&EPA
              United States
              Environmental Protection
              Agency
              Office of Acid Deposition,
              Environmental Monitoring and
              Quality Assurance
              Washington DC 20460
EPA/600/4-88/018
April 1988
              Research and Development
National  Stream
Survey -  Phase I

Quality Assurance
Report

-------
     SUBREGIONS  OF  THE  NATIONAL  STREAM SURVEY-PHASE I
                               Northern
                            Appalachians (2Cn)
                                              Valley and Ridge (2Bn)
     Southern Blue Ridge (2As)
       (Pilot Study)
   Poconos/Catskills (ID)

         NY\
                                 v  r >    I fPK-yVf'-
                                 ^ i Jill
                                   7   J&P-1-!**! "•
                                   /- Tl^^PW
                                   / Ji^S >:i/«:-:-:K
  Mid-Atlantic
Coastal Plain (3B)
Ozarks/Ouachitas (2D)
                                                          Piedmont
                                                             (3A)
                          Florida (3C)
Southern Appalachians (2X)

-------
                                             EPA/600/4-88/018
                                             April 1988
National  Stream Survey
             Phase  I
    Quality Assurance Report
             A Contribution to the
   National Acid Precipitation Assessment Program
                    U.S. Environmental Protection Agency
                    Office of Research and Development
                        Washington, DC 20460
          Environmental Monitoring Systems Laboratory - Las Vegas, NV 89119
              Environmental Research Laboratory - Corvallls, OR  97333
                                            Protection Agency
                                         / (5PL-16)
                                       r Street, Hoom 1670
                                       oOo04

-------
                                      NOTICE


     The information in this document has been funded wholly or in part by the United States
Environmental Protection Agency under contract number 68-03-3249 to Lockheed Engineering and
Management Services Company, Inc., and contract number 68-03-3246 to Northrop Services, Inc.
It has been subject to the Agency's peer and administrative review, and it has been approved for
publication as an EPA document.

     The mention of trade names or commercial products does not constitute endorsement or
recommendation for use.

     This document  is one volume of a  set which fully describes the  National Stream Survey.
The complete document set includes the major data  report,  pilot survey data report, quality
assurance plan, analytical methods manual,  field operations report, processing laboratory
operations report,  and  quality assurance report.   Similar sets are  being produced for each
Aquatic Effects  Research Program component  project.  Colored covers, artwork, and use of the
project name in the document title serve to identify each companion document set.

     Proper citation of this document is:

     Cougan, K. A., D. W. Sutton, D. V. Peck, V.  J. Miller, and J. E. Pollard. 1988.  National Stream
     Survey -  Phase I:  Quality Assurance Report.   EPA/600/4-88/018,   U.S.  Environmental
     Protection Agency, Environmental Monitoring Systems Laboratory, Las Vegas, Nevada.

-------
                                     ABSTRACT


     The National Stream Survey - Phase I, conducted during the spring of 1986, was designed
to assess quantitatively the present chemical status of streams in regions of the eastern United
States where aquatic resources are potentially at risk as a result of acidic deposition. A quality
assurance program was implemented to ensure consistency in the collection and analysis of
water samples and to verify the reported results.  In addition, the quality assurance program
provides data users with  quantitative and qualitative documentation of the quality of the data
base in terms of representativeness, completeness,  and comparability and the quality of the
analytical results in terms  of detectability, accuracy, and precision. This quality assurance report
describes the major design and operational aspects of the quality assurance program and the
final assessment of the quality of the  National Stream Survey  data  base.   This report also
describes sampling and analytical problems that occurred during the survey and the corrective
actions implemented.

      The survey data  base  is sufficiently  representative and  complete so  that population
estimates based on chemical characteristics  can be computed and  interpreted.  The results of
the survey can be compared to the results of the Phase I Pilot Survey, to other data bases of the
National Surface Water Survey, and to  other existing or future water quality data bases with
similar design, methodology, reporting units, and quality assurance.

      There are only a few cases in which data interpretation may be limited by data quality in
terms of detectability,  accuracy, and precision.  In most of these cases, the limitations affect
only  interpretation of  measurements  at  low concentrations.   A  model-based  approach to
evaluating systematic errors is presented as an appendix to this report.  Suggestions for future
surveys include performing on-site inspections of all operations earlier in the survey so that most
potential problems can be identified before they affect data quality and modifying the procedures
for preparation of synthetic audit samples to facilitate improved estimates of accuracy.

       This report is submitted in partial fulfillment of contract number 68-03-3249 by Lockheed
 Engineering  and Management Services Company,  Inc., under the sponsorship of the U.S.
 Environmental Protection  Agency.  This report covers a field work period from February 1986
 through May 1986; data  verification was completed in March 1987 and  data  evaluation was
 completed in March I988.
                                             in

-------
                                         CONTENTS
Abstract	'  •	viii
Figures	•	   jx
Tables	
Abbreviations and Acronyms	•	
Acknowledgment	XIV

      1.    Introduction	
                 Background	^
                 National Stream Survey	~
                 Survey Participants	3
                 National Stream Survey Documents	4
      2.    Conclusions and Recommendations	5
                 Conclusions	^
                      Detectability	 5
                      Accuracy	6
                      Precision	jj
                 Recommendations	~
                      Field and Processing Laboratory Activities	6
                      Analytical Laboratory Activities	7
                      Data Evaluation	7
                      Design of Quality Assurance Program	7
      3. Design of the Quality Assurance Program for the
            National Stream Survey - Phase I	9
                 Stream Characteristics and Data Quality Objectives	9
                 Statistical Design of the National Stream Survey	13
                 Sample Collection and Analyses-Quality
                  Assurance  and Quality Control	13
                       Quality Assurance Samples	13
                       Quality Control Samples	18
                 Data Base Management	20
                       Raw Data Set (Data Set 1)	20
                       Verified Data Set (Data Set 2)	21
                       Validated Data Set (Data Set 3)	21
                       Enhanced Data Set (Data Set 4)	23
                 Differences Between NSS-I and the Pilot Survey	23
                       Processing Laboratory Sample Holding Times	23
                       Processing Laboratory Location	23
                       Field pH Measurement	24
                       Fractionation and Determination of Aluminum Species
                       Matrix Spike Samples	25
                       Phosphorus Measurements	25
                       Specific Conductance Measurements	25
       4. Operations of the Quality Assurance Program	27
                 Selection of Analytical Laboratories	27
                 Training of Field, Processing Laboratory, and
                   Quality Assurance Personnel	28
                  Field Sampling Operations	28
                  Processing  Laboratory Operations	29

-------
                               CONTENTS (Continued)


              Analytical Laboratory Operations	                   33
              Monitoring	    	36
                   Communications	         	3»
                   On-Site Evaluations	       	36
                   Data Base Management and Data Verification	          37
                   Data Validation and Data Base Management......          45
                   Enhanced Data Set	     	46
   5. Results and Discussion - Assessment of Operations.  .  .  .  .  .   .  .  .  '.'.'' 47
              Field Sampling Operations and Protocol Changes  .......        47
              Processing Laboratory Operations and
               Protocol Changes	                   47
                   Specific Conductance Measuring	.47
                   Nitrate Contamination	     	88
                   Total Monomeric and Nonexchangeable Monomeric
                    Aluminum Measurements	              51
              Analytical Laboratory Operations and Protocol
               Changes	                     52
                   Effect of Large Sample Loads	52
                   Centrifuge Tubes for Extractable Aluminum Analyses    	53
                   Laboratory pH Data	                 53
                   ANC and BNC Recalculations	'.'.'.'.'.	53
                   Sample Holding Time and Reanalysis for Metals  ....           54
                   Reanalyses of Nitrate, Sulfate, and Chloride	 55
                   Total Extractable Aluminum Values Greater than
                    Total Aluminum Values	            55
                   Data Reporting Errors	5g
             Data Verification Activities	    	56
                   Review of Field Data Forms	'.'.'.'.'.'.'''' 57
                   Review of Processing Laboratory Forms  ......     .  .      57
                   Review of Analytical Data Forms and
                    Correction of Data	                    co
                  Changes to Analytical Data Applied at EMSL-LV   ......    . 58
                  Modifications to the Exception-Generating
                    Programs and New Data Qualifier Flags	       59
                  Delivery of Verification Tapes	      	5g
                  Data Base Audit	    	5g
6.  Assessment of Data Quality	61
             Introduction	       	g^
             Completeness.  .	61
             Comparability	     	65
             Representativeness	66
             Detectability	    	66
                 Assessment of Method-Level Limits of
                   Detection  .	              gg
                 Assessment of System-Level Detectability
                   (Background)	                72
                 Discussion and Summary: Detectability	.74
                                      VI

-------
                                CONTENTS (Continued)

                                                                                7fi
               Accuracy	Qn
                    Percent Accuracy Estimates for Laboratory 1	»"
                    Percent Accuracy Estimates for Laboratory 2	83
                    Percent Accuracy Estimates for the Processing
                      Laboratory	^
                    Interlaboratory Bias	°°
                    Discussion and Summary	™
               Precision	
                    Precision Estimates Derived from Field
                      and Laboratory Duplicate Samples	.  .  . 92
                    Precision Estimates Derived from Audit
                      Sample Measurements	
                    Comparison of Precision Estimates
                    Discussion and Summary: Precision
               Summary of Data Quality Assessment	
                    Charge Balances	J^
                     Specific Conductance Check	^
                     Comparison of ANC Values	113

References	
Appendices
                                                                                 1^0
     A. Preparation of Audit Samples	
     B. Data Qualifiers	•	
     C. Acceptance Criteria	
     D. Estimating Relative Interlaboratory Bias for
           the National Stream Survey	

                                                                   	194
Glossary	
                                           VII

-------
                                       FIGURES


 Number
  2      National Stream Survey study areas
         Organization of the National Surface Water Survey, Showing
            the two major components, the National Lake Survey and
            the National Stream Survey	                           2
  3      The relationship of the quality assurance and quality control
           samples to the collection and analysis process	          17

  4      Routine and quality assurance samples collected during the
               "	18

  5      Data Base Management System	                         03

  6      Field sampling activities for the National Stream Survey -
           Phasel	    	30

  7      Data form flow, National Stream Survey - Phase I „	            31

  8      Flow of samples and data from the field through the
           processing laboratory	                         32

  9       Preparation and preservation procedures for each aliquot
           at the preparation laboratory for the National Stream
           Survey - Phase  I	                          _.

 10     Analytical data verification, initial verification for the
           National Stream Survey - Phase 1	                 38

 11      Analytical data verification, final verification for the
        National Stream Survey - Phase I	                     39

 12      Percentage of National Stream Survey - Phase I routine
           samples with measured concentrations below the system
           decision limit	                            75

 13      Sum of cations versus sum of anions for routine samples,
           National Stream Survey-Phase I	'                       113

 14      Measured versus calculated specific conductance at 25"C
           for routine samples, National Stream Survey - Phase 1	114

15      Calculated carbonate alkalinity versus measured ANC for
           Routine Samples, National Stream Survey - Phase 1	114
                                         VIII

-------
                                      TABLES
Number
 1      Chemical and Physical Variables Measured During the
           National Stream Survey - Phase I	 11

 2      Quality Assurance and Quality Control Samples Used in the
           National Stream Survey - Phase I	15

 3      Number of Routine and Quality Assurance Samples Collected
           and Analyzed During the National National Stream Survey -
           Phase I	"	19

 4      Differences Between the National Stream Survey - Phase I
           and the NSS-I Pilot Survey	25

 5      Maximum Holding Time Requirements Before Sample Analyses
           at Analytical Laboratories, National Stream Survey -
           Phase I	35

 6      Exception-Generating Programs Within the AQUARIUS II
           Data Review and Verification System	42

  7      Chemical Reanalysis Criteria for Sample Ion Balance
           Difference and Percent Specific Conductance Difference	42

  8      Factors for Determining the Conductances of Ions
           (juS/cm at 25 *C)	44

  9      Significant Findings Concerning Field Sampling Operations
           and Their Effect on Data, National Stream Survey -
           Phase I	•	48

 10      Significant Findings Concerning Processing Laboratory
           Operations and Their Effect on Data, National Stream
           Survey - Phase I	49

  11      Recommended Number of Decimal Places	57

  12      Changes to Sample Numeric Data Incorporated in the Verified
            Data Set, National Stream Survey - Phase I	    	58

  13      Analytical Data Quality Objectives for Dectectibility,
            Precision, and Accuracy for the National Stream Survey -
            Phase I	•	62
                                            IX

-------
                                TABLES (continued)

Number
                                                                                Page
 14     Summary of Streams Visited During the National Stream
           Survey-Phase I	                            6g

 15     Estimates of Limits of Detection Based On Analyses of
           Laboratory Blank Samples, National Stream Survey -
           Phase I	                 6g

 16     Estimates of Limits of Detection Based On Analyses of Field
           Blank Samples, National Stream Survey - Phase 1	      70

 17     Estimates of System Decision Limits Based on Analyses of
           Field Blank Samples Pooled Across Laboratories, National
           Stream Survey - Phase 1	                 73

 18     Range and Central Tendency Statistics for Analyte Concentra-
           tions in Routine Stream Samples, National Stream Survey -
           Phase I	                      76

 19     Summary Statistics for Selected Variables for EPA Reference
           Samples Measured at the Support Laboratory, National
           Stream Survey - Phase I	     78

 20     Theoretical and Index Values for Analyses of Synthetic and
           Natural Audit Samples, National Stream Survey -  Phase I.    	79

 21      Percent Accuracy Estimates for Laboratory 1 Measurements
           of Synthetic Audit Samples, National Stream Survey -
           Phase I	
                                                                                 81
22      Percent Accuracy Estimates for Laboratory 1 Measurements
          of Selected Variables in Natural Audit Samples, National
          Stream Survey - Phase 1	          82

23      Percent Accuracy Estimates for Laboratory 2 Measurements
          of Synthetic Audit Samples, National Stream Survey -
          Phase I	                84

24      Percent Accuracy Estimates for Laboratory 2 Measurements
          of Selected Variables, in Natural Audit Samples,
          National Stream Survey - Phase I	85

25      Percent Accuracy Estimates for Processing Laboratory
          Measurements of Synthetic Audit Samples, National
          Stream Survey -  Phase 1	                       88

-------
                               TABLES (continued)
Number
 26      Estimates of Percent Accuracy for Analytes Measured at the
           Processing Laboratory Based on Natural Audit Samples,
           National Stream Survey - Phase I	89

 27      Components of Variance Included in Precision Estimates
           from Routine-Duplicate Pairs and Audit Samples,
           National Stream Survey - Phase -1	92

 28      Method-Level and System-Level Precision Estimates by
           Concentration Ranges of Variables (Laboratories
           Pooled), National Stream Survey - Phase 1	94

 29      Summary Statistics for Among-Batch Precision Based On
           Pooled Audit Sample Data, National Stream Survey -
           Phasel	102

 30     Comparison of Method-Level, System-Level, and Among-Batch
           Precision Estimates, National Stream Survey - Phase 1	105

 31      Summary of Data  Quality Assessments for Chemical Variables
           with Respect to Dectectiblity, Accuracy, and Precision,
           National Stream Survey - Phase I	111
                                           XI

-------
                            Abbreviations and Acronyms
  Acroyms

  AERP
  AQUARIUS

  AQUARIUS II

  CLP
  DQO
  ELS-I
  EMSL-LV

  EPA
  ERL-C

  IFB
  IDL
  NAPAP
 NSS-I
 NSWS
 NTU
 ORNL
 PCU
 QA
 QC
 QCCS
 %RSD
 SAS
 SDL
 SOW
  Aquatic Effects Research Program
  Aquatics Quality Assurance Review, Interactive Users'
   System
  modification of the AQUARIUS system developed for the
   National Stream Survey - Phase I
  Contract Laboratory Program
  data quality objective
  Eastern Lake Survey - Phase I
  U.S. Environmental Protection Agency, Environmental
   Monitoring Systems Laboratory, Las Vegas, Nevada
  U.S. Environmental Protection Agency
  U.S. Environmental Protection Agency, Environmental
   Research Laboratory, Corvallis, Oregon
  invitation for bid
  instrument detection limit
  National Acid Precipitation Assessment  Program
 National Stream Survey - Phase I
 National Surface Water Survey
 nephelometric turbidity unit
 Oak Ridge National Laboratory
 platinum cobalt unit
 quality assurance
 quality control
 quality control check sample
 percent relative  standard deviation
 Statistical Analysis System
 system decision limit
 statement of work
Variables and Units
Al-ext
Al-mono
Al-nex
Al-total
ANC
BNC
Ca
cr
co32-
aluminum, total extractable
aluminum, total monomeric
aluminum, nonexchangeable monomeric
aluminum, total
acid-neutralizing capacity
base-neutralizing capacity
calcium
chloride
carbonate
                                         XI!

-------
                   Abbreviations and Acronyms (continued)
Variables and Units (continued)
Cond-in situ
Cond-lab
Cond-PL
DIC-closed
DIC-eq
DIC-open
DO
DOC
F"
Fe
H+
HOD,'

 Mg
 mg/L
 Mn
 Na
 NH4+
 NO '
 OH3
 P
 pH
 pH-ANC
 pH-BNC
 pH-closed
 pH-eq
 pH-field
 SiO9
 so4%-
 /neq/L
 /L/S/cm
specific conductance measured in the field
specific conductance measured in the analytical laboratory
specific conductance measured in the processing laboratory
dissolved inorganic carbon, closed
dissolved inorganic carbon, equilibrated
dissolved inorganic carbon, open system
dissolved oxygen
dissolved organic carbon
fluoride, total dissolved
iron
hydrogen ion
bicarbonate
potassium
magnesium
milligrams per liter
manganese
sodium
 ammonium
 nitrate
 hydroxyl
 phosphorus, total dissolved
.negative logarithm of the hydrogen-ion concentration
 pH, initial (acid titration for ANC)
 pH, initial (acid titration for BNC)
 pH, closed system
 pH, equilibrated
 pH, measured in the field
 silica
 sulfate
 microequivalents per liter
 microsiemens per centimeter
                                            xiii

-------
                               ACKNOWLEDGMENTS
      G.  M.  Aubertin and M. P. Aubertin (Southern Illinois University), J.  K.  Taylor (Quality

 ^pU±nC?th0nSUltanVnd K  *  Tonnessen (California Air Resources Board) provded peer
 reviews of this report. The,r thoroughness in reviewing the report under a tight time constraint Is

 *™
                                         wh°
                       -
 K N pS (     ,? £ Un!frS'ty)> M' J" Sale and J- M. Coe (Oak Ridge National Laboratory)  and
 K. N. Eshleman (Northrop Services, Inc., Corvallis, Oregon).



 i ac v A """I? 6r °1 emP|ovees of Lockheed Engineering and Management Services Company Inc
 Las Vegas, Nevada  assisted in preparing this report. The authors greatly appreciate the write

 contnbut.ons of D. J. Chaloud, J. L Engels, R. Hoenicke, and C. M. Monaco. J  E  Teberg provided


 for SFasZa^  "? *"* f "^ ^ 8Uth0r8 a'S° acknow^9e the foliowln'indSals
          5186 and constructive input: J.  K.  Bartz, R. Corse, S. K. Drouse R E  Enwall
M  ,                                       .   .       ,   .       ,  .   .   o


L ' K  £L  MC' i0!^' T ' c\.Ha9ley'  M'  L  HOPPUS' D- C' Hillman' J- D- Hunter J. I
LK  Marks
    K         M  i        T  c.'   '           '   -  '        '  -   -  uner  .    au
    K  Marks  MJ. Miah, T.  E.Mitchell-Hall, S. L. Pierett,  P.P.  Showers, M. E. Silverstein

 L. A. Stanley, M. A. Stapanian, and A. D. Tansey.                                   onvws>iein,
    aMM    astsoc?2d  witn other organizations  also provided assistance and advice-


   «t     A( r'V!rSlty °Tf NeVada' LaS Ve9as): S' D' Edland-  T- J- Per"iutt, and T. S. Stock ng

             vCaNrS'vn^ San "afa0el' California); E- Canelli (New York State  DepartmeTS

                                      Shepard and R" °- •*««*» (Global Geochemistry
nmnr/mna^?Uid.ar?Ce ''" devel°Pin9.  implementing, and  administering the quality assurance

Directo? and     Nat'°nal Stream Survev '  Phase l was provided by R A. Linthurst,  Program



qnr^^nH^n' "5a"fmann (Utah state University), Technical Directors forThe NaSl Stream'

Spn^'r3  H\    Jchonbrod and D- T- He99em  (U.S. Environmental Protection Agency, Las

Vegas, Nevada), who served as Technical  Monitors for the National Stream Survey - Phase  I

quality assurance program.                                                    y
                                         XIV

-------
                                     Section 1

                                   Introduction
     The  National Stream Survey - Phase  I
(NSS-I) was designed to determine  the
present chemical status of streams in regions
of the  eastern  United States where  aquatic
resources  are potentially at risk as a result of
acidic deposition.  This report describes  the
quality assurance (QA) program employed
during  the NSS-I.   The  QA program  was
designed  to  ensure  consistency  in the
collection  and analysis of  samples, to verify
the reported results, and to inform data users
of the  quality and potential limitations of the
resultant data base. This document evaluates
the QA program itself as well as the quality of
the analytical data base.

     Section 2 presents  conclusions  about
NSS-I  data quality and recommendations
regarding the  QA  program.    Section  3
describes  the design of the QA program and
Section  4 describes the QA  operations.
Section  5  discusses the  results of  the
operational aspects of the QA program and
Section 6 assesses NSS-I data quality.

 Background

      The National Stream Survey is one of  a
 series  of  surveys conducted  as part of the
 National  Acid  Precipitation  Assessment
 Program  (NAPAP).   This  program  is an
 interagency research, monitoring, and
 assessment effort initiated to address  a
 growing concern about the possible effects of
 acidic  deposition on the natural  resources of
 the United States and neighboring countries.
 Congress  established the  NAPAP  as  part
 of the Acid  Precipitation  Act of 1980 to pro-
 vide policymakers with technical information
concerning the extent and the severity of the
effects of acidic deposition.

    The NAPAP  is composed of seven  task
groups. Task Group VI oversees the Aquatic
Effects  Research Program  (AERP),  which is
administrated by the  U.S.  Environmental
Protection Agency (EPA) through its Office of
Acid Deposition, Environmental Monitoring, and
Quality Assurance. One objective of the AERP
is to identify subpopuiations of surface waters
and the associated biota  at risk from acidic
deposition.  The AERP consists of five large-
scale projects that address chronic (long-term)
and acute  (short-term)  exposure  of  aquatic
systems to acidic deposition.

     The  National  Surface Water  Survey
(NSWS), one of the AERP projects, consists of
two components:  the  National Lake Survey
and the National Stream Survey.    Figure 1
shows the relationship of the regional surveys
and monitoring  projects  that  make up the
NSWS. Each component of the NSWS began
with  a  synoptic  survey designed  to
characterize  and quantify the  chemistry of
lakes and streams throughout  the United
States. The focus was on areas  expected to
contain the  majority of low-alkalinity waters.
The National Lake Survey was initiated with a
pilot survey in 1983. Lake surveys took place in
1984 in the eastern United States (Linthurst et
al.,  1986) and in 1985 in  the western United
States  (Landers et  al., 1987).  The National
Stream Survey was initiated with a pilot survey
in  1985 in  the  southern  Appalachian region
 (Messer et al., 1986).  The full-scale synoptic
survey was  conducted in the eastern United
States in the spring of 1986.

-------
                            NATIONAL ACID PRECIPITATION
                           ASSESSMENT PROGRAM (NAPAP)
                                 AQUATIC EFFECTS RESEARCH
                                      PROGRAM (AERP)
                           [NATIONAL SURFACE WATER SURVEY (NSWS)
                    NATIONAL LAKE SURVEY (NLS) |    | NATIONAL STREAM SURVEY
                                                    PHASE I
                                                  PILOT SURVEY
                                                  MID-ATLANTIC
                                               SOUTHEAST SCREENING
                     EPISODES PILOT
       All AERP surveys are designed to yield
 data bases of known  quality through  the
 standardized collection of data from regionally
 typical study sites.   Each AERP  project
 includes an extensive QA program.   Such a
 program is required  of every EPA-funded
 monitoring and measurement effort (Stanley
 and Verner, 1985).

 National  Stream Survey

      The major goals  of the NSS-I were to
 describe and classify streams in the eastern
 United States  target population.    Figure  2
 shows the regions studied during the NSS-I.
 The  NSS-I activities  were  initiated  during  a
 pilot study conducted  in  the Southern  Blue
 Ridge province of the  United States (Messer et
ai.,  1986; Drouse, 1987).  The purpose of the
Phase  I pilot  survey  was  to  evaluate  the
adequacy of the logistics plan, the statistical
sampling design, and the methods proposed
 for the full-scale Phase I study as well as to
 finalize QA and quality control (QC) guidelines
 and data quality objectives.  As a result of the
 pilot study,  which was conducted from  mid-
 March to mid-July of 1985, some changes (see
 Section  3)  were  made  in the  design  and
 operations for the full-scale survey conducted
 in 1986.

    The major collection efforts of the NSS-I
 were conducted in the mid-Atlantic  region of
 the eastern United States,  where survey
 personnel collected more than 1,000  samples
 from approximately 270 streams.  This Mid-
 Atlantic Survey was designed to estimate the
 present degree of acidity of streams  in areas
 that  are characterized  by low  surface-water
 alkalinity, high rates of acidic deposition,  and
 few  lakes.   In  addition,   the  survey  was
designed to  determine for future study which
streams are representative of  stream
subpopulations.   The    Mid-Atlantic  Survey

-------
 El NSS-I PILOT SURVEY
   Figure 2.  National Stream Survey study areas.

covered stream reaches in an area bounded by
the Catskill and  Pocono Mountains  to  the
north, the  North Carolina-Virginia boundary to
the  south,  the approximate  western
boundaries of  Pennsylvania and West Virginia
to the west,  and the  Atlantic Ocean to the
east.   Each  stream reach (segment of the
stream network between  two   tributary
confluences) was sampled twice  during spring
baseflow conditions, March 15 through May 15.
Two sampling  points on each of these reaches
were located just  above the downstream point
of confluence  and just below the  upstream
point of confluence.

      A less  intensive collection effort was
conducted concurrently in the  southeastern
United States.   The  Southeast Screening
Survey was designed to  evaluate specific
areas  for intensive study in the future.   The
screening survey was  conducted in parts of
Virginia,  North  Carolina,  South  Carolina,
Kentucky,  Tennessee,  Mississippi,  Alabama,
Georgia,  Oklahoma,  Arkansas,  and  Florida
(Figure 2) that were  identified by the National
Lake Survey as having a large number of acidic
lakes (Linthurst et al., 1986). One sample was
collected from the upstream and downstream
ends of 180 stream reaches.
    A small-scale episodes pilot survey for a
proposed study of episodic events in streams
(related to  weather conditions that produced
snowmelt  and rainfall)  was conducted in
conjunction with the mid-Atlantic field sampling
effort.  This survey was designed to test the
feasibility  of  using  a  probability-based
sampling  design to  assess  the  extent,
magnitude, duration, and frequency of acidic
episodes on a regional scale.  This study also
tested specific physical and chemical sampling
protocols proposed for the full-scale Episodic
Response Project, another NAPAP project.

     Results  of  the episodes pilot  study
indicated that a synoptic approach to sampling
streams  during episodes  would  not  be
logistically feasible (Hagley  et al., in press).
Although collection  of  30 sets  of  episode
samples was anticipated, dry weather allowed
collection of only 2 complete sets and 7 partial
sets of samples. Based on the results of the
episodes pilot survey, the Episodic Response
Project will use a model-based approach  to
assess the regional importance of episodes to
stream chemistry and biota (Eshleman, 1988).
The results of the episodes pilot survey will not
be discussed further in this report.

     NSS-I sampling  activities included
 locating stream  sites  and  collecting water
 samples and associated data on the physical
 and chemical characteristics of the  streams.
 After collection, the samples were sent  to a
 processing  laboratory where they  were
 organized  into sample batches, analyzed for
 selected chemical and physical variables, split
 into aliquots,  preserved, packed, and shipped
 to the analytical laboratories.   After the
 samples  were  analyzed,   the analytical
 laboratories prepared a report on the analytical
 data produced.   Copies of  this report  were
 distributed by overnight  courier  to  the  data
 management  staff for entry of the information
 into the NSS-I data base and to the QA staff
 for verification of the reported results.

 Survey Participants

     A number of organizations  were involved
 in various  aspects of the NSS-I.  The National

-------

  bye  EP7 Of'ioA^ ^^   "* ^^ <1987' sum™rt** *• °* data









  Environmental Monitoring Systems Laboratory
  in Las Vegas, Nevada, was responsible for QA
  and QC activities,   sampling  and  logistical
  operations, communications coordination, and
  analytical support. The Las Vegas laboratory
  received assistance in these areas from
  Lockheed  Engineering  and  Management
  Services Company,  Inc.   The  U.S. Soil
  Conservation Service and  other federal and
  state agencies helped to determine land
  ownership and to obtain access to field sites.
  Global  Geochemistry Corporation  (Canoga
  Park,  California)  and the  New York State
  Department  of Health  (Albany,  New York)
  provided analytical laboratory services. Radian
  Corporation  (Austin, Texas), the  support
  laboratory,   provided  performance  audit
  samples. The Oak Ridge National Laboratory
  in Oak Ridge, Tennessee, was responsible for
 developing and managing the  data base for
 the  survey.    Personnel at the Oak  Ridge
 laboratory also participated in data inter-
 pretation  and  provided statistical program-
 ming, mapping,  and  other  geographical
 analyses.  Systems Applications,  Inc. (San
 Rafael,  California),  provided support  in
 analysis of analytical laboratory bias and
 audited the data  base.    The  EPA  Sample
 Management Office in Alexandria, Virginia, was
 responsible for sample tracking and  assess-
 ment of analytical laboratory performance to
 determine financial compensation.

 National Stream Survey
 Documents

      This QA report is one of  a number of
publications that describe the NSS-I.   The QA
plan  for the NSS-I is documented in Drous6 et
al. (1986a).    Messer  et  al.  (1986) describe
findings of the NSS-I pilot survey conducted in

-------
                                    Section 2

                     Conclusions and Recommendations
     This  section presents conclusions  and
recommendations drawn from the description
and evaluation of the NSS-I quality assurance
program found in Sections 3 through 6 of this
report.  For an explanation of the premises on
which the following statements are made, as
well  as additional  detail,  the reader should
refer to the appropriate section.
Conclusions

     The  success  of  the quality assurance
program depends on how  well  the data
generated  by the survey met the data quality
objectives.  Overall, the program was able to
assure that the quality  of the NSS-I data was
known and  acceptable  and  that  the  data
quality issues  were  documented.   The
following general conclusions can be drawn:

»    The  representativeness,  completeness,
     and  comparability of the  data  are
     adequate for project objectives.

«    In a few cases, data interpretation may
     be  limited by considerations  of  data
     quality in  terms  of precision, accuracy,
     and  detectability. Table 31 defines the
     status of overall  results  for each
     analysis in terms of each data quality
     objective  (DQO).    Table  13  lists the
     DQOs.

•     Checks of the  internal consistency of
      results for  each  sample  generally
      indicate excellent agreement,  although
      some unmeasured ions or noncarbonate
     protolytes are apparently present in
      some of the streams sampled.
   The following  subsections list specific
conclusions regarding the three primary DQOs.

Detectability

•   For most variables,  instrumental and
    methodological performance and
    background levels of analyte did not
    produce any serious problems with data
    quality.

•   Method-level limits of detection  for all
    measurements met the DQOs, except for
    phosphorus and silica  measurements
    from one analytical laboratory and specific
    conductance measurements from the
    processing laboratory for the  first half of
    the survey.

•   No DQOs were established for  system-
    level detectability.  However,  comparison
    of system-level  limits of  detection to the
    method-level DQOs showed that  results
    for acid-neutralizing capacity  (ANC),
    specific  conductance, magnesium,
    potassium, sodium,  and sulfate  met or
    nearly met the method-level DQO; results
    for extractable  aluminum, calcium,  fluo-
    ride, and manganese were less than twice
    the method-level DQO;  and results for
    total aluminum, equilibrated  and initial
    dissolved inorganic carbon  (DIC),
    dissolved organic carbon (DOC), and
    nitrate  exceeded twice the method-level
    DQO.

•   Samples for which background levels of
    nonexchangeable monomeric aluminum
    exceed total monomeric aluminum have an
    increased level of uncertainty associated

-------
       with their  values for exchangeable
       monomeric aluminum, especially at low
       concentrations.
  Accuracy

  •    For all variables except base-neutralizing
       capacity (BNC)  at  low  concentrations
       (less  than  30 jueq/L),  specific
       conductance in dilute samples (less than
       25 //S/cm),  and total  dissolved
       phosphorus during the latter half of the
       survey,  accuracy  estimates  for
       measurements at Laboratory 1 were
       within  the  DQOs.   Measurements  of
       several  variables  at Laboratory 2
       showed potential systematic error;  of
       these,  only DOC at low concentrations,
       silica,  and  BNC exhibited a degree  of
       potential error that  might affect data
       interpretation.

 •    The synthetic audit samples provided a
      reliable means to assess  the accuracy
      of most measurements.  However, the
      potential for loss of  aluminum and iron
      as well as the dependency of ANC, BNC,
      DIG, and pH values on dissolved carbon
      dioxide concentration and on the correct
      addition of other analytes to  the formu-
      lation could affect the composition of
      the synthetic audit sample. Therefore, in
      some cases, accuracy estimates based
      on synthetic audit samples may  not
      reflect the true quality  of the data for
      those analytes.

 •    During  the  last  half  of  the  survey,
      phosphorus data from Laboratory 1 may
      be affected by a low-level negative
      calibration bias.

 Precision

 •     Random errors occurring during sample
      preparation and analysis contribute only
      a  small proportion  to the  overall
      measurement error.

•     For all variables except  total aluminum,
     DOC, iron, and  ammonium,  variability
      among batches contributes more to
      overall measurement error than does
      sample collection; therefore, among-batch
      precision  estimates should be  used to
      evaluate measurement uncertainty.

      For  total aluminum,  DOC,  iron, and
      ammonium,  sample-to-sample  variability
      or collection effects are more important
      than  day-to-day or among-laboratory
      variability  in  determining  overall
      measurement error;  therefore,  system-
      level precision estimates should be used
      to evaluate measurement uncertainty.

 Recommendations

 Field and  Processing Laboratory
 Activities                             y

 •   For  the  specific  conductance
     measurement made at the processing
     laboratory, use a water bath to maintain
     sample temperature at 25 *C rather than
     calculating the temperature-corrected
     measurement.

 •   Test  all instrumentation and  protocols
     used for analytical measurements before
     the survey begins  to be certain they will
     perform as anticipated.

 •    Minimize  contamination  of non-acid-
     washed apparatus by shielding it from the
     acid washing of filtration equipment.

 •    Provide longer and more comprehensive
    training programs for all processing
    laboratory personnel  and  include an
    emphasis on  proper completion of  data
    forms.

•   Designate  an assistant to the  base
    coordinator for the purpose  of reviewing
    and correcting all field data forms before
    shipment.

«   Develop efficient filtration procedures for
    samples containing large quantities of
    particulate  material for future studies of
    stream chemistry.

-------
Analytical Laboratory Activities

•     Conduct  on-site evaluations at  the
      analytical laboratories early in the course
      of analyses, as directed by the QA plan.
      Make follow-up evaluations if possible.

«     Make confirmation and reanalysis of
      questionable analytical  values within a   «
      specific time frame  a  contractual
      requirement.

•     Specify clearly the required number of
      decimal places to which analytical
      results must be reported throughout the
      survey.

•     Use  identical software  at  each
      laboratory to calculate ANC and BNC in
      future surveys.
                                           •
Data Evaluation

•     Consider  using  the  closed-system
      measurements of DIG  and pH  in  the
      analysis of the NSS-I data because they
      provide  a  better  estimate of  in situ
      conditions at the time of sampling than
      do open-system measurements.           •

•     Use analytical measurements of specific
      conductance because  measurements
      made  at  the  processing laboratory
      during the first  half of the NSS-I may be   •
      subject to systematic error.

Design of Quality  Assurance Programs
                                           •
•     Perform preliminary statistical evaluation
      of pooled  raw data early enough to
      identify problems and  to allow
      reanalyses within  holding  time
      limitations.

•     Raise the detection  limit objective  for   •
     equilibrated and  initial  DIC
     measurements  if the  detection limit is to
     be estimated from blank samples that
     are exposed to  the atmosphere.

•    Clearly  define the  approach for   •
     determining the detection limits and
 specify the approach in the QA plan.  Low-
 level QCCS or audit samples may be of
 more  use  in assessing laboratory
 performance in terms of detectability than
 laboratory blank samples.  Delineate
 approaches to assessing data  quality in
 the QA plan.

 Whenever more  than  one  laboratory is
 involved in analyses and interlaboratory
 bias is a concern, consider more stringent
 within-laboratory control  limits and  use
 audit samples (or  collected samples)
 representing  a  wide   range  of
 concentrations to monitor,  assess,  and
 possibly correct for any biases that occur.
 This approach will also allow for a more
 rigorous assessment of accuracy within a
 laboratory.

 Allow  the synthetic audit samples to
 equilibrate for a period of time before use;
 subject  both synthetic and  natural  audit
 samples  to rigorous  verification
 measurements against certified standards
 so their composition is known with a high
 level of certainty.

 Consider preparing synthetic samples on
 an analyte-by-analyte basis or as aliquots
 that include chemically  compatible
 variables.

 Select  audit  sample compositions that
 bracket the  expected  concentrations of
 analytes in the stream samples.

 Provide  the  analytical laboratories with
 known performance standards  from a
 single source so  that all laboratories can
 calibrate their measurement systems to a
 given  target  value  to  reduce
 interlaboratory bias.

 Consider using a series of split samples,
 prepared from a  composite bulk  routine
 sample, rather than duplicate samples to
 assess precision and identify components
of error more discretely.

Consider taking  chemically  well-
characterized  natural audit samples into

-------
the field  and processing them through
the sampling device to allow estimates
of the total uncertainty due to sampling
and measurement.

If reliable  BNC data  is  required to
differentiate weak  and  strong acid
concentrations in natural water samples,
modify the  analytical methodology  so
that titration is conducted under an inert
atmosphere free of carbon dioxide.
                                     8

-------
                                     Section 3

              Design of the Quality Assurance Program for the
                       National Stream Survey-Phase I
     An important design criterion of the NSS-
I was that the  data  collected must  be
scientifically sound and of known quality.   To
meet  these  requirements,  standardized
collection of data was  implemented and  a
rigorous QA program was  established.  This
program has two separate but integrated
components that cover operations and data
management.  The operations component
included QA and QC procedures to ensure that
all samples were collected and analyzed
consistently and to estimate the accuracy and
precision of the reported  values with  a known
degree of confidence.  The data  management
component  established a program that stored
and tracked the data;  identified and corrected
entry, reporting, and analytical errors;  and kept
a record  of such changes.  These procedures
produced documented files that contained
data of known quality and that are accessible
to project scientists and extramural users.  The
NSS-I QA plan (Drous6 et  al., 1986a) defines
the activities needed to meet the requirements
of the QA program and to guide the operations
and data management components.  The plan
also presents QA protocols for collecting,
processing, shipping,  and  analyzing  samples
as well as for reporting and verifying analytical
results.

 Stream  Characteristics  and  Data
 Quality Objectives

      One of the first  steps  in the design of
the NSS-I was to identify  the variables to be
 measured  and  to define the analytical data
quality objectives (DQOs) for measuring each
variable.  Twenty-seven chemical and physical
characteristics of stream water were selected
for in-situ or laboratory measurement. Table 1
 lists these  characteristics along with
abbreviations used in this report and analytical
methods.  These variables were selected
because measurements of their concentration
in  stream waters  should  provide  sufficient
information to determine the  chemical  and
physical quality of the streams with respect to
fish habitat and the geochemical nature of the
waters  with respect  to  past and future
susceptibility to acidic deposition.  Some
variables are of primary interest with respect
to these survey objectives (e.g., pH and acid-
neutralizing capacity).    Other variables are
important  in interpreting the primary variable
data (e.g.,  dissolved organic carbon (DOC),
color, and fluoride are useful in understanding
the speciation of aluminum).  Variables such
as nitrate,  sulfate,  and DOC are needed to
describe the ionic composition of waters, and
some   may be useful  indicators of
nonatmospheric pollution (e.g.,  chloride, total
dissolved  phosphorus,  and  ammonium).
Finally, some variables may provide clues to
the geochemical  processes controlling water
chemistry in a region and may also be useful in
classification  of  stream reaches  for further
study (e.g.,  silica,   sodium,  potassium,  and
calcium).  Complete chemical analysis for all
major ions is needed to conduct  verification
checks on the accuracy of chemical analyses
on  the  basis  of cation/anion  balances and
specific conductance checks.   Messer et al.
(1986)  and Hillman  et al.  (1987)  give  brief
descriptions of each variable.

    Table 1 lists the instrument or method used
in the field and in the laboratory to measure
each variable.   Some variables  (dissolved
inorganic  carbon,   pH,  and  specific
conductance) were measured more than once
in each sample, either with  different methods
or at different locations (field and laboratory),
                                          9

-------
 Table 1.  Chemical and Physical Variables Measured During the National Stream Survey - Phase
 Variable (units)
 Abbreviation4
                                                                  Instrument or analytical method*
                                                    FIELD SITE
 pH, field (pH units)
 Specific conductance
   (fiS/cm)

 Dissolved oxygen
   (mg/L)
 Temperature
 pH-field
 cond-in situ
 DO
 Portable pH meter (Beckman pHI-21); glass
   combination electrode (Orion-Ross Model 8104)

 Portable conductivity meter (YSI Model 33 S-C-T)
   with probe (YSI Model 3310)

 Portable dissolved oxygen  meter (YSI Model 54A);
   pressure-compensating oxygen-temperature probe
   (YSI 5739)

 Portable conductivity meter (YSI Model S-C-T)
   with probe (YSI Model 3310)
                                             PROCESSING LABORATORY
 Aluminum (mg/L)
     Total monomeric
     Nonexchangeable
      monomeric

Specific conductance
  (/jS/cm)

pH, closed system
  (pH units)

Dissolved Inorganic carbon,
  closed system (mg/L)

True color (PCU)

Turbidity (NTU)
                                     Al-mono
 Al-nex
Cond-PL
pH-closed
DIC-closed
Colorimetry (complexation with pyrocatechol
  violet, automated flow injection analyzer),
  La Chat  Quick Chem System IV Colorimeter

Same as total monomeric
Conductivity meter (YSI Model 32); probes (YSI
  Model 3417 and Model 3401); NBS thermometer

pH  meter (Orion-Ross Model 611), and glass
  combination electrode (Orion-Ross Model 8104)

Infrared spectrophotometry (carbon analyzer)
  (Dohrmann  Model  DC-80)

Comparator (Hach Model CO-1)

Nephelometer (Monitek Model 21)
                                            ANALYTICAL LABORATORY
Acid-neutralizing capacity
  U/eq/L)

Aluminum, (mg/L)
    Total extractable
    Total
ANC
                                    Al-ext
                                    Al-total
                    Acidimetric titration, modified Gran analysis
                   Atomic absorption spectroscopy (furnace) on
                     methyl isobutyl ketone extract

                   Atomic absorption spectroscopy (furnace)
                                                                                             (Continued)
                                                10

-------
Table 1.  (Continued)
Variable (units)
                                     Abbreviation*
                            Instrument or analytical method
                                       ANALYTICAL LABORATORY (Continued)
Ammonium (mg/L)

Base-neutralizing capacity
  Ojeq/L)

Calcium (mg/L)

Chloride (mg/L)

Specific conductance
  (/jS/cm)

Dissolved inorganic carbon
  (mg/L)
     Open system

     Equilibrated
NH4+

BNC


Ca

CI"

Cond-lab
DIC-open

DIC-eq
Colorimetry (phenate, automated)

Alkalimetric titration, modified Gran analysis


Atomic absorption spectroscopy (flame)

Ion chromatography

Conductivity cell and meter
Infrared spectrophotometry

Infrared spectrophotometry
 Dissolved organic carbon
   (mg/L)

 Fluoride, total dissolved
   (mg/L)

 Iron (mg/L)

 Magnesium (mg/L)

 Manganese (mg/L)

 Nitrate (mg/L)

 pH (pH units)
      Equilibrated
     Initial (acid titration
       for ANC)

     Initial (base titration
       for BNC)

  Phosphorus, total dissolved
    (mg/L)
                                     DOC
 Fe

 Mg

 Mn

 N03-


 pH-eq


 pH-ANC


 pH-BNC
 Infrared spectrophotometry


 Ion-specific electrode


 Atomic absorption spectroscopy (flame)

 Atomic absorption spectroscopy (flame)

 Atomic absorption spectroscopy (flame)

 Ion chromotography
 pH electrode and meter; sample equilibrated with
   300 ppm CO2  in air

 pH electrode and meter
 pH electrode and meter
                     Automated colorimetry (phosphomolybdate or
                       modification)
                                                                                               (Continued)
                                                      11

-------
Table 1. (Continued)
Variable (units)
Potassium (mg/L)
Silica (mg/L)
Sodium (mg/L)
Sulfate (mg/L)
Abbreviation*
K
Si02
Na
SO^
Instrument or analytical method*
Atomic absorption spectroscopy (flame)
Automated colorimetry (molybdate blue)
Atomic absorption spectroscopy (flame)
Ion chromatography
 * These abbreviations for variables will be used throughout this report.
 * Methods and instruments are described in Hagley et al. (in press) for field activities and in Hillman et al.
   (1987) for processing and analytical laboratory analyses.  The analytical laboratories met the instrument
   requirements as defined in the Statement of Work.
 for a total of 35 measurements. The analytical
 laboratories made 24 of these measurements.
 Three operationally defined aluminum fractions
 were measured:   total  monomeric aluminum,
 nonexchangeable  monomeric  aluminum, and
 extractable aluminum.  Nonexchangeable
 monomeric  aluminum was  determined  after
 passing the sample through a cation exchange
 column (Hillman  et al., 1987). Closed-system
 measurements of  pH and DIG were made  on
 samples collected in sealed syringes without
 exposure to  atmospheric carbon dioxide.
 Equilibriated measurements of  pH and DIG
 were conducted after sparging the samples
 with 300  ppm carbon dioxide in  air mixture
 (Hillman et al., 1987).

      Other  stream  characteristics that  were
 measured or estimated at sampling sites
 included  watershed disturbances,  land use,
 bank vegetative cover, stream substrate, and
 stream width, depth, and flow velocity.  The
 site information recorded  at  each sampling
 location was intended to assist in the initial
 interpretation of physical ahd chemical  data
 from each site and to aid in locating the site in
 future studies.  This site information was not
 subjected to the full scope of QA activities and,
 although it  is recorded in the data base,  it
 should  not be used to draw  quantitative
 inferences about the other chemical or physical
data. These data  are not discussed further in
this report.
     Researchers involved in any monitoring or
 measurement study funded by the EPA must
 establish DQOs based on the proposed end
 use of the data.   These objectives are set
 before the  research  begins.   The expected
 range of sample concentrations and the objec-
 tives for detection  limits,  precision,  and
 accuracy were developed for each parameter
 by  using data from the published literature,
 from statistical error simulation, and from the
 results of Phase I of the Eastern and Western
 Lake Surveys. Equipment, sampling protocols,
 and analytical methodologies were selected
 and were standardized in order to  achieve the
 DQOs.  These objectives were also applied to
 the statistical assessment of sampling,  pro-
 cessing  laboratory, and analytical laboratory
 performance.   The objectives set  criteria for
 detectability, accuracy,  precision,  representa-
 tiveness, completeness, and comparability.

    Measures of detectability, accuracy,  and
 precision are estimated by analyzing data from
 QA and QC samples. Detectability is the ability
 of an  instrument or  method to determine  a
 measured value  for  an analyte above
 background levels  with a specified degree of
 confidence.  Accuracy describes the closeness
 of a measured value to the true (or index) value
of the variable concentration in the  sample.
 Precision describes the  closeness of values
derived  by  repeated  measurements  of  the
same quantity under specified conditions.  The
                                           12

-------
values and ranges that were established for
these  three  DQOs  are given in  Section 6,
Table 10.    For  most of the 35 variables
measured, the  analytical results were
evaluated to determine if they met these
analytical  DQOs.   In addition to these six
DQOs,   relative interlaboratory  bias,
operationally defined  as a systematic
difference in analytical performance  between
laboratories, is evaluated in this report.

     The requirements for survey  data to be
representative, complete, and comparable
were addressed  by the  NSS-I  statistical
sampling  design  (Kaufmann et al., in press)
and the  QA plan (Drouse et al.,  I986a).
Completeness is  a measure of data quality
that is the quantity of acceptable data actually
collected relative to the total quantity that was
expected to be  collected.  Comparability
expresses the confidence with which one data
 set  can be compared to another.   Represen-
 tativeness is a measure of the degree to which
 sample data accurately and precisely reflect
 the characteristics  of  a population.
 Representativeness also relates to the degree
 to which QA and QC samples represent routine
 stream samples.

      As the survey progressed, measure-
 ments of the QA and QC samples that were
 made at the stream  sites,  processing
 laboratory,  and  analytical laboratories were
 compared to the DQOs  and  concentration
 ranges.   These comparisons provided a
 mechanism to identify and correct  sampling,
 analytical,  and reporting  errors before  data
 quality was affected.

 Statistical  Design  of the National
 Stream Survey

       To characterize stream  chemistry  and
 associated physiographic attributes accurately
 and confidently,  a statistically based scheme
 was developed  to  ensure  that  the stream
 reaches  sampled would  be representative of
 the target  population (i.e., those streams of
 interest with respect to the primary objectives
 of the Aquatic Effects Research Program and
 the National Acid  Precipitation  Assessment
 Program).  The detailed rationale behind
stream selection and sampling during Phase I
is  described by Blick  et  al.  (I987),  Overton
(1985,1987), and Kaufmann et al. (in press).

Sample Collection and Analyses--
Quality Assurance and Quality
Control

    For the NSS-I,  a routine sample was
collected from the stream in a 3.8-L container
and four syringes.   In addition, QA and QC
samples, described in  the QA plan (Drouse et
al., 1986a) and in Table 2, were employed in the
field,  in the processing laboratory,  and at the
analytical laboratories to  maintain the  quality
of the survey data and  to ensure that data
quality could  be characterized.   Stringent
requirements for  instrument calibration also
helped to provide reliable measurements.

     Figure  3  shows  the relationship  of the
different QA and QC samples to the collection
and  analysis process.   The results from
analyses of QA samples were used to evaluate
the   performance  of sampling  methods,
laboratory  analyses,  and overall  data quality
for the survey.   Analyses of the  QC samples
allowed field samplers and  laboratory
personnel  (in  both the  processing  and
 analytical laboratories) to identify and correct
 specific problems such  as  poor  instrument
 performance or reagent contamination before
 and during routine sample analyses. Although
 it was not a requirement of the NSS-I, each
 laboratory followed its own internal good
 laboratory practices  and  measured  QC
 samples that were independent of the survey
 QA samples.

 Quality Assurance Samples

      Of the 1,654  NSS-I  samples analyzed at
 the analytical laboratories, 273 (16.5 percent)
 were QA samples (Table 3).  These field and
 processing laboratory blank,  field duplicate,
 and   field  and   laboratory audit samples
 (Table 2)  were added to a group of routine
 stream samples either at the stream site or at
 the processing laboratory. They were analyzed
 at the processing  laboratory  (except  for
 laboratory audit samples) and the analytical
 laboratories.   Because  analytical  laboratory
                                           13

-------
 Table 2.   Quality Assurance and Quality Control Samples Used In the National Stream Survey - Phase
 Sample type
       Description
                                                                   Function
                                                            Frequency of use*
 QUALITY ASSURANCE
 Field blank
 Processing laboratory
   blank
 Field duplicate
 Reagent-grade deionized
   water* subjected to
   sample collection,
   processing, and
   analysis

 Reagent-grade deionized
   water* subjected to
   sample processing and
   analysis
 Duplicate sample
  collected immediately
  after the routine
  stream sample
 To assess detectability
   and identify possible
   sample contamination
   resulting from collec-
   tion and processing

 To estimate  background
   effects due to
   sample processing
   and analysis
                                                             To estimate system
                                                               precision
                                                                                          One per batch
In lieu of field
  blank when
  logistical con-
  straints prevented
  its collection

One per  batch
Performance  audit
    Field
Laboratory
QUALITY CONTROL
Calibration blank
Reagent blank
Synthetic or natural
  lake sample; prepared
  at support laboratory
  and processed at
  processing laboratory
Synthetic or natural
  lake sample; prepared
  and processed at
  support laboratory
                              Reagent-grade deionized
                                water*
Reagent-grade deionized
  water* plus reagents
  for total aluminum and
  silica  analyses
To estimate analytical
  precision of processing
  and analytical labora-
  tory measurements; to
  estimate relative
  accuracy and relative
  interlaboratory bias

To estimate analytical
  precision of analytical
  laboratory measurements;
  to estimate relative
  accuracy and relative
  interlaboratory bias
                              To identify signal drift
                                                            To identify contamination
                                                              due to reagents
As scheduled
                                                                                          As scheduled
                             One per batch
                               for applicable
                               variables

                             One per batch for
                               total  aluminum
                               and silica
                                                                                         (continued)
                                                      14

-------
Table 2.  (Continued)
Sample type
                               Description
                                                         Function
                                                   Frequency of use*
Quality Control (continued)

Quality control
  check sample
  (QCCS)
Standard solution from
  source other than
  calibration standard
To determine accuracy
  and consistency of
  instrument calibration;
  to check statistical
  control of measurement
  process
Before the first
  measurement,
  after the last,
  and at specified
  intervals in
  between for each
  batch
 Detection limit
  QCCS
 Processing
   laboratory
   duplicate

 Analytical
   laboratory
   duplicate
Standard solution at 2
  to 3 times the
  required detection
  limit
 Split of stream sample
 Split of sample aliquot
To determine precision
  and accuracy at lower
  end of linear dynamic
  range of measurement
  method; to verify
  instrument detection
  limits

To monitor analytical
  precision of processing
  laboratory measurements

To monitor analytical
  precision of analytical
  laboratory measurements
One per batch
  for applicable
  variables
                                                                             One per batch
                                                   One per batch
 a  Planned frequency for use of QA samples was not always possible due to logistical constraints.
 *  ASTM (1984).
 personnel did not know the origin, identity, or
 chemical composition of the samples, the QA
 samples were analyzed as if they were routine
 stream samples. These samples were used to
 evaluate the overall per- formance of sampling
 and analytical activities and to estimate data
 quality.  Figure 4 gives a  graphic presentation
 of the numbers and kinds of samples collected
 during the NSS-I.

 Blank Samples-

       Field  blank samples were prepared at
 the processing laboratory from deionized
 water that  met American Society for Testing
 and  Materials  specifications  for Type  I
 reagent-grade  water  (ASTM, 1984).  Sampling
                      crews transported  the deionized water to the
                      stream sites and processed the blank sample
                      through sampling equipment as if it  were a
                      routine stream sample.  Because closed-
                      system DIG and  pH  analyses  were  not
                      performed on  field blank samples  in  the
                      processing laboratory,  only two  syringes of
                      stream water  were   collected for  these
                      samples.  These two  syringes were used to
                      prepare an aliquot for analysis of extractable
                      aluminum  and determination  of  total
                      monomeric and nonexchangeable monomeric
                      aluminum.

                            Field blanks were processed along with
                      routine  samples at the processing  laboratory
                      and  were included in the sample batches that
                                               15

-------
                      FIELD
                       SITE
                     FIELD BLANK
                   FIELD DUPLICATE
                      ROUTINE
                      STREAM
                       QCCS
                     (DO, pH-field,
                     cond-in situ)
     BIG MOOSE LAKE
     NATURAL AUDIT
      BAGLEY LAKE
     NATURAL AUDIT
                   PROCESSING
                   LABORATORY
                                                  FIELD BLANK
                                                 PROCESSING
                                               LABORATORY BLANK
                                                (In lieu of field blank)
                                                FIELD DUPLICATE
                    PROCESSING
                    LABORATORY
                     DUPLICATE
                   (Split of a randomly
                    selected routine
                    stream sample)
                                               CALIBRATION BLANK
                                                 (AI-Nex, AI-Mono,
                                                cond-PL, DIC-closed)
                       QCCS
                 (AI-Nex, AI-Mono, cond-PL,
                  DIC-closed, pH-closed,
                      turbidity)
                                                  FIELD AUDITS
                                                   PROCESSED
                                               LABORATORY AUDITS
                                                 RELABELED ONLY
   AUDIT SAMPLE
     SUPPORT
   LABORATORY
PREPARED NATURAL AND
  SYNTHETIC AUDITS
ANALYTICAL
LABORATORY
                                                                               FIELD BLANK
                                                 PROCESSING
                                                 LABORATORY
                                                   BLANK
                                                                             FIELD DUPLICATE
  ANALYTICAL
  LABORATORY
  DUPLICATE
 (Split of a sample
  from the batch)
                                               CALIBRATION AND
                                               REAGENT BLANKS
     QCCS
                                                  FIELD AND
                                               LABORATORY AUDITS
Figure 3.     The relationship of the quality assurance and quality control samples to the collection and
            analysis process.
were sent to  the analytical laboratories.
Analytical data  for these QA samples  in each
batch  were  used  to identify  possible
contamination problems during sampling and
analyses.

   Occasionally, due  to logistical constraints,
field blanks were not available for processing
at a  stream site on a particular sampling day.
In  such  instances  (on  five  occasions),
processing laboratory personnel  substituted a
deionized  water sample for the missing field
blank.   Although  this  processing  laboratory
blank sample was not  processed through the
sampling equipment,  it took the place of the
missing field blank sample in the sample batch
sent to the analytical laboratory.  These  pro-
cessing laboratory blanks  were used  only to
                   detect  contamination  and  were not used in
                   statistical QA analyses because they did not
                   go through  the  entire  sampling and analysis
                   system  from the  field through  the analytical
                   laboratory.

                   Field Duplicate Samples-

                        A field  duplicate  is  a  second set  of
                   stream water samples  collected  immediately
                   after the routine sample.  The sampling  crew
                   used the same procedure to collect  both the
                   routine and duplicate samples.  Pairs of field
                   routine and  duplicate samples were used to
                   assess  the   precision  of the field  sampling
                   techniques and the processing and analytical
                   laboratory procedures.
                                              16

-------
 1400-


 1200-


 1000-


 800-

 600-


 400-
O  80-1
o;
UJ
|  70-


Z  60-


   50-


   40-


   30-


   20-


   10-
             ^^^^^s^^
             - « , v^;^
             .»•** -
-------
Table 3.  Number of Routine and Quality Assurance
        Samples Collected amd Analyzed During
        the National Stream Survey - Phase I

                                     Percent
                                     of total
                        Number of     samples
Sample type                samples     collected

Quality assurance samples
  Field blank                  63
  Processing  laboratory blank      5
  Field duplicate               66*
  Laboratory  synthetic audit      42
  Field synthetic audit           14
  Laboratory  natural audit        24
  Field natural audit            54
  Special studies            	5

  Total quality assurance
    samples                  273         16.5

Routine samples
  Mid-Atlantic               1,017
  Screening                  343
  Episodes                 	21
Total routine samples
Total samples collected
1381
1,654
83.5
100.0
 a Includes one sample that was not processed
  correctly and cannot be used to estimate data
  quality.
 stored at 4 *C to minimize changes in chemical
 composition.

       Field synthetic  audit  samples,  which
 were prepared  at  the  support laboratory to
 simulate natural water,  included a matrix of
 analytes at specified  theoretical concen-
 trations.   The  synthetic  sample represented
 surface water  with low concentrations of
 analytes.  Because the first lot of synthetic
 material was exhausted before the end of the
 survey,  the support  laboratory  prepared  a
 second lot with the same theoretical concen-
 tration.  Field synthetic audit  samples were
prepared as concentrates  and  diluted just
before they were  sent  in 2-L bottles  to  the
processing laboratory.   The chemical
composition and preparation of  the synthetic
audit samples is described in Appendix A.

     Data  obtained  from  analyses  of
laboratory audit samples identified problems
encountered during the analytical process that
may affect data quality.  In addition to their
use in determining relative interlaboratory bias
and  the precision of measurements  of  the
same sample type, laboratory audit samples
helped  to  verify  the accuracy  of analytical
procedures.  Natural and synthetic laboratory
audit  samples came from the same sources
as did the field audit samples.   The support
laboratory supplied audit samples already split
into seven  aliquots  to the  processing
laboratory.  The laboratory audit samples were
labeled  at  the  processing laboratory  in  the
same manner as routine samples and were
indistinguishable from any field  sample.
However, they were not  processed or analyzed
at the  processing  laboratory.    They were
included in  a  batch with  routine  stream
samples that were processed and shipped on
the same day to an analytical laboratory.

Quality Control Samples

     The QC samples (Table 2)  were used in
the field and at the processing and analytical
laboratories.  In general, QC samples are used
to ensure proper instrument performance and
sample  analysis.  QC samples are defined as
control  samples for which the analyst knows
the  true  analyte concentration  or  value.
Analytical  data for these samples must  fall
within control limits specified in  the QA plan
(Drouse et al., 1986a).

Field Quality Control Samples-

    Quality  control check samples  (QCCSs)
were  used by the field crews to check  the
calibration  of the  pH,  conductivity, and
dissolved oxygen meters before sampling and
to check for instrument drift during and after
field measurements.  Daily QC  checks were
made   before  and  after  sampling.   If  the
                                            18

-------
measurement of the QCCS did not fall within
the control limits,  the meter was recalibrated
or checked for proper operation.

Processing  Laboratory  Quality
Control Samples--

      Processing  laboratory personnel
analyzed calibration blank samples,  QCCSs,
and processing  laboratory duplicate  QC
samples.   A calibration blank was analyzed
before any samples in the batch to check for
baseline drift and for  contamination  of  the
carbon  analyzer, flow injection  analyzer, and
conductivity meters.  Calibration and drift of
the carbon  analyzer and of the instruments
used to  measure pH,  turbidity,  specific
conductance, and the aluminum species were
also  checked with  QCCSs  at  specified
intervals.  The QA plan (Drouse et  al., 1986a)
required observed concentrations to be within
the specified control limits.   When an unac-
ceptable value was obtained,  the  instrument
was  recalibrated  and all samples  that were
analyzed after the last  acceptable QC sample
were reanalyzed. Each day one routine stream
sample was selected  randomly as  the pro-
cessing laboratory duplicate; this sample was
split and analyzed in duplicate for pH, DIG, true
color,  turbidity,   specific conductance,  total
monomeric  aluminum,  and  nonexchangeable
monomeric aluminum.  Immediately after
analyses, precision estimates were calculated
from these  analyses  and compared to the
DQOs for precision. If the  calculated values
did not meet the DQOs, then another duplicate
sample was analyzed.   If the calculated  pre-
cision estimates from this analysis  still did not
meet the  DQOs, the data were qualified with a
tag (Appendix B).

 Analytical Laboratory Quality Control
 Samples-

       The  analytical laboratories   used  five
 types  of QC samples-calibration blanks,
 reagent blanks,  detection limit QCCSs,  low-
 and high-concentration QCCSs, and laboratory
 duplicates.  For each analytical procedure, the
 calibration blank was analyzed  after the initial
 instrument  calibration to check for drift in the
 measurement signal.    For  silica  and total
 aluminum measurements,  the laboratory was
 required to analyze a reagent blank.   The
 reagent blank,  containing al! the  reagents in
 the same volumes that were used to prepare a
 real sample for analysis,  was prepared in the
 same  manner  as a  routine sample.    The
 observed analyte concentration for calibration
' and reagent blanks could not exceed twice the
 required detection limit (Section 6, Table 13) for
 each analyte.  If the concentration exceeded
 this limit, the source of the contamination had
 to be  investigated and  eliminated.    If the
 source of contamination could not  be identified
 before reanalysis, the data were qualified.

     The QCCSs were either commercially
 prepared or laboratory-prepared samples that
 were  made from stock solutions  independent
 from  those  used  to prepare  calibration
 standards.   The  analyst  was  required to
 choose a QCCS for a particular variable such
 that its  theoretical concentration fell in the
 mid-calibration range  for that variable.  The
 QCCS was analyzed to verify  instrument
 calibration  at  the   beginning  of  sample
 analysis, at specified  intervals during sample
 analyses, and  after the final sample in the
 batch  was  analyzed.   The observed
 concentrations had to be within the specified
 control limits (Drous<§ et  al., 1986a).  When an
 unacceptable  value for  the   QCCS  was
 obtained, the instrument  was recalibrated and
 all samples that were analyzed after the last
 acceptable QCCS were reanalyzed.    In
 addition, the analytical laboratories were
 required  to demonstrate statistical control  by
 plotting  the observed concentrations of the
 QCCS on a QC chart.  To ensure  continuity of
 QC charts,  QCCSs of  the same  theoretical
 concentration were used  throughout the
 plotting  process.   Both 99 percent and  95
 percent confidence intervals were developed
 and  used as control and warning limits,
 respectively.   If the  99  percent  control limit
 differed  from  the  theoretical value by  more
 than  the limits given in the QA plan (Drous6 et
 al., 1986a),  the  laboratories were required to
 consult the QA  staff  in Las Vegas regarding
 corrective action (i.e.,  sample reanalysis). On
 a weekly basis,  QC charts were  updated,
 cumulative means were  calculated, and new
 warning and control limits (95 percent and 99
                                           19

-------
 percent,  respectively)  were determined.   In
 addition to QC charts developed  with survey
 data, each laboratory prepared QC charts for
 the internal QC samples.

      A detection limit QCCS is a low-level QC
 sample that contains the analyte of interest at
 a  concentration of two  to three times the
 required detection limit. A QCCS was analyzed
 once per batch before routine stream samples
 were analyzed for specified variables.  These
 QC samples were used to verify the low end of
 the calibration curve  and  the values for the
 low-concentration samples  near the detection
 limits.  The concentration of the detection  limit
 QCCS had to be between two and three times
 the required detection limit  and the measured
 value had  to be  within  20 percent  of  the
 theoretical value.  If  it was not,  the analyst
 was required to  identify and  correct  the
 problem before sample analysis.

      A duplicate analysis  (laboratory
 duplicate) for each   specified variable  was
 performed on one sample  in each batch to
 estimate and monitor analytical precision. If
 the observed  precision did not meet the DQOs
 established for these  variables,  then another
 duplicate sample had  to be analyzed  (Drousd
 et al.,  1986a).   Data  for  which the precision
 estimate did not  meet the DQOs were qualified
 with a flag (Appendix B).

 Data        Management

      The NSWS  data base  management
 system  incorporates the results from data col-
 lection, evaluation, verification, validation,  and
 enhancement  activities.  This system
 assembles, stores, and edits data generated
 during the  NSS-I  and other NSWS surveys.
 The system also provides basic reports of the
 survey results,  performs  certain statistical
 analyses,   and  provides  data  security.   A
 detailed description of the system is given in
 Sate (in press).

     An important too! in the development of
 the NSS-I data  base was the  use of data
 qualifiers to mark an  individual value or even
 an entire stream  as having a particular feature
that may be useful in data interpretation.  Two
types of data qualifiers,  tags and flags, are
used in the NSS-I data base (Appendix B).  A
tag is a code that was added to a value at the
time of sample collection or analysis to qualify
the  value.    A flag is  a qualifier  that was
assigned during the verification and validation
procedures to  data that did not  meet the
established acceptance criteria or that were  in
some way unusual.   These  qualifiers alert
future data  users to values identified as
questionable or unusual by the verification and
validation  process.  These qualifiers also
provide a method for identifying and removing
clearly  erroneous  data and  retaining
questionable  data  with appropriate tags and
flags.

     The NSS-I data base  was  subjected to
four levels of QA evaluation to  ensure that the
data collected  during  the survey  are
representative of  the  physical and chemical
characteristics of the samples  taken from the
streams.   Each  level of quality  assurance
produced a new  and  more refined working
data set. These working data sets are defined
as:  raw (Data Set 1), verified (Data Set 2),
validated (Data Set 3), and enhanced (Data Set
4).   Data  Set 4  Is the final  product of the
refinement  process.    Alt data  sets  are
protected from unauthorized or accidental
access by individual,  system,  and  file
password protection.    The development of
these working data sets is  summarized  in
Figure 5. The data sets are further described
in the following subsections.

Raw Data Set (Data Set 1)

     The data from ai! components  of  the
sampling and analysis process make up the
raw  data set.  The raw data set  includes all
analytical results and data  qualifiers.
Appendix B  lists the data  qualifiers.  The data
forms used for reporting the raw data can be
found in the QA plan (Drousd et al., 1986a). All
field and processing laboratory forms on which
data were recorded received a preliminary QA
review at EMSL-LV before the  data were
reviewed and  entered into the raw data set at
ORNL.   Data from the analytical  laboratory
forms were entered into the  raw  data set
before the QA review, which took place during
                                          20

-------
data verification.   To  ensure  accurate  data
transfer from field and  laboratory reports, the
information  was entered into  two  computer
files and  subjected  to automated checking
procedures  to  minimize  transcription errors.
The raw data set was used to screen the data
for problems,  perform exploratory  data
analyses, and evaluate the need for any
adjustments in the data analysis plans.

Verified Data Set (Data Set 2)

      The objectives of the data  verification
process were to identify, correct, and flag raw
data of questionable or  unacceptable quality
and to identify data that might need to be
corrected during  or after  data validation.
These objectives were met by reviewing the QA
and QC data measured  and recorded  at the
sampling  site,  at  the  processing  laboratory,
and at the analytical laboratories and by
examining all sample data in terms of chemical
charge balance.   Verification  determines the
quality of  the analytical  data  through a
rigorous protocol based on known principles of
chemistry.    It  scrutinizes the internal
consistency of chemical  concentrations as a
result of  cation/anion  balances, conductance
balance, or protolyte analysis for each sample.

       Computer programs automated much of
the verification process and generated reports
for evaluating intra- and interlaboratory bias as
 well as discrepancies in blank,  audit, and other
 QA  and  QC samples. The Automated Quality
 Assurance Review, Interactive Users System
 (AQUARIUS), which was used  to process data
 during the NSS-I Pilot, was modified for use
 during the remaining  stream  surveys.    The
 Aquatics Analysis System (AQUARIUS II)
 generates data  changes in the form of
 transaction records.  The records  are derived
 from  exception-generating  programs that
 identify or flag analytical results that do not
 meet the  expected QA or QC criteria.

       The final product of the verification
 process is the verified data set in  which each
 sample batch and each sample value has been
 reviewed individually and all questionable
 values are either corrected or identified with an
 appropriate  flag. Data verification  takes place
in two parts:  a preliminary evaluation which
incorporates the majority of  numeric changes
and a final evaluation which  includes any final
numeric changes and  the addition of data
qualifier flags. The verified data set was used
as the basis for data validation.

Validated Data Set (Data Set 3)

     While verification  procedures evaluated
data at the sample  and batch level, validation
procedures examine the plausibility of  sample
data  in the context of  a subregional  set of
samples.   NSS-I  subregional boundaries
generally group streams  of similar
geochemistry together.  The validation process
identified  unusual data that would need
special attention when used in statistical
analysis, particularly in regional estimates
concerning the target population of streams.
Observations  identified as atypical during
review of data  at subregional levels are
considered outliers  from the rest of the data.
Two components of the validation process are
the  identification of statistical outliers from
subregional distributions of chemistry  and the
evaluation of possible systematic errors in the
measurement process.    Such  outliers  may
result from the natural variability of streams in
the set of stream reaches, from anthropogenic
disturbances in the natural  environment, and
from errors in the sampling design, as well as
from sampling and  analytical errors.
Conditions that may cause outliers include:

1.   Sample collection during an eposidic event
     for a given reach.

2.   Factors other  than normal geochemical
     processes (e.g.,  pollution or watershed
     disturbance, including acid mine drainage,
     brine, or other nonpoint sources).

3.   Unusual geochemical properties within  a
     given subregion.

4.   Impossible datum, clearly erroneous when
     reviewing chemistry for that reach.

     Although outliers  may  represent  unusual
data in comparison  with  other data, such
                                           21

-------
                 FIELD SAMPLING AND
               PROCESSING LABORATORY
Figure 5. Data base management system.

                  22

-------
values  are not necessarily inaccurate in their
representation of  a stream reach.   The
validation process is, therefore, not meant to
be a stringent pass or fail test, but rather a
way to search  for observations  that  may
represent entry or analytical errors or unusual
water chemistry. These unusual observations
become apparent when the data are viewed as
a set of information using univariate, bivariate,
and multivariate  analyses.   All outliers
identified during the validation procedures
were investigated further to confirm that they
were entered into the data base correctly.  Any
values determined to be erroneous were
corrected  in  the validated data set.   Values
identified as  unusual as a result of validation
analyses were flagged  in the validated data
set. This data set  retains the values for field
blank, field duplicate,  and performance audit
samples.    A detailed  description of the
validation  process  is  given  in the  QA plan
 (Drouse et  al., 1986a);  validation is also
described by Kaufmann et al. (in press).

 Enhanced  Data Set (Data Set 4)

       Calculations of population estimates are
 difficult if  values are missing from the data
 set.   To  avoid such problems,  an enhanced
 data  set was  prepared  by  substituting
 erroneous or missing values according to
 specified criteria (Kaufmann et al., in press).
 Negative  concentrations reported by the
 analytical laboratories were set equal to zero
 (except for ANC and BNC). An index value for
 each  chemical variable for each sampling site
 was  calculated by averaging the values of
 routine-duplicate pairs and  the values  from
 multiple observations for a sample site. The
 enhanced data set contains a single value for
 each  variable for each sampling location (i.e.,
 one  observation  for each upstream and
 downstream location) and therefore does not
 include  values for  QA samples or  data
 qualifiers.

  Differences  Between  the NSS-I
  and the NSS-I Pilot Survey

        A number  of  changes were made in
  methods and procedures for the  mid-Atlantic
  and southeast screening surveys as a result of
the  pilot survey.    These changes  are
summarized in Table 4 and are described in the
following subsections.

Processing Laboratory Sample Holding
Times

     Sample holding times for water samples
(the period after sample collection and before
aliquot preparation  and sample analyses  at
the processing laboratory) were  increased
from 12 hours in the pilot survey to 30 hours in
the  mid-Atlantic and screening  surveys.  The
decision to increase sample holding times was
based on the results of two experiments:  (1) a
laboratory study testing whether or not carbon
dioxide  can permeate syringes  over  time
 (Burke and Hillman, 1987) and (2) a field study
of bulk samples held in Cubitainers (Stapanian
et  al.,  1987).   The  syringe experiments
determined that holding times for DIG and pH
 held in syringes at 4 °C could be increased to
 30 hours without a measurable effect on these
 variables.   These experiments  did not deter-
 mine the effects  of holding time for aluminum
 speciation.   However,  because pH changes
 that result  from  changes in dissolved carbon
 dioxide  appear  to  be the most significant
 cause of changes  in aluminum speciation, it
 was assumed that syringe aliquots can also
 be  held  for at least 30 hours before aluminum
 extraction.    Bulk  sample experiments  also
 demonstrated that  increasing holding times to
 as  much as 30 hours would have no important
 effect on  analyte concentration.   These
 conclusions may only be applicable to low-
 ionic-strength  natural streamwater samples
 such  as those  of  the NSS-I and  may not be
 universally  applicable  to other sample types
 (e.g., ground  water, polluted  waters, or
 industrial wastes).

 Processing Laboratory Location

      In the pilot survey,  mobile processing
 laboratories were located in the  field in order
 to  meet the 12-hour holding time  requirements
 for aliquot preparation, preservation,  and
 preliminary  analyses.    As a result of the
 holding time  experiments  conducted  for
  syringes and bulk samples  during the  pilot
  survey,  sample  processing was centralized in
                                           23

-------
  Table 4.  Differences Between the National Stream Survey - Phase I and the NSS-I Pilot Survey
       Technique
     Pilot
                                                                Phase I
  Sample holding time



  Processing laboratory location

  Field pH
  Methods of fractionation
   and determination of
   aluminum species
  Matrix spike quality
   assurance samples


  Phosphorus measurement

  Specific conductance in
  processing laboratory
12 hours


Decentralized
Closed-system and
  open-system


8-hydroxyquinoline
  method
Used
Total phosphorus
  (unfiltered)


Not  measured
30 hours


Centralized



Open-system
8-hydroxyquinoline
  method and colorimetric
  method with pyrocatechol
  violet


Not used
Dissolved phosphorus
  (filtered)

Measured
Las Vegas, resulting in better quality control as
well as reduced costs.

Field pH Measurement

      During the  pilot  survey,  comparisons
were  made  between  two techniques for field
pH measurements (Messer et  al., 1986).   The
pH of samples  collected  in  a  syringe  was
measured in a closed system (in a custom-
made sample chamber without exposure to the
atmosphere) and the  pH of samples collected
in  an open container (beaker) was measured in
an open system.  Both methods are described
in  the analytical methods manual (Hiilman et
al., 1987).  When the data resulting from these
two  measurements were compared,   no
significant difference  (p =  0.05) was found
             between the open-system measurement and
             the  closed-system measurement.  Thus, the
             logistically simple open-system measurement
             was chosen to determine field pH during the
             remainder of the NSS-I.

             Fractionation  and Determination of
             Aluminum Species

                 An  experimental semiautomated  colori-
             metric method for fractionation  and determina-
             tion of aluminum species by complexation with
             pyrocatechol violet (Hiilman et  al., 1987) was
             used  during  the NSS-I to measure  total
             monomeric and nonexchangeable monomeric
             aluminum.  This method was expected to be
             less expensive, less time consuming, and more
             reproducible  than  the 8-hydroxyquinoline
                                           24

-------
method  used  to  measure  total  extractable
aluminum  during  the  pilot  survey.
Measurement of total monomerie aluminum
using the  pyrocatechol violet  method is
expected to yield data similar to data obtained
by measurement of total extractable aluminum.
The automated method  should reduce
variability due to different  analysts  and
eliminate problems related  to reproducibility
and precise  timing inherent  in the manual 8-
hydroxyquinoline method.  However,  because
application  of  this method  on a large scale
was  in the developmental  stages, total
extractable aluminum  measurements using the
8-hydroxyquinoline method were continued
throughout the NSS-I to permit comparison of
the two methods.

Matrix Spike Samples

      The purpose of matrix spike samples  is
to  establish a matrix that  is similar to the
matrix of the samples collected and that can
be used to verify the accuracy of an analysis.
The analyst adds a  known  quantity of an
analyte to  a sample of known concentration
and then analyzes the spiked sample.  The
percentage  of spiked  analyte recovered
(percent recovery) determines whether or not
there was  a matrix effect on the analysis  of
the original  sample.   During the pilot survey,
the limits for spike recovery were met for every
batch  and no matrix interferences were
observed (Drouse, 1987).  The  matrix spike
 samples were not included in the 1986 NSS-I
 surveys because  they  did not  provide any
 additional information about the quality of the
 data.    Elimination  of  these samples  also
 reduced costs.

 Phosphorus Measurements

       According to recent studies (e.g., Young
 et al.,   I985),  particulate-bound  phosphorus
 tends to have  a wide range of bioavailability,
 depending on  its source.   Measurement  of
 total dissolved  (filtered)  phosphorus was
 selected for  the NSS-I  rather  than the
 measurement of  total (unfiltered) phosphorus
 made in the pilot and previous NSWS surveys
 because  measurement  of  total  dissolved
phosphorus provides a better estimate of
biologically active phosphorus.

Specific Conductance Measurements

     During the pilot survey and the NSS-I,
specific conductance was  measured in the
field and by the analytical  laboratories.   An
additional conductance measurement was
made in the processing laboratory during the
NSS-I to provide another comparison for the
data user.   Comparison of these  measure-
ments  was also used in data verification and
validation.
                                          25

-------
THIS PAGE INTENTIONALLY LEFT BLANK
               26

-------
                                     Section 4
                Operations of the Quality Assurance Program
     Quality assurance was  an integral
component of all aspects of the NSS-I includ-
ing  (1)  selecting laboratories to analyze the
samples; (2) providing QA-related information
for  training  field and processing  laboratory
personnel; (3)  collecting, processing, and
shipping  the samples;  (4) analyzing  the
samples; (5) managing the data base; and (6)
monitoring sample collection and analyses.

Selection of Analytical
Laboratories

     The Contract Laboratory Program  (CLP),
established  to  support  the  EPA  hazardous
waste monitoring activities,  provided the
mechanism  for choosing  the  analytical
laboratories.  Under the CLP, an invitation for
bid  (IFB)  is advertised.  The IFB  includes  a
statement  of  work  (SOW) that  defines
analytical and QA and QC requirements in  a
contractual format.   Each laboratory submitt-
ing  a bid in response to the IFB is appraised
on the basis of the analysis of performance
evaluation samples and an on-site evaluation.
The laboratory analyses had  to be conducted
according to handling,  analytical,  and QA
protocols detailed in the SOW and published  in
the methods manual (Hillman et al., 1986) and
in the QA plan (Drousi et al., 1986a).

     The NSS-I analyses were performed
under three separate SOWs.  Laboratory 1 and
Laboratory 2 previously had been awarded
SOWs to analyze  samples  for the Eastern
Lake Survey, Phase I (ELS-I), and the  NSS-I
Pilot, respectively.  Analyses  during the ELS-I
and the NSS-I Pilot did not exhaust the bid lots
(600 samples analyzed for each bid  lot) that
had been awarded to the laboratories, and it
was decided to use up these bid lots during
the NSS-I activities.  The SOWs for the two
surveys were basically identical: each required
the laboratories to  analyze up to 30 samples
per day. The SOWs for the ELS-I and the NSS-
I Pilot were modified for use in the NSS-I by
eliminating analyses of matrix spike samples
(see Section 3).

     Because the remaining samples in the bid
lots for  those two  laboratories were  not
sufficient to complete the NSS-I  survey,  a
revised SOW was advertised before the survey
began.   Laboratory  2  passed the selection
process and was awarded three additional bid
lots to complete  analyses of  the NSS-I
samples.  This SOW became effective when
the survey was two-thirds  complete (with
batch  2147).   During the  NSS  activities,
Laboratory 2 analyzed about 66 percent of the
total number of samples  and Laboratory 1
analyzed the  remaining 34 percent.  The
revised SOW differed from the previous SOWs
in the following ways:

1.  The laboratory  was required to analyze as
    many as 60 samples per day rather than
    the  previous  maximum  of  30,  thus
    eliminating  the  need  to  use two
    laboratories In  the  latter part of  the
    survey.

2.  Analyses of detection level QCCSs were
    required for chloride, sulfate, nitrate,
    ammonium,  and silica in addition to the
    variables listed in the original SOWs as
    described in the ELS-I QA plan (Drous6 et
    al., 1986b). This requirement provided an
    additional  check on the low end of the
    linear dynamic  range for these analytes.
                                         27

-------
3.    The method for determining specific
      conductance was modified to include a
      step  to  equilibrate samples  at 25  "C
      before analysis.   This additional  step
      uses a  constant-temperature water
      bath.    The  initial SOW  allowed  the
      laboratory to correct the  sample
      measurement to 25  °C after analysis.
      The  modification  minimized  errors
      associated with the calculation.

4.    The number of decimal places recom-
      mended for reporting each measurement
      was increased  by one place  for most
      variables.   This change ensured  con-
      sistent reporting of even low-level con-
      centrations and it minimized rounding
      errors.

Training of Field, Processing
Laboratory, and Quality Assurance
Personnel

      Training provided to the NSS-I field and
processing laboratory personnel and QA staff
members ensured consistency in sample col-
lection, processing, and analysis and for QA
and data  verification.

      Field  personnel  were introduced to  the
NSS-I project design, given safety training, and
issued equipment at  the Environmental
Monitoring Systems Laboratory in Las Vegas,
Nevada (EMSL-LV)  (Hagley et al.,  in press).
Training  that  covered  NSS-I logistics and
operations, instrumentation,  stream sample
collection and  measurement  techniques,  QA
and QC procedures, and proper data recording
continued at the  Oak  Ridge  National
Laboratory (ORNL).  Training was completed at
the Nantahala  Outdoor Center in Bryson City,
North Carolina, where map reading, outdoor
skills, and safety were emphasized and where
a dry run to  practice sample collection and
stream  measurement techniques was
conducted.

      Laboratory supervisors gave  individual
training to processing laboratory personnel for
as long  as 10 days  in  Las  Vegas, Nevada
(Arent et  al.,  in preparation).   This training
covered  all technical  aspects  of  laboratory
operations,  including  QA and safety
procedures.

    QA auditors received a week-long training
session in Las  Vegas,  Nevada.   Training
covered all  aspects  of QA and QC  as
described in the QA plan (Drouse et al. 1986a).
Auditors worked under close supervision of the
QA  supervisory staff  throughout the
verification process.

Field Sampling Operations

    There were two  separate  sample
collection operations for both the  mid-Atlantic
and the southeast screening regions.  For each
mid-Atlantic operation,  five teams composed
of two samplers each collected samples and
associated field  data.   For each  screening
operation there were  two teams of samplers.
Each  group of teams was supervised  by a
base  coordinator who was assisted  by  a
logistics coordinator.   The  field operations
report  (Hagley et al.,  in press)  gives an  in-
depth description  of logistics and procedures
of sampling.

     Each  group  of teams for an  assigned
sampling  area operated from  base sites
selected on the  basis of their  proximity  to
sampling sites and the  availability of required
shipping and support services. Each base site
occupied as many as 8 to 15 locations in a
sampling area.   The  teams obtained access
information for each stream reach before field
operations began. Stream sites were reached
by vehicle or by foot.

     Each  team sampled one or two reaches
(at upstream and downstream sites) per day
between mid-March and mid-May of 1986. The
samplers  calibrated the pH and dissolved
oxygen field meters each morning at the base
site.    The pH,   dissolved oxygen,  and
conductivity meters were checked with QCCSs
before leaving the base site.   Samplers also
calibrated the dissolved oxygen meter at each
stream site  and  checked the  pH and
conductivity meters with QCCSs  before and
after measurements were made.   Activities of
the field teams and measurement techniques
are described in  the field operations report
                                         28

-------
(Hagley et al., in press).  Figure 6 shows the
flow of field activities.  At each sampling site,
the samplers recorded watershed distur-
bances and  substrate characteristics  on
standardized forms.  They also made in-situ
measurements  of specific  conductance,
temperature,  and dissolved  oxygen and
determined stream pH  at  streamside  on  an
aliquot (beaker) of water collected  by using
Tygon tubing and  a portable  peristaltic pump.
Hydrological data  collected  at downstream
sites included  stream  width,  depth, velocity,
and discharge.

     The  sampling  team  collected  a
streamwater  sample  (routine sample) from
each stream by pumping water through 1/4-
inch Tygon tubing.  The water samples were
pumped  from  the  midchannel of  the  stream
into a 3.8-L polyethylene Cubitainer by a
portable, battery-driven peristaltic pump. The
samplers also filled  four  gastight 60-mL
syringes for the  analyses performed in  the
processing laboratory (i.e., pH, DIG, and total
monomeric and nonexchangeable  aluminum)
and for  the  preparation of  the extractable
aluminum aliquot.  More detailed discussions
of  these techniques are available in the field
operations reports (Knapp et  al., 1987; Hagley
et al., in press).

      Two  types of QA  samples were
collected. Each day, one team at each of two
base sites collected a field blank sample at the
first site visited.  The reagent-grade water for
this field blank sample was carried from the
processing laboratory to the  sample site and
pumped through  all sampling equipment and
into clean sample containers.    In addition,
using identical techniques, one team at each of
the two  remaining base sites collected a field
duplicate sample, a  second  set  of  sample
containers (Cubitainer and syringes) filled with
 stream water from the pump immediately after
the routine sample was collected.

      All sample containers  were transported
to the team vehicle in portable soft  coolers
that contained  frozen-gel  packs.   Team
 members transferred  the  samples and
 associated data  forms to  insulated shipping
containers  with frozen-gel packs.    The
temperatures of the coolers were checked by
inserting a thermometer whenever the samples
were  transferred from one  container to
another.  The containers were shipped on the
same day by overnight courier to ensure  their
arrival  at  the  processing  laboratory in  Las
Vegas, Nevada, on the morning after collection.
Because it was  necessary to meet overnight
courier deadlines, only the stream data  form
(Form  4)  was  enclosed with the samples.
Other field forms were sent to the EMSL-LV QA
staff as soon as the forms were reviewed by
the base coordinators.  If the review identified
any   changes   necessary on   Stream  Data
Form 4, the coordinator notified the  EMSL-LV
QA staff  by telephone and provided paper
documentation with the next form shipment to
Las  Vegas.  Titles of  all field forms can be
found in Figure 7 and the forms are reproduced
in the QAplan (Drouse et al., 1986a).
 Processing Laboratory Operations

     The processing laboratory provided  a
 controlled environment in which to process and
 preserve water  samples and  to  measure
 variables  that tend to become unstable over
 time.    Processing  laboratory personnel
 included a laboratory coordinator,  laboratory
 supervisor, and as many as 20 analysts. The
 processing laboratory personnel:

 1.   randomly selected and organized the
     stream,  blank,  and audit  samples into
     batches;

 2.   divided the samples into aliquots;

 3.   prepared  the sample aliquots for sub-
     sequent analytical laboratory analysis;

 4.   prepared  and shipped the sample batches
     to the analytical laboratories;

 5.   measured seven variables (pH,  total
     monomeric aluminum,  nonexchangeable
     monomeric aluminum,  specific conduc-
     tance, dissolved inorganic carbon,
     turbidity, and true color);
                                          29

-------
                                  ARRIVE AT

                                 STREAM SITE
     PHOTOGRAPH SAMPLE SITE

     AND RECORD WATERSHED

       CHARACTERISTICS
                                  r
         MAKE IN SITU

        MEASUREMENTS
           RECORD

       HYDROLOGICAL DATA
FIELD BLANK SAMPLE

 (Deionized Water)
                                       TWO 60-mL

                                       SYRINGES
                                     3.8-L CONTAINER
         COMPLETE FIELD

          DATA FORMS
                       COLLECT
                        WATER
                       SAMPLES'
ROUTINE

SAMPLE
                       FOUR 60-mL

                       SYRINGES
                                                        3.8-L CONTAINER
                                                        STORE AT 4 °C
                    TRAVEL TO NEXT
                    SAMPLE SITE OR

                    RETURN TO  BASE
FIELD DUPLICATE
   SAMPLE
                 FOUR 60-mL

                  SYRINGES
                                                                           3.8-L CONTAINER
    * ONLY SPECIFIED SAMPLING TEAMS COLLECTED FIELD BLANK
      AND FIELD DUPLtCATE SAMPLES TO ENSURE THAT ONE OF EACH
      WOULD  BE AVAILABLE FOR EACH SAMPLE BATCH
                Figure 6. Field sampling activities for the National Stream Survey - Phase 1.
6.     checked data forms before transfer to
      the EMSL-LV QA staff; and

7.     prepared  and  shipped  reagents  and
      supplies to the field base  sites.

      Arent et  al.  (in  preparation)  give a
detailed discussion of  processing  laboratory
protocols for NSS-I.   Figure 8 shows the flow
of samples and data from  the field through the
processing laboratory, and a brief description
of processing laboratory activities follows.
Samples were processed on  the  same  day
           they  were  received.   When the shipment
           arrived, the analysts inspected the samples for
           proper identification and for shipping damage,
           and noted comments concerning the samples
           on the sample log-in sheet.  Each sample was
           assigned  a  unique  batch and sample number
           combination to distinguish  it from any other
           sample in the  survey.  There were 68 batches
           of samples (numbered from  2100 to  2167)
           analyzed  during the NSS-I.    Each batch of
           samples  contained  routine samples, one field
           (or  processing  laboratory)  blank, one field
           duplicate, and  at ieast one audit sample.  Each
                                            30

-------
                 Form 4
              (3 copies)
                          PROCESSING
                          LABORATORY
              Forms 4A. 6. and 7
Form 3 (2 copies)
                                        Form 3
                                       (1 copy)
         EPA
       SAMPLE
    MANAGEMENT
       OFFICE
                                 Forms 4, and 5
                                 (2 copies each)
                                   and Form 3
                                    (1 copy)
               (2  copies each)
        EMSL-LV
          QA
         STAFF
 Form 3
(1 copy) and
                                                                      Sample Data
                                                                        Package
                                                                       (1 copy)
  Sample Date
   Package
   (1 copy)
                                                            Forms 4, 4A, 5, 6, and
                                                                 (1 copy each)
                                                          ORNL
                                                          DATA
                                                          BASE
                                                      MANAGEMENT
                       Sample Dat
                        Package
                        (1  copy)
      Field and Processing Laboratory
       Form      Description
        3   Shipping
        4   Stream Data
        4A  Hydrologic Data
        5   Batch/QC Processing
              Laboratory Data
        6   Stream Episode Data
        7   Watershed
              Characteristics
NSWS FORMS
Analytical Laboratory Sample Data Package
Form
11
13

14a

15a

•3
16a

Description
Summary of Sample
Results
ANC and BNC
Analyses Results
QC Data for ANC
and BNC Analyses
Specific Conductance
(Measured and
Calculated)
Anion-Cation Balance
Calculations
Form
17
18
19

20

21
22


Description
Ion Chromatography
Resolution Test
Detection Limits
Sample Holding Time
Summary
Blanks and QCCS
Results
Dilution Factors
Duplicates Results


                                       aForm not required to be submitted with
                                         data package but recommended for internal
                                         QC requirements.
EMSL-LV - U.S. EPA, Environmental Monitoring Systems Laboratory, Las Vegas, Nevada
ORNL - Oak Ridge National Laboratory, Oak Ridge, Tennessee
                    Figure 7.  Data form flow, National Stream Survey • Phaee I.

                                             31

-------
                            _SAMPLES FROM
                              FIELD SITES  ~
                   SAMPLES FROM
                     SUPPORT
                   LABORATORY
                                            PROCESSING LABORATORY (Next Day)
                            Samples Organized
                              into Batch
                        .. Total -
                        Monomeric
                         Aluminum
                        and Nonex-
                        changeable
                        Monomeric
                        Aluminum
                       Specific
                     Conductance
              Aluminum Extraction
                                                             ALIQUOT PREPARATION
                                                              1. Filtration
                                                              2. Preservation
                                                              3. Storage at 4 'C
                                 Quality Assurance
  Seven Aliquots Packed for
   Shipment and Sent to
   Analytical Laboratories
   via Overnight Courier
             Figure 8. Flow of samples and data from the fleld through the processing laboratory.
routine, blank, duplicate, and audit sample was
randomly numbered within  the batch.  Each
batch   was  sent as  a  unit  to  a specific
analytical laboratory.  A batch contained  up to
40 samples; the fewest number of samples in
a batch was 8.

      Generally, samples from the mid-Atlantic
and southeast screening sites were  grouped in
the same batch.  However, if the total number
of incoming samples (including duplicates,
audits,  and blanks) exceeded  the  number of
sample analyses required of a laboratory  in the
SOW,  then separate  batches  were prepared
for mid-Atlantic and southeast  screening
samples.  In this case, each batch contained a
blank, a duplicate, and an  audit sample.  The
communications  center personnel at EMSL-LV
informed the base site coordinators whenever
it was necessary to collect  more than one field
blank and duplicate to accomodate the large
sample load.   Each of the two batches were
sent to different  analytical  laboratories.  After
the new contract  with Laboratory 2 became
effective,   both  batches were  sent  to  this
laboratory.

     After  sorting  the  samples into batches,
the four syringes  collected in  the field were
distributed to  analysts to  measure pH,  DIG,
                                             32

-------
and  total and  nonexchangeable  monomeric
aluminum species  and to prepare the total
extractable aluminum  aliquot.   The samples
collected in sealed  syringes allowed
measurements at the processing laboratory,
within a short holding  time, of some variables
(pH, DIG, and the aluminum species) that tend
to become unstable  over time.   The  sealed
syringes  minimized chemical changes  before
analysis. The contents of the Cubitainers were
divided  into six  additional  aliquots and
subsamples from each Cubitainer were used
to obtain specific conductance, turbidity, and
true color measurements. Figure 9 shows the
preparation  and preservation procedures for
each aliquot.

      The instruments and methods used for
analyses  are listed in Table 1.   Processing
laboratory analytical methods are described by
Hillman et al. (1987).  Quality control check
samples  used  in  the processing laboratory
were  measured  as specified in the QA plan
(Drouse et al., 1986a).

      At the processing  laboratory  the
procedures for preparing and preserving each
of  the  seven aliquots taken  from each
Cubitainer  sample  were  specific  for  the
variable  to be  measured  at  the analytical
laboratories. The  aliquots were stabilized by
using filtration, acid preservation, refrigeration,
or  some combination of these  procedures.
Filtration removed suspended material  in order
to reduce biological  activity and  to eliminate
surfaces  that  could  adsorb  or  release
dissolved chemical species.   Acid was added
to some aliquots to prevent loss of dissolved
analytes through precipitation, chemical
reaction,  or biological  activity.   All  aliquots
were  stored and  shipped  at  4°C to inhibit
biological activity  and,  in  the case   of total
extractable aluminum aliquots, to  reduce
volatilization of solvent.

      Once the samples were preserved, the
aliquots were  prepared and packed in a
shipping container with frozen  gel packs and
sent  by  overnight courier  to  the analytical
laboratories.  Extractable aluminum   aliquots
were separated from the other aliquots. These
aliquots were inserted  into a Styrofoam rack
and packed  in a separate shipping container
that contained frozen gel packs.

    A shipping form, Form 3, was completed
and copies were sent with the aliquots to the
analytical laboratories and to the EPA Sample
Management Office (Figure 7).   As soon as
shipping  activities were  completed,   the
processing laboratory personnel notified the
EMSL-LV communications center which
tracked custody of the samples from the field
to the  processing laboratory to the analytical
laboratories.

    Analytical data, QC data, and comments
pertinent to sample analyses were recorded in
bound  laboratory logbooks and then on the
batch/QC Form  5 (Figure 7).  All logbook data
and  forms were reviewed by the processing
laboratory supervisor or coordinator to ensure
that calibration and QC checks were within the
required  limits  and that all  comments  and
qualifiers were complete and understandable.
All forms were then sent to the QA staff in Las
Vegas  for review of data consistency before
transmittal to ORNL for data entry.
 Analytical Laboratory Operations

     Analytical laboratory  personnel were
 responsible for  inspecting  the  samples
 received from the processing laboratory  for
 damage, logging in the sample  batches,
 analyzing  the  samples  according  to
 procedures described in the statement of work
 and published in the NSS-1 analytical methods
 manual (Hillman et al., 1987), and preparing and
 distributing data  packages  (Figure 7)
 containing the  analytical results.   For each
 shipment,  laboratory personnel recorded all
 notes concerning sample  condition  on  the
 shipping form  and sent  a  copy  of  the
 annotated  form  to   the  EPA Sample
 Management Office.

     As part of the contract  requirements,  the
 analytical laboratories agreed  to  follow good
 laboratory practices related to laboratory
 cleanliness and the  use  and  storage of
 reagents, solvents, and  gases.  For standard
 guidelines  regarding    general   laboratory
                                          33

-------
          UNFILTERED;
       7  NITRIC ACID
  f FILTERED;
1  | NITRIC ACID
   PRESERVED
        ( FILTERED; METHYL-
       2J ISOBUTYL KETONE
          EXTRACTION   "
       3 | FILTERED^
        j FILTERED;
       4] SULFURIC ACID
          PRESERVED
       5\  UNFILTERED
        ( FILTERED;
       6| SULFURIC ACID
          PRESERVED
                        15
          PRESERVED
                                  Ca, Mg. K, Na,
                                  Fe, Mn
                            Al-ext
                                       2-
                           CI-, F , SO4
                            NO3~, SiO


                           DOC, NH4f
                           DIC-init, DIC-eq,
                           pH-ANC, pH-BNC,
                           Cond-lab, pH-eq,
                           ANC, BNC
                           Al-total
  4 °C   ANALYTICAL
OVERNIGHT LABORATORY
 COURIER
Figure 9.     Preparation and preaervation procedurea for each aliquot at the preparation laboratory for
           the National Stream Survey - Phase I.
practices,  the analytical laboratories   were
directed to follow procedures in the Handbook
for  Analytical Quality  Control jn  Water and
Wastewater Laboratories (U.S. EPA, 1979). The
analytical laboratories  also  were  required  to
operate according to a uniform set of internal
QC  procedures,  as  described in the QA plan
(Drouse  et  at.,  1986a),   to check  data
consistency,  and  to  document method
performance.    Table  1 lists the analytical
instrument or method used for each variable.

     A maximum sample holding time, deter-
mined from the  time of  sample preservation
to sample analysis, was established for each
variable  measured  in   the  analytical
laboratories (Table 5).   These holding times
                                    were based  upon  information  from  the
                                    literature, the best scientific judgment related
                                    to the defined  needs, and the logistical
                                    demands and limitations of the NSS-I. After all
                                    initial analyses were completed, the analytical
                                    laboratories  refrigerated the  samples at  a
                                    temperature  of 4 "C in  case  reanalyses were
                                    necessary.    The samples  remained at  the
                                    laboratories  for approximately 6 to 12 months
                                    or until notice was received from the EMSL-LV
                                    QA manager to dispose of the samples or ship
                                    them to EMSL-LV for storage.

                                        Each data  package  prepared  at  the
                                    analytical laboratories included a set of NSWS
                                    forms  (Drous6 et al.,  I986a) containing  the
                                    following information:
                                         34

-------
1.     Measured sample  concentration  in  the
      appropriate units for each variable.

2.     Titrant concentration and titration data
      points for each sample for  ANC and
      BNC.

3.     Percent  conductance difference  calcu-
      lation for each  sample  (optional;  this
      calculation is an initial check made in the
      analytical laboratory to ensure data
      consistency,   but it  is  also performed
      during  data  verification  under  the
      direction of the EMSL-LV QA manager).

4.     Percent  ion  balance  difference
      calculation for  each  sample  (optional;
      this calculation is an initial  check made
      in the  analytical laboratory to  ensure
      data  consistency,  but  it is also
      performed during data  verification under
      the  direction   of  the  EMSL-LV  QA
      manager).

5.    Ion chromatograph specifications.
6.   Instrument detection limits.

7.   Date of  sample analysis and sample
    holding time.

8.   Calibration and reagent  blank values and
    QCCS values.

9.   Internal  (laboratory)  duplicate precision
    calculated  as percent  relative standard
    deviation.

     Each data package included a cover letter
from  the analytical laboratory manager to the
QA group at EMSL-LV.  The letter specified the
batch ID number and the number of samples
analyzed,  identified  all problems  associated
with  the analyses, described  any deviations
from  protocol, and contained other information
that  the  laboratory  manager  considered
pertinent to a particular sample or  to the entire
batch.  Copies of the completed data package
were sent to the EMSL-LV QA staff for review,
to ORNL for data entry, and to the EPA Sample
Management Office for sample tracking.
Table 5.   Maximum Holding Time Requirements* Before Sample Analysis at Analytical Laboratories,
          National Stream Survey - Phase I
              Variable
                                                                             Holding time
 Nitrate, total extractable aluminum6

 ANC, BNC, specific conductance, DIG, DOC, pHc

 Phosphorus, ammonium, chloride, sulfate, fluoride, silica

 Calcium, Iron, potassium, magnesium, manganese, sodium, total monomerlc aluminum
                                7 days

                               14 days

                               28 days

                               28 days'
 * Number of days between sample preservation and sample analysis.
 * Although the EPA (U.S. EPA, 1983) recommends that nitrate In unpreserved samples (unacidifled) be
   determined within 48 hours of collection, evidence exists (Peden, I98I, and APHA et al., 1985) that
   nitrate Is stable for 2 to 4 weeks If stored In the dark at 4 *C.
 0 Although the EPA (U.S. EPA, I983) recommends that pH be measured immediately after sample collection,
   evidence exists (McQuaker et al., 1983) that pH is stable for as long as 15 days if the sample is
   stored at 4 *C and sealed from the atmosphere.  The pH  is also measured in a sealed sample at the
   processing laboratory the day after sample collection.
 d Although the EPA (U.S. EPA, 1983) recommends a 6-month holding time for these metals, the NSS-I
   required that all of the metals  be determined within 28 days. This requirement ensured that significant
   changes would not occur and that data would be obtained in a timely manner.
                                              35

-------
  Monitoring

  Communication*
  Communications
               .
       Monitoring QA activities required
  continuous communication among the many

  XriSiorT^nd" '"£  }°' T C°"eCti°n'
  verification, and validation.  These communi-
  cations were centralized through the QA staff

  SSf m0mhmUni?a£0nS C6nter  at EMSI-LV-
 Staff members at the  commumcations center

        t   h     I*'""1?8 indUdin9 Samp'e
 wa      « mT  ' °f StreamS  SamP'ed-
 weather, sampling  projections, supply
 requests, and  miscellaneous problems.   The

 ^hnLf^,38  f- P,°int °f  °°ntaCt f°r a"
 techn,cal and og.stical questions, provided a
 backup contact for samphng teams when base
 unlLlht logis^cs+ Coordinators were
 duoNc^HN  J nated,the  assi9nment of
 duphcate and blank samples to the base sites,

 ODera^onf and *  t™ F*™ **  "**
 £E£t                    93S Processin9
 .aooraiory.

      -    „„     .  ..
      The  communications center  personnel
 were responsible for tracking   sample ship-
 ments  from  the  field to the  processing
 laboratory  to the analytical laboratories  and
 wiJTSS 'rf Shipm^tsi if necfsary- They
 were also respons.ble for ordering audit
 samp es  and  communicating with  the audit
 sample support laboratory  Any appropriate
 nformation  from  the  field,   processing
 abora ory,  analytical laboratories, or support
 laboratory was relayed to the QA staff.
    2   the EPA Samp|e  Management Offjce
        concerning sample tracking and analytical
        laboratory compliance with contractual
        requirements,

    3.   the support  laboratory personnel  to
        addressquestions rela'tin^ to samp e
        preparation  and

    4'   the data base management group  at
        ORNL to clarify the meaning of comments
        t0  deciPher  ille9ible data for data  entry
        and t0 diSCUSS data base desi9n and data
        entry progress.
              «« ^  ^A  * «        .
dirertiv     t  ?  w        !, c°m™mc**d
directly w^h the field crews and the processing
and analytical  laboratory personnel on a daily

nPr«LJ £SH-    y teleph|.one calls were
necessary to discuss sampling,  processing,

mo!*? "S?" !!?!?• re'ated t0 '°9istics.
h« lc ,' ^ 3" u,QC S^ thfj  problems could
be resolved quickly and  efficiently,  and to
obtam current  sample data and QA and QC
information.  The QA staff also communicated
periodically with:
,        ....    iu  .
1.    analytical methods experts at EMSL-LV
     to  resolve issues related to analytical
     methodology,
        Durin9 the N88-1-  conference calls were
    held regularly (either weekly  or  every  two
    weeks,  depending  on need)  for all  survey
    participants to aid  in  efficient exchange of
    information,  problem  solving,  and  improve-
    ments. Discussions  during these calls covered
    survev P"^*, protocol changes, and issues
    related to  sample collection, sample  load and
    analyses,  raw  data  set development,
    resolutions of problems relating to QA issues,
    data evaluation, and  progress of report writina
                                  P   writing.

    On-Site Inspections

       Qn-site inspections of field and laboratory
   activities  were conducted to  ensure that
   sampling and analytical procedures were being
   performed  according to the survey protocol
   The  two  mid-Atlantic  field  base sites,  the
   processing laboratory, and Laboratory 2 were
   evaluated  during  the  NSS-I.    QA  and  QC
   sample data  were reviewed thoroughly and
   used «n conjunction with on-site  evaluations to
   confirm proper operations and to identify any
   necessary  changes in protocol or the need for
   reanalysis.   The  findings  from these
   evaluations were documented in  on-site
   inspection reports.  Because  of  budget
   constraints at the time of sample analyses it
   was not possible to evaluate Laboratory 1 on-
   site during the analyses of NSS-I samples
   However, an on-site  inspection  performed  at
   thia |aboratory during  tne  ELS_T determined
   tnat all  analytical operations followed correct
   protocol.
36

-------
Data  Base  Management and Data
Verification

     The creation of the four NSS-I data sets
(raw, verified, validated, and enhanced)
invoived numerous operational steps  as well
as several NSS-I participants. To create a raw
data set,   all data  were entered into two
separate data sets by two different operators
at Oak Ridge National Laboratory (ORNL).  A
custom program developed using the
Statistical Analysis System (SAS Institute,
Inc., 1985)  compared the two data sets and
identified any inconsistencies  in numeric and
alphabetic  variables.    Any  errors  were
corrected by referring to the original forms. All
NSS-I data sets were created and maintained
at  ORNL  by using  the  Statistical Analysis
System.  When the data sets were complete,
they were transferred via magnetic tape to the
National  Computer  Center  at  Research
Triangle Park, North Carolina. There, scientists
at the Las  Vegas and  Corvallis laboratories
could gain  access to the data sets.  The QA
staff members in Las Vegas were primarily
responsible for data verification and personnel
at  Corvallis  were responsible  for  data
validation.

      In order to meet  the objectives of the
data verification process and  to identify raw
data of questionable or unacceptable quality
that might need to be corrected during or after
data validation, the QA auditors examined the
data for internal consistency and reviewed the
QA and QC data measured and recorded at the
sampling sites, the processing laboratory, and
the analytical laboratories. Geographical data
were verified and validated at the  Corvallis
laboratory; analytical data  were verified  at the
Las Vegas  laboratory.  Computer programs
provided a mechanism to automate  much of
the verification procedure.  Redundant values,
calculated by computer and measured at more
than one place,  were compared.   The veri-
fication process  evaluated data at both the
sample and batch  levels.   Data  verification
took place in two parts:  initial verification of
the  numerical changes  and  final  verification
that involved the  final numerical changes  as
well as the addition of data  qualifier  flags.
These verification activities are  identified  in
figures 10 and 11.

Review  of   Field  and   Processing
Laboratory Data Forms--

    Verification began  when the data forms
from  the  field and processing laboratory
(Forms 4,  4A,  5,  6, and 7;  Figure  7) were
received at EMSL-LV.   A QA auditor reviewed
the  data  forms  for completeness,   for
agreement of stream identification codes given
on the field and processing laboratory forms,
and for  proper  assignment  of  sample
identification codes and  data  qualifier tags.
Specific conductance and pH  measurements
recorded on field and  processing laboratory
forms  were compared in  order to identify
possible measurement or  reporting errors. The
auditors calculated  precision estimates using
field  and  processing laboratory  routine-
duplicate  data and evaluated  the estimates
using  the data quality objective for each
variable (see  Section 6) as a reference.
Measurements for field  audit  samples were
evaluated  using measurement data  provided
by  the support laboratory  as  a reference.
Measurements for  the audit  samples were
compared with results from previous NSWS
surveys for the same audit types.

     Data  anomalies were reported to the field
base site  and  processing  laboratory coor-
dinators  for  corrective  action,  and data
reporting errors were corrected  before the data
were entered  into  the raw  data set.  After
reviewing  the  information on  all field  and
processing  laboratory data forms  for
completeness  and accuracy, the  QA staff sent
the  forms to  the  Oak  Ridge laboratory by
overnight courier.

     At times,  data errors were identified by
communications between the  QA staff, field
and processing laboratory personnel, and the
ORNL staff after the data forms were sent to
Oak Ridge. If the data in question were not
yet entered into the raw data set, the QA staff
sent documentation with instructions to make
changes to the erroneous data.  Most trans-
cription errors were identified  and corrected
                                          37

-------
   RECEIVE FIELD.
/ PROCESSING LAB. AND

\ANALYTICAL LAB DATA

        _EMSL-H
I 	 *

REVIEW OF RAW DATA
FORMS WORKSHEET
EMSL-LV

x-\.


INFORM DATA
PRODUCER
(U, FIELD CREW.
PROCESSING LAS. OR
ANALYTICAL LAB }
EMSL-LV


ENTER RAW DATA
(DOUBLE ENTRY)
ORNL


CORRECT
DISCREPANCIES ON
THE FORMS PRIOR TO
DATA ENTRY
EMSL-LV



COM
ON
'
LETE
TASET
APE
ORNL

EXECUTE
EXCEPTION -
GENERATING
PROGRAMS
EMSL-LV
I
   COMPLETE NSS-
 fRST PASS VERIFICATION
 EPORT WORKSHEET AND
 EXCEPTION RECORD
       EMSL-LV
                                                                                                           National Stream Survey -
                                                                                                                    Phase  E
                                                                                                        ANALYTICAL DATA VERIFICATION

                                                                                                                Initial Verification
                                                        EANA
                                                 YES /  VALUE OF
                                                      BETTER
                                                       HAN (
Figure 10.  Analytical data verification, Initial verification for the National Stream Survey - Phase I.

-------
                          EDIT A COPY OF THE RAW DATA
                           SET TO INCORPORATE VALUE
                              OR FLAG CHANGES
                                          EMSU-LV
                National Straam Survay -
                        Pha«a I

             ANALYTICAL DATA VERIFICATION
                    Final Verification
Figure 11.  Analytical data verification, final verification for the National Stream Survey - Phase I.

                                              39

-------
 before data entry.  All changes made at Oak
 Ridge were initialed and dated by Oak  Ridge
 personnel.  A copy of this documentation was
 sent to the QA staff as confirmation that the
 changes  had been made.   Any data errors
 identified after the  original data were entered
 into the raw  data  set were  corrected during
 verification  and  were incorporated  into  the
 verified data set.

 Review  of  Preliminary  Results from
 the Analytical Laboratories-

      In response to inquiries from the EMSL-
 LV QA staff, personnel  at the analytical
 laboratories  reported  preliminary results  of
 analyses on a daily basis, either by telephone
 or by electronic data transfer,  depending on
 available  resources.    When  requesting
 information,   the QA  staff  members  would
 inquire  about selected QA  samples and a
 group of randomly chosen routine samples to
 minimize  the  possibility  that the laboratory
 personnel would identify the QA samples.
 These preliminary results  were  evaluated  to
 ensure that the acceptance criteria (Drous6 et
 al., 1986a) for the QA samples were  met and
 that data  analyses were performed according
 to protocol.   Whenever  any problems  were
 noted, the QA auditor conferred with laboratory
 personnel to  determine  the  source  of the
 problem and  to  implement corrective action.
 The  primary  objective of this preliminary
 evaluation was to identify and resolve issues
 quickly  before they affected  data quality or
 interfered with the completion of the survey.

 Initial Review  of Analytical
 Laboratory Sample Data  Package--

      The analytical laboratories sent  the
 analytical  data packages  by overnight courier
 to Oak  Ridge  and  to  Las  Vegas  for
 simultaneous QA  review.   While  personnel at
 Oak  Ridge were entering the  data packages
 into   the   raw data  set,  the  QA auditors
 reviewed  the  analytical  laboratory data
 packages manually.  They reviewed the sample
data packages for  completeness,    internal
QC compliance,  and appropriate use of data
qualifiers.  Each  auditor used a  checklist  to
 ensure consistency in this data review process
 (Drous<§ et al., 1986a).

      At the  Las  Vegas laboratory,  the QA
 sample data for each batch were tabulated for
 each laboratory in a computer file before the
 raw data set was available.  The field blank
 data were evaluated to determine if the values
 fell within expected criteria (Appendix C).
 Because  the sampling protocol for obtaining
 blank  samples  for the  mid-Atlantic  and
 southeast screening surveys was identical to
 that used  during  the  NSS-I Pilot,   the
 concentrations found  in the  blank  samples
 from the pilot survey were used to calculate
 control limits for the  NSS-I blank samples.
 These control limits were used to (1) check for
 evidence of sample  contamination,   (2)
 determine the necessity of data confirmation
 or  reanalysis, and (3) generate  data qualifier
 flags to indicate  potential contamination by
 batch or by sample.  The 95th percentile of the
 NSS-I Pilot  field  blanks was chosen for the
 upper control limit, except for a  few variables
 for which the value  of the required detection
 limit (see Section 6) was used.  The negative
 of the value for the required detection limit was
 used for the lower control limit to account for
 background  noise and  minor fluctuations  in
 instrument performance. Concentration values
 below  this  limit  would  indicate  possible
 negative  bias.     Histograms  were  also
 developed with the field blank data for  each
 laboratory to  aid  in detecting  unacceptable
 data for each variable.

     The values for the variables measured in
 a field routine-duplicate  pair were considered
 an  exception  when the concentration of  both
 samples  in a pair exceeded the  required
 detection  limit  by a  factor  of  10 and  the
 precision estimate exceeded  the acceptance
 criteria  (Appendix  C) developed  by the QA
 staff.   Precision  was calculated as percent
 relative  standard  deviation.  The acceptance
criteria  were established to meet the precision
 DQOs (see Section 6) although some flexibility
was allowed  to compensate for  the variance
due to sample handling.

    Bar histograms were developed by the QA
staff  with the audit sample data  for each
                                          40

-------
variable.  Auditors compared the results from
each laboratory for each audit sample in order
to detect any problems inherent within a
laboratory.    Again,  reported  concentrations
from  the audit samples analyzed in the
support laboratory  or  historical NSWS data
were used as a reference before the raw data
set was  available  in  order  to identify  and
correct unacceptable data early in  the survey.
Quality control charts were developed with the
audit  sample data in order to detect suspect
data points.   The  auditor requested that the
analytical laboratory manager confirm all  data
points outside control limits.

      Any repeated measurements  made in the
field,  processing  laboratory,   and analytical
laboratory  such as for specific conductance,
dissolved inorganic carbon,  and  pH were
compared. Any discrepancies were reported to
the field base sites, processing laboratory, or
analytical laboratory for corrective action.

Automated Review of  NSS-I  Data
 Base-

      Once the development  of the  raw  data
 set was completed at  Oak Ridge and the data
 set was available to the EMSL-LV QA staff, the
 data were reviewed by using the AQUARIUS II
 exception-generating and data review
 programs  (Table  6).    These computer
 programs  were used to identify or flag results
 that were  exceptions,  i.e., results  that did not
 meet the  expected  QA  and  QC  limits.   The
 auditors used the output from these programs,
 along with original  data  and  field notebooks,
 to complete the NSWS verification report
 specified in the QA plan (Drouse et al., I986a).
 The verification report  is a worksheet designed
 to guide the auditor systematically through the
 verification process by  listing the steps that
 lead  to identification  of  QA  exceptions,
 explaining  how to  flag  data,  tracking data
 resubmissions and  requests for confirma-
 tions and reanalyses, and  summarizing any
 required modifications to the raw data set (i.e.,
 preparing records of  numerical  and flag
 changes required to  create the  verified data
 set).   Flags  and  their definitions  are  given in
 Appendix B.
     Each  sample  value  was  verified
individually and by  analytical batch.   All
samples had  to  meet  internal  consistency
checks for percent ion balance difference and
for percent  conductance difference.   Percent
ion balance  difference (%IBD) is calculated as
follows:
        Z anions - Z cations + ANC
   Z  anions  + Z cations +  ANC + 2  [H+]
                                      MOO
where:
  Z  anions
Z  cations
[Cl'l +
[S042']

[Na+]  +
[Mg2+]
[N0~]
                       [NH4+]
 [Ca2+]
     ANC    Z. alkalinity (the  ANC value is
              included  in the calculation to
              account for the presence of
              unmeasured ions  such as organic
              ions)

        [H+] =  (10'pH) x 106 peq/L

 Note:    Brackets  indicate  concentration of an ion
         in  microequivalents per  liter.

      A list  of factors for converting mg/L  to
 /jeq/L for each variable is given in the analytical
 methods manual (Hillman et al., 1986).  After
 confirmation of  the  original  suspect  values,
 samples which had a poor ion balance were
 flagged or  reanalyzed  unless  a  high  DOC
 measurement  accounted for  the  difference.
 Table 7 lists the acceptance criteria for the ion
 balance difference.

     The calculation program was modified  so
 that whenever the absolute value of ANC was
 less than or equal to 10 /Lieq/L, the value zero
 was substituted  for ANC in the equation. The
 equation is sensitive to slight variations in ANC
 for samples that have very low ionic strength.

      The percent  conductance  balance  is
 determined as follows:
                                            41

-------
Table 6.    Exception-Generating Programs WHhln the Aquarius I! Data Review and Verification System
           Program
                                                          Sample (data) type
Exception-generating programs:
   Audit  Sample Summary

   Field Blank Summary
   Field Duplicate  Precision Summary
   Instrumental Detection Limit Summary
   Holding Time Summary
   % Conductance Difference Calculations
   Anion/Cation Balance Calculations
   Internal Laboratory Duplicates
   Protolyte Analysis
   Reagent/Calibration  Blanks and QCCS

Data review programs:
   Comparison of Total Aluminum and Extractable
      Aluminum
   Raw Data Listing
   Comparison of Form 4 and Form 5
   Comparison of Form 5 and Form 11
   QA/QC Flag Summary
   Modified Gran Analysis Program
   Audit Sample Window Generation
                                           Field natural and synthetic and laboratory
                                              natural and synthetic
                                           Blank
                                           Routine-duplicate pairs
                                           All species
                                           All species
                                           All spades
                                           All species
                                           Analytical laboratory duplicate data
                                           DIC,  DOC, pH, ANC, and BNC data evaluation
                                           All species except pH
                                          Total aluminum and extractable aluminum

                                          All field and laboratory data
                                          pH, DIC, and specific conductance
                                          pH, DIC, and specific conductance
                                          All exceptions
                                          ANC and BNC
                                          All species
 Table 7.
Chemical Reanalysis Criteria for Sampie Ion Balance Difference and Percent
Specific Conductance Difference
 A.       Anion-Cation Balance

         Total Ion strength fueg/Ll
                <50
                >50 and IOO

 B.       Specific Conductance

         Measured specific
         conductance  (uS/em)
                <5
                >.5 and <30
                >30
                                      Maximum % jon balance difference*
                                                     60
                                                     30
                                                     15
                                            Maximum % specific
                                          eonductanet difference*
                                                     SO
                                                     30
                                                     20
* If the absolute value of the percent difference exceeds these values, the sample is reanalyzed.
  When reanalysis is indicated, the data for each parameter are examined for possible analytical
  error. Suspect results  are then redetermined  and the above percent differences are recalculated
  (Peden,  I98I).  If the percent differences for reanalyzed  samples are still unacceptable or no
  suspect data are identified,  the QA manager  must be contacted for guidance.
                                                  42

-------
/calculated conductance - measured conductance
                                      -J100
            measured conductance
      The ions used to calculate conductance
 are Ca,  Cl",  CO*-, H+, HCO3', K+,  Mg, Na,
 NO3", OH", and SO42~. Calculated conductance
 is determined by multiplying the concentration
 of  each  ion by the appropriate factor given  in
 Table 8.    All  three  measured  specific
 conductance values from the field, processing
 laboratory,  and the analytical laboratory are
 compared  to the  calculated value.   The
 acceptance  criteria for the  differences are
 listed in  Table 7. Any routine stream sample  or
 QA sample  that did  not  fall  within the
 applicable criterion was qualified with a flag  or
 reanalyzed.

      On the basis of  the analytical results
 reported for QA and QC samples, the QA staff
 directed the analytical laboratory to  confirm
 reported values or  to reanalyze selected
 samples or sample batches, if necessary.

      Generally,   reanalyses  were requested
 when at least three  different  QA  and QC
 samples generated  flags  for a particular
 variable in a particular batch or when incorrect
 methodology was  used during the  original
 analysis.  In such cases, sample reanalysis
 usually was requested for a given variable on a
 per-batch basis.   A  tracking  form  for  data
 confirmation and sample reanalysis provided a
 standard format for data transfer between the
 QA staff and the analytical laboratories.
 Suspect data that were not corrected through
 confirmation or reanalysis were flagged with
 an appropriate  data  qualifier  when the
 exception-generating programs were  rerun
 during  final verification.   Additional data
 qualifiers were added to a given variable if the
 QA samples (field blanks, field duplicates,  or
 audit samples)  within  the   same  analytical
 batch did  not  meet  the acceptance criteria.
 Acceptance criteria  were calculated for the
 audit samples on two different occasions. The
 first calculation was  with Data Set 1 when it
 was received from ORNL and before the initial
 verification.  The second calculation was after
 all the  numerical changes were made to the
 raw data and before final verification. The QA
plan (Drous§ et al.,  1986a) gives a detailed
description of the method for calculation of
acceptance criteria for audit samples.   Every
batch that contained an audit sample with an
unacceptable  analyte concentration was
flagged accordingly.   Acceptance criteria for
the audit samples are found in Appendix C.

     Data were also given qualifier flags if
internal QC  checks  (such as calibration and
reagent blank analyses, internal duplicate
precision, required instrumental detection limit,
QCCS central limit criteria, and  maximum
allowable holding times) were not  met.  The
protolyte analysis program identified discre-
pancies  related to  processing and analytical
laboratory measurements of pH,  DIG,  ANC,
BNC,  and  DOC when carbonate equilibria,
corrected for organic  species,   did  not have
internal agreement.  Flags were produced for
data that  were  questionable.    The  overall
process involved:

1.   performing  redundant alkalinity calcula-
     tions using three different measurements
     for both pH (pH-closed, pH-ANC, and pH-
     eq)  and DIG (DIC-closed,   DIC-init,  and
     DIC-eq),

2.   verifying measured ANC and BNC, and

3.   determining  whether  the  system  is  car-
     bonate  or mixed.  Empirical relationships
     defined by Oliver et a!. (I983) were used to
     estimate the contribution of organic proto-
     types to the measured ANC.  A data quali-
     fier flag  was  assigned for  samples in
     which the  presence  of organic species
     resulted in an  ion  imbalance for the
     sample.

     Another  program compared the extract-
able aluminum and total aluminum values for
each sample.   By definition,  the extractable
aluminum concentration for a sample could not
exceed the total aluminum concentration.  The
program generated a flag  when the value for
extractable aluminum was at least 0.015 ring/l-
and when it was higher than the value for total
aluminum by more than 0.010 mg/L (twice the
required detection  limit).   This qualification
was intended to account for background noise
                                            43

-------
Table 8.    Factors for Determining the Conductances of Ions foiS/cm at 25 'C)a'b
Ion
Calcium
Chloride
Hydrogen (H+)

Hydroxyl (OH')

Bicarbonate (HCO3~)
Carbonate (CCj2")
Factor
per mg/L
2.60
2.14
3.5 x 10s
(per mole/L)
1.92 x 105
(per mole/L)
0.715
2.82
Ion
Magnesium
Sodium
Ammonium

Sulfate
Nitrate
Potassium

Factor
per mg/L
3.82
2.13
4.13

1.54
1.15
1.84

  Taken from APHA et al. (1985) and Weast (1972).  Ion concentration is multiplied  by the listed factor
  to obtain the conductance value.  The concentrations of the  ions that  are not measured directly are
  calculated by means of the following equations:

             [H+]  =  10"pH

     where:     pH  =  initial pH measured before BNC titration.  (Brackets represent molar concentrations.)
    where:     Kw  =  1 x 10"13-8 moles/L

                      5.080 (DIC(mg/L)) [H+]
     urn - tmnn ^
     HC03  (mg/L)
         2- lmnl\ \     4.996
        3   (m9/L)  "
  where:       K,  =  4.4463 x 10"7 moles/L, ICj = 4.6881 x 10~11 moles/L, and
               K1  =  first ionization constants for carbonic acid
               ^  =  second ionization constants for carbonic acid
               Kw =  ionization constant for water

  Conductance factors are not given for ionic aluminum, iron, or manganese because these ions are rarely
  present in concentrations great enough to affect the percent conductance  difference.
                                                    44

-------
(especially  at low  levels)  and for minor
fluctuations  in  instrument reading  and
calibration.  In all cases, each flag generated
by the  AQUARIUS  II  system was evaluated
by  an  auditor  for  reasonableness  and
consistency  before  it  was  entered  into the
verified data set.

      Comparison of  upstream  and down-
stream data for all streams and comparison of
first  and second site visits for streams in the
mid-Atlantic region provided an additional
method to  identify anomalies.   If  the dif-
ferences in the data appeared suspicious and
there were  no obvious reasons  for the dif-
ferences such as a point source of  pollution
(e.g., mine tailings) or a recent rain event noted
on the  field forms, then  confirmation of the
suspect data  was requested from the  analy-
tical laboratories.

      Changes and additions to the raw data
set  resulting  from corrections to  analytical
data or from  reanalysis of  samples (Form 4
from the  field,  Form 5 from the processing
laboratory, and the sample data package from
the analytical laboratory) were made by the QA
staff during initial verification.  The  QA staff
also added  all appropriate data qualifiers not
marked on  the original  data forms.    This
verified data  set was  sent to Oak  Ridge
National  Laboratory on  magnetic  tape.
Personnel  at Oak Ridge  added changes
resulting from  corrections to nonanalytical
forms (Forms 4A, 6,  and 7) provided by the
EMSL-LV  QA  staff on a modification sheet
(Drouse et  al.,  1986a) to create the official
verified data set.

Preparation and Delivery of
Verification Tapes--

      Three separate versions of the verified
data were delivered on magnetic tape to ORNL
where they  were made available to the ERL-C
staff for validation.   The first tape  was
delivered after the initial run of the exception-
generating  programs.   All  numeric changes
identified at this time were incorporated into
the  changed data   set,   and the  initial
verification  was complete.   An  intermediate
tape was delivered midway through final
verification  in order to aid the ERL-C staff in
data assessment.   The third and final tape
was delivered after  all the reanalyzed  and
corrected data were incorporated into the data
set,  the exception-generating programs were
rerun and evaluated, and all the data qualifiers
were applied to the data set.

     The third and final verified data set was
generated by the  EMSL-LV QA computer
support staff. The tape was sent to Oak Ridge
where  it was checked for consistency  before
use  during data validation.  Oak Ridge was
responsible for generating the official verified
data set (Data Set 2) and for archiving the
tape.

Data Validation and Data Base
Management

     Data  validation took  place  primarily at
ERL-C.  The validation process  is  a way to
search for observations that  may represent
entry or analytical errors  or  unusual  water
chemistry.   This process incorporates
univariate,  bivariate, and multivariate analyses
(Drouse et  al., I986a).

     Validation of NSS-I data began with the
raw unverified data. During validation, a matrix
for  each  subregion was  constructed  that
depicted the results of validation checks on
each individual water sample and datum.
Using this  matrix, outliers were identified and
sent to the QA staff at  EMSL-LV to be checked
for possible entry errors. Sites having atypical
chemistry when compared  to other sites in a
subregion  (unique  multivariate relationships)
were identified and evaluated as unusual sites
(e.g.,  sites affected by acid  mine drainage,
agricultural impact, or tidal influence).   Data
from unusual sites generally appeared as
outliers in several statistical  analyses (e.g.,
regression and univariate statistics).

     In addition to  identifying outliers,
validation identified stream  samples that might
have been collected during a precipitation  or
snowmelt  episode  (or  influenced by it) and
therefore  would  not provide  an acceptable
index of base flow chemistry (Kaufmann et al.,
1988).  Although great care was  taken not to
                                           45

-------
 collect  samples during an episode,  such
 conditions might not have been apparent to a
 sampling team.

      Sites considered to be of noninterest for
 calculation of NSS-I population estimates were
 identified as those satisfying certain criteria
 (Kaufmann et al., 1988).   When  data for an
 entire reach were considered unacceptable for
 the intended  use—to make  population
 estimates,   for  example-a disturbance  flag
 code was placed in the data set. Disturbances
 that might affect stream reach  data include
 acid mine sites, pollution, tidal influence,  and
 watershed disturbances.

 Enhanced Data Set

      After data validation, the enhanced data
 set  was  prepared to  resolve problems  with
 erroneous  data and  missing  values in  the
 validated data set. (Kaufmann et al., 1988). In
 cases where  it  was deemed  necessary,
 substitutions were performed according to the
 following criteria:

 1.    Values from duplicate samples were
      used whenever possible.

 2.    If  a duplicate measurement  was  not
      available, a value from an alternate visit
      to the site was used.

 3.    If  a duplicate  measurement  or a
      measurement from an alternate visit
      was not available,  a  substitution value
      was  calculated  by  means  of  a linear
      regression model.  This was done by (I)
      calculating a predicted value based on
      observed  relationships with other
      chemical variables  or  (2) predicting a
      value based on  relationships  between
      upstream and   downstream  observa-
      tions of the same chemical variable.

4.    The  last  option  for identifying a
      substitution value  was  to  use   the
      subregional mean.

     All  substitute values were examined for
acceptability before they were included in  the
final  data set.  In addition to the substitute
values that were calculated,  negative values
for parameters other than ANC and BNC that
resulted from analytical calibration bias were
set equal to zero.  Streams considered to be of
noninterest were flagged in a manner by which
they  could be excluded when using  the
enhanced  data  set  for  making  regional
estimates of the target population.

     An index value for each chemical variable
for each NSS-I sampling site was calculated
by  averaging data from  duplicate pairs and
multiple sample visits.  The resulting data set
contains  a  single value for each variable for
each sampling  site (i.e.,  one  observation for
each upstream and downstream site).
                                          46

-------
                                    Section 5
           Results and Discussion - Assessment of Operations
Field Sampling Operations
and Protocol Changes

     The EMSL-LV QA staff conducted on-site
inspections of field sampling operations  at
both of  the mid-Atlantic survey base  sites.
Observations of presampling  calibration, QC
procedures,  sampling  methods,   sample
handling, and sample shipment indicated that
the proper  protocols were followed.   Field
personnel  strictly adhered  to  QA and QC
protocols, accurately documented problems,
and took corrective action when necessary.
Due to time constraints, the QA staff did not
visit and inspect the southeast screening field
operations;  however, NSS-I management and
supervisory personnel did visit these field
operations.  Their observations of the sampling
activities,  sample  handling,  and  sample
shipments indicated that all required protocols
were followed for activities of the southeast
screening survey. All relevant findings from the
QA inspection at the mid-Atlantic sites were
forwarded  to  the  screening  operations.
Protocol changes and  problems encountered
during the field sampling operations for the
NSS-I are discussed in detail  in Hagley et al.
(in press). QA issues identified during the on-
site evaluations and in the course of the survey
are described in Table 9.  Further information
relating to the data  base  variable fields
mentioned in this section can  be found in the
data base dictionary (Sale, in press).

Processing Laboratory
Operations and Protocol
Changes

     Two  on-site  QA  inspections  were
performed at the processing laboratory during
the survey.   The  laboratory staff followed
protocols and all activities  were generally
satisfactory.   Some QA related  issues were
resolved as  a  result of  the  inspections.
Protocol changes  were  implemented  in
response to  the  sample load,  concentration
values,  and  recommendations  for improve-
ments  from  the  EMSL-LV  methods develop-
ment group, QA staff, or processing laboratory
staff.  These issues  are listed in Table 10 and
the most complex are discussed further in the
following paragraphs.

Specific Conductance Measurements

    Processing  laboratory conductance
measurements are not reliable from batch 2100
through batch 2127.  The conductance probe
was thought to  be  faulty early in the survey
because increasing values for field blank
samples  were noted with  each new batch.
Attempts to repair the probe were unsuccess-
ful.  An alternative  probe was obtained from
the EMSL-LV methods laboratory and was put
into use with batch 2104. This probe appeared
to operate initially,  but  beginning with batch
2118,  high values were  again noted for field
blanks.  The processing  laboratory water
systems, which were the sources for the field
blank  samples,   were suspected to  be  the
probable cause of the increasing values for the
blank samples.  After further investigation, an
additional new probe was  obtained and was
put into use  beginning with batch 2128.  This
newer  probe provided acceptable field blank
measurements and  it was determined that
there was not a problem with the processing
laboratory water system.

Nitrate Contamination

    At the  beginning of survey operations,
preliminary data from   the analytical
laboratories indicated nitrate contamination in
                                         47

-------
 Table 9.
Slgnlfleant Findings Concerning Field Sampling Operations and Their Effect on Data,
National Stream Survey - Phase I
           Finding
                              Corrective action
                                                                               Effect on NSS-I data
 On two occasions, samplers
 realized on the second visit that
 they had sampled the wrong
 stream on the first visit.
 Since 1986 was an unusually
 dry year,  a number of streams
 that might normally be flow-
 ing during the spring were
 completely dry or stagnant.
 A number of streams could
 only be sampled at one site
 because more than 90 percent
 of the reach  was dry.

 Temperatures in the shipping
 coolers sometimes deviated
 significantly from the rec-
 ommended 4 "C  upon arrival
 at the processing laboratory.

 Three sample shipments (8
 samples total) were misrouted
 by the overnight courier
 service.
Some shipping containers were
damaged during shipment caus-
ing leaks in 31 of 1,512
Cubitainers that were shipped
from the field.  Of the 5,912
syringes that were shipped,
13 were received with broken
syringe  tips.

Measured values for specific
conductance QCCSs used in the
field were consistently outside
acceptance criteria (Hagley et ai.,
in press) in the beginning of the
survey.
                          Data for the streams that should
                          not have been sampled were
                          qualified with an XO flag
                          (Appendix  B) on the sample
                          identification code  in the verified
                          and validated data sets.

                          None.
                          Numbers and types of frozen-gel
                          packs were adjusted.
                         Extra effort was expended by
                         the field crew to track and
                         recover the samples.
                         The use of hard plastic con-
                         tainers or the taping of the
                         more fragile Styrofoam con-
                         tainers for reinforcement pre-
                         vented any further breakage.
                         Processing laboratory personnel
                         changed the original protocol
                         which called for daily preparation
                         using volume dilution techniques
                         to a protoco! which called for
                         using weight  dilution techniques
                         and preparing the solution in
                         bulk quantities.  More careful
                         attention to good laboratory
                         techniques also resolved this
                         problem.
 No apparent effect.  First-visit
 data for these two streams were
 not available for the creation of
 the enhanced data set. Second
 visit  data  were used.
                                                              None.  Enough water samples were
                                                              collected from the target stream
                                                              population so that the DQO
                                                              for completeness was met (see
                                                              Section 6).
 No apparent effect.  Temperature
 of the cooler in which the sample
 was shipped is noted in  the
 variable field, COOLR, in  the
 verified and validated data sets.

 No apparent effect.  Samples
 processed after the 30-hour
 holding time were qualified on the
 sample identification code in the
 raw, verified, and  validated data
 sets.

 No apparent effect.  Data for
 samples from these damaged con-
 tainers  are qualified in the tag
 field in  the raw, verified, and
 validated data  sets.
Comparisons of specific
conductance measurements made
in the field and  at the analytical
laboratories indicated that field
measurements were  acceptable.
Data for QCCSs that were not
acceptable were qualified in the
tag field in the raw  and verified
data sets.
                                                     48

-------
Table 10.   Significant Findings Concerning Processing Laboratory Operations and Their EHect on  Data,
           National Stream Survey - Phase I
         Finding
                                          Corrective action
                                                                             Effect  on NSS-I data
Underestimation of daily sample
loads required that additional
processing laboratory analysts be
hired during survey operations.
The  necessity of  a shortened
training period covering only one
or two methods resulted  in a loss
of four specific conductance
values and one turbidity value;
these samples were inadvertently
discarded before measurement.

Field duplicate samples were not
assigned randomly in 16  batches.
 Early in the survey, one field
 routine-duplicate pair was
 inadvertently split between two
 batches and hence between two
 analytical  laboratories.

 An initial comparison between
 processing laboratory and
 analytical  laboratory specific
 conductance measurements
 indicated that the  processing
 laboratory values were not
 temperature compensated to
 25 *C  as  required.
 Processing laboratory specific
 conductance measurements for the
 first 28 batches often did not
 agree with the field and analytical
 laboratory measurements.  See text
 for further discussion.
Careful monitoring of the newly
trained analysts eliminated this
problem.
Processing laboratory personnel
were notified of this incorrect
practice and the problem was
resolved.
 Processing laboratory personnel
 were notified.  The problem did
 not occur again.
 After the survey was completed,
 all processing laboratory specific
 conductance data were  corrected
 to 25 "C by using the temperature
 data recorded in the analysts'
 logbooks. These corrected values
 are included in the verified,
 validated, and enhanced data sets.
 The original, uncorrected values
 are included in the raw  data set.

 The use of a new conductance
 probe provided acceptable data.
 See text for further discussion.
A turbidity measurement for one
sample was lost.  For specific
conductance, the field or
analytical laboratory measurements
could be used in place of the
processing laboratory
measurements.
Apparently, the analytical
laboratory personnel did not
identify these consecutive  samples
as duplicates and there was no
effect on data quality.

Batches 2104 and 2105 do not have
field  routine-duplicate pairs that
can be used as QA samples.
 None.
 For batches 2100 through 2127,
 specific conductance data from  the
 processing laboratory are
 considered unreliable.  See text
 for further discussion.
                                                                          (continued)
                                                     49

-------
Table 10.  (Continued)
         Finding
     Corrective action
                                                                            Effect on NSS-I data
In previous NSWS surveys, only
one pH  meter was used  per batch
of samples; more than one was
used during the  NSS-I to complete
daily analyses of the  large sample
load.
A dilute-buffer check solution was
developed and was measured daily
to ensure that the meters
produced comparable results (Arent
et al., in preparation).  Field
routine-duplicate pairs always were
analyzed on the  same meter.
None.  The meters produced
comparable results.  The
identification number of the meter
that was used to analyze each
sample is recorded in the raw and
verified data sets in the variable
field PHMID.
At the beginning of the survey,
nitrate contamination was
identified in some  QA samples.
See text for further discussion.
A change  in the filtration
procedure and more care during
the preservation process eliminated
the problem.  See  text for futher
discussion.
Eight samples were  identified as
nitrate contaminated and are
identified by a data  qualifier flag
in the verified  and validated data
sets. See text for further
discussion.
Filtration of the stream samples
was time consuming and labor
intensive.  Membranes had to be
changed frequently during
filtration.
For future stream sample analyses,
a two-stage filtration procedure
that employs a coarse filter in
addition to the 0.45-/jm filter that
was used during the NSS-I is
recommended.
None.
The protocol for measuring
turbidity was originally developed
for the  Eastern Lake Survey -
Phase I and was based on an
expected range of 0 to 20
nephelometric turbidity
measurments exceeded this range.

In previous NSWS surveys, the
buffer used in the extracable
aluminum procedure for Aliquot 2
was ammonium chloride/ammonia.
It is thought that volatile
ammonium chloride fumes
generated  by this buffer can coat
the surface of labware and the
laminar flowhood,  resulting in
chloride and amonium
contamination.
The protocol  was modified to
include a  procedure for high-level
samples.  Modifications included a
separate calibration, QCCS, and
dilution procedures  (Hillman et  al.,
1987).
The  buffer was changed to
amonium  acetate/ammonia to
eliminate  the potential ammonium
chloride contamination problem.
None.
Possible improvement  in data
quality.
                                                                        (Continued)
                                                      50

-------
Table 10.  (Continued)
        Finding
                                   Corrective action
                                                                Effect on NSS-I data
Unexpected colored precipitates
(black, brown, purple, yellow,
green) developed with the
aluminum extraction for some
sample including some of the
performance audit samples. In
all pH and  aluminum ranges.

The method that uses flow
injection analysis for the
measurement of total monomeric
aluminum and nonexchangeable
monomeric aluminum was under
development in the processing
laboratory at the time of the
NSS-I.  Numerous hardware and
software problems occured. See
text for further descussion.
None.  The EMSL-LV methods
development group analyzed these
samples for metals that may have
produced such colors, but no
identifiable trend was indicated.
The protocol was mofified with
development of the method.  See
text for further discussion.
                                                             None.
Many samples were analyzed for
these aluminum species weeks
after the batch was processed.
See text for further discussion.
some  field blank and performance audit
samples.   Aliquot 3 is the aliquot in which
anions, including nitrate, are measured in the
analytical laboratory  (Figure 9). During an on-
site  processing laboratory evaluation,  it  was
observed  that  the non-acid-washed filtration
apparatus used for Aliquot  3  was located in
the middle of a series of nitric acid-washed fil-
tration  apparatus  in order to  expedite
processing. This practice ailowed two techni-
cians to filter samples  simultaneously so that
deadlines for shipment to the analytical labora-
tories could be  met.

      It was suspected that the cause of the
nitrate contamination was twofold.  Contami-
nation of  less than  1.00 mg/L was probably
due  to nitric acid splashing  from the acid-
washed units during rinsing steps of the filtra-
tion  procedure.  For higher levels of contamina-
tion, it was suspected that the analysts mis-
takenly preserved  Aliquot  3 with nitric  acid
during the sample preservation  procedure.
More care during the  preservation  procedure
and the  installation  of a Plexiglas shield
around the non-acid-washed filtration
apparatus eliminated the  contamination
                problem.   Of  the 68  field and processing
                laboratory blank samples, 4 are suspected to
                be nitrate contaminated. Of the 68 field audit
                samples, 4 are suspected to be contaminated.

                    Of the 1,381 routine stream samples, only
                one  is  suspected  to be  contaminated.
                Contaminated samples  are qualified with  an
                X3 data qualifier flag  (Appendix  B) in the
                verified and validated data sets.

                Total Monomeric and
                Nonexchangeabl® Monomeric
                Aluminum Measurements

                    The protocol for measuring total  mono-
                meric  and   nonexchangeable  monomeric
                aluminum by flow injection analysis (FIA) for a
                large-scale operation was  under development
                at the time of the NSS-I. The QA plan (Drous<§
                et al., 1986a) specifies that the  QCCS control
                limits for all  aluminum  determinations  in the
                NSS-I must be within +20 percent.  However,
                in the development of the  protocol,  attempts
                were  made  to  determine   if more  stringent
                limits (±10 percent)  could be established for
                these two  aluminum measurements.   With
                                            51

-------
development of the method, it was clear that
the 10 percent limit would be too stringent.
The control limits were successfully widened
to ±15 percent for the total monomeric fraction
and to ±20 percent for the nonexchangeable
fraction  for the remainder of the survey.  The
required frequency of QCCS analysis was also
decreased from every 5 samples to every 10
samples.

      Finally,  because  the  sample
concentration for  these variables often
exceeded the  expected range (0 to 1.50 mg/L)
during the NSS-I, a calibration procedure for
samples that  contained  high concentrations
was included  in the  protocol.  This calibration
pro- cedure included analysis of QCCSs appro-
priate for the concentration range and a
requirement for  duplicate analysis for  each
separate calibration. Further detail is provided
in Arent  et al. (in preparation) and Hillman et al.
(1987).

      There were numerous  hardware and
software problems  with the FIA in the
developmental stage of the protocol. Because
of these problems, several batches of samples
could  not be  analyzed on the day of receipt as
the QA  plan required.   A total of 13 batches
(285   samples) were  refrigerated   and  were
analyzed as  long as  4 weeks  after sample
collection.

      The quality of  data from the backlogged
samples is uncertain for several reasons. The
effects  of  holding time and atmospheric
carbon dioxide exposure on aluminum specia-
tion are  not fully known.  It was necessary to
recycle  syringe valves  back  to  the  field
because  the  ongoing  surveys exhausted the
manufacturer's supply.  Because the syringes
were no longer airtight throughout the holding
time  before analysis and the samples  were
thus exposed  to atmospheric carbon dioxide, a
modification to the protocol was implemented
to hasten analyses.  The samples were filtered
through  syringe filters into  sample  cups and
then were analyzed by using an open-air  auto-
sampler rather than  by direct syringe injection.
A data qualifier and comment indicating that
the backlogged  samples  were  analyzed by
modified analytical protocol or out of  holding
time  protocol were  applied to the variables
(ALDSVLjr, total  (monomeric alumi- num  tag
field, and ALORVL_T,  nonexchange- able
monomeric aluminum  tag field)  in  the raw,
verified, and validated data sets.  The asso-
ciated comments  for these tags are found in
the variable field COMMO5.

Analytical Laboratory
Operations and Protocol
Changes

    Through preliminary evaluation of  the
data, on-site evaluations, and data verification,
the QA program was instrumental in identifying
and resolving several significant problems at
the analytical  laboratories.   Appropriate
changes were incorporated in the verified data
set. The most significant issues are discussed
below.

Effect of Large Sample Loads

    The consistent incoming sample load of
40 samples or more each day was a hardship
on Laboratory 2,  which  accommodated  this
situation  by adding a  second work shift.
Instrument malfunctions,  especially for  the
carbon analyzer,  resulted  in  a  backlog  of
samples which then exceeded sample holding
time requirements for dissolved inorganic
carbon and dissolved organic carbon measure-
ments by a few days.   All measurements for
which the  analyses exceeded sample holding
time allowances are qualified with a  tag in  the
verified and validated data sets. Of the 1,083
samples measured for both initial  dissolved
inorganic carbon  and  equilibrated  dissolved
inorganic carbon at Laboratory 2, 434 samples
(40.1 percent) were analyzed outside  the
holding time requirement.    Of  the 1,079
samples  analyzed for  dissolved  organic
carbon, 398 samples  (36.9 percent) were
analyzed  outside  the  holding  time require-
ments.   Twenty-five of the  initial  dissolved
inorganic carbon measurements and six of  the
equilibrated  dissolved  inorganic carbon
measurements were identified as questionable
during data verification.  A data qualifier flag
(Appendix  B) was applied to these  question-
able values.
                                         52

-------
     Laboratory 1 measured 567 samples for
both initial dissolved organic carbon and equili-
brated dissolved inorganic carbon. Twenty-six
analyses (4.6  percent) exceeded  the holding
time requirements  for  both measurements.
This laboratory measured 568 samples for dis-
solved organic carbon.  Fifteen (2.6 percent) of
these analyses exceeded holding time require-
ments.    Holding  times were exceeded  at
Laboratory 1 due to an error in holding time
calculations and were  not due to instrument
malfunction. None of the measurements made
at Laboratory  1  for these late analyses were
identified as questionable in the veri- fication
process.  No  apparent data quality problem
exists because of these late analyses.

Centrifuge Tubes for Extractable
Aluminum Analyses

     Early in the survey,  four  plastic cen-
trifuge  tubes  for  extractable  aluminum
analyses  (Aliquot  2) were  damaged  during
sample  shipment.    These  tubes were com-
posed of a different material than tubes used
in previous NSWS surveys.  Tests con-  ducted
at  the  processing   laboratory  indicated the
tubes  were  extremely fragile.   By  simply
placing  the tube in  a test-tube  rack, the tube
could crack.   The fragility was thought to be
caused  by the acid-washing procedure used to
clean the tubes. Because it was  not possible
to procure tubes similar to those used in the
previous surveys,  the  processing  laboratory
staff packed the centrifuge tubes in Styrofoam
racks  and  in shipping containers  separate
from the other six aliquots to prevent  further
breakage.  For future surveys, attempts  should
be  made to  procure  less  fragile  centrifuge
tubes.

 Laboratory pH Data

      One of the QA checks that  the auditors
performed during the verification process was
a comparison of the initial pH values recorded
on the ANC and BNC titration data form (Form
13) with  the pH values  recorded on the
analytical data form (Form 11).   On the ANC
and BNC titration data form, the laboratories
were required to report measured  pH  values
and pH values that  were calculated as a result
of applying electrode calibration factors to the
measured values.  The pH values  reported on
the analytical data form and then entered into
the data  base  should be those  that  were
measured and  not calculated.   However,
Laboratory 2 incorrectly  reported  the calcu-
lated  pH values  on the analytical data  form.
The QA staff replaced the calculated values
with the measured pH values in  the verified
data set.  Because the QA program identified
this problem, the data were not affected.

ANC and BNC Recalculations

    As allowed  by  the  contracts,  both
analytical  laboratories  developed their own
software for the calculation of ANC and BNC
and used  the Gran analysis  algorithm des-
cribed in the statement of work  as a guide.
The laboratories submitted the values calcu-
lated  by using their own software, and  these
values were included as part of the raw data
set used for data verification.  During the verifi-
cation process,  certain inconsistencies  in the
values reported  by  the  two laboratories
became apparent.  Further analysis revealed
shortcomings in  both calculation methods
used  by the laboratories;  therefore,  the QA
staff  recalculated  all  ANC and BNC values by
using software prepared at EMSL-LV.   These
recalculations not only corrected the identified
shortcomings in the  software used by  the
analytical  laboratories,  but  also  eliminated
interlaboratory bias that could be  attributed to
the differences in software.

    A new program,  GRANNI.EXE,  made it
possible to do these  recalculations.  This pro-
gram  is a noninteractive Gran analysis pro-
gram that includes a  consistent  point  selec-
tion routine and uses the algorithm given in the
statement of work.

    All ANC and BNC values submitted  by the
laboratories were  recalculated.    The  values
originally submitted  were replaced with the
new values in the verified data set.  Almost all
of the new values  were calculated  using
GRANNI.EXE. This algorithm did fail in certain
cases (poor titration data) and interactive soft-
ware was used when necessary.
                                           53

-------
      After the verified data set was delivered
 to  ORNL,  EMSL-LV  developed  an  improved
 data point selection algorithm. The values for
 the NSS-I data set were recalculated by using
 a new program, GRAN.EXE, and were delivered
 to ORNL after review by ERL-C.   These new
 values are used in  the  analysis  of  the  QA
 results presented in this report and are includ-
 ed in the official verified data set generated by
 ORNL.

 Sample Holding Time and Reanalysis
 for Metals

      The allowed sample holding time for the
 metal analyses (Aliquot 1) in the NSWS was 28
 days.   Sample analyses  within this holding
 time allowed  data bases to be created within
 time frames set by the EPA  Because the EPA-
 recommended holding time  for these metals is
 6 months (U.S. EPA,  1983), and the samples
 are considered stable for that period of time,
 reanalysis during the NSS-I was requested for
 some  metals  even  though the 28-day  limit
 specified in the QA plan had been exceeded.
 All  results   from analyses performed  on
 samples outside sample holding time require-
 ments are qualified  with  an H  in the tag
 variable field  in the verified  and validated data
 sets.

 Calcium Reanalysis-

      During   preliminary  data evaluation,
 histograms and QC  charts were  developed
 with the performance  audit  sample  data.  The
 QA staff compared these histograms and QC
 charts to those  developed  by  laboratories
 involved in other NSWS programs.  A positive
 bias was identified in the analysis  of calcium
 by Laboratory 2 and was traced to a difference
 in nitric acid content of standards (calibration
 and QC) and  survey samples.  The  standards
 prepared by the analytical laboratory contained
 1.25 percent nitric acid and  the samples, after
 preservation in the processing laboratory, con-
 tained approximately 0.15 percent.  The higher
 concentration of nitric acid  in the  standards
 suppressed  the  calcium  analytical signal
 resulting  in a positive bias  of approximately
twenty-five percent in the sample results.  The
 bias was not identified because the NSS-I QC
 standards also were prepared  with the incor-
 rect quantity of acid. The acid concentration in
 the samples was clearly marked on the aliquot
 bottles.  Therefore the analytical laboratory did
 not  follow  protocol when  preparing  its
 standards.   The laboratory was directed  to
 reanalyze all affected samples  (batches 2104
 through  2147, 917 samples) using calibration
 and  QC  standards containing  0.15  percent
 nitric acid. After the reanalyses, no significant
 interlaboratory bias was identified (see Section
 6). All  samples  were  analyzed within the
 6-month  holding time recommended by EPA for
 metals.  The identification of this problem by
 the QA  staff demonstrates  the success  of
 maintaining a QA program in which the use of
 QC charts and performance audit samples is
 standard protocol.

 Total Aluminum Reanalysis--

    During analyses of NSS-I samples, an on-
 site inspection was performed at Laboratory 2.
 During the inspection, the QA staff discovered
 that  laboratory analysts were  using the  pro-
 tocol  for total recoverable  aluminum  rather
 than the  protocol  for total aluminum that was
 required  for the NSS-I analyses.  The method
 for total  recoverable aluminum calls for a less
 rigorous  digestion procedure than the method
 for total  aluminum.   Also,  the total aluminum
 results would be biased  lower than  those
 obtained   by using  the correct  methodology.
 The  laboratory  was directed to  use  the
 designated  digestion procedure to  redigest
 and reanalyze all samples  (42  batches) that
 were  originally analyzed with  the  incorrect
 methodology. The laboratory reanalyzed 1,083
 samples within  the 6-month  holding time
 recommended by EPA for  metals.    Due to
 funding restrictions, an on-site evaluation was
 not performed  at  Laboratory  2 during  the
 analyses  for NSS-I  samples until analyses of
two-thirds  of  the samples had  been
completed.   Therefore,   the request  for
reanalyses involved many samples.  For future
surveys,  thorough on-site evaluations should
be performed early in a survey; follow-up eval-
uations are also recommended.
                                          54

-------
Magnesium Reana.ys.s-

     Fifty-four  samples  (batches  2113 and   for which this value exceeded  the total
2116) were  reanalyzed for magnesium by   aluminum concentration by 0.010 mg/L were
Labo a^y 2 because QC charts indicated con-   reanalyzed for total aluminum after conf ,rma-
trol values slightly outside the 95 percent con-   tion  of  the originally  reported values by the
trol limits. The new results are slightly higher   laboratory. The samples were not  reanayzed
than the original values, and the negative bias   for total extractable alum.num because (1) the
indicated from the original control charts is eli-   seven-day holding time was exceedecI and1 (2)
  .   t d                                     the problem was thought to have occurred in
                                            the digestion procedure for total aluminum.  If
Reanalyses of Nitrate, Sulfate, and       reanalysis did not provide    improved results
Chloride                                  the values for total extractable aluminum and
                                            total aluminum were qualified with an  X1 flag in
      Some  stream  samples  contained very  the verified data set.
hiah  sulfate  concentrations  that  made
analyses of  low-concentration nitrate samples       This practice  of  requesting  reanalyses
difficult with ion chromatography.   Data eval-  using  the above  guidelines  's  probablyr  oo
uation of results from Laboratory 1 established  stringent  considermg  that  the deviation
that if the  samples were  not diluted,  these  allowed by  the  control  limits  for QCCS
high-level sulfate samples  produced chroma-  analyses was ±20 percent for both alum.num
togram peaks  that  overwhelmed the  nitrate  measurements.    For future  surveys,  error
peaks and sometimes the chloride peaks. The  bounds of ±20 percent for each measuremen
result was off-scale sulfate readings that   may be  appropriate  cntena to determine if
 masked  actual nitrate and chloride peaks and   data are unacceptable and  reanalysis is
yielded 0 mg/L values for these analytes.        required.

      This problem was  not identified  during   Data Reporting Errors
 initial  analyses  at  the laboratory.   After  the
 problem  was identified,   the  laboratory was       Data reporting errors that were  ident.fied
 requested to reanalyze  the  10 questionable   during  the survey included variable concen-
 samples for sulfate, nitrate, and chloride after   trations incorrectly reported as 0 mg/L, reagent
 sample  dilution,  although the sample holding   blanks  subtracted in error,  and the  incorrect
 time  was   exceeded.   Because the original   number of decimal places reported.
 analyses yielded  such  poor  results,  it was
 thought  that reanalysis  outside the holding   Total Dissolved Fluoride-
 time requirement would  be an improvement
 over the zero  values originally reported  and       Total dissolved fluoride is determined by
 would provide more information to the users,   the ion-selective electrode technique.  The con-
 All data resulting from the reanalysis of these   tracts  awarded to the laboratories suggest the
 samples are qualified by both H  and  R tags   use of a  digital potentiometer with  expanded
 (Appendix B) in the verified and validated data   mV scale capable  of  reading   in 0.1  mV
 gjtg                                        increments.  The required  detection limit for
                                             fluoride was 0.005 mg/L.
  Total Extractable Aluminum Values                                 ..„•.*
  Greater than Total Aluminum Values          For  the determination  of  fluoride,
                                              Laboratory 2 used an instrument  that did not
       During the data verification process,   have the capability of measuring low-concen-
 both analytical laboratories identified  several   tration samples (less than 0.010  mg/L) if call-
  samples for which the  total  extractable   brated for  higher  concentration  samples
  aluminum concentration was  greater than the   (greater than 0.010 mg/L).  The laboratory did
  total aluminum  concentration.   Samples for   not  recalibrate or  use  two  different
                                            55

-------
 instruments, one for high concentrations and
 one for low concentrations.   Because sample
 concentrations below 0.010 mg/L could not be
 detected, all values below this threshold con-
 centration were recorded as 0.000 mg/L.

      During QA data analyses,  the QA staff
 observed that Laboratory 2 had reported
 fluoride values for all field and laboratory blank
 samples  and for stream  samples  that were
 less than 0.010 mg/L as zero.   Although the
 laboratory consistently  reported a  calculated
 instrument detection limit (IDL)  of less than
 0.005  mg/L,  which  is  within  contract
 specifications, the laboratory did not follow the
 contract guidelines in determining the IDL The
 laboratory calculated the  IDL as three times
 the standard deviation of ten nonconsecutive
 low-concentration standards  which  were
 greater than  0.010  mg/L,   instead of using
 values for laboratory blank samples for these
 calculations.  In the verified and validated data
 sets (Data Sets 2  and 3),  an M1 flag was
 applied to all zero values for fluoride reported
 by Laboratory 2.  This  flag indicates that the
 value was not actually  measured and may be
 inaccurate.

      Preliminary QA data analyses using the
 raw data  set, prior to data verification,  would
 have  indicated  problems early enough  to
 permit fluoride reanalyses. For future surveys,
 preliminary evaluations  should  be  a project
 priority.

 Total Dissolved Phosphorus,
 Ammonium, and  Silica-

     During QA data analyses, the QA staff
 discovered several  instances where values
 were misreported by Laboratory 2.  The labora-
 tory had reported the theoretical value, 0 mg/L,
 for  all  total  dissolved  phosphorous, ammo-
 nium,  and silica calibration blanks rather than
the measured value.

     The inspection of raw data also revealed
that  Laboratory  2  reported  total  dissolved
phosphorous, ammonium,  and silica  concen-
trations that originally were measured as
negative during analyses as 0 mg/L.  This error
affected field  blank and stream sample values.
 The laboratory submitted corrected values
 which are included  in the verified, validated,
 and enhanced data  sets and are used in the
 QA data analyses.

 Subtraction of Values for Silica and
 Ammonium Reagent Blanks-

     During  QA analysis, the  QA staff
 discovered that silica and ammonium reagent
 blank values originally were subtracted from all
 sample  concentrations  measured  by
 Laboratory 2.   Because this practice did  not
 follow protocol, values for reagent blanks for
 both  variables  were  added to  the  reported
 values.  The corrected values are  included in
 the official verified,  validated, and enhanced
 data sets.

 Decimal Place Reporting-

    The initial contracts with both laboratories
 recommended that values be reported to  the
 number of decimal places in the  instrument
 detection limit (IDL).  The second  contract with
 Laboratory 2  recommended  that  values  be
 reported to the IDL, plus one, or to a maximum
 of four significant figures (Table 11).

    Laboratory 1 consistently provided values
 with more decimal places than recommended
 in the first contract.  Laboratory 2 consistently
 reported values as they do in their standard
 laboratory procedure, that is, to the number of
 significant figures that they considered mean-
 ingful for that concentration of analyte (usually
 to three significant figures).   Although data
 interpretation  and  population estimates
 (Kaufmann et al., 1988) were not  affected, this
 inconsistency created difficulty  in  the statis-
 tical  analysis of QA data.  This inconsistency
 could be prevented if future contracts "require"
 rather than "recommend" the  number of deci-
 mal places to be reported.

 Data Verification Activities

    The QA  staff  reviewed field  and
processing laboratory  forms and analytical
data packages to identify and to correct data
reporting errors, to evaluate data trends, and
to identify  which  samples needed reanalysis.
                                          56

-------
Table 11.  Recommended Number of Decimal Place*
Variable
   Original
contracts for
 laboratories
   1 and 2
  New
 contract
  for
laboratory 2
Acid-neutralizing capacity       1
Aluminum, total               3
Aluminum, total extractable      3
Ammonium                   2
Base-neutralizing capacity       1
Calcium                     2
Chloride                     2
Dissolved inorganic
  carbon, equilibrated          2
Dissolved inorganic
  carbon, initial               2
Dissolved organic carbon       1
Fluoride                     3
Iron                        2
Magnesium                   2
Manganese                   2
pH, equilibrated               2
pH, initial acid-
  neutralizing capacity         2
pH, initial base-
  neutralizing capacity         2
Phosphorus, total dissolved      3
Potassium                   2
Silica                       2
Sodium                     2
Specific conductance,
  analytical laboratory         1
Sulfate                     2
Nitrate                     3
                1
                4
                4
                3
                1
                3
                3
                3
                2
                4
                3
                3
                3
                2
                2
                4
                3
                3
                3


                1
                3
                4
    All forms used in the NSS-I are given in the QA
    plan  (Drous6  et al.,  1986a).    Any required
    changes to the data on these forms resulted in
    changes  to  the raw data set  and  were
    reflected  in the verified data set.   The  types
    and numbers of changes made to create the
    verified data set are given in Table 12.

    Review of Field Data Forms

         The review of field  data forms and the
    subsequent additions or corrections made to
them required a great deal of time and effort
during  the NSS-I.  Due to sample shipment
deadlines,  the base coordinators made only
cursory   review  of the stream data  forms
(Form  4)  before  packing and  shipping them
with the  samples.   When the field  crews
shipped the  samples from a remote  site,  it
was not possible for the base coordinator to
review the forms. The hydrology and site char-
acteristic forms (Forms 4A, 6, and 7) were not
sent to the QA staff until the base coordinator
reviewed them after sample shipment.  In the
beginning of  the  survey,  this  process took
several days.

    When the base coordinator found errors
on the forms after they had been sent to the
QA staff, the changes were submitted over the
telephone, and were followed  by hard copy
documentation.   If the ORNL  staff had not
already entered the original data, the changes
were forwarded to ORNL for entry.  If the data
had been entered into the raw data set, any
changes were made to a copy of the raw data
set by the EMSL-LV QA staff in order to create
the verified data set.  The concept of  a "raw
data  set" was  that no  changes  would be
applied after  the  data were  entered.   All
changes  after initial  data entry are made  to
subsequent data sets.

     There were two significant problems with
this system.   The hydrology  and site  des-
cription forms were received by the QA staff
piecemeal, and there  were numerous changes
to the forms after they  were shipped from the
field.  In the future, more emphasis on correct
data form completion would minimize changes.
The number  of  NSS-I  forms that the base
coordinators  were  required  to review  took
 more time than was available.  Designating  a
fieid member to assist the base coordinator in
 reviewing forms  may expedite the  review pro-
cess  and at  least make it possible  for  all
forms to be shipped by the following day.

 Review of Processing Laboratory
 Forms

     The  magnitude of the  NSS-I  sample
 processing effort, in conjunction with the
 concurrent sample processing for the Eastern
                                               57

-------
Table 12.  Changes to Sample Numeric Data Incorporated In the Verified Data Set, National Stream
         Survey - Phase I
Data source
(form number)
Field Data Forms
(Form 4)6
Number of
changes made
from raw to veri-
fied data sets*
2 (5,618)
Percent of
changes to
total values
from data source
<0.1
Comments
Most changes to the field
forms were not numerical.
 Processing
  Laboratory Forms
  (Form 5)^

 Analytical
  Laboratory
  Stream and QA
  Data Forms
  (Form 11) °

 Total
 1.600  (10.831)
10,613  (56,019)
14.8
                            22.8
19.0
Most of these changes are
the result of temperature
corrections for Cond-PL.

Most of these changes
resulted from corrections
to Al-total, ANC, BNC, Ca,
pH-ANC, and pH-BNC.
 3 Number in parentheses is the number of numerical observations in the verified data set.
 b Changes to flags, tags, and QC data are not included in this table.
Lake Survey - Phase II, made daily form com-
pletion difficult  in  the beginning of these
surveys.  A backlog of forms developed at the
processing laboratory before they were avail-
able for the QA staff review.  After the proces-
sing laboratory was operating more efficiently,
the forms were delivered daily for QA  review.

     The processing laboratory analytical
data form (Form  5)  was completed correctly
most  of the time.    It was  less difficult to
correct these  forms because the processing
laboratory and the QA staff were both located
in Las Vegas.

Review of Analytic®! Data Forms and
Correction of Data

     Review and  verification of the sample
data packages  submitted by the analytical
laboratories was a bigger task than review of
the field and processing  laboratory data.  The
QA staff  always  requested confirmation  of
suspect data before reanalysis was requested.
After confirmation was requested, a response
usually took two  to  five weeks.   Due to the
reanalyses requested  for calcium  and  total
aluminum, Laboratory 2 required three months
to  complete  the  task.  These analyses were
performed  within  the  6-rnonth holding  time
                      recommended by EPA.  There was no specific
                      requirement for response time in the contracts
                      for NSS-I.  The contracts required a response
                      within a "reasonable"  amount  of  time.    A
                      prompt response to the reanalysis request
                      was necessary to meet sample holding time
                      requirements.  All  reanalyzed sample values
                      incorporated in the verified  data  set are
                      qualified with an R tag (Appendix B) in the
                      variable tag field.  All changes to the analytical
                      laboratory data were documented on the
                      Confirmation/Reanalysis Request Form,  Form
                      26  (Drouse et al., 1986a),  or  in  revised data
                      packages submitted by the laboratories.

                      Changes to Analytical Data Applied at
                      EMSL-LV

                          All hydrology and site description changes
                      for the verified data set were  made at ORNL,
                      and all changes  to the analytical data  were
                      madeby the EMSL-LV QA staff. These changes
                      originated from all three data sources:  field
                      forms, processing laboratory  forms, and
                      analytical laboratory forms.  The EMSL-LV QA
                      staff made changes using transaction records
                      that were applied to  a  copy of  the raw data
                      set. The ORNL staff made changes by editing
                      directly into a copy of the raw data set using
                      the SAS full-screen edit facility.  Each change
                                           58

-------
went through a series of checks before it was
considered final.  The changes were entered
from the modification sheets (Drous6 et a!.,
1986a) or  from the revised analytical  data
packages by a data entry technician.  A dif-
ferent technician checked the values for
accuracy before moving them into the changed
data set.  After the update of the transaction
records,  the  changed data set was checked
point  by point  to  confirm that the intended
changes were made correctly.  The changes
consisted of sub-stituting  correct  values or
adding data  qualifiers.   All changes to the
analytical data  and flags from the raw to the
verified data  set are documented in a history
file of changes that was included with the
verified data set on the magnetic tape sent to
ORNL

Modifications to the Exception-
Generating Programs and New Data
Qualifier Flags

      The  data qualifier  flags used in the
NSS-I (Appendix B) were similar to those  used
in all the  NSWS surveys with a few modifi-
cations:

1.    For the anion and cation balance check
      program, an A9  flag  was used to
      indicate a possible analytical error with
      the ANC measurement.

2.    For the conductance balance program,
      the original NSWS C7 qualifier indicated
      a conductance  imbalance  due to
      unmeasured protolyte  anions.  For the
      NSS-I,  the definition of the C7 flag was
      changed to indicate an imbalance due to
      the influence of other anions and cations
      that are not included in the conductance
      balance calculation.  In addition, a  new
      flag, F6,  was created for the NSS-I to
      indicate a problem with  the processing
      laboratory  specific  conductance
      measurement.

3.    For comparison of field and processing
      laboratory data in the protolyte analysis
      program,  changes were made to the
      definitions of the original NSWS F  flags
      to reflect the difference  in stream field
    instrumentation from that used during
    lake surveys.

4.   An M1 flag was created for the NSS-I and
    applied to all fluoride samples measured
    by Laboratory 2 that were  reported as
    zero.  This flag indicates that the value
    was not actually measured and therefore
    may be inaccurate.

5.   Additional miscellaneous X flags were
    used in the NSS-I.  The  flag,  X3, indicates
    a potential gross  contamination  of  the
    aliquot.  The X7 flags were added to the
    flag field of the sample identification
    number  to indicate a  site  disturbance,
    such as a strip mine or  sewage treatment
    plant, in the watershed.

Delivery of Verification Tapes

    The  original  intention  of  the  NSS-I
verification and  validation  process was  to
deliver only  two  verification  data tapes  to
ORNL: the  first with numerical changes and
the  final with  the  data qualifier  flags.
However,  due to the magnitude  of the value
changes resulting from reanalyses, three data
tapes were delivered.   It was not possible to
include all reanalyzed data  in  an intermediate
tape  needed  by  the  ERL-C  staff  for  data
assessment and validation issues.  Therefore,
the value changes  submitted by the analytical
laboratories  up to that time were included in
the intermediate data tape,  and the remainder
of the value changes were included in the fina!
verified data  set with the data qualifier flags
delivered to ORNL on magnetic tape.  At ORNL,
the tape  was checked  for consistency before
the changes to  the nonanalytical data  were
made. ORNL then created the official  verified
data set that was used in data validation.

Data Base Audit

    At the  conclusion of the  verification
process, a data base audit  was performed by
an independent  organization.   The audit
consisted of reviewing the verification records,
evaluating for accuracy the results generated
by AQUARIUS  II and  other computer
programs, reviewing the procedures used to
                                          59

-------
substitute for missing values, and determining
the error rates associated with each aspect of
the verification procedure.  No incorrect value
changes were detected in the verified data set
and all  value changes were well documented
(Grosser and Pollack, in preparation).
                                           60

-------
                                     Section 6
                          Assessment of Data Quality
Introduction

     The quality assurance program  of  the
National Stream Survey - Phase I (NSS-I) was
successful  in reducing  to acceptable levels
errors  associated  with  the acquisition and
subsequent reporting of data.   The program
was also successful in identifying and correct-
ing potential problems related to data quality
that occurred over the course of the NSS-I.

     One purpose of the QA program was to
determine  if any  corrective actions (e.g.,
reanalyses,  qualifying  unacceptable  values)
would  improve  the quality of  the analytical
data and, if so, to implement those actions.
The second purpose of the assessments was
to  identify  possible  limitations  of the  data
base  that  might affect  data  interpretation.
This second  purpose was accomplished  by
viewing the data  in  terms  of  repre-
sentativeness,  completeness,   comparability,
detectability, accuracy, and precision.

     The six aspects of NSS-I data quality fal!
essentially  into two groups.  Completeness,
representativeness, and comparability apply to
the sampling design and to the verified data
set.   Detectability, accuracy,  and precision
quantify  the  performance of  one  or  several
components of the collection and  measure-
ment system.  These properties are evaluated
by comparing the data acquired from analysis
of QA  samples to the established data quality
objectives.  Data quality objectives for detec-
tability, accuracy, and precision  are presented
in Table 13. The values given are performance
targets that the analytical  laboratories were
expected to meet.  Objectives were not estab-
lished  for field measurements, although these
measurements were subject to QC protocols
(see Section  4).  For most variables, within-
laboratory precision  goals  were established
only for  measured  values  greater  than  ten
times the value of the detection limit objective
(Drous6 et a!., 1986a). For other variables (e.g.,
total monomeric aluminum), objectives were
set for specified ranges.

     Some evaluations of detectability, accu-
racy,   and  precision were  improved  by  the
elimination  of  a small  number of  extreme
values that were considered outliers.  When-
ever outliers  were removed for a  particular
assessment, that fact is included in the appro-
priate text discussion. The removal of outlying
values sometimes resulted in a  difference be-
tween the  number  of  samples collected or
processed and the number of measured values
for a particular analyte.

Completeness

     Trie completeness of the NSS-I data base
was a critical  aspect of data quality. If an
insufficient number of streams  were sampled
or if a large number of analytical results were
invalid, the  representativeness  and com-
parability of  the NSS-I  data base could be
compromised.   The DQO  for  completeness
was established as 90 percent before the start
of the NSS-I.  That  is, 90 percent or more of
the  streams  initially selected  for sampling
were expected to yield data that would meet
QA criteria  and that  could thus be used for
estimating the number of stream reaches  with
a particular chemical characterization of inter-
est (e.g., measured ANC less than Oyueq/L).

     Completeness  of the  data  base  was
evaluated based on  the  overall number of
stream reaches from which samples were col-
lected and on  the  percentage  of  acceptable
data generated from these samples. A total of
                                          61

-------
 Table 13.     Analytical Data Quality Objectives For Detectablllty, Precision, and Accuracy For The
              National Stream Survey - Phase I
 Variable (units)
Detection
  limit
objective
 (units)
  Within-
laboratory
 precision
  (%RSD)*
                                                                                         Within-
                                                                                       laboratory
                                                                                      accuracy  (%)
                                                    FIELD SITE
 pH, field (pH units)

 Specific conductance
 Dissolved oxygen
   (mg/L)

 Current velocity
   (m/s)
                                                         ±o.r
                                             PROCESSING LABORATORY
Aluminum (mg/L)
  Total monomeric
  Nonexchangeable
     monomeric

Specific conductance
  (/jS/cm)

pH, closed system
  (pH units)

Dissolved inorganic
  carbon, closed system
  (mg/L)

True color (PCU)

Turbidity (NTU)
                                  0.01
                                  0.01
                       10 (>0.01 mg/L)
                       20 (£0.01 mg/L)

                       10 (>0.01 mg/L)
                       20 (S0.01 mg/L)
 0.05

 0

 2
  0.1£



 10

  5*

 10
                           10 (>0.01  mg/L)
                           20 (S0.01  mg/L)

                           10 (>0.01  mg/L)
                           20 (<0.01  mg/L)
                                                                                         ±0.1£
                                                                                         10
                                                                                         10
                                            ANALYTICAL LABORATORY
Acid-neutralizing
  capacity
Aluminum  (mg/L)

     Total extractabla
                                 0.005
                                                           10
                                                       10 (>0.01 mg/L)
                                                       20 (<0.01 mg/L)
                                                                                         10
                                                    10 (>0.01 mg/L)
                                                    20 (<0.01 mg/L)
                                                                                             (Continued)
                                                   62

-------
Table 13.   (Continued)
Variable (units)
Ammonium (mg/L)
Base-neutralizing
capacity (/jeq/L)
Calcium (mg/L)
Chloride (mg/L)
Specific conductance
(/jS/cm)
Dissolved inorganic
carbon (mg/L)
Initial
Equilibrated
Dissolved organic
carbon (mg/L)
Fluoride, total
dissolved (mg/L)
Iron (mg/L)
Magnesium (mg/L)
Manganese (mg/L)
Nitrate (mg/L)
Detection
limit
objective
(units)
0.01
d
0.01
0.01

c


0.05
0.05
0.1


0.005
0.01
0.01
0.01
0.005
Within-
laboratory
precision
(%RSD)*
5
10
5
5

2


10
10
5 (>5.0 mg/L)
10 (£5.0 mg/L)

5
10
5
10
10
Within-
laboratory
accuracy (%)
10
10
10
10

5


10
10
10


10
10
10
10
10
 pH (pH units)

   Equilibrated

   Initial ANC

   Initial BNC

 Phosphorus, total
   dissolved  (mg/L)

 Potassium (mg/L)
0.002
0.01
     0.05°

     0.05*

     0.05*

10 (>0.010 mg/L)
20 (sO.010 mg/L)
    ±0.1°

    ±0.1*

    ±0.1d

10 (>0.010 mg/L)
20 (=£0.010 mg/L)
                                   10
                                                                                                (Continued)
                                                     63

-------
  Table 13.   (Continued)
  Variable (units)

  Silica (mg/L)

  Sodium (mg/L)

  Sulfate (mg/L)
Detection
  limit
objective
 (units)

  0.05

  0.01

  0.05
Within-
laboratory
precision
(%RSD)*
5
5
5

Within-
laboratory
accuracy (%)
10
10
10
   mn.antra..           u           	  Unless otherwise noted, this is the precision goal at
   concentrations greater than or equal to 10 times the required detection limit
   Precision or accuracy goal in terms of applicable units.

   ~  I?™! SlX,."0n!0nSe!UtiVe b'ank measurements must not exceed 0.9 pS/cm.
                                                         was required to be less than or equal
 479 stream reaches were initially selected for
 sampling in the mid-Atlantic and  southeast
 screening  regions.   In addition to  those
 reaches selected for inclusion in the probability
 sample, the total included a number of reaches
 on "special interest" streams where research
 programs  independent  of the NSS-I were  in
 progress.

      Table 14 presents the number of streams
 by region from which samples were collected.
 Of all reaches individually  identified for sam-
 pling,  samples were collected from  429 (90
 percent) at both upstream  and downstream
 sites on  every visit.    Of  the  1,406  visits
 scheduled to upstream and downstream sites,
 water samples were collected from  1,328 sites
 (95 percent). Only eight of the sites could not
 be sampled because of  access permission dif-
 ficulties or because of physical inaccessibility.
 Water samples were not collected from  the
 other 70 sites because they were classified as
 nontarget reaches as  specified  in  the  NSS-I
 sampling design (e.g., the stream sites were
 influenced  by salt  water or  no  water was
 present in the streambed).

      In the mid-Atlantic region, nine streams
 were  not  sampled:   two streams  were  not
sampled because of tidal  influence, four
streams were  not  sampled  because  the
specific conductance  measurement exceeded
                  the 500 fjs/cm criteria, and three streams were
                  dry.  Five streams were partially sampled:  two
                  streams   were  sampled at  only  one  site
                  because of tidal influence,  one  stream was
                  sampled at only one  site  because of the high
                  conductance measurement at one of its sites,
                  and  two streams were sampled at only one
                  site  because  access  permission  was not
                  obtained for the other site.   In the southeast
                  screening region, twenty-three  streams  were
                  not sampled: 20  streams that  might normally
                  be flowing during the spring were completely
                  dry or stagnant, one stream  was not sampled
                  because it was inaccessible as a  result of
                  hazardous  conditions, one  stream  was  not
                  sampled because access  permission was not
                 obtained,  and one  stream was not sampled
                 because it was inundated by a major  water
                 project.  Thirteen  streams were only sampled
                 at one site:  twelve streams were sampled at
                 only  one site because more than 90 percent of
                 the reach  was dry  and one stream  was
                 sampled at only one site because access per-
                 mission was denied at the  other site.

                      Reported values are given for all physical
                 and chemical variables for  1,613 (97.7 percent)
                 of the 1,651 stream  samples and QA samples
                 listed in the NSS-I verified  data  set.  Of these
                 1,613 samples,  the  verification  process  Iden-
                 tified 97 samples  (6 percent) with values for
                 one or more variables that should be used only
                                           64

-------
Table 14.   Summary of Streams Visited During the National Stream Survey - Phase I

Region
Mid-Atlantic
Southeast
screening
Total
Total
streams
targeted
(includes
special
interest)
276
203
479

Total
special
interest
streams
26
_3
29

Total
streams
sampled0
267
180
447

Number of
streams
sampled at
only one site
5
13
18

Number of
streams
not
sampled
9
23
32
 a Includes streams sampled at only one site.
 b Missing upper or lower sites on one or both visits.
 with caution.  These values were qualified with
 an M1, XO, X1, or X3 flag (Appendix B) in the
 verified and validated data sets because of the
 likelihood of a contami- nation or an analytical
 method problem.  During  the data validation
 process and the creation of the enhanced data
 set,  missing or unacceptable values were re-
 placed with  the values from field duplicate
 samples (if available) or by an estimate com-
 puted by one of several approaches described
 by Kaufmann et  al. (in press).  Not all values
 identified  as suspect during  the verification
 process were replaced in the enhanced data
 set because these values  may not have  been
 identified  as statistical  outliers on the  sub-
 regional level during validation.

       Overall, the completeness of the data
 base exceeded the DQO of 90 percent, and the
 representativeness and comparability of the
 NSS-I data base were not affected by incom-
 pleteness. The data base is sufficiently com-
 plete to  provide  representative  spring  base
 flow chemical indices with which to estimate
 the chemical status of streams in these areas.

 Comparability

        For the  NSS-I,  the confidence  in the
 comparability (or  compatibility)  of  data from
 samples collected and analyzed by many dif-
 ferent  individuals  and   organizations was
 maximized  by  the use  of  standardized
 protocols for sample collection,  processing,
and measurement.  When the QA staff mem-
bers identified deviations from the protocols
during on-site evaluations or during  daily data
verification activities (see Section 4), prompt
corrective  actions  helped  to   improve  the
comparability of data  within the NSS-I data
base.  The comparability of NSS-I data could
also be affected by systematic differences in
performance    between  the participating
laboratories (interlaboratory bias).  Inter-
laboratory bias is  evaluated as  part of  the
assessments of accuracy and precision.

     In addition, the NSS-I data base is com-
parable to data  bases from other AERP
programs, a critical objective of the NAPAP.
The use and documentation of standard sam-
pling and analytical  methodologies and  the
large volume of QA and QC data present in the
verified data set allow quantitative evaluations
of data comparability  to past and future
studies.  Data  from analyses of  performance
audit samples  from laboratories participating
in the NSS-I and  other  NSWS programs are
presented in Appendix A. These data may be
useful in evaluating the comparability of the
NSS-I  data base  to  those of other NSWS
projects.  In addition, data  from the  special
interest streams are included in the NSS-I data
base.  These data will be useful in classifying
these sites. The data are also potentially  use-
ful in making regional extrapolations based on
information gained from  intensive study at
these sites.
                                             65

-------
 Representativeness

       The statistical frame for NSS-I sampling
 was designed to ensure that analytical results
 would  represent the stream  chemical  condi-
 tions in the subregions sampled. Standardized
 protocols defined  the  appropriate weather
 conditions for sampling activities, the criteria
 for selecting a sampling site  on a reach, and
 the criteria for selecting a sampling location in
 the  stream (Hagley et  al.,  in press).   These
 protocols helped to ensure that each sample
 collected was representative  of  spring  base
 flow  chemical conditions existing in the stream
 at the time of sampling.

 Detectabflity

      Two  aspects of detectability were
 assessed for the NSS-I.  Laboratory perfor-
 mance  was assessed by estimating the  mini-
 mum limit  of  detection for  each  analytical
 method  except  for  pH,  color, and turbidity.
 This  "method-level"  limit of  detection repre-
 sented  the  smallest quantity of a  chemical
 variable  that a method (or  instrument)  could
 measure reliably. The second aspect of detec-
 tability  assessed  was "background" or  the
 quantity of a chemical variable that was intro-
 duced into streamwater samples during  their
 collection,  handling, and preparation for
 analysis.  The assessment of background is
 especially important to data interpretation.
 Background quantities serve  as decision
 points; they are the lowest  concentration of a
 given chemical variable that can be identified
 (with specified statistical confidence) as
 having been present in streamwater samples
 at the time  of collection.   The estimation  of
 background represents  a  "system-level"
 assessment of detectability.

 Assessment of Method-Level Limits of
 Detection

      Numerous operational definitions  and
computational approaches exist for estimating
detection limits (Currie, 1968; Hubaux and Vos,
 1970;  American Chemical Society, 1980; Glaser
et  al., 1981; Keith  et al.,   1983;  Long  and
Winefordner,  1983; Oppenheimer et al., 1983;
Clayton   et   al.,   1987).  The data  quality
 objectives for detection  established  for the
 NSS-I were  based on the "limit of detection"
 advocated by the American  Chemical Society
 (American Chemical Society,  1980; Keith et al.,
 1983). The limit of detection is defined as 3s_,
 where SQ represents the standard deviation at
 the lowest level of measurement (Taylor, 1987),
 which is usually zero.  This expression of the
 limit of detection does not specify the probabil-
 ity of falsely concluding that  a  chemical  vari-
 able  is absent (termed a false negative,  /3, or
 Type II error; Clayton et al., 1987).  Specifying
 3sQ provides Type I and Type  II error rates of
 approximately 7 percent each.  This limitation
 is  not  critical  when assessing NSS-I  data
 quality, although it may be important in  work
 that tests for the presence (or absence)  of a
 toxic substance.

     For the  NSS-I,  SQ  was  estimated  from
 laboratory blank samples (i.e., calibration
 blanks or reagent blanks) and from field blank
 samples, rather than from the analyses of low
 concentration standard solutions.  The use of
 blanks is advocated by Campbell and Scott
 (1985) and Hunt and Wilson (1986), while the
 use of  low-level  standards is advocated by
 Taylor (1987).  When the limit of detection is
 estimated from the analyses of blank samples,
 it is operationally similar to an analytical or an
 instrument detection limit (Keith et al.,  1983;
 Taylor,  1987) because  the samples  are  only
 prepared for  analysis and are not subjected to
 collection or processing.

    During the course of the NSS-I, personnel
 at the processing and analytical laboratories
 calculated detection limits weekly and reported
 them  to the  QA staff at EMSL in Las Vegas.
 These limits  were based on  the  analyses of
 either laboratory  blank samples or detection
 limit quality control check samples  (Table 2).
 At the processing  laboratory,  detection limits
 were calculated for total monomeric and non-
 exchangeable monomeric  aluminum  and  for
 closed-system  DIG.  The detection limits
 reported by the  laboratories met the  require-
 ment that  they be less than the detection  limit
objectives  (Table 13). For this report method-
level limits of  detection were calculated to con-
firm the reported detection limits. The results
are presented in the following sections.
                                          68

-------
Limits of Detection Based on
Laboratory Blank Samples-

     Laboratory 1  and Laboratory 2 analyzed
26 and 42 laboratory blank samples, respec-
tively. Laboratory blanks were not used to cal-
culate limits of detection for ANC, BNC, or pH
measurements.    Laboratory blank  measure-
ment data for ANC and BNC were not reported
on  the standard  reporting form  for QC
samples in the analytical data package (Form
20) but were included in the titration data files
for ANC and BNC.  The titration data are not
included in the verified data set.  The process-
ing laboratory analyzed 68 laboratory blanks
for monomeric aluminum, 66 for nonexchange-
able monomeric  aluminum,  58 for closed-
system  DIG, and 46 for specific conductance.
During the first half of the  NSS-I, specific con-
ductance  measurements  were  made  on
laboratory  blank  samples  for only  seven
batches. On several occasions, closed-system
DIG measurements for more than one batch
were made  using a single calibration.  There-
fore, one laboratory blank may have been used
in the measurement of more than one batch of
samples.

      Summary statistics  and limits of detec-
tion based on  laboratory  blank sample
measurements are presented in Table 15. For
measurements   made   at the  analytical
laboratories, values qualified with  an  X flag
(Appendix  B)  were  excluded from detection
limit calculations.  Measured values from the
processing  laboratory that were identified as
statistical outliers  (Grubbs1  test,  p <. 0.05;
Grubbs,  1969) were excluded from the detec-
tion limit calculations.   Laboratory blank
measurements  made in  the  processing
laboratory for total monomeric aluminum, non-
exchangeable monomeric  aluminum, specific
conductance, and  closed-system DIG are not
included In the NSS-I data base and thus were
not subject to the same verification procedures
as analytical laboratory blanks.

      Laboratory  /--Limits of detection  esti-
mated  from laboratory blank samples were
within the detection limit objective for all vari-
ables (Table 15).  Mean values for all variables
were at or  very near to zero (Table 15).  All
values for chloride, specific conductance, DIG,
sodium, and  nitrate were  reported as zero,
indicating that a low-level QCCS may be more
suitable than  a laboratory blank sample to cal-
culate an instrumental detection limit for these
variables.

    Laboratory ^-Limits  of  detection  were
less than or near to the detection limit objec-
tive for all variables  except  total  aluminum
(0.017 mg/L).  In addition, the limit of detection
for fluoride was estimated   as 0.010  mg/L,
because laboratory personnel  did not calibrate
the instrument to measure concentrations less
than 0.010 mg/L (see Section 5).  Mean values
of laboratory blank measurements were at or
near  zero for  all variables  except  total
aluminum (0.010 mg/L) and silica (0.04 mg/L),
indicating the possibility of sporadic reagent
contamination or calibration bias.

    During data verification, the QA staff set
control limits  for  the reported  values of
laboratory  blank  measurements.   The  lower
control  limit was established as the negative
value and the upper control limit as twice the
value of the  detection limit objective for each
variable as required by the contracts with the
analytical laboratories. The lower control limit
allowed for minor fluctuations in  instrument
performance.   Of the 42 total  aluminum
measurements, 15 were greater than twice the
detection limit  objective, which  resulted in  a
large mean value and a large standard devia-
tion.

     For silica measurements, only two values
were greater  than   twice the detection limit
objective.  Altogether  the  distribution of silica
measurements  included four values (both posi-
tive and negative) that were statistical outliers
(Grubbs'  test,  p <  0.05;  Grubbs,  1969).  It
appears that silica measurements  in a small
number of batches may have  been affected by
low-level reagent contamination  or  a negative
calibration bias.

     Processing Laboratory-Tine limit of detec-
tion for closed-system DIG measurements
(0.03 mg/L) was less than the detection limit
objective, and the mean value (0.02 mg/L) was
near  zero  (Table 15).  For  total monomeric
                                           67

-------
  Table 15.
Estimates of Umlts of Detection Based on Analyse, of Laboratory Blank Sample*, National
Steam Survey - Phase I
                                                             Analytical laboratories4
Laboratory '
Variable
Al-ext
Ai-total
Ca
cr
Cond-lab
DICf
DOC
F-
Fe
K
Mg
Mn
Na
NH4+
N03-
P
SiO2
S042-
Units
mg/L
mg/L
mg/L
mg/L
pS/cm
mg/L
mg/L
mg/L
mg/L
mg/L
mg/L
mg/L
mg/L
mg/L
mg/L
mg/L
mg/L
mg/L
Detection
limit
objective
0.005
0.005
0.01
0.01
X <0.9
0.05
0.1
0.005
0.01
0.01
0.01
0.01
0.01
0.01
0.005
0.002
0.05
0.05
n
26
26
26
26
26
26
26
26
26
26
26
26
26
26
26
26
26
26
Mean
<0.001
0.002
<0.01
0.00d
0.00rf
0.00*
<0.1
0.002
<0.01
<0.01
<0.01
<0.01
0.00*
<0.01
o.ooorf
<0.001
<0.01
<0.01
s»
_
0.0017
—
0.00
0.00
0.000
-
0.0007
—
..
-
—
0.000
-
0.000
_
—
-
I
Estimated
limit of
detect iontf
<0.001
0.005
0.01
0.00*
__«
0.00<*
<0.1
0.002
0.003
<0.01
<0.01
<0.01
o.oorf
<0.01
o.ooorf
0.001
<0.01
<0.01
Laboratory 2
n
42
42
42
42
42
42
42
42
42
42
42
42
42
42
42
42
38
42
Mean
0.001
0.010
<0.01
<0.01
0.6
<0.01
<0.1
0.000^
<0.01
<0.01'
<0.01
<0.01
<0.01
0.00
<0.001
<0.01
0.04
<0.01
S*
0.0007
0.0057


0.20


0.0000





0.001


0.022

Estimated
limit of
detection0
0.002
0.017
0.02
0.01
_JB
0.07
0.1
0.010^
<0.01
0.02
0.01
<0.01
0.01
0.003
0.009
0.001
0.07
0.04
                                           Processing laboratory*



Al-mono
Al-nex
Cond-PL"
Cond-PL7
DIC-closed



mg/L
mg/L
pS/cm
/uS/cm
mg/L



0.010
0.010
X <0.9
X <0.9
0.05


n
65
65
7
39
58


Mean
0.009
0.014
5.6
0.6
0.02


s"
0.0047
0.0064
1.48
0.27
0.011
Estimated
limit of
detect ionff
0.012
0.019
_»
__«
0.03
                 	.—. _    standard deviation.
  Dashes indicate that the standard deviation is nearly zero.
0 Estimated limit of detection - 3s.
d All measurements reported as zero by laboratory.
* Detection limit objective expressed as the mean of the  blank sample measurements
  DIG detection limits apply to both equilibrated and initial measurements.
a Laboratory reported all concentrations less than 0.010 mg/L as zero.  Limit of detection set at 0.010.
  Values for blanks measured in the first 28 batches.
  Values for blanks measured in the remaining 40 batches.
                                                   68

-------
aluminum,  three statistical  outliers  (Grubbs'
test, p <. 0.05; Grubbs, 1969) were not included
in  assessing method-level  detectability.   All
values except for the outliers were less than
twice  the detection limit objectives.  The limit
of detection for  total monomeric aluminum
(0.012 mg/L) measurements was very near  the
detection limit objective (Table 15).  The mean
value  for total monomeric aluminum measure-
ments (0.009 mg/L)  indicates a possibility of
reagent contamination or a calibration bias.

      For the nonexchangeable  monomeric
aluminum measurements, one  statistical out-
lier was not included in assessing method-level
detectability. Fifteen values were greater than
twice the detection limit objective.  The limit of
detection  for nonexchangeable  monomeric
aluminum (0.019 mg/L)  was less than twice the
detection limit objective (Table  15).  The mean
value (0.014 mg/L) suggests that the in-line cat-
ion exchange column  in one  channel of  the
flow  injection  system caused  a positive
calibration bias for this channel.

      Because of  the  malfunctioning specific
conductance probe  used  in  the  processing
laboratory during the  first half of the NSS-I
(Section 5), detectability was evaluated  for
each half of the survey. The mean value before
the probe was replaced was high  (5.6 pS/cm).
The mean  value after  the probe was replaced
 (0.6 A/S/cm) indicated that the problem  had
been corrected.

 Limits of Detection Based on Field
 Blank Samples--

      Field  blank  samples  offered  several
 advantages over laboratory blank samples in
 assessing  method-level detectability.   Field
 blanks were blind  .Samples   (except at  the
 processing laboratory) inserted at random into
 sample batches.   The  values\ obtained from
 measuring field  blanks were  not subject  to
 control limits  at  the laboratory, as were
 laboratory blanks or detection limit quality con-
 tro! check  solutions.  Field blank samples for
 the NSS-I were prepared from  a single source
 of reagent water (the processing laboratory)
 and  were thus independent of blank samples
 used in calibrations (except at the processing
laboratory).  Finally, field blank measurements
were subjected to the same data verification
procedures as strearnwater samples; this was
not always true for laboratory blank measure-
ments.

    The  limit of  detection calculated from
measurements of  field blank samples  should
be similar to the instrumental limit of detection
calculated from laboratory blank samples. The
us© of field blank samples to assess limits of
detection does have some limitations, but has
been recommended for wet precipitation
samples (Campbell and Scott, 1985).  Concep-
tually,  field blanks are similar to laboratory
blanks, except that the sources of variability
are different.   Field  blanks were processed
through the entire collection and measurement
system of the NSS-I; thus,  they  were poten-
tially subject  to  more sources of error  than
were laboratory blank samples and provide a
more representative  estimate of  a  detection
limit for the entire collection and measurement
system.  Unless mean levels of chemical vari-
ables  measured  in field blank samples are
substantial,  however, the overall  variance
should not be affected by relationships of vari-
ance to  concentration;  hence,  the  variance
for  field blank  samples should be  similar to
the variance  expected from a laboratory blank
sample or from a low-concentration standard
solution.

     Before limits of detection were estimated
from  field blank measurements,  all values
qualified  with an X  flag (Appendix B)  were
eliminated.  In addition, all values identified as
 significant outliers by Grubbs' test (p  <. 0.05;
 Grubbs, 1969) were eliminated.  No more than
 three values were identified as outliers for any
 variable.   Removal of the outliers provided a
 more  representative estimate of  variance that
 was not influenced  by occasional  cases  of
 possible sample  contamination during collec-
 tion and processing.  Summary statistics and
 limits  of detection based  on  field blank
 measurements are presented in Table 16.

     For ANC and specific conductance,  mean
 values from both analytical laboratories were
 less than or near the  detection  limit objec-
 tives. For magnesium, potassium, sodium,
                                            69

-------
 Table 16.
Estimates of limits of Detection Based on Analyses of Field Blank Samples, National
Stream Survey - Phase I
Analytical laboratories*
Laboratory •
Variable
Al-ext
Al-tota!
ANC
BNC
Ca
cr
Cond-lab
DIC-eq
DIC-init
DOC
F
Fe
K
Mg
Mn
Na
NH4+
N03-
P
SiO2
SO42-
Units
mg/L
mg/L
peq/L
A»q/L
mg/L
mg/L
pS/cm
mg/L
mg/L
mg/L
mg/L
mg/L
mg/L
mg/L
mg/L
mg/L
mg/L
mg/L
mg/L
mg/L
mg/L
Detection
limit
objective
0.005
0.005
|Xj * 10
|X| s 10
0.01
0.01
X < 0.9
0.05
0.05
0.1
0.005
0.01
0.01
0.01
0.01
0.01
0.01
0.005
0.002
0.05
0.05
n
22
21
24
24
24
23
24
24
24
24
24
24
23
24
23
24
24
23
24
24
24
Mean
0.004
0.005
-0.4
14.5
0.01
<0.01
<0.1
0.08
0.15
0.1
0.003
<0.01
<0.01
<0.01
<0.01
<0.01
0.01
0.010
<0.001
0.02
0.01
s*
0.0037
0.0053
0.56
2.03
0.006
-
_
0.053
0.048
0.13
0.0013
..
-
-
-
-
0.010
0.012
-
0.027
0.007
1
Estimated
limit of
detection"
0.011
0.016
1.7
6.1
0.02
0.02
0.7
0.16
0.14
0.4
0.004
0.03
0.01
0.01
0.02
0.01
0.03
0.036
0.011
0.08
0.02
Laboratory 2
n
39
39
34
37
37
39
39
39
37
37
39
37
36
38
39
38
39
34
36
38
39
Mean
0.001
0.013
4.4
23.3
0.01
0.01
1.0
0.13
0.10
0.2
NO4*
<0.01
<0.01
<0.01
<0.01
<0.01
<0.01
0.004
<0.001
0.03
0.02
S*
0.0008
0.0063
2.76
5.02
0.004
0.009
0.12
0.081
0.069
0.12
MC*
M
—
__
'mm
__
mm
0.0048
_
0.045
0.016
Estimated
limit of
detection"
0.002
0.019
8.3
15.1
0.01
0.03
0.4
0.24
0.27
0.4
0.010rf
<0.01
0.01
<0.01
<0.01
0.01
0.01
0.014
0.003
0.14
0.05
Processing laboratory*



Al-mono
Al-nex
Cond-PL*
DIC-closed



mg/L
mg/L
//S/cm
mg/L



0.010
0.010
X < 0.9
0.05


n
60
61
46
58'


Mean
0.010
0.014
1.3
0.02'


S*
0.0031
0.0061
0.27
0.011'
Estimated
limit of
detection"
0.009
0.018
0.8
0.03'




























* n = number of samples;  s  - standard deviation.
6 Dashes indicate that the standard deviation is nearly zero.
0 Limit of detection = 3s.
d NC = Not calculated.   Laboratory reported all values less than 0.010 mg/L as 0 mg/L.  Limit of detection
  estimated as 0.010 mg/L
* Laboratory blank data  indicate measurements from batches 2100 through 2127  may be inaccurate as a result of
  a faulty probe.  These  data  were not included in this estimate.
  Field blanks were not  measured.  Calibration blank measurements were used to estimate the limit of detection.
                                                   70

-------
and sulfate, limit of detection estimates from
both analytical  laboratories were near or less
than the detection limit objectives.  The limit of
detection  for  total  monomeric  aluminum
analyses at  the processing laboratory  was
less than the detection limit objective.

      From both  analytical laboratories, limits
of detection were near or less than twice the
detection  limit  objective  for extractable
aluminum (0.011 and 0.002 mg/L), calcium (0.02
and 0.01 mg/L), fluoride (0.004 and 0.010 mg/L),
arid manganese (0.01 and <0.01 mg/L) (Table
16).  The limit of detection for nonexchangeable
monomeric aluminum  (0.018  mg/L) was  near
the detection limit based  on laboratory blank
samples (Table 15).   Measurements of these
variables are apparently subject to sources of
variability from sample collection and process-
ing,  but  the  field blank measurements for
these variables do not indicate a data quality
problem  related to method-level detectability.
Mean values for all  of these  variables  were
near zero at all laboratories.

      For  five variables,  limit  of  detection
estimates  from  both analytical laboratories
were greater than  twice the  detection limit
objective. These variables were total aluminum
(0.016 mg/L and 0.019 mg/L), equilibrated DIG
(0.16 and 0.24 mg/L),  initial DIG (0.14 mg/L and
6.27 mg/L), DOC (0.4 and 0.4 mg/L), and nitrate
(0.036 and 0.014 mg/L).   The detection limits
estimated  for  DIG measurements are derived
from a different sample matrix (analyte-free
water versus  a natural  water sample),  but
provide an indication of the amount of change
expected in samples undersaturated with dis-
solved carbon  dioxide. Measurements of DOC
appear to have been  affected by sporadic low-
 level  contamination during  collection or
processing, as 5 values from Laboratory 1 and
31 values from Laboratory 2 were greater than
twice the detection limit objective.  Addition of
nitrate  to samples  also appears  to  have
occurred, especially during the  early stages of
the NSS-I, as shown by the mean (0.010 mg/L)
 and  limit of detection estimate (0.036 mg/L)
from Laboratory 1 (Table 16).  This laboratory
 did not  analyze  samples in the latter part of
 the survey. The most likely source of the addi-
 tion  appears  to have been the  processing
laboratory (see  Section  5).   The additional
precautions taken at the processing laboratory
appear to have lowered the levels of nitrate in
field blank samples during the latter part of the
survey as shown by the mean (0.004 mg/L) and
limit of detection (0.014 mg/L) from  Laboratory
2 (Table 16).

    For total aluminum analyses, examination
of the measured field blank values for total
aluminum from Laboratory 1 did not indicate a
data  quality  problem.   Only one  value was
greater than twice the detection limit objective.
At Laboratory 2,  19 values were greater than
twice the detection limit objective.  Variability
among batches that occurred during the diges-
tion procedure and sporadic additions of low
concentrations to field blank samples  in the
field  or at the  processing   laboratory would
increase the  standard deviation of the blank
measurements and thus the limit of detection
estimate at both laboratories.   However,  for
Laboratory 2 the analyses of laboratory blank
samples  (Table  15)  suggest  that total
aluminum  measurements   may   have  been
affected by reagent contamination.

     For Laboratory 1, limits of detection for
iron  (0.03 mg/L),   manganese  (0.02  mg/L),
ammonium (0.03 mg/L), and phosphorus (0.011
mg/L) were equal to or greater than twice the
detection limit objective (Table 16).  For  iron,
manganese.and ammonium, all measurements
(except one  for  ammonium) were within  the
control limits established for laboratory blanks.
No data  quality problems  are indicated  for
these  variables.     For   phosphorus,   12
measurements were outside the control limits
for laboratory blanks  (four  were greater  than
twice the  detection limit objective and eight
were less than the negative  value of the detec-
tion limit objective).  It is possible that, for a
few batches, the phosphorus measurements
at Laboratory 1 may be affected by a very low-
level negative calibration bias.

      For Laboratory 2, limits of  detection for
chloride (0.03 mg/L) and silica (0.14 mg/L) were
greater than  twice the detection limit objective
 (Table  16).   In   addition,   the  mean  value
for BNC measurements (23.3 //eq/L) was more
than twice   the  detection   limit objective.
                                            71

-------
  Examination of chloride measurements
  showed only one value greater than twice the
  detection  limit objective,  and no data quality
  problem  is  indicated.    For  silica,   two
  measurements  were greater than twice the
  detection  limit objective.  However, the mean
  value (0.03  mg/L,  Table 16)  indicated con-
  tamination caused by a  sporadic addition of
  silica to blank samples. This addition probably
  occurred  at Laboratory 2, based on the
  analysis of  laboratory blank samples  (Table
  15).  For  BNC measurements, the BNC of a
  field blank sample should be due totally to dis-
  solved carbon dioxide. A high background level
  and  considerable  variability in  BNC  can be
 expected if the sample is not protected from
 the atmosphere during the base titration.  The
 high variability of DIG in field blank samples at
 this laboratory (Table 16) would result in a high
 variability in BNC.

 Assessment of System-Level
 Detectability (Background)

      Background quantities of chemical vari-
 ables were assessed by examining the values
 of  field  blank measurements pooled across
 both analytical laboratories.  Values qualified
 with an X flag (Appendix B) were not included
 in the assessment, but statistical outliers were
 not removed as they were  for the estimates of
 detection limits.  The two statistics of interest
 in  assessing  background are the mean  (or
 median)  and the system  decision limit.  The
 mean  (or  median)  can provide  an  average
 estimate of the amount  of  background con-
 tamination added  during  collection  and
 processing. The system decision limit repre-
 sents the lowest measured value of a chemical
 variable that is distinguishable from field blank
 measurements  at a specified  level  of con-
 fidence. It is a critical value when testing the
 null hypothesis that a single measured value is
 not greater than the  average of field blank
 measurements. System decision limits should
 not be confused with detection limits.

      System decision limits were calculated
from  measurements of  field blank samples
based on both parametric  and nonparametric
statistics.  For many variables, distributions of
field blank measurements may be non-normal,
  and the  use of  nonparametric statistics
  provides a  more  representative  estimate  of
  background that is less sensitive to outlying
  measurements. A parametric system decision
  limit (SDLp) was calculated as follows:
SDL =
Lp =
                          1.65s
 where \ is the mean of field blank measure-
 ments and s is the standard deviation of field
 blank measurements.  The constant 1.65 rep-
 resents the number of  standard deviations
 from the mean of blank  samples within which
 approximately 95 percent of the measurements
 would be expected to lie if they belonged to a
 normal distribution.  A nonparametric  system
 decision limit (SDL ) was calculated using the
 approach of Permuft and Pollack (1986):
                SDLnp -
     = P
         95
 where Pg5 is the the 95th percentile of the field
 blank measurements.

     Summary statistics and system decision
 limits are presented in Table 17. For chemical
 variables whose field blank measurements are
 distributed more  or  less normally, parametric
 and nonparametric decision limits are approx-
 imately equal.    For  almost  all variables,
 parametric and  nonparametric system-level
 decision limits were nearly equal (Table  17).
 Data from  streamwater samples that contain
 chemical variables in quantities less than the
 system-level decision limit should be compared
 and interpreted cautiously, because the source
 of the variability is confounded between what
 was present in the stream at the time of col-
 lection  and what  may have been  added as
 background during collection and processing.
 System-level  decision  limits  were not calcu-
 lated for closed-system DIG or closed-system
 pH because field blank  samples  were not
 measured for these variables.

    For nearly all  variables  the system  deci-
 sion limits did not indicate  serious data quality
 problems related to detection of the analyte or
to background levels.    The percentage of
routine  samples collected  during the  NSS-I
with measured concentrations for a variable
that  were below  the system decision limit is
                                          72

-------
Table 17.    Estimates of System Decision  Umfts Based on Analyses of Field Blank Samples Pooled Across
            Laboratories, National Stream Survey - Phase I
Parametric
Variable
Al-ext
Al-total
Al-mono
Al-nex
ANC
BNC
Ca
cr
Cond-PLrf
Cond-PL*
Cond-lab
DIC-eq
DIC-init
DOC
F
Fe
K
Mg
Mn
Na
NH4+
N03-
P
pH-ANC
pH-BNC
pH-eq
Si02
S042'
True color
Turbidity
Units
mg/L
mg/L
mg/L
mg/L
peq/L
/wq/L
mg/L
mg/L
pS/cm
/uS/cm
pS/cm
mg/L
mg/L
mg/L
mg/L
mg/L
mg/L
mg/L
mg/L
mg/L
mg/L
mg/L
mg/L
pH units
pH units
pH units
mg/L
mg/L
PCU
NTU
n
61
61
61
61
58
63
62
63
26
33
63
63
63
63
63
63
62
62
63
62
63
62
63
58
58
63
63
63
63
63
Mean
0.002
0.011
0.010
0.014
1.8
20.8
0.01
0.01
3.5
1.5
0.6
0.11
0.13
0.2
NC
<0.01
<0.01
<0.01
<0.01
<0.01
<0.01
0.008
0.001
5.73
5.77
5.96
0.03
0.01
6
0.3
s*
0.0028
0.0081
0.0038
0.0061
4.10
7.99
0.006
0.010
2.59
1.08
0.49
0.075
0.085
0.25
NC
-
~
-
-
-
0.008
0.0112
0.0029
0.155
0.131
0.233
0.043
0.014
3.7
1.36
System
decision
limit (SDLp)*
0.007
0.024
0.016
0.024
8.6
34.0
0.020
0.026
7.8
3.2
1.4
0.236
0.276
0.6
0.010'
0.01
0.01
<0.01
0.010
0.01
0.02
0.026
0.006
NC
NC
NC
0.10
0.033
12.1
2.5
n
61
61
61
61
58
63
62
63
26
33
63
63
63
63
63
63
62
62
63
62
63
62
63
58
58
63
63
63
63
63
Nonparametric
Median
0.001
0.010
0.010
0.015
1.5
19.3
<0.01
<0.01
2.0
1.2
0.9
0.10
0.13
0.2
NC
<0.01
<0.01
<0.01
<0.01
<0.01
0.01
0.005
<0.001
5.69
5.74
5.89
0.02
0.01
5
0.1
System
decision
limit (SDLnp)c
0.007
0.027
0.015
0.023
8.7
34.9
0.02
0.03
8.0
2.2
1.2
0.23
0.23
0.5
0.010/
0.02
0.02
<0.01
0.012
0.012
0.32
0.02
0.006
NC
NC
NC
0.09
0.05
10
0.2
* Dashes indicate that the standard deviation is nearly zero.
h SDLp = mean +  1.65s.
0 SDLnp = 95th percentile of distribution of field blank measurements.
d Measurements for first half of survey.
6 Measurements for second half of survey.  NC  = Not calculated.
f Laboratory 2 reported all measured concentrations less than 0.010 mg/L as zero.  System decision limit
  estimated as 0.010 mg/L.
                                                     73

-------
 presented in Figure 12,  and  the  ranges of
 measured values are presented in Table 18. In
 some cases, the extreme values (e.g., extreme-
 ly high specific conductance or extremely low
 ANC)  represent rare cases of  extreme  condi-
 tions that are not representative of the
 streams of interest. The data shown in  Figure
 12 indicate that for total aluminum, silica, and
 fluoride,  90  to 100 percent of the routine
 samples  had  measured  concentrations  that
 were greater than the respective system deci-
 sion limits. Thus, the potential problems iden-
 tified for  these  three variables will not have a
 large  effect  on the interpretation of routine
 sample data.

 Discussion and Summary:
 Detectability

      For most variables,  method dectection
 limits and system decision limits indicated no
 serious problems with data quality that were
 related to either instrument  performance,
 methodological  performance,  or  background
 levels of analytes.  Background levels of total
 aluminum and silica  (SDL   = 0.027 mg/L and
 0.09 mg/L, Table 17)  should not confound data
 interpretation as nearly all routine samples had
 measured  concentrations above the system
 decision limit (Figure 12).  The error in fluoride
 measurements at Laboratory 2 (not measuring
 values less than 0.010 mg/L) also should not
 confound  data  interpretation,  as  nearly all
 routine samples had measured concentrations
 greater than the system decision limit.

      The  observed  background levels of total
 monomeric aluminum (SDL  = 0.015 mg/L)
 and nonexchangeable monoYneric  aluminum
 (SDL   =  0.023 mg/L) are the result of instru-
 ment variability or calibration bias rather than
 contamination.  The usefulness of these data
 may be limited at  low concentrations, espe-
 cially since the variable of interest,  exchange-
 able (or labile) monomeric aluminum,  is calcu-
 lated as  the  difference  between the two
 measured   fractions.   As a  result of  low
concentrations,  the  difference calculated  for
exchangeable monomeric aluminum some-
times results in negative values.
     The background level of BNC (SDL    =
 39.14 ijeq/L) suggests that there can be "con-
 siderable variability in dilute  streamwater
 samples with low BNC.   The BNC measure-
 ment  protocol for the  NSS-I  was designed
 primarily to assist in verifying  the measure-
 ments of ANC on a  sample-by-sampie basis
 (Hillman et  al.,  1987).    The usefulness of
 routine stream sample  data to determine the
 presence of weak versus strong acids may be
 limited, because the  samples  were  not
 protected  from  atmospheric carbon  dioxide
 during collection, handling, or titration.

     Specific conductance  measurements at
 the  processing laboratory  were  affected by a
 faulty  probe for  the first 28 sample batches,
 and thus we recommend  that the analytical
 laboratory  measurement of  specific conduc-
 tance  be used for data interpretation.   The
 primary purpose  of the  processing laboratory
 measurement of  specific conductance was to
 check on the stability  of streamwater samples
 between the time of collection and the time of
 processing.

     Certain other types of  data  interpretation
 activities, not related to acidification or stream
 classification, may be  limited by  the  back-
 ground levels introduced into NSS-I samples.
 For example, examination of nutrient relation-
 ships or productivity may be limited by back-
 ground levels of nitrate and phosphorus. The
 measurement program for  blank samples that
 was used during the  NSS-I was designed to
 control calibration  biases and   background
 contamination,  rather than to correct routine
 sample measurements at trace concentrations
 (see Taylor, 1984 and  1987).  Conceivably,  the
 data from field or laboratory blank measure-
 ments  presented  in this report could be used
 to develop  a correction factor by  using  the
 equations presented  by Taylor (1984,  1987).
 However,  the  uncertainty  associated with
 the blank measurements will be conservative
 because it will be based on an among-batch
 rather  than a   within-batch  estimate  of
 measurement variability.

     The concept  of detection limits  may need
to be more clearly defined  for  future  AERP
                                          74

-------
JUU -
Samples
OO
O
I.I.
6
S 60-
t.
c
~ 40-
o
IT
O
Percent
ro
3 O
I.I.I
CATIONS AND METALS
a 80-
"5.
eg —
09
i 60-
0
k.



f|H!
:?':*


m
Si



1
i
w
«
c
i 40-
0
cc
o
| 20-
o
a. ~
ANIONS AND DOC





Ca Mg Na K Fe Mo NH4* S042- N03" CT F' DOC
0.02 <0.01 0.01 0.02 0.02 0.01 0.02 0.05 0.04 0.03 0.01 0.4
                  System Decision Limit (mg/L)
               System Decision Limit (mg/L)
      100-
           ALUMINUM FRACTIONS
                                                         100-
          Al-total  Al-ext  Al-mono Al-nex

           0.027  0.008  0.017  0.025

          System Delsion Limit (mg/L)
                                                       £  80-
                                                          60-
                                                       a
                                                       c
                                                       1  «H
                                                       o
                                                       
-------
Table 18.   Rang* and Central Tendency Statistic* for Analyte Concentrations In Routine Stream Sample*,
          National Stream Survey - Phase I
Variable'
Al-ext
Al-total
Al-mono
Al-nex
ANC (/jeq/L)
BNC (jieq/L)
Ca
cr
Cond-PL O^S/cm)
Cond-lab (pS/cm)
DIC-closed
DIC-eq
DIC-init
DOC
F
Fe
K
Mg
Mn
Na
NH4+
NOg
P
pH-closed (pH units)
pH-ANC (pH units)
pH-BNC (pH units)
pH-eq (pH units)
Si°i
S04*
True color (PCU)
Turbidity (NTU)
Number
1,345
1,343
1,353
1,349
1.378
1,378
1,375
1.375
1,366
1,378
1.378
1,378
1,378
1.374
1,375
1,376
1.376
1.376
1.376
1,376
1.374
1,373
1,374
1.378
1,378
1,378
1,378
1.375
1,375
1,367
1.364
Minimum
value0
-0.002
-0.006
0.001
0.000
-1,750.600
-85.400
0.063
0.075
10.500
10.500
0.036
-0.102
0.059
0.000
0.000
-0.010
0.001
0.098
-0.010
0.129
-0.011
-0.001
-0.008
3.270
3.000
3.000
3.050
-0.006
0.046
0.000
0.030
Median
value
0.009
0.111
0.017
0.018
176.550
59.250
4.850
2.980
65.700
63.750
3.204
1.878
2.358
1.610
0.040
0.041
0.940
1.591
0.023
2.401
0.023
0.842
0.004
6.840
6.850
6.865
7.340
5.300
8.126
15.000
2.200
Mean
value
0.080
0.345
0.098
0.026
448.039
95.011
10.365
6.725
111.708
109.612
6.261
5.191
5.720
3.338
0.050
0.262
1.185
3.341
0.136
4.263
0.051
3.147
0.016
6.633
6.631
6.644
7.077
6.506
15.610
31.137
5.776
Maximum
value
10.100
37.100
12.223
0.375
7,602.800
2,421.400
96.623
380.000
1,376.300
1,294.000
92.440
71.200
92.706
171.000
0.520
32.800
8.842
37.345
12.100
185.000
3.035
70.000
1.420
9.360
8.890
8.920
8.860
34.125
340.000
900.000
650.000
a Concentrations are in mg/L unless otherwise indicated.
* Negative values are a result of analytical laboratory instrument calibration bias.
(such as  audit  samples) may be of more use
when assessing  laboratory  performance in
terms of detectability.

Accuracy

      Accuracy within the processing lab-
oratory and each analytical laboratory involved
in the NSS-I  was evaluated by examining the
results  from analyses of performance  audit
samples.  During the data verification process,
audit sample results were compared to accep-
tance criteria (ranges) calculated from perfor-
mance  audit  data from  previous  NSWS
projects.  Further detail about the calculation
of acceptance criteria is found in Drouse et al.
(1987).  The control limits for the acceptance
criteria are presented in Appendix C.  Batches
                                            76

-------
of samples  containing  audit  sample values
that  were outside the critiera were  qualified
with an N flag (Appendix B) for more  intensive
review.

      Accuracy was evaluated primarily on the
basis of the results of analyses of  the  syn-
thetic audit  samples.   However, other AERP
studies that have  used  the  synthetic  audit
samples (Drouse, 1987;  Silverstein et  al., 1987)
have reported that  the  concentrations of
several variables are affected either  by
changes in dissolved carbon  dioxide  con-
centration between the time of preparation and
measurement (e.g., BNC,  DIG, and pH) or by
concentrations of other variables (e.g., ANC).
For these variables, theoretical concentrations
are  inaccurate or not  available.   For  these
selected variables, data from the analysis of
natural audit samples are  presented in this
section to  assist in assessing accuracy.  The
natural audit samples appear to be less sensi-
tive to changes  in dissolved carbon dioxide,
and their chemical composition is not affected
by preparation errors.  Data  for all variables
from measurements of natural audit samples
 are presented in Appendix A.

      The accuracy estimates presented in this
 section provide  an indication  of the presence
 of  systematic   errors in  measurement.  An
 accuracy estimate for an  analyte derived from
 a theoretical value for a synthetic audit sample
 can  be used as an  estimate of  absolute
 analytical  method bias  only if  preparation
 errors  are  assumed to  be  negligible.   This
 assumption  is  not  valid  for  all  variables,
 because measured values from  the support
 laboratory did not always agree with theoreti-
 cal  values (see Appendix A).  Thus,  accuracy
 estimates from  synthetic audit samples are
 probably conservative.  Accuracy estimates for
 variables not having defined theoretical con-
 centrations in the synthetic audit samples, or
 estimates  based on natural audit  samples,
 only  provide an  estimate  of  relative   bias,
 because the index value  (the value  measured
 by  the analytical  laboratories) represents a
 measured  value obtained by a  specific
 methodology.   In addition,  the audit sample
 concentrations  do not bracket the  range of
 sample concentrations or many variables. For
these reasons, accuracy estimates should not
be  used  as quantitative estimates of  sys-
tematic measurement  errors.    Estimates of
among-batch precision presented later in this
report include the effects of both systematic
and random measurement errors and thus can
provide estimates of measurement uncertainty,
subject to the limitations mentioned above.

     The following  equation provides an esti-
mate of  percent accuracy for each laboratory
for all analytes except pH:

      Percent accuracy = [(x- R) + R]100
 where x is the mean of measured values and R
 is the theoretical value or an index value based
 on measurements from  one  or  more  lab-
 oratories. For pH, accuracy was expressed as
 the difference between  the mean measured
 value and the theoretical or index value.

     For the  synthetic audit  samples,  index
 values were developed for all analytes based
 on verification  measurements  made at the
 support laboratory immediately after prepara-
 tion of a sample  lot.  Six replicate measure-
 ments were made for each variable  for each
 lot.  All replicates for each lot were measured
 in a  single batch, providing  an estimate  of
 within-batch variability. Index values  for ANC,
 BNC,  DIG,  and pH are presented in  this sec-
 tion,  while  values calculated  for all  variables
 are presented in Appendix A.  For all  variables
 except BNC,  standard solutions at two con-
 centrations  were obtained  from  the  EPA
 Environmental Monitoring Systems Laboratory
 in Cincinnati and were  analyzed  with each
 batch of verification  samples.  The  measure-
 ment of these standards served to validate the
 analytical measurements made at the support
 laboratory.   Data from these standards were
 obtained for six different batches of  synthetic
 audit  verification  samples to  provide an esti-
 mate  of the  among-batch variability of  the
 support  laboratory measurements.  These
 data,   based  on measurements  of  the EPA
 standard solutions, are  presented  in Table 19
 for ANC, pH, and DIG.  Data for these stand-
 ards  for all  other variables  except  BNC are
 presented in Appendix A.
                                            77

-------
Table 19.
Summary Statistics for Selected Variables for EPA Reference Samples Measured at the Support
Laboratory, National Stream Survey - Phase I


Variable
ANC Oueq/L)
DIG (mg/L)
pH (pH units)
True
value of
standard
68.8
1.13
7.8

Num-
ber
5
6
6


X
67.1
1.20
7.76


s
0.77
0.048
0.042

x - True
value
-1.7
0.07
-0.04

Accuracy
percent
-2.5
6.2
•

Precision
(%RSD)
1.1
4.0
™
      For ANC, the  within-batch standard
deviation estimates for Lot 14 (2.90 peq/L) and
Lot 15 (2.93 //eq/L),  presented in Appendix A,
were greater than the among-batch standard
deviations  estimated from the standard solu-
tions (0.77 jueq/L; Table 19).  For BNC,  within-
batch standard deviations (Appendix A)  for Lot
14 (4.37/L/eq/L) and Lot 15 (3.85 A/eq/L) were the
only estimates available. Index values for ANC
and  BNC were estimated  as  the  mean
measured  values,  with  the 95 percent con-
fidence  intervals calculated by using the
within-batch   standard  deviation   estimate.
Index values  for  DIG  and pH  also were
estimated as the mean measured values from
the support laboratory, but the 95 percent con-
fidence intervals were  calculated from  the
among-batch  standard deviations  estimated
from  the EPA standards (Table 19).

      Index  values  for  the  natural  audit
samples were developed  from measurements
made at different laboratories.  The  Bagley
Lake  audit sample used during the NSS-I was
also  analyzed at two other laboratories during
the Western Lake Survey-Phase I (Silverstein
et al., 1987) in late 1985.  The Big Moose Lake
audit  sample  was   analyzed  by two  other
laboratories during Phase II of the Eastern
Lake  Survey, which was conducted during and
after  the NSS-I.     Data  from  all  these
laboratories  (including the  two laboratories
involved  in the  NSS-I) were used to estimate
index values.  Index values were calculated as
a  grand  mean,  based  on  weighted  mean
values from each laboratory, by the following
equation (Taylor, 1987):
                                                      L
                                                      I w.x.
                                                         '  '
                                              s  =
                                    where    w

                                         Y   =
                                          L
                                          wi   =
                                          s   =
                                          X.   =

                                          a*--
                                                      L
                                                      Z w.
grand  mean
number of  laboratories
weighting factor for
laboratory i
standard deviation of
weighted means
mean  value from
laboratory i
variance from laboratory i
number of  measurements
from laboratory  i
                                          Ninety-five percent confidence  intervals
                                    about  the  index value  were also calculated.
                                    Theoretical and index values for synthetic and
                                    natural audit samples are  presented in Table
                                    20.

                                          Accuracy estimates were calculated for
                                    each laboratory and for each lot of synthetic
                                    audit samples. Synthetic audit samples from
                                    Lot 14 were used to evaluate accuracy during
                                    the initial part of  the NSS-I, while  samples
                                    from Lot  15 were used during the latter part.
                                              at
                                          For those analytes that were measured
                                        the  analytical   laboratories,  systematic
                                            78

-------
Table 20.    Theoretical and Index Value* for Analyses of Synthetic and Natural Audit Samples, National
            Stream Survey - Phase I
Theoretical values* of synthetic audit samples
Variable
M*
Ca
cr
Cond-lab
Die*"
DOC
F-
Fe
K
Units Theoretical value
mg/L
mg/L
mg/L
pS/cm
mg/L
mg/L
mg/L
mg/L
mg/L
0.020
0.194
0.343
17.5
0.959rf
1.0
0.042
0.059
0.203
Variable
Mg
Mn
Na
NH4+
N03-
P-
SiO,
SO4*

Units
mg/L
mg/L
mg/L
mg/L
mg/L
mg/L
mg/L
mg/L

Theoretical value
0.447
0.098
2.75
0.168
0.467
0.0273
1.070
2.280

                                                Index values
Synthetic audit samples*
Lot 14 Lot 15
Variable n Mean Cl0 n Mean Cla
ANC 6 101.7 3.04 6 109.2 3.07
BNC 6 32.0 4.58 6 22.8 4.04
DICC 6 1.14 0.050 6 1.26 0.050
pH7 6 7.22 0.044 6 7.29 0.044










Natural audit samples'
Bagley Lake
Variable
Al-ext
Al-total
Al-mono
Al-nex
ANC
BNC
DIC-closed
DIC-eq
DIC-initial
Fe
pH-ANC
pH-BNC
pH-eq
pH-closed
n
..
4
-
-
4
4
2
4
4
~
4
4
4
2
Grand.
mean
..
0.016
-
-
121.0
29.7
1.67
1.52
1.52
~
7.06
7.06
7.29
7.04
CI*
-
0.0006
-
-
0.16
0.52
0.068
0.015
0.012
~
(0.014)
(0.014)
(0.017)
(0.066)
Big Moose
n
4
4
2
2
4
4
2
4
4
4
4
4
4
2
Grand
mean"
0.197
0.272
0.193
0.053
-3.1
72.9
0.55
0.11
0.36
0.05
5.10
5.15
5.17
5.14
Lake
CI*
0.0052
0.0045
0.0174
0.0178
0.45
1.36
0.061
0.009
0.009
0.002
0.010
0.013
0.006
0.060
 * Assuming no preparation error or external effect.
 b Value applicable to all aluminum fractions.
 0 Value applicable to all DIC measurements.
 d Value does not  include carbon dioxide added during air equilibration procedure.
 e Index values based on support laboratory measurements.
 f Index values based on analytical laboratory measurements from NSWS programs, including NSS-I.
 9 One-sided  95% confidence  interval.
 h Grand mean calculated from weighted means from four analytical laboratories or from two processing
   laboratories.
 1 Value applicable to all pH  measurements.
                                                    79

-------
errors may have been introduced  at  the
processing laboratory during the preparation
and processing of the several  aliquots from
bulk  streamwater samples  (or   syringe
samples in the case of extractable aluminum).
These errors could result from contamination,
improper filtration or preservation, or changes
in the sample composition between the time of
collection and the time of processing.

        For synthetic audit samples,  results
for  each  laboratory  of  field  audit  and
laboratory audit sample analyses were com-
pared by using  a single-classification analysis
of  variance.   For ail variables  (except total
aluminum,  extractable aluminum,  and iron),
mean values for field  audit samples were not
significantly  different  (p <.  0.05) from cor-
responding laboratory audit samples. For total
aluminum, extractable  aluminum, and iron, the
mean concentrations  in field audit samples
were  substantially  lower  than in laboratory
audit  samples at  both laboratories.  Loss of
total  aluminum and iron from field audit
samples may result from adsorption of these
species onto  container  surfaces.   Sample
filtration at the processing laboratory also may
remove  iron as well  as the  aluminum that
would otherwise transfer to the MIBK extract
and   be   measured  as  total extractable
aluminum, especially if iron or aluminum-
containing  precipitates  have  formed
(Silverstein et al.,  1987; Drous3, 1987).  There-
fore, for these three variables, only laboratory
audit  samples  were used to estimate accu-
racy.  For  all other variables, measured values
of  field and  laboratory audit samples were
combined  for each laboratory when both types
of samples were measured.   Laboratory 1 did
not  measure any field  audit samples from Lot
15; accuracy estimates for all variables for this
lot  are  based  on laboratory audit samples
only.  For each variable,  outlying values were
identified by using Grubbs'  test (p = 0.005;
Grubbs, 1969)  and were not included in the
analyses.  No more than one outlier was iden-
tified  and  removed for  any single variable.
Removing  outliers  served to improve the preci-
sion estimates for the  values and thus  the
confidence  in  the estimated mean value.
Removal of outliers did not necessarily improve
the estimate of percent accuracy.
Percent Accuracy Estimates for
Laboratory 1

     Summary statistics  and percent accu-
racy estimates for synthetic and  natural audit
samples  measured  at  Laboratory  1 are  pre-
sented in Tables 21 and 22.  For  the synthetic
audit samples, percent accuracy estimates for
eleven  variables for  which theoretical  values
were available were  within or near the data
quality objective for both lots:  chloride, DOC,
fluoride, potassium,  magnesium, manganese,
sodium,ammonium, nitrate, silica, and sulfate.
The observed percent accuracy for calcium for
Lot 14 (16 percent)  represents  an apparent
bias of 0.03 mg/L, is barely significant  at the
95 percent level of confidence, and does not
indicate a data quality problem.

     For specific conductance, percent accu-
racy estimates for both lots were less than 10
percent (Table 21) and  represent an apparent
negative bias of less than 2 juS/cm. Conduc-
tance measurements in the natural audit
samples from Laboratory 1 also  indicated the
potential for negative bias, compared to index
values (see Appendix A).

      Percent  accuracy   estimates for  four
variables for which  theoretical  values were
available   were  well  outside  data  quality
objectives for one or both lots:   extractable
aluminum,  total aluminum,  iron,  and  phos-
phorus. For the synthetic audit samples, mean
values   for extractable and  total  aluminum
were not significantly different from theoreti-
cal  values because  of  the  large  variability
(Table 21). Data from the  Bagley  Lake and Big
Moose Lake samples (Table 22), for these two
variables were not significantly different from
the index values. There is no  evidence  for
systematic error in extractable aluminum  and
total aluminum measurements.

     Accuracy  estimates  for  iron based on
the synthetic  audit sample may  be  unreliable
because of a change in the sample between
the time of sample  preparation and the time
that preserved aliquots were  prepared  at the
support laboratory.  Mean  values for both lots
(0.03 mg/L, Table 21) were in agreement  with
verification values from the support laboratory
                                          80

-------
Table 21.    Percent Accuracy Estimates for  Laboratory 1 Measurments of Synthetic Audit Samples,
            National Stream Survey - Phase I
Lot 14*

Variable
Al-ext
Al-total
Ca
cr
Cond-lab
DOC
F.
Fe
K
Mg
Mn
Na
NH4+
N03-
P
SiO,
S04^




ANC
BNC
DIC-eq
DIC-init
pH-ANC
pH-BNC
pH-eq

Units
mg/L
mg/L
mg/L
mg/L
j/S/cm
mg/L
mg/L
mg/L
mg/L
mg/L
mg/L
mg/L
mg/L
mg/L
mg/L
mg/L
mg/L

Accuracy
objective
(%)
10
10
10
10
0.1rf
0.1rf
0.1*
Accuracy
objec-
tive (%)
10
10
10
10
5
10
10
10
10
10
10
10
10
10
10
10
10


Index
value(CI)c
101.7 (3.04)
32.0 (4.58)
1.14 (0.050)
1.14 (0.050)
7.22 (0.044)
7.22 (0.044)
7.22 (0.044)
Theo-
retical
value6
0.020
0.020
0.19
0.34
17.5
1.0
0.042
0.06
0.20
0.45
0.10
2.75
0.17
0.467
0.027
1.07
2.28
Lot


n
4
5
14
13
14
13
13
9
13
14
14
14
13
13
13
14
14
14*


Mean
0.027
0.018
0.22
0.33
16.3
1.0
0.040
0.03
0.20
0.43
0.09
2.82
0.17
0.477
0.021
1.20
2.21

CI
0.0254
0.0063
0.026
0.006
0.18
0.06
0.0013
0.011
0.001
0.002
0.004
0.023
0.003
0.0243
0.0025
0.015
0.029
Percent
accu-
racy
35™
no
-10
16*
-3*
-7*
0
-5*
-50*
0
-4
-10*
2*
0
2
-22*
ns
12
-3

n
3
3
4
4
4
4
4
4
4
4
4
4
4
3
3
4
4
Lot

Mean
0.015
0.029
0.20
0.33
16.0
1.0
0.039
0.03
0.19
0.43
0.09
2.78
0.16
0.487
0.015
1.19
2.25
15*

CI
0.0086
0.0226
0.047
0.012
0.20
0.42
0.0043
0.041
0.004
0.008
0.007
0.036
0.011
0.0027
0.0087
0.036
0.028

Percent
accu-
racy
-25nS
ns
-21
5
-3
-9*
0
-7
ns
-50
-5*
-4
-10*
1
-6
4*
-44*
ns
10
-1
Lot 15*


Percent
n
Mean
14 104.7
13
14
14
13
13
14
22.6
1.33
1.41
7.04
7.07
7.17
CI accuracy
1.56
3
2.10 -29*
0.049
0.030
0.035
0.035
0.159
17*
24*
ff
-0.18 '
* 6
-0.15 '
-0.05*




Index
value(CI)c
109.2
22.8
1.26
1.26
7.29
7.29
7.29
(3-07)
(4.04)
(0.050)
(0.050)
(0.044)
(0.044)
(0.044)
n Mean
3 104.2
4 18.8
4 1.23
4 1.39
4 7.06
4 7.08
4 7.31
CI
1.54
2.42
0.081
0.099
0.083
0.078
0.166

Percent
accuracy
-5*
ns
-18
-2
10
*,e
-0.23
*e
-0.21
0.02*
 * n  = number of measurements.
   CI = one-sided 95 percent confidence intervals.
   ns = not significantly different from the theoretical or index value at p = 0.05.
   *  = significantly different from the theoretical or index value at p < 0.05.
 b The theoretical value is the expected value of the synthetic audit sample assuming no preparation
   error and no external effects.
 0 Index value  is based on the mean values from  the support laboratory measurements.
 a Objective expressed in pH units.
 e Accuracy expressed as the difference between the index value and the mean analytical laboratory value.
                                                    81

-------
Table 22.   Percent Accuracy Estimates for Laboratory 1 Measurements of Selected Variables In Natural
           Audit Samples, National Stream Survey - Phase I
Index value
(n* = 4)
Variable
Baqley Lake samples
Al-total
ANC
BNC
DIC-eq
DIC-initial
pH-ANC
pH-BNC
pH-eq
Big Moose Lake samples
Al-ext
Al-total
ANC
BNC
DIC-eq
DIC-initial
Fe
pH-ANC
pH-BNC
pH-eq
Accuracy
objec-
Units live (%)

mg/L
peq/L
peq/L
mg/L
mg/L
pH units
pH units
pH units

mg/L
mg/L
peq/L
peq/L
mg/L
mg/L
mg/L
pH units
pH units
pH units

10
10
10
10
10
0.1"
0.1*
0.1*

10
10
10
10
10
10
10
0.1*
0.1*
0.1*
Grand*
mean

0.016
121.0
29.7
1.52
1.52
7.06
7.06
7.29

0.197
0.272
-3.1
72.9
0.11
0.36
0.05
5.10
5.15
5.17
±CIC

0.0006
0.16
0.52
0.015
0.012
0.014
0.014
0.014

0.0052
0.0045
0.45
1.36
0.009
0.009
0.002
0.010
0.013
0.006
n*

14
14
14
14
14
14
14
14

13
13
15
15
15
15
15
15
15
15
Laboratory 1
measured values
Mean

0.019
120.7
23.0
1.56
1.68
7.08
7.13
7.30

0.223
0.259
-2.9
64.4
0.07
0.25
0.04
5.14
5.16
5.17
±Clc

0.0152
1.83
2.42
0.032
0.029
0.051
0.068
0.103

0.0296
0.0196
1.24
3.68
0.026
0.022
0.010
0.041
0.052
0.008
Percent
accuracy

19"5
<1
-22*
3
0*
0.02f
0.07nS-f
O-O/

13"*
-5
-6
12*
-36*
-31*
0
0.04^
O.O/
0
* n  = number of measurements.
b Grand mean based on weighted mean values from four laboratories.
0 ±CI = 95 percent confidence interval.
d For pH, accuracy is expressed as the absolute difference between the index value and the measured value.
  *  = mean value significantly different from the index value at p  s 0.05.
  ns = accuracy estimate outside data quality objective, but mean  value is not significantly different from
       the index value at p = 0.05.
* Objective expressed in pH units.
f Accuracy expressed as the difference between the index value and the mean analytical laboratory value.
(0.04 mg/L, see Appendix A) and those of Lab-
oratory 2 (0.02 to 0.04 mg/L, discussed  in the
next subsection). Measurements of iron in the
synthetic audit samples were imprecise, with
the  95 percent confidence intervals repre-
senting from  +30 percent of the mean  (0.011
mg/L for Lot 14) to ±136 percent of the  mean
(0.041 mg/L) for Lot 15 (Table 21).  Data from
the Big Moose Lake sample (Table 22),  which
had an iron concentration similar to that  of the
synthetic audit sample, did not indicate a rela-
tive bias with respect to the index value.  There
is  no strong reason to  suspect systematic
errors in  iron measurements within the  range
of  concentrations  represented  by  the  audit
samples.
                                               82

-------
        Data for phosphorus  measurements
from lot 14 (Table 21) indicate the potential for
negative bias at high phosphorus concentra-
tions (greater than  0.20 mg/L).  For Lot 14, the
mean value was in agreement  with that from
Laboratory 2 (0.022 mg/L; discussed in the fol-
lowing subsection),  and thus  the  observed
error may represent a loss of phosphorus from
the synthetic audit during the  preparation of
preserved aliquots at the support laboratory.
However, the mean value from Lot  15  (0.015
mg/L)  is not  in  agreement with that  from
Laboratory 2  (0.023 mg/L);  see  the next
subsection), indicating that measurements
made at Laboratory  1 during the  last half of
the survey may be underestimates of the true
sample concentrations. The accuracy estimate
for Lot 15,  however,  is based on only  three
measurements.   One outlying  value (-0.0050
mg/L) was not included in estimating accuracy.

        The accuracy estimates for ANC, DIC-
eq, and pH-eq for both lots of synthetic audit
samples were within or near the data quality
objectives based on comparisons  to the index
values (Table 21). For DIC-initial, pH-ANC, and
pH-BNC measurements,  the observed biases
(positive for DIG in  Lot 14), indicate that the
dissolved carbon dioxide concentration  in the
audit samples generally increased between the
time of sample preparation and measurement.
This change could result from  a  decrease  in
temperature  (increasing  the solubility of
carbon dioxide), a higher ambient  atmospheric
concentration of carbon dioxide at the analyti-
cal  laboratory,  or  a  combination  of  both
effects.  Equilibrated pH and  DIG  measure-
ments, although  within accuracy objectives,
were imprecise, providing additional evidence
of the sensitivity  of  the synthetic  audit
samples  to changes in dissolved carbon
dioxide concentrations.  Data from the Bagley
Lake sample,  with index values for ANC, DIG,
and pH that were similar to those calculated
for the synthetic audit  sample, yielded accu-
racy  estimates that were within or near the
data  quality objectives for  ANC, ail DIG
measurements, and ail pH measurements
(Table 22).   Data from the Big  Moose Lake
sample (Table 22)  had index values for pH that
were between  5.10 and 5.20, index  values for
 DIG less than 0.4 mg/L, and an index value for
ANC  of  approximately -3  //eq/L   Accuracy
estimates for ANC and pH from  Laboratory 1
for  the Big Moose Lake sample were within
data quality objectives.   Accuracy estimates
for DIG were outside the data quality objective
but the  observed  differences  were small
(approximately 0.1 mg/L or less).

      For BNC,  accuracy estimates for the
synthetic audit samples are outside the data
quality objective for Lot 14 (Table  21),  when
compared to the index value based  on support
laboratory  measurements.    Data  from the
Bagley Lake sample (Table 22), which has an
index value for BNC similar to that  of the syn-
thetic audit samples,  suggest that measure-
ments of BNC  from  Laboratory  1 may be
underestimates of sample concentrations, but
the observed  magnitude of the relative bias is
small (5  to 10 /^eq/L) and occurs only at very
low concentrations of BNC (30 /jeq/L).   Data
from the Big Moose Lake audit sample (Table
22) which had a higher index value for BNC
(72.9 Afeq/L), indicate that BNC measurements
from  Laboratory 1  exceeded the DQO of 10
percent  by  2  percent  relative to  the  index
values.

      In  conclusion, the only data quality
problems observed for Laboratory 1 that  are
related to accuracy appear to be the potential
for underestimating BNC at low concentrations
(less than  30 jueq/L),  specific conductance in
dilute samples (less than 25 juS/cm), and phos-
phorus during the latter half of the survey.  For
all other  variables, percent accuracy estimates
were within or near the data quality objectives.

Percent Accuracy Estimates for
Laboratory 2

      Accuracy  estimates for synthetic and
natural  audit  samples  for Laboratory 2  are
presented in  Tables 23 and  24.  For the syn-
thetic audit sample, accuracy estimates for ten
variables for  which theoretical  values  were
available were  within or near the data quality
objectives for both lots (Table 23):  calcium,
chloride, fluoride, potassium,  magnesium,
manganese,  sodium, ammonium, nitrate, and
sulfate.
                                           83

-------
Table 23.    Percent Accuracy Estimate* for Laboratory 2 Measurement* of Synthetic Audit Samples, National
            Stream Survey - Phase I
Lot 14*
Variable
Al-ext
Al-total
Ca
cr
Cond-lab
DOC
F
Fe
K
Mg
Mn
Na
NH4+
N03-
P
SiO2
S04a-
Variable
ANC
BNC
DIC-eq
DIC-init
pH-ANC
pH-BNC
PH-eq
Units
mg/L
mg/L
mg/L
mg/L
pS/cm
mg/L
mg/L
mg/L
mg/L
mg/L
mg/L
mg/L
mg/L
mg/L
mg/L
mg/L
mg/L
Accuracy
objective
10
10
10
10
0.1*
0.1*
0.1rf
Accuracy Theo-
objective retical.
(%) value*
10
10
10
10
5
10
10
10
10
10
10
10
10
10
10
10
10

0.020
0.020
0.19
0.34
17.5
1.0
0.042
0.06
0.20
0.45
0.10
2.75
0.17
0.467
0.027
1.07
2.28

Index
value(CI)tf n
101.7
32.0
1.14
1.14
7.22
7.22
7.22
(3.04) 9
(4.58) 8
(0.050) 9
(0.050) 9
(0.044) 9
(0.044) 8
(0.044) 9
n
5
4
9
8
9
9
9
5
9
9
9
8
9
8
9
9
9

Mean
114.4
49.1
1.37
1.62
6.73
6.71
7.25
Mean
0.021
0.035
0.19
0.32
19.5
1.3
0.043
0.04
0.20
0.44
0.11
2.76
0.19
0.465
0.022
0.92
2.15
Lot 14
CI
0.0024
0.0103
0.017
0.018
0.21
0.20
0.0012
0.003
0.006
0.016
0.002
0.062
0.010
0.0140
0.0026
0.044
0.157

Percent
CI accuracy
5.73
2.59
0.202
0.074
0.049
0.033
0.069
7*
53*
20/79
42*
0.49**
0.51**
0.03*
Percent
accu-
racy
5
75*
0*
-6*
11*
30*
2
-33*
0
-2
10
0
12*
0
-18
-10*
-6

n
23
23
29
28
28
28
28
23
29
28
29
29
28
28
28
28
29

Index
value(CI)c
109.2
22.8
1.26
1.26
7.29
7.29
7.29
(3.07)
(4.04)
(0.050)
(0.050)
(0.044)
(0.044)
(0.044)
Lot 15*
Mean
0.015
0.031
0.18
0.33
19.6
1.1
0.043
0.02
0.20
0.44
0.10
2.70
0.17
0.473
0.023
0.92
2.30
Lot 15
n Mean
29 119.0
29 58.9
29 1.32
29 1.51
29 6.69
29 6.69
29 7.23
CI
0.0008
0.0032
0.004
0.011
0.07
0.08
0.0004
0.006
0.004
0.004
0.001
0.038
0.005
0.0132
0.0012
0.024
0.046

CI
2.71
4.74
0.082
0.102
0.048
0.049
0.056
Percent
accu-
racy
-25*
55*
-5*
-3
12*
10*
2*
-67*
0
-2*
0
-2*
0
1
15*
-14*
1

Percent
accuracy
10*
158*
5
20*
-0.6 *e
-0.6 **
0.06"
* n  = number of  measurements.
  CI = one-sided 95 percent confidence interval.
  ns = not significantly different from the theoretical or index value  at p = 0.05.
   * = mean value significantly different from the index value at p <. 0.05.
b The theoretical value is the expected value of the synthetic audit sample assuming no preparation
  error and no external effect.
0 Measured mean  values from  the support laboratory for  Lot  14 and Lot 15.
d Objective expressed  in pH units.
* Accuracy expressed as the difference between the index value and mean analytical laboratory values.
                                                    84

-------
Table 24.     Percent Accuracy Estimate* for Laboratory 2 Measurement* of Selected Variable*, In Natural
            Audit Sample*, National Stream Survey - Pha«* I.
Variable
Accuracy
objective
Units (%)
Index value
(n* = 4)
Grand*
mean ±Cl°
Laboratory 2 measured
values (n* = 24)
Percent
Mean +CIC accuracy*
                                                                                         ns
Baalev Lake samples
Al-total                 mg/L        10         0.016       0.0006      0.031    0.0191      94
ANC                  /Jeq/L        10       121.0         0.16       132.0     5.45        9*
BNC                  peq/L        10        29.7         0.52       43.1     11.38       45*
DIC-eq                 mg/L        10         1.52        0.015        1.47    0,105       -3
DIC-init                 mg/L        10         1.52        0.012        1.46    0.097       -4
PH-ANC                 pH units      0.1*       7.06        0.014       6.97    0.036       0.09*
pH-BNC                 pH units      0.1*       7.06        0.014       6.98    0.036       0.08
pH-eq                  pH units      0.1*       7.29        0.014       7.31    0.027       O.O/
                                                                                            *f
Big Moose Lake samples
Al-ext
Al-total
ANC
BNC
DIC-eq
DIC-init
Fe
pH-ANC
pH-BNC
pH-eq
mg/L
mg/L
Aieq/L
/wq/L
mg/L
mg/L
mg/L
pH units
pH units
pH units
10
10
10
10
10
10
10
0.1*
0.1*
0.1*
0.197
0.272
-3.1
72.9
0.11
0.36
0.05
5.10
5.15
5.17
0.0052
0.0045
0.45
1.36
0.009
0.009
0.002
0.010
0.013
0.006
0.210
0.281
-1.4
79.8
0.10
0.23
0.06
5.22
5.24
5.24
0.0082
0.0071
2.50
4.01
0.038
0.035
0.005
0.079
0.079
0.032
7ns
3
-48"s
8
-9
-38*
20*
*/
0.12 '
* f
0.09 '
* f
0.07 '
 * n  = number of measurements.
 * Grand mean based on weighted mean values from four laboratories.
 0 ± CI =  95  percent confidence interval.
 d For pH,  accuracy is expressed as  the absolute difference between the index value and the measured value.
   *  =  mean value significantly different from the index value at p s 0.05.
   ns =  Accuracy estimate outside data  quality objective, but mean value is not significantly different
        from  the index value at p = 0.05.
 * Objective expressed in pH units.
 ' Accuracy expressed as the difference between the index value and the mean analytical laboratory values.
           For  three  variables  (extractable
  aluminum,  specific  conductance,   and  phos-
  phorus), accuracy estimates were outside the
  data quality objectives for one or both lots, but
  these estimates do not represent data quality
  problems.    The mean  value for  extractable
  aluminum in Lot 15  (0.015  mg/L) is in agree-
  ment  with the  mean  value  measured  at
  Laboratory 1 (0.015 mg/L; Table 21).  The accur-
  acy estimate  based on the  Big  Moose  Lake
  sample (7 percent;  Table  24),  which has a
                                                   higher concentration  of  extractable  aluminum,
                                                   was not statistically significant and  is  less
                                                   than  the  apparent  bias  observed  for Lab-
                                                   oratory 1 (13 percent; Table 22).  At low con-
                                                   centrations, systematic errors  in extractable
                                                   aluminum  measurements will  probably be
                                                   masked by  imprecision  due  to  the  extraction
                                                   process.

                                                         The  apparent bias  in  specific  conduc-
                                                   tance measurements  (11 to 12 percent; Table
                                                 85

-------
23) represents a difference of approximately 2
juS/cm.  This difference is about equal in mag-
nitude,  but  in the opposite direction, to that
observed  for Laboratory 1  (approximately  16
/jS/cm, Table 21).  However, measurements of
specific conductance  from  both  types of the
natural  audit samples are  in agreement with
index values (see Appendix A).  Thus, evidence
for systematic errors in specific conductance
measurements is inconclusive.

        For phosphorus measurements,
mean values observed in the  synthetic audit
samples were consistent for both lots (0.022
and 0.023 mg/L, Table 23)  and similar to the
mean from Laboratory 1 for Lot 14 (0.021 mg/L,
Table 20).  The magnitude of the bias is small
(approximately 0.005 mg/L)  and  may partially
result from  an error in sample preparation  at
the support laboratory.  There  is no strong
evidence to suggest  a data quality problem
with  respect to accuracy for  phosphorus
measurements.

        The mean value for DOC was slightly
greater  (1.1  to 1.3 mg/L) than  the theoretical
value in the synthetic audit sample.   For the
Big Moose Lake sample, which  had  a higher
index value  for DOC (approximately 3.6  mg/L;
see Appendix A), the mean  value for measure-
ments made at Laboratory  2 (n = 24) was 4.1
mg/L (Appendix A).  It appears that measured
values of  DOC from Laboratory 2 may be an
overestimate of  true  sample concentrations.
The magnitude of the apparent bias  is within
the range of background levels  measured  in
field blank samples (Table 17).

        For iron, mean values in  the synthetic
audit  sample (0.04 and 0.02 mg/L) were  lower
than the theoretical value (0.06 mg/L, Table 23)
but were  in agreement with the  mean  value
from Laboratory 1 (0.030 mg/L, Table 21). Data
from  the  Big  Moose  Lake sample (Table 24)
indicate a potential for a  positive bias; the
mean  value (0.06 mg/L) was larger  than the
index value (0.05 mg/L).  The magnitude of the
apparent bias (0.01 mg/L) is small and may be
masked by  measurement  imprecision at low
concentrations.
      For total aluminum and silica, sys-
tematic errors may  have an  effect on data
interpretation. Mean values for total aluminum
measure- ments of the synthetic audit sample
were much greater than the theoretical  for
both lots  (Table 23).   Data from the  Bagley
Lake sample (Table 24), which had an index
value  for  total  aluminum  similar to the
theoretical concentration of the synthetic audit
sample, also indicates a potential for positive
bias of approxi- mately 0.01 mg/L.   The mag-
nitude  of this bias (0.01 to 0.015 mg/L) appears
to  be consistent  over  a  range  of
concentrations; the mean value  for measure-
ments  of the Big Moose Lake sample (0.281
mg/L;  Table 24), which has a much higher in-
dex value for total aluminum (0.272 mg/L), also
indicates a positive bias of approximately 0.01
mg/L.  The bias may result from reagent con-
tamination, as was suggested by the analysis
of blank samples (Table 15) and should only
affect  the use of values of total aluminum at
low concentrations (less than 0.1 mg/L).

       For silica measurements,  an apparent
negative bias was observed for both lots of
synthetic audits (0.11 mg/L to 0.15 mg/L; Table
23).  Data from both  types of natural audit
samples, with  index values for  silica greater
than the theoretical concentration of the syn-
thetic audit sample, also indicated a potential
negative bias  in silica  measurements relative
to the  index values. For the Big Moose Lake
audit sample the index value for silica was 4.48
mg/L,  while the mean  measured  value  for
Laboratory 2 was 3.95 mg/L (see Appendix A).
For the Bagley Lake sample,  the index value
was 9.48  mg/L,  while the mean  measured
value  for Laboratory 2 was  8.90  mg/L (see
Appendix A).

      Accuracy estimates for  ANC  and equi-
librated pH relative to index values were within
or near to the data quality objectives for both
lots (Table 21).  For equilibrated DIG, the mean
value for Lot  14 was not significantly different
from the index value, while the accuracy esti-
mate for Lot 15  was  within the data quality
objective.
                                          86

-------
         Data for DIG  and pH  indicate that
the   synthetic   audit  samples   (Table  23)
may have  increased their  dissolved  carbon
dioxide concentration to a greater degree at
Laboratory 2 than  at  Laboratory  1  (Table
21). Mean values for initial DIG measurements
made at Laboratory 2 were generally 0.1 to 0.2
mg/L  greater than those from Laboratory 1,
while pH-ANC and pH-BNC  measurements
were approximately 0.3 pH units lower.

        Data from the Bagley  Lake audit
sample  (Table 24) indicate that accuracy es-
timates  for initial  DIG   (-4 percent),   equi-
librated   DIG (-3 per- cent), pH-ANC (-0.09 pH
units), pH-BNC (-0.08 pH units) and pH-eq (0.0
pH units) measurements were within the data
quality objectives for accuracy with respect to
the index values.   For the  Big  Moose Lake
sample,  mean values   for equilibrated  DIG
(0.10 mg/L; Table 24) were within the accuracy
objectives. The mean for initial DIG measure-
ments (0.23 mg/L), although outside the data
quality  objective  with  respect  to the  index
value, was in agreement with the mean value
for Laboratory 1 (0.26 mg/L; Table 22).  Mean
values  for  pH  measurements from the Big
Moose  Lake sample (Table  24) were 0.07 to
0.12 units higher than the index value, although
they were  still near or within the data quality
objectives with respect to the index value.

         BNC measurements from Laboratory
2 appear to overestimate actual sample  con-
centrations significantly at low concentrations.
For the synthetic audit  sample,  mean values
for both lots were  approximately 1.5 and 1.8
times greater than index values based on sup-
port laboratory measurements (Table 23). For
the Bagley Lake  sample, which had an index
value for BNC similar to that of  the  synthetic
audit sample, the mean value (43.1 jueq/L) was
much greater than the index value (29.7 /Lieq/L;
Table 24).  For  the Big  Moose Lake  sample,
which had a higher index value for BNC (72.9
/jeq/L), the accuracy estimate for Laboratory 2
of 8  percent (Table 24)  was within  the  data
quality objective.

         In conclusion,  systematic errors  in
measurements of total  aluminum, BNC, DOC,
iron, and silica may affect the interpretation of
analytical data from Laboratory 2.  For all of
these  variables, the  suspected  systematic
errors will be most influential at low concentra-
tions.

Percent Accuracy Estimates for the
Processing Laboratory

     Accuracy estimates  for  six variables
measured  in synthetic and  natural audit
samples at the processing  laboratory are
presented in Tables 25 and 26. For the syn-
thetic audit  samples, the theoretical concentra-
tion of total monomeric  and nonexchangeable
monomeric  aluminum should be equal to the
value for extractable aluminum (0.020 mg/L).
However,  the  processing laboratory only
analyzed  field  audit samples,  and accuracy
estimates for the two aluminum fractions will
not be representative, because of the possible
loss of aluminum from the field  audit samples
that was  noted in the introduction  to the
accuracy section. For both lots, mean values
for the two aluminum fractions were less than
the theoretical values.  Mean values for nonex-
changeable monomeric aluminum were greater
than those for total  monomeric aluminum,
probably resulting  from the  high  instrument
background observed in the analysis of blank
samples (Table 15).

       Aluminum concentrations in the Bagley
Lake sample were below the limit of detection.
Data  from  the  Big   Moose Lake  sample
(Table  26)  show  that  the percent accuracy
estimates were  within the  data quality objec-
tives for both aluminum fractions (1 percent for
Al-mono;  5 percent  for Al-nex).  Systematic
errors  do not appear to be evident at higher
concentrations  of  total monomeric or nonex-
changeable monomeric aluminum.

       Specific conductance  measurements
were not accurate during the first half of the
NSS-I,  as  evidenced  by  the   mean  value
observed for Lot 14 (22.7 /^S/cm).  The  mal-
functioning  probe  that  was  used during the
first half of the NSS-I was the source of this
error.

      For  the  closed-system DIG  and pH
measurements,  data   from   both   lots  of
                                           87

-------
  Table 25.
Percent Accuracy Estimates for Processing Laboratory Measurements of Synthetic Audit
Samples, National Stream Survey - Phase I
                                                          Field synthetic audit samples
Lot 14
Accuracy
objectives
Variable
Al-mono
Al-nex
Cond-PL
True color
Units
mg/L
mg/L
pS/cm
PCU
(%)
10
10
5
-
Theo-
retical
value*
0.020
0.020
17.5
NC


n*
7
7
6
8


Mean
0.008
0.014
22.7
5


CIC
0.0031
0.0034
2.08
2


%Accrf
-60*
-30*
30*
-


n*
5
5
4
4


Mean
0.009
0.013
18.8
4
Lot 15


Cl°
0.0065
0.0071
1.88
8



%Acc°'
55*
-35"s
7ns
__
                                         Lot 14
                                                                             Lot 15

Variable
(Units)
DICf(mg/L)
pHT(pH units)
Accuracy Index
objective value* (CI)C
(%) (n = 6)
10 1.14 (0.050)
±0.1^ 7.22 (0.044)

Mean (Cl)c
(n = 9)
1.38 (0.030)
6.96 (0.074)
Index
. value* (Cl)°
%Acc.fl' (n = 6)
21* 1.26 (0.050)
-0.26* 7.29 (0.044)

Mean (Cl)c
(n = 5)
1.44 (0.041)
6.92 (0.078)

%Acc°'
14*,A
-0.37/7
  a The theoretical value is the expected value of the synthetic audit sample assuming no preparation error
   and no external effect.  NC = theoretical value not calculated.
  * n = number of measurements.
  0 One-sided 95% confidence interval.
  d Percent accuracy.
   *  =  mean value significantly different from the theoretical or index value  at p <. 0.05.
   ns =  percent accuracy outside accuracy objective, but mean value is not significantly different from
         the theoretical or index value at P = 0.05.
  e Index  value  = mean value from support  laboratory measurements.
   Closed-system measurement.
  9 Accuracy expressed in pH units.
  h Accuracy expressed as the difference between the index value  and the mean analytical laboratory values.
synthetic audit samples show the same trend
as was observed for the analytical laboratory
measurements of DIG and pH (Tables 21 and
23):   a  DIG  concentration that  was greater
than the index value and a pH value that was
lower than the index value.  For DIG and pH,
the index values for the Bagley  Lake sample
were  calculated based on  measurements col-
lected at  the  processing laboratory during the
NSS-I and the Western Lake Survey-Phase  I.
For the  Big  Moose Lake  sample, the  index
value was based on measurements conducted
during the NSS-I and  four seasonal  studies
that  were conducted  during Phase II  of the
Eastern Lake  Survey.  Data for closed-system
DIG and pH measurements from both natural
audit  samples  do  not indicate systematic
errors that exceed data  quality  objectives,
based on comparisons to index values.
                                    Interlaboratory Bias

                                         An evaluation by Edland et al. of the rela-
                                    tive bias  between  analytical  measurements
                                    from the two laboratories involved in the NSS-I
                                    is  presented  in  Appendix  D.    Their  report
                                    presents four different variations of a  linear
                                    model to describe possible functional relation-
                                    ships of interlaboratory bias to concentration.
                                    The general form of the model is:

                                        Laboratory 1 measurements =
                                        (1 + j3)  (Laboratory 2 measurements)  + a.

                                    where a is a constant, analogous to the inter-
                                    cept of a  regression equation,  and j8 is a
                                    proportionality term, which represents the
                                    deviation of the slope of a regression equation
                                    from 1.0.
                                             88

-------
  Table 26.  Estimates of Percent Accuracy for Analytes Measured at the Processing Laboratory Based on
           Natural Audit Samples, National Stream Survey - Phase I

                                                          Bagley Lake
Index value*

Variable
Al-mono
Al-nex
DIG'
PH'

Units
(mg/L)
(mg/L)
(mg/L)
(pH units)
Accuracy
objective
10
10
10
±OAg

n*
2
2
2
2

Grand
mean0
-
-
1.67
7.04

Cld
-
-
0.068
0.066

n*
-
—
27
27
Measured values

Mean
-
—
1.70
6.99

CIrf %Acca
..
»
0.024 2
*
0.019 -0.05
                                                           Big Moose Lake
Index value*

Variable
Al-mono
Al-nex
DIG'
pH'

Units
(mg/L)
(mg/L)
(mg/L)
(pH units)
Accuracy
objective
10
10
10
±0.1*

n*
2
2
2
2

Grand
mean"
0.193
0.053
0.55
5.14

CI«
0.0174
0.0178
0.061
0.060

7*
24
24
27
27
Measured values

Mean
0.195
0.058
0.024
5.15

CI*
0.0049
0.0054
0.53
0.023

%Acce
1
5
4
0.01
   *  Index values = mean value from support laboratory measurements.
   b  n = number of measurements.
   c  Grand mean based on weighted means of measurements from the processing lab during two surveys (NSS-I
     and Phase  II of the Eastern Lake Survey).
   d  One-sided 95% confidence interval.
   0  %Acc = percent accuracy expressed in pH units.
     * = mean value significantly different  from the index value at p <. 0.05.
   '  Closed-system measurements.
   3  Accuracy expressed in pH units.
         The model was evaluated using  the
seven  groups  of  audit  samples  that  were
measured by   both   analytical  laboratories
(two   lots  of  synthetic  samples and two
natural audit samples,  each prepared  both
as field  and as laboratory samples).
Laboratory  1 did  not  analyze any  field  audit
samples from Lot 15.   For the models, a and
j3  were  derived by using maximum likelihood
estimation techniques. The four  variations of
the model evaluated were:
      1. No bias (a = 0, ]8 = 0)
      2. Constant bias (a + 0, /3 = 0)
      3. Proportional bias(cr = 0, j3 ± 0)
      4. Bias that includes both a constant
        and proportionality term
        (a±Q,p* 0).

The assumption  of linearity was also
evaluated.

      The  results   of these  evaluations,
presented  in  Appendix  D,   indicate   that
                                             89

-------
interlaboratory bias of some type  is present
for most of the variables;  the report provides
the information required to transform the data
so that the two  laboratories are  calibrated.
However,  the authors present  several con-
siderations which  should  be weighed before
applying  the  transformation procedures.   In
many cases, the audit samples do not bracket
a  large  portion  of  the  range of  values
measured  in routine  streamwater  samples;
thus  there may  be considerable uncertainty
introduced if the model is  used to extrapolate
findings beyond the range represented by the
audit samples. A second consideration is that
for a number of  variables the assumption of
linearity was not confirmed. Finally, any gain
in  accuracy achieved by transforming the data
will  be  accompanied by a loss in precision
because  of the uncertainty associated with
estimating the model parameters a and jS.

        Because accuracy estimates for most
variables from both laboratories were  within
the data quality objectives, transformation of
the data  is not necessary  to make  population
estimates.   The  estimates of among-batch
precision presented later in this report include
the effects of interlaboratory bias,  and thus
they can be used to evaluate uncertainty.

        Other data interpretation  activities,
however, may require laboratory measure-
ments  to be  intercaiibrated.  For such activ-
ities,  Edland et ai. (Appendix  D) provide the
appropriate transformation coefficients and
also the  deviation  of the maximum likelihood
estimates.   Summary data for  audit sample
measurements are presented in Appendix A.
Estimates for accuracy presented in  the
preceding sections provide information that
may be useful in deciding how to proceed with
intercalibration (e.g., whether to  correct data
from one laboratory to be more equivalent to
data from the other, or to correct  both data
sets to some intermediate value).

Discussion and Summary; Accuracy

        For nearly a!l variables, accuracy esti-
mates  for measurements at Laboratory 1 were
.within the data quality objectives. Phosphorus
measurements made  by  Laboratory 1 during
the last  half of the NSS-I may be underes-
timates of the true sample concentrations.  It
is  not known  whether  this  apparent bias
occurs at  lower concentrations or if it  is
present only at higher concentrations. In addi-
tion,  specific conductance  measurements
throughout the NSS-I may be underestimates
of true sample values.   The apparent magni-
tude of the bias in audit samples was approxi-
mately -2 /L/S/cm.  It is not known if the bias is
constant  over the entire range of measured
values,  because  the audit samples all  had
specific  conductances  between  10  and  25
/^S/cm.    Interpretation of data  from  dilute
samples  should  consider  the potential  for
negative bias.

      For Laboratory  2, measurements of
several variables  in the  audit samples  had
potential  systematic errors.  In each case, the
bias appeared to be constant over the range of
values measured  in the audit samples;  thus,
the bias  will have the greatest impact in inter-
preting data at low concentrations.  For total
aluminum, a positive bias on the order of 0.010
mg/L  was observed.   This should have little
impact on data interpretation, as most  of the
streamwater  samples  will have  much  higher
concentrations.

      For iron,  the observed bias  in  audit
samples  was  on  the order  of 0.02  mg/L.
At low concentrations, iron is not an important
contribution to the overall ion balance, and the
potential  systematic error does not affect data
interpretation.  For DOC, the observed bias in
audit  sample measurements was between 0.1
and 0.5 mg/L  and may only need to be con-
sidered when  comparing groups of samples
having low  concentrations of DOC.   Silica
measurements from Laboratory 2 may be
underestimates of true  sample values,  based
on audit sample measurements. The observed
magnitude of the bias was on the order of 0.15
mg/L  at  low  concentrations (1 mg/L)  to  0.5
mg/L  at higher concentrations (3 to 10  mg/L).
Interpretation  of  silica  data should consider
the possibility of systematic errors.

      Measurements of BNC in audit samples
having low concentrations (less than 30 fieq/L)
were subject to systematic errors and also to
                                          90

-------
poor precision,  particularly at Laboratory 2.
The usefulness of the BNC data may be limited
because of the method of measurement.  For
the NSS-I, the primary purpose  of  the BNC
measurement  was  to provide a means  of
verifying the ANC measurement on a sample-
by-sample basis.  Because BNC is affected by
dissolved  carbon dioxide,  special precautions
must be taken to yield accurate results.  Titra-
tion with a strong base in  a vessel exposed to
the atmosphere will cause the sample to con-
tinually increase its  dissolved carbon dioxide
concentration. If future studies require reliable
BNC data  to  differentiate weak  and  strong
acids in natural water samples,  the method
should be modified so that the titration is con-
ducted  under an inert,    carbon-dioxide-free
atmosphere.

        The synthetic audit samples used for
the NSS-I  provided a reliable means to assess
accuracy for most variables.   However, this
sample, if not filtered and preserved imme-
diately after preparation, was affected by the
loss of aluminum  and  iron.    In  addition,
theoretical  values for variables  affected  by
dissolved carbon dioxide,  such as ANC, BNC,
DIC, and pH, are difficult to calculate and may
not be  valid  because the  preparation of the
synthetic  audit  sample composition  is  not
necessarily  based  on  the  assumption of
carbonate equilibrium. In addition, the concen-
trations of these four variables can be affected
by preparation errors arising from adding other
chemical constituents to  the synthetic audit
sample.  It may  be desirable to allow the
synthetic  audit  sample to  equilibrate  for a
period  of  time after preparation,  but before
use. Following equilibration, the sample would
be subjected to rigorous verification measure-
ments  that  include  comparisons  to certified
standards so that the audit sample composi-
tion is  known with a high level of  certainty for
all variables.   This kind of verification  should
also be made for the natural audit  samples.
Verified natural audit samples would offer a
means to  assess overall measurement accu-
racy including systematic errors resulting from
sample collection and handling.

        As an alternative,  synthetic  audit
samples could be prepared on  an analyte-by-
analyte basis or as  aSiquots that include
chemically compatible  variables.    Such
samples would provide reliable  concentrations
for all variables to assess absolute accuracy,
but they also  would have  the disadvantage
that they could only be introduced at the point
of analysis.  In addition, assessments of ac-
curacy for measurements  of these kinds of
samples would not include  errors  due to  col-
lection and handling.

      Finally, if interiaboratory  bias is a con-
cern,  analytical objectives for accuracy should
be  developed  to control interiaboratory  dif-
ferences.   An accuracy objective of ±10 per-
cent allows for interiaboratory  differences of
up to 20 percent.  It may be desirable to estab-
lish an accuracy objective at +5 percent, thus
reducing  the tolerable  interiaboratory dif-
ference to ±10 percent.  Coincident with such
a reduction,  the  laboratories should  be  peri-
odically provided with known performance
standards  from a single source so  that  all
laboratories can  calibrate their measurement
systems to a given target value, rather  than
relying on  internally developed standards to
monitor their  performance.   Samples  of
unknown composition  would then provide  an
independent check on laboratory performance.

Precision

      The precision (or variation)  associated
with various components of the collection and
measurement system was assessed by using
the results from  analyses of processing and
analytical laboratory duplicate,  field duplicate,
and audit samples.  Table 27  illustrates  the
potential sources  of variation for each kind of
sample and  those  components  of variation
that are included  in an estimate of  precision
based on each type of QA sample.  Processing
and analytical  laboratory duplicate samples
provided an estimate of within-batch analytical
precision.   The DQOs for precision (Table 13)
were based on the expected analytical perfor-
mance at a single laboratory; laboratory dupli-
cate samples were  used to assess  precision
relative to the DQOs.  Field duplicate samples
provided estimates of overall within-batch
precision,   including  sample collection,  han-
dling, and processing and analytical errors.
                                           91

-------
  Table 27.
Components of Variance Included In Precision Estimates From Routine-Duplicate Pairs and
Audit Samples, National Stream Survey - Phase I
                                                       Type of sample
      Sources of
  measurement error
                      Field
                     duplicate
Laboratory
 duplicate
Field
audit*
Laboratory
  audit*
 Sampling (system-level)
      Among crews
      Day-to-day
      Among samples
      Sampling variance
                       X
                       X
Processing (aliquot preparation
and preservation)
Among laboratories
Among batches
Within a batch x
Subsampling x
Analysis (method-level)
Among laboratories
Among batches
Within a batch x
Within a sample x


X
X

X

X
X
X
x x







x
x

X
 * Field and laboratory audits also have variance components associated with their preparation
Audit samples provide  estimates  of among-
batch precision that include the effects of day-
to-day differences within  a  single laboratory,
effects  from processing and  transport,  and
also the effects due to interlaboratory biases.

         None of the samples listed in Table
27  provide  an  estimate  of  total overall
variability due to the collection and  measure-
ment of samples.  Precision estimates based
on  duplicate  samples  do  not account for
among-batch variation.  Audit samples provide
an  estimate  of among-batch  and among-
laboratory variation,  but this  estimate does
not include  any effects of sample collection.
Precision esti-mates from routine-duplicate
sample  pairs  are based on pooled  measure-
ments for ranges of concentrations; estimates
from audit samples  are  based on  repeated
measurements at a single concentration.
                                  Nevertheless, qualitative comparisons are
                                  possible to discern the major sources of varia-
                                  tion in the NSS-I collection and measurement
                                  system and to evaluate the effect of impreci-
                                  sion on data interpretation.

                                  Precision Estimates Derived from Field
                                  and Laboratory Duplicate Samples

                                        Because of the  range  of  values for
                                  chemical variables encountered  during the
                                  NSS-I and concentration-dependent effects
                                  (Mericas and Schonbrod, 1987), a  single esti-
                                  mate of precision based on all pairs of routine-
                                  duplicate samples may be misleading.   This
                                  analysis did not use model-based approaches
                                  to predict precision as a function of concentra-
                                  tion  (e.g.,   Mericas   and  Schonbrod,  1987)
                                  because data from duplicated measurements
                                  violated several   assumptions  of  regression
                                           92

-------
analysis  (e.g.,  concentrations were not
measured without error, measured pairs were
distributed  over the  entire measurement
range).   Because precision  varies with  con-
centration,  precision  was  estimated from
routine-duplicate measurements for several
ranges of values. Examination of scatterplots
of  the standard deviation versus the mean
concentration  of sample  pairs  within each
range subset indicated  that there are no rela-
tion ships between variance and concentration.

         A total of 65  field routine-duplicate
pairs and 68 laboratory routine-duplicate pairs
were analyzed for most  variables.  Oc-
casionally, a routine-duplicate pair was not in-
cluded for a batch of samples.  On  some oc-
casions, an analytical laboratory analyzed
more than one sample in duplicate. In the lat-
ter cases the first pair was  used in precision
estimates. Measurements that were qualified
with an X flag were not included  in precision
estimates.

         Precision  estimates for  each range
subset  were  based on a  pooled  standard
deviation that was  calculated from  the mean
and  var-iances of the individual sample pairs.
The  following formula was used to calculate a
pooled standard deviation (Taylor, 1987) with
the number of replicates in each case equal to
two:
                                RSDp =
                            1100
 where   s.
          2   _
pooled standard
deviation,
number of routine-
duplicate pairs,
variance of a routine-
duplicate pair.
 For ANC and BNC, expressing precision in rela-
 tive terms can be misleading for values less
 than 100 //eq/L Except for ANC, BNC, pH, and
 true color, precision was also expressed as a
 relative pooled standard deviation (%RSD ) by
 dividing s  by the grand mean of the sample
 pairs and multiplying by 100:
                                                n_
                                                Zx/n
  where  s  =    pooled standard
                 deviation,
         Xj  =    mean of a sample pair,
         n  =    number of sample pairs.

Method-Level Precision Estimates-

   Estimates  of  pooled standard  deviations
and pooled  relative standard  deviations for
subranges of variables are presented in Table
28. These estimates are for combined meas-
urements from  both analytical laboratories.
Examination of method-level (within-batch)
precision  estimates  for  each  laboratory,
although not presented in detail here, indicated
that for all variables neither  laboratory was
outside the within-laboratory precision objec-
tives (Table 13). The DQOs  for precision were
established initially for  measured values
greater than 10 times the detection limit objec-
tive (Table 13). For all subranges greater than
the detection limit objective, within-batch
precision   estimates   for  the  combined
measurements  (Table   28)  were  also  within
the DQOs for precision for all variables.

System-Level  Precision Estimates
from  Field Duplicate Samples-

   Because  DQOs  were not established for
system-level precision for the  NSS-I, the
method-level DQOs were used as  a gauge.
Measurements  of field  routine-duplicate
sample pairs from both analytical laboratories
were   combined  to estimate system-level
(overall within-batch) precision for the  NSS-I
Table 28). Eighteen variables had sp or %RSDp
estimates that were within  or near the within-
laboratory precision goal for all subset ranges
except the  lowest,  which represented values
below the system decision  limit for most vari-
ables.  These variables  were total monomeric
aluminum (largest %RSD = 12.4),  ANC  (larg-
est s   = 11.22 jueq/L), BNC (largest s = 12.25
      , calcium (largest %RSD  = 5.6), chloride
                                           93

-------
Table 28.    Method-Level and System-Level Precision Estimates by Concentration Ranges of Variables
            (Laboratories Pooled), National Stream Survey - Phase I
Method-level precision
(processing and analytical
laboratory duplicate pairs)
Variable (units)
and measurement
range
Al-ext (mg/L)
<0.007
0.007 to 0.050
0.050 to 0.100
>0.100
All data
Al-total (mg/L)
<0.027
0.027 to 0.100
0.100 to 0.500
0.500 to 1.000
> 1.000
All data
Al-mono (mg/L)
<0.015
0.015 to 0.100
0.100 to 0.500
0.500 to 1.000
> 1.000
All data
Al-nex (mg/L)
<0.023
0.023 to 0.100
0.100 to 0.500
All data
ANC (fJ&q/L)
<0
0 to 50
>50
All data
BNC (peq/L)
0 to 50
>50
All data
Number
of pairs

3
35
15
15
68

9
6
51
1
1
68

33
24
7
2
1
67

52
12
3
67

2
4
62
68

24
44
68
Grand
mean

0.004
0.019
0.071
0.269
0.085

0.014
0.056
0.204
0.791
1.235
0.189

0.010
0.026
0.273
0.534
1.908
0.087

0.014
0.043
0.225
0.029

-6.8
21.9
401.2
366.9

32.6
93.4
71.9
Pooled
s

0.0002
0.0009
0.0026
0.0055
0.0029

0.0018
0.0034
0.0042
0.0304
0.0573
0.0087

0.0011
0.0017
0.0033
0.0045
0.0030
0.0019

0.0022
0.0033
0.0139
0.0039

0.70
1.05
4.42
4.23

2.68
6.72
5.64
%RSDp*

5.3
4.6
3.6
2.0
3.4

13.1
6.1
2.0
3.8
4.6
4.6

11.0
6.4
1.2
0.8
0.2
2.1

15.8
7.6
6.2
13.4

..
—
..
-

-
-
—
System-level precision
(field routine-duplicate pairs)
Number
of pairs

29
25
4
7
65

4
16
38
7
0
65

30
29
6
0
0
65

47
18
0
65

7
9
49
65

28
37
65
Grand
mean

0.003
0.016
0.072
0.278
0.042

0.019
0.056
0.214
0.767
..
0.222

0.010
0.031
0.283
..
	
0.045

0.016
0.042
..
0.023

-42.4
27.9
425.0
319.7

36.2
111.1
78.8
Pooled
s

0.0016
0.0059
0.0021
0.0166
0.0067

0.0035
0.0079
0.0807
0.1953
__
0.0905

0.0022
0.0038
0.0049
„
„
0.0033

0.0031
0.0068
..
0.0045

2.94
7.83
11.22
10.22

6.38
12.25
10.14
%RSDp*

51.5
37.2
2.9
6.0
16.1

18.7
14.1
37.8
25.5
„
40.7

21.4
12.4
1.7
	
__
7.4

19.2
16.2
__
19.2

„
__
„
..

__
„
»
                                                                                             (Continued)
                                                  94

-------
Table 28.    (Continued)
Method-level precision
(processing and analytical
laboratory duplicate pairs)
Variable (units)
and measurement
range
Ca (mg/L)
0.02 to 1.00
1.00 to 5.00
5.00 to 10.00
>10.00
All data
Of (mg/L)
<0.03
0.03 to 1.00
1.00 to 2.00
2.00 to 5.00
5.00 to 10.00
> 10.00
All data
Cond-PL (/^S/cm)
<25.0
25.0 to 50.0
50.0 to 100.0
> 100.0
All data
Cond-lab (/jS/cm)
<25.0
25.0 to 50.0
50.0 to 100.0
>100.0
All data
DIC-closed (mg/L)
<1.00
1.00 to 2.00
2.00 to 5.00
5.00 to 10.00
> 10.00
All data
DIC-eq (mg/L)
<0.23
0.23 to 1.00
1.00 to 2.00
2.00 to 5.00
5.00 to 10.00
> 10.00
All data
Number
of pairs

6
35
12
15
68

2
13
13
23
11
6
68

5
22
21
19
67

14
20
17
17
68

9
8
26
18
7
68

1
4
8
24
18
13
68
Grand
mean

0.59
2.37
7.06
25.59
8.16

0.00
0.57
1.49
3.18
6.73
15.89
3.96

20.1
37.8
75.0
244.3
106.7

19.9
35.6
69.2
206.9
83.6

0.60
1.56
3.42
6.50
31.79
6.56

0.07
0.74
1.69
3.60
7.27
19.21
7.11
Pooled
s

0.009
0.041
0.034
0.529
0.251

0.000
0.006
0.022
0.045
0.077
0.336
0.108

1.11
1.94
5.27
16.99
9.59

0.12
0.23
0.29
0.49
0.31

0.025
0.040
0.068
0.168
0.236
0.123

0.002
0.029
0.031
0.067
0.086
0.639
0.286
%RSDpa

1.5
1.7
0.5
2.1
3.1

-
1.1
1.5
1.4
1.1
2.1
2.7

5.5
5.1
7.0
7.0
9.0

0.6
0.6
0.4
0.2
0.4

4.2
2.5
2.0
2.6
0.7
1.9

2.9
4.0
1.8
1.9
1.2
3.3
4.0
System-level precision
(field routine-duplicate pairs)
Number
of pairs

4
35
19
7
65

0
8
18
24
6
9
65

5
24
24
12
65

6
24
24
11
65

7
9
31
13
5
65

10
12
12
22
5
4
65
Grand
mean

0.56
2.55
6.82
28.90
6.51

-
0.69
1.53
3.05
7.46
18.21
4.85

20.4
37.7
70.1
192.0
76.8

20.6
36.9
69.9
197.4
74.7

0.46
1.43
3.37
6.65
24.47
5.07

0.12
0.55
1.50
3.49
6.90
26.22
3.72
Pooled
s

0.015
0.057
0.090
1.604
0.530

—
0.018
0.033
0.043
0.073
0.734
0.276

0.26
1.26
1.32
3.12
1.75

0.87
0.27
0.42
1.22
0.64

0.020
0.063
0.092
0.184
0.365
0.147

0.054
0.099
0.162
0.261
0.184
1.050
0.317
%RSDpa

2.7
2.2
1.3
5.6
8.1

—
2.6
2.2
1.4
1.0
4.0
5.7

1.3
3.3
1.9
1.6
2.3

4.2
0.7
0.6
0.6
0.9

4.3
4,4
2.7
2.8
1.5
2.9

45.3
18.1
10.8
7.5
2.7
4.0
8.5
                                                                                                (Continued)
                                                    95

-------
Table 28.    (Continued)
Method-level precision
(processing and analytical
laboratory duplicate pairs)
Variable (units)
and measurement
range
DIC-init (mg/L)
<0.23
0.23 to 1.00
1.00 to 2.00
2.00 to 5.00
5.00 to 10.00
> 10.00
All data
DOC (mg/L)
<0.5
0.5 to 2.0
2.0 to 5.0
5.0 to 10.0
>10.0
All data
F- (mg/L)
0.010 to 0.050
>0.050
All data
Fe (mg/L)
<0.02
0.02 to 0.05
0.05 to 0.10
0.10 to 0.50
0.50 to 1.00
>1.00
All data
K (mg/L)
<0.15
0.15 to 0.35
0.35 to 0.45
>0.45
All data
Mg (mg/L)
<1.00
1.00 to 2.00
2.00 to 5.00
>5.00
All data
Number
of pairs

0
1
17
18
18
14
68

4
27
16
14
7
68

29
39
68

2
3
11
36
3
12
67

1
10
2
55
68

26
21
15
6
68
Grand
mean

-
0.24
1.60
3.42
7.22
18.82
7.09

0.3
1.2
3.3
7.2
17.2
4.5

0.037
0.121
0.085

0.01
0.02
0.08
0.19
0.72
6.85
1.38

0.11
0.26
0.39
1.47
1.24

0.54
1.47
3.42
18.15
3.02
Pooled
s

—
0.007
0.036
0.111
0.139
0.780
0.366

0.02
0.05
0.11
0.15
0.15
0.10

0.0000
0.0021
0.0016

0.001
0.001
0.002
0.004
0.008
0.114
0.048

0.001
0.006
0.002
0.015
0.014

0.004
0.013
0.033
0.158
0.050
%RSDp*

_
2.9
2.3
3.2
1.9
4.1
5.2

6.5
4.0
3.3
2.1
0.9
2.3

0.0
1.7
1.9

7.4
2.7
3.0
2.2
1.2
1.7
3.5

1.3
2.5
0.6
1.0
1.1

0.8
0.9
1.0
0.9
1.7
System-level precision
(field routine-duplicate pairs)
Number
of pairs

3
10
15
26
6
5
65

2
41
15
7
0
65

51
14
65

14
18
10
18
3
2
65

1
5
6
53
65

19
23
18
5
65
Grand
mean

0.17
0.56
1.50
3.39
6.17
24.46
4.24

0.30
1.20
3.10
7.20
_
2.23

0.033
0.082
0.044

0.01
0.03
0.08
0.22
0.64
1.55
0.16

0.03
0.26
0.40
1.45
1.24

0.68
1.48
3.04
9.39
2.29
Pooled
s

0.014
0.091
0.211
0.250
0.050
1.009
0.339

0.02
0.24
0.44
0.58

0.34

0.0011
0.0029
0.0017

0.003
0.009
0.029
0.117
0.163
0.377
0.098

0.003
0.003
0.005
0.063
0.056

0.013
0.014
0.058
0.242
0.075
%RSDp*

8.2
16.3
14.0
7.4
0.8
4.1
8.0

6.1
20.8
14.3
8.1

15.4

3.4
3.5
3.8

66.0
34.0
34.3
53.3
25.6
24.4
61.4

8.3
1.0
1.3
4.3
4.5

2.0
0.9
1.9
2.6
3.3
                                                                                              (Continued)

-------
Table 28.    (Continued)
Method-level precision
(processing and analytical
laboratory duplicate pairs)
Variable (units)
and measurement
range
Mn (mg/L)
<0.01
0.01 to 0.05
0.05 to 0.10
>0.10
All data
Na (mg/L)
<0.50
0.50 to 1.00
1.00 to 2.00
2.00 to 5.00
>5.00
All data
NH4+ (mg/L)
<0.02
0.02 to 0.05
0.05 to 0.10
>0.10
All data
N03- (mg/L)
o.ooo
>3.000
All data
P (mg/L)
<0.001
0.001 to 0.005
0.005 to 0.015
>0.015
All data
pH-closed (pH units)
<4.00
4.00 to 5.00
5.00 to 6.00
6.00 to 7.00
7.00 to 8.00
>8.00
All data
Number
of pairs
2
9
12
44
67
4
8
11
33
12
68
12
20
14
22
68
55
13
68
2
24
20
22
68

4
4
7
25
26
2
68
Grand
mean
<0.01
0.03
0.08
1.00
0.68
0.23
0.73
1.47
3.40
10.27
3.80
0.02
0.03
0.07
0.35
0.14
0.722
8.182
2.148
-0.001
0.003
0.008
0.105
0.037

3.75
4.51
5.61
6.63
7.43
8.26
6.59
Pooled
8
0.001
0.001
0.003
0.017
0.014
0.001
0.027
0.017
0.028
0.174
0.076
0.000
0.001
0.002
0.004
0.003
0.0162
0.1212
0.0549
0.0003
0.0003
0.0004
0.0018
0.0010

0.000
0.012
0.041
0.012
0.028
0.010
0.023
%RSDpfl
11.8
2.1
3.5
1.7
2.0
0.6
3.7
1.1
0.8
1.7
2.0
0.0
3.2
3.1
1.2
1.9
2.2
1.5
2.6
30.5
8.6
4.6
1.7
2.8

-
-
-
-
-
-
—
System-level precision
(field routine-duplicate pairs)
Number
of pairs
19
19
7
20
65
4
4
24
25
8
65
33
21
7
4
65
53
12
65
7
31
20
7
65

0
6
13
24
19
3
65
Grand
mean
<0.01
0.02
0.07
0.24
0.09
0.28
0.84
1.42
3.52
9.66
3.14
0.01
0.03
0.07
0.17
0.03
0.686
14.885
3.307
-0.001
0.003
0.008
0.025
0.006

-
4.55
5.69
6.66
7.26
8.45
6.53
Pooled
s

0.015
0.002
0,023
0.015
0.001
0.030
0.028
0.037
0.097
0.045
0.006
0.007
0.014
0.011
0.008
0.0422
0.3239
0.1443
0.0009
0.0016
0.0034
0.0040
0.0026

-
0.032
0.036
0.036
0.036
0.026
0.035
%RSDps
56.6
65.0
2.7
9.5
16.8
0.5
3.6
1.9
1.1
1.0
1.4
42.0
23.4
19.7
6.6
22.8
6.2
2.2
4.4
62.9
53.7
43.8
. 16.1
40.5

-
..
-
-
-
-
—
                                                                                               (Continued)
                                                    97

-------
Table 28.    (Continued)
Method-level precision
(processing and analytical
laboratory duplicate pairs)
Variable (units)
and measurement
range
pH-ANC (pH units)
4.00 to 5.00
5.00 to 6.00
6.00 to 7.00
7.00 to 8.00
>8.00
All data
pH-BNC (pH units)
4.00 to 5.00
5.00 to 6.00
6.00 to 7.00
7.00 to 8.00
>8.00
All data
pH-eq (pH units)
<4.00
4.00 to 5.00
5.00 to 6.00
6.00 to 7.00
7.00 to 8.00
>8.00
All data
Si02 (mg/L)
<0.50
0,50 to 1.50
1.50 to 5.50
5.50 to 10.50
>10,50
Ail data
SO/1 (mg/L)
<0.05
0.05 to 2.50
2.50 to 5.00
5.00 to 12.50
>12.50
Ail data
True color (PCU)
<30
>30
All data
Number
of pairs

1
5
34
28
0
68

0
6
34
28
0
68

2
^
4
3
34
18
68

1
1
23
18
18
63

2
13
18
26
9
68

50
17
67
Grand
mean

4.98
5.73
6.69
7.25
—
6.82

-
5.61
6.70
7.25
-
6.83

3.73
4.55
5.55
6.86
7.55
8.45
7.22

0.44
1.03
3.71
7.73
16.38
8.10

<0,01
1.43
3.60
7.95
22.56
7.21

13
129
43
Pooled
s

0.014
0.021
0.053
0.039
~
0.046

-
0.034
0.042
0.040
-
0.041

0.005
0.009
0.000
0.004
0.019
0.028
0.020

0.007
0.039
0.052
0.141
0,185
0.129

0.002
0.019
0.048
0.117
0.239
0.116

0.9
8.7
4.4
%RSDp*

-
_
...
..
-
..

..
..
-
-
-
-

„
--
-
..
..
-
..

1.6
3.4
1.4
1.8
1.2
1.6

66.7
1.3
1.3
1.5
1.1
1.6

6.5
6.8
10.5
System-level precision
(field routine-duplicate pairs)
Number
of pairs

6
8
27
22
2
65

6
8
26
23
2
65

0
6
4
11
40
4
65

0
1
25
21
18
65

0
18
10
24
13
65

44
21
65
Grand
mean

4.49
5.70
6.63
7.31
8.10
6.59

4.50
5.69
6.63
7.31
8.12
6.60

—
4.53
5.73
6.73
7.48
8.37
7.03

..
0.57
3.31
7.70
14.51
7.78

_
1.40
3.86
8.38
21.28
8.36

13
50
25
Pooled
s

0.022
0.104
0.073
0.091
0.034
0.080

0.035
0.092
0.065
0.086
0.062
0.075

—
0.017
0.041
0.113
0.171
0.083
0.144

..
0.064
0.071
0.122
0.730
0.393

..
0.054
0.081
0.128
0.243
0.140

2.6
1.9
2.4
%RSDp*

..
..
..
„
..
--

..
—
—
..
—
-

..
—
..
..
..
..
..

..
11.3
2.2
1.6
5.0
5.0

„„
3.6
2.1
1.5
1.1
1.7

19.6
3.9
9.6
                                                   98
(Continued)

-------
Table 28.   (Continued)
                            Method-level precision
                           (processing and  analytical
                           laboratory duplicate pairs)
                  System-level precision
              (field routine-duplicate pairs)
Variable (units)
and measurement
range
Turbidity (NTU)
<2.0
2.0 to 20.0
20.0 to 100.0
All data
Number
of pairs
28
38
1
67
Grand
mean
1.1
8.1
33.5
5.6
Pooled
s
0.07
0.26
0.71
0.22
%RSDp*
6.5
3.2
2.1
3.9
Number
of pairs
27
36
2
65
Grand
mean
1.0
7.7
24.0
5.4
Pooled
s
0.08
0.39
1.41
0.38
»SDp.
8.6
5,0
5.9
7.1
 * %RSD = relative pooled standard deviation, calculated as (pooled s grand mean) x 100.
(largest %RSD  = 4.0), specific conductance at
the processing laboratory (largest %RSD   =
3.3),  specific conductance  at  the analytical
laboratory (largest %RSD  = 4.2), closed sys-
tem DIG (largest  %RSD   = 4.4),  fluoride
(largest %RSD  =  3.5),  potassium  (largest
%RSD = 8.3), magnesium (largest %RSDp =
2.6),  sodium  (largest %RSD   = 3.6),  nitrate
(largest  %RSD   =  6.2),  cfosed-system  pH
(largest s   = 0.036 pH units),  silica  (largest
%RSD =11.3), sulfate (largest %RSD   = 3.6),
true color (largest s  = 2.6 PCU), and turbidity
(largest %RSD  =  8.6).  For these variables,
random  variations due  to collection  and
processing  appear to  be minimal at  all con-
centrations.

    For six other variables, estimates  of s  or
%RSD  were not near the within-laboratory
precisian goal for one subrange.   These vari-
ables were extractable aluminum (%RSD  -
37.2  for the  subrange  0.007 to 0.050 mg/L),
nonexchangeable monomeric aluminum
(%RSD  - 16.2 for the subrange 0.023 to 0.10
mg/L), equilibrated DIG (%RSD  - 18.1 for the
subrange 0.23 to 1.00 mg/L), initial DIG (%RSD
»  16.3 for  the subrange 0.23 to 1.00 mg/Lj,
DOC (%RSD  -  20.8 for the subrange 0.5 to
2.0 mg/L), arid manganese (%RSDp - 65.0 for
the subrange 0.01 to 0.05 mg/L).

    For the extractable aluminum values  out-
 side the goal, the mean value (0.016 mg/L) indi-
cates  that  routine-duplicate  pairs   in  this
subrange tended  to be of low concentration.
The s estimate (0.0059 mg/L) is approximately
12 percent of the  upper limit of the subrange,
and thus no data quality problem should exist
for measurements approaching 0.050 mg/L At
lower  concentrations,  precision is probably
affected  by sample-to-sample differences  in
extraction  efficiency  at  the  processing
laboratory.

   For nonexchangeable monomeric alumi-
num, the  majority  of  the duplicate  pairs
appears  to   have  mean  concentrations  in
the  lower  portion of the subrange between
0.023 and  0.100  mg/L (grand mean value  =
0.042   mg/L).   The observed   precision is
similar to  that obtained for other aluminum
measurements in the  same general subrange
of concentrations, and no data quality problem
is indicated.

   For the subranges of the two DIG  deter-
minations that were outside the precision
goals, precision estimates at both laboratories
for low concentrations (less than 1 mg/L) were
outside the precision objective (%RSD  - 16 to
22 for Laboratory 1;  %RSD =  16 for Labo-
ratory 2 for both variables).  However, the sp
value was small (less  than 0.1 mg/L),  and a
data quality problem is not indicated for this
 subrange.

    For DOC,  precision estimates  at both
 laboratories (%RSDp for Laboratory  1  = 27;
                                            99

-------
        for Laboratory 2 = 17) were outside the   outlying pairs are removed, the value for S  for
 Precision  objectives  for  concentrations  less   the subrange improved from 0.171 to 0 001 pH
  han 2.0 mg/L, suggesting an effect of collec-   units.  When all measurements except the two
  ion,  processing,  or  sample preparation,   outlying pairs were considered (n = 63), the s
 htr!±  V" d'gestlon efficiency, sporadic   value  based on  all measurements  improveS
 background contamination,  or carry-over from   from 0.144 to 0.090.  This variation is probably
 «mn11P h     ° ?  9n^°C COntent **  *   ** ***"* of sample-to-sample differences  in
 sample  having a low DOC content would all   the carbon  dioxide equilibration  procedure
 increase the variation of measurements within   Interpretation  of equilibrated PH measure-'
  •eld  routine-duplicate pair  measurements,   ments in the circumneutral range should con-
 interpretatsons   of  DOC  measurements of   sider this variability
 approximately 2.0 mg/L or less should be con-
 ducted with this potential imprecision in mind.      For total aluminum (largest %RSD   = 37.8),
    Por t.                                      iron (largest %RSD  =  53.3), ammonium
 nnt   thte+man9anese measurements that did   (largest %RSDp =  £.4), and  phosphorus
 not meet  trie precision  objective,  the grand   (largest %RSDD P= 53.7), %RSD  estimates for
 d.tPnt-r  w-(th°  T-9/L) 1S "ear the Hmit °f   most or a" s&ranges were  not near within-
 detection  Within this subrange (0.01 to 0.05,   laboratory precision objectives.  For total
 Table  27)  one pair  of measurements had  a   aluminum,  %RSDn estimates  for both
 large  d.fference (0.019 mg/L) between the   laboratories for concentrations  less  than 05
 measurements.  Although removal of this pair   mg/L were not near the precision objectives
 improved precision estimates by approximately   Sample-to-sample    differences  in  digestion
 25 percent, the precision within this subrange   appear to  be the most likely source of the
 was still  well outside the  within-laboratory   variance for total aluminum.  For iron  %RSD
 precision goals, indicating that measurements   values from both laboratories were  also no?
 in this subrange are not reliable. Interpretation   near the precision objectives.  For these two
 of  streamwater  concentrations  in this  sub-   variables there appears  to be a substantial
 range  should be conducted with this variability  effect from sample preparation, processing or
 tn mind-                                      collection on precision.

   The three pH measurements determined at     Both laboratories also had  %RSD  esti-
 Ir^fnao£'Cai laboratories nad SP estimates   mates  for ammonium measurements  f8r con-
 (Table 28) that  were  not  near  the within-   centrations less than 0.05 mg/L that  were not
 laboratory  precision  goals  for several sub-   near  precision  objectives.  At concentrations
 ranges. For pH-ANC and  pH-BNC,  s values   between 0.05 and 0.10 mg/L, both laboratories
 were generally less than ± 0.1 pH unif Other   had a %RSDD estimate of approximately 23 to
 studies of pH measurements  (Tyree,  1981;   24 percent.  Precision estimates of %RSD  for
 Davison and  Gardner,  1986) report errors of   phosphorus  measurements at  both la'bor-
 this  magnitude or  greater within a laboratory in   atories were not near precision objectives for
 samples of low ionic strength unless stringent   any subrange. Measured values between 0 001
 protocols  are  followed to minimize pH  elec-   and 0.005 are close  to the limit  of detection
 trode errors.   Thus,  there  is no reason to   and below  the  system decision limit (0007
 SU!Pe?0aK,    quality problem in  the pH"ANC   m9/L)-  At concentrations between 0.005 and
 and  PH-BNC measurements.  Equilibrated pH   0.015 mg/L, precision was probably affected by
 measurements in  the circumneutral range  (pH   background contamination or  sample car-
   ? t0^8'°°Lhad relative|y h|9h s  esti-   ryover.   Good  measurement  precision for
mates  (0.113 to 0.171 pH units). Two  sample   ammonium  and phosphorus may  not be
pairs in the 7.00 to 8.00 subrange had large dif-   achievable  at these concentrations unless
ferences  (0.8 to  approximately 1.0 pH units)   special  precautions are taken during  sample
between the routine  and duplicate measure-   collection,  processing,  and preparation  for
ments, resulting in large estimates of variance   analysis.  Users of phosphorus data for the
(5 to 8 times the next largest value). When the   NSS-I should consider this variability
                                          100

-------
Precision Estimates Derived From
Audit Sample Measurements

      For each type of audit sample (synthetic,
Bagley Lake,  and Big Moose Lake), individual
measurements  from both  laboratories  were
combined to  provide  estimates of among-
batch precision.   For most variables, among-
batch precision was estimated as  the  stan-
dard deviation of the combined audit sample
measurements.   Precision  estimates for tur-
bidity were not determined  because the  audit
samples were  filtered.   Summary statistics
based on the combined audit sample meas-
urements are presented in Table 29.

      For the synthetic audit samples, meas-
urements of field and laboratory audit samples
from   both preparation lots were combined,
except  for  extractable  aluminum,   total
aluminum, and iron.  Field audit and laboratory
audit  measurements were not significantly dif-
ferent for  other  variables  (see  preceding
Accuracy  subsection) and  thus  processing
effects were minimal and did not contribute to
among-batch variability.  For extractable and
total  aluminum,  the difference between lots
was large enough to possibly inflate the  preci-
sion estimate artificially; thus, precision esti-
mates were calculated for each lot. For extrac-
table  aluminum, total aluminum, and iron, only
laboratory  audits were used because of the
loss  of aluminum and  iron from field  audit
samples due to  adsorption or  precipitation.
For the natural audit samples,  field and
laboratory samples were combined for all vari-
ables.

      For each type of audit sample,  several
variables were present in very low concentra-
tions, and analytical  imprecision near the limit
of detection  is probably more influential than
among-batch precision. For the synthetic audit
sample,  mean  values  for total monomeric
(0.010 mg/L) and  nonexchangeable monomeric
aluminum (0.015 mg/L) were near the limits of
detection  (Table  16).  In addition,  the  mean
value for iron (0.03  mg/L) was also low, near
the system decision limit  (0.017 mg/L,   Table
17).  For the Bagley Lake audit sample,  mean
values for extractable aluminum  (0.007 mg/L),
total monomeric aluminum (0.01 mg/L), nonex-
changeable aluminum (0.014 mg/L),  iron (0.01
mg/L),  manganese (<0.01 mg/L), ammonium
(0.01 mg/L),  nitrate (0.012 mg/L), and phos-
phorus (0.001 mg/L) were near or below limits
of detection  (Tabie 16).  Mean values for total
aluminum (0.027 mg/L), BNC (35.7 jueq/L), and
DOC (0.4 mg/L) were near to the system deci-
sion limits (Table 17).  In the Big Moose Lake
sample, equilibrated DIG (mean  = 0.09 mg/L)
and  phosphorus  (mean  = 0.002 mg/L) con-
centrations were near to or  below  limits of
detection (Table 16).

     As shown in Table 29,  the major  com-
ponents of among-batch variability, assuming
minimal  preparation  errors of the audit
samples,  will be interlaboratory differences
and  batch-to-batch differences  within labor-
atories.  Among-batch precision estimates for
total monomeric aluminum, ANC, all DIG meas-
urements, fluoride,  potassium,  magnesium,
sodium, ammonium, nitrate, closed-system pH,
and sulfate were  within  or near the within lab-
oratory  precision  objectives  for  all audit
sample types  having mean concentrations of
these variables greater  than the system deci-
sion limit.  Among-batch variability  for these
variables is  not substantial for  the  range of
concentrations  represented by the audit
samples.

   Among-batch variability appears to be large
for extractable aluminum at low concentrations
(less than 0.030  mg/L).  Examination of data
from each laboratory (Tables  21  and 23) indi-
cated that variability at  both laboratories was
large, and the most likely source of  the varia-
tion  is among-batch differences in extraction
efficiency  at low concentrations  at the
processing laboratory.

   For total aluminum, among-batch  vari-
ability  is large   at concentrations  less than
approximately 0.050 mg/L, and was observed
for both synthetic and Bagley Lake audit
samples  (Table 29). Examination of data from
both laboratories (Tables 21 through 24) indi-
cates that variation in these measurements is
large at  both  laboratories; among-batch dif-
ferences  due to the digestion procedure is the
probable source of the variation.
                                          101

-------
 Table 29.
Summary Statistic* for Among-Batch Precision Based on Pooled Audit Sample Data, National
Stream Survey - Phase I

Synthetic
Variable
Al-ext, Lot 14
Al-ext, Lot 15
Al-total, Lot 14
Al-total, Lot 15
Al-mono
Al-nex
ANC
BNC
Ca
cr
Cond-PL
Cond-lab
OlC-closed
DIC-eq
DIC-initial
DOC
F-
Fe
K
Mg
Mn
Na
NH4+
N03-
P
pH-closed
pH-ANC
pH-BNC
pH-eq
an.
S042"
True color
Units
mg/L
mg/L
mg/L
mg/L
mg/L
mg/L
peq/l
peq/l
mg/L
mg/L
pS/cm
pS/cm
mg/L
mg/L
mg/L
mg/L
mg/L
mg/L
mg/L
mg/L
mg/L
mg/L
mg/L
mg/L
mg/L
pH units
pH units
pH units
pH units
mg/L
mg/L
PCU
n*
9°
2BC
19*
32*
13*
13*
56
56
56
56
12*
56
14*
56
56
56
56
42"
56
56
56
56
56
55
53'
14*
56
56
56
56
56
12*
Mean
0.024
0.015
0.034
0.030
0.010
0.015
113.9
45.6
0.19
0.33
23.3
18.5
1.40
1.32
1.49
1.1
0.042
0.03
0.20
0.43
0.10
2.74
0.17
0.469
0.022
6.94
6.80
6.82
7.23
1.01
2.25
5
s"
0.0106
0.0030
0.0222
0.0075
0.0074
0.0080
9.24
20.15
0.031
0.029
8.35
1.58
0.046
0.192
0.211
0.27
0.0026
0.018
0.008
0.032
0.009
0.099
0.015
0.040
0.0039
0.085
0.187
0.200
0.179
0.143
0.131
3.3
Audit samples

n*
mm
38
_
38
25*
25*
38
38
39
39
26
38
27
38
38
39
39
39
39
39
39
39
39
38
39
27
38
38
38
39
39
26
bagiey Lake
Mean
_
0.007
w
0.027
0.013
0.014
127.8
35.7
1.57
0.18
15.5
13.1
1.70
1.50
1.54
0.4
0.025
0.01
0.30
0.17
0.01
0.85
0.01
0.012
0.001
6.99
7.01
7.04
7.30
9.32
0.63
6
s*

0.0034
_
0.0399
0.0048
0.0045
11.73
23.55
0.065
0.045
2.45
1.91
0.061
0.203
0.214
0.42
0.0033
0.017
0.016
0.016
0.004
0.042
0.010
0.0189
0.0014
0.049
0.100
0.122
0.117
0.537
0.046
3.2
Big Moose Lake
n*

37

37
24*
24*
39
39
39
39
26
39
27
39
39
39
39
39
39
39
39
39
39
39
39
27
39
39
39
39
39
25
Mean

0.215

0.274
0.195
0.058
-2.1
73.9
1.90
0.43
25.5
24.8
0.53
0.09
0.24
3.9
0.074
0.05
0.43
0.32
0.08
0.62
0.06
1.228
0.002
5.15
5.19
5.21
5.22
4.25
6.33
18
S*

0.0329

0.0255
0.0117
0.0127
4.88
11.34
0.077
0.051
1.76
0.89
0.062
0.078
0.070
0.28
0.0024
0.017
0.022
0.021
0.020
0.038
0.009
0.0553
0.0022
0.059
0.159
0.161
0.071
0.419
0.167
4.6
  n = number of measurements.
b s = standard deviation.
0 Only laboratory audit samples were used for precision estimation.
* One significant outlier (Grubbs' test, p <. 0.05;  Grubbs, 1969) was excluded from the precision estimation.
* Only field audit samples were used for precision estimation.
' Three significant  outliers (Grubbs test, p < 0.05; Grubbs, 1969) were excluded from the precision estimation.
                                                  102

-------
   For nonexchangeable  monomeric alu-
minum,  the Big Moose Lake audit samples
provide  the only  estimate  of among-batch
precision (s = 0.0127 mg/L; Table 29) at a con-
centration greater  than the system decision
limit.  The relative precision, even after remov-
ing three statistical outliers, was 22 percent at
a mean concentration of 0.058 mg/L.  Precision
estimates  derived  from  this methodology
appear to be  affected by  day-to-day dif-
ferences in performance.  Further evaluations
are needed to  provide the  level  of precison
required to interpret data in this range of con-
centrations on a routine basis.

   Measurements  of  BNC  have large esti-
mates of  among-batch variability (s =  11.34 to
23.55 jueq/L) for all three  audit sample types
(Table 29).   The  source of this  variability is
primarily  a result of   interlaboratory  bias,
as Laboratory 2 had measured values for BNC
approximately  twice  those   of Laboratory 1
for all  types  of  audit samples (Tables 21
through 24).

   Calcium measurements at low concen- tra-
tions  (approximately  0.2  mg/L)  may be af-
fected by interlaboratory bias.  Examination of
data  from each laboratory indicated that there
was  a consistent difference  of approximately
0.02   mg/L  between the  two analytical
 laboratories for both  lots of synthetic audit
 samples  (Tables  21 and  23).  Users of data
from meas-urements  of calcium at  low con-
 centrations should consider among-batch
 precision.

    Chloride measurements at  low concentra-
 tions (less  than  0.2 mg/L,  represented by
 the Bagley Lake sample) appear to be affected
 by among-batch variations  (s  = 0.045 mg/L).
 The source of this variation appears to be dif-
 ferences among batches  within  the lab-
 oratories and not interlaboratory bias  (Tables
 22 and 24).

    The among-batch precision of specific con-
 ductance  measurements  at the processing
 laboratory was undoubtedly affected by the
 malfunctioning  probe  used  for  the  first  28
 batches.   The  majority  of synthetic  audit
 sample measurements were conducted by
using the faulty  probe.    Measurements of
natural audit samples had improved precision
estimates (s  = 1.76 to   2.45 juS/cm),  which
were close to the among-batch precision of
analytical laboratory measurements (Table 26).

   For specific conductance measurements at
concentrations less than 20 /^S/cm, among-
batch precision estimates  at both  analytical
laboratories were  outside the within-laboratory
precision objectives.  The among-batch varia-
tion was also affected by interlaboratory bias
(Tables 21 through 24).

   Among-batch precision for laboratory  DOC
measurements  at  low concentrations  (near
1 mg/L)  appears  to  be  affected by among-
batch differences within the  laboratories
(especially Laboratory 1 for Lot 15, Table 21). A
possible source of the variation is differences
resulting from the sample digestion process at
low concentrations.

   For iron, among-batch precision  of all con-
centrations represented by the  audit samples
(mean value  = 0.05 mg/L or less)  was  poor.
For the Big Moose Lake sample, which had the
highest measured iron concentrations (mean =
0.05 mg/L), there were  consistent differences
between the mean values  from the two
laboratories (approximately 0.02 mg/L,  Tables
22 and 24).  Among-batch precision for syn-
thetic field audit  samples within  Laboratory 2
 (Table  23) was  better than for Laboratory 1
 (Table 21) as indicated by the larger confidence
 intervals for Laboratory 1.

    Among-batch precision of  manganese
 measurements is acceptable for all concentra-
 tions represented by the audit samples.  In the
 Big Moose Lake audit  sample, the precision
 estimate (s  = 0.020 mg/L)  was outside the
 within-laboratory precision  objective.
 Measurements of manganese in  the synthetic
 audit sample had a similar mean concentration
 (0.10 mg/L),  but the precision estimate (s  =
 0.009) was within  the within-laboratory  preci-
 sion objective.    Two  outlying  values  were
 present for  the Big  Moose  Lake sample
 measurements (0.016 mg/L from a laboratory
 audit sample at  Laboratory 2 and  0.172 mg/L
 from a field audit sample at Laboratory 1). The
                                           103

-------
 among-batch precision estimated without
 these  two values (n = 37, mean = 0.08 mg/L,
 s = 0.009  mg/L) provided  a relative precision
 estimate (11%) that was near the within-
 laboratory precision objective.

    Only the synthetic audit sample had con-
 centrations of phosphorus above background
 levels.   Initially, among-batch precision was
 large (n = 56, mean = 0.023 mg/L, s = 0.0069
 mg/L,  %RSD  = 30.6).   Examination of the
 measurements revealed that one measurement
 value  was -0.0005 mg/L,  while  two  other
 measurements were greater than 0.040 mg/L.
 Because  these samples  were all  laboratory
 audit samples, the sample concentrations
 could have been  affected during  sample
 preparation at either the support laboratory or
 the  analytical laboratory.   When these three
 significant outliers  (Grubbs test,  p  <^ 0.05;
 Grubbs, 1969) were excluded, the among-batch
 precision  estimate improved by almost 50 per-
 cent (mean = 0.022 mg/L, s = 0.0039, %RSD =
 17.7).

   Estimates of among-batch precision for all
 of the analytical laboratory pH measurements
 were between 0.1  and 0.2 pH units.   Dif-
 ferences between mean values from the two
 laboratories ranged  from 0.1 to 0.3 pH  units
 (Tables 21 through 24) and standard deviations
 of measurement within each laboratory ranged
 from approximately 0.1 to 0.15 pH units.  This
 level of variation has  been observed during
 other multilaboratory  studies   (Davison  and
 Gardner, 1986) unless measurement protocols
 are stringently defined and followed.   The
 primary purpose of the analytical laboratory pH
 measurements was to serve as a check on the
 ANC and BNC measurements on a sample-by-
 sample basis.

   For silica,  among-batch  precision  at  low
concentrations  (=^1  mg/L,  represented by the
synthetic audit sample) was slightly  greater
(%RSD  = 14.1) than the  within-laboratory
precision objectives.   Among-batch  precision
within each laboratory was good  for  both
types of natural  audit samples (Tables 22
and 24).   The negative bias  observed for
Laboratory 2 (Tables 23 and 24) appears to be
the primary source of among-batch variation.
  Comparison of Precision Estimates

    Estimates of method-level (within-batch
  analytical), system-level (overall within-batch),
  and among-batch (including among-laboratory)
  precision were compiled from Tables 28 and 29
  and are presented in Table 30.  For subranges
  where more than one  audit sample type was
  represented, the  sample type having the
  largest standard deviation is presented  in the
  table.

    For all variables,  the within-batch precision
 estimates presented here   indicate that ran-
 dom errors occurring during  sample  prep-
 aration and analysis contributed only a small
 proportion to the overall measurement  error
 during the NSS-I. For all variables except total
 aluminum, DOC, iron, and ammonium,  among-
 batch precision estimates for most or  all sub-
 ranges  were greater than corresponding
 system-level within-batch estimates. This dif-
 ference is  not  totally unexpected,  as  the
 among-batch  precision estimates are
 analogous to long-term standard  deviations
 (or reproducibility), which are expected to be
 greater than a short-term  standard deviation
 (or repeatability) (Taylor, I987). This difference
 suggests that variation among batches within
 a  laboratory or  among laboratories  (due to
 differences in calibrations or processing)  con-
 tributes  more to overall  measurement  error
 than  does sample collection (including natural
 spatial variability  in the stream over the 10-to-
 20  minute period during which samples were
 collected).

   The  most conservative estimates of
 measurement  precision should be  used  for
 data  interpretation activities.   For total
 aluminum,   DOC,   iron,  and  ammonium
 measurements, system-level precision esti-
 mates were nearly the same as or larger than
 corresponding among-batch precision esti-
 mates.  Sample-to-sample variability or collec-
 tion effects are more important for these vari-
 ables  than  day-to-day  or among-laboratory
 variability in determining overall  measurement
error.   For these  four variables,  system-level
precision provides the most conservative esti-
mate of overall measurement error and  should
be used  for data interpretation.  Among-batch
                                         104

-------
TABLE 30.  Comparison of Method-Level, System-Level, and Among-Batch Precision Estimates, National
           Stream Survey - Phase I
Within-batch
Variable (units)
and measurement
range
Al-ext (mg/L)
<0.007
0.007 to 0.050
0.050 to 0.100
>0.100
Al-total (mg/L)
<0.027
0.027 to 0.100
0.100 to 0.500
0.500 to 1.000
> 1.000
Al-mono (mg/L)
<0.015
0.015 to 0.10
0.10 to 0.50
0.50 to 1.00
>1.00
Al-nex (mg/L)
<0.023
0.023 to 0.10
0.10 to 0.50
ANC 0/eq/L)
<0
0 to 50
>50
BNC (/Jeq/L)
0 to 50
>50
Ca (mg/L)
0.02 to 1.00
1.00 to 5.00
5.00 to 10.00
>10.00
Method-level*
Number Pooled
of pairs sa

3
35
15
15

9
6
51
1
1

33
24
7
2
1

52
12
3

2
4
62

24
44

6
35
12
15

0.0002
0.0009
0.0026
0.0055

0.0018
0.0034
0.0042
0.0304
0.0573

0.0011
0.0017
0.0033
0.0045
0.0030

0.0022
0.0033
0.0139

0.70
1.05
4.42

2.68
6.72

0.009
0.041
0.034
0.529
System-level*
Number Pooled
of pairs sd

29
25
4
7

4
16
38
7
0

30
29
6
0
0

47
18
0

7
9
49

28
37

4
35
19
7

0.0016
0.0059
0.0021
0.0166

0.0035
0.0079
0.0807
0.1953
-

0.0022
0.0038
0.0049
-
-

0.0031
0.0068
-

2.94
7.83
11.22

6.38
12.25

0.015
0.057
0.090
1.604
Among-batchc
Number of Sample
samples s0 type

38
9
..
37

38
19
37
-
-

13
-
24
--
-

13
24
--

39
-
38

38
39

56
39
-
"

0.0034
0.0106
-
0.0329

0.0399
0.0222
0.0255
-
-

0.0074
-
0.0117
-
--

0.0080
0.0127
-

4.88
-
11.73

23.55
11.34

0.031
0.077
-
"

BL
S

BM

BL
S
BM



S

BM



S
BM


BM

BL

BL
BM

S
BM


                                                                                    (Continued)
                                                 105

-------
Table 30.  (Continued)
Within-batch
Variable (units)
and measurement
range
Cr (mg/L)
<0.03
0.03 to 1.00
1.00 to 2.00
2.00 to 5.00
5.00 to 10.00
> 10.00
Cond-PL (/;S/cm)
<25.0
25.0 to 50.0
50.0 to 100.0
> 100.0
Cond-lab (pS/cm)
<25.0
25.0 to 50.0
50.0 to 100.0
> 100.0
DIC-closed (mg/L)
<1.00
1.00 to 2.00
2.00 to 5.00
5.00 to 10.00
> 10.00
DIC-eq (mg/L)
<0.23
0.23 to 1.00
1.00 to 2.00
2.00 to 5.00
5.00 to 10.00
>10.00
DIC-lnIt (mg/L)
<0.23
0.23 to 1.00
1.00 to 2.00
2.00 to 5.00
5.00 to 10.00
> 10.00
Method-level3
Number
of pairs

2
13
13
23
11
6

5
22
21
19

14
20
17
17

9
8
26
18
7

1
4
8
24
18
13

0
1
17
18
18
14
Pooled
stf

0.000
0.006
0.022
0.045
0.077
0.336

1.11
1.94
5.27
16.99

0.12
0.23
0.29
0.49

0.025
0.040
0.068
0.168
0.236

0.002
0.029
0.031
0.087
0.086
0.639

.-
0.007
0.036
0.111
0.139
0.780
System-level*
Number
of pairs

0
8
18
24
6
9

5
24
24
12

6
24
24
11

7
9
31
13
5

10
12
12
22
5
4

3
10
15
26
6
5
Pooled
s*

«
0.018
0.033
0.043
0.073
0.734

0.26
1.26
1.32
3.12

0.87
0.27
0.42
1.22

0.020
0.063
0.092
0.184
0.365

0.054
0.099
0.162
0.261
0.184
1.050

0.014
0.091
0.211
0.250
0.050
1.009
Among-batchc
Number of
samples s*


39 0.051
..
..
..
_

12 8.35
26 1.76

-

38 1.91
39 0.89
-
-

27 0.062
27 0.061
mm ..
-
-

39 0.078
mm
38 0.203

..
..

..
39 0.070
38 0.214
•» mm
•• mm
..
Sample
type'


BM





s
BM



BL
BM



BM
BL




BM

BL





BM
BL



                                                                                    (Continued)
                                                 106

-------
Table 30.  (Continued)
Within-batch
Variable (units)
and measurement
range
DOC (mg/L)
<0.5
0.5 to 2.0
2.0 to 5.0
5.0 to 10.0
>10.0
F (mg/L)
0.010 to 0.050
>0.050
Fe (mg/L)
<0.02
0.02 to 0.05
0.05 to 0.10
0.10 to 0.50
0.50 to 1.00
>1.00
K (mg/L)
<0.15
0.15 to 0.35
0.35 to 0.45
>0.45
Mg (mg/L)
<1.00
1.00 to 2.00
2.00 to 5.00
>5.00
Mn (mg/L)
<0.01
0.01 to 0.05
0.05 to 0.10
>0.10
Na (mg/L)
<0.50
0.50 to 1.00
1.00 to 2.00
2.00 to 5.00
>5.00
Method-level3
Number Pooled
of pairs sa

4
27
16
14
7

29
39

2
3
11
36
3
12

1
10
2
55

26
21
15
6

2
9
12
44

4
8
11
33
12

0.02
0.05
0.11
0.15
0.15

0.0000
0.0021

0.001
0.001
0.002
0.004
0.008
0.114

0.001
0.006
0.002
0.015

0.004
0.013
0.033
0.158

0.001
0.001
0.003
0.017

0.001
0.027
0.017
0.028
0.174
System-level^
Number
of pairs

2
41
15
7
0

51
14

14
18
10
18
3
2

1
5
6
53

19
23
18
5

19
19
7
20

4
4
24
25
8
Pooled
s"

0.02
0.24
0.44
0.58
—

0.0011
0.0029

0.003
0.009
0.029
0.117
0.163
0.377

0.003
0.003
0.005
0.063

0.013
0.014
0.058
0.242

-
0.015
0.002
0.023

0.001
0.030
0.028
0.037
0.097
Among-batchc
Number of Sample
samples se type

39
56
39
-
—

39
39

39
42
39
-
-
—

-
39
39
—

56
-
-
—

39
-
39
56

-
39
-
56
"

0.42
0.27
0.28
—
••

0.0033
0.0024

0.017
0.018
0.017
-
-
••

-
0.016
0.022
—

0.032
-
-
—

0.004
-
0.020
0.009

-
0.042
-
0.099
~

BL
S
BM



BL
BM

BL
S
BM





BL
BM


S




BL

BM
S


BL

S

                                                                                      (Continued)
                                                   107

-------
Table 30. (Continued)
Within-batch
Variable (units)
and measurement
range
NH4+ (mg/L)
<0.02
0.02 to 0.05
0.05 to 0.10
>0.10
N03- (mg/L)
O.OOO
>3.000
P (mg/L)
<0.001
0.001 to 0.005
0.005 to 0.015
>0.015
pH-closed (pH units)
<4.00
4.00 to 5.00
5.00 to 6.00
6.00 to 7.00
7.00 to 8.00
>8.00
pH-ANC (pH units)
4.00 to 5.00
5.00 to 6.00
6.00 to 7.00
7.00 to 8.00
>8.00
pH-BNC (pH units)
4.00 to 5.00
5.00 to 6.00
6.00 to 7.00
7.00 to 8.00
>8.00
pH-eq (pH units)
<4.00
4.00 to 5.00
5.00 to 6.00
6.00 to 7.00
7.00 to 8.00
>8.00
Method-level4
Number Pooled
of pairs &d
12
20
14
22
55
13

2
24
20
22

4
4
7
25
26
2

1
5
34
28
0

0
6
34
28
0

2
7
4
3
34
18
0.000
0.001
0.002
0.004
0.0162
0.1212

0.0003
0.0003
0.0004
0.0018

0.000
0.012
0.041
0.012
0.028
0.010

0.014
0.021
0.053
0.039
-

-
0.034
0.042
0.040
-

0.005
0.009
0.000
0.004
0.019
0.028
System-level6
Number Pooled
of pairs srf
33
21
7
4
53
12

7
31
20
7

0
6
13
24
19
3

6
8
27
22
2

6
8
26
23
2

0
6
4
11
40
4
0.006
0.007
0.014
0.011
0.0422
0.3239

0.0009
0.0016
0.0034
0.0040

..
0.032
0.036
0.036
0.036
0.026

0.022
0.104
0.073
0.091
0.034

0.035
0.092
0.065
0.086
0.062

-
0.017
0.041
0.113
0.171
0.083
Among-batch"
Number of Sample
samples s* type'
39
..
39
56
39
~

39
39
_
53

_
..
27
14
27
-

_
39
56
38
-

—
39
56
38
-

-
_
39
—
56
-
0.010
..
0.009
0.015
0.0553
-

0.0014
0.0022
..
0.0039

—
„
0.059
0.085
0.049
-

..
0.159
0.187
0.100
-

—
0.161
0.200
0.122
-

-
—
0.071
..
0.179
-
BL

BM
S
BM


BL
BM

S



BM
S
BL



BM
S
BL



BM
S
BL




BM

S

                                                  108
                                                                                     (Continued)

-------
Table 30. (Continued)
Within-batch
Variable (units)
and measurement
range
SiO2 (mg/L)
<0.50
0.50 to 1.50
1.50 to 5.50
5.50 to 10.50
>10.50
SO42" (mg/L)
<0.05
0.05 to 2.50
2.50 to 5.00
5.00 to 12.50
>12.50
True color (PCU)
<30
>30
Turbidity (NTU)
<2.0
2.0 to 20.0
20.0 to 100.0
Method-level*
Number
of pairs

1
1
29
19
18

2
13
18
26
9

50
17

28
38
1
Pooled

0.007
0.035
0.052
0.141
0.195

0.002
0.019
0.048
0.117
0.239

0.9
8.7

0.07
0.26
0.71
System-level"
Number
of pairs

0
1
25
21
18

0
18
10
24
13

44
21

27
36
2
Pooled

—
0.064
0.071
0.122
0.730

—
0.054
0.081
0.128
0.243

2.6
1.9

0.08
0.39
1.41
Among-batch"
Number of Sample
samples sa type'

—
56 0.143 S
39 0.419 BM
39 0.537 BL
™ —

—
56 0.131 S
-
39 0.167 BM
« -•>

25 4.6 BM
™ ™

—
~
.- —
 * Method-level = within-batch precision calculated from laboratory routine-duplicate sample pairs.
 b System-level = overall within-batch precision calculated from field routine-duplicate sample pairs.
 0 Among-batch = among-batch precision (across laboratories) calculated from audit sample measurements.
 d Pooled s = pooled standard deviation.
 6 s = standard deviation.
 ' Audit  sample types from which precision estimate was derived: S = synthetic audit sample. BL = Bagley
   Lake sample, and BM = Big Moose Lake sample.
 precision estimates for all other variables
 provide the  most  conservative estimate  of
 overall measurement error available for the
 NSS-I.

 Discussion and Summary:  Precision

       The  grouping of  measurements  of
 routine-duplicate pairs into concentration sub-
 ranges generally provides more useful informa-
 tion related to data quality than a  single esti-
mate  of relative   precision.   However,  there
appear to be two limitations to this approach
that must be considered.  Unless there is  exist-
ing  information  to  allow field  duplicate
samples to be selected so that each subrange
is represented  by  an adequate  number of
pairs, data for some subranges may be
scarce.  However, if  samples are  duplicated
more or  less at random, then the distribution
of pairs  among subranges should  reflect the
distribution  of routine sample  concentrations.
                                             109

-------
A more serious concern is that precision esti-
mates based on pooled variances of sample
pairs can be extremely sensitive to  outlying
values, as  was observed for the equilibrated
pH  measurements,   where  removal  of two
values   improved  the precision estimate by
almost 50 percent.

     The utility of duplicate measurements is
limited in that they do not provide an estimate
of among-batch variability, as  do the audit
samples.  Likewise, audit  samples do not pro-
vide an estimate of random errors associated
with  sample collection and handling  and are
concentration-specific.  A series of split sam-
ples prepared  from a single bulk sample may
provide a  more useful design to monitor and
assess total  measurement uncertainty.  The
split  samples  could then  be used to assess
various components of within-batch precision,
among-batch  precision (by withholding a split
until  the next  batch of samples is analyzed),
and even among-laboratory precision (by send-
ing  splits  to  other  laboratories  for  analysis
within holding times).   Natural audit samples
of appropriate concentrations that have been
well  characterized  chemically could  also be
taken to the field and processed through the
sampling device to provide estimates of total
measurement uncertainty.

     For many variables,  among-batch preci-
sion estimates are not available for subranges
at high concentrations.   The  audit sample
program used during  the NSS-I was designed
to control  measurement processes at concen-
trations representative of  surface waters sen-
sitive to or impacted by acidic deposition.  It
cannot be  said with certainty that measure-
ment error will be  reduced at concentrations
greater than  those represented by the NSS-I
audit samples. Among-batch precision within
a laboratory would probably improve at higher
concentrations, but interlaboratory differences
may be present at all concentrations.   For
those variables for which  interlaboratory bias
was indicated as  a primary  influence on
among-batch   precision   (e.g.,  specific
conductance),  the   among-batch  precision
estimate available  for  a  lower  concentration
may pro-vide  a more  conservative estimate of
overall  measurement  precision  at  higher
concentrations than a  system-level  precision
estimate.   For future studies, the  audit
samples chosen should bracket the range of
sample concentrations  to the greatest extent
feasible.
Summary of Data Quality
Assessment

    The  overall  data  quality  of  the  NSS-I
analytical results,  in terms of the data quality
objectives established for the project, is
adequate  to  achieve  the  project objectives.
The samples collected are representative of
the types  of stream  resources of interest to
the project. The data base is sufficiently com-
plete,   in  terms of  the number  of samples
providing valid data, to allow the estimation of
population sizes.  Based on a  comparison of
audit sample measurements (see Appendix A),
there is high confidence that the data collected
and analyzed during the NSS-I are comparable
and appear to be compatible with other NSWS
data bases. A general summary of the detec-
tability, accuracy, and precision  of the ana-
lytical measurements is presented in Table 31.
There  are  only a few cases where use of the
data to meet  the project  objectives  may be
limited by data quality considerations. In most
of these  cases,  the  limitation  affects  only
interpretation of measurements at  low con-
centrations.  More general limitations  are
summarized below:

1.   Specific conductance measurements
    made at the processing laboratory during
    the first  half of the NSS-I are suspect,
    because of systematic  errors related  to a
    faulty probe.    Analytical  laboratory
    measurements  of specific conductance
    should be used  for all interpretative ac-
    tivities.

2.   Estimates of exchangeable monomeric
    aluminum,  calculated as the difference
    between  total  monomeric  and non-
    exchangeable monomeric aluminum,  may
    be negative for  some  samples because
    the background levels of nonexchangeable
    monomeric aluminum are higher than
    those  observed   for   total  monomeric
                                          110

-------
Table 31.   Summary of Data Quality Assessment for Chemical Variables with  Respect to Detectablllty, Accuracy,
            and Precision, National Stream Survey - Phase I*
Variable
                                Detectability
Accuracy
                                                                                           Precision
Al-ext +
Al-total +
Al-mono +
Al-nex -b
ANC +
BNC -b
Ca +
cr +
Cond-PL •"
Cond-lab +
DIC-closed +
DIC-eq +
DIC-init +
DOC +
F +
Fe +
K +
Mg +
Mn +
Na +
NH4+ +
NOg" +
P +
pH-closed +
pH-ANC +
pH-BNC +
pH-eq +
SiO, +
so/- +
True color NE
Turbidity NE
_D ,o
+ mb
+ •"•"
_b ,b
b m&»G
_b _b.c
+ mb
_e J3,o
-* +
NE +
 * +  = acceptable in terms of data quality objective or primary project objectives.
  -  = estimate not near data quality objective.
  NE = not evaluated.
 b Possible limitations at low concentrations.
 c Overall  within-batch precision estimate is larger than among-batch precision estimate.
 ''Estimates not of acceptable quality.
 * Possible limitations at high concentrations.
                                                     111

-------
     aluminum.   These  negative values will
     result in an increased uncertainty in the
     estimate of exchangeable monomeric
     aluminum, especially at lower concentra-
     tions.  In addition, the among-batch pre-
     cision  of nonexchangeable monomeric
     aluminum measurements may not be of
     adequate quality to effectively  interpret
     patterns in aluminum chemistry.

3.    At  low  concentrations  of  BNC,  data
     interpretations  may be  limited  because
     of random and  systematic errors caused
     by changes in  dissolved carbon dioxide
     concentration in the sample during col-
     lection, transport, or the titration proce-
     dure.

4.    Although all DIG and pH measurements
     provided acceptable results, the closed-
     system measurements are less subject
     to changes in  dissolved carbon dioxide
     concentration between the time of col-
     lection  and analysis, and thus provide
     the best estimates  of in situ conditions
     at the time of sampling.

5.    Data users  interested  in phosphorus
     measurements  should consider  the pos-
     sibility of  a   negative bias from
     Laboratory 1 during  the last half of the
     NSS-I.

6.    For  total  aluminum,  DOC,  iron, and
     ammonium, sample-to-sample varia-
     bility within  a  batch  was apparently
     larger  than  sample-to-sample variability
     among batches.   For these four vari-
     ables,   the system-level  precision es-
     timates  provide the most conservative
     estimate of  meas-urement uncertainty
     available for the NSS-I.   For  all other
     variables, the estimates of among-batch
     precision are the best available for the
     evaluation of measurement uncertainty.

     In  addition to  the variable-by-variable
evaluations, three  checks of overall  data
quality were related to interpretation  of
acidification  effects:   comparison of  the total
ionic charge of  cations   and  anions;  com-
parison of calculated and measured  specific
conductance values; and comparison of calcu-
lated carbonate alkalinity and measured values
of ANC.

Charge Balances

    For each sample the  sum  of  concentra-
tions (expressed in jueq/L)  was  calculated for
both major cations and major anions.  For cat-
ions,  the summary  value was the total of
calcium,   magnesium,  sodium,  potassium,
ammonium, and hydrogen  ion concentrations.
For anions, the summary value  was the total
of sulfate, nitrate, chloride, fluoride, ANC, and
hydrogen ion concentrations. Use of ANC and
hydrogen  ion concentrations in the equation
accounted for bicarbonate  and carbonate ions,
hydroxide ions, and noncarbonate anions such
as organic anions and metal oxides.

    Because of the electroneutrality con-
straint, the sum of cations must equal the sum
of anions. Realistically, the charge balance will
never equal zero, because analytical errors in
each of the measured  cations or anions are
additive in the summation process.  In addi-
tion,  the analytical  errors due to different
methodologies will be different for the various
cations and anions.

    The sum of cations is  plotted against the
sum of  anions  for each sample in Figure 13.
Values from the enhanced  data set were used
for  this comparison.  The  agreement between
the sum of cations and the sum of anions is
excellent, with 94.7 percent of the samples (n
= 1,342) falling within 10 percent of the line of
1:1 correspondence (representing the values at
which the sum  of cations  equals the  sum of
anions). A total of 88.4  percent of the samples
fell  within 5  percent of the line  of  1:1  cor-
respondence.  The observed pattern of devia-
tion from the identity line  suggests either an
underestimate of the sum of anions (due to
systematic errors  in  measurement  or to
unmeasured anions) or an  overestimate of the
sum of cations.  As  no  serious systematic
errors were observed for measured anions or
cations,  the   most  likely  reason  for the
observed pattern is  unmeasured anions such
as  organics  (indirectly measured  as  DOC),
aluminum,  or   metals   present  in samples
                                          112

-------
 collected from polluted  streams (e.g.,  those
 receiving mine effluents).
   2000-f
-z.
o
<
CJ
1000-
                       1000
                 SUM OF ANIONS (,ueq L
                                       2000
 Figure 13.   Sum of cations versus sum of anlons, for
 routine samples,  National Stream Survey -  Phase I.
 (From Kaufmann et al., In press; line represents 1:1 cor-
 respondence,  where sum of cations  equals sum  of
 anlons).

 Specific Conductance Check

      Comparison  of the  calculated and
 measured values for specific conductance
 provides   an additional  check  on  analytical
 errors  in  measurements  or the presence of
 unmeasured ionic species.    The calculated
 conductance for each major ion was estimated
 using the Debye-Huckel-Onsager equation
 (Atkins, 1978), which corrects for concentration
 effects:
 where
      Mc

      Mc

      Mc

      A
      B
      C
=  M°
                           BM° )xC*
                               C
=  corrected molar conduc-
   tance at 25 "C
=  molar conductance at
   infinite dilution
=  60.2
=  0.229
=  molar concentration of ion
                                    Figure  14  presents  a  plot  of  the
                               measured  versus  calculated specific  conduc-
                               tance for each routine sample.  In general, the
                               agreement of the two estimates is excellent for
                               most samples. The linear regression equation
                               that best described the observed relationship
                               was:

                                 Measured conductance = 0.981 (calculated
                                        conductance) + 4.87

                               This  equation explained  98.7 percent of the
                               total variance.  Both the slope (0.981, standard
                               error is ± 0.0031) and intercept (4.87, standard
                               error is +. 0.522) indicated a deviation from the
                               line of  1:1 correspondence  (where  measured
                               conductance equals calculated conductance).
                               Some of the deviation is probably due to sys-
                               tematic errors in measurement, at least at low
                               values  of specific conductance (see Accuracy
                               section).   Unmeasured ions,  as indicated by
                               the charge balance comparisons  (Figure 13)
                               also  apparently  affected   the  estimate of
                               calculated   conductance   in  some   cases,
                               because the measured conductance  was
                               larger than the calculated conductance.

                               Comparison ofANC Values

                                    Comparison of an  ANC value measured
                               in  a  sample to that predicted  assuming  a
                               carbonate system provides  an indication of
                               the reliability of pH, DIG, and ANC measure-
                               ments,  as well as the presence of unmeasured
                               noncarbonate protolytes.  Carbonate alkalinity
                               represents the contributions  to  ANC from
                               bicarbonate,  carbonate,   hydroxide,  and
                               hydrogen ions.  Carbonate alkalinity, or calcu-
                               lated ANC,  was estimated  from  measured
                               values  of pH-BNC and  initial DIG (Hillman et
                               al., 1987):
ANC =
    C
                                                      [DIG]
                                                                  K1 + 2 K1 K2
                                                      12,011  \ [H+]2
                                                               w
                                                           K1 + K1 K2
                                                                          x 10*
                                           113

-------
 where ANC    =
       DIG
       "
         v
          calculated acid-neutralizing
          capacity
          DIC-initial value in mg/L
          10.-(PH) Where pH = value
          from pH-BNC measurement
          4.4463 x 10'7 at 25 *C
          4.6881 x 10'11 at25"C
          1.0023 x 10'14 at 25 °C
 For this relationship, the measured ANC
 should be greater than or equal to the car-
 bonate alkalinity; otherwise, analytical errors
 may  have  influenced  measurements of  pH,
 DIG, or ANC.

      Carbonate alkalinity values are plotted
 against  measured ANC values  in Figure 15.
 The majority of the data points fell below  the
 line of 1:1 correspondence  which shows that
 measured  ANC is greater than carbonate
 alkalinity.  Thus, measurements of  pH-BNC,
 DIC-initial, and ANC appear to be reliable for
 most  samples.  The  observed differences
    500 -{
    400-
1/1
-   300-
O
O
    200-
     100-
      0-
    n	1	'	1	'	1	'	1—
0     100    200    300    400
    CALCULATED CONDUCTIVITY (/iS cm"1)
between carbonate  alkalinity  values  and
measured ANC values are apparently due to
the presence of noncarbonate protolytes, such
as  DOC,  aluminum,   dissolved  metal  com-
plexes, or particulate metal oxides.
                                                      600-f
  a.

  >-
  i—
  z:
  _i
  <
  _i
  <
                                          o
                                          CD
                                          o
                                          o
                                          o
                                          _l
                                          <
                                              500-
                                              400-
                                              300-
                                              200-
                                              100-
                                                       n—i'i|—i  i i  | i  i i  | i—i  i |
                                                       100   200   300   400   500   600
                                                           MEASURED ANC (,ueq L~1)
                                        Figure 15.    Calculated carbonate akallnlty versus
                                                   measured   ANC  for  routine  samples,
                                                   National Stream Survey -  Phase I  (from
                                                   Kaufmann et al., In press; line represents
                                                   1:1 correspondence line, where carbonate
                                                   alkalinity equals measured ANC).
                                          500
Figure 14.    Measured versus calculated specific con-
            ductance at 25°C for routine samples, Na-
            tional Stream Survey-  Phase I  (from
            Kaufmann et a!., In press; line represents
            1:1  correspondence,   where measured
            conductance  equals calculated conduc-
            tance).
                                             114

-------
                                    REFERENCES
American Chemical Society, Committee on Environmental Improvement. 1980. Guidelines for Data
     Acquisition and Data Quality Evaluation in Environmental Chemistry.  Analytical Chemistry,
     52:2242-2249.

APHA (American Public  Health Association), American Water Works Association,  and Water
     Pollution Control Federation.  1985.  Standard Methods for the Examination of Water and
     Wastewater, 16th Ed. APHA, Washington, D. C.

Arent, L J., M. O. Morison, and C. S. Soong. In preparation. Eastern Lake Survey- Phase II and
     National Stream Survey - Phase I: Laboratory Operations Report.  EPA/600/4-88/025. U. S.
     Environmental Protection Agency, Las Vegas, Nevada.

ASTM (American Society for Testing and Materials).   1984.  Annual Book of ASTM Standards.
     Standard Specification for Reagent Water.  D1193-77 (reapproved 1983). Vol. 11.01. ASTM,
     Philadelphia, Pennsylvania.

Atkins, P. W., 1978. Physical Chemistry. W. H. Freeman and Co., San Francisco, California.

Blick, D. J., J. J. Messer, D. H. Landers, and W.  S. Overton.  1987.  Statistical Basis for the Design
     and Interpretation of the National Surface  Water Survey,  Phase I:  Lakes and Streams.
     Lake and Reservation Management. Vol. 3:470-475.

Burke,  E. M., and D. C. Hillman.  1987. Syringe  Sample  Holding Time  Study.  Appendix A in
     C. M. Knapp, C. L. Mayer, D. V. Peck, J. R. Baker, and G. J.  Filbin.   National Surface Water
     Survey, National Stream Survey, Pilot Survey.  Field Operations Report. EPA/600/8-87/019.
     U. S. Environmental Protection Agency, Las Vegas, Nevada.

Campbell, S., and H. Scott.  1985.  Quality  Assurance in  Acid  Precipitation  Measurements.
     Pp. 272-283. In: J. K. Taylor and T. W. Stanley (eds.). Quality Assurance for Environmental
     Measurements.  ASTM STP 867. American Society for Testing and Materials,  Philadelphia,
      Pennsylvania.

Clayton, C.  A., J.  W. Hines, and  P. D. ESkins.   1987. Detection Limits with Specified Assurance
      Properties.  Analytical Chemistry, 59:2506-2514.

Currie, L A 1968.  Limits of Qualitative Detection  and Quantitative Determination: Application to
      Radiochemistry. Analytical Chemistry, 40:586-593.

Davison, W., and  M. J. Gardner.  1986. Interlaboratory Comparisons of the Determinations of pH
      in Poorly-Buffered  Fresh Waters. Analytica Chimica Acta, 182:17-31.
                                           115

-------
Drous6, S. K., D. C. Hillman, L W. Creelman, J. F. Potter, and S. J. Simon. 1986a. National Surface
      Water Survey, Eastern Lake Survey - Phase I:  Quality Assurance Plan.  EPA/600/4-86/008,
      U. S. Environmental Protection  Agency, Environmental  Monitoring  Systems  Laboratory,
      Las Vegas, Nevada.

Drouse, S.  K., D. C. Hillman, J. L. Engels, L  W. Creelman, and S. J. Simon, 1986b.  National
      Surface  Water Survey,  National Stream Survey (Phase  I -  Pilot,  Mid-Atlantic Phase I,
      Southeast Screening, and Episodes Pilot):  Quality Assurance Plan.  EPA/600/4-86/044.
      U.S. Environmental Protection  Agency,  Environmental  Monitoring Systems Laboratory,
      Las Vegas, Nevada.

Drouse, S. K.  1987. The National Surface Water Survey,  National Stream Survey, Phase I - Pilot
      Survey:  Summary   of    Quality  Assurance   Data   Results. EPA/600/8-87/057.  U.S.
      Environmental  Protection Agency,  Environmental  Monitoring Systems Laboratory,  Las
      Vegas, Nevada.

Eshleman, K. N.  In press.  Predicting Regional Episodic Acidification of Surface Water Using
      Empirical Techniques. Water Resources Research.

Glaser, J. A, D. L Forest, G. D. McKee, S. A Quae, and W. L. Budde.  1981.  Trace Analyses of
      Wastewaters. Environmental Science and Technology.  15:1426-1435.

Grosser, S.  C., A.  K. Pollack.   1987.   National Stream Survey  Data Base Audit.   Final Report.
      Systems Applications, Inc., San Rafael, California

Grubbs, F. E.  1969. Procedures for Detecting Outlying Observations in Samples. Technometrics,
      11:1-21.

Hagley, C. A,  C. L Mayer, and  R. Hoenicke.  In press.  National Stream Survey,  Phase I, Field
      Operations  Report.  EPA/600/4-88/023. U.S. Environmental Protection Agency, Las Vegas,
      Nevada.

Hillman, D. C., S. H. Pia, and S.  J. Simon.  1987. National Surface Water Survey, Stream Survey
      (Pilot,  Middle-Atlantic Phase I,  Southeast Screening,  and Middle-Atlantic Episode Pilot):
      Analytical  Methods   Manual. ,  EPA/600/8-87/005. U.S. Environmental Protection Agency,
      Las Vegas, Nevada.

Hubaux,  A,  and G. Vos.   1970.  Decision and Detection Limits for Linear Calibration  Curves.
      Analytical Chemistry.  42:849-855.

Hunt, D. T. E., and A. L. Wilson.  1986. The Chemical Analysis of Water:  General  Principles and
      Techniques.  The Royal Society of Chemistry, Burlington House, London.

Kaufmann, P., A Herlihy, J.  Elwood,  M.  Mitch, S.  Overton,  M. Sale,  J.  Messer, K.  Reckhow,
      K. Cougan, D. Peck, J. Coe, A Kinney, S. Christie, D. Brown, C. Hagley, and Y. Jager.  1988.
      Chemical Characteristics  of Streams in the Mid-Atlantic and  Southeastern  United States.
      Volume  I:  Population Descriptions  and  Physico-Chemical Relationships.   EPA/600/3-
      88/021a  U. S. Environmental Protection Agency, Washington, D. C.

Keith, L H., W. Crummett, J. Deegan, Jr., R. A Libby, J. K. Taylor, and G. Wentler. 1983.  Principles
      of Environmental Analysis. Analytical Chemistry. 55:2210-2218.
                                           116

-------
Knapp, C. M., C. L. Mayer, D. V. Peck, J. R. Baker, and G. J. Filbin. 1987. National Surface Water
     Survey: National Stream Survey, Pilot Survey:  Field Operations Report. EPA/600/8-87/019.
     U.S. Environmental Protection Agency, Las Vegas, Nevada. 94 pp.


Landers, D.  H., J. M. Eilers,  D. F. Brakke,  W. S. Overton, P. E. Kellar, M. E. Silverstein, R. D.
     Schonbrod,  R. E. Crowe, R. A. Linthurst, J. M. Omernik, S. A. league, and E. P. Meier.  1987.
     Characteristics of Lakes in the Western United States. Volume I: Population Descriptions
     and Physico-Chemical Relationships.   EPA/600/3-86/054a.  U.S.  Environmental Protection
     Agency, Washington,  D.C. 176 pp.

Linthurst, R. A., D. H.  Landers, J. M.   Eilers, D. F. Brakke, W. S. Overton,  E. P.   Meier, and
     R.  E. Crowe.  1986.  Characteristics of Lakes in the Eastern United States.  Volume I:
     Population  Descriptions and Physico-Chemical Relationships.   EPA/600/4-86/007a.   U.S.
     Environmental Protection Agency, Washington, D.C.  136 pp.

Long, G. L,  and  J. D. Winefordner.  1983.  Limit of Detection: A Closer Look at the  IUPAC
     Definition. Analytical  Chemistry. 55:712A-724A.

McQuaker, N.  R.,  P.  D.   Kluckner,  and D.  K.  Sandberg,  1983.   Chemical Analysis of Acid
     Precipitation:  pH and Acidity Determinations.   Environmental  Science and  Technology.
     17(7):431-435.

Mericas, C.  E., and R.D. Schonbrod. 1987.  Measurement Uncertainty in the National Surface
     Water Survey. Lake and Reservoir Management. 3:488-497.

Messer,  J. J., C. W. Ariss, J. R.  Baker, S. K.  Drouse,  K. N.  Eshleman,  P.  R.   Kaufmann,
     R. A. Linthurst, J. M. Omernik, W. S. Overton, M. J. Sale, R. D. Schonbrod, S. M. Stambaugh,
     and J. R. Tuschall, Jr. 1986. National Surface Water Survey, National Stream Survey, Phase
     I  - Pilot Survey.   EPA/600/4-86/026.  U.S. Environmental Protection Agency, Washington,
     D.C.

Miller, J. C., and J. N. Miller.  1984.  Statistics for Analytical Chemistry.  Ellis  Horwood Limited,
      Chichester, England.  201 pp.

Oak Ridge National Laboratory. 1984. National Surface Water Survey Project - Data Management
      Proposal. Environmental Sciences Division and  Computer   Sciences,   UCC-ND,  ORNL,
      Oak Ridge,  Tennessee.

Oliver,  B. G., E. M. Thurman, and R. K. Malcolm. 1983. The Contribution of Humic Substances to
      the Acidity of Colored Natural Waters. Geochemica et Cosmochemica Acta.  47:2031-2035.

Oppenheimer, L, T. P. Capizzi, R. M. Weppelman, and Hina Mehta. 1983. Determining the Lowest
      Limit of Reliable Assay Measurement. Analytical Chemistry. 55:638-643.

Overton,  W. S.   1986.  A Sampling Plan  for Streams in the  National Surface Water Survey.
      Technical Report 114. Department of Statistics, Oregon State University, Corvallis, Oregon.

Overton, W. S.  1987.  A Sampling and Analysis Plan for Streams in the National  Surface  Water
      Survey Conducted by EPA   Technical Report  117. Department of Statistics, Oregon State
      University, Corvallis, Oregon.


                                            117

-------
Peden, M.  E.   1981.   Sampling Analytical and Quality Assurance Protocols for the National
     Atmospheric Deposition Program. ASTM D-22 Symposium and Workshop on Sampling and
     Analysis of Rain.  ASTM, Philadelphia, Pennsylvania.

Permutt, T. J. and A. K. Pollack.  1986.  Analysis of Quality Assurance Data for the Eastern Lake
     Survey.   Appendix A in M. D.  Best,  S. K. Drous6, L W.  Creelman and  D. J. Chaloud.
     National Surface Water Survey,  Eastern  Lake Survey (Phase I  -  Synoptic Chemistry):
     Quality Assurance  Report.   EPA 600/4-86/011.  U.S.  Environmental  Protection Agency,
     Environmental Monitoring Systems Laboratory, Las Vegas, Nevada.

Royset, O.   1986. Flow-injection Spectrophotometric Determination of Aluminum in Water with
     Pyrocatechol Violet. Analytica Chimica Acta. 185:75-81.

Rogeberg,  E.  J.  S., and A.  Henriksen.   1985.  An Automatic Method for Fractionation and
     Determination of Aluminum Species in Freshwaters. Vatten.  41:48-53.

Sale, M. J. (ed.). In press.  Data Management and Analysis Procedures for the National Stream
     Survey. ORNL/TM-XXXX. Oak Ridge National Laboratory, Oak Ridge, Tennessee.

SAS Institute, Inc.  1985.  SAS User's Guide:  Statistics, Version  5 Edition.  SAS Institute, Inc.,
     Gary, North Carolina.  956 pp.

Skougstad,  M.  W., J.  Fishman,  L  C. Friedman, D.  E. Erdmary and S. S. Duncan (eds.).   1979.
      Methods  for  Determination  of  Inorganic Substances in Water  and Fluvial Sediments:
     Techniques of Water-Resources Investigations of the United States Geological Survey,
      Book 5, Chapter A1.  U.S. Government Printing Office, Washington, D.C.

Silverstein, M. E., M. L Faber, S.  K. Drous6, and T. E. Mitchell-Hall. 1987.  National Surface Water
      Survey, Western Lake Survey (Phase I-Synoptic  Chemistry):  Quality Assurance Report.
      U.S.  Environmental Protection Agency, Environmental  Monitoring  Systems Laboratory,
      Las Vegas, Nevada. 292 pp.

Stanley, T.  W., and S. S. Verner.   1985.  The U.S. Environmental Protection Agency's Quality
      Assurance Program.  In:  Quality Assurance  for Environmental Measurements, ASTM STP
      867, J. K. Taylor and T. W.  Stanley (eds.).   American Society for Testing and Materials,
      Philadelphia, Pennsylvania, pp. 12-19.

Stapanian, M. A., A. K. Pollack, and B. C. Hess. 1987. Container Holding Time Study.  Appendix B
      in C. M. Knapp, C. L. Mayer, D. V. Peck, J. R. Baker, and G. L  Filbin.  National Surface Water
      Survey, National Stream Survey: Pilot Survey, Field Operations Report. EPA/600/8-87/019.
      U.S. Environmental Protection Agency, Las Vegas, Nevada.

Taylor, J. K.   1984.  Guidelines fqr  Evaluating  the Blank Correction.  Journal  of Testing and
      Evaluation. 12:54-5.

Taylor, J.  K. 1985.  Principles of Quality Assurance of Chemical Measurements.  NSBIR 85-3105.
      U.S.  Department of  Commerce, National Bureau of Standards, Gaithersburg, Maryland.
      81pp.

Taylor,  J.  K.   1987.   Quality Assurance of Chemical Measurements.   Lewis  Publishers, Inc.,
      Chelsea, Michigan. 328 pp.


                                           118

-------
Tyree,  S. Y., Jr.  1981.  Rainwater Acidity Measurement Problems.  Atmospheric Environment,
     5:57-60.

U S  EPA (Environmental Protection Agency).  1979.  Handbook for Analytical Quality Control in
     Water and Wastewater Laboratories.  EPA/600/4-79/019.  U.S. Environmental Protection
     Agency, Environmental Monitoring Systems Laboratory, Cincinnati, Ohio.

U.S. EPA (Environmental Protection Agency). 1983. Methods for Chemical Analysis of Water and
     Wastes.  EPA/600/4-79/020.  U.S. Environmental Protection Agency, Cincinnati, Ohio.

U S  EPA (Environmental Protection Agency).  1987.  Handbook of  Methods for Acid  Deposition
     Studies:  Laboratory Analyses  for Surface Water Chemistry.  EPA/600/4-87/026.  U.S.
     Environmental Protection Agency, Office of Research and Development, Washington, D.C.
     342 pp.

Weast, R. C.  (ed.).   1972.  CRC Handbook of Chemistry and Physics, 53rd Ed.,  CRC Press,
      Cleveland, Ohio.

Young, T.  C.,  J.  V.  DePinto,  S.  C.  Martin, J.  S. Bonner.  1985.  Algal Available  Particulate
      Phosphorous in the Great Lakes Basin. Journal of Great Lakes  Research. 11(4):434-436.
                                            119

-------
                                    Appendix A

                          Prepara tion of Audit Samples


Preparation of Natural Audit Samples

      To ensure that all field natural audit samples of a particular lot were uniform, EMSL-LV
instructed the support laboratory to follow the protocol specified below:

      1.    Clearly label the field and laboratory natural stock barrels with the lot number.

      2.    Label the 2-L bottles to be filled.

      3.    Operating in a  clean environment, flush the Tygon tubing  lines with  lake water.
           Discard the water.

      4.    Pump 20 to 25 mL lake water into the audit bottle, cap the bottle, rinse the bottle to
           get complete coverage, and discard the rinse.

           NOTE: The Tygon tubing must not touch the sidewalls of the bottle.

      5.    Perform  step 4 two more times.  Discard the rinse water each time.

      6.    Fill the bottle to the top  (no head  space) with lake water that is filtered through a
           0.40-micron filter.

           NOTE:   The  bottle must be capped  immediately after it is  filled to minimize  the
           possibility of  contamination.

      7.    Secure the cap to the bottle with tape.

      8.    Log in the total number of samples  prepared, the date prepared, and the name of the
           analyst or technician.

      9.    Place samples in storage at 4 °C by lot and ID number to await shipment.

      10.   Discard any water remaining in Tygon tubing.  Do not drain residual lake water into
           the stock barrel.

      For  laboratory  natural audit samples,  the  contents of the 2-L bottle were divided into
analytical  aliquots and preserved (Figure 9) by the support laboratory before shipment to  the
processing laboratory.

Preparation of Synthetic Audit Samples

      To prepare  the field  synthetic  audit  samples  of  the desired concentrations,  support
laboratory technicians diluted the lot stock concentrates with ASTM Type I reagent-grade water.

                                         120

-------
Each diluted 2-L synthetic audit sample was  prepared for shipment on the same day to  the
processing laboratory as follows:

     1.    Fill a 2-L volumetric flask with 1.5-L deionized water.

     2.    Add a predetermined volume of each of the four stock concentrates (see Table A-1)
           to the flask.

     3.    Fill the flask to volume and mix the solution thoroughly.

     4.    When  the dilution is complete, transfer the 2-L sample to a carboy.  (If 10 sample
           were prepared in one day, the carboy would eventually contain 20-L of diluted stock,
           prepared 2-L at a time.)

     5.    When  these dilution and transfer steps  are  completed, sparge the audit sample
           solution in the carboy with 300 ppm CO2 and equilibrate.   (The equilibration raises
           the acidity of the sample, thereby counteracting the effect of adding the strong base
           Na.SiOg.  It also restores any DIG lost  during sample preparation  steps and, by
           stabilizing the  sample, it minimizes day-to-day sample variation caused by shipping
           and handling).


     Table A-1.  Composition of the Field and Laboratory Synthetic Audit Sample Concentrates,
               National Stream Survey - Phase  I
Stock concentrate Chemical formula
1 AI2(S04)3-(NH4)2S04-24H20
2 FeNH4(SO4)2-12H20
3 NagSiOg
4 CaCI2
NaHC03
C6H4(COOH)2
MgSO4
NaF
MnSO4-H2O
NH4NO3
Na2HPO4
KHC8H4O4
Anaiytes to be measured
Al-ext, Al-total
NH4+, S04*
Fe, NH4+, SO42"
Ha, SiO2
ca, cr
DIG, Na
DOC
Mg, S042-
F, Na
Mn, SO42'
NH4+. NO3"
Na, P
DOC, K
                                            121

-------
      6.    After the sample is sparged, transfer the sample to 2-L bottles and ship the same
           day to the processing laboratory.

      For laboratory synthetic samples, aliquots were prepared from a 2-L bottle, preserved, and
shipped on the same day to the processing laboratory.


Summary of Audit Sample Measurements

      The data tables presented in this section provide information concerning the verification of
the composition of the synthetic audit samples used for the NSS-I and also measured values for
chemical variables  in both synthetic and natural audit samples reported by other laboratories
during other NSWS programs besides the NSS-I.  Table A-2 summarizes the verification analyses
that  were  conducted at the support laboratory for the synthetic  audit samples.   Table A-3
presents results of analyses of EPA reference samples  that  were analyzed at the support
laboratory during the verification analyses of synthetic and natural audit samples (including other
lots of synthetic audit samples used in other NSWS programs).  These data  provide an indication
of the reliability of the  verification  analyses,  as well as estimates of  among-batch  standard
deviations. Tables A-4 through A-7 provide summary statistics from all laboratories (including
the NSS-I participants) that analyzed the particular audit samples used during the NSS-I. These
data provide an indication of  the compatibility of the NSS-I  data base  with the data bases of
other NSWS programs.  Tables A-8 and A-9 present summary statistics associated with the
calculation of index values for each audit sample type. These  index values were developed based
on the data reported from the various laboratories that analyzed each type  of audit sample.
These index values can  be used to evaluate the performance  of a particular analytical laboratory
with  respect to other laboratories that analyzed the same sample.  Index values were calculated
using weighted mean values from each laboratory because of the differences in sample size and
precision.
                                          122

-------
Table A-2.    Levels of Analytes In Synthetic Audit Samples Measured at the Support Laboratory, National
             Stream Survey - Phase I
                                                      Lot 14 (n  = 6)*
Lot 15 (n = 6)a
Variable
Al-ext
Al-total
ANC
BNC
Ca
cr
Cond-lab
DIG
DOC
F
Fe
K
Mg
Mn
Na
NH4+
N03-
P
pH
SiOo
so/-
Units
mg/L
mg/L
A/eq/L
/wq/L
mg/L
mg/L
pS/cm
mg/L
mg/L
mg/L
mg/L
mg/L
mg/L
mg/L
mg/L
mg/L
mg/L
mg/L
pH units
mg/L
mg/L
Theoretical
value*
0.020
0.0199
NC
NC
0.194
0.343
17.5
0.959C
1.00
0.042
0.059
0.203
0.447
0.098
2.75
0.168
0.467
0.0273
NC
1.070
2.280
Mean
0.024*
if
0.026
102
32
0.179
£
0.315
15.2
1.141*
1.10\
0.045*
il
0.039
0.199
i,
0.41
0.097
3.1*
1t
0.14
0.460
A
0.025
if
7.22
1.065
2.207*
s
0.0038
0.0006
2.90
4.37
0.0179
0.003
0.03
0.0720
0.064
0.0012
0.0018
0.0077
0.024
0.0068
0.098
0.013
0.0074
0.0015
0.035
0.0139
0.0534
Mean
0.009*
it
0.016
109
22.8
#
0.208
it
0.300
15.23
1.263*
0.98
0.044*
it
0.041
*
0.217
*
0.43
*
0.093
2.9
*
0.14
0.456
*
0.023
it
7.29
it
1.057
2.172*
S
0.0005
0.0032
2.93
3.85
0.0098
0.0068
0.034
0.0197
0.062
0.0008
0.0024
0.012
0.0103
0.0046
0.20
0.009
0.0154
0.0016
0.024
0.0082
0.0287
 * n - number of  measurements.
   * - mean value significantly different from theoretical at P <. 0.05.
   s « wlthln-batch standard deviation.
 " The theoretical  value is the expected  value of the synthetic audit sample assuming no preparation error
   or external effect.  NC -  theoretical value was  not calculated.
 0 Theoretical value does not Include carbon dioxide added to the solution during the air-equilibration process.
                                                   123

-------
Table A-3.  Summary Statistics for EPA Reference Standards Measured With Audit Samples at the Support
           Laboratory, National Stream Survey - Phase \a
Low concentration
Variable
(units)
Al-ext (mg/L)
Al-total (mg/L)
ANC Oieq/L)
BNC faeq/L)
Ca (mg/L)
Cr (mg/L)
Cond-lab (pS/cm)
DIC (mg/L)
DOC (mg/L)
Fe (mg/L)
F- (mg/L)
K (mg/L)
Mg (mg/L)
Mn (mg/L)
Na (mg/L)
NH4+ (mg/L)
N03- (mg/L)
PH
P (mg/L)
SiO- (mg/L)
SO/' (mg/L)
True
value
0.021
0.021
35.0
..
0.53
1.15
9.27
1.13
1.02
0.04
0.043
0.21
0.18
0.026
0.82
0.19
0.354
5.70
0.130
4.28
0.72

Mean
0.023
0.024
34.2
—
0.53
1.18
8.7
1.20
1.02
0.04
0.047
0.21
0.18
0.025
1.05
0.18
0.353
5.62
0.120
4.45
0.73

s*
0.0052
0.0024
4.18
-
0.026
0.169
0.49
0.048
0.024
0.003
0.058
0.013
0.009
0.004
0.122
0.013
0.0066
0.034
0.0069
0.348
0.021
* Source: Radian Corporation, 1987. Quality
Water Survey Field Programs.
Nevada. DCN
87-203-023-87-02
Final Report

n°
5
4
6
0
6
5
6
6
5
4 •
6
5
6
4
6
5
3
6
5
6
6

Bias"
+0.002
+0.003
-0.8
-„
0.00
+0.03
-0.57
+0.07
0.00

+0.004
0.00
+0.00
-0.001
+0.23
-0.01
-0.001
-0.08
-0.01
+0.17
+0.01
Assurance Audit
True
value
0.146
0.146
68.8
..
4.06
8.08
55.2
5.68
4.10
1.56
0.130
0.98
0.84
0.696
4.65
1.94
-
7.80
1.030
21.40
9.53
Sample
High

Mean
0.151
0.143
67.1
..
3.85
8.22
45.9
5.19
4.08
1.39
0.152
1.04
0.80
0.681
5.72
1.77
-
7.76
0.897
22.24
9.29
Support for
concentration

s
0.0235
0.0170
0.77
„
0.179
0.186
1.69
0.426
0.050
0.244
0.0285
0.092
0.025
0.051
0.608
0.136
-
0.042
0.093
1.187
0.250

n
5
5
5
0
4
3
5
3
3
3
4
4
4
3
4
4
0
6
4
4
4

Biasrf
+0.005
-0.003
-1.7
..
-0.21
+0.14
-9.3
-0.49
-0.02
-0.17
+0.022
+0.06
-0.04
-0.015
+ 1.07
-0.17
—
-0.04
+0.133
+0.84
-0.24
the National Surface
submitted to U.S. Environmental Protection Agency,
. Contract 68-02-3994, Work
Assignment Numbers 47 and
87.
Las Vegas,

b s = within-batch standard deviation.
0 n = number of
measurements.
d Bias - true value - mean value.
                                                124

-------
Table A-4.   Summary Statistics for Measurements of Synthetic Audit Sample Lot 14, From Four Analytical
            Laboratories and Processing Laboratory (Field and Laboratory Audit Samples Combined)3


Variable (units)
Al-ext (mg/L)
Al-total (mg/L)
ANC Qjeq/L)
BNC Oraq/L)
Ca (mg/L)
Cr (mg/L)
Cond-lab (pS/cm)
DIC-eq (mg/L)
DIC-init (mg/L)
DOC (mg/L)
Fe (mg/L)
F (mg/L)
K (mg/L)
Mg (mg/L)
Mn (mg/L)
Na (mg/L)
NH4+ (mg/L)
N03- (mg/L)
pH-ANC (pH units)
pH-BNC (pH units)
pH-eq (pH units)
P (mg/L)
SiO* (mg/L)
S04* (mg/L)


N
4
6
14
14
14
14
14
14
14
14
9
14
14
14
14
14
14
14
14
14
14
14
14
14

LAB 1
Mean
0.027
0.026
104.707
20.850
0.218
0.335
16.329
1.327
1.406
1.063
0.030
0.041
0.195
0.426
0.089
2.819
0.167
0.467
7.021
7.049
7.171
0.023
1.203
2.214
NSS-I

SD N
0.0160 5
0.0212 5
2.7008 9
5.6349 9
0.0455 9
0.0171 9
0.3197 9
0.0846 9
0.0524 9
0.2768 9
0.0138 5
0.0033 9
0.0048 9
0.0036 9
0.0068 9
0.0390 9
0.0073 9
0.0544 9
0.0804 9
0.0866 9
0.2757 9
0.0066 9
0.0267 9
0.0495 9
ELS-II
LAB 2
Mean
0.021
0.039
114.378
50.656
0.193
0.326
19.544
1.374
1.624
1.281
0.040
0.043
0.203
0.436
0.109
2.729
0.188
0.454
6.734
6.740
7.253
0.022
0.957
2.151

SD
0.0019
0.0102
7.4547
4.8706
0.0216
0.0349
0.2698
0.2625
0.0965
0.1253
0.0028
0.0015
0.0074
0.0211
0.0022
0.1162
0.0130
0.0378
0.0635
0.1007
0.0892
0.0034
0.0568
0.2048

N
1
1
6
6
6
6
6
6
6
6
1
6
6
6
6
6
6
6
6
6
6
6
6
LAB 1
Mean
0.017
0.022
113.500
36.817
0.189
0.248
18.950
1.746
1.716
0.820
0.045
0.050
0.189
0.427
0.089
2.717
0.137
0.432
6.853
6.910
7.087
0.020
0.963
6 2.313
LAB 2
SD
-
-
7.8656
12.1733
0.0081
0.0497
0.6473
0.3101
0.2083
0.1942
-
0.0220
0.0081
0.0078
0.0012
0.1177
0.0132
0.0807
0.0647
0.0603
0.0848
0.0044
0.0327
0.0495
N
1
1
2
2
2
2
2
2
2
2
1
2
2
2
2
2
2
2
2
2
2
2
2
2
Mean
0.028
0.023
113.200
18.850
0.198
0.452
19.500
1.237
1.345
1.335
0.044
0.043
0.246
0.445
0.095
2.850
0.162
0.493
7.100
7.145
7.240
0.022
1.089
1.722
SD
-
—
2.8284
1.0607
0.0156
0.1570
0.1414
0.0339
0.0240
0.1202
—
0.0023
0.0099
0.0057
0.0000
0.0396
0.0099
0.0156
0.0424
0.0495
0.0141
0.0004
0.0148
0.8436
Processing Laboratory


Al-mono (mg/L)
Al-nex (mg/L)
Cond-PL (f/S/cm)
DIC-closed (mg/L)
pH-closed (pH units)
True color (PCU)
Turbidity (NTU)










N
8
8
7
9
9
8
8
NSS-I
Mean
0.011
0.017
26.400
1.379
6.958
5.000
0.115

SD
0.0089
0.0091
10.0411
0.0394
0.0964
2.6726
0.0288


ELS-II

N Mean

SD




5 0.009 0.0037
5 0.016 0.0038


0

6 1.370 0.
—
1345




6 6.863 0.0787
5 11.000 13.4164
6 0.147 0.0413
 * For Al-ext, Al-total, and Fe, only laboratory audit samples are included.  For processing laboratory
   measurements, only field audit samples are included.
                                                    125

-------
Table A-5.   Summary Statistics for Measurements of Synthetic Audit Sample Lot 15, From Three Analytical
            Laboratories and Processing Laboratory (Field and Laboratory Audit Samples Combined)4
Variable (units)
Al-ext (mg/L)
Al-total (mg/L)
ANC (/jeq/L)
BNC (jueq/L)
Ca (mg/L)
Cf (mg/L)
Cond-lab (/jS/cm)
DIC-eq (mg/L)
DIC-init (mg/L)
DOC (mg/L)
Fe (mg/L)
F (mg/L)
K (mg/L)
Mg (mg/L)
Mn (mg/L)
Na (mg/L)
NH4+ (mg/L)
N03- (mg/L)
pH-ANC (pH units)
pH-BNC (pH units)
pH-eq (pH units)
P (mg/L)
Si02 (mg/L)
S042- (mg/L)
Al-mono (mg/L)
Al-nex (mg/L)
Cond-PL (/jS/cm)
DIC-closed (mg/L)
pH-closed (pH units)
True color (PCU)
Turbidity (NTU)


N
3
3
4
4
4
4
4
4
4
4
4
4
4
4
4
4
4
4
4
4
4
4
4
4







NSS-I
LAB 1
Mean SD N
0.015 0.0035 23
, 0.029 0.0092 24
105.650 3.5977 29
18.800 1.5188 29
0.199 0.0293 29
0.329 0.0075 29
15.950 0.1291 29
1.233 0.0507 29
1.386 0.0621 29
0.995 0.2651 29
0.031 0.0259 24
0.039 0.0027 29
0.194 0.0025 29
0.430 0.0049 29
0.090 0.0046 29
2.778 0.0228 29
0.155 0.0067 29
0.477 0.0197 28
7.055 0.0520 29
7.075 0.0493 29
7.307 0.1040 29
0.011 0.0083 29
1.189 0.0228 29
2.250 0.0173 29
Processing
NSS-I
N Mean SD
5 0.009 0.0052
5 0.013 0.0058
5 19.060 1.1238
5 1.439 0.0326
5 6.920 0.0628
4 3.750 4.7871
4 0.102 0.0457
ELS-II
LAB 2
Mean
0.015
0.036
118.993

SD
0.0017
0.0263
7.1318
58.890 12.4716
0.177
0.336
19.562
1.321
1.511
1.092
0.024
0.043
0.200
0.437
0.103
2.703
0.174
0.473
6.687
6.689
7.234
0.024
0.974
2.297
Laboratory

N
4
4
0
4
4
4
4
0.0105
0.0334
0.2194
0.2169
0.2674
0.2930
0.0195
0.0017
0.0098
0.0426
0.0032
0.0989
0.0156
0.0341
0.1250
0.1299
0.1464
0.0064
0.0955
0.1198

ELS-II
Mean
0.013
0.013
-
1.431
6.927
3.750
0.092
LAB 1
N Mean
2 0.008
2 0.018
6 135.333
6 40.100
6 0.179
6 0.381
6 18.967
6 1.795
6 1.906
6 0.843
2 0.026
6 0.044
6 0.183
6 0.439
6 0.103
6 2.670
6 0.099
6 0.365
6 6.902
6 6.907
6 7.157
6 0.020
6 1000
6 2.081


SD
0.0035
0.0021
-
0.0802
0.0556
2.5000
0.0287

SD
0.0045
0.0081
65.3334
11.3434
0.0097
0.1519
0.1633
0.3757
0.3993
0.2496
0.0269
0.0024
0.0145
0.0072
0.0034
0.0874
0.0344
0.1387
0.2022
0.1878
0.1033
0.0062
0.0245
0.0724







a For Al-ext, Al-total, and Fe, only laboratory audit  samples are included.  For processing laboratory
  measurements, only field  audit samples are included.
                                                126

-------
Table A-6.    Summary Statistics for Measurements of Bagley Uke Natural Audit Sample, From Four Analytical
             Laboratories and Processing Laboratory (Field and Laboratory Audit Samples Combined)5


Variable (units)
Al-ext (mg/L)
Al-total (mg/L)
ANC (A«q/L)
BNC (peq/L)
Ca (mg/L)
Cr (mg/L)
Cond-lab (pS/cm)
DIC-eq (mg/L)
DIC-init (mg/L)
DOC (mg/L)
Fe (mg/L)
F (mg/L)
K (mg/L)
Mg (mg/L)
Mn (mg/L)
Na (mg/L)
NH4+ (mg/L)
N03- (mg/L)
pH-ANC (pH units)
pH-BNC (pH units)
pH-eq (pH units)
P (mg/L)
SiOj, (mg/L)
S04^ (mg/L)


N
29
29
29
29
29
29
29
29
29
29
29
29
29
29
29
29
29
28
29
29
29
29
29
29

LAB 1
Mean
0.006
0.016
120.934
31.303
1.574
0.162
14.290
1.472
1.424
0.196
0.014
0.021
0.291
0.173
0.004
0.815
0.000
0.007
7.095
7.046
7.263
0.002
9.650
0.639
NSS-I
ELS-II
LAB 2
SD N
0.0024 8
0.0023 8
0.5440 8
1.9126 8
0.0317 8
0.0066 8
0.4798 8
0.0923 8
0.0508 8
0.0864 8
0.0080 8
0.0011 8
0.0076 8
0.0028 8
0.0097 8
0.0175 8
0.0069 8
0.0054 8
0.0693 8
0.0693 8
0.1569 8
0.0033 8
0.3869 8
0.0248 8
Mean
0.008
0.012
122.450
18.537
1.643
0.169
13.975
1.524
1.626
0.407
0.002
0.022
0.294
0.176
0.000
0.812
-0.007
0.050
7.094
7.141
7.216
0.001
8.329
0.615
SD
0.0021
0.0026
4.0733
3.3517
0.0202
0.0147
0.4950
0.0488
0.0836
0.0785
0.0019
0.0041
0.0109
0.0048
0.0007
0.0179
0.0146
0.0747
0.0746
0.0522
0.1046
0.0013
0.4407
0.0339
N
14
14
14
14
15
15
14
14
14
15
15
15
15
15
15
15
15
14
14
14
14
15
15
15
LAB 1
Mean
0.010
0.019
120.664
23.000
1.500
0.175
10.743
1.557
1.680
0.274
0.014
0.021
0.298
0.163
0.002
0.835
0.007
0.007
7.077
7.134
7.295
0.000
9.906
0.631
LAB 2
SD
0.0038
0.0266
3.1750
4.1870
0.0354
0.0349
0.5639
0.0549
0.0500
0.1418
0.0252
0.0015
0.0077
0.0231
0.0063
0.0249
0.0111
0.0180
0.0882
0.1173
0.1782
0.0024
0.2019
0.0127
N
24
24
24
24
24
24
24
24
24
24
24
24
24
24
24
24
24
24
24
24
24
24
24
24
Mean
0.006
0.031
131.958
43.146
1.610
0.185
14.517
1.470
1.464
0.540
0.004
0.027
0.310
0.178
0.001
0.853
0.018
0.015
6.969
6.981
7.307
0.002
8.953
0.634
SD
0.0022
0.0452
12.9136
26.9478
0.0374
0.0507
0.4603
0.2480
0.2332
0.5043
0.0019
0.0016
0.0178
0.0033
0.0009
0.0487
0.0061
0.0186
0.0847
0.0851
0.0634
0.0016
0.2939
0.0588
Processing Laboratory


Al-mono (mg/L)
Al-nex (mg/L)
Cond-PL (pS/cm)
DIC-closed (mg/L)
pH-closed (pH units)
True color (PCU)
Turbidity (NTU)










N
25
25
26
27
27
26
27
NSS-I
Mean
0.013
0.014
15.505
1.703
6.987
5.615
0.175
ELS-II
SD
0.0046
0.0046
2.4486
0.0612
0.0493
3.2010
0.3332








N
0
0
0
37
36
32
35
Mean
-
—
-



SD
-
-
- •








1.647 0.0593
7.126 0.0688
1.250 2.1997
0.066 0.0591
   For processing laboratory measurements, only field audit samples were included.
                                                   127

-------
Table A-7.   Summary Statistics for Measurements of Big Moose Lake Natural Audit Sample, From Four Analytical
            Laboratories and Processing Laboratory (Field and Laboratory Audit Samples Combined)*
NSS-I

Variable (units)
Al-ext (mg/L)
Al-total (mg/L)
ANC fciaq/l)
BNC (peq/l)
Ca (mg/L)
Cr (mg/L)
Cond-lab (pS/cm)
DIC-eq (mg/L)
DIC-init (mg/L)
DOC (mg/L)
Fe (mg/L)
F (mg/L)
K (mg/L)
Mg (mg/L)
Mn (mg/L)
Na (mg/L)
NH4+ (mg/L)
N03- (mg/L)
pH-BNC (pH units)
pH-ANC (pH units)
pH-Eq (pH units)
P (mg/L)
Si02 (mg/L)
SO42' (mg/L)

N
13
13
15
15
15
15
15
15
15
15
15
15
15
15
15
15
15
15
15
15
15
15
15
15
LAB 1
Mean
0.223
0.259
-2.907
64.413
1.829
0.430
23.753
0.069
0.252
3.696
0.037
0.074
0.425
0.321
0.079
0.632
0.062
1.222
5.160
5.139
5.169
0.002
4.668
6.391
I
SD
0.0490
0.0325
2.2381
6.6485
0.0770
0.0492
0.3399
0.0469
0.0392
0.1552
0.0188
0.0036
0.0106
0.0049
0.0263
0.0444
0.0084
0.0341
0.0939
0.0748
0.0144
0.0028
0.1127
0.0478
LAB 2
N Mean
24 0.210
24 0.281
24 -1.625
24 79.846
24 1.937
24 0.434
24 25.479
24 0.103
24 0.230
24 4.065
24 0.057
24 0.074
24 0.427
24 0.326
24 0.086
24 0.613
24 0.067
24 1.232
24 5.235
24 5.224
24 5.245
24 0.002
24 3.987
24 6.290
SD
0.0194
0.0168
5.9768
9.4930
0.0332
0.0536
0.2187
0.0906
0.0829
0.2366
0.0108
0.0015
0.0264
0.0266
0.0154
0.0316
0.0086
0.0656
0.1879
0.1879
0.0767
0.0013
0.3106
0.2018
N
29
35
38
38
38
38
38
38
38
38
38
38
38
38
38
38
38
38
38
38
38
38
38
38
LAB 1
Mean
0.141
0.259
-1.987
ELS-II
LAB 2
SD
0.0562
0.0633
4.2976
72.768 14.5541
1.983
0.406
25.380
0.267
0.428
3.408
0.078
0.086
0.404
0.346
0.072
0.609
0.056
1.205
5.143
5.120
5.180
0.002
4.418
6.567
0.6425
0.1930
1.0147
0.1231
0.1253
0.2587
0.1159
0.0754
0.0177
0.1334
0.0041
0.0270
0.0717
0.3054
0.0773
0.0892
0.3357
0.0018
0.2728
1.3722
N
24
27
31
31
31
31
31
31
31
31
31
31
31
31
31
31
31
31
31
31
31
31
31
31
Mean
0.162
0.246
-3.677
74.677
1.925
0.417
25.023
0.096
0.454
3.419
0.044
0.073
0.393
0.322
0.071
0.566
0.059
1.199
5.137
5.081
5.135
0.002
4.364
6.364
SD
0.0420
0.0388
2.1180
7.2473
0.0551
0.0233
0.4938
0.0439
0.0472
0.1512
0.0067
0.0194
0.0345
0.0121
0.0040
0.1660
0.0322
0.0320
0.0727
0.0457
0.3797
0.0031
0.2541
0.3626
Processing Laboratory
NSS-I

Al-mono (mg/L)
Al-nex (mg/L)
Cond-PL O^S/cm)
DIC-closed (mg/L)
pH-closed (pH units)
True color (PCU)
Turbidity (NTU)








N
24
24
26
27
27
25
25
Mean SD
0.195
0.058
25.473
0.526
5.148
18.000
0.607
0.0117
0.0127
1.7644
0.0620
0.0594
4.5644
2.1666







N
39
39
0
39
39
38
38
ELS-II
Mean SD
0.189
0.046
..
0.564
5.129
15.789
0.143
0.0205
0.0190
„



0.0513
0.0508
4.8666
0.0934
 For processing laboratory measurements, only field audit samples were included.
                                                 128

-------
Table A-8.   Calculated Index Values for Measurements of Synthetic Audit Samples (Lots 14 and 15)
            (Field and Laboratory Audit Samples Combined)5
Lot 14
Index Value

Variable (units)
Al-ext (mg/L)
Al-total (mg/L)
ANC (^eq/L)
BNC (Aieq/L)
Ca (mg/L)
Cr (mg/L)
Cond-lab (^iS/cm)
DIC-eq (mg/L)
DIC-init (mg/L)
DOC (mg/L)
Fe (mg/L)
F (mg/L)
K (mg/L)
Mg (mg/L)
Mn (mg/L)
Na (mg/L)
NH4+ (mg/L)
N03- (mg/L)
pH-BNC (pH units)
pH-ANC (pH units)
pH-eq (pH units)
P (mg/L)
SiO2 (mg/L)
S042'(mg/L)


N
4
4
4
4
4
4
4
4
4
4
4
4
4
4
4
4
4
4
4
4
4
4
4
4

Grand
mean6
0.021
0.036
106.578
24.033
0.192
0.330
18.335
1.295
1.410
1.181
0.039
0.043
0.197
0.427
0.095
2.813
0.167
0.472
6.968
6.907
7.230
0.022
1.120
2.241


SDC
0.00086
0.00403
0.64174
0.61562
0.00283
0.00416
0.05164
0.01605
0.01018
0.03082
0.00120
0.00042
0.00106
0.00088
0.00040
0.00929
0.00165
0.00705
0.01384
0.01200
0.00907
0.00024
0.00519
0.01092


CIrf
0.00136
0.00642
1.02115
0.97960
0.00451
0.00662
0.08217
0.02554
0.01619
0.04903
0.00191
0.00067
0.00168
0.00140
0.00064
0.01479
0.00262
0.01122
0.02203
0.01909
0.01444
0.00038
0.00826
0.01737
Processing

N
3
3
3
3
3
3
3
3
3
3
3
3
3
3
3
3
3
3
3
3
3
3
3
3
Laboratory
Index Value

Variable (units)
Al-mono (mg/L)
Al-nex (mg/L)
Cond-PL OuS/cm)
DIC-closed (mg/L)
pH-closed (pH units)
True color (PCU)
Turbidity (NTU)

N
2
2
1
2
2
2
2
Grand
mean*
0.010
0.016
26.400
1.379
6.911
5.145
0.123

SD°
0.00145
0.00152
3.79517
0.01277
0.02271
0.93341
0.00871

CIrf
0.01300
0.01363
-
0.11473
0.20408
8.38633
0.07829

N
2
2
1
2
2
2
2
Lot 15
Index Value3
Grand
mean*
0.015
0.028
114.337
23.105
0.178
0.331
18.624
1.268
1.433
1.032
0.024
0.043
0.196
0.434
0.103
2.751
0.164
0.473
6.879
6.853
7.227
0.023
1.066
2.244


SDC
0.00035
0.00315
1.06563
0.71298
0.00174
0.00322
0.03061
0.02124
0.02599
0.04512
0.00373
0.00030
0.00102
0.00182
0.00053
0.00935
0.00217
0.00537
0.01682
0.01694
0.02092
0.00105
0.00692
0.00779


CIrf
0.00088
0.00781
2.64718
1.77114
0.00431
0.00800
0.07604
0.05275
0.06457
0.11210
0.00926
0.00074
0.00253
0.00453
0.00131
0.02324
0.00540
0.01334
0.04179
0.04209
0.05198
0.00260
0.01719
0.01936

Index Value*
Grand
mean*
0.012
0.013
19.060
1.438
6.924
3.750
0.095

SD"
0.00139
0.00098
0.50259
0.01372
0.01977
1.10801
0.01216

Cld
0.01245
0.00880
--
0.12326
0.17759
9.95504
0.10927
 a For Al-ext, Al-total, and Fe, only laboratory audit samples were used.  For processing laboratory
   measurements, only field  audit samples were used.
 * Grand mean calculated from weighted  mean from four analytical laboratories.
 0 SD =  standard deviation, estimated as the square root of the reciprocal of the sum of individual
   laboratory weighting factors.
 d CI = One-sided 95% confidence interval.
                                                    129

-------
 Table A-9.
Index Values for Measurements of Bagley Lake and Big  Moose Lake  Natural Audit Samples
(Field and Laboratory Audit Sample Combined)3
Bagley Lake
Index Value

Variable (units)
Al-ext (rng/L)
Al-total (mg/L)
ANC (peq/L)
BNC (//eq/L)
Ca (mg/L)
Cr (mg/L)
Cond-lab (fjSlcm)
DIC-eq (mg/L)
DIC-init (mg/L)
DOC (mg/L)
Fe (mg/L)
F (mg/L)
K (mg/L)
Mg (mg/L)
Mn (mg/L)
Na (mg/L)
NH4+ (mg/L)
NOg' (mg/L)
pH-ANC (pH units)
pH-BNC (pH units)
pH-eq (pH units)
P (mg/L)
SiO2 (mg/L)
S042' (mg/L)


N
4
4
4
4
4
4
4
4
4
4
4
4
4
4
4
4
4
4
4
4
4
4
4
4

Grand
mean*
0.006
0.016
120.954
29.685
1.588
0.163
13.844
1.520
1.515
0.257
0.004
0.023
0.295
0.175
0.001
0.820
0.008
0.007
7.059
7.057
7.293
0.002
9.476
0.633


SDff
0.00028
0.00039
0.09999
0.32494
0.00359
0.00118
0.05626
0.00920
0.00736
0.01288
0.00032
0.00016
0.00106
0.00040
0.00014
0.00254
0.00084
0.00096
0.00890
0.00867
0.01096
0.00022
0.03370
0.00255


CI*
0.00044
0.00062
0.15911
0.51705
0.00571
0.00188
0.08953
0.01465
0.01172
0.02050
0.00051
0.00025
0.00168
0.00063
0.00023
0.00405
0.00134
0.00153
0.01417
0.01379
0.01744
0.00036
0.05362
0.00406
Processing

N
4
4
4
4
4
4
4
4
4
4
4
4
4
4
4
4
4
4
4
4
4
4
4
4
Laboratory
Bagley Lake
Index Value

Variable (units)
Al-rnono (mg/L)
Al-nex (mg/L)
Cond-PL OuS/cm)
DIC-closed (mg/L)
pH-elosed (pH units)
True color (PCD)
Turbidity (NTU)

N
1
1
1
2
2
2
2
Grand
mean*
0.013
0.014
15.505
1.670
7.043
2.461
0,068

SD"
0.00093
0.00092
0.48021
0.00751
0.00731
0.33057
0.00987

CI*
«»
..
..
0.06746
0.06567
2.97009
0.08872

N
2
2
1
2
2
2
2
Big Moose Lake
Index Value5
Grand
mean*
0.197
0.272
-3.110
72.897
1.924
0.420
25.120
0.106
0.359
3.568
0.046
0.074
0.414
0.322
0.072
0.612
0.065
1.209
5.100
5.147
5.173
0.002
4.485
6.382

Big

SDC
0.00329
0.00284
0.28133
0.85272
0.00584
0.00371
0.03545
0.00594
0.00581
0.01833
0.00103
0.00029
0.00178
0.00107
0.00048
0.00343
0.00132
0.00451
0.00659
0.00827
0.00360
0.00018
0.02032
0.01162


Cld
0.00524
0.00451
0.44766
1.35687
0.00929
0.00590
0.05640
0.00945
0.00925
0.02917
0.00164
0.00046
0.00283
0.00170
0.00077
0.00546
0.00210
0.00718
0.01049
0.01317
0.00573
0.00029
0.03234
0.01849

Moose Lake
Index Value a
Grand
mean*
0.193
0.053
25.473
0.552
5.136
16.735
0.144

BDC
0.00193
0.00198
0.34604
0.00677
0.00663
0.59714
0.01514

CIrf
0.01735
0.01776

0.06080
0.05954
5.36510
0.13607
* For processing laboratory measurements, only field audit samples were used.
  Grand mean calculated from weighted means from four analytical laboratories.
0 SD  = standard deviation, estimated as the square root of the  reciprocal of the sum of individual laboratory
  weighting  factors.
d CI = One-sided 95% confidence interval.
                                               130

-------
                                           Appendix B
                                        Data Qualifiers
      Qualifiers were assigned to data points in the data sets to  identify unusual conditions or
results outside the expected criteria or  limits.  Tags  assigned during collection and analyses
activities are listed in Table B-1.   Flags assigned as a result of verification activities are listed in
Table B-2.
  Table B-1.   Field and Laboratory Data Qualifiers (Tags), National Stream Survey - Phase
     Qualifier
                                               Indicates
        A
        B
        C
        D
        E
        F
        G
        H

        J
        K
        L
        M
        N
        P
        Q
        R
        S
        T
        U
        V
        w

        X,  Y, Z
Instrument unstable.
Redone,  first reading not acceptable.
Instruments,  sampling gear not vertical in water column.
Slow stabilization.
Result not available;  sample destroyed during shipment.
Result outside QA criteria (with consent of QA manager).
Atypical result; already reanalyzed and confirmed by the laboratory manager.
Holding time  exceeded criteria.
Result not available; insufficient sample volume shipped to analytical
   laboratory from the processing laboratory.
Result  not available; entire aliquot not shipped.
Not analyzed because of Interference.
Result  not available; sample  lost or destroyed by laboratory.
Not required.
Result outside QA criteria,  but  insufficient volume for reanalysis.
Result outside QA criteria.
Result from reanalysis.
Contamination suspected.
Leaking  container.
Result not required by procedure; unnecessary.
% ion balance difference value (Form  16) outside criteria because of high  DOC.
% difference  calculation  for calculated  ANC  (Form 14) outside criteria because
    of high DOC.
 Available for miscellaneous comments in the field and processing laboratory only.
                                                   131

-------
Table B-2.    Verification Data Qualifiers  (Flags), National  Stream
               Survey - Phase I

  FLAGS USED WITH ANION/CATION BALANCE CHECK PROGRAM:

    AO  Anion/Cation % Ion Balance Difference is outside criteria due  to  an
         unknown cause.

    A1   Anion/Cation % Ion Balance Difference is outside criteria due  to  unmeasured
         anions/cations (other anions/cations not  considered  in % ion balance
         difference calculation).

    A2   Anion/Cation % Ion Balance Difference is outside criteria due  to  anion (flag
         suspect anion) contamination.

    A3   Anion/Cation % Ion Balance Difference is outside criteria due  to  to cation
         contamination.

    A4   Anion/Cation % Ion Balance Difference is outside criteria due  to  unmeasured
         organic protolvtes (fits Oliver ModeH.

    A5   Anion/Cation % Ion Balance Difference is outside criteria due  to  possible
         analytical error -  anion concentration too high (flag suspect anion).

    A6   Anion/Cation % Ion Balance Difference is outside criteria due  to  possible
         analytical error -  cation concentration too low (flag suspect cation).

    A7   Anion/Cation % Ion Balance Difference is outside criteria due  to  possible
         analytical error -  anion concentration too low (flag suspect anion).

    A8   Anion/Cation % Ion Balance Difference is outside criteria due  to  possible
         analytical error -  cation concentration too high  (flag  suspect cation).

    A9   Anion/Cation % Ion Balance Difference is outside criteria due  to  possible
         analytical error -  alkalinity (ANC) measurement.


  FLAGS GENERATED BY APPROPRIATE BLANK EXCEPTION PROGRAM:

    BO   External (field) blank is above  expected criteria for pH, DIG, DOC, specific
         conductance, ANC,  and  BNC determinations.

    B1   Internal (laboratory) blank is >2 x required detection limit for  DIC,  DOC,
         and specific conductance determinations.

    B2   External (field) blank is above  expected criteria and contributed >20% to
         sample concentrations.  (This flag is not used for pH,  DIC, DOC, specific
         conductance, ANC,  and BNC determinations.)

    B3   Internal (laboratory) blank is 2  x required detection limit  and contributes
         >10% to the sample concentrations. (This flag is not  used for DIC, DOC,
         and specific conductance determinations.)

                                                                          (Continued)

                                       132

-------
Table B-2.   (Continued.)
  FLAGS GENERATED BY APPROPRIATE BLANK EXCEPTION PROGRAM (continued):

    B4  Potential negative sample bias based on internal (laboratory) blank data.

    B5  Potential negative sample bias based on external (field) blank data.


  FLAGS USED WITH CONDUCTANCE BALANCE CHECK PROGRAM:

    CO  % Conductance Difference is outside criteria due to unknown cause.

    C1  % Conductance Difference is outside criteria due io possible analytical
         error-anion concentration too high (flag suspect anion).

    C2  % Conductance Difference is outside criteria due to anion contamination.

    C3  % Conductance Difference is outside criteria due to cation contamination.

    C4  % Conductance Difference is outside criteria due to unmeasured organic ions
          (fits  Oliver Model).

    C5  % Conductance Difference is outside criteria due to possible analytical error
          in specific conductance measurement.

    C6  % Conductance Difference is outside criteria due to possible analytical
         error-anion concentration too low  (flag suspect anion).

    C7   % Conductance Difference is outside criteria due to unmeasured
          anions/cations (other anions/cations not measured in % conductance
          difference calculation).

    C8   % Conductance Difference is outside criteria due to possible analytical
          error-cation concentration too low (flag suspect cation).

    C9   % Conductance Difference is outside criteria due to possible analytical
          error-cation concentration too high (flag suspect cation).

   FLAGS GENERATED BY DUPLICATE PRECISION EXCEPTION PROGRAM:

     D2   External (field) duplicate precision  exceeded the maximum expected %
          relative standard  deviation, and both the routine and duplicate sample
          concentrations were >.10 X required detection limit.

     D3   Internal (laboratory) duplicate precision exceeded the maximum required %
          relative standard  deviation, and both the routine and duplicate sample
          concentrations were >.10 x required detection limit.


                                                                           (Continued)
                                         133

-------
Table B-2.  (Continued.)
  FLAGS USED WHEN FIELD DATA ARE OUTSIDE CRITERIA:

    FO   % Conductance difference exceeded criteria when in situ field conductance
         value was substituted.

    F1   Hillman/Kramer protolyte analysis program indicated field pH problem when
         stream site pH value was substituted.

    F2   Hillman/Kramer protolyte analysis program indicated unexplained problem
         with stream site pH or processing laboratory PIC values when stream site
         pH  value was substituted.

    F3   Hillman/Kramer protolyte analysis program indicated field problem
         processing  laboratory pH.

    F4   Hillman/Kramer protolyte analysis program indicated field problem
         processing  laboratory DIG.

    F5   Hillman/Kramer protolyte analysis program indicated unexplained problem
         with processing laboratory pH or DIG values when processing laboratory pH
         value was substituted.

    F6   % Conductance Difference exceeded criteria when processing laboratory
         specific conductance value was  substituted.

  FLAGS GENERATED BY HOLDING  TIME EXCEPTION PROGRAM:

    HO   The maximum holding time criteria were not met.

    H1   No  "Date Analyzed" data were submitted for reanalysis data.

  FLAG GENERATED BY DETECTION LIMIT EXCEPTION PROGRAM:

    L1   Instrumental Detection Limit exceeded required detection limit and  sample
         concentration was  <10 x instrumental detection limit.

  FLAGS GENERATED BY AUDIT CHECK PROGRAM:

    NO   Audit sample value exceeded upper control limit.

    N1   Audit sample value was below control limit.

  FLAGS GENERATED BY HILLMAN/KRAMER PROTOLYTE ANALYSIS PROGRAM:

    PO   Laboratory problem-initial pH from alkalinity (ANC)  titration.

    P1   Laboratory problem-initial pH from acidity (BNC) titration.

    P2   Laboratory problem-unexplained - initial  pH from ANC or BNC titration.

                                                                        (Continued)


                                      134

-------
Table B-2.   (Continued.)
  FLAGS GENERATED BY HILLMAN/KRAMER PROTOLYTE ANALYSIS PROGRAM
  (Continued):

    P3   Laboratory problem-initial DIG determination.

    P4   Laboratory problem-air-equilibrated pH or DIG determinations.

    P5   Laboratory problem-unexplained - initial pH from  ANC or BNC titrations
         or initial DIG determinations.

    P6   Laboratory problem-alkalinity (ANC) determination.

    P7   Laboratory problem»CO2-acidity (BNC) determination.

  FLAGS GENERATED BY QCCS EXCEPTION PROGRAM(S):

    Q1   Quality  Control Check Sample was above contractual criteria.

    Q2   Quality  Control Check Sample was below contractual criteria.

    Q3   Insufficient number of Quality Control Check Samples were  measured.

    Q4   No Quality Control Check Sample was analyzed.

    Q5   Detection Limit Quality Control Check Sample was not 2 to  3 x Required
         Detection Limit and measured value was  not within  20% of  the theoretical
         concentration.

  MISCELLANEOUS FLAGS:

    MO   Value obtained using a method which is unacceptable as specified in the
         Invitation for Bid  contract.

    M1   Value reported is  questionable due to limitations  of  the laboratory
         methodology.

    XO   Invalid but confirmed data based on QA review.

    X1   Extractable aluminum concentration is greater than total aluminum
         concentration  by 0.010 mg/L where extractable aluminum >_  0.015 mg/L.

    X2   Invalid but confirmed data-potential  aliquot switch.

    X3   Invalid but confirmed data-potential  gross contamination of aliquot or
         parameter.

    X4   Invalid but confirmed data-potential  sample (all aliquots) switch.

         Values for flags XO through X4 should not be included in any statistical
         analysis.

                                                                         (Continued)
                                       135

-------
Table B-2.  (Continued.)
  MISCELLANEOUS FLAGS (continued):



    X7   Site disturbance in watershed (e.g., strip mine).



  MISSING CODE VALUE



    "."    Value never reported.



         (Note:  This code appears in numeric fields only.)
                                      136

-------
                                   APPENDIX C

                            ACCEPTANCE CRITERIA
     Appendix C consists of control limits for field blank, performance audit, and field duplicate
pair samples used in the exception-generating programs found in AQUARIUS II.


CALCULATION OF FIELD BLANK SAMPLE CONTROL LIMITS

     Criteria for determining  contamination  were needed in order to check for systematic
contamination problems during sample collection and analysis and before preparing a verified
data set.  Some control limits were established on the basis of specifications provided by the
instrument manufacturer; others reflected DQOs (i.e., the level of detectability needed to meet the
goals of the survey). Control limits for some analytes could be defined only in terms of analytical
experience and  intuitive assumptions based on that experience, because there  were not any
acceptable precedents.

     Upper control limits for the NSS-I blank samples were determined statistically on the basis
of NSS-I Pilot experience  and the analytical results obtained for NSS-I Pilot field blanks.  The
same type of sampling apparatus was used to collect field blanks for both surveys.  The 95th
percentile (Pgs)  nonparametric test was used to calculate the upper limit at which blank values
would be flagged.  The value of the required detection limit was used when the Pg5 statistic was
below the required detection limit. The lower limit was designated as the negative value of the
required detection  limit, except for pH measurements for which the lower  limit  was the 5th
percentile of the  NSS-I  Pilot  field  blanks.   Anything less than  this negative value was
unacceptable and was attributed to excessive instrumental drift  or to inaccurate  calibration of
the instrument.

     Table C-l presents the field blank control limits for the  NSS-I.  Field blank concentrations
that  were outside these limits were considered suspect and were flagged.  Establishing these
limits prior to a full-scale statistical  analysis was  essential  in order to identify contamination
trends  as they occurred.  The  detailed statistical analysis of the NSS-I field blank values for
estimates of detectability was performed after data verification was completed.
                                         137

-------
TABLE C-1.   Field Blank Control Limits,  National Stream Survey - Phase
Variable*
Al-ext
Al-total
ANC (peq/L)
BNC fjjeq/L)
Ca
cr
Cond-lab (pS/cm)
DIC-eq
DIC-init
DOC
F-
Fe
K
Mg
Mn
Na
NH4+
N03-
P
pH-ANC (pH units)
pH-BNC (pH units)
pH-eq (pH units)
Si02
SO42-
Low limit*
-0.0050
-0.0050
-5.0000
-5.0000
-0.0100
-0.0100
-0.9000
-0.0500
-0.0500
-0.1000
-0.0050
-0.0100
-0.0100
-0.0100
-0.0100
-0.0100
-0.0100
-0.0050
-0.0020
5.4880*
5.5600*
-5.6420e
-0.0500
-0.0500
High limitc
0.0100
0.0619
5.6160
26.3000
0.0400
0.0632
1.0000
0.3620
0.2040
0.5400
0.0050
0.0100rf
0.0100rf
0.0100rf
0.0100tf
0.0114
0.0210
0.0354
0.0084
5.9000
5.9680
6.7120
0.0622
0.0500rf
a Units are in mg/L unless otherwise noted.
b The low limit is the negative value of the required detection limit.
0 The high limit is the 95th percentile of the NSS-I Pilot field blanks.
d The required detection limit is substituted for the 95th percentile of the NSS-I Pilot  field  blanks.
6 The 5th percentile is substituted for the negative value of the required detection limit.
                                                      138

-------
PERFORMANCE AUDIT SAMPLE CONTROL LIMITS


     Final audit sample control limits were generated after all analytical laboratory data (68
batches) had been entered into the raw data set. The QA plan (Drouse et al., 1986a) provides
information about how to calculate the control limits.  Values that were outside the control limits
were considered suspect and were the basis for requesting confirmation of the values reported
by the analytical laboratories. Tables C-2 through C-5 give the control limits for the two natural
and two synthetic samples used during the NSS-I.
 TABLE C-2 Control Limit* for Performance Audit Samples From Bagley Lake, National Stream
          Survey - Phase I

                                            Control limits
Variable
Al-ext (mg/L)
Al-total (mg/L)
ANC Oraq/L)
BNC 0*»q/L)
Ca (mg/L)
Of (mg/L)
Cond-lab (fiSlcm)
DIC-eq (mg/L)
DIC-init (mg/L)
DOC (mg/L)
F (mg/L)
Fe (mg/L)
K (mg/L)
Mg (mg/L)
Mn (mg/L)
Na (mg/L)
NH4+ (mg/L)
N03- (mg/L)
P (mg/L)
pH-ANC (pH units)
pH-BNC (pH units)
pH-eq (pH units)
Si02 (mg/L)
SO42- (mg/L)
Field
Lower
limit
0.0010
-0.0010
113.5960
13.5300
1.4202
0.0841
9.3027
1.0590
1.0370
0.0590
0.0186
-0.0060
0.2846
0.1666
-0.0020
0.7509
-0.0030
-0.0260
-0.0040
6.7789
6.7898
7.0722
8.0746
0.5660
audit
Upper
limit
0.0105
0.0330
133.6830
48.1790
1.6990
0.2976
17.2895
1.8913
2.0243
0.6909
0.0312
0.0115
0.3347
0.1836
0.0036
0.9428
0.0340
0.0518
0.0066
7.2233
7.2565
7.5293
10.5190
0.6788
Laboratory
Lower
limit
0.0030
0.0010
113.5220
10.4010
1.4448
0.1386
8.2771
1.3650
1.3480
0.1640
0.0155
-0.0080
0.2752
0.1658
-0.0020
0.7687
-0.0150
-0.0070
0.0000
6.9026
6.8278
7.2024
8.3863
0.5637
audit
Upper
limit
0.0136
0.0290
139.2050
53.7490
1.7258
0.1790
17.2394
1.7007
1.8676
0.3786
0.0340
0.0199
0.3260
0.1859
0.0031
0.9173
0.0339
0.0124
0.0037
7.0813
7.2387
7.3812
10.3577
0.6945
                                         139

-------
TABLE C-3  Control Limits for Performance Audit Samples From Big Moose Lake, National Stream
           Survey - Phase I
Control limits
Variable
Al-ext (mg/L)
Al-total (mg/L)
ANC (/wq/L)
BNC 0/eq/L)
Ca (mg/L)
cr (mg/L)
Cond-lab foiS/cm)
DIC-eq (mg/L)
DIC-init (mg/L)
DOC (mg/L)
F (mg/L)
Fe (mg/L)
K (mg/L)
Mg (mg/L)
Mn (mg/L)
Na (mg/L)
NH4+ (mg/L)
N03- (mg/L)
P (mg/L)
pH-ANC (pH units)
pH-BNC (pH units)
pH-eq (pH units)
SiO2 (mg/L)
SO42- (mg/L)
Field
Lower
limit
0.1400
0.2120
-6.7820
51.0900
1.7282
0.3243
22.8934
-0.0540
0.0970
3.3390
0.0704
0.0110
0.3839
0.3100
0.0610
0.5482
0.0460
1.1480
-0.0010
4.9903
5.0226
5.0818
3.3512
5.9858
audit
Upper
limit
0.2719
0.3241
-0.8270
96.2952
2.0497
0.5501
26.7436
0.2149
0.3571
4.4967
0.0774
0.0889
0.4737
0.3441
0.1023
0.6949
0.0840
1.3045
0.0052
5.3424
5.3069
5.3304
5.1708
6.6752
Laboratory
Lower
limit
0.1840
0.2370
-9.3400
44.6450
1.7949
0.3677
22.8268
-0.1220
0.1100
3.2830
0.0691
0.0420
0.3709
0.3128
0.0640
0.5807
0.0470
1.0870
-0.0010
4.9817
4.8747
5.0834
3.3070
5.9118
audit
Upper
limit
0.3004
0.3260
1.5040
104.1552
2.0581
0.4388
26.7897
0.3457
0.3760
4.5851
0.0793
0.0620
0.4680
0.3448
0.1054
0.6398
0.0841
1.3434
0.0033
5.2787
5.5468
5.3237
5.1361
6.7373
                                                 140

-------
TABLE C-4  Control Umlto for Performance Audit Samples From Synthetic Lot 14, National Stream
           Survey -  Phase I
Control limits
Variable
Al-ext (mg/L)
Al-total (mg/L)
ANC (peq/L)
BNC (peq/L)
Ca (mg/L)
Cr (mg/L)
Cond-lab (juS/cm)
DIC-eq (mg/L)
DIC-init (mg/L)
DOC (mg/L)
F (mg/L)
Fe (mg/L)
K (mg/L)
Mg (mg/L)
Mn (mg/L)
Na (mg/L)
NH4+ (mg/L)
NO3- (mg/L)
P (mg/L)
pH-ANC (pH units)
pH-BNC (pH units)
pH-eq (pH units)
SiO2 (mg/L)
SO42- (mg/L)
Field audit
Lower
limit
0.0000
-0.0340
87.6412
-1.7160
0.1227
0.2635
13.5135
0.8790
1.1890
0.8550
0.0341
-0.0170
0.1852
0.4078
0.0630
2.7904
0.1470
0.3790
0.0140
6.4931
6.4869
6.8688
0.7382
1.7440

Upper
limit
0.0179
0.1079
130.6250
71.6280
0.2715
0.4017
22.0642
1.7018
1.8158
1.4353
0.0514
0.0317
0.2172
0.4426
0.1284
2.8668
0.2036
0.5638
0.0234
7.2957
7.3419
7.4534
1.4360
2.6095
Laboratory
Lower
limit
0.0020
-0.0100
97.6354
-6.0570
0.1179
0.3143
13.8833
1.1280
1.1820
0.6240
0.0355
0.0150
0.1925
0.4215
0.0750
2.7060
0.1390
0.3350
0.0160
6.5537
6.5200
6.8165
0.8442
1.9712
audit
Upper
limit
0.0517
0.0664
116.3490
67.9430
0.3127
0.3386
21.0309
1.5440
1.7862
1.5587
0.0469
0.0561
0.1996
0.4332
0.1207
2.8847
0.2106
0.5757
0.0310
7.2834
7.3542
7.7434
1.3932
2.4235
                                                 141

-------
TABLE C-5  Control Limits for Performance Audit Sample* From Synthetic Lot 15, National Stream
           Survey - Phase I
Control limits
Variable
Al-ext (mg/L)
Al-total (mg/L)
ANC (jieq/L)
BNC (jueq/L)
Ca (mg/L)
Cr (mg/L)
Cond-lab (/jS/cm)
DIC-eq (mg/L)
DIC-init (mg/L)
DOC (mg/L)
F (mg/L)
Fe (mg/L)
K (mg/L)
Mg (mg/L)
Mn (mg/L)
Na (mg/L)
NH4+ (mg/L)
N03- (mg/L)
P (mg/L)
pH-ANC (pH units)
pH-BNC (pH units)
pH-eq (pH units)
Si02 (mg/L)
S042- (mg/L)
Field
Lower
limit
0.0000
0.0110
114.6760
29.4950
0.1469
0.2744
19.3235
0.80SO
1.0060
-0.8090
0.0382
-0.0100
0.1806
0.4074
0.0970
2.2153
0.1270
0.4170
0.0020
6.1860
6.1841
6.6952
0.6790
1.8906
audit
Upper
limit
0.0125
0.0421
125.6740
98.8250
0.1914
0.3803
20.1164
1.8036
2.3103
2.6054
0.0472
0.0161
0.2233
0.4801
0.1097
3.1446
0.2120
0.4886
0.0454
7.1059
7.0878
7.8487
1.2451
2.5333
Laboratory
Lower
limit
0.0110
0.0140
99.0350
14.3340
0.1556
0.2752
16.3188
0.8630
0.9380
0.7810
0.0419
-0.0170
0.1788
0.4204
0.0920
2.5396
0.1430
0.4230
0.0190
6.3907
6.3810
6.9524
0.7600
2.0754
audit
Upper
limit
0.0186
0.0465
134.1080
90.1090
0.2039
0.3910
21.7168
1.7594
1.9948
1.3563
0.0452
0.0667
0.2197
0.4654
0.1126
2.8954
0.1966
0.5173
0.0273
7.1035
7.1254
7.5225
1.2530
2.5360
                                                142

-------
FIELD ROUTINE-DUPLICATE PAIR PRECISION LIMITS

      Precision  limits  for field  routine-duplicate  pairs  were designed using the  DQO for the
intralaboratory precision estimates as  a reference.  Some flexibility was allowed since these
samples pass through the system from the field to the analytical laboratory, whereas laboratory
duplicates do not.

      A data qualifier flag  (Appendix B)  was given to all  the  samples  in the batch when the
precision estimates  for the  field routine-duplicate pair exceeded  the  allowable limit.    The
measurements of the field routine-duplicate pairs for this variable are considered suspect and
the analytical laboratories are asked to confirm the value.
 TABLE C-6  Precision Limits for Field Routine-Duplicate Pairs
                                             Field Routine-Duplicate Pair Precision Limit
       Variabie                                Percent Relative Standard Deviation (%RSD)*

                                                10 (if Ai-ext concentration > 0.01 mg/L)
                                                20 (if Al-ext concentration =s 0.01 mg/L)
       A|.tota)                                    10 (if Al-total concentration > 0.01 mg/L)
                                                20 (if Al-total concentration «s 0.01 mg/L)
       ANC                                      10
       BNC                                      10
       Ca                                       5
       cr                                       5
       Cond-lab                                  3
       DIC-eq                                    10
       DIC-init                                   10
       DOC                                     10
       F-                                       10
       Fe                                       10
       K                                       10
       Mg                                       5
       Mn                                      10
       Na                                      10
NH4+


                                         20 (if P concentration s 0.01 mg/L)
        N03-                                     10
        p                                       10 (if P concentration > 0.01 mg/L)
        pH-ANC                                  ±0-1 (PH unit)
        pH-BNC                                  ±0-1 (PH unit)
        pH-eq                                    ±0-1 (PH unit)
        SiO2                                      5
        SO42-                                     5
  e This limit was the %RSD at all concentrations, unless otherwise noted.
                                             143

-------
                                 APPENDIX D
       ESTIMATING RELATIVE INTERLABORATORY BIAS FOR THE
                        NATIONAL STREAM SURVEY
     The following report assesses relative interlaboratory bias during Phase I of the National
Stream Survey.  The report was submitted to Lockheed Engineering and Management Services
Company by Systems Applications, Inc.
                         ESTIMATING RELATIVE INTERLABORATORY BIAS
                             FOR THE  NATIONAL STREAMS SURVEY

                                    SYSAPP-87/093

                                     15 June 1987
                                     Prepared for

                                     Mohammed Mi ah

                                    Lockheed EMSCO
                            1050 East Flamingo Road, Suite 209
                              Las Vegas, Nevada  89119-7432
                            (EPA Contract 68-03-3249, Task 2)
                                      Prepared by

                                     S. 0. Edland
                                     T. J. Permutt
                                    T. S. Stocking

                               Systems Applications, Inc.
                                  101 Lucas Valley Road
                                  San Rafael, CA  94903
                                       144

-------
                                 CONTENTS



1    INTRODUCTION	      1

2    STATISTICAL METHODS	      3

3    RESULTS	      8

4    APPLYING THE RESULTS	     12

5    CONCLUSIONS	     13

Appendix A:  SCATTERPLOTS OF THE MEANS OF PERFORMANCE AUDIT DATA

Appendix B:  MODELS CONSIDERED AND DERIVATION OF MAXIMUM
             LIKELIHOOD ESTIMATES
                                    145

-------
                                 Errata
Page 8, line 20:  s,3 should be i±



Page 10, line 2:  z1 should be 13



Page 16, Table 4:  Column head "vaiable" should be "variable"



App. B, page 3, line 7:  y should be ^



App. B, page 4, line 9:  "m.l.e" should be m.l.e."




App. B, page 5, line 1:  "Yi = 1, 2, ...7" should read "Yl, i = 1, 2, ...7,"
                                      146

-------
                                 INTRODUCTION
During the course of the National Streams Survey over 1300 water samples
were collected from 450 streams in the mid-Atlantic and Southeast
regions.  These samples were sent to one of two laboratories contracted
for the survey, New York State Department of Health (NYSDOH) or Global,
for analysis of 24 critical chemical constituents (Table 1).  In addition,
134 "performance audit" water samples were sent to the laboratories to aid
in assessing the quality of the data they produced.  We analyzed this per-
formance audit data set to assess and estimate the relative interlabora-
tory bias of the measurements.

Relative interlaboratory bias can be defined as the mean difference in
measurement by two laboratories of identical water samples.  Different
laboratory facilities, personnel, and instrumentation provide many oppor-
tunities for inconsistent treatment of the samples, and hence for the
introduction of some interlaboratory bias.  This inter!aboratory bias may
then confound the analysis of the data.  Differences between two streams
ascribed to properties of the water may actually be the result of measure-
ment bias.  Apparent regional differences in stream water quality may
actually be due to the fact that streams from different regions were
analyzed by different laboratories.  For these reasons it is important to
assess and, if possible, correct for any interlaboratory bias within the
survey.
                                  147

-------
TABLE 1.  Variables measured in
National Streams Survey.
Calcium                      CA
Magnesium                    MG
Potassium                    K
Sodium                       NA
Manganese                    MN
Iron                         FE
Aluminum (extractable)       ALEX
Chloride                     CL
Sulfate                      S04
Nitrate                      N03
Silica                       SI02
Fluoride (total)             FTL
Dissolved Organic Carbon     DOC
Ammonium                     NH4
pH (equilibrated)            PHEQ
pH (alkalinity)              PHAL
pH (acidity)                 PHAC
Acidity                      ACCO
Alkalinity (ueq/1)           ALKA
Conductivity (yS/cm)         COND
Dissolved Inorganic
   Carbon (equilibrated)     DICE
Dissolved Inorganic
   Carbon (initial)          DICI
Phosphorus (total)           PTL
Aluminum (total)             ALTL
                 148

-------
                             STATISTICAL METHODS
Interlaboratory bias can be estimated 1f we have similar  water samples
measured by both of the labs,  this 1s exactly the case for the perform-
ance audit data 1n the National Streams Survey (NSS).  The performance
audit data were grouped into eight types of samples; of these, seven were
analyzed by both NYSDOH and Global.
Audit
Group
FL-14
LL-14
FL-15
LL-15
FN-6
LN-6
FN-8
LN-8

Number
Global
5
9
0
4
10
5
11
4
48
of Samples
NYSDOH
4
5
5
24
17
7
16
8
86
These audit groups are distinguished by the source of the samp1es~L14 and
L15 are  low concentration synthetic audit samples (produced by a contract
laboratory), and N6 and N8 are natural audit samples (Bagley Lake, Wash-
ington,  and Big Moose Lake, New York, respectively).  The groups are fur-
ther distinguished by the sample processing protocol used—the F prefix
Indicates samples preprocessed at the NSS field lab, and the L prefix
indicates samples preprocessed at the contract laboratory.

The samples within each of the natural audit groups were produced by sub-
dividing a single large, homogeneous, and presumably chemically stable
water sample into several two-liter aliquots.  The synthetic audit samples
                                    149

-------
were produced from stock solutions using a recipe designed to give concen-
trations close to the limits of quantitation for most analytes.  Measure-
ments will be considered as repeated measures of the same solution.

Ideally, the mean measurements by Global and NYSDOH should approximately
agree for each of the seven data groups.  Consistent or large deviations
between these pairs of means would be an indication of Interlaboratory
Mas.  These seven pairs of means, for each of the 24 parameters measured
in the survey, are summarized in the scatterplots in Appendix A.  Each of
the scatterplots includes the line of identity (of slope 1 and Intercept
0) about which we would expect the data points to be clustered if there
were no interlaboratory bias.  The scatterplots will be discussed in more
detail in Section 5.  Deviations from the line of Identity may be the
result of random error, or they may be the result of bias.

In order to quantify these deviations from the line of identity and thus
be able to test hypotheses about the deviations, we need models that
describe these deviations as a function of the concentrations of the
analytes.  Since there are only seven data points with which to estimate
these functional relationships, we considered only the simplest linear
functions.  Fitting more complex functions would risk overfittlng the
data, and would be appropriate only 1f we had previous knowledge of the
functional relationships to expect, as for example, if a more complex
relationship had been suggested by data on the laboratory instrumentation
1n an earlier study.  The linearity assumption is, however, a practical
alternative.  It is often appropriate in statistical applications, since
functions are often approximately linear when considered over a limited
range.  Also, it may be that many of the factors that lead to measurement
error, such as sample contamination or errors in instrument calibration,
result in biases that are linear functions of concentration.

We considered four linear models:  (1) no bias, (2) constant bias, (3)
bias proportional to concentration, and (4) the linear model with a con-
stant term and a proportionality term.  These models are described in
                                  150

-------
detail 1n Appendix B.  They translate into the  following  functional  rela-
tionships between the expected values of measurements  by  Global  and
NYSDOH:

     (1) NYSDOH measurements = Global measurements
     (2) NYSDOH measurements = Global measurements + a
     (3) NYSDOH measurements = (1 + e) (Global  measurements)
     (4) NYSDOH measurements = (1 + e) (Global  measurements) + a

where a represents the constant term and B represents the proportionality
term.

The positions of Global and NYSDOH in these relationships is completely
arbitrary and is by no means intended to suggest that one lab has per-
formed better than the other.  The roles of NYSDOH and Global could be
switched in the following analyses and none of the conclusions would be
changed.

If the audit samples are representative of the water samples encountered
in the survey, these relationships can be used to adjust the NSS data base
and correct for interlaboratory bias.  Our goal is then to  estimate a
and/or B (for model  numbers 2, 3, and 4) and provide statistical proce-
dures for comparing  the four models.

Assuming that the audit samples are  measured with independent normal
errors, this can be  treated as a maximum  likelihood estimation  problem.*
The maximum likelihood estimation procedure identifies for  each model the
parameters that are  most likely to have produced  the observed audit
 * Permutt,  T., M.  Moezzi,  and  S.  C.  Grosser.   1986.   "Relative  Inter-
   laboratory  Bias  in  the Western  Lake  Survey."   Systems Applications,
   Inc.,  San Rafael, California (SYSAPP-86/173).
                                  151

-------
data.  The maximum likelihood estimates (m.l.e.) can then be used to test
various hypotheses about the four models.  The derivation of the m.l.e. is
described in Appendix B.  The hypotheses tests are as follows.

First, we evaluate the probability density functions corresponding to each
of the models at their maximum likelihood estimates.  Call these values
Lj, 1-2, L3, and L^.

Second, we define three statistics a., i», and !u as:
                                          L.
                               11 = -2 In -j-3-
                                          L4
                               a, = -2 In -pi
                                i         4
                                          L.
                               t, = -2 In -r-5-
                                J         L3
Finally, we note that these statistics have approximately chi-square dis-
tributions (ap and 13 are chi-square with one degree of freedom, t,  with
two degrees of freedom) under the appropriate hypotheses, and so provide
test statistics for these hypotheses.

For example, 13 can be used to compare model 4  (o f 0 and & j* 0) with
model 3 (o = 0).  Under the assumption that a actually 1s equal to  0,  n3
1s distributed as chi-square.  If in fact a is  not equal to 0,  then
observed values of t^ would be tend to be larger than we would  expect
under the hypothesis.   If n3 is  large enough  (e.g., greater than 3.85  at
the 95 percent confidence level) we conclude that a -f 0.  Otherwise we can
assume that the simpler model with a = 0  1s adequate to explain the func-
tional relationship of  Global and NYSDOH measurements.   In  an analogous
fashion, a-, and «,- can  be used to test the models with a  =  6  =  0 and  6 =  0
against the general linear model.

We can test for our basic assumption of  linearity in a similar  fashion.
If we define a general  bias model with seven  distinct bias  terms to
                                    152

-------
describe the bias among our seven pairs of samples  (see  Appendix B for the
derivation of m.l.e.), and we define L5 as this  bias  model's density func-
tion evaluated at its m.l.e. values, then,

                              *4 = -2 log ^

1s distributed as a chi-square with five degrees of freedom under the
hypothesis that bias is indeed linear.  High values of s,4 (greater than
10.07 at the 95 percent confidence level) cause  rejection of the linear
bias model.

The observed values of the statistics 14 through 14 and an Interpretation
of their meaning relative to our problem are given in the following sec-
tion.
                                    153

-------
                                3   RESULTS
We tested the following four hypotheses about the  functional  relationship
of interlaboratory bias to concentration.   (Observed  test statistics  a^
through a^ corresponding to these hypotheses are given,  for each of the 24
analytes in the study, in Table 2.)

To choose among the linear functions considered, we test:

(1)  H0:  no bias (i.e., a = s = 0)  (versus HA:  a y  0,  & j 0);  this
     hypothesis is accepted at the 95 percent confidence level  if ij  <
     5.99.

(2)  H0:  bias is constant (i.e., e = 0) (versus HA:   a  f 0, s  j 0);  this
     hypothesis is accepted at the 95 percent confidence level  if s,2  <
     3.85.

(3)  H0:  bias is proportional to concentration (I.e., a = 0) (versus
     HA:  a f 0, 6 t 0); this hypothesis is accepted  at  the 95  percent
     confidence level if  10.07.

Using the chi square scores in Table 2 as a guide, we can choose from
among the linear models considered.

Based on iy the no bias model is acceptable for CL,  S04, N03,  DICE,  Did.
and PTL.  Of the remaining variables, the s = 0 model is acceptable for
                                    154

-------
TABLE 2.  Chi square statistics.
Variable 4^ ^ a3 "4
CA
MG
K
NA
MN
FE
ALEX
CL
S04
N03
SI04
FTL
DOC
NH4
PHEQ
PHAL
PHAC
ACCO
ALKA
CONO
DICE
DICI
PTL
ALTL
57.444
18.781
18.223
12.316
46.279
17.288
25.273
1.150
2.348
1.097
193.560
76.548
40.050
8.265
17.904
28.428
36.253
93.509
40.402
379.168
2.791
0.832
2.201
12.778
46.794
0.415
0.001
9.187
60.129
15.031
7.094
1.146
2.162
1.075
67.310
21.332
5.901
3.401
0.285
17.114
21.804
26.108
24.106
73.498
0.603
0.295
0.412
0.490
10.211
6.188
1.441
1.197
0.629
4.636
11.161
1.023
1.616
0.684
45.345
48.723
7.315
0.029
1.586
13.282
17.953
50.790
1.555
90.049
2.733
0.018
2.138
5.481
3.591
3.503
3.853
14.015
7.040
1.714
6.605
4.830
10.416
2.426
11.227
5.496
7.719
6.368
7.629
36.951
31.901
19.497
1.461
20.262
6.400
37.994
13.396
5.271
                    155

-------
MG, K, NH4, PHEQ and ALTL (see *2),  while the a = 0 model  1s  acceptable
for K, NA, MN, NH4, PHEQ, and ALKA (see i±).   Both the s = 0  and the a = 0
models are acceptable for K, NH4, and PHEQ.   The final choice of models
for these three variables can rest on considerations of which form of
bias, constant or proportional to concentrations, is more likely to have
been introduced during the respective measurement procedures.  The maximum
likelihood estimates of a and e for the chosen models are listed in Table
3.

We use the statistic n4 to test hypothesis (4).  Rejection of hypothesis
(4), indicated in Table 3 by an asterisk, means that our basic assumption
of linearity is in doubt.  In practical terms this means that we are less
certain of the functional relationships suggested by hypotheses 1 through
3 and hence of the associated transformations in Table 3.  The implica-
tions of this are discussed in detail in Section 5.
                                    156

-------
TABLE 3.  Estimated transformation coefficients.
Variable
CA
MG
Ka
Ka
NA
MN
FE
ALEX
CL
S04
N03
SI02
FTL
DOC
NH4a
NH4a
PHEQa
PHEQa
PHAL
PHAC
ACCO
ALKA
COND
DICE
DICI
PTL
ALTL
a
-0.042
0.0070
0.0071
0.0
0.0
0.0
-0.0205
-0.0038
0.0
0.0
0.0
-0.18
0.0081
0.17
0.005
0.0
0.071
0.0
0.661
1.13
37.7
0.0
5.91
0.0
0.0
0.0
0.014
6 Nonlinear5
0.096
0.0
0.0
0.030
-0.02.0
0.187
1.053
-0.122
0.0
0.0
0.0
-0.083
-0.112
0.064
0.0
0.064
0.0
0.012
-0.116
-0.208
-0.394
0.106
-0.174
0.0
0.0
0.0
0.0
.._
~
__
~
*
~
—
~
—
*
~
*
—
—
—
—
—
~
*
*
*
__
*
~
*
*
—
 a  Two models are acceptable for these variables.

 b  An asterisk Indicates rejection of the linearity
    hypothesis at the 95 percent confidence level.
                            157

-------
                         4   APPLYING THE RESULTS


There are several ways in which the models in Table 3 can be applied.
Where bias is identified, the measurements by the two labs need to be in
effect calibrated so that they are expressed on the same scale.  Hence if
one has reason to believe that NYSDOH's measurements are more accurate
than Global's, then the Global measurements would be transformed to the
NYSDOH measurement scale:

                  GlobalNEW = (1 + s) (GlobalOLD) + a  .

Conversely, if one believes that the Global measurements are more
accurate, the NYSDOH measurements would be transformed to the Global mea-
surement scale:

                  NYSDOHNEW = (NYSDOHOLD - «)/(! + B]  .

In the absence of knowledge of absolute bias, a reasonable alternative
would be to split the difference between the measurement scales for the
two labs and transform the measurements by each lab to this new scale:

                  GlobalNEW = (1 + f) (GlobalOLD) + f  ,


                  NYSDOHNEW=(NYSDOHOLD-f)/(l+f)  .
                                  158

-------
                             5   CONCLUSIONS
DISCUSSION

The coefficients listed 1n Table 3 can be used to transform the measure-
ment data from one laboratory to the measurement scale of the other when
Interlaboratory bias 1s Identified.  Even given non-zero coefficients we
cannot immediately assume that the NSS data would be Improved by applying
the transformations.  Instead, we must decide whether the expected
Improvements 1n relative accuracy justify the loss 1n precision that will
result from applying these transformations.  Loss of precision results
from the fact that the values of a and 6 in Table 3 are estimates only;
the uncertainty of the a and B estimates (I.e., the uncertainty 1n the
estimated functional relationship) would produce uncertainty 1n the trans-
formed NSS data.

The figures 1n Appendix A  Illustrate the potential gains and losses from
calibration.  The distance from the best fitting line to the line of
Identity  at a given  concentration  1s our best estimate of  the  bias at that
concentration.  Other things  being equal,  the larger this  estimated bias,
the more  important the correction  transformation  (I.e.,  calibration)
becomes.  On the other hand,  the  curved  band  represents  the  uncertainty of
this  estimate.  If,  for example,  this  band straddles  the line  of  identity
at a  relevant concentration,  there is  a  significant doubt  that the pro-
posed correction  is  even  1n  the right  direction.  This  happens for potas-
sium  (K)  for example.   In less  extreme cases,  the direction  of the  bias is
clear, e.g., extractable  aluminum (ALEX),  but there  is  still  substantial
uncertainty  about  its  magnitude.   In  such  cases calibration  therefore
 Introduces a  substantial  uncertainty  in  exchange for  eliminating  a likely
bias.
                                     159

-------
The figures show 1n each case the line of best fit.  In many cases,
Identified 1n Table 3, the fit 1s not significantly worse for a line con-
strained to have zero intercept or unit slope.  In these cases, if cali-
bration seems desirable, the simpler forms are probably to be preferred.

An additional factor to consider before applying the transformations 1n
Table 2 1s the representativeness of the range of the audit data as com-
pared to the range of the NSS data.  As an illustration of the importance
of this consider the calcium audit data.  The range from zero to the
highest calcium audit group mean (1.95 mg/1) contains only 23.0 percent of
the NSS calcium measurements.  This 1s the range for which we can be most
certain that the audit data has represented the NSS data, and hence most
certain of the validity of the estimated functional relationship.  Apply-
ing this transformation beyond the range of the audit data relies heavily
on the assumption of linearity, especially when the NSS data extend well
beyond the range of the audit data.  For calcium, for example, the range
of the audit data represents only 2.0 percent of the entire range of the
NSS data.  Even If we eliminate the top 5 percent, and thus any outliers,
the audit data still only represents 4.6 percent of lowest 95 percent of
the NSS data.  This 1s dramatically Illustrated in Figure 1, which shows
the estimated transformation and 95 percent confidence bounds extrapolated
over approximately 95 percent of the range of the NSS calcium data.  (The
audit data means are clustered 1n the bottom left corner of the graph.)

The measures of representativeness outlined above are provided, for each
of the variables in the study, in Table 4.

Figure 1 emphasizes the Importance of the linearity assumption when the
audit data do not cover the range of the NSS data.  Even given this
assumption we see how the 95 percent confidence bounds widen substantially
when we extrapolate the estimated transformation beyond the range of the
audit data.  If this assumption is not true, these bounds should be even
wider than they are now represented.  If the relationship is nearly
                                   160

-------
                                             50.0
0>
                                                                                          -Best linear transformation
                                                                                           95* confidence bounds
                                                             10.0
  20.0          30.0
GLOBAL MEASUREMENT
                                                                                                      40.0
                                                                                                                   50.0
                                      FIGURE 1     Comparison of means  of GLOBAL vs. NYSDOH measurements for CAll.
                                              Uncertainty shown by bars (standard deviation) and ellipses (standard error).

-------
TABLE 4.  Representativeness of audit data.
Coverage by Audit Data
Vaiable
CA
MG
K
NA
MN
FE
ALEX
CL
S04
N03
SI02
FTL
DOC
NH4
PHEQ
PHAL
PHAC
ACCO
ALKA
COND
DICE
DICI
PTL
ALT1
a The %
audit
h „._.,..„
% of Data3
23.0
5.2
14.5
57.3
81.6
56.5
94.0
1.8
38.4
58.2
83.1
86.8
81.4
96.1
39.7
53.7
58.5
67.0
31.8
12.0
44.3
38.5
91.7
76.2
% of Truncated
Rangeb %
4.6
3.2
14.8
23.0
27.8
7.3
82.3
1.9
12.0
6.7
67.1
64.5
35.8
146.0
54.9d
56. 5d
58. 6d
33.3
6.1d
6.8
6.4
6.7
62.6
26.1
of Rangec
2.0
1.2
4.9
1.5
0.9
0.2
2.7
0.1
1.9
0.9
29.1
14.3
2.4
5.9
38. 2e
32.9e
34.0e
3.4
1.5e
2.0
2.2
1.8
1.9
0.8
of the survey data $ the maximum
mean.
maximum audit mean ^ ,„„ „
fSthquantHe of the
maximum audit mean
maximum survey value
. maximum audit mean -
§5"th quant 11 e -
maximum audit mean
survey data '
* 100 %
minimum audit mean
5th quant 11 e
- minimum audit mei
V
* 100 %
IIL_ * inn
                            162

-------
linear, but only wavers slightly about  this  straight  line,  then  the  con-
fidence bounds are approximately correct.   If,  however,  the relationship
could be a quadratic or more complicated function,  then  these widths are
significantly underestimated particularly when we must extrapolate over a
relatively large range, as with the calcium data.
SUMMARY

The decision to transform the NSS data depends then on a careful weighing
of the expected improvements in accuracy against the possible losses in
precision  that can result.  This information  is summarized  in the graphs
1n Appendix A.  The width of the confidence bounds  in these graphs however
depends  heavily on the assumption of  the linear relationship of  bias to
concentration  across  the range  of NSS concentrations.  Hence, this assump-
tion should be carefully considered for  those variables  for which  the
 linearity hypothesis  was rejected  (Indicated  by  an asterisk in  Table  2)
 and for those variables for which  there  1s no audit data to test this
 assumption for the entire  range of  the NSS data (as indicated  by Table 4).

 These are decisions that should be made by someone very familiar with the
 laboratory analytical procedures Involved and with the ways bias can be
 Introduced.  However, we can make the following specific recommendations:

      (1)   Measurements  of those analytes  for which the no  bias model was
            recommended,  Cl. S04, N03, DICI, and PTL cannot  be improved by
            transformation.

       (2)   If  policy  decisions  are  based on a measurement  range  of the data
            that 1s within the range of the performance audit data, for
            instance  in the  low  concentrations for  most of  the variables  in
            the study, then  transformations may be  in order.

       (3)  Transformations  that Involve  extrapolation over much of the
            range of the data should only be carried out after careful  con-
            sideration of the linearity assumption.
                                        163

-------
                    Appendix A



SCATTERPLOTS OF THE MEANS OF PERFORMANCE  AUDIT  DATA
                 164

-------
  0.30
35
o
a
CO
  0.20 -
                                                   Identity line

                                                   Best linear transformation

                                                   95J confidence bounds
      fr.OO
0.10                   0.20

   GLOBAL MEASUREMENT
                                                                         0.30
       Comparison of means of GLOBAL vs. NYSDOH measurements for ALEX11.

    Uncertainty shown by bars (standard deviation) and ellipses (standard error).
                                     165

-------
   0.40
   0.30 -
Gfi
  0.20 -
SB
o
Q
  O.Oj
                                               -Best linear transformation

                                                95X confidence bounds
                                      0.20


                              GLOBAL MEASUREMENT
  0.10 -
0.40
      Conolparison of means of GLOBAL vs. NYSDOH measurements for ALTL11.

   Uncertainty shown by bars (standard deviation) and ellipses (standard error).
                                  166

-------
    200
§
                                                Identity line
                                                Best linear transformation
                                                951 confidence bounds
     -50
                                             100
                              GLOBAL MEASUREMENT
150
             800
      Comparison of means of GLOBAL vs. NYSDOH measurements for ALKA11.
  Uncertainty shown by bars (standard deviation) and ellipses (standard error).
                                  167

-------
100.0
                                           -Best linear transformation
                                             951 confidence bounds
                80.0
  40.0          60.0
GLOBAL MEASUREMENT
                                                       00.0
                                                                    100.0
   Comparison of means of GLOBAL vs. NYSDOH measurements for ACC011.
Uncertainty shown by bars (standard deviation) and ellipses (standard error).
                             168

-------
4.0
                                          -Best linear transformation
                                            93% confidence bounds
 °'%0      0.5      1.0      1.5      2.0      2.5      3.0     3.5      4.0
                           GLOBAL MEASUREMENT
    Comparison of means of GLOBAL vs. NYSDOH measurements for CAli.
Uncertainty shown by bars (standard deviation) and ellipses (standard error).
                               169

-------
u
K
§
                                                 -Best linear transformation

                                                  951 confidence bounds
                            0.2         0.3         0.4

                              GLOBAL MEASUREMENT
0.5
           0.6
       Comparison of means of GLOBAL vs. NYSDOH measurements for CLU.

   Uncertainty shown by bars (standard  deviation) and ellipses (standard error).
                                170

-------
  30.0
  25.0 -
  20.0 -
M
ae
8
S
o
Q
                                               -Best linear transformation
                                                 9SS confidence bounds
   15.0 -
   10.0  -
                 5.0
10.0       15.0       80.0
   GLOBAL MEASUREMENT
                                                                        30.0
       Comparison of means of GLOBAL vs. NYSDOH measurements for CONDI 1.
    Uncertainty shown by bars (standard deviation) and ellipses (standard error).
                                  171

-------
    2.0
    1.8
    i.a
    1.4
    1.2
    i.o
K
o
S   0.8
    0.8
    0.4
    0.2
                                                  - Identity line
                                                  -Best linear transformation
                                                  - 951 confidence bounds
    °'%.0     0.2     0.4     0.6    0.8     1.0    1.2    1.4    1.6    1.8    2.0
                               GLOBAL MEASUREMENT

       Comparison of means of GLOBAL vs. NYSDOH measurements for DICE11.
   Uncertainty shown by bars  (standard deviation) and ellipses (standard error).
                                   172

-------
                                 	-Best linear transformation
                                             95X confidence bounds
 H.O    0.3    0.4    0.6    0.8    1.0    1.3    1.4     1.6    1.8    3.0
                           GLOBAL MEASUREMENT
    Comparison of means of GLOBAL vs. NYSDOH measurements for DICI11.
Uncertainty shown by bars (standard deviation) and ellipses (standard error).
                               173

-------
                                  	Best linear transformation
                                              95J confidence bounds
                             2.0          3.0
                           GLOBAL MEASUREMENT
4.0
    Comparison of means of GLOBAL vs. NYSDOH measurements for DOCU.
Uncertainty shown by bars (standard deviation) and ellipses (standard error).
                               174

-------
   0.09
   0,08
   0.07
it
a  o.oe
3
   0.05
§0.04
  0.03
  0.08
  0.01
  0.0(
 Identity Hne
-Best linear transformation
 95S confidence bounds
                                                             J	I
      .00   0.01   0.02   0.03   0.04   0.05   0.06   0.07   0.08   0.09
                               GLOBAL MEASUREMENT
       Comparison of means of GLOBAL vs. NYSDOH measurements for FTL11.
   Uncertainty shown by bars (standard deviation) and ellipses (standard error).
                                 175

-------
  0.08
  0,09
M
S 0.04
w
OS
D

3
u


n

9 0.02
   0.00
-D.02
                                                    Identity line
                                        	Best linear transformation

                                        	 95S confidence bounds
                   0.00          0.02          0.04

                               GLOBAL MEASUREMENT
0.06
0.08
       Comparison of means of GLOBAL vs. NYSDOH measurements for FE11.

   Uncertainty shown by bars (standard deviation) and ellipses (standard error).
                                  176

-------
0.50
                                               Identity line
                                               Best linear transformation
                                               95X confidence bounds
   .10
0.20
                                    0.30
                           GLOBAL MEASUREMENT
                                  0.40
0.50
     Comparison of means of GLOBAL vs. NYSDOH measurements for Kll.
Uncertainty shown by bars (standard deviation) and ellipses (standard error).
                              177

-------
0.6
                                             -Best linear transformation
                                              95* confidence bounds
              0.1
0.2        0.3        0.4
   GLOBAL MEASUREMENT
    Comparison of means of GLOBAL vs. NYS00H measurements for MG11.
Uncertainty shown by bars (standard deviation) and ellipses (standard error).
                               178

-------
0.12
                                               -Best linear transformation
                                                95* confidence bounds
  •-11.02     0.00
0.02      0.04      0.08      0.08
      GLOBAL MEASUREMENT
                                                             0.10     0.12
     Comparison of means of GLOBAL vs.  NYSDOH measurements for MN11.
 Uncertainty shown by bars (standard deviation) and ellipses (standard error).
                                179

-------
                                             -Best linear transformation
                                              95J confidence bounds
                               1.5       3.0
                           GLOBAL MEASUREMENT
    Comparison of means of GLOBAL vs. NYSDOH measurements for NA11.
Uncertainty shown by bars (standard deviation) and ellipses (standard error).
                              180

-------
 0.20
 0.15 -
 0.10
 0.05
 0.00
-0.0
84.
                                                Identity line
                                                Best linear transformation
                                                95J confidence bounds
      05         0.00         0.05          0.10
                             GLOBAL MEASUREMENT
0.15
                                                                       0.80
     Comparison of means of GLOBAL vs. NYSDOH measurements for NH411.
  Uncertainty shown by bars (standard deviation) and ellipses (standard error).
                                 181

-------
    1.5
    1.0
3
(d
K
O
Q
s
   0.5
   0.0
  -0.5.
                                                   Identity line
                                                 -Best linear transformation
                                                   95% confidence bounds
     -0.5
0.0              0.5
        GLOBAL MEASUREMENT
1.0
1.5
      Comparison of means of GLOBAL vs. NYSDOH measurements for N0311.
   Uncertainty shown by bars (standard deviation) and ellipses (standard error).
                                  182

-------
 0.04
                                             -Best linear transformation
                                              95S confidence bounds
-0.01
     .01
0.00
  0.01          0.08
GLOBAL MEASUREMENT
                                                        0.03
                                                     0.04
     Comparison of means of GLOBAL vs. NYSDOH measurements for PTL11.
  Uncertainty shown by bars (standard deviation) and ellipses (standard error).
                                183

-------
w
a.
K
O
o
                                              -Best linear transformation

                                                95% confidence bounds
   4.0.
                                       8.0

                              GLOBAL MEASUREMENT
8.0
      Comparison of means of GLOBAL vs. NYSDOH measurements for PHEQ11.

   Uncertainty shown by bars (standard deviation) and ellipses (standard error).
                                  184

-------
w
K
D
CO
33
o
a
    7.5
    7.0
    6.5
    8.0
    5.5
    5.0
    4.5
      4.5
5.0
                              I
                                                    Identity line
                                        	--Best linear transformation

                                        	 95X confidence bounds
5.5         6.0         6.5


   GLOBAL MEASUREMENT
                                                                7.0
                                                                           7.5
      Comparison of means of GLOBAL vs. NYSDOH measurements for PHAL11.

   Uncertainty shown by bars (standard deviation) and ellipses (standard error).
                                   185

-------
u
w
a:
o
Q
                                                -Best linear transformation

                                                 951 confidence bounds
   5.5 -
   5.0 -
   4.5.
     41.5
                            5.5        6.0         6.5


                              GLOBAL MEASUREMENT
7.0
7.5
      Comparison of means of GLOBAL vs. NYSDOH measurements for PHAC11.

   Uncertainty shown by bars (standard deviation) and ellipses (standard error).
                                  186

-------
  12.0
  10.0 -
u
U
9S
o
Q
                                                    Identity line

                                                    Best linear transformation

                                                    95X confidence bounds
2.0
                             4.0         6.0        8.0

                               GLOBAL MEASUREMENT
                                                                          12.0
       Comparison of means of GLOBAL vs. NYSDOH measurements for SI0211.

    Uncertainty shown by bars (standard deviation) and ellipses (standard error).
                                  187

-------
                                               Identity line
                                               Best linear transformation
                                               951 confidence bounds
                      3-0       3.0       4.0       5.0
                           GLOBAL MEASUREMENT
6.0
          7.0
    Comparison of means of GLOBAL vs. NYSDOH measurements for S0411.
Uncertainty shown by bars (standard deviation) and ellipses (standard error).
                              188

-------
                                Appendix B
     MODELS CONSIDERED AND DERIVATION OF MAXIMUM LIKELIHOOD ESTIMATES


MODELS CONSIDERED

Formally stated, the four linear models and the general  bias model  out-
lined 1n Section 2 are as follows.

If
then the models are

                      (i)  x21>j

                      (2) X21tJ

                      (3) X
                       (5) X21J = p. + YI + EIJ;

where
 6.J. - N(0, o_j) are the errors  1n measurement by Global;
 •44 ~ N(°« °2i) are the errors 1n measurement by NYSDOH; X21-1 j
                                189

-------
1s the jth measurement by Global on the 1th group; X21 j 1s the jth mea-
surement by NYSDOH on the 1th group; and ^ 1s the long-run mean concen-
tration measured by Global.  The YI 1n the fifth model are the long-run
mean differences 1n concentration by each laboratory for each audit
group.  The measurement error terms « and e are Independent and normally
distributed, with variances that may vary from audit group to audit group
as well as from laboratory to laboratory.
DERIVATION OF MAXIMUM LIKELIHOOD ESTIMATES

As above, we use the assumption of random normal errors and assign the
parameters yi (1 = 1, 2, .... 7) to the means of Global's measurements of
each audit group.  We get the following density functions for the observa-
tions at each of the audit groups for the linear bias model.

For Global:
For NYSDOH:
where
     n21-l B aud1t subgroup sample size corresponding to Global
      2
     °21-1 = var^ances corresponding to Global

      n21  - audit subgroup sample size corresponding to NYSDOH
                                 190

-------
             variances corresponding to NYSDOH
The density function for the linear bias model  1s  .then:
           14
      1-4 ' II  f 1
       4        J
Maximum likelihood estimates (m.l.e.) for the linear bias model are
obtained by maximizing this equation as a function of the parameters
v, .... v7't oj, ..., o^4; o; and 8.   Or, more conveniently, we
could minimize -2 log L4.

Many computer algorithms are available for solving minimization problems
such as this.  A "grid search" type of algorithm, however, provides the
most consistently reliable answer.  A grid search Involves literally
evaluating the likelihood function at every possible value of the
parameter, or at least on a reasonably fine mesh over all the possible
values, and empirically determining which parameter values maximize the
likelihood function.  This approach has the additional advantage of pro-
viding the range of o and e that decides acceptance of the o = 6 = 0
hypothesis (hypothesis number 1 1n Section 3), I.e., the subset of all a
and & for which the (logged) general linear bias likelihood equation  1s
within one-half (5.99) of Its (logged) value at the m.l.e. solution.  This
range provides, by definition, the 95 percent confidence Intervals for the
best fit m.l.e.'s of linear Inter laboratory bias.

Our grid search algorithm takes advantage of the fact that y and a can be
defined Implicitly 1n terms of each other and by a and B at the m.l.e.
solution as follows:
                                  191

-------
       "1
                                .,  (i + flr
                                   °21/n21
     2
    °21-1
              21-1
        5  (X2i-l,
        J
                                       2i-l
"21
              21
                          _
                        - X2l     n2i-l
for 1 = 1, 2, .... 7.

This system of equations quickly converges to a single answer by  1tera-
tlvely recalculating the first 1n terms of the next two and  the last two
1n terms of the first.  These equations can be obtained by taking first
partial derivatives of -2 log L4, setting them equal to 0, and solving.

This 1s saying that for any given line drawn through the data there 1s
just one set of y and. a that maximizes the likelihood function.   By taking
advantage of this, the problem of maximizing L4 Is reduced to that of a
simple grid search over the possible range of a and e.

The m.l.e for Lj, l^, and L^ in Section 2 can be found by simply  replacing
a and/or B with zero as appropriate 1n each of the above equations.

For the general bias model the density functions for the Global audit
groups are unchanged, but for NYSDOH they become:
  21
                 _L\Z1 J-l/2
                         CXP
                                  192

-------
where y^ - 1, 2, ..., 7 are the seven bias terms used to account for the


deviations between the NYSDOH and Global means.  Maximum likelihood esti-


mates for this density are obtained simply as:




          y.) = observed Global means,




          YJ = observed difference between the Global arid NYSDOH means,
                                           "21-1
and
         ^O4 =  * *  \^O4  A ™ "94/  / "O4
          i\    S~,  * cl»J    tl      tl
                              193

-------
                                      GLOSSARY
absolute bias
acceptance
  criteria
accuracy
The  difference  between  a  measured  value  and  the  true  value.
(See "accuracy.")

The range in which the analytical measurement of a quality  assurance
or quality control sample is  expected to be;  measurements outside
that  range are considered suspect.

The closeness  of a measured value to the true  value  of an analyte.
For this report, accuracy  is calculated as:
acid-neutralizing
  capacity (ANC)
air equilibration
aliquot
among-batch
  precision
analyte

analytical
   laboratory
analytical
   laboratory
   duplicate

anion
where the X is the mean of all measured values and R = the theoretical
or index value.

Total acid-combining  capacity of  a  water  sample  determined  by
titration  with  a  strong acid.   Acid-neutralizing  capacity  includes
alkalinity  (carbonate  species)  as well  as other basic  species  (e.g.,
borates, dissociated organic acids, alumino-hydroxy complexes).

The  process of  bringing a sample aliquot to equilibrium with standard
air (300  ppm CO2) before analysis;  used with some pH and dissolved
inorganic carbon measurements.

Fraction of  a  sample prepared for  the analysis  of particular con-
stituents; sent  in a separate container to the analytical laboratory.

The  estimate of precision that  includes effects of different  laboratories
and  day-to-day  difference within a  single laboratory,  calculated from
audit sample data.

A chemical  species that is measured  in a water sample.

In this  report, a laboratory under contract with the U.S.  Environmental
Protection  Agency  to  analyze water  samples  shipped  from  the
processing  laboratory.

Aliquot   of   a   sample  that  is  split  in the  analytical   laboratory.
The  aliquots are analyzed in the same batch.
A negatively charged ion.
                                               194

-------
anion-cation
  balance
In an  electrically  neutral solution  such as water, the total  charge
of positive  ions  (cations)  equals the total  charge of negative ions
(anions).    In  this  report,  anion-cation  balance  is  expressed  as
percent ion balance difference  and Is calculated  as follows:
 anion deficit
 ASTM Type I
   reagent-grade
   water
 audit sample
 base site
 base-neutralizing
   capacity (BNC)
 batch
                 Z anions - Z cations + ANC

                 I anions + Z cations  + ANC +  2[H+]
                                                                                  100
                       where:
 batch ID
         I anions  = [CP] + [F'] + [NO3-] +  [SO42-],

         I cations  = [Na+] +  [K+] + [Ca2+] +  [Mg2+] + [NH4+],

                   ANC ^    alkalinity  (the  ANC value  is  included
                              in the  calculation  to  account  for the
                              presence  of unmeasured ions  such as
                              organic ions), and

                    [H+]  «    (10'PH) x 106 /jeq/L

The concentration  (in microequivalents per liter) of measured anions
less the measured cations.

Deionized water that meets American Society for Testing and Materials
(ASTM)  specifications for Type  I  reagent-grade  water (ASTM,  1984)
and that has  a  measured conductance of less than  1 ^S/cm at 25
°C. This water is used to prepare blank samples  and reagents.

In this survey, a  standardized water sample submitted to an analytical
laboratory for  the  purpose of  checking overall performance in sample
analysis. Natural audit samples in the NSS-I were lake water;  synthetic
audit samples were prepared by diluting concentrates of known chemical
composition in ASTM Type  I reagent-grade water.

A  location  providing  support  for sampling  personnel  during  field
sampling operations.

Total  base-combining capacity of  a water sample  due to dissolved
CC>2, hydronium, and hydroxide;  determined by titration with  a strong
base,

A group of samples processed  and  analyzed together. A batch consists
of all samples (including quality assurance and quality control  samples)
that  are assigned a unique batch identification  number  and that are
processed and sent to one analytical laboratory in  one day.  In the
 NSS-I, a batch did not exceed 40 samples.

The numeric identifier for each batch.
                                               195

-------
 bias

 blank sample
 calculated
   conductance

 calibration blank
   sample
calibration curve



carryover



cation

circumneutral

closed system
comparability


completeness
component (of a
   sampling  system)
 The systematic difference between values or sets of values.

 A  sample  of  ASTM Type  I  reagent-grade  water  analyzed  as  a
 quality assurance  or quality control  sample  during  the NSS-I (see
 calibration,  reagent, processing  laboratory, and field blanks).

 The sum (as microsiemens per centimeter) of the theoretical specific
 conductances of all measured ions in  a sample.

 A sample of  ASTM Type I reagent-grade water  defined  as  a 0 mg/L
 standard used In standardizing  or checking the calibration of analytical
 instruments; also used to determine instrument detection limits.

 The linear  regression  of  the  analytical  instrument  response  to  a
 set  of calibration  standards (varying  in concentrations)  from  which
 the linear dynamic  range  is determined.

 An artifact of the analyte carried from a  sample  of high concentration
 to a subsequent sample or samples as a result  of incomplete rinsing
 of an  instrument or apparatus.

 A positively charged ion.

 Close  to neutrality  in pH (near pH 7).

 Method of  measurement  in which a  water  sample  is collected and
 analyzed  for  pH and dissolved  inorganic carbon without  exposure
 to  the atmosphere.   These  samples  were collected  in   syringes
 directly from  the  sampling  apparatus  and  were  analyzed  in the
 processing laboratory.

 A measure  of  data quality that allows the  similarity  within  and
 among data sets to be established confidently.

 A measure  of data quality  that  is the quantity of  acceptable  data
 actually collected  relative  to the total  quantity  that  was  expected
 to be collected.

 For this report, any of the  sets of procedures used to get a sample
 from  the  stream  to analysis.    Major  components  include sample
collection, sample processing, and  sample analysis.  Other components
 include sample  transport,  sample shipment,  and  data  reporting.
Together, these components are  the system.
                                              196

-------
conductance  balance
confidence interval
   (95% and 99%)
confidence limits
control limits
 Cubitainer
 data base
 data base
   audit
 data package
 data qualifier
 data quality
   objectives
A comparison of the  measured  conductance  of  a water  sample (in
microsiemens  per  centimeter)  to the  equivalent  conductances (in
microsiemens  per centimeter)  of each  ion measured  in  that  water
sample at infinite dilution.  In  this report,  conductance  balance is
expressed as percent conductance difference and is
calculated as follows:

  (calculated conductance - measured conductance\
  	1100
               measured conductance            /
                       The ions used to  calculate conductance are  Ca,  CI", CO32',
                       HC03-, K, Mg, Na, NO3-, OH', and SO42-.
                                                               H+,
The  range of values,  calculated from  an estimate of the mean and
standard  deviation, between  the confidence  limits.    This  interval
has  a  high  probability  (a  95  or 99  percent level  of confidence)
of containing  the true  population value.

Two  statistically  derived values  or  points,  one  below  and one
above  a  statistic,  that provide a  given degree  of  confidence that
a population parameter falls between them.

Two values between  which  the analytical measurement  of a quality
assurance or  quality control  sample is expected to be; measurements
outside these limits are suspect.

A  3.8-L container made  of  semirigid polyethylene used  to transport
field  samples (routine, duplicate,  blank)  from  the  stream  site  to
the processing laboratory.

All  computerized   results of  the  survey,  which include  the  raw,
verified,  validated,  and  enhanced  data  sets  as well  as  back-up
and historical data sets.

An accounting of  the data  and  of  the data  changes  in the data
base;  includes changes made within  a data set and  among  all data
sets.

A  report,  generated  by  an analytical  laboratory for  each batch of
samples  analyzed, that  includes  analytical results,   acid-neutralizing
capacity  titration  data,  ion  chromatography  specifications,  analysis
dates, calibration and  reagent blank data, quality control  check sample
results, and analytical laboratory duplicate results.

Annotation  applied to  a  field  or  analytical  measurement  related
to  possible effects of  the  quality of  the  datum.    (See  definitions
for "flags" and "tags".)

Accuracy,  detectability,  and precision  limits  established  before  a
sampling  effort.    Also  include  comparability,  completeness,  and
representativeness.
                                               197

-------
Data Set 1

Data Set 2


Data Set 3


Data Set 4


detectability
detection  limit
  quality control
  check  sample

dissolved
  inorganic carbon
  (DIG)

dissolved organic
  carbon  (DOC)
enhanced data  set
equivalent


exception


exception program
extractable
  aluminum
field audit sample
Set of files containing raw data.  (See definition for "raw data set").

Set  of files  containing  verified  data.    (See  definition for "verified
data  set.")

Set of  files  containing validated data.   (See definition for "validated
data  set.")

Set  of files  containing  final,  enhanced  stream  data. (See definition
for "enhanced data set.")

The capability of  an instrument or method to determine a measured
value for an  analyte above either zero or background levels.

A quality  control  check sample  that  has  a  specified  theoretical
concentration and that   is designed  to check  instrument  calibration
at the low end of the linear dynamic range.

A measure of the dissolved carbon dioxide, carbonic acid, bicarbonate
and  carbonate  anions that constitute  the major  part of  ANC  in  a
stream.

In a  water sample, the  organic  fraction  of carbon that is dissolved
or unalterable (for  this  report, the  fraction that will pass  a  filter
of 0.45-ju m pore size).

Data  Set  4.   Missing  values or  errors  in  the validated data  set
are replaced  by substitution  values; duplicate  values are  averaged;
negative values (except  for ANC and BNC)  are set to zero.  Values
for field blank, field duplicate,  and performance  audit  samples  are
not included  in this data  set.

Unit  of ionic charge;  the quantity  of  a substance that either  gains
or loses one mole of protons or electrons.

An analytical result  that does not  meet  the  expected QA or  QC
criteria and for which a data flag is generated.

A computer program  in AQUARIUS-II that identifies or  flags analytical
results classified as exceptions.

Operationally  defined  aluminum  fraction  that  Is extracted  by  the
procedure  used during the  NSS-I;  this measurement  is intended  to
provide an indication  of  the concentration of the aluminum species
that may be  available  in a form toxic to fish.

A standardized  water  sample  submitted  to  field  laboratories  to
check  overall performance in  sample  analysis  by field laboratories
and by analytical laboratories.  Field natural audit  samples  were lake
water;  field   synthetic  audit   samples  were  prepared  by  diluting
concentrates of known chemical composition into ASTM  Type I reagent-
grade water.
                                               198

-------
field blank sample
field duplicate
  sample
field duplicate
  pair
field natural
   audit sample

field synthetic
   audit sample

flag
 Gran analysis
 holding  time
 imprecision
 index value
 in situ
 initial dissolved
   inorganic carbon
   (DIG)
A sample prepared at the processing laboratory consisting  of  ASTM
Type  I  reagent-grade  water  and  transported to the  stream site  by
the  field sampling crews.  At the stream site, the blank was processed
through the sampling  apparatus.   These samples were analyzed  at
the  processing  laboratory (except for pH  and  DIG)  and  analytical
laboratories and were  employed in the calculation of system decision
and system detection limits and instrument  detection limits.

The second sample  of stream water collected by the  sampling crew
at  the  same  location and  depth  at the  stream  site  immediately
after  the routine sample, in accordance with standardized protocols.

A routine stream water sample and a second sample (field duplicate
sample)  collected from the  same stream,  by  the  same  sampling
crew, during the same visit, and according to the same procedure.

See field audit sample.
See field audit sample.
 Qualifier of a  data point  that  did  not  meet established acceptance
 criteria or  that  was  unusual.    Flags were  assigned during  the
 verification and validation procedures.

 A  mathematical  procedure  used  to  identify the  equivalence  point
 or  points  of  the titration of a  carbonate system  and  subsequently
 for ANC and BNC of that system (Hillman et al, 1987).

 (1)  In the processing laboratory, the elapsed time between sample
 collection and  sample preservation.  (2)   In  the analytical laboratory,
 the elapsed time between sample processing in the processing laboratory
 and final sample analysis or reanalysis.

 For  a  particular   analyte,  the  degree  of  irreproducibility  of  or
 deviation of a measurement from the average of a set of measurements;
 the variation about the mean.

 A  mean  value for  measurements of  a  performance  audit sample
 made at either  the support laboratory (synthetic  audit  samples) or
 by a number of analytical laboratories (natural audit samples).

 For this  survey, any measurements  made  within  the water column
 of a stream.

 A  measurement of  dissolved  inorganic carbon made on  an  aliquot
 immediately before  it is titrated for ANC.
                                               199

-------
 instrument
   detection limit
 interlaboratory
   bias
 For  each  chemical  variable,  a  value  calculated  from  laboratory
 calibration blank,  reagent blank, or field blank samples that indicates
 the minimum  concentration  reliably detectable by the instrument(s)
 used; calculated  as  three  times  the  standard  deviation  of 10  or
 more  nonconsecutive  (i.e., from  different  calibrations)  blank  sample
 analyses.

 Systematic differences in performance between laboratories estimated
 from analysis  of  the  same  type  of samples.
 ionic strength
laboratory blank
   sample

laboratory
   duplicate sample

linear dynamic
   range

management team
matrix

matrix spike
method-level
  precision
nephelometric
  turbidity unit
  (NTU)

on-site evaluation
open  system
 A  measure  of  the  interionic  effect  resulting  from  the  electrical
 attraction  and repulsion  between different ions.  In very dilute solu-
 tions,  ions  behave  independently,  and  the  ionic  strength can be
 recalculated  from  the  measured concentrations of  anions and cations
 present in the  solution.

 A sample of ASTM Type  I reagent-grade water prepared and analyzed
 by analytical laboratories. (See calibration blank, reagent blank.)

 Sample aliquot that is split and prepared  at the analytical laboratories
 and that is analyzed once in a batch.

 The range of  analyte concentration  for  which the calibration curve
 is best fitted by a straight line.

 EPA personnel  responsible  for  overseeing the NSS-I sampling  and
 QA design and the subsequent interpretation of  stream data results.

 The physical and chemical composition of a sample  being analyzed.

 A QC sample, analyzed at an analytical laboratory, that was prepared
 by adding a known concentration  of analyte to  a  sample.   Matrix
 spike samples can be used to determine possible chemical interferences
 within  a sample that might affect the analytical  result.

 Precision estimates based on pooled  standard deviations and pooled
 relative standard  deviations of processing and analytical laboratory
 duplicate samples.

 A measure  of light  scatter by a  solution of  suspended  materials
 detected at an angle of 90 degrees  to an incident  light source.
A  formal  on-site  review  of  field  sampling,   field  laboratory,  or
analytical laboratory  activities to  verify  that standardized protocols
are being followed.

A  measurement of pH or dissolved inorganic  carbon obtained  from
a  sample  collected  in  a beaker  and  exposed to the  atmosphere
during collection, processing, and preparation before  measurement.
                                               200

-------
outlier
P95
Observation not  typical of  the  population  from  which  the  sample
is drawn.

The 95th percentile of a distribution of blank sample measurements.
percent conduc-
  tance difference
  calculation

percent ion
  balance difference
  calculation

percent relative
  standard
  deviation  (%RSD)

pH
pH-ANC
pH-BNC
platinum cobalt
  unit (PCU)

population estimate
precision
processing
   laboratory

processing
   laboratory blank

processing
   laboratory
   duplicate
A QA procedure used to check that the measured specific conductance
does  not  differ  significantly (outside  the acceptance criteria) from
the calculated specific conductance value.

A QA procedure used to check  that the sum of the  anion equivalents
equals the sum of the cation equivalents (see anion-cation balance).
The  standard  deviation  divided  by  the  mean,  multiplied by 100,
expressed  as percent  (sometimes  referred to as  the coefficient  of
variation).

The  negative  logarithm   of  the   hydrogen-ion  activity.    The  pH
scale runs from  1  (most acidic) to 14 (most  alkaline); the difference
of 1 pH unit  indicates a tenfold change in hydrogen-ton activity.

A measurement of pH made in the analytical  laboratory immediately
before the ANC titration procedure and before the potassium chloride
spike has been added.

A measurement of pH made in the analytical  laboratory immediately
before the BNC titration procedure and before the potassium chloride
spike has been added.

A measure  of  the  color of a  water sample defined  by a  potassium
hexachloroplatinate and cobalt  chloride standard color  series.

A statistical estimate of  the  number of streams  (target  streams)
that have  a particular set of chemical characteristics (i.e., alkalinity
class within a subregion)  extrapolated from the number  of streams
sampled (probability sample).

A measure  of  the capability  of a method to  provide  reproducible
measurements  of a particular analyte.

The laboratory that processed samples and measured selected variables.
The  NSS-I processing laboratory was  located in Las Vegas, Nevada.

An ASTM Type I reagent-grade  water  sample prepared and processed
at the processing laboratory but analyzed  at an analytical laboratory.

A split sample  prepared and analyzed at  the processing laboratory.
                                               201

-------
 protolyte
protolyte analysis
   program
QC chart
quality assurance
   (QA)
quality assurance
  sample
quality control
  (QC)
quality control
  check sample
  (QCCS)

quality control
  sample
raw  data set
reagent
reagent blank
  sample

representativeness
required
  detection limit
 That portion  of a  molecule  that reacts with either  H+ or  OH'  in
 solution.

 An  exception-generating  computer program  of  AQUARIUS  II that
 evaluates  in  situ,  processing  laboratory,  and  analytical  laboratory
 measurements  of pH,  DIG, ANC, BNC,  and  DOC  in  light  of known
 characteristics of  carbonate equilibria.

 A graphical  plot  of  test  results with respect to  time or sequence
 of  measurement,  together with  limits within  which the results are
 expected  to  lie when the system is  in a state  of  statistical  control
 (Taylor, 1987).

 Steps taken  to  ensure  that  a  study  is  adequately  planned and
 implemented to provide data of the highest  quality and that adequate
 information is provided to determine  the quality of  the  data  base
 resulting from the study.
                                                   I
 A sample (other  than  the routine  stream  sample)  that is  analyzed
 in the analytical laboratory and that  has an origin and composition
 unknown  to the analyst.

 Steps taken  during  sample  collection, processing,  and analysis  to
 ensure that the  data quality meets the minimum standards established
 by the QA plan.

 A sample of known concentration used to verify  continued calibration
 of an instrument.
Any sample used by  the  analyst  to  check  immediate  instrument
calibration or response;  the measurement  obtained from  a  quality
control  sample is expected  to fall within  specific acceptance  criteria
or control limits.

Data  Set  1.   The  initial  data  set  that  received  a cursory  review
to confirm that data  are  provided in proper format and are complete
and legible.

A  substance  (because of its  chemical reactivity)  added  to water
to determine the concentration of a specific analyte.

A  laboratory blank sample that contained all  the  reagents required
to prepare a sample for analysis.

The degree to which  sample data  accurately  and precisely  reflect
the characteristics of a population.

For each  chemical  variable,  the highest  instrument detection  limit
based  on analyses  of laboratory  blanks or  detection  limit  check
standards allowable in the analytical  laboratory contract.
                                              202

-------
routine sample


sample ID


sampling  crew


SAS
specific
   conductance
 spike


 split sample
 standard
   deviation
 statistical
   (significant)
   difference

 stream  ID
 synoptic
 system decision
    limit
  system-level
    precision
The  first stream sample  collected at  a  site  in  accordance with
standardized protocols.

The  numeric identifier given to each stream sample  and to each  QA
sample in each batch.

A team of stream sampling personnel who gained access to the stream
site  on foot or by vehicle.

Statistical  Analysis  System, Inc.   A statistical  data  file manipulation
package that  has data management, statistical,  and graphical analysis
abilities.  The NSS-I data  base was developed and analyzed primarily
using SAS software and  is distributed  in SAS format.

A measure  of the electrical conductance (the reciprocal of the electrical
resistance)  or total  ionic strength of  a water  sample  expressed as
microsiemens per centimeter at! 25 °C.

A known  concentration  of  an analyte introduced into a  sample or
aliquot.

A replicate portion or  subsample of  a  total sample  obtained in
such  a manner that  it  is not  believed  to differ significantly from
other  portions of the same sample (Taylor, 1987).

The  square  root  of  the variance of  a  given  statistic,  calculated
by the equation:
                                standard deviation = yZ(x - x)2(h - 1)
 A conclusion based on a stated probability that two sets of measure-
 ments did not come from the same population of measurements.


 An   identification  code,  assigned  to  each  stream  in  the  survey,
 which  indicates  subregion,  alkalinity   characteristics,   and  map
 coordinates.

 Relating to  or displaying  conditions  as  they exist  simultaneously
 over a broad area.

 For  each chemical variable except  pH, a value  that  reliably indicates
 a  concentration  above  background, estimated  as  either  the 95th
 percentile  (Pgs)  or  as  1.65  times the standard  deviation  of  the
 field blank sample measurements.

 Cumulative  variability  associated with  sample collection,  transport,
 processing,  preservation,  shipment, analysis,  and  data  reporting.
 An  estimate  of  data variability for  each  analyte  associated  with
 analyte concentration; the estimate is based on the statistical evalua-
 tion of field routine-duplicate pairs.
                                                203

-------
systematic error


tag


target population


target stream

theoretical value


titration data

true color



turbidity


validated data  set
validation



verification


verified data set
within-batch
  precision
withheld  sample
A consistent difference introduced in the measuring process.   Such
differences commonly result in biased estimations.

Code on a data point that  is added at the  time of  sample collection
or analysis to qualify the datum.

In this survey, the stream  population  of  interest that  was sampled.
This population was defined by the sampling protocol.

A stream of interest in the target  population.

The  expected  value  of the  synthetic audit sample  assuming  no
preparation error and no external effect.

Individual data  points from the Gran  analysis of  ANC and BNC.

The  color  of water  that has been  filtered  or centrifuged to remove
particles  that  may impart  an  apparent  color;  true  color   ranges
from clear blue to  blackish-brown.

A measure of light scattering by suspended particles in an unfiltered
water sample.

Data  Set  3.   Final  product  of  the  validation  process  in  which
sample data are  examined in  the  context  of  a  subregional  set of
samples,  rather than at the  batch  and sample level.   Outliers are
identified and  flagged.   Data  for  field  blank,   field  duplicate,  and
performance audit  samples are included in this data  set.

Process by  which data  are evaluated  for  quality  with reference to
the intended data use; includes identification  of outliers and evaluation
of potential systematic error after  data verification.

Process of   ascertaining  the  quality  of the   data in  accordance
with the minimum standards established by the QA plan.

Data  Set  2.   Final  product  of  the  verification process  in  which
each sample batch and each sample value has been reviewed individually
and  all questionable  values  are corrected  or identified with  an
appropriate flag.

The  estimate of precision expected  in the  analysis of  samples in
a batch by the same  laboratory on  any single  day.  In this  report,
overall within-batch precision includes the effects of sample collection,
processing,  and analysis;  analytical within-batch precision  includes
the  effects of sample analysis within  the analytical  laboratories.

An  additional  duplicate  sample  collected  from a   stream  by  the
sampling  crew  as  part of  a  holding-time experiment.   This  sample
was held in the dark at 4 *C  for a 24-hour period prior  to processing
and preservation.
                                               204

-------
within-laboratory         A  precision  goal  based  on  the  data   quality  objectives  for  the

  precision goal         analysis of laboratory duplicate samples within a single laboratory.
                                                  205
                                                   «U.S.GOV£RNMENTPR1OTNGOFFICE: 19 B 8- 5 „« 5 ./ 870 3 S

-------
                 SUBREGIONS  OF THE  NATIONAL STREAM SURVEY-PHASE I
C>
O
  cc nJ 4
  c+ f O
  -  I  e+
    M CD
  ;a3 GJ o
  o — d-
  o   H-
  B   o
  -3   G»J
  O   CD
      3
      o
                                            Northern

                                        Appalachians (2Cn)
                                                           Valley and Ridge (2Bn)
                 Southern Blue Ridge (2As)

                   (Pilot Study)
   Poconos/Catskills (ID)

         NY\
             Ozarks/Ouachitas (2D)
  Mid-Atlantic
Coastal Plain (3B)
           Southern Appalachians (2X)

-------