Data. Proceedings of the 1991
U.S. EPA/A&WMA International Symposium
on the Measurement of Toxic and Related Air
                                     12-15

-------
                                                                      TR-4423-99-03
 Pollutants,  Air  &   Waste   Management
 Association, Pittsburgh, PA, pp. 704-706.

 Minnich,  T.R.,   R.L.  Scotto,  M.R.  Leo,
 B.C. Sanders,  S.H. Perry, and T.H. Pritchett.
 1992. A Practical Methodology Using Open-
 Path FTIR Spectroscopy to Generate Gaseous
 Fugitive-Source Emission Factors at Industrial
 Facilities. SP-81  Optical Remote Sensing.
 Applications to Environmental and Industrial
 Safety Problems, Air & Waste  Management
 Association, Pittsburgh, PA, pp. 513-518.

 Minnich, T.R., R.L. Scotto, R.H. Kagann, and
 O.A.   Simpson.   1990.  Optical  Remote
 Sensors  Ready to Tackle Superfund, RCRA
 Emissions Monitoring  Tasks. Hazmat World
 3:42.

 Minnich, T.R., and R.L. Scotto. 1994. The
 Role of Open-Path FTIR Spectroscopy in the
 Development  of  a Successful Accidental
 Release  Detection Program. SP-89  Optical
 Sensing for  Environmental Monitoring, Air &
 Waste Management Association, Pittsburgh,
 PA, pp. 340-348.

 Mosebach,  H., T.  Eisenmann, Y. Schulz-
 Spahr, I.  Neureither,  H. Bittner,  H.  Rippel,
 K. Schafer,  D. Wehner, and R.  Haus. 1992.
 Remote Sensing of Smoke Stack Emissions
 Using  a  Mobile  Environmental Laboratory.
 Proc.  SPIE 171 7:149-1 58.

 Muthusubramanian, P., and S.P.M. Deborrah.
 1989. Design  Considerations  of  a 32  m
 Optical Path Multiple Reflection Cell for Air
 Pollution Studies.  Indian J. Environ. Prot.
 9(2):144-146.

 Nelson, C.M.,  K.S. Knight, A.S. Bonanno,
 M.A. Serio,  P.R. Solomon, and  M.A.  Halter.
 1994. On-Line FT-IR Analysis of Fossil Fuel-
 Fired Power Plants. SP-89  Optical Sensing
.for Environmental Monitoring, Air  & Waste
 Management Association,  Pittsburgh, PA,
 pp. 826-840.
 Nicolet Analytical Instruments. 1986. FT-IR
 Theory.  Nicolet   Analytical   Instruments,
 Madison, Wl.

 Niki, H., and P.O. Maker. 1990. Atmospheric
 Reactions  Involving  Hydrocarbons:  Long
 Path-FTIR  Studies. Adv.  Photochem.  15:
 69-137.

 Nyden, M.R.,  W.  Grosshandler, D. Lowe,
 R. Harris, and E. Braun. 1 994. Application of
 FTIR   Remote  Sensing  Spectroscopy  in
 Environmental Impact  Assessments of Oil
 Fires.   SP-89   Optical  Sensing   for
 Environmental  Monitoring, Air  &  Waste
 Management Association,  Pittsburgh,  PA,
 pp. 767-779.

 Oppenheimer,  C.,  P.  Francis,  M.  Burton,
 A.J.H. Maciejewski, and L. Boardman. 1 998.
 Remote Measurement of Volcanic Gases by
 Fourier Transform   Infrared Spectroscopy.
 Appl.  Phys. B. 76(4):505-51 5.

 Paine, R.J., J.O. Zwicker,  and  H. Feldman.
 1997.  Project  OPTEX: Field  Study at a
 Petrochemical  Facility to  Assess  Optical
 Remote Sensing and  Dispersion Modeling
 Techniques.  Proceedings  of  the  AWMA
 Annual Meeting in Toronto, Air &  Waste
 Management Association, Pittsburgh, PA.

 Park,  D.Y., S.  Levine, and  G.M. Russwurm
 1998. Fourier Transform  Infrared  Optical
 Remote   Sensing.   Encyclopedia   of
 Environmental Analysis and  Remediation.
 Vol.7, Chapter 7, pp. 4137-4149.

 Pawloski, J.N.  and  D.G. Iverson. 1998. The
 Use of Optical Remote Sensing to Monitor
 Facility Releases.   Hydrocarbon Process.
 77(9)  September.

 Penner, S.S. 1959. Quantitative Molecular
Spectroscopy and Gas Emissivities. Addison-
Wesley, Reading, MA.
                                      12-16

-------
                                                                     TR-4423-99-03
 Perry,  S.H., P.L. McKane,  D.E. Pescatore,
 A.E.   Dubois,  and   R.J.  Kricks.  1996.
 Maximizing the Use of Open-Path FTIR for 24
 Hour Monitoring Around the Process Area of
 an  Industrial Chemical Facility. Proceedings
 of the A WMA Specialty Conference in San
 Francisco,   Air   &   Waste  Management
 Association, Pittsburgh, PA.

 Pescatore, D.E., S.H. Perry, R.J. Kricks, and
 G.R. Pollack. 1994. Use of  Open-Path FTIR
 Spectroscopy   to  Investigate  Unknown
 Airborne  Contaminants   at   a   Sludge
 Dewatering Facility. SP-89  Optical Sensing
 for Environmental Monitoring, Air  & Waste
 Management  Association,  Pittsburgh,  PA,
 pp. 93-104.

 Pfeiffer, H.G., and H.A. Liebhafsky. 1951.
 The Origins of Beer's  Law.  J. Chem. Educ.
 28:123-125.

 Phillips, B.  and Roy Brandon.  1993. FTIR
 Remote Sensing  Data Reduction Technique
 for Elevated Temperature Gas Emissions from
 an  Aluminum  Smelting Plant.  Proceedings
 AWMA Annual  Meeting.  Air &  Waste
 Management Association, Pittsburgh, PA.

 Phillips, B.,  D.  Brown,  G.M. Russwurm,
 J. Childers,  E. Thompson, and  L.T. Lay,
 1996.   Innovative Open-Path  FTIR  Data
 Reduction Algorithm.  Proceedings of  the
AWMA  Specialty   Conference   in  San
 Francisco,  Air   &   Waste  Management
 Association, Pittsburgh, PA.

 Phillips, B.,  R.  Moyers, and  L.T. Lay. 1995.
 Improved FTIR Open-Path Remote Sensing
 Data Reduction. Proceedings of the AWMA
 Specialty Conference in San Francisco, Air &
Waste  Management Association, Pittsburgh,
PA.

Piccot, S.D., A. Chadha, D.A. Kirchgessner,
R.   Kagann,   M.J..   Czerniawski,   and
T. Minnich.  1991. Measurement of Methane
 Emissions in the Plume of a Large Surface
 Coal  Mine   Using   Open-Path   FTIR
 Spectroscopy. Paper 91-57.6, Proceedings
 of the 84th Annual Meeting and Exhibition,
 Air  &  Waste  Management   Association,
 Pittsburgh, PA.

 Piccot,  S.D.,   D.A.  Kirchgessner,
 S.S. Masemore, W.F. Herget, and E. Ringler.
 1994. Validation of a Method for Estimating
 Pollution  Emission Rates Using Open-Path
 FTIR  Spectroscopy   and  Modeling
 Techniques.  SP-89  Optical  Sensing  for
 Environmental  Monitoring,  Air &  Waste
 Management Association, Pittsburgh, PA,
 pp. 322-339.

 Piccot, S.D., S.S.  Masemore, E.S. Ringler,
 S.  Srinivasan,   D.A.  Kirchgessner,  and
 W.F.  Herget. 1994. Validation of a Method
 for Estimating Pollution Emission Rates from
 Area   Sources    Using   Open-Path  FTIR
 Spectroscopy   and  Dispersion   Modeling
 Techniques. J.  Air Waste Manage. Assoc.
 44:271-279.

 Pitts,  J.N.,  B.J.  Finlayson-Pitts,   and
 A.M. Winer. 1977. Optical Systems Unravel
 Smog  Chemistry.  Environ.  Sci.  Techno/.
 11(61:568-573.

 Plummer,  G.M.  1994. Current  and  Future
 Spectroscopic   Emission  Test  Method
 Development Topics.  SP-89 Optical Sensing
 for Environmental Monitoring,  Air  & Waste
 Management Association, Pittsburgh, PA,
 pp. 492-502.

 Polak,  M.L., J.L.  Hall,  G.J.  Sherer,  and
 K.C. Herr. 1 996. Proceedings of the AWMA
 Specialty Conference  in San Francisco, Air &
 Waste Management Association, Pittsburgh,
 PA.

 Powell,   J.R.,   F.M.   Wasacz,   and
 R.J. Jakobsen.  1986. An Algorithm for the
 Reproducible Spectral Substraction  of Water
from the  FT-IR Spectra of Proteins in Dilute
                                      12-17

-------
                                                                    TR-4423-99-03
 Solutions and Adsorbed Monolayers. Appl.
 Spectrosc. 40(3):339-344.

 Prengle,  H.W., Jr., C.A. Morgan, C-S. Fang,
 L-K. Huang,  P.  Campani,  and W.W.  Wu.
 1973.   Infrared   Remote   Sensing   and
 Determination of Pollutants in Gas Plumes.
 Environ.  Sci.  Techno/. 7(5):417-423.

 Pruitt, G.R.  1992.  Status  of Cryogenic
 Coolers  for Remote  Sensing Applications.
 SP-81 Optical Remote Sensing. Applications
 to  Environmental and Industrial  Safety
 Problems,  Air  &  Waste   Management
 Association, Pittsburgh, PA, pp. 313-324.

 Reid,    S.A.,   J.I.L.   Hughes,   P.T.
 Roberts,  I. Archibald, and K. Gregory. 1994.
 Development of  Open Path  Systems for
 Emission Rate Measurements. SP-89 Optical
 Sensing for Environmental Monitoring, Air &
 Waste Management Association, Pittsburgh,
 PA, pp. 305-321.

 Rennilson,   J.J.   1980.   Retroreflection
 Measurements:  A  Review.  Appl.   Opt.
 19:1234.

 Ropertz,  A., T. Lamp, K. Weber, M. Douard,
 A. Gaertner, and R. Nitz. 1 997. Calibration of
 High   Resolution   OP-FTIR  Spectrometer
 Combined  with   an  Adjustable   100  m
 Multipass Cell. Proc SPIE 3107:1 37-147.

 Rossi,  B.  1957.  Optics.  Addison-Wesley
 Publishing Company, Inc., Reading, MA.

 Russwurm, G.M. 1992. Quality Assurance,
 Water Vapor,  and the Analysis of FTIR Data.
 Paper  92-73.03, Proceedings of  the 85th
Annual Meeting and Exhibition, Air & Waste
 Management Association, Pittsburgh, PA.

 Russwurm, G.M. 1992. Quality Assurance
and  the  Effects  of  Spectral  Shifts  and
 Interfering Species in FTIR  Analysis. SP-81
 Optical Remote  Sensing.  Applications  to
Environmental   and   Industrial  Safety
 Problems,  Air  &   Waste   Management
 Association, Pittsburgh, PA, pp. 105-111.

 Russwurm, G.M. 1994. A Technique for the
 Quantitative Comparison  of FT-IR  Spectra.
 SP-89 Optical Sensing for Environmental
 Monitoring,  Air  &  Waste   Management
 Association, Pittsburgh, PA, pp. 15-25.

 Russwurm, G.M. 1997. Long-Path Open-Path
 Fourier  Transform  Infrared Monitoring of
 Atmospheric Gases.  Compendium Method
 TO-16. Proc. SP/£3107.

 Russwurm,  G.M.  1997.  Methodological
 Questions   in   the   Application   of
 Spectroscopic Techniques. Proc. SPIE 3107.

 Russwurm  G.M.  1997. Quality Assurance/
 Quality  Control Issues in  Open-Path  FTIR
 Monitoring. Proc. SP/£3107.

 Russwurm, G.M., and J.W. Childers. 1994.
 An Overview  of the  Fourier  Transform
 Infrared Remote Sensing Technique. Environ.
 Lab. June/July:! 5-1 8.

 Russwurm, G.M., and J.W. Childers. 1995
 FT-IR  Spectrometers Perform Environmental
 Monitoring. Laser Focus World, April.

 Russwurm, G.M. and J.W. Childers. 1996.
 Compendium  Method TO-16:  Long  Path
 Open Path Fourier Transform Infrared Method
 for Monitoring  Ambient Air. Proceedings of
 the 89th Annual Meeting & Exhibition of the
Air& Waste Management Association. Paper
 96-MP5.04, Air &  Waste  Management
 Association, Pittsburgh, PA.

 Russwurm, G.M., and J.W. Childers. 1997.
 Quality Assurance/Quality Control Issues in
 Open-Path  FT-IR  Monitoring.  Proc.   SPIE
 3107.

 Russwurm,  G.M.,  and  W.A.  McClenny.
 1990.  A Comparison of  FTIR  Open  Path
Ambient Data with Method TO-14  Canister
                                      12-18

-------

                                                                    TR-4423-99-03
 Data.  Proceedings  of  the  1990  U.S.
 EPA/A&WMA  International Symposium on
 the Measurement of Toxic and Related Air
 Pollutants,   Air   &  Waste   Management
 Association, Pittsburgh, PA, pp. 248-253.

 Russwurm,  G.M., and  B.  Phillips.  1998.
 Uncertainties in FTIR Data Due to a Nonlinear
 Response. Presented at the  SPIE Specialty
 Conference on Environmental  Monitoring and
 Remediation Techniques, Boston, November.

 Russwurm,   G.M.,    R.H.   Kagann,
 O.A.   Simpson,   W.A.  McClenny,   and
 W.F.   Herget.   1991.   Long-Path   FTIR
 Measurements   of   Volatile   Organic
 Compounds  in an  Industrial  Setting. J.  Air
 Waste Manage. Assoc.  41 (8): 1062-1066.

 Russwurm,   G.M.,    R.H.   Kagann,
 O.A. Simpson, and W.A. McClenny.  1991.
 Use of a Fourier Transform Spectrometer as
 a Remote Sensor at Superfund  Sites. Proc.
 SPIE 1433:302-314.

 Sandridge, R.L. 1994. Calibration of Bistatic
 Long-Path FTIR Monitoring Systems.  SP-89
 Optical   Sensing   for  Environmental
 Monitoring,  Air   &  Waste   Management
 Association, Pittsburgh, PA, pp. 255-262.

 Schafer,  K., R.  Haus,  D.  Wehner,  and
 H. Mosebach. 1992.  Modelling of Radiative
 Transfer for  FTIR Remote Sensing  of Smoke
 Stack  Emissions.  Paper  IU-9B.05, Waste
 Management, Proceedings of the 9th World
 Clean  Air  Congress,  Air  &  Waste
 Management Association, Pittsburgh, PA.

 Schafer,   K.,   R.   Haus,   D.  Wehner,
 H. Mosebach, H. Bittner, and  N. Neureither.
 1 992.   Fourier-Transform-lnfrared-
 Spectroscopy of Air  Pollution by a Mobile
 System. Proceedings of the International
Symposium on Environmental  Contamination
in Central and Eastern Europe, pp. 843-845.
 Schafer, K., R. Haus, and J. Heland. 1994.
 Measurements of Gaseous Compounds from
 Emissions Sources and in  Ambient Air by
 Fourier  Transform Infrared Spectroscopy:
 Method and Results of FTIS-MAPS. SP-89
 Optical   Sensing   for  Environmental
 Monitoring,  Air   &  Waste   Management
 Association, Pittsburgh, PA, pp. 455-465.

 Schafer K., J. Heland, R. Haus, C. Werner,
 and F. Koepp. 1995. Contribution of Remote
 Sensing for  Diagnostics of Aircraft Engine
 Combustion. Proc. SPIE 2506:55-61.

 Schafer,  K.,  R.Haus,  J.  Heland,   and
 C. Werner. 1995. Determination of Emission
 Sources by Remote'Sensing. Proceedings of
 the 1995 3rd International Conference on Air
 Pollution,  Vol. 2, Comp. Mechanics Inc.,
 Billerica, MA.

 Schafer, K.,  R. Haus, and S. Heinz.  1996.
 Determination  of  Emission   Rates  from
 Diffuse Sources by Ambient FTIR-absorption-
 spectroscopy  and   Adapted   Dispersion
 Modeling.   Proceedings  of   the  AWMA
 Specialty Conferences in San Francisco, Air
 &   Waste   Management   Association,
 Pittsburgh, PA.

 Schlosser, R.L., and R.W.  Brandon.  1992.
 FTIR  Remote  Sensor  Measurements  of
 Hydrogen  Fluoride   and   Other  Fugitive
 Emissions  at Aluminum Smelting  Facilities.
 SP-81 Optical Remote Sensing. Applications
 to  Environmental  and  Industrial  Safety
 Problems,.  Air   &  Waste   Management
 Association,  Pittsburgh, PA, pp. 286-295.

 Scotto,  R.L., T.R. Minnich, and M.R. Leo.
 1991.   A  Method  for  Estimating  VOC
 Emission Rates  from Area Sources Using
 Remote  Optical Sensing. Proceedings of the
 1991   U.S.   EPA/A&WMA   International
Symposium on the Measurement of Toxic
and Related Air  Pollutants, Air  & Waste
 Management  Association,  Pittsburgh, PA,
pp. 698-703.
                                      12-19

-------
 jECHm^mm
                                                                     TR-4423-99-03
 Scotto, R.L., B.C. Sanders, R.H. Kagann, and
 O.A.   Simpson.   1992.  Validation  of  a
 Gaussian Plume  Dispersion Model Based on
 Data   from  the  Kansas Open-Path  FTIR
 Intercomparison  Study.  Paper 92-73.04,
 Proceedings of the 85th Annual Meeting and
 Exhibition,   Air   &  Waste   Management
 Association, Pittsburgh, PA.

 Shepherd,  0.,  A.G. Hurd,  R.B. Wattson,
 H.J.P.  Smith, and  G.A. Vanasse.  1981.
 Spectral Measurements  of Stack Effluents
 Using  a Double  Beam  Interferometer  with
 Background   Suppression.   Appl.   Opt.
 20(22):3972-3980.

 Simonds-Malachowski,   M.,   S.P.  Levine,
 G. Herrin, R.C. Spear, M. Yost, and Z. Yi.
 1994.  Workplace  and  Environmental  Air
 Contaminant  Concentrations  Measured by
 Open   Path  Fourier  Transform   Infrared
 Spectroscopy: A Statistical Process Control
 Technique to Detect Changes from Normal
 Operating Conditions. J. Air Waste Manage.
 Assoc. 44:673-682.

 Simpson, O.A.,  and R.H. Kagann.  1990.
 Measurements of Emissions at a Chemical
 Waste Water Treatment Site with an Open
 Path    Remote   Fourier    Transform
 Interferometer. Proceedings of the 1990 U.S.
 EPA/A&WMA International Symposium on
 the Measurement of Toxic and Related Air
 Pollutants,  Air  &  Waste   Management
 Association, Pittsburgh, PA, pp. 937-940.

 Small,  G.W.,  R.T.  Kroutil, J.T. Ditillo,  and
W.R.    Loerop.   1988.  Detection   of
Atmospheric Pollutants by Direct Analysis of
 Passive   Fourier   Transform  Infrared
 Interferograms./lna/. Chem. 60(3):264-269.

Small,   G.W.,  A.C.  Harms,   R.T.  Kroutil,
J.T. Ditillo,  and W.R. Loerop. 1990. Design
of Optimized Finite Impulse Response Digital
Filters for Use with Passive Fourier Transform
Infrared  Interferograms.  Anal.   Chem.
62(17):1768-1777.
 Small,   G.W.,   T.F.   Kaltenbach,   and
 R.T. Kroutil. 1991. Rapid Signal Processing
 Techniques  for Fourier Transform  Infrared
 Remote  Sensing.  Trends  Anal.   Chem.
 10(51:149-155.

 Solomon,   P.R.,   P.W.   Morrison,  Jr.,
 M.A. Serio,  R.M Carangelo, J.R.  Markham,
 S.C. Bates, and J.E. Cosgrove. 1 992. Fourier
 Transform Infrared Spectroscopy for Open-
 Path  Monitoring.  SP 81  Optical  Remote
 Sensing. Applications to Environmental and
 Industrial Safety  Problems,  Air  &  Waste
 Management Association, Pittsburgh,  PA,
 pp. 479-489.

 Spartz, M.L., M.R. Witkowski, J.H. Fateley,
 J.M. Jarvis, J.S. White, J.V. Paukstelis, R.M.
 Hammaker,  W.G. Fateley, R.E. Carter,  M.
 Thomas,   D.D.   Lane,   G.A.   Marotz,
 B.J. Fairless, T. Holloway, J.L. Hudson, and
 D.F. Gurka.  1989. Evaluation  of a Mobile
 FT-IR System for Rapid VOC Determination.
 Part  1:   Preliminary   Qualitative   and
 Quantitative   Calibration   Results.  Am.
 Environ. Lab.  1(2): 15-30.

 Spartz, M.L., M.R. Witkowski, J.H. Fateley,
 R.M. Hammaker, W.G. Fateley, R.E. Carter,
 M.   Thomas,  D.D.   Lane,   B.J.  Fairless,
 T. Holloway, J.L. Hudson, J. Arello,  and
 D.F. Gurka. 1990. Comparison of Long Path
 FT-IR Data to Whole Air Canister Data from
 a   Controlled   Upwind   Point   Source.
 Proceedings   of  the  1990  EPA/A&WMA
 International    Symposium   on   the
 Measurement  of  Toxic and Related Air
 Pollutants,   Air   &   Waste   Management
 Association, Pittsburgh, PA, pp. 685-692.

 Spear, R.D., G.M.  Rojek, P.O. Greenlaw, and
 R.J.  Bath.   1991. A Case  Study  Using
 Remote   Sensing  in  the  Region  II  Site
 Assessment  FASP Program. Proceedings of
 the 1991 U.S.  EPA/A&WMA International
 Symposium on  the Measurement of Toxic
and  Related Air  Pollutants,  Air  & Waste
                                      12-20

-------
                                                                     TR-4423-99-03
 Management  Association,
 pp. 729-740.
Pittsburgh,  PA,
 Spear, R.D.,  P. Boone, P.D. Greenlaw, and
 R.J. Bath. 1991. Standard Field Operating
 Guidelines for the  Determination  of  Air
 Toxics  in the  Region  II Site  Assessment
 FASP Program. Proceedings of the 1991 U.S.
 EPA/A&WMA International Symposium on
 the Measurement of Toxic  and Related Air
 Pollutants,  Air  &  Waste  Management
 Association, Pittsburgh, PA, pp. 712-718.

 Spear, R.D., P. Greenlaw, and R. Bath. 1 991.
 System  Selection  and  Design  for the
 Determination of Air Toxics in the Region II
 Site Assessment FASP Program. Proceedings
 of the  1991 U.S. EPA/A&WMA  International
 Symposium on the Measurement of Toxic
 and Related  Air  Pollutants, Air &  Waste
 Management  Association,  Pittsburgh, PA,
 pp. 719-728.

 Spellicy,  R.L.,  W.L.  Crow, J.A.  Draves,
 W.F.  Buchholtz,  and  W.F. Herget.  1991.
 Spectroscopic Remote Sensing: Addressing
 Requirements   of   the   Clean  Air Act.
 Spectroscopy 6(9):24-34.

 Spellicy,  R.L.,  J.A. Draves,  W.L.  Crow,
 W.F. Herget,  and  W.F. Buchholtz. 1992. A
 Demonstration of Optical Remote Sensing in
 a Petrochemical Environment. SP-81 Optical
 Remote  Sensing.   Applications  to
 Environmental   and   Industrial  Safety
 Problems,  Air  &  Waste  Management
 Association, Pittsburgh, PA, pp. 273-285.

 Spellicy,   R.L.,  J.A.  Draves,   W.  Crow,
 W.F. Herget,  and  W. Buchholtz.  1992.  A
 Demonstration of  Open-Path Fenceline and
 Extractive Workplace Monitoring with FTIR
 and UV-DOAS  Systems. Paper 92-73.05,
Proceedings of the 85th Annual  Meeting and
Exhibition,  Air  &  Waste  Management
Association, Pittsburgh, PA.
 Spellicy, R.L., W.F. Herget, and J.A. Draves.
 1994.  Integrated  Stack  and Open-Path
 Measurements of Industrial Facilities Using
 FTIR.   SP-89   Optical   Sensing   for
 Environmental Monitoring, Air  &  Waste
 Management Association, Pittsburgh, PA,
 pp. 105-122.

 Spellicy, R., and  S.E. Hall.  1997. Early Test
 Results  on  a  FTIR   Industrial   Process
 Monitoring  System.  Proceedings  of  the
 AWMA Annual Meeting in Toronto.  Air &
 Waste Management Association, Pittsburgh,
 PA.

 Spellicy. R. 1997.  Analytical  Methods for
 Optical Remote Sensing. Proceedings of the
 AWMA Annual Meeting in Toronto.  Air &
 Waste Management Association, Pittsburgh,
 PA.

 Spence,   J.W.,  and  P.L.  Hanst.  1978.
 Oxidation of Chlorinated  Ethanes.  J.  Air
 Pollut. ControlAssoc. 28(3):250-253.

 Spence, J.W., P.L. Hanst, and B.W. Gay, Jr.
 1976. Atmospheric Oxidation of   Methyl
 Chloride,   Methylene   Chloride,   and
 Chloroform. J. Air  Pollut. Control Assoc.
 26(10):994-996.

 Spence, J.W.,  E.O.  Edney, and P.L.  Hanst.
 1 978. Peroxychloroformyl Nitrate: Synthesis
 and  Thermal Stability. Chem.  Phys. Lett.
 56(3):478-483.

 Spence, J.W.,  E.O.  Edney, and P.L.  Hanst.
 1 980. Infrared and Ultraviolet Spectral Study
 of HOCI. J. Air  Pollut.  Control  Assoc.
 30(1):50-52.

 Stedman, D.H., and S.E. McLaren.  1990.
 Flux Measurements Using Simultaneous Long
 Path Ultraviolet and Infrared Spectroscopy.
 Paper  90-86.6,  Proceedings of the 83rd
Annual Meeting and Exhibition,  Air & Waste
 Management Association, Pittsburgh, PA.
                                      12-21

-------
                                                                     TR-4423-99-03
 Stephens,  E.R.  1958.  Long-Path  Infrared
 Spectroscopy  for  Air  Pollution  Research.
 Appl. Spectrosc. 12:80.

 Stephens, E.R., P.L. Hanst, and R.C. Doerr.
 1957.   Infrared   Spectra  of   Aliphatic
 Peroxyacids. Anal.  Chem. 29:776-777.

 Stephens,  R.D.,  and  S.H.  Cadle.  1991.
 Remote Sensing Measurements of Carbon
 Monoxide Emissions from On-Road Vehicles.
 J. Air Waste Manage. Assoc. 41:39.

 Stone, J.M. 1963. Radiation  and Optics.
 McGraw-Hill, New York.

 Strang, C.R., and  S.P. Levine.  1989. The
 Limits  of  Detection for the Monitoring  of
 Semiconductor Manufacturing Gas and Vapor
 Emissions  by  Fourier  Transform  Infrared
 (FTIR)   Spectroscopy.   Am.  Ind.   Hyg.
 Assoc. J.  50(2):78-84.

 Strang, C.R., S.P. Levine, and W.F. Herget.
 1 989. A Preliminary Evaluation of the Fourier
 Transform Infrared (FTIR) Spectrometer as a
 Quantitative Air Monitor for Semiconductor
 Manufacturing  Process  Emissions. Am. Ind.
 Hyg. Assoc. J. 50(2):70-77.

 Stuffier, T., V. Klein, and M. Resch. 1995.
 C02 LIDAR System  for the  Detection  of
 Atmospheric Pollutants  and Comparisons  of
 the Measurement Results with Those  of a
 Passive Remote Sensing System:  An FTIR
 Spectrometer. SPIE Specialty Conference in
 Munich, Vol. 2505, International Society for
 Optical Engineering, Bellingham, WA.

 Tannenbaum, H.,  R.L.  Byer, S.F. Clifford,
 K.S. Fu, E.D. Hinkley, T.  Jaefer, A.G. Kjelass,
 K.W. Mill, M. Slatkine, and  A. Wood. 1976.
 Long-Path  Monitoring   of   Atmospheric
 Pollutant Gases.  Opt.  Quantum Electron.
 8:194-196.

Tate,  J.D., J.P.  Chauvel,  and  K.  Taylor.
 1994.  Application  of Industrial  Open-Path
 FTIR  for  Continuous  Monitoring. SP-89
 Optical   Sensing   for   Environmental
 Monitoring,  Air  &  Waste  Management
 Association, Pittsburgh, PA, pp. 791-806.

 Theriault, J.M., C. Bradette, and J. Gilbert.
 1996. Atmospheric Remote Sensing with a
 Ground Based Spectrometer System. Proc.
 SPIE 2744.

 Theriault, J.M., L. Bissonnette, and G. Roy.
 1997. Monitoring Cloud Parameters Using a
 Ground  Based  FTIR  System. Proc. SPIE
 3106:7-16

 Thomas,  M.J.,  J.L. Hudson,  G.A.  Marotz,
 D.D. Lane, and R.E. Carter, Jr. 1992. Data
 Variability and Use Factors Compared Based
 on Open-Path  FTIR and  Canister Sampling
 Techniques. SP-81 Optical Remote Sensing.
 Applications to Environmental and Industrial
 Safety Problems, Air & Waste Management
 Association, Pittsburgh, PA, pp. 133-144.

 Thomas,   M.J., J.L.  Hudson, D.D. Lane,
 R.E. Carter, Jr., and G.A. Marotz.  1992. A
 Statistical and Data Use Comparison of a
 Canister-Based  Sampling  Technique with
 That of Open-Path Remote Sensing. Paper
 92-73.07, Proceedings of the 85th Annual
 Meeting   and  Exhibition,   Air  &   Waste
 Management Association, Pittsburgh, PA.

 Todd,    L.A.   1992.   Optical   Remote
 Sensing/Computed Tomography Systems for
 Workplace Exposure  Assessments.  SP-81
 Optical Remote  Sensing.  Applications   to
 Environmental   and  Industrial   Safety
 Problems,  Air   &  Waste   Management
 Association, Pittsburgh, PA, pp. 356-360.

Todd,  L., and  D. Leith.   1990.   Remote
 Sensing  and  Computed  Tomography   in
 Industrial Hygiene. Am. Ind. Hyg. Assoc.  J.
 51(4):224-233.

Tolansky, S.  1962.  An  Introduction   to
Interferometry. John Wiley and Sons, New
York.
                                      12-22

-------
                                                                     TR-4423-99-03
Toth,  R.A.,  R.L.  Abrams,  G.  Birnbaum,
S.T. Eng, R.A. McClatchey, P.E. Nordal, and
S.O.   Olsen.   1976.   Infrared  Spectral
Properties of Atmospheric Molecules. Opt.
Quantum Electron. 8:1 91 -1 94.

Tso, T.L., W.C. Liao, and S.I. Chang.  1992.
Portable Long Open Path FTIR Applied in In-
Situ  Measurement  of  Trace  Gases  of
Ambient   Air   Pollution.   Proc.   SPIE
1637:95-106.

Tuazon, E.G., R.A. Graham, A.M. Winer, R.R.
Easton, J.N. Pitts, Jr., and P.L. Hanst.  1 978.
A Kilometer Pathlength  Fourier-Transform
Infrared System  for the  Study of  Trace
Pollutants   in  Ambient  and  Synthetic
Atmospheres. Atmos. Environ. 12:865-875.

Tuazon, E.G., A.M. Winer, R.A.  Graham, and
J.N.   Pitts,   Jr.   1980.  Atmospheric
Measurements  of  Trace  Pollutants  by
Kilometer-Pathlength FT-IR  Spectroscopy.
Advances in Environmental  Science and
Technology, Vol. 10 (J.N. Pitts, Jr., and R.L.
Metcalf, Eds.), John Wiley and Sons, New
York, pp. 259-299.

University of  South  Florida.  1993.  USF
HITRAN-PC. University of South  Florida,
Tampa, FL.

Vandaele, A.C.,  M. Carleer, R. Colin, and
P.C.  Simon. 1993. Detection  of  Urban
Ozone, Nitrogen Dioxide, Formaldehyde, and
Sulfur  Dioxide  Using  Fourier Transform
Spectroscopy. Proc. SPIE 1 71 5:288-92.

van de Hulst, H.C. 1981. Light Scattering by
Small  Particles.  Dover  Publications,  New
York.

Vaughan,  W.M.  1991.  Remote Sensing
Terminology. J.  Air Waste Manage. Assoc.
41(111:1489-1493.

Vaughan,   W.M.,    J.O.   Zwicker,
R.H. Dunaway, and P.  Martino. 1994. The
 Feasibility of Using Optical Remote Sensing
 Systems to  Determine  Fugitive  Emissions
 from Petroleum Refineries: A Review of the
 Present State  of the  Technology. SP-89
 Optical   Sensing   for  Environmental
 Monitoring,   Air  &  Waste  Management
 Association, Pittsburgh, PA, pp. 517-528.

 Vazquez, G.J. 1995. FTIR Remote Sensing
 of Atmospheric Species: Application to Air
 Pollution. Proc.  SPIE 2365:438-463.

 Walsh,  J.W.T.  1965.  Photometry. Dover
 Publications,  New York.

 Walter, B. 1889. Ann. Physik 36:502-518

 Wang, C. D., W.T. Walter, and R.H. Kagann.
 1995. Neural Network  and Classical Least
 Squares Methods for Quantitative Analysis in
 Remote Sensing FTIR Systems.  Proc. SPIE
 2366:251-262.

 Ward, T.V., and H.H. Zwick 1975. Gas Cell
 Correlation  Spectrometer:  GASPEC. Appl.
 Opt. 14:2896.

 Wassell, P.T., R.P. Wayne, J. Ballard, and
 W.B.    Johnston.   1989.   Laboratory
 Spectroscopic Studies  of  Atmospherically
 Important Radicals Using Fourier Transform
 Spectroscopy, J. Atmos. Chem. 8(1):63-85.

Webb,  J.D.,  K.R. Loos,  C.L.  Yao,  D.C.
 Kreuger,  S.A.   Reid,   S.  Williamson,
 M.J. DeLong, and P.I. Chipman. 1995. Initial
Applications of Open Path FTIR to Three
Shell  Oil  Facilities—a  Chemical   Plant,  a
Refinery, and  a  Gas  Processing  Plant.
Proceedings  of  the  AWMA   Specialty
Conference in San Francisco.  Air & Waste
Management  Association, Pittsburgh, PA.

Weber, K., H.J.  van de Wiel, A.C.F. Junker,
and C.  de  LaRiva.   1992.  Definition  and
Determination of Performance Characteristics
of Air Quality Measuring Methods  as Given
by  the  International   Organization   for
                                      12-23

-------

                                                                     TR-4423-99-03
 Standardization (ISO): Applicability to Optical
 Remote  Sensing.  SP-81.  Optical  Remote
 Sensing. Applications to  Environmental and
 Industrial Safety  Problems,  Air & Waste
 Management  Association,  Pittsburgh,  PA,
 pp. 30-42.

 Weber, K., V. Klein, and  C. Werner. 1992.
 The Measurement  of Gaseous Air Pollutants
 in  the Troposphere  by Optical  Remote
 Sensing: Developments and Applications in
 Germany. SP-81  Optical Remote Sensing.
 Applications to Environmental and Industrial
 Safety Problems, Air & Waste Management
 Association, Pittsburgh, PA, pp.  44-54.

 Werner,   C.,  R.   Haus,   F.  Koepp,   and
 K. Schafer. 1995.  Remote Sensing  of Mass
 Fluxes of Trace Gases in the Boundary.Layer.
 Appl. Phys. B 61:249-251.

 White,  J.U.  1942.  Long  Paths of  Large
 Apertures. J.  Opt.  Soc. Am. 32:285.

 Whitecraft, W.K., and K.N. Wood. 1 990. Use
 of Remote Sensing to Measure Wastewater
 Treatment Plant Emissions. Paper 90-86.4,
 Proceedings of the  83rd Annual Meeting and
 Exhibition, Air  &  Waste  Management
 Association, Pittsburgh, PA.

 Willard, H.H.,  L.L.  Merritt,  and  J.A. Dean.
 1 974. InstrumentalMethods of Analysis, 5th
 Ed., D. Van Nostrand, Princeton, NJ.

 Winkler, C., H. DeWitt, T.  Eisenmann, and
 H. Mosebach. 1 997. Surveillance System for
 Air Pollutants by Combination of the Decision
 Support  System COMPAS and the FTIR
 Optical Remote  Sensing  System  K300.
 Proceedings of the 1997  1st International
 Conference on Measurements  and Mode/ing
 in Environmental Pollution.  Madrid,  Spain.
 Computational    Mechanics   Publ.,
Southhampton, England.

 Witkowski, M.R., T.L. Marshall, C.T. Chaffin,
 R.M.  Hammaker, and W.G. Fateley. 1992.
 Field   Monitoring   of  Volatile   Organic
 Compounds (VOCs)  with  a  Mobile  Fourier
 Transform  Infrared  (FT-IR)   Spectrometer
 System.   1992.  SP-81   Optical  Remote
 Sensing.  Applications to Environmental and
 Industrial Safety Problems,  Air &  Waste
 Management Association, Pittsburgh, PA,
 pp. 157-168.

 Woods,  P.T.,  R.H.  Partridge, M.J.  Milton,
 B.W.   Jolliffe,  and  N.R.  Swann.  1992.
 Remote  Sensing   Techniques  for  Gas
 Detection,  Air Pollution and Fugitive  Loss
 Monitoring. SP-81 Optical  Remote Sensing.
 Applications to Environmental and Industrial
 Safety Problems, Air & Waste Management
 Association, Pittsburgh, PA, pp.  3-29.

 Wu,  R.T.,  S.Y.  Chang,  Y.W.  Chung,
 H.C. Tzou, and T.L. Tso. 1995. FTIR Remote
 Sensor Measurements of Air Pollutants in the
 Petrochemical Industrial Park. Proceedings of
 the SPIE  Conference on Infrared  Technology
 XXI in San Diego.  International  Society for
 Optical Engineering, Bellingham,  WA.

 Xiao,  H-K., S.P. Levine, and J.B. D'Arcy.
 1 989. Iterative Least-Squares Fit Procedures
 for the  Identification  of  Organic  Vapor
 Mixtures   by  Fourier  Transform Infrared
 Spectro-photometry.  Anal. Chem. 61(24):
 2708-2714.

 Xiao,   H-K.,  S.P.   Levine,  J.B. D'Arcy,
 G.  Kinnes,  and  D.  Almaguer.   1990.
 Comparison of the Fourier Transform Infrared
 (FTIR) Spectrophotometer and the Miniature
 Infrared   Analyzer   (MIRAN)   for  the
 Determination of Trichloroethylene (TCE) in
the Presence of Freon-11 3 in Workplace Air.
Am. Ind.  Hyg. Assoc. J. 51 (7):395-401.

 Xiao,   H-K.,  S.P.  Levine,  W.F.  Herget,
J.B. D'Arcy,  R. Spear, and  T. Pritchett.
 1991.  A Transportable, Remote Sensing,
 Infrared Air  Monitoring System. Am.  Ind.
Hyg. Assoc. J. 52(11 ):449-457.
                                      12-24

-------
                                                                     TR-4423-99-03
Ying, L-S., and S.P. Levine. 1989. Evaluation
of the  Applicability  of  Fourier Transform
Infrared (FTIR) Spectroscopyfor Quantitation
of the  Compounds  of  Airborne  Solvent
Vapors  in  Air. Am.  Ind.  Hyg. Assoc. J.
50(7):360-365.

Ying, L-S., and S.P.  Levine. 1989. Fourier
Transform  Infrared Least-Squares  Methods
for   the    Quantitative   Analysis   of
Multicomponent Mixtures of Airborne Vapors
of Industrial Hygiene  Concern. Anal. Chem.
61(71:677-683.

Ying, L-S.,  S.P.  Levine, C.R.  Strang,  and
W.F.  Herget.  1989.  Fourier  Transform
Infrared (FTIR) Spectroscopy for Monitoring
Airborne Gases  and  Vapors  of Industrial
Hygiene Concern. Am. Ind. Hyg. Assoc. J.
50(7):354-359.

Yokelson,   R.J.,   D.W.T.   Griffith,
J.B.  Burkholder,  and D.E. Ward.  1996.
Accuracy  and  Advantages  of  Synthetic
Calibration of Smoke Spectra. Proceedings of
the AWMA Specialty Conference  in  San
Francisco.    Air  &   Waste   Management
Association, Pittsburgh, PA.
Yost, M.G.,  H.K.  Xiao,  R.C.  Spear,  and
S.P. Levine. 1992. Comparative Testing  of
an  FTIR Remote Optical Sensor with Area
Samplers   in   a  Controlled  . Ventilation
Chamber.   Am.   Ind.   Hyg.  Assoc.   J.
53(101:611-616.

Zhang,  Y.,  D.H. Stedman,  G.A.  Bishop,
P.L.  Guenther,   S.P.   Beaton,   and
J.E. Peterson.  1993. On-Road Hydrocarbon
Remote  Sensing in the Denver Area. Environ.
Sci. Technol. 27(9): 1885-1 891.

Zwicker,   J.O.,   W.M.    Vaughan,
R.H. Dunaway,  and R.H. Kagann.  1994.
Open-Path   FTIR   Measurements   of
Carpet/Water   Sealant   Emissions   to
Determine Indoor Air  Quality.  SP-89 Optical
Sensing for Environmental Monitoring, Air &
Waste Management Association, Pittsburgh,
PA, pp. 225-241.

Zwicker, J.O. 1 995. Intercomparison of Data
from Open-Path Fourier Transform Infrared
Spectrometers Collected During the EPA/API
Cooperative Remote Sensing Dispersion Field
Study  at   Duke  Forest,  North  Carolina,
January 9  to 27, 1995. Proceedings of the
AWMA  Specialty   Conference   in   San
Francisco.   Air  &   Waste  Management
Association, Pittsburgh, PA.
                                      12-25

-------