vvEPA
United States
Environmental Protection
Agency
Office of Acid Deposition,
Environmental Monitoring and
Quality Assurance
Washington DC 20460
EPA/600/4-86/011
August 1987
Research and Development
Eastern Lake Survey
Phase I
Quality Assurance
Report
-------
EPA 600/4-86/011
August 1987
Eastern Lake 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 89114
Environmental Research Laboratory - Corvallls. OR 97333
-------
NOTICE
The information in this document has been funded wholly or in part by the U.S. Environmental Protection
Agency (EPA) under Contract No. 68-03-3249 and 68-03-3050 to Lockheed Engineering and Management Services
Company, Inc., No. 68-02-3889 to Radian Corporation, No. 68-03-3246 to Northrop Services, Inc., and Interagency
Agreement No. 40-1441-84 with the U.S. Department of Energy. It has been approved for publication as an EPA
document.
Mention of corporation names, trade names or commercial products does not constitute endorsement or rec-
ommendation for use.
This document has been published previously. As part of the AERP Technical Information Program, this docu-
ment has been repackaged and retitled to clearly identify its relationship to other documents produced for the
Eastern Lake Survey. The document contents and reference number have not changed. Proper citation of this
document remains:
Best, M. D., S. K. Drouse, L. W. Creelman, and D. J. Chalous. National Surface Water Survey, Eastern Lake Survey
(Phase I - Pynoptic Chemistry) Quality Assurance Report. EPA600/4-86/011, U.S. Environmental Protection
Agency, Las Vegas, NV. 1986.
-------
ABSTRACT
The National Surface Water Survey (NSWS) is a three-phase project of the National Acid Precipitation Assess-
ment Program (NAPAP). The objectives of the survey are to evaluate the current water chemistry of acid-sensi-
tive lakes and streams in the U.S., to determine the status of fisheries and other biotic resources in those
waters, and to quantify changes in a representative subset of lakes and streams through a long-term monitoring
program. Phase I of the Eastern Lake Survey is the first part of the NSWS lake study. This Quality Assurance
Report is a retrospective, comprehensive overview of the quality assurance activities and results of the Eastern
Lake Survey Phase I. The report describes the chemical parameters measured, the sampling and analytical
methods used, and the quality assurance procedures required for field, laboratory, and data base operations.
The report also discusses the rationales and testing that led to the implementation of specific protocols. These
protocols were extensively reevaluated during and after the survey, as described in this document. The statisti-
cal testing of the analytical and quality assurance data is explained, and the results of these tests are pre-
sented.
Overall, Phase I of the Eastern Lake Survey was successful in achieving its objectives. The quality assurance
requirements proved adequate to ensure that all samples were collected and analyzed consistently, and that the
resulting data were scientifically sound and of known quality. This report was submitted in partial fulfillment of
contract numbers 68-03-3050 and 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 planning, imple-
mentation, and data review period from March 1983, to January 1986, and work was completed as of March 1986.
-------
Contents
Abstract iii
Figures vi
Tables vii
Acknowledgement ix
1. Introduction 1
Eastern Lake Survey 1
Survey Participants 1
Measurement Requirements 1
Data Quality Objectives 1
Development of Documents for the Eastern Lake Survey - Phase I 6
Pilot Studies 6
Sampling and Analytical Methodologies 6
Data Comparability Studies 7
2 Operational Quality Assurance-Quality Control Program 8
Field Station Organization and Responsibility 8
Selection of Contract Analytical Laboratories 8
Statement of Work 8
Invitation for Bid 9
Analytical Laboratory Evaluation 9
Training 9
Communications 9
Field Communications 9
Daily Communications with Contract Analytical Laboratories 10
Sampling and Field Laboratory Quality Control Protocols 10
Sample Preservation 10
Field Laboratory Analyses 10
Contract Analytical Laboratory Quality Control Protocols 13
Contract Analytical Laboratory Sample Holding Times 13
Reporting Requirements 14
Internal Quality Control 14
Field Laboratory and Contract Analytical Laboratory On-Site Evaluations 14
3. Data Base Quality Assurance 17
Data and Sample Transfer 17
Transfer of Samples and Data From Field Stations 17
Transfer of Contract Analytical Laboratory Data "17
Raw Data Set 22
Data Verification 22
Daily Communication 23
IV
-------
Data Receipt 23
Quality Assurance and Quality Control Data 27
Follow-up with Contract Analytical Laboratories 30
Preparation and Delivery of Verification Tapes 30
Data Validation 31
4. Results 35
Operations Evaluation 35
Lake and Sample Information 35
Field Problems and Their Resolutions 35
Contract Analytical Laboratory Problems and Resolutions 37
Data Verification Problems and Resolutions 37
Methods Evaluation 38
Fluoride Determinations 38
Acid-Neutralizing Capacity and Base-Neutralizing Capacity 38
Total Extractable Aluminum 38
pH Sample Chamber 39
Formation of Filterable Iron and Aluminum Complexes in Synthetic Field Audit
Samples 39
Nitrate Contamination 39
Evaluation of Quality Assurance Data 42
Blank Data 42
Duplicate Data 44
Audit Sample Data 48
5. Data Variability in the Eastern Lake Survey • Phase I 59
Comparisons of Precision Estimates 59
Expected Relationships Between Precision Estimates 59
Summary 61
References 62
Appendices
A - Analysis of Quality Assurance Data for the Eastern Lake Survey 65
B - Instrumental Detection Limits, System Detection Limits, and System Decision Limits by
Laboratory, Eastern Lake Survey - Phase 1 117
C - Overall Within-Batch Precision by Laboratory for 23 Parameters, Eastern Lake Survey -
Phase I 123
D - Analytical Within-Batch Precision by Laboratory for 23 Parameters, Eastern Lake Survey -
Phase I 129
E - Overall and Analytical Among-Batch Precision by Laboratory for 23 Parameters, Eastern
Lake Survey - Phase I 135
-------
Figures
Number Page
1 Regions and subregions sampled during the Eastern Lake Survey - Phase 1 2
2 Flow of samples and data through field and analytical laboratories, Eastern Lake
Survey - Phase I 11
3 NSWS Form 3 - Shipping 18
4 NSWS Form 1 - Lake Data 19
5 NSWS Form 2 - Field Laboratory Data 20
6 Eastern Lake Survey - Phase I data flow scheme 21
7 Data verification process for the Eastern Lake Survey - Phase I 26
8 NSWS Data Package Completeness Checklist 28
9 Data validation process for the Eastern Lake Survey - Phase I 34
-------
Tables
Number Page
1 Parameters Measured, Quality Assurance Criteria, and Analytical Methods for the
Eastern Lake Survey - Phase I 3
2 Sample Preservation Requirements 12
3 Descriptions, Applications, and Frequencies of Quality Assurance and Quality Con-
trol Samples, Eastern Lake Survey - Phase I 15
4 Descriptions and Applications of Quality Control Samples, Eastern Lake Survey -
Phase I 16
5 List of Data Forms Used by the Contract Analytical Laboratory, Eastern Lake Survey
- Phase 1 22
6 Eastern Lake Survey - Phase I Field and Laboratory Data Qualifiers (Tags') 23
7 Eastern Lake Survey - Phase I Verification 24
8 Exception Generating and Data Review Programs, Eastern Lake Survey - Phase I.. 30
9 NSWS Missing Value Codes 31
10 Data Validation for the Eastern Lake Survey - Phase I: Comparison of Variables Used
to Check for Random and Systematic Errors 32
11 Physical Parameters Subject to Validation, Eastern Lake Survey - Phase I 33
12 Numbers of Samples Received and Analyzed by Contract Laboratories, Eastern Lake
Survey - Phase I 36
13 Numbers of Samples Delivered by Field Stations, Eastern Lake Survey - Phase I .. 36
14 Cleaning Procedure Used for Aliquot 3 Sample Containers, Eastern Lake Survey -
Phase I Pilot Study 40
15 Filtration Procedure Originally Used by Field Personnel, Eastern Lake Survey -
Phase I 40
16 Field Laboratory Filtration Procedures Used in Nitrate Contamination Experiment,
Eastern Lake Survey - Phase I 41
17 Description of Field Laboratory Blank Samples Collected in Nitrate Contamination
Experiment, Eastern Lake Survey - Phase I 42
18 Nitrate Concentrations in Field Laboratory Blank Samples in Nitrate Contamination
Experiment, Eastern Lake Survey - Phase I 43
19 Instrumental Detection Limits, System Detection Limits, and System Decision Lim-
its for 20 Parameters, Eastern Lake Survey - Phase I 45
20 Overall Within-Batch Precision Estimated from Field Duplicate Data and Field Blank
Data for Measurements of 23 Parameters, Eastern Lake Survey - Phase I 46
21 Overall and Analytical Within-Batch Precision Estimated from Field Duplicate and
Trailer Duplicate Data for Measurements of Four Parameters, Eastern Lake Survey -
Phase I 49
22 Analytical Within-Batch Precision Estimated from Contract Laboratory Duplicate
Data and Calibration Blank Data for Measurements of 23 Parameters, Eastern Lake
Survey - Phase I 50
vu
-------
23 Overall Among-Batch Precision Estimated from Field Natural, Lot 2 (FN2, Big Moose
Lake) and Field Natural, Lot 3 (FN3, Lake Superior) Audit Sample Data For Measure-
ments of 23 Parameters, Eastern Lake Survey - Phase I 52
24 Mean Measured Values, Overall and Analytical Among-Batch Precision Estimates,
and Theoretical Concentrations of High Synthetic Audit Samples, Eastern Lake Sur-
vey - Phase I 53
25 Mean Measured Values, Overall and Analytical Among-Batch Precision Estimates,
and Theoretical Concentrations of Low Synthetic Audit Samples, Eastern Lake Sur-
vey - Phase I 55
26 Contract Analytical Laboratory Performance Windows for Audit Sample Measure-
ments, Eastern Lake Survey - Phase 1 58
-------
Acknowledgments
The completion of this report would not have been possible without the efforts of many individuals who are
associated with the National Surface Water Survey. In particular, the authors wish to acknowledge the technical
assistance, contributions, and editorial comments that were provided by the following people: R. Linthurst, E.
Meier, and R. Schonbrod (U.S. Environmental Protection Agency); D. Landers (State University of New York); P.
Kanciruk(Oak Ridge National Laboratory); C. Macleod, D. Hillman, D. Peck, T. Lewis, M. Faber, J. Engels, J. Villa,
M. Stapanian, J. Lau, C. Mericas, and S. Simon (Lockheed-EMSCO); J. Eilers (Northrop Services, Inc.); D. Brakke
(Western Washington University); T. Permutt and A. Pollack (Systems Applications Inc.); and the word process-
ing staff of Computer Sciences Corporation.
IX
-------
Section 1
Introduction
The National Surface Water Survey (NSWS) is a three-
phase project within the National Acid Precipitation
Assessment Program (NAPAP). The NSWS was initi-
ated by the U.S. Environmental Protection Agency
(EPA) in 1983. The purpose of Phase I of the NSWS is
to document the present chemical status of lakes and
streams in areas of the U.S. that are potentially sus-
ceptible to the effects of acidic deposition (Linthurst
et al., 1986). Phase II of the survey is intended to deter-
mine the present status of biotic resources and to
assess the chemical variability within and among sur-
face waters characterized during Phase I. Phase III, a
long-term monitoring program, will quantify changes
in the aquatic resources of a subset of Phase II sur-
face waters.
The scientific and legislative decisions that will be
based on the data from the NSWS must be well sup-
ported. Therefore, an extensive quality assurance
(QA) program has been established to ensure that the
best possible data are collected and that the quality
of the data can be defined and defended at the com-
pletion of the field surveys.
Eastern Lake Survey
This report summarizes the results of the QA program
for Phase I of the Eastern Lake Survey (ELS-i). The QA
program, including data verification and validation
procedures, is discussed in greater detail in Drouseet
al. (1986). Analytical methods and field sampling pro-
tocols for the ELS-I are discussed in detail in Hillman
et al. (1986) and Morris et al. (1986), respectively. In
total, 1800 lakes were sampled in the ELS-I. The lakes
were selected from three regions east of the Missis-
sippi River that are potentially susceptible to acidifi-
cation (Figure 1).
Survey Participants
Planning, conducting, and interpreting a study of the
magnitude of the NSWS required the cooperation of
numerous government agencies and private organi-
zations. Development of the NSWS included defini-
tion of measurement requirements and survey objec-
tives, which was accomplished through discussions
at meetings and workshops. Participants included
representatives of several government agencies as
well as other scientists involved in acidic deposition
research (U.S. EPA 1984a and 1984b).
The U.S. EPA Environmental Monitoring Systems Lab-
oratory-Las Vegas, Nevada (EMSL-LV) had primary
responsibility for the ELS-I QA program and sampling
operations. The Agency receives assistance in this
area from Lockheed Engineering and Management
Services Company, Inc. (Lockheed-EMSCO) which is
the prime contractor for EMSL-LV. Sampling and qual-
ity assurance activities were performed by Lockheed-
EMSCO personnel. The relevant state agencies and
EPA regional offices were also involved in the sam-
pling activities. The data base operations were per-
formed by Oak Ridge National Laboratory (ORNL).
The EPA Environmental Research Laboratory-Corval-
lis, Oregon (ERL-Corvallis) had primary responsibility
for design and planning of the ELS-I, as well as for
data validation and interpretation.
Measurement Requirements
The first step in design of the QA program was to
define the measurement requirements for the ELS-I.
Thirty-two parameters were selected for in situ or lab-
oratory measurement during Phase I (Table 1). The
EPA data users (see Linthurst, et al. 1986) reasoned
that measurement of these chemical and physical
parameters would provide adequate information from
single lake samples for the evaluation of the present
status of lakes and thus would meet the objectives of
Phase I. A brief description of each parameter is pre-
sented in Overton et al. (1986).
Data Quality Objectives
Data quality objectives (DQO's) were defined early in
1984 in terms of anticipated value range, detection
limits, and precision for each measurement parame-
ter. These objectives were developed using data from
published literature and from statistical error simula-
tion. Equipment, sampling protocols, and analytical
methodologies were selected and standardized in
order to achieve the DQO's. The pilot studies then
afforded the opportunity to evaluate and revise the
analytical methodologies, equipment, and DQO's.
One change to the DQO's was in the detection limit for
total phosphorus which was lowered from 3 to 2 ppb
because experience indicated the lower limit was
-------
2C
2A
2D
UPPER
MIDWEST
REGION
NORTHEAST
REGION
\ ili
\ / v
1 \ >
v>—A /
^ \^~r
\ —
SOUTHEAST
REGION
Figure 1. Regions and Subregions Sampled During the Eastern Lake Survey - Phase I.
-------
Observed
Parameter3 Range
IN SITU
pH 360-10.46
Conductance pS crrf1 00-1267.0
Lake Temperature. *C
Secchi disk
Transparency, m
FIELD LABORATORY
Laboratory pH 3.81 - 9.36
(closed system)
Dissolved inorganic carbon 0 16 - 48.99
(closed system), mg L'1
True color, PCU 0 - 345
Turbidity, NTU 0 - 290
CONTRACT ANALYTICAL LABORATORY
Acid-neutralizing capacity -209.1
(ANC).Meq L"1 +40466
Intra- Maximum
Required Laboratory Sample Holding
Detection Precision Time - Days Instrument
Limit Goal (%)" (Analytical Lab) or Method0
Potentiometer
(Hydrolab)
Conductivity cell
(Hydrolab)
Thermistor
(Hydrolab)
Secchi disk
0.1 g — pH meter
Orion Model 611
0.05 10 — Infrared spectroscopy
Dohrmann DC-80
carbon analyzer
0 5 g — Comparator Hach
Model CO-1
2 10 — Nephelometer Monitek
Model 21
5 10 14 Acidimetric titration,
modified Gran analysis
Reference
(Laboratory
Methods)''
Morris et al.
(1986)
Morris et al
(1986)
Morris et al.
(1986)
Morris et al.
(1986)
EPA 150.1
EPA 41 5 2
(modified)
EPA 110. 2
(modified)
EPA 180.1
Hillman et al.
(1986);
Kramer (1984)
Table 1. Parameters Measured, Quality Assurance Criteria, and Analytical Methods for the Eastern Lake Survey - Phase I.
-------
Parameter3
Total Extractable
Total
Base-neutralizing capacity
Calcium (ca),
mgr1«>
Chloride (C1"),
mg L~1(l)
Conductance, /uS cm"1
Dissolved inorganic carbon
(DIG) (air-equilibrated), mg L"1
Dissolved inorganic carbon
(DIG) (initial ANC), mg L"1
Dissolved organic carbon
(DOC).mg L"1
Fluoride (F~), total dissolved,
mg L
Iron (fe), mg L"1(e)
Magnesium (Mg), mg L"1(e)
Manganese (Mn), mg L~1(d)
Observed
Range
-0.009-3.594
-0.002-9.678
0.19-60.94
0.01-609
7.8-3613.3
-0.07-46.91
0.15-4983
0.0-48.2
0 001 -0.839
-0.034-2.64
0.10-39 76
-0 02 - 2 03
Required
Detection
Limit
0.005
0.005
5
0.01
0.01
-h
0.05
0.05
0 1
0.005
001
0.01
0.01
Intra-
Laboratory
Precision
Goal(%)b
10 (X). 010)
20 (<0.010)
10(>0.010)
20(<0.010)
10
5
5
1
10
10
5 (>5.0)
10 (<5 0)
5
10
5
10
Maximum
Sample Holding
Time - Days
(Analytical Lab)
7
28
14
28
28
14
14
14
14
28
28
28
28
Instrument
or Method11
Furnace AAS on MIBK
extract
Furnace AAS
Alkalimetric titration,
modified Gran analysis
Flame AAS, ICPES
1C
Conductivity cell
IR
IR
IR
ISE
Flame AAS, ICPES
Flame AAS, ICPES
Flame AAS
Reference
(Laboratory
Methods/
Hillman et al.
(1986)
EPA 202.2
Hillman et al.
(1986); Kramer
(1984)
EPA 215.1
ASTM (1984);
O'Dell etal.
(1984)
EPA 120.1
EPA 415. 2
(modified)
EPA 415. 2
(modified)
EPA 415. 2
EPA 340.2
(modified)
EPA 236
EPA 242.1
EPA 243.1
Table 1. Parameters Measured, Quality Assurance Criteria, and Analytical Methods for the Eastern Lake Survey - Phase I (Continued).
-------
Observed
Parameter3 Range
Required
Detection
Limit
Intra-
Laboratory
Precision
Goal (%)"
Maximum
Sample Holding
Time - Days
(Analytical Lab)
Instrument
or Method1
Reference
(Laboratory
Methods/
"1(e>
Nitrate (N03~, mg L
pH (air-equilibrated)
pH (initial ANC)
pH (initial (BNC)
Phosphorus (P), total, mg
-'(e)
Potassium (K), mg L~1(f)
Silica (Si02), mg L"1(e)
Sodium (Na), mg L"1|f)
Sulfate (S042~), mg L'1"1
-0.106-30 6
3.82-893
380-8.78
3.81 -882
-0.006- 0.833
000-24.98
-1.14-4353
0.06-323
00-119
0.005
0.002
0.01
0.05
0.01
0.05
10
0.5
0.05
0.05
10 (>0 010)
20(<0.101)
5
5
5
5
7 1C
7 pH meter
7 pH meter
7 pH meter
28 Colorimetry
(phosphomolybdate, or
modification,
automated)
28 Flame AAS
28 Colorimetry
(automated)
7 Flame AAS, ICPES
28 1C
ASTM (1984),
O'Dell etal.
(1984)
EPA 150.1
EPA 150 1
EPA 150.1
USGS I-4600
EPA 258 1
USGSi-2700
EPA 273 1
ASTM (1984),
O'Dell etal.
(1984)
a Dissolved ions and metals were determined, except where noted
Relative precision was calculated for samples at concentrations above 10 times instrumental detection limits, except where noted
c AAS = atomic absorption spectroscopy, MIBK = methyl isobutyl ketone, ICPES = mductilely coupled plasma emission spectroscopy, 1C = ion chromatography, IR
infrared spectrophotometry; ISE = ion-selective electrode
d In situ measurements are outlined in Morris et al. (1986), EPA methods are from U.S. EPA 1983); USGS methods are from Skougstad et al (1979)
e Values converted to/*g L"1 for data analysis Required detection limits are in mg L"'.
Values converted to Meq L~1 for data analysis Required detection limits are in mg L"'.
9 Absolute precision goal in applicable units
The mean of six nonconsecutive blank measurements was required to be less than 0 9*iS cm"'.
Table 1. Parameters Measured, Quality Assurance Criteria, and Analytical Methods for the Eastern Lake Survey - Phase I (Concluded).
-------
necessary. The final DQO's were provided in a docu-
ment delivered to the quality assurance management
staff of the EPA in October 1984. The observed ranges
of values in the verified data set, required detection
limits, intralaboratory precision goals, maximum
sample holding times, and measurement methods
used for each variable are listed in Table 1. The
instances where extreme values affected the esti-
mated precision are listed in Section 5.
Development of Documents for the
Eastern Lake Survey (Phase I)
Protocols for sampling, chemical analysis, and data
processing were based on the best available pub-
lished methods. A draft QA plan and a draft analytical
methods manual were used during pilot studies con-
ducted to test all aspects of the ELS-I research plan,
including logistics, methods, and QA. The evaluation
and modification of the data quality objectives, sam-
pling and analytical methodologies, and verification
and validation procedures were of particular impor-
tance to the QA program. Prior to commencement of
the ELS-I, the draft QA plan and the draft analytical
methods manual were revised on the basis of results
obtained from the pilot studies. The final QA plan and
the final methods manual incorporated both those
revisions and any changes implemented during the
ELS-I.
Pilot Studies
Two pilot studies for the ELS-I were undertaken dur-
ing the winter and spring of 1984. The winter pilot
study was conducted in January 1984 and consisted
of sampling 50 ice-covered lakes in Maine, New
Hampshire, and Vermont. During the spring pilot
study in May-June, 137 lakes in Maine, New Hamp-
shire, Vermont, Massachusetts, and New York were
sampled. These studies were implemented to evalu-
ate all aspects of the National Surface Water Survey
research plan including lake selection, proposed
sampling protocols, the QA program, and data man-
agement. Objectives of the pilot studies are outlined
in greater detail in Drouse et al., 1986. The process of
lake selection is discussed in Linthurst et al. (1986).
Sampling and Analytical Methodologies
Modifications to the original sampling and analytical
protocols were made based on the experience gained
during the pilot studies. These modifications inclu-
ded elimination of the dissolved oxygen measure-
ments which were deemed to be time consuming and
unnecessary, and revision of the sample-bottle wash-
ing procedure which was discovered to introduce
nitrate contamination (see Section 4). In addition,
four experiments were conducted prior to the ELS-I to
investigate (1) the possibility of contamination by hel-
icopter exhaust, (2) representativeness of single-
point sampling, (3) variability of measurements made
using the Hydrolab units, and (4) effects of increased
pressure (depth) on Hydrolab accuracy (Morris et al.
1986).
The primary goals of base site operations were to
obtain accurate physicochemical and geographical
data at each lake site, to collect representative lake
samples without introducing contamination, to pre-
serve the integrity of samples until their analysis at
contract laboratories, and to perform selected chemi-
cal analyses. The objectives of the field laboratory
are defined as follows:
• receive lake and QA samples and field data from
each sampling team and assess sample condi-
tion upon receipt
• review lake data forms for accuracy and com-
pleteness
• incorporate audit samples with lake samples to
form a batch
• analyze the batch samples for pH, DIG, true color,
and turbidity
• perform aluminum extraction
• filter, preserve, and ship samples to contract
analytical laboratories for detailed analysis
• coordinate sample shipment information with
the Sample Management Office and EMSL-LV
• distribute copies of NSWS Forms 1 and 2 to the
appropriate offices
The protocols for collection of field data and water
samples during the ELS-I were implemented in three
phases: preflight preparation, lake site activities
(including sampling methodologies), and postflight
operations (Morris et al. 1986).
Standardized forms were developed to record meas-
urements made at each lake and at the field and con-
tract analytical laboratories. The multicopy field and
field laboratory forms were checked for complete-
ness and internal consistency at the field station.
One copy of each form was sent to Oak Ridge
National Laboratory for entry into the NSWS data
base and a second copy was sent to quality assur-
ance personnel in Las Vegas; both copies were sent
by overnight mail service. Transfer of samples and
data is discussed in Section 3.
Analytical methodologies were selected for the ELS-I
on the basis of guidelines from the DQO's. For some
analytes, two or more techniques were considered to
be equivalent based on published literature (Hillman
et al. 1986). The pilot studies and field and laboratory
experiments provided opportunities to evaluate the
relative merits of each technique. Two techniques
were judged to be equivalent for determining the con-
-------
centrations of total phosphorus and of dissolved
metals (Ca, Fe, K, Mg, Mn, Na), and either technique
was permitted. However, measurements of free
(uncomplexed) dissolved fluoride by ion-selective
electrode and of total dissolved fluoride by ion chro-
matography were found to lack reproducibility, and
these methods were eliminated. The field measure-
ment and analytical methods used for the ELS-I are
summarized in Table 1.
Data Comparability Studies
Standardized techniques were specified for sampling
and chemical analysis of water samples from the
ELS-I. Standardization ensured that we could identify
the effect of variance (if any) of sampling and analysis
on differences identified between lakes.
A study was also undertaken to determine whether
ELS-I data could be compared to survey data from
other countries. Subsamples were obtained from 215
ELS-I samples collected in the southern Blue Ridge
Mountains (NSWS subregion 3A, see Figure 1) and
shipped via commercial courier to Norway for chemi-
cal analysis of 14 parameters. Similarly, 105 subsam-
ples from the Adirondack Mountains (NSWS subre-
gion 1A) were analyzed in Canada for 18 chemical
parameters. Results of this study are summarized in
Stapanian et al. (1986).
A second study utilized 2047 split samples from the
ELS-I to compare chemical analyses by flame atomic
absorption spectroscopy (AAS) and inductively coup-
led plasma emission spectrscopy (ICPES). The ICPES
analyses were performed at ERL-Corvallis and by
ELS-I laboratories, and flame AAS analysis by ELS-I
laboratories only. These analyses will be presented in
Drouse and Best (1986).
-------
Section 2
Operational Quality Assurance - Quality Control Program
The QA plan (Drouse et al. 1986) describes the quality
assurance-quality control (QA-QC) program of the
ELS-I, and the analytical methods manual (Hillman et
al. 1986) documents all methodologies used in the
survey at the standard-operating-procedure (SOP)
level of detail. The major aspects of the operational
QA-QC program are summarized in the following sec-
tions. Quality assurance aspects related to the data
base are discussed in Section 3. More detail is pro-
vided in the documents just referenced.
Field Station Organization and
Responsibility
The operation of a field station required the establish-
ment of a field laboratory, a calibration room, and a
local communications center to coordinate field
activities. Fifteen people were based at each field sta-
tion, including two helicopter pilots and a mechanic,
an EPA base coordinator, an EPA duty officer, five lab-
oratory personnel, and five field samplers. Field labo-
ratory personnel included a laboratory coordinator, a
laboratory supervisor, and three analysts. All labora-
tory personnel were cross-trained in sample prepara-
tion and analysis. At some sites, individuals rotated
analyst positions on a weekly basis. All personnel
reported to the EPA base coordinator who was
responsible for the overall operation of the base site.
The duties of the field station personnel are summa-
rized below and are discussed in greater detail in Mor-
ris et al. (1986).
Selection of Contract Analytical
Laboratories
The analytical requirements and QA approach were
defined during the development of the QA program.
The estimated number of samples to be analyzed and
the estimated rate of sample collection were defined
in the logistics planning. It was recognized early in
the planning that no single analytical laboratory
could analyze the number of expected samples at the
rate they were to be collected and still meet the QA-
QC requirements (especially the required holding
times) that had been established. There was not a sin-
gle EPA laboratory that had all of the analytical capa-
bilities or resources to provide the required analytical
support. This meant that the analytical support
would have to be obtained via contracts with commer-
cial analytical laboratories. A Contract Laboratory
Program (CLP) had already been established to sup-
port the hazardous waste monitoring activities of the
Environmental Protection Agency across the United
States. Use of the CLP to obtain contract analytical
laboratories for the NSWS was reasonably inexpen-
sive. Use of multiple contract analytical laboratories
also meant that additional care would have to be
placed in the selection and documentation of analyti-
cal methods and QA activities to assure 9 consistent
and adequate performance in all laboratories. The
contracting process required the following activities:
• preparation of a statement of work (SOW) that
defined the analytical and QA-QC requirements
in a contractual format
• preparation and advertisement of an invitation
for bids (IFB) to solicit contractor support
• evaluation of the lowest bidders to assure that
qualified laboratories were selected
Statement of Work
Control of data quality within each contract analyti-
cal laboratory was necessary in order to control data
quality among the different laboratories and to mini-
mize the data variability. The methods manual and QA
plan had been drafted in the early phases of the plan-
ning process. However, in order to obtain analytical
support contracts, the methods and the QA-QC
requirements had to be prepared in a contractual for-
mat or statement of work (SOW). This involved careful
review of the analytical and logistics requirements to
assure that they were clearly stated and could be
enforced under the terms of the contracts. Every
effort was made to assure that the reporting and QC
requirements were clearly stated in the SOW. The
major contractual requirements in the SOW were the
following:
• A contractor could bid on the analysis of one or
more bid lots (500 samples per bid lot) that would
be delivered to the analytical laboratory at a
maximum rate of 30 samples per day per bid lot.
• Maximum holding times were specified. Failure
to meet this requirement resulted in a penalty to
the contractor of 20 percent per day per analyti-
-------
cal subunit (7-, 14-, or 28-day holding time group)
to a maximum of 100 percent of the bid price.
• Delivery of the completed data package was
required within 35 days of sample receipt by the
contractor. An incentive fee for early delivery of
data and a penalty for late delivery of data (1 per-
cent per day up to a maximum of 10 percent of
the bid price in each case) were established.
• Failureof thecontractorto provide adequate QA-
QC data and deliver ables as required by the
SOW resulted in a penalty of up to 15 percent of
the bid price.
The contractor laboratories were required to follow
the methods exactly as specified in the SOW. The QA
manager was allowed to make interpretations for the
contract laboratory, but contractual changes were
only made with the approval of the EPA contract offi-
cer.
Invitation for Bid
The SOW and IFB reflected the experience gained
from using contract analytical laboratories during the
pilot studies. The IFB was prepared and advertised in
Commerce Business Daily. Approximately 200 labora-
tories responded to the advertisement and were sent
copies of the SOW. Twenty-five laboratories submit-
ted bids. The twelve lowest bidders were considered
to have submitted reasonable bids and were selected
to receive the preaward performance evaluation (PE)
samples for the second step of the selection process.
Analytical Laboratory Evaluation
The lowest bidders were required to analyze PE sam-
ples and report the results within 15 days after sam-
ple receipt. The PE samples were prepared to repre-
sent lake samples at both the low and high
concentrations expected for the survey. Each bidder's
data and data package was evaluated and scored.
The scoring was based on evaluation of the quality of
the analytical data as well as the quality and com-
pleteness of the data package itself. This process
eliminated those laboratories that could not perform
the analytical and data reporting requirements. A
more detailed description of analytical laboratory
performance evaluations is provided in Drouse et al.
(1986).
Each of the laboratories that passed the PE sample
evaluation was visited by an EPA team in order to ver-
ify the qualifications and capabilities of the laborato-
ries to meet the contractual requirements. The EPA
team determined whether the analytical laboratory
had adequate equipment, personnel, and facMities to
analyze samples in accordance with the SOW. These
visits also provided an opportunity to clarify contract
requirements with the analytical staff and to identify
deficiencies that were observed in the PE sample
data evaluation. The results of these onsite evalua-
tions are on file with the QA manager. The lowest bid-
ders who passed both the PE sample and onsite eval-
uations were awarded contracts to provide analytical
services for the ELS-I.
Training
Data quality depended on the ability of the field and
laboratory personnel to properly collect, process, and
analyze the samples. Operation of the ELS-I required
a large staff composed of Lockheed-EMSCO employ-
ees, EPA regional and EMSL personnel, and the con-
tract analytical laboratories. Training was essential
to ensure consistent application of all operational
and quality assurance procedures. Lockheed-
EMSCO field sampling personnel received six days of
intensive technical and safety training during a 15-
day orientation program in September 1984 at the U.S.
EPA EMSL-LV. Personnel from the regional EPA
offices who would be involved in sampling were
trained at the field stations prior to commencement
of sampling. Details of these training sessions are
contained in Morris et al. (1986).
A meeting was held in early August 1984 to review and
discuss the objectives and requirements for the ELS-
I. Participants included representatives from the
NSWS management team, EMSL-LV QA staff, data
base management team, contract analytical labora-
tories, and the analytical support laboratory. The
objective of the meeting was to ensure that all parties
would implement both 11 the analytical methods and
the QA and QC procedures accurately and consist-
ently. This meeting also provided an opportunity for
the participants to clarify the survey requirements.
Communications
Coordination of the ELS-I operations required close
communication to ensure all program objectives
were met. During the actual sampling phase, the most
critical lines of communication were between Las
Vegas and the field stations and between Las Vegas
and the contract analytical laboratories. Daily com-
munication was required concerning both logistical
and quality assurance topics.
Field Communications
Field sampling activities were closely monitored
each day to ensure safety and logistical coordination.
Regular communication among the field stations and
Las Vegas was also necessary. A local communica-
tions center staffed by the EPA base coordinator and
the duty officer was established at each field station,
and a primary communications center was estab-
lished in Las Vegas for these purposes.
The logistics communications center in Las Vegas
was a clearinghouse for information about the num-
-------
her and type of lakes sampled, sample shipment
schedules, helicopter flight hours, and long-range
weather forecasts. Logistics personnel monitored
the survey by coordinating and tracking shipment of
QA and analytical samples to contract laboratories,
and by coordinating the shipment of supplies to field
stations.
Computer software utilized in tracking the progress
of lake sampling activities was developed before
sampling began. Maps for the daily tracking of field
activities were inventoried and were displayed by
region. Bulletin boards and chalkboards were
installed to effectively monitor field activities. All
communications were logged on a field communica-
tion form. Sampling progress was graphically dis-
played on regional maps with color-coded flags to
indicate lakes sampled and lakes remaining to be
sampled Progress reports were made by phone, and
a written report was made twice weekly to the NSWS
management team.
The establishment of communications centers and
the implementation of communications plans
enabled field operations to proceed in a coordinated
and consistent manner although field stations were
located over a wide geographic area.
Daily Communications with Contract Analytical
Laboratories
Daily calls were made to each contract analytical lab-
oratory by the QA staff. The primary objective of
these calls was to ensure that QA and QC procedures
were being implemented according to the survey
requirements and that the samples were being han-
dled and analyzed properly. Othertechmcal and logis-
tical issues were addressed as they arose. Examples
of issues that were quickly identified or resolved as a
result of these calls include:
• aluminum contamination in aliquot 7 during the
digestion step of the analysis for total aluminum
• incorrect calculations for reporting nitrate and
silica data
• sample overload at one laboratory
• nitrate contamination in aliquot 3
• illegible data reporting
The daily QA contact with each laboratory continued
until sample analysis was completed Preliminary
sample data were obtained either verbally, via com-
puter, or via TELEFAX, the method of data collection
depended on the resources available to the contract
analytical laboratory. The preliminary data were eval-
uated by comparing QA sample values with prelimi-
nary acceptance criteria, calculated from pilot study
data or early QA sample data.
Sampling and Field Laboratory Quality
Control Protocols
The ELSI-I also included (QC) procedures for sam-
pling and analytical activities. Specific procedures
are outlined in the QA plan, the methods manual, and
the field operations report. The flow of samples and
data through the field and analytical laboratories is
illustrated in Figure 2.
Sample Preservation
For each chemical parameter measured during the
ELS-I, it was necessary to identify the procedures
required for sample preservation. The objectives of
sample preservation were to (1) inhibit chemical and
biological activity, (2) prevent changes due to volatil-
ity, and (3) prevent precipitation or adsorption effects.
These considerations led to a sample preparation
process in the field laboratory in which seven pre-
served aliquots were prepared from each bulk (rou-
tine, field duplicate, field blank, or fie Id audit) sample.
The preservation process used for each aliquot is
listed in Table 2.
Filtration through a 0.45Mmmembrane filter was used
to remove suspended particulate and large colloidal
material. This process provided subsamples that rep-
resented the dissolved fraction of analytes. Sus-
pended material was filtered from these aliquots at
the field laboratory because such material may have
been a source of biological activity or, through disso-
lution, of additional analyte. It may also have pro-
vided surface area for adsorption or precipitation of
dissolved analytes which would serve as a transport
mechanism for removing these analytes from solu-
tion. Acid was added to some aliquots to prevent loss
of dissolved analytes caused by precipitation or by
chemical or biological reactions. Storage at 4°C was
specified for aliquot 2 to reduce volatilization of the
solvent, and for aliquots 3 through 6 to reduce biologi-
cal activity.
Field Laboratory Analyses
After the sample preparation and preservation steps
were identified, it was necessary to establish maxi-
mum holding times to assure that the samples were
analyzed before any significant degradation
occurred.
Four parameters (pH, DIG, true color, and turbidity)
were identified as requiring immediate analysis. Use
of the mobile field laboratory for these analyses per-
mitted all measurements to be completed within 16
hours of sample collection.
To assure that reliable measurements of DIG and pH
were obtained, samples for those two field laboratory
measurements were collected and were analyzed in a
closed system within the shortest possible holding
times. This procedure was followed to avoid prob-
10
-------
FIELD BLANK
SAMPLES
FIELD DUPLICATE
SAMPLE
FIELD/LABORATORY
AUDIT SAMPLE(S)
FIELD LABORATORY
DAILY BATCH
OF SAMPLES
1
RELABELING
'
ANALYSIS
(DIG, pH, Turbidity, True Color)
QCCHECK
SAMPLES
BATCH
SAMPLES
TRAILER
DUPLICATE
SAMPLE
DATA
ALIQUOT PREPARATION
ALUMINUM EXTRACTION
PRESERVATION
SHIPMENT TO
ANALYTICAL LABORATORY
ANALYTICAL LABORATORY
ANALYSIS
INTERNAL
QC SAMPLES
BATCH
SAMPLES
LABORATORY
DUPLICATE
LABORATORY BLANK,
MATRIC SPIKE,
QCCHECKSAMP1E
I
DATA
1 |
Figure 2. Flow of Samples and Data Through Field and Analytical Laboratories, Eastern Lake Survey
- Phase I.
ll
-------
Aliquota"
Container Size
Preservation Required
Parameters to be Measured
a Aliquots 2, 3, 4, 5, and 6 were
1 2 3
250mL 10 ml 250 ml
Filtered Filtered
ph<2 MIBK-HQ Filtered
w/HN03 Extract
Ca Total Extractable C1~
A1
Mg F"
K S042~
Na N03~
Mn Si02
Fe
stored in the dark at 4°C.
4567
125 ml 500ml 125 ml 125mL
Filtered Not Filtered Not Filtered
pH<2 Not Filtered pH<2 pH<2
w/H2S04 w/H2S04 w/HN03
DOC pH Total P Total A1
NH4+ ANC
BNC
Conductance
DIG
Table 2. Sample Preservation Requirements.
-------
lems with changes in dissolved inorganic carbon
(DIG) content and thereby to represent accurately the
in situ values.
True color and turbidity were determined in the field
laboratory because the recommended holding time
was 48 hours (U.S. EPA 1983). Turbidity had to be mea-
sured as soon as possible to avoid settling of sus-
pended matter. Both true color and turbidity are sub-
ject to changes due to chemical activity and
adsorption effects.
There is evidence that dissolved aluminum species
change rapidly after sample collection (Driscoll et al.
1983); therefore, it was recommended that the alumi-
num extraction be performed in the field as soon as
possible. It was not possible to perform the extrac-
tion in the helicopter because of concerns about
exposure of the crew to MIBK vapors and the
increased possibility of sample contamination. All
samples were extracted in a laminar-flow hood in the
field laboratory within 16 hours of collection.
Contract Analytical Laboratory Quality
Control Protocols
Analytical methods and quality control protocols
were developed for use by the contract laboratories in
accordance with the ELS-I research plan and the pilot
study results. These methods and protocols are
described below and are outlined in greater detail in
the analytical methods manual and the QA plan.
In general, specific models of instruments were not
required for the contract laboratories although the
analytical methods manual contains particular rec-
ommendations for instruments in some methodolo-
gies. Instrumentation for all of the methods required
some form of calibration. For all methods except con-
ductance, a series of standards were analyzed and a
calibration curve was calculated.
Several requirements for sample concentration
ranges were specified for the analytical methods. The
concentration range of the standards used to calcu-
late the calibration curve was required to bracket the
range of concentrations observed in the samples ana-
lyzed. If the concentration of analyte was at or below
the detection limit, the concentration of the detection
limit QC standard had to be within two to three times
the detection limit. This QC requirement assured that
the reported results were based upon interpolation 16
within the calibration curve and not on extrapolation
outside the curve which could result in significant
error. The calibration required for conductance meas-
urements was dependent upon the type of conductiv-
ity meter used and was either a single point calibra-
tion or was internally set by the factory.
Calibration of laboratory equipment was initially veri-
fied prior to sample analysis by analyzing an indepen-
dent QC sample (either commercially or internally
prepared). If the measured value for the QC sample
was not within established control limits, new cali-
bration standards were prepared, and, where applica-
ble, a new calibration curve was generated. Calibra-
tion was also reverified on a routine basis by
reanalyzing the QC sample after every 10 samples
analyzed and after the last sample for the batch. If the
measured value was not within established control
limits, a new calibration was required, and all sam-
ples after the previous acceptable QC check had to be
reanalyzed.
Contract Analytical Laboratory Sample Holding
Times
A maximum sample holding time (determined from
the time of sample collection to sample analysis,
Table 1) was established for each parameter mea-
sured in the contract laboratories. These holding
times were based upon information from the litera-
ture, the best scientific judgment related to the
defined needs, and the logistical demands and limita-
tions of the ELS-I.
A 7-day holding time was specified for the contract
analytical laboratory measurement of pH, nitrate,
and total extractable aluminum. McQuaker et al.
(1979) reported that the pH of a sample remains sta-
ble for up to 15 days if the sample is kept at 4°C and
sealed from the atmosphere; however, in the EPA
manual Methods for Chemical Analysis of Water and
Wastes (U.S. EPA 1983), it is recommended that pH be
measured immediately after sampling. The same
manual also specifies a 48-hour holding time for
nitrate in unpreserved (not acidified) samples of
water and wastes. Other sources (Williams, 1979;
Quave, 1980) indicate that nitrate is stable for 2 to 4
weeks if the sample is stored in the dark at 4°C. The 7-
day holding time for pH and nitrate was selected as a
limit that was practical with respect to the logistical
constraints of the survey and that was conservative in
relation to the available guidelines.
Barnes (1975) reported that the MIBK-aluminum
extract is stable for several weeks following extrac-
tion. However, for the ELS-I, a 7-day holding time was
specified for total extractable aluminum in order to
minimize potential changes in sample composition
due to volatilization of the MIBK.
A 14-day holding time was selected for ANC, BNC,
conductance, DIG, and DOC; selection of this holding
time was based on U.S. EPA (1983) and on practical
considerations with regard to the logistical design of
the survey. In most cases, the DIG analyses were per-
formed almost simultaneously with the correspond-
ing pH measurements because of specific require-
ments in the analytical laboratory contract.
A 28-day maximum holding time was specified for the
remaining contract analytical laboratory measure-
13
-------
ments. In many cases, especially for the metals, hold-
ing times up to 6 months are acceptable with proper
sample handling and preservation (U.S. EPA 1983).
However, the analytical laboratory contract 17
required that the final data package be submitted
within 35 days after receipt of the samples. A 28-day
holding time was specified as a conservative limit
both to meet that requirement and to protect sample
quality.
Reporting Requirements
As noted above, the contract analytical laboratories
were required to analyze the samples and report the
results within 35 days after sample receipt. The
reporting requirements included the submission of
both analytical results and specific information
related to the maximum sample holding times dis-
cussed above This information was used by the QA
auditors to judge the performance of the laboratory
and the quality of the data. The requirement for a spe-
cific delivery time assured that the data packages
were delivered in time to allow for preliminary review
by the QA auditors and for corrective action, if neces-
sary.
Internal Quality Control
The QA program utilized a variety of QA and QC sam-
ples. The numbers of QA and QC samples used in the
survey were based on the need both to keep program
costs within reason and to provide a maximum
amount of information. The QC samples were used by
field samplers, field laboratory personnel (Morris et
al. 1986), and contract analytical laboratory person-
nel (Drouse et ai, 1986); QA samples were used by the
QA staff to evaluate data quality and to judge overall
field and laboratory performance. Descriptions,
applications, and frequencies of QA and QC samples
are provided in Tables 3 and 4.
Field Laboratory and Contract Analytical
Laboratory On-site Evaluations
Onsite evaluations of the contract analytical labora-
tories and the field stations were performed by a QA
audit team to assure that the sampling and analysis
activities were being implemented as planned. These
evaluations are described in Drouse et al. (1986).
14
-------
Sample Type
Field Blank
Laboratory Blank a
Field Duplicate
Trailer Duplicate a
Contract Laboratory
Duplicate3
Field Audit
Laboratory Audit
a Serves as both a QA
Description
Deionized water (ASTM
Type 1) treated as a
lake sample
Zero analyte standard
Duplicate lake sample
Lake sample, split
Sample aliquot, split
Synthetic sample or
natural lake sample
Synthetic sample
sample and a QC sample.
Application
Estimate system decision limit
and quantitation limit
Identify sample contamination
Estimate overall withm-batch
precision
Estimate analytical
withm-batch precision
Estimate analytical
within-batch precision
Estimate overall among-batch
precision, estimate laboratory
bias
Estimate analytical
among-batch precision,
estimate laboratory bias
Frequency
One per sampling crew per day
One per laboratory batch
One per field station per day
One per field batch
One per laboratory batch
A minimum of one field or
laboratory audit per field batch
A minimum of one field or
laboratory audit per field batch
Table 3. Descriptions, Applications, and Frequencies of Quality Assurance (QA) and Quality Control (QC)
Samples, Eastern Lake Survey - Phase I.
15
-------
Sample Type
Description
Application
Frequency
Trailer Duplicate
Contract Laboratory
Duplicate
Laboratory Calibration
Blank
Matrix Spike
Quality control Check
Sample
Lake sample; split
Sample aliquot, split
Zero analyte standard
Batch sample plus
known quantity of
analyte
Standard, source other
than calibration
standard
Field lab, determine analytical
within-batch precision
Contract lab, determine
analytical within-batch
precision
Field and contract lab; identify
signal drift and sample
contamination
Contract lab, determine
sample matrix effect on
analysis
Field and contract labs,
determine accuracy and
consistency of calibration
One per field batch
One per laboratory batch
One per laboratory batch
One per laboratory batch
Before, after every 10, and after
final sample in batch
Table 4. Descriptions, Applications, and Frequencies of Quality Control Samples, Eastern Lake Survey - Phase I.
16
-------
Section 3
Data Base Quality Assurance
The data base for the ELS-I was managed by Oak
Ridge National Laboratory (ORNL) which has consid-
erate expertise in managing large data bases, in
manipulating data, and in restructuring data bases to
satisfy data analysis needs. The data are stored in
four major data sets: (1) a raw data set of field and
analytical laboratory data, (2) a verified data set, (3) a
validated data set, and (4) an enhanced data set.
Development of the data base began when samples
and data were transferred from the field stations to
the analytical laboratories and continued with receipt
of analytical results and verification reports. The
process was complete after final validation of the
data by ERL-Corvallis. The enhanced data set was
generated from the validated data by using known
relationships between physicochemical parameters
(Linthurst et al. 1986). The enhanced data set provided
a representative summary of sample values for use in
generating population estimates.
Data and Sample Transfer
Data from in situ, field laboratory, and contract ana-
lytical laboratory measurements were transferred to
ORNL and EMSL-LV initially in hardcopy form. Docu-
mentation related to shipment of samples from the
field stations to the contract analytical laboratories
was also transferred in this manner. After the appro-
priate information was entered into the data base, a
magnetic tape containing raw data was sent to the
EPA IBM 3081 computer at the National Computer
Center (NCC), Research Triangle Park, North Caro-
lina. Each tape received by the NCC tape library was
assigned a volume serial number and a BIN number
that indicated the physical location of the tape. The
EMSL-LV QA staff then remotely loaded the tape to
disk files and reviewed the data.
Transfer of Samples and Data From Field Stations
Following sample processing at the field laboratory,
the aliquots were shipped in Styrofoam-lined ship-
ping cartons with frozen freeze-gel packs to maintain
a temperature of 4°C. A 4-part carbonless shipping
form (Form 3, Figure 3) was completed for each ship-
ping container. One copy was retained at the field sta-
tion, one copy was sent to the U.S. EPA Sample Man-
agement Office (SMO), and two copies (sealed in a
plastic bag) were enclosed in the carton. The ship-
ping container was then sealed and shipped by over-
night delivery to the contract analytical laboratory.
Preserved splits of all samples were sent to ERL-Cor-
vallis for chemical analysis by ICPES (Drouse and
Best 1986). Additionally, splits of samples from
selected regions were sent to two Canadian laborato-
ries and to a Norwegian laboratory forchemical anal-
ysis (Stapanian et al. 1986).
When the samples arrived at the destination contract
laboratory, a receiving clerk recorded the date
received on each shipping form, verified that the sam-
ple contents matched those listed on the shipping
form, and completed the "sample condition" portion
of the shipping form. The sample condition notation
included such information as leakage, insufficient
sample, noticeable suspended particulates, partially
frozen samples, and internal temperable conditions
ordiscrepancies in the listed contents. Upon comple-
tion of sample inspection, the receiving clerk mailed
one copy of each shipping form to SMO and retained
the other copy.
In addition to shipping samples, the field laboratory
personnel transferred data forms to various loca-
tions. Copies of the lake data form (Form 1, Figure 4)
and the field laboratory data form (Form 2, Figure 5)
were mailed to the QA manager and to ORNL. Upon
receipt of Forms 1 and 2, ORNL personnel entered the
data from these forms into the raw data set. This data
flow scheme is summarized in Figure 6. Verification
of the field data is discussed below.
Transfer of Contract Analytical Laboratory Data
After all analyses for a single batch of samples were
completed, the contract laboratory personnel pre-
pared an analytical report called a sample data pack-
age. All analytical results were recorded on the forms
listed in Table 5. The laboratory manager was
required to sign each form signifying that he or she
had reviewed the data and that the samples were ana-
lyzed exactly as described in the contract. Any devia-
tions from the contract required the authorization of
the ELS-I QA manager prior to sample analysis. The
data package also contained a narrative description
of any difficulties encountered. All original raw data
were retained by the laboratory manager until
17
-------
NATIONAL SURFACE WATER SURVEY
SAMPLE MANAGEMENT OFFICE
P.O. BOX 8 I 8
ALEXANDRIA, VA 22314
NSWS
FORM 3
SHIPPING
RECEIVED BY
IF INCOMPLETE IMMEDIATELY NOTIFY:
SAMPLE MANAGEMENT OFFICE
(703)557-2490
FROM
(STATION ID):
SAMPLE
ID
0!
02
03
04
05
06
07
08
09
10
1 1
12
1 3
14
15
16
1 7
18
19
20
2 1
22
23
24
25
26
27
28
29
30
TO
(LAB):
3ATCH
ID
DATE SAMPLED
ALIQUOTS SHIPPED
(FOR STATION USE ONLY)
1
2
3
1
4
5
6
7
DATE SHIPPED DATE RECEIVED
AIR — BILL N1"1,
SAMPLE CONDITION UPON LAB RECEIPT
(FOR LAB USE ONLY)
QUALIFIERS:
V: ALIQUOT SHIPPED
M: ALIQUOT MISSING DUE TO DESTROYED SAMPLE
WHITE - FIELD COPY
PINK - LAB COPY
YELLOW - SMO COPY
GOLD - LAB COPY FOR RETURN TO SMO
Figure3. NSWS Form 3 - Shipping.
18
-------
NATIONAL SURFACE WATER SURVEY
FORM1
LAKE DATA
D D M M M Y Y
STATE
LAKE ID
LAKE NAME
LORAN READINGS }
DATE
SAMPLING TIME
HYDROLAB ID
INITIAL
FINAL
INITIAL i
i pH
LATITUDE i n ii i°i ii i.i ii .LONGITUDE i u n i°i 11 i.i n i FINAL. — . i — n — .. — . IJS v--/
PHOTOGRAPHS^)
FRAME ID AZIMUTH
^^ LAP CARD
DISTURBANCES WITHIN 100 METERS OF SHORE
EH ROADS EH LIVESTOCK EH MINES/QUARRIES
EH DWELLINGS EH INDUSTRY EH LOGGING LH OTHER
SITE DEPTH (ft) x 0 3048 m/ft - i n i.i i m
AIR TEMP i ii ii i°C
SITE DEPTH
SECCHI DEPTH DISAPPEAR i
LAKE STRATIFICATION DATA
1 5mV_y
T°C
BOTTOM -1 5m
AT°C (1 5, B-1 5m) i n—I.L
06 DEPTH T°C
IF A > 4° C PROCEED
IF NOT, STOP HERE |
pH
PH
AT°C (1 5, 06 DEPTH) i — n — i.i — i (^J
[ IF AT >4°C FILL IN j
FOLLOWING DATA BLOCK
LAKE DIAGRAM # I 1
Flpvatinn ft Ontlpts LAKE DEPTH
CHECK ONE
TN InlPK D-20m D 20m T°C pS
* 4
6
8
10
12
14
16
18
20
c ,_,,_,, ,Q ,__, ,__,,_, 1_J
10 ^i^.^Q ,_,,_,,_,,_,
15 i_n_i.i iQ i i I—" L— > i— i
20 i — i i — i.i — i \^) i — i i — i i — i ' — i
25 i , i tll i O i i , i , , , i
30 i i i i . i i O i i i i i i i i
35 ,_,,_,. i iQ i— i •—! i— ' i i
40 ^_,^.^_,O ^^^^
45 L-JL^.I—.O L-IU-J ,_,,_,
50 __._0 ^^^^
0
O
o
o
o
o
o
o
o
o
COMMENTS D NOT SAMPLED, SEE BELOW
DATA QUALIFIERS
® INSTRUMENT UNSTABLE
@ REDONE FIRST READING NOT
ACCEPTABLE
© INSTRUMENTS SAMPLING GEAR
NOT VERTICAL IN WATER COLUMN
@ SLOW STABILIZATION
(D HYDROLAB CABLE TOO SHORT
(N) DID NOT MEET OCC
® OTHER (explain)
FLOWING WATER D INACCESSIBLE D NO ACCESS PERMIT DURBAN/INDUSTRIAL
DHIGHCOND (>1500/jS) GNON-LAKE DTOOSHALLOW DOTHER-
FIELD LAB USE ONLY
TRAII FR in
RATP.H in
SAMPI F in
FIELD CREW DATA
P.HFW in
ORSERVER
SAMPLER
nRR SIGN
GROUND CREW MEMBER
SIGN
WHITE COPY— ORNL
PINK COPY— EMSL-LV
Figure 4. NSWS Form 1 - Lake Data.
19
-------
NATIONAL SURFACE WATER SURVEY
FORM 2
BATCH/QC FIELD DATA
DATE RECEIVED
BY DATA MGT
ENTERED
RE-ENTERED
BATCH ID
NO SAMPLES
IN BATCH
BASE SITE ID
SAMPLE ID
01
02
03
04
05
06
07
08
09
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
DUP
FIELD
CREW
ID
LAKE
ID
(XXX-XXX)
LAB TO WHICH
BATCH SENT
DATE SHIPPED
LAB CREW ID
SAMPLE
CODE
TD
DIG (mg/L)
QCCS LIMITS
UCL - 2.2
LCL — 1.8
VALUE
QCCS
STATION pH
QCCS LIMITS
UCL - 4.1
LCL - 3.9
VALUE
QCCS
DATE PRO(
AIR-BILL N
FIELD LAB
SUPERVIS
MESSED
0
ORATORY
->R
TURBIDITY (NTU)
OCCS LIMITS
UCL - 5.5
LCL — 4.5
VALUE
OCCS
COLOR
(APHA
UNITS)
VALUE
SPLIT
CODES
(E,L)
COMMENTS
WHITE — ORNL COPY YELLOW — FIELD COPY PINK — EMSL LV COPY
Figure 5. NSWS Form 2 - Field Laboratory Data.
20
-------
CONTRACT
ANALYTICAL
LABORATORY
2 copies
Form 3
FIELD STATION
(One copy
each of Forms
1, 2, and 3
retained)
1 copy
Form 3
SAMPLE
MANAGEMENT
OFFICE
1 copy
Form 3
ELS-I QA
MANAGER
1 copy each
Forms 1 and 2
ORNL
DATA BASE
1 copy each
Forms 1 and 2
Figure 6. Eastern Lake Survey - Phase I Data Flow Scheme.
21
-------
Data Form
11
13
14"
15b
16b
17
18
19
20
21
22
23
a These forms are shown in Drouse et al.
b Form not required to be submitted with
Description
Summary of Sample Results
ANCand BNC Results
QC Data for ANC and BNC Analyses
Conductance (Measured and Calculated)
Anion-Cation Balance Calculations
Ion Chromatography Resolution Test
Instrumental Detection Limits
Sample Holding Time Summary
Results of Blank Sample and QCCS Analyses
Results of Matrix Spike Analyses
Results of Duplicate Sample Analyses
Results of Standard Additions Analyses
(1986).
data package but recommended for internal QC requirements.
Table 5. List of Data Forms Used by the Contract Analytical Laboratory, Eastern Lake Survey (Phase l)a.
otherwise notified. Raw Data included data system
printouts, chromatograms, notebooks, QC charts,
standard preparation data, and any other information
pertinent to sample analysis. The original data pack-
age was retained by the laboratory, and copies were
mailed to EMSL-LV, SMO, and ORNL The analytical
results were double-entered into the raw data set by
ORNL
Raw Data Set
At ORNL, the field and laboratory data and data quali-
fiers ('tags', Table 6) reported on Forms 1, 2, 11, 13,
and 18 through 23 were directly entered into the data
base using Statistical System software (SAS Insti-
tute, Inc. 1985).
All data were entered independently by two different
operators. A computer program (COMPARE) was
developed to identify any inconsistencies between
the two data sets (Rosen and Kanciruk 1985). The
inconsistencies were then corrected using the SAS
full-screen editing procedure. The purpose of this
double entry and comparison process was to mini-
mize data entry errors.
DATA VERIFICATION
Verification procedures for the raw data set were
developed and implemented by the EMSL-LV QA staff.
The objective of data verification was to identify and
correct, qualify, or eliminate data of unacceptable
quality. Data qualifiers added during the verification
process ('flags') are listed in Table 7. This objective
was accomplished using the following organized
process to examine the data: (1) establish daily com-
munication with the field and analytical laborato-
ries;^) verify completeness and consistency of the
data package on receipt and review any comments or
questions associated with the batch or sample under
evaluation (i.e., tags and narrative comments); (3)
evaluate preliminary QA sample data and routine
sample data; (4) obtain confirmation, correction, or
reanalysis data from the laboratories as needed to
address atypical values; and (5) provide correcting
entries to ORNL for establishing the verified data set.
A computer software package (AQUARIUS) was
developed to automate this procedure as much as
possible (Fountain and Hoff 1985). The AQUARIUS
package was tested during the ELS-I and is being
modified to provide additional verification proce-
dures for other phases of the NSWS.
Data verification procedures are summarized below
and are illustrated in Figure 7. Additional details are
provided in Drouse et al. (1986).
22
-------
Qualifier
Indicates
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
Hydrolab cable too short
Result outside QA criteria (with consent of QA manager)
Result obtained from method of standard additions
Holding time exceeded criteria (Form 19 only)
Result not available, insufficient sample volume shipped to laboratory from the field
Result not available; entire aliquot not shipped
Not analyzed due to interference
Result not available; sample lost or destroyed by lab
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
Anion-cation balance outside criteria due to high DOC
Percent Difference (%D) calculation (Form 14) outside criteria due to high DOC
Available for miscellaneous comments in the field only
Available for miscellaneous comments in the field only
Available for miscellaneous comments in the field only
Measurements taken at <1.5 m
Table 6. Eastern Lake Survey (Phase I) Field and Laboratory Data Qualifiers ('Tags').
Daily Communication
Daily communication was maintained between the
EMSL-LV QA staff and each contract analytical labo-
ratory during the periods when samples were being
analyzed. The objectives of daily communication
were to assure that the laboratory was implementing
the QC requirements and to obtain a preliminary eval-
uation of data quality and laboratory performance.
Performance was judged by reviewing the analytical
results for QA samples (e.g., field blanks, field dupli-
cates, and audit samples). Daily contact enabled the
QA auditors to be familiar with analytical problems
and some results so that data analysis was already
partially underway prior to receipt of the data pack-
age. The bulk of the data evaluation occurred concur-
rently with input of the raw data by ORNL.
Data Receipt
As noted above, the QA staff at EMSL-LV received
copies of the field data forms (Forms 1 and 2) from
each field laboratory coordinator. Upon receipt, the
auditor checked for the following items:
Lake ID. The lake data form (Form 1) was com-
pared with the field laboratory data form (Form 2)
for transcription errors.
Trailer Duplicate. The duplicate lake sample ID
recorded on Form 2 had to match a routine lake
sample ID, and the precision criteria for pH, DIG,
true color, and turbidity between those two sam-
ples had to be achieved.
Hydrolab Calibration Data. The Hydrolab pH and
conductance QCCS values from Form 1 were
compared with the data from the Hydrolab cali-
bration forms to assure that initial calibration
criteria were met or that data qualifiers were
added.
Hydrolab pH. The Form 1 pH value at 1.5 meters
was compared with the field laboratory (Form 2)
pH values. Values were expected to agree within
0.5 pH unit.
Field laboratory pH and DIG. Form 2 values for
field audit samples were compared with accept-
ance criteria. Routine/field duplicate pairs and
23
-------
AO Anion-Cation Percent Ion Balance Difference (%IBD) is outside criteria due to unknown cause
A1 Anion-Cation Percent Ion Balance Difference (%IBD) is outside criteria due to Nitrate contamination.
A2 Anion-Cation Percent Ion Balance Difference (%IBD) is outside criteria due to anion (other than nitrate) contamination.
A3 Anion-Cation Percent Ion Balance Difference (%IBD) is outside criteria due to cation contamination.
A4 Anion-Cation Percent Ion Balance Difference (%IBD) is outside criteria due to unmeasured organic protolytes (fits Oliver
Model).
A5 Anion-Cation Percent Ion Balance Difference (%IBD) is outside criteria due to possible analytical error - anion
concentration too high (list suspect anion}.
A6 Anion-Cation Percent Ion Balance Difference (%IBD) is outside criteria due to possible analytical error - cation
concentration too low (list suspect cation).
A7 Anion-Cation Percent Ion Balance Difference (%IBD) is outside criteria due to possible analytical error - anion
concentration too low (list suspect anion).
A8 Anion-Cation Percent Ion Balance Difference (%IBD) is outside criteria due to possible analytical error - cation
concentration too high (list suspect cation).
BO External (field) blank is above expected criteria for pH, DIG, DOC, conductance, ANC, and BNC determinations
B1 Internal (lab) blank is >2 x RDL for PH, DIG, DOC, conductance, AJ\IC, and BNC determinations.
B2 External (field) blank is above expected criteria and contributed >20% to sample value. (This flag is not used for pH,
DIG, DOC, ANC, or BNC determination.)
83 Internal (lab) blank is >2 x RDL and contributes >10% to the sample concentrations. (This flag is not used for pH,
DIG, DOC, ANC, or BNC determinations.)
B4 Potential negative sample bias based on internal (lab) blank data.
B5 Potential negative sample bias on external (field) blank data.
CO Percent Conductance Difference (%CD) is outside criteria due to unknown cause (possible analytical error - ion
concentration too high).
C1 Percent Conductance Difference (%CD) is outside criteria due to possible analytical error - anion concentration too high
(list suspect anion).
C2 Percent Conductance Difference (%CD) is outside criteria due to anion contamination.
C3 Percent Conductance Difference (%CD) is outside criteria due to cation contamination.
C4 Percent Conductance Difference (%CD) is outside criteria due to unmeasured organic anion (fits Oliver Model).
C5 Percent Conductance Difference (%CD) is outside criteria due to possible analytical error in conductance measurement
C6 Percent Conductance Difference (%CD) is outside criteria due to possible analytical error - anion concentration too low
(list suspect anion).
C7 Percent Conductance Difference (%CD) is outside criteria due to unmeasured protolyle anions (does not fit Oliver
Model).
C8 Percent Conductance Difference (%CD) is outside criteria due to possible analytical error - cation concentration too low
(list suspect cation)
C9 Percent Conductance Difference (%CD) is outside criteria due to possible analytical error - cation concentration too high
(list suspect cation).
DO External (field) duplicate precision exceeded the maximum expected percent relative standard deviation (%RSD), but
either the routine or the duplicate concentrations were >10 x RDL.
D2 External (field) duplicate precision exceeded the maximum expected percent relative standard deviation (%RSD), and
both the routine and duplicate sample concentrations were >10 x RDL
Table 7. Eastern Lake Survey - Phase I Verification Data Qualifiers ('Flags').
24
-------
03 Internal (lab) duplicate precision exceeded the maximum allowable percent relative standard deviation (%RSD), and
both the routine and duplicate sample concentrations were >10 x RDL
FO Percent Conductance Difference (%CD) exceeded criteria when Hydrolab conductance value was substituted.
F1 Hillman-Kramer protolyte analysis program indicated field pH problem when Hydrolab pH value was substituted.
F2 Hillman-Kramer protolyte analysis program indicated unexplained field pH or PIC problem when Hydrolab pH value was
substituted.
HO The maximum holding time criteria were not met.
H1 No "Date Analyzed" data were submitted for reanalysis data.
L1 Instrumental Detection Limit (IDL) exceeded RDL and sample concentration was <10 x IDL.
MO Value was obtained using a method which is unacceptable by the contract.
NO Audit sample value exceeded upper control limit.
N1 Audit sample value was below control limit.
N2 Audit sample value exceeded control limits due to suspect audit sample preparation.
N5 N03~ data obtained from analysis of aliquot 5.
PO Field problem - station pH.
P1 Field problem - station DIG.
P2 Field problem - unexplained pH or DIG.
P3 Lab problem - initial ANC pH.
P4 Lab problem-initial BNC pH.
P5 Lab problem - unexplained - initial pH (ANC or BNC).
P6 Lab problem - initial DIG.
P7 Lab problem - air-equilibrated pH or DIG.
P8 Lab problem - unexplained - initial pH or DIG.
P9 Lab problem - ANC determination.
Q1 Quality control check sample (QCCS) was above contractual criteria.
Q2 Quality control check sample (QCCS) was below contractual criteria.
Q3 Number of quality control check samples (QCCS) measured was insufficient.
Q4 No quality control check sample (QCCS) analysis was performed).
SO Matrix spike percent recovery (%REC) was above contractual criteria.
S1 Matrix spike percent recovery (%REC) was below contractual criteria
V Data verified.
Table 7. Eastern Lake Survey - Phase I Verification Data Qualifiers ('Flags') (Concluded).
25
-------
Field Data
Analytical Data
Preliminary
Acceptance
Criteria in
QA Samples
LEMSCO
Preliminary
Review
Fix Data
Initiate
Corrections
Yes
Generation of
Raw Data Set
Daily Communications
With the Field and
Analytical Laboratory
Generate Performances
Based on Acceptance
Criteria
Magnetic Tape
Mailed From
ORNLtoNCC-IBM
Automated QA Review
1. Identify Exceptions
to QA Criteria
2. Preliminary Statistics
to Identify Outliers
±.
Fix Mistakes
and Set Flags
Using SAS Full-Screen
Editor
Apply
Correcting
Entries to Raw
Data Set
Magnetic Tape Mailed
From NCC-IBM to
ORNL
Generation of
Verified Data
Set
No
Yes
Figure 7. Data Verification Process for the Eastern Lake Survey - Phase I.
26
-------
routine/trailer duplicate pairs were evaluated for
precision.
• pH and DIG QCCS Data. Form 2 QCCS data were
compared with acceptance criteria.
Data anomalies were reported to the field laboratory
coordinator for review. Continual review by the field
laboratory staff minimized the number of data tran-
scription errors encountered by the EMSL-LV QA
staff. Data reporting errors were reported to ORNL for
correction before values were entered into the data
base. All telephone communications were docu-
mented in a bound notebook. Data changes were
annotated on the appropriate form.
The contract laboratory data packages were deliv-
ered to the EMSL-LV QA staff upon completion of
analysis or within 35 days following sample receipt.
As they were received, data packages were reviewed
for completeness, internal QC compliance, and
appropriate use of data qualifiers. A data package
completeness checklist was used by each EMSL-LV
QA auditor to assure consistency in the review of all
data packages (Figure 8). Problems were reported to
the appropriate contract laboratory manager for cor-
rective action. Comments provided by the laboratory
with the data package were also reviewed to deter-
mine their impact on data quality and the need for any
followup action by the laboratory. Completion of this
checklist was important in verifying that the labora-
tory had met all contractual requirements for the pur-
pose of payment.
Quality Assurance and Quality Control Data
ORNL personnel entered the data into the raw data
set as the data packages were received (see Section
3). During the data entry period, QA sample data were
evaluated by the EMSL-LV QA staff to establish per-
formance-based acceptance criteria. The review
process utilized the computer programs listed in
Table 8 to identify or flag results that were exceptions
to the expected QA-QC limits. These programs auto-
mated much of the QA review process and enabled
the auditor to concentrate more effort on the correc-
tion or flagging of questionable data.
The QA auditor used the output from these programs
(along with the original data and field notebooks) to
evaluate the data and to complete the NSWS verifica-
tion report contained in the QA plan. The verification
report was actually a worksheet designed to system-
atically guide the auditor through the verification
process. It listed procedures for qualifying ('flagging')
data and for tracking both requests for confirmation
reanalysis and for data resubmissions. It also listed
the steps used to help explain the QA exceptions and
to summarize all modifications to the raw data set.
These auditing procedures are described in detail in
Drouse et al. (1986).
After being entered into the raw data set, the data
were reviewed on a batch basis by making use of both
computer programs and manual review. These efforts
checked internal consistency (e.g., ion balance differ-
ence and conductance difference) for each sample
and the acceptability of both the QA sample data and
the laboratory QC data for each batch.. Examination
of univariate distributions was also performed on QA
sample data to identify statistical outliers and to
establish or update performance acceptance criteria.
Samples which met acceptance criteria for these
checks were transferred into the verified data set.
When exceptions could be explained by the presence
of organic compounds based on the Oliver et al. (1983)
model or on a correctable reporting error, these val-
ues associated with these samples were qualified or
corrected and entered into the verified data set. This
was accomplished through the Hillman-Kramer Pro-
tolyte Analysis Program as described in Drouse et al.
(1986). When exceptions to the ioin balance differ-
ence or conductance difference criteria could not be
explained by calculated organic ion concentrations,
that sample was reanalyzed to determine if the result
was due to reporting or analytical error.
Suspected analytical errors were referred to the ana-
lytical laboratory for reanalysis. Acceptable values
from reanalysis were qualified and substituted for the
original values in the verified data set. For each
parameter, samples that were not analyzed within
maximum allowable holding times or that were asso-
ciated with unacceptable QC or QA sample results
were flagged before entry into the verified data set.
When the QA sample for a given parameter did not
meet the acceptance criteria, that parameter was
flagged for all samples in the batch. A parameter was
also flagged when internal QC checks were not met.
Those checks included matrix spike recovery, calibra-
tion and reagent blank analysis, internal duplicate
precision, required instrumental detection limit,
QCCS percent recovery, and maximum allowable
holding times. In all cases, each flag generated by the
computer was reviewed by the QA auditor for reason-
ableness and consistency before it was added to the
data base.
Less than 3 percent of the raw data reported for lake
samples was classified as reporting errors and was
corrected before transfer to the verified data set.
Sample reanalysis was requested for less than 4 per-
cent of the originally reported raw data values. Less
than 1 percent of the reported data required correc-
tion because of transcription or data entry errors. The
overall error rate for data entry and updating of the
raw data set to the verified data set was estimated to
be less than 0.03 percent.
27
-------
Lab Name:
Date:
NATIONAL SURFACE WATER SURVEY
Data Package Completeness Checklist
Batch ID:
Page 1 of 2
Auditor's Initials:
All major difficulties during analyses have
been discussed with the QA manager or designee.
Anion-cation balance and conductance balance
checks exceeding criteria are reported on cover
letter.
a. Required forms (11, 13, 17, 18-23) submitted.
b. Lab name, batch ID, and lab manager's
signature submitted on all forms.
c. Sample ID reported on Forms 13, 21, 22, and 23.
d. Analyst's signature on Form 13.
e. Correct units indicated on all forms.
Form 11:
a. Correct number of samples analyzed and
results for each parameter tabulated.
b. Data qualifiers (J, K, M, or U) reported when
results are missing.
c. Data qualifier R is reported when a sample
is reanalyzed for QC purposes.
d. F is reported as a data qualifier when a
result is outside criteria (with consent
of QA manager).
e. G is reported as a data qualifier when the
method of standard additions is used and
Form 23 is submitted.
f. ANC initial pH and BNC initial pH are within
±0.1 pH unit.
Percent 1C resolution reported as greater than 60%
on Form 17.
Form 18:
a. Instrumental detection limits and associated
dates of determination tabulated.
b. Instrumental detection limits less than or
equal to the required detection limits
(Table 1-1).
Form 19:
a. Date sampled, date received, holding time
plus date sampled, and dates of analyses for
the correct number of samples are tab-
ulated.
Yes
Par-
tial
No
Comments
(continued)
Figure 8. NSWS Data Package Completeness Checklist.
28
-------
Page 2 of 2
10.
11.
b. Date analyzed is less than or equal to the
reported holding time plus date sampled.
c. The data qualifier H is reported for dates
of analyses which exceed the holding time
plus date sampled with consent of QA manager.
Form 20:
a. Calibration blanks, reagent blanks, and DL
QCCS are reported where required.
b. Calibration blanks and reagent blanks are
less than 2 times the required DL.
c. DL QCCS is approximately 2 times the required
DL and the measured values are within 20% of
the theoretical values.
d. QCCS true values are in the midrange of
sample values.
e. If high QCCS true values are reported, the
samples analyzed on high range are discussed in
the cover letter.
f. Diluted samples and their dilution factors
are discussed in the cover letter.
Percent recovery of matrix spikes is reported
on Form 21 for each required analysis, and
the values are within the range of 85% to 115%.
Duplicate precision results are reported for
each parameter and are less than or equal
to the maximum %RSD (Table 1-1).
Standard additions are performed and Form 23 is
submitted when the matrix spike analyses do
not meet contract requirements.
Yes
Par-
tial
No
Comments
Note: Checklist is not required in data package but is recommended
to be included in the raw data.
Figure 8. (continued).
Figure 8. NSWS Data Package Completeness Checklist (Concluded).
29
-------
Title
Sample Type
Audit Sample Summary
Lab and File Blank Summary
Field Duplicate Precision Summary
Instrumental Detection Limit Summary
Holding Time Summary
DIG Check Calculations
ANC Check Calculations
Conductivity Check Calculations
Anion-Cation Balance Calculations
Batch QA-QC Summary
Comparison of Form 1 and Form 2
Comparison of Form 2 and Form 11
Protolyte Analysis - (DIG, DOC, pH, ANC, and BMC Data Evaluation)
Audit Sample Window Generation
Raw Data Listing - Format for QA Manager
Complete Raw Data Listing - Format for Audit Staff
Reagent-Calibration Blanks and QCCS
Calculation of Laboratory Penalties
Matrix Spike Summary
(LH, LL, FH, FL, FN)
(B, LB, FB)
(R, D Pairs)
(All)
(All)
(All)
(All)
(All)
(All)
(All)
(pH and DIG)
(pH and DIG)
Table 8. Exception Generating and Data Review Programs, Eastern Lake Survey (Phase I).
Follow-up with Contract Analytical Laboratories
Completion of Step 2 (data receipt) and Step 3 (QA-QC
review of data) included communications with the
appropriate contract analytical laboratory. This fol-
low-up (Step 4) was needed to obtain corrections to
the data package, confirmation or correction of
reported data, and sample reanalysis when required.
Step 4 was the most difficult and time-consuming
process in data verification, especially if the requests
to the laboratory were not clearly supported by con-
tract requirements. Typically, responses to requests
for confirmation or correction of reported data were
completed within 2 to 4 weeks. Reanalyses were
either completed within 2 to 3 months or were not per-
formed because of negotiations with the contractors
for additional payment. Although every effort was
made to identify samples for reanalysis within the
maximum allowable holding times, the contract labo-
ratory often had to choose between samples in pro-
gress and samples for reanalysis. Therefore, many
reanalyses, especially for nitrate in aliquot 5, were
performed outside the maximum allowable holding
time.
Preparation and Delivery of Verification Tapes
After the first four verification steps were completed
by the EMSL-LV QA staff, the data were adequate for
establishing the verified data set. In order to translate
the raw data set into the verified data set, a method
forquickly transferring the information from EMSL-LV
to ORNL was required. Changes to the raw data set
were made using computer entries called tuples. A
tuple is an ordered group of elements (e.g., batch ID,
sample ID, variable, old flag, new flag, old value, new
value). For the ELS-I, tuples identified changes into a
data set such as adding, changing, or deleting sam-
ple values or data qualifiers. A tuple was also created
when a computer program generated a flag for a spe-
cific parameter.
The system that was used for modifying the ORNL
raw data set was designed to minimize data entry
errors. This system used separate areas for each pro-
30
-------
gram-generated tuple and each manual tuple (value
change or deletion), which facilitated searching,
modifying, and checking of tuple listings. When a
tuple listing was ready to send to ORNL, a computer
program (Database Administrator Program) was used
to combine all of the tuple areas including flags, tags
(data qualifiers added to a value in the field), and
value changes, and to append the listing to the data
base. This listing included only those tuples for which
the batch ID, sample ID, variable name, and old value
from EMSL-LV matched those from ORNL. The com-
bined tuble was then written on magnetic tape and
mailed to ORNL from the NCC, RTP, North Carolina.
ORNL processed the combined tuble listing and
returned a magnetic tape to RTP with a listing of
unmatched tubles which were mistakes that could
not be applied to the data base until they were cor-
rected. This cycle took about 10 days.
The overall outcome of the five steps described above
was a verified data set in which all values which did to
meet criteria were flagged, replaced with either cor-
rected or reanalyzed data or replaced with missing
value codes (Table 9). Confirmation (C) and reanalysis
(R) codes were removed from the data set when the
correct values were received from the contract labo-
ratory.
Data Validation
The data validation process identified potential
errors in chemical analyses that could not be
revealed by verification procedures. These values
included potential outliers and systematic errors
because data were evaluated on a regional basis
(Table 10). The quality of non-chemical variables was
also evaluated during the validation process (Table
11).
Data validation was a joint activity performed by ERL-
Corvallis and EMSL-LV. While interpreting the results,
the EPA data users identified questionable data
ponts (Figure 9). The atypical data points were
reported to the EMSL-LV QA staff who then reviewed
the data packages and QA information to reverify the
quality of that particular value. The final decisions
regarding data quality were made by the EPA data
users at ERL-Corvallis on the basis of all available
information.
The data validation process included the identifica-
tion of possible outliers and the evaluation of possi-
ble systematic error in the measurement process.
Both of these aspects were exploratory (as opposed
to test-oriented). Thus, the validation methods
stressed visual presentation and subjective,
although conservative, data selection procedures.
The objective was to identify data values or sets of
data values that warranted special attention or cau-
tion when used for analysis of survey results when
used for model-building based upon survey data. The
methods selected for detection of outliers and sys-
tematic errors were chosen for simplicity of computa-
tion by using pre-existing software whenever possi-
ble (see Figure 9). Data validation is discussed in
greater detail in Drouse et al. (1986) and the Eilers et
al. (1986).
A - Carbonate alkalinity, C02-Acidity and mineral acidity data are eliminated from data base due to method inconsistencies.
C - Temporary flag indicating raw data incomplete pending CONFIRMATION by analytical laboratory.
N - Eliminate from data base pending review of aliquot 5 nitrate data.
R - Temporary flag indicating raw data incomplete pending REANALYSIS.
X - Permanent flag indicating IMVALID data based on QA review.
- Value never reported.
(NOTE: These codes appear in [nuJNUMERIC fields only.)
Table 9. NSWS Missing Value Codes.
31
-------
Conductance (contract analytical laboratory
AND
pH (field laboratory)
DOC
True color
Turbidity
Ca
Na
DIG (equilibrated)
Al (total)
vs. Conductance (field)
Calculated conductance
Sum of cations
Sum of anions
Ca
Ca plus Na
pH (field laboratory)
ANC
DIG
vs. Ca
pH (all measures)
Sum of cations
Sum of anions
Calculated ANC [sum of base cations minus (S042~ and Cl
vs. pH (all measures)
Sum of cations
Sum of anions
Al (extractable)
S042~ (expressed as percent of total anions)
vs. True color
Anion deficit
Secchi disk transparency
Turbidity
vs. Anion deficit
Secchi disk transparency
Turbidity
vs. Secchi disk transparency
vs. Wig
Si02
vs. Cf
vs. DIC (initial)
vs. Al (total extractable)
Table 10. Data Validation for the Eastern Lake Survey - Phase I: Comparison of Variables Used to Check for
Random and Systematic Errors.
32
-------
Variable General Description of Validation Checks
Latitude, Lake location as measured by LORAN was compared against location on U.S.G.S. maps.
Longitude
Lake Elevation, Lake and watershed characteristics were checked against state records, where available, to
Lake Area, confirm lake identification.
Watershed Area,
Site Depth,
Stream Inlets and Outlets,
Lake Hydrologic Type
Shoreline Land Use Compared against aerial photographs
Water Temperature Compared against range of appropriate temperature
Secchi Disk Transparency, Compared with each other for internal consistency
True Color,
Turbidity
Table 11. Physical Parameters Subject to Validation, Eastern Lake Survey - Phase I.
33
-------
c
Data Set No. 2
(Verified)
UNIVARIATE
• Box Plots
• Probability Plots
MULTIVARIATE
• PCA
• Cluster Analysis
• Trilinear Plots
• MLR
Outliers
BIVARIATE
• Scatter Plots
• Regression
Relational
Comparative
Modify or
Delete Value
Yes
/ValueV
\ in /
\Error/
Systematic
Differences
(
U
Hag or
flodify Values
No
No
Data Set
No. 3
(Validated)
Figure 9. Data Validation Process for the Eastern Lake Survey - Phase I.
34
-------
Section 4
Results
Evaluation of the quality assurance and quality con-
trol data was an ongoing process during and follow-
ing the ELS-I. A substantial part of this evaluation
process involved the statistical analysis of the QA
and QC sample results which are presented below
and in Appendix A. Evaluations were also made of the
field and laboratory operations, the verification pro-
cedures, and the analytical methods. Results of the
chemical analyses of samples split between ELS-I
analytical laboratories and ERL-Corvallis, Canadian,
or Norwegian laboratories are summarized in Drouse
and Best (1986), Stapanian et al. (1986), and Yfantis et
al. (1986).
Operations Evaluation
Overall, operation of the ELS-I proceeded smoothly.
The QA-QC program was strictly adhered to through-
out the period of operations. Some specific problems
were encountered that were expected for a study of
this magnitude, and these were usually detected and
resolved quickly. Some protocol changes were imple-
mented during the survey; others were made after the
survey as a result of debriefing and evaluation recom-
mendations. All changes were incorporated into the
final QA plan (Drouse et al. 1986) and were imple-
mented during subsequent NSWS operations.
The ELS-I was completed in a timely manner, and data
of known quality were collected consistently through-
out the period of operations. The statistical require-
ments for adequate sample size were achieved at all
field stations (Linthurst et al. 1986). There were-no
major interruptions in field operations due to acci-
dents, weather, or equipment failure. The sampling
and laboratory protocols were very successful for
most procedures and should serve as a model for
future field studies of a similar nature.
Lake and Sample Information
Field operations were a successful means for obtain-
ing samples and field data consistent with the ELS-I
research plan. Approximately 90 percent of the lakes
initially selected for sampling were visited by sam-
pling crews, and of those lakes visited, 96 percent
(1,612) were actually sampled. In addition to those
lakes, 188 special interest lakes were sampled for a
total sample of 1,800 lakes. The numbers of regular
lakes selected, visited, and sampled during the ELS-I
are listed by subregion in Morris et al. (1986).
Less than 20 percent of the lakes visited were sam-
pled at a depth other than the one originally specified.
Only 5 percent of the lakes sampled were thermally
stratified; thus, 95 percent of all samples were
acceptable in terms of the research plan requirement
that a single water sample was to be collected from
each lake during a period when the lakes were iso-
thermal.
In total, 2,389 routine, field duplicate, and field blank
samples were delivered from the field laboratories to
the contract analytical laboratories (Table 12). This
resulted in 2,639 sets of analyses because the con-
tract required that one matrix spike and one labora-
tory dupicate be analyzed per batch of samples. The
distribution of samples by field station and analytical
laboratory is provided in Table 13.
Field Problems and Their Resolutions
In general, the sampling teams performed as planned
during the ELS-I with an average yield of 20 processed
samples per operating day from each field laboratory.
Three problems were identified and corrected during
field operations. These concerned an inaccurate pH
meter, contamination of blanks from one field labora-
tory, and destruction of one set of split samples. The
identification of the problems and the corrective
actions taken are discussed below.
Prior to the beginning of routine lake sampling from
the Duluth and Rhinelander field stations, a practice
sampling excursion was undertaken. Samples col-
lected during that excursion were split between the
Duluth and Rhinelander field laboratories for proc-
essing and chemical analyses. When the data were
evaluated, it was found that the Duluth pH measure-
ments differed from the Rhinelander measurements
by one pH unit at pH values greater than 6.0. Measure-
ments of the buffers used for calibration and the
QCCS were found to be accurate. After changing the
pH electrodes, checking the buffers, and reviewing
the measurement technique, the problem was eventu-
ally traced to a faulty electronic display on the Duluth
pH meter. A replacement meter was obtained and was
used throughout the survey. The problem with the
Duluth pH meter led to implementation of an addi-
35
-------
Laboratory
Versar
EMS!
Global
USGS-CO
Total
a Includes analysis of matrix spikes
Samples Received
872
983
332
202
2,389
and duplicates (11% lab QC)
Samples Analyzed3
970
1,081
366
222
2,639
Table 12. Number of Samples Received and Analyzed by Contract Laboratories, Eastern Lake Survey - Phase I
tional, mid-level QC check at all field stations. For
each batch of samples, the laboratory supervisor was
required to compare the Hydrolab pH measurements
with the field laboratory measurement and to take
corrective action if the two values differed by more
than 0.5 pH unit.
Some field blanks associated with the Lexington field
station were found to have high levels of several
major cations and anions. Further investigation
revealed that, because of failure of the reverse osmo-
sis-deionization (RO-DI) water system in the field lab-
oratory, the field personnel had been instructed to
take distilled water for two batches from the EPA
regional laboratory still. Each field station was sub-
sequently informed of the problem and cautioned to
use only field laboratory RO-DI water for field blanks.
No routine samples were lost during the ELS-I
although one batch of samples was temporarily mis-
routed in shipment to the contract analytical labora-
tory. These samples were located using the sample
tracking procedures forthe survey and were analyzed
within the required sample holding time. The Norwe-
gian split samples collected from Region 1E were
inadvertently destroyed by the commercial courier
service. A second set of split samples was collected
from Region 3A for shipment to Norway (see Section
1).
Several general recommendations to improve field
operations of future NSWS activities were obtained
from summaries provided by each EPA base coordina-
tor. Many of the temporary employees hired by Lock-
heed-EMSCO as field sampling and field laboratory
Field Station
Bangor
Lexington
Lake Placid
Mt Pocono
Duluth
Rhinelander
Asheville
Lakeland
Total Samples
Versar
—
77
—
99
343
353
—
—
872
Analytical Laboratories
EM SI Global
213
332
332
106
54
116
162
—
983 332
USGS-CO Total
213
409
332
205
397
469
162
202 202
202 2,389
Table 13. Number of Samples Delivered by Field Stations, Eastern Lake Survey - Phase I.
36
-------
personnel also provided debriefing letters before or
shortly after the ELS-I was completed. In addition, all
field laboratory notebooks were turned over to the QA
manager. A debriefing was held for all EPA base coor-
dinators and duty officers and members of the NSWS
management team in Plant City, Florida, in December
1984. Topics from that meeting related to the quality
assurance program included communications, sam-
ple shipment, and training. These topics are dis-
cussed below.
Efficient and complete information transfer between
the NSWS management team and the field stations
was necessary to ensure that all new developments
or modifications in operational protocols were con-
sistently disseminated and understood.
Shipment of samples to the contract analytical labo-
ratories was a problem on weekends especially when
there was no service by overnight courier. Weekend
shipments using commercial airlines required close
coordination between the field laboratory and con-
tract laboratory personnel to assure that the samples
were received at the contract laboratories within the
appropriate time. The contract laboratories later sug-
gested holding weekend shipments until Monday,
since it was easier for them to meet the holding time
constraints than to pick up samples at the airport.
More comprehensive instruction to field samplers in
the completion of field data forms was recommended
to ensure clarity and consistency. It was suggested
that training at each field station be lengthened,
since some field sampling personnel lacked experi-
ence and an understanding of limnology and certain
types of sampling equipment. At some field stations,
state or regional EPA sampling personnel were
rotated and were replaced by new people on a fre-
quent basis. It was suggested that this practice be
discouraged if possible because it added to inconsis-
tencies in data reporting and required that additional
time and effort be spent on training activities.
Contract Analytical Laboratory Problems and
Resolutions
During the ELS-I, there was one occurrence of a sam-
ple volume overload at one contract laboratory. This
problem was immediately resolved by distributing
some samples to another contract laboratory.
Several analytical problems were identified and cor-
rected during contract laboratory operations. Two
significant difficulties centered on the calibration
method used with silica measurement at one labora-
tory and aluminum contamination problems at three
laboratories. These are discussed below.
The daily QA checks and audit sample data compari-
sons by the QA staff indicated that the silica results
from one contract laboratory had a negative bias.
Field blank data with highly negative values were also
being reported by the same laboratory. The bias was
eventually traced to a calibration problem. The other
contract laboratories did not experience the same dif-
ficulty. The laboratory in error was calibrating for sil-
ica in the 10 to 60 ppm range which was specified for
the method. However, the silica method also speci-
fied that the calibration standards bracket the
expected sample concentrations which were well
below the 10 to 60 ppm range. The laboratory involved
calibrated its instruments for silica analyses from 0
to 10 ppm thereafter and reanalyzed the affected sam-
ples. Data from reanalysis for silica were thoroughly
reviewed for technical merit before substitution into
the raw data set.
During the pilot study, laboratory blank values were
reported for total aluminum that were equal to or
greater than the values found in routine samples. The
high aluminum concentrations, believed to be caused
by either airborne dust contamination, reagent and
glassware contamination, or both, occurred at two
laboratories. The major source of contamination was
traced to the use of borosilicate glassware which
containsaluminumoxide. Duringthedigestion proce-
dure aluminum was being leached from the glass.
Teflon beakers were substituted for glass in each lab-
oratory.
The initial sample batches analyzed at one laboratory
also exhibited sporadic aluminum contamination.
Through the daily QA contact with this laboratory, the
problem was identified and was quickly traced to a
building maintenance operation. The floors in an
adjacent room had been sanded during the weekend
while several sample batches were being digested for
measurement of total aluminum. As a result, all labo-
ratories were instructed to improve contamination
control procedures by performing digestions in a
laminar-flow hood or in an isolated station within a
standard fume hood to prevent airborne contamina-
tion. This procedure was implemented at all partici-
pating laboratories.
Data Verification Problems and Resolutions
Several instances of misreported data were uncov-
ered through daily QA checks and the data verifica-
tion process. Although uncommon, errors were iden-
tified and traced to data transcription errors,
switched aliquots in the field laboratory, mislabeled
samples, and data entry errors at the ORNL data cen-
ter. Through careful QA evaluation and follow-up with
all participants, the incidents were identified and cor-
rected.
As discussed in Section 3, general problems existed
with the system used for modifying the ORNL raw
data base. The system was slow; the entire cycle took
about 10 days. A second system was implemented for
a short time which was faster (3 days) but required
manual entry error checking and, thus, was labor-
37
-------
intensive. The modification procedure has since been
redesigned to facilitate error detection and data
tracking while providing an efficient turnaround time.
Methods Evaluation
Several analytical questions arose during the plan-
ning for the ELS-I; others arose as problems during
the pilot studies. Several methods studies were con-
ducted at EMSL-LV to address and resolve these
issues. Questions that arose during the planning
stages included the practicality of measuring free
dissolved fluoride, the relative merits of ion chroma-
tography (1C) versus ion-selective electrodes (ISE) for
measurement of total dissolved fluoride, and the
accuracy of Gran analysis for determination of ANC
and BNC. These issues were evaluated during and fol-
lowing the pilot study. Other issues that were identi-
fied during the pilot study concerned the procedure
for measuring total extractable aluminum, the design
of the pH sample chamber, the effects of filtration on
iron and aluminum in the audit samples, and nitrate
contamination in blank samples. These methods
evaluations are discussed below.
eluteatthesametimeasfluoride(Hillmanet al. 1986).
It was concluded that despite its advantages 1C is not
a suitable method for measuring total dissolved fluo-
ride in natural waters under standard 1C conditions.
Total dissolved fluoride was therefore measured with
ISE during the ELS-I.
Acid-Neutralizing Capacity and Base-Neutralizing
Capacity
In order to maximize the accuracy and precision of
ANC and BNC determinations, a full Gran analysis
was specified for interpreting titration data. Because
Gran analysis is not a standard technique for measur-
ing ANC and BNC and because complex calculations
were involved, several samples were analyzed prior to
the ELS-I to ensure that the calculations were correct
and to provide detailed examples to the contract labo-
ratories. These examples were included in both the
IFB and the analytical methods manual. Technical dif-
ficulties were encountered with the interpretation of
BNC titration data which are presently unresolved
This parameter is not discussed further in the present
report.
Fluoride Determinations
Prior to the start of the ELS-I, free dissolved fluoride
was chosen as one of the parameters of interest. It is
an important parameter to consider in modeling alu-
minum chemistry, but it is difficult to measure. No
satisfactory method could be found in the literature,
and none was recommended by the analytical
experts at a workshop which was held in Denver, Col-
orado, during the planning stage of the ELS-I.
On a trial basis, a simple ion-selective electrode
method was used during the pilot study. Pilot study
results for free dissolved fluoride tended to be varia-
ble and often were higher than those for total dis-
solved fluoride. Also, the measurement was very
time-consuming. Based on the pilot study results,
free dissolved fluoride was dropped from the list of
parameters for the NSWS until an acceptable method
for the determination could be developed.
At the Denver Analytical Workshop, two methods
were suggested for the determination of total dis-
solved fluoride: ion-selective electrode (ISE) and ion
chromatography (1C). The ISE method is a standard
analytical technique, but the electrode is character-
ized by low sensitivity and slow response especially
when the fluoride concentration is below 0.1 mg L .
The 1C method is not commonly used for fluoride
determinations, but it is very sensitive (detection
limit <0.005 mg L~1), rapid (analysis time less than 8
minutes), and easily automated. It is also possible to
simultaneously measure anions in addition to fluo-
ride using 1C. However, the determination of fluoride
by 1C is subject to interferences such as the "water
dip" and measurement of small organic species that
Total Extractable Aluminum
Although the method specified for the determination
of total extractable aluminum (Al) is based on meth-
ods in the literature (Barnes 197979), questions arose
during the pilot study regarding the effects of filtra-
tion and filter type (brand and composition), sample
temperature, and extraction technique. Estimates of
precision based on both single- and multi-analyst
measurements were desired. In order to obtain this
information, a series of experiments was performed
using the natural audit sample from Big Moose Lake.
The effect of filter type on analytical results was
examined by determining total extractable Al in an
unfiltered sample and in three filtered samples, each
obtained using a different type of membrane filter.
The analytical results indicated that neither filtration
(0.45-fj.m membrane) nor filter type had an effect on
total extractable Al concentrations. The membrane
filter was selected for use during the ELS-I.
The question of sample temperature arose because
an effort was not made during the pilot study to
ensure that samples were at the same temperature
prior to Al extraction. If sample temperature were
important, then the comparability of results among
lakes would be affected. To determine whether tem-
perature had an effect on total extractable Al concen-
tration, portions of a sample were equilibrated at two
different temperatures (4°C and 20°C) and then
extracted. The results of this experiment indicated
that sample temperature had no effect on total
extractable Al concentration.
The experimental protocol for total extractable Al
called for a rapid extraction with 8-hydroxyquinoline
38
-------
into methyl isobutyl ketone (MIBK). Because the com-
plexation time and the method for this extraction
were not specified, it was necesary to estimate the
effect of small variations in the extraction technique.
Portions of the natural audit sample were extracted
using two mixing styles (vigorous shaking and no
deliberate shaking after hydroxyquinoline addition)
and two complexation times (5 seconds and 20 sec-
onds). The results indicated that equivalent results
were obtained for each of the experiments, i.e., that
small variations in extraction technique did not affect
measured sample concentrations.
The lack of effect from variations in extraction tech-
nique was seen when determining single- and multi-
analyst precision. The mean concentrations and
standard deviations of total extractable Al measure-
ments by a single analyst (0.192 ± 0.011 mg L~1,
n = 24) and by several analysts (0,193 ± 0.010 mg L~1,
11 analysts, 2 extractions each) are essentially identi-
cal. This experiment demonstrated that, with ade-
quate personnel training, total extractable Al results
were not affected by any among-analyst variations in
extraction technique.
pH Sample Chamber
During the ELS-I, sealed syringe samples were
obtained at each lake for pH determinations. Sealing
samples in syringes was expected to preserve sample
pH by minimizing exchange of dissolved C02 with
the atmosphere. A sample chamber was also devel-
oped for the field laboratory pH measurements to pre-
vent exposure of the sample to air (Hillman et al.
1986).
Formation of Filterable Iron and Aluminum
Complexes in Synthetic Field Audit Samples
During the pilot study, concentrations of dissolved Fe
and total extractable Al concentrations in field syn-
thetic audit samples were consistently lower than
those in laboratory synthetic audit samples. The pri-
mary difference between laboratory synthetic and
field synthetic audit samples was that laboratory
audits were processed (acidified for Fe analysis and
extracted into MIBK for Al analysis) immediately after
preparation without filtering, and field audits were
processed (with filtration) 24 to 72 hours after prepa-
ration. The probable cause for lower concentrations
of Fe and total extractable Al in the field audit sam-
ples was that, during the time delay, hydroxides or
other complexes were formed that did not pass
through the 0.45-Mm membrane filter.
An experiment was performed to determine the cause
of the low concentrations of Fe and total extractable
Al in field audit samples. Three aliquots each from
both the low and high field synthetic audit samples
were acidified to pH 2 with analytical-grade nitric
acid and analyzed for Fe and total extractable Al.
Three additional aliquots from each audit sample
were similarly prepared, except that they were ana-
lyzed after being filtered through a 0.45-fxm membrane.
The analytical results suggested that total extract-
able Al was lost from the field audit samples during
filtration. The purpose of sample filtration was to
remove suspended materials which at pH 6 to 8 may
have included hydroxyaluminum complexes. This
explanation is consistent with the observation that
laboratory audit values for total extractable Al were
higher than field audit values. By extracting the labo-
ratory audit sample immediately after preparation
(within 1 hour), the formation of large polymeric Al
species may have been avoided, and, hence, extract-
able Al was not lost from the samples.
Furthermore, Fe was not detected in either filtered or
unfiltered field audit samples. In those samples, Fe
probably formed a precipitate which was adsorbed to
the sample bottle between the time of preparation at
the analytical support laboratory and the time of
processing (filtration and acidification) in the field
laboratory. Because the samples were filtered prior to
being acidified, the adsorbed Fe precipitate would
have been lost from these samples. In laboratory
audit samples which were processed by the analyti-
cal support laboratory immediately after preparation,
the precipitate may not have had time to form. Field
audit results for Fe and total extractable Al were
therefore not included in statistical evaluations.
Nitrate Contamination
To meet contract analytical laboratory quality control
specifications, nitrate concentrations in field blank
samples were required to be less than 0.01 mg L~1.
However, during the ELS-I pilot study, up to 18 mg L~1
nitrate was detected in field blanks, and this sug-
gested that a serious contamination problem existed.
This contamination was not present in contract ana-
lytical laboratory blank samples. After discussions
with the managers of the contract laboratories per-
forming the nitrate analyses, it was concluded that
the source of the contamination was not in the con-
tract laboratory but was in the cleaning procedure for
aliquot 3 sample containers. That cleaning procedure
which included a nitric acid rinse is outlined in Table
14. As a first attempt at eliminating the contamina-
tion, the nitric acid rinse was omitted from the clean-
ing procedure for aliquot 3 containers during the ELS-
I.
Another potential source of contamination was the
sample filtration procedure at the field laboratory
which included a nitric acid rinse of the filter holder
and membrane followed by copious rinsing with RO-
Dl water. This step was not considered to be a prob-
lem at the time that the bottle cleaning procedure was
changed, and during the ELS-I field laboratory train-
ing sessions, the need for copious rinsing of the filtra-
tion apparatus was stressed.
39
-------
1. Rinse container three times with deionized water.
2. Rinse container three times with 3 N HN03.
3. Rinse container six times with deionized water.
4. Fill container with deionized water and allow to stand for 48 hours.
5. Empty container, air dry in a laminar-flow hood (class 100 air).
6. Cap containers and place in clean plastic bags.
NOTE: Deionized water must meet ASTM specifications for Type I Reagent Water. Nitric acid is Baker Instra-Analyzed or
equivalent
Table 14. Cleaning Procedure Used for Aliquot 3 Sample Containers, ELS-I Pilot Study.
The change in the cleaning procedure for the sample
containers and the rinsing step in the filtration proc-
ess reduced but did not eliminate the nitrate contami-
nation problem. Field blank samples obtained during
the early part of the ELS-I still contained up to 3.5 mg
L~1 nitrate. Overall, the mean nitrate concentration in
field blank samples during the early part of the ELS-I
was 0.3 ± 0.6 mg L~1 (n = 146). It became apparent
that the filtration procedure was a source of more
contamination than was previously expected.
In the original filtration procedure (Table 15), the filter
holder and membrane were rinsed with 5 percent
HNC>3 between samples. In a further attempt to elimi-
nate the nitrate contamination, the nitric acid rinse
was eliminated for aliquot 3. However, the same filtra-
tor was still used for all aliquots, and the filtration pro-
cedure during the processing of aliquots 1 and 4
included a nitric acid rinse of the filter holders and
membranes. Blanks processed in this manner still
contained about 0.05 mg L~1 nitrate. Further investi-
gation revealed that the design of the filter holder per-
mitted intermittent nitrate contamination (i.e., nitrate
was not completely removed by successive deionized
water rinses). To avoid this occurrence, the filtration
procedure for the ELS-I was modified to include the
use of separate filter holders and membranes to
obtain the unacidified aliquots. New filter holders
were dedicated to the aliquot 3 filtration and were
never allowed to contact nitric acid.
To determine whether the new filtration procedure
eliminated nitrate contamination of aliquot 3, a series
of experiments was performed at the field laborato-
ries in Duluth, Lexington, and Rhinelander. At each
laboratory, 10 4-liter deionized water samples (two
samples every hour for 5 hours) were collected and
processed, each in three different ways (Table 16) at
the times and dates of sample preparation given in
Table 17. The processed samples were then analyzed
Filtered Aliquots - Filtered sample is obtained by vacuum filtration through a 0.45-jum membrane filter into a clean 500-mL
container The filtered sample is then transferred into the containers for aliquots 1, 3, and 4. the filtration is performed as
follows1
a. Assemble the filtration apparatus with a waste container as a collection vessel. Thoroughly rinse the filter holder and
membrane filter in succession with 20 to 40 ml of deionized water, 20 ml of 5% HN03 (Baker Instra-Analyzed grade),
and 40 to 50 ml of deionized water (it is CRUCIAL that all traces of the HN03 rinse be removed)
b. Rinse the filter holder and membrane with 10 to 15 mL of the sample to be filtered.
c. Replace the waste container with a clean 500-mL plastic bottle Reapply vacuum (vacuum pressure must not exceed
12 inches Hg), and filter 10 to 15 mL of sample into the bottle. Rinse the container by slowly rotating the bottle so that
the sample touches all surfaces. Discard the rinse and place the 500-mL bottle back under the filter holder.
d. Filter the sample into the 500-mL bottle until the bottle is full.
e. Transfer filtered sample into aliquot containers 1, 3, and 4 (previously labeled) after first rinsing containers with 10 to
15 mL of filtered sample (as described in "c").
f If necessary return the 500-mL bottle to the filtration apparatus and collect additional filtered sample [about 700 mL of
filtered sample is required for aliquots 1 (250 mL), 3 (250 ml), and 4 (125 mL)].
Table 15. Filtration Procedure Originally Used by Field Personnel, Eastern Lake Survey - Phase I.
40
-------
Process 1. (Raw, unfiltered sample - no treatment )
a. Rinse two aliquot 3 bottles with sample (4-lrter deionized water) and discard rise water
b. Fill with fresh sample and seal.
c. Label samples as A and B
Process 2.a (Filtered sample, filter holder rinsed with acid and then with deionized water.)
a. Assemble filtration apparatus, rinse once with 5% nitric acid, followed by three deionized water rinses.
b. Insert filter membrane, rinse once with nitric acid, followed by three deionized water rinses.
c Filter 300 to 400 ml deionized water sample into waste container
d. Replace filter membrane with new membrane Rinse three times with deionized water and once with 20 ml
sample.
e. Place clean aliquot 5 bottle (500 ml) under filter holder Filter 20 ml sample into bottle. Rinse bottle with this
initial portion, then discard rinse. Filter and collect 500 ml of sample.
f. Transfer filtered sample to two aliquot 3 bottles. Label as C and D
Process 3. (Filtered sample, filter holder rinsed with deionized, water only.)
a Assemble filtration apparatus using a new filter holder. Rinse three times with deionized water
b. Insert filter membrane. Rinse three times with deionized water, followed by one 20-mL rinse with sample.
c. Place a clean aliquot 5 bottle (500 mL) under filter holder. Filter 20 mL sample into bottle. Rinse bottle with
this initial portion, then discard rinse. Filter and collect 500 mL sample.
d. Transfer filtered sample to two aliquot 3 bottles. Label as E and F.
Step d of process 2 performed at Duluth field station only
Table 16. Field Laboratory Filtration Procedure Used in Nitrate Contamination Experiment, Eastern Lake Survey
- Phase I.
for nitrate at EMSL-LV. The Lexington samples were
also split between EMSL-LV and Western Washington
University (WWU) for analysis. Detection limits for
nitrate at EMSL-LV and WWU were 0.001 and 0.002 mg
L~1, respectively.
A comparison of the nitrate data from EMSL-LV and
WWU is given in Table 18. The data from EMSL-LV indi-
cate that the nitrate concentrations in deionized
water used to prepare the field laboratory blank sam-
ples were 0.005 mg L'1 or less in 28of 30 samples. The
two exceptions (both at Duluth) had concentrations
less than or equal to 0.01 mg L~1 nitrate. With one
exception, the nitrate concentration in samples pre-
pared by process 3 (filtered sample, filter holder
rinsed with deionized water only) was 0.005 mg L"1 or
less. That exception (Duluth 5) contained less than
0.02 mg L"1 nitrate.
Both the WWU and EMSL-LV data indicated that proc-
ess 2 (filtered sample, filter holder rinsed with HNO3,
then with deionized water) produced field laboratory
blank samples with high concentrations of nitrate
(values ranging from 0.004 to 1.65 mg L"1 with an aver-
age of 0.2 = 0.5 mg L"1). Again, field blank samples
filtered using process 3 contained low concentra-
tions of nitrate. WWU reported no nitrate values
above their instrumental detection limit (0.002 mg L~
11 nitrate) using process 3.
By making a comparison of the results using process
2 with those using process 3, it was concluded that
the nitric acid rinse of filter holders and filter mem-
branes was responsible for the nitrate contamination
seen in field blanks (and, by implication, in routine
and duplicate samples). By using process 3 to pre-
pare aliquot 3, nitrate concentrations in field blanks
were reduced to levels less than 0.01 mg L~1 for the
remainder of the ELS-I.
It should be noted that the sources of contamination
can be traced to the deionized water, sample contain-
ers, and filtration apparatus, which contributed up to
1.65 mg L~1 of nitrate to field laboratory blank sam-
ples in this experiment. Two raw unfiltered samples
(Duluth 8-1 and 9-1) contained 0.009 and 0.010 mg L~1
nitrate, respectively, and the two corresponding fil-
tered samples (deionized water only, Duluth 8-3 and 9-
41
-------
Dates3 and Times'1 of Sample Processing
Sample
1
2
3
4
5
6
7
8
9
10
11e
12f
Lexington0
10/28/84,1000
10/28/84,1000
10/28/84,1100
10/28/84,1100
10/28/84,1200
10/28/84,1200
10/28/84,1300
10/28/84,1300
10/28/84,1400
10/28/84,1400
10/28/84,1400
10/28/84,1400
Rhinelander0
10/29/84,1325
10/29/84,1330
10/29/84,1420
10/29/84,1420
10/29/84,1530
10/29/84,1530
10/29/84,1630
10/29/84,1630
10/29/84,0900
10/29/84,0900
—
—
Duluth"
10/28/84
10/28/84
10/28/84
10/28/84
10/28/84
10/28/84
10/28/84
10/28/84
10/28/84
10/28/84
—
—
3 Day the sample was collected and processed (filtered).
b Time the water sample was collected from the Millipore
0 Times and dates
d Time unknown.
e This sample was
' This sample was
of sample collection. All samples were
taken from the still in the EPA regional
system.
processed on 10/30/84.
lab in Lexington, Massachussetts.
taken from the deionized water spigot in the EPA regional lab in Lexington, Massachussetts.
Table 17. Description of Field Laboratory Blank Samples Collected in Nitrate Contamination Experiment, Eastern
Lake Survey - Phase I.
3) contained 0.003 and 0.005 mg Lr1 nitrate. The ali-
quot bottles (rather than the deionized water) were
the most likely source of the slight contamination in
Duluth samples 8-1 and 9-1. Similarly, a raw unfiltered
sample (Duluth 5-1) contained 0.005 mg L~1 nitrate
while a corresponding filtered aliquot (deionized
water only, Duluth 5-3) contained 0.018 mg Lr1 nitrate.
In this case, the contamination was probably from
either the aliquot bottle or the filtration apparatus.
Both of these examples indicate the extreme care
that was necessary to reduce nitrate contamination
to less than 0.005 mg L~1. The ELS was about halfway
completed (November 1, 1984) before the cause of
nitrate contamination was isolated and corrected.
The QA manager immediately asked the contract lab-
oratories to reanalyze nitrate in aliquot 5, the unfil-
tered aliquot. Unfortunately, most aliquot 5 nitrate
determinations were performed well outside the max-
imum allowable holding time of seven days. After
extensive statistical review, it was decided to replace
all nitrate data processed on or before November 1,
1984, with the aliquot 5 data.
Evaluation of Quality Assurance Data
The QA program used a combination of blanks, dupli-
cates, and audit samples to provide an external check
on the quality of the ELS-I data and to allow early
detection of problems in sample collection and analy-
sis. Analytical data and associated QA and QC infor-
mation were collected to define the overall and ana-
lytical within-batch precision and the overall and
analytical among-batch precision for each parameter,
the normal background contamination that occurred
during the sampling and analytical process, and,
where possible, the bias among the laboratories that
performed the analyses. This section presents a sum-
mary of the QA results. Permutt and Pollack (1986)
provided additional statistical evaluation of ELS-I
data which are included in this report as Appendix A.
Blank Data
Both field blanks and analytical laboratory (calibra-
tion or reagent) blanks were analyzed during the ELS-
I. Field blanks were analyzed for turbidity at the field
laboratory and for 21 physicochemical parameters at
the contract analytical laboratory. The contract labo-
ratories used calibration and reagent blanks to con-
trol their background levels and to calculate their
instrumental detection limits. These were reported
weekly for each parameter except pH, conductance,
ANC, and BNC.
42
-------
Sample
1
2
3
4
5
6
7
8
9
10
11
12
1
2
3
4
5
6
7
8
9
10
1
2
3
4
5
6
7
8
9
10
Process
1
1
1
1
1
1
1
1
1
1
1
1
2
2
2
2
2
2
2
2
2
2
3
3
3
3
3
3
3
3
3
3
a All samples were analyzed within 7 days of
b ND = not detected (less than 0.002 mg L"1
Duluth
EMSL-LV
0003
0.005
0005
0.005
0.005
0.003
0.003
0009
0.009
0.010
0.003
—
—
0.033
0.033
0.033
0036
0059
0.224
0.051
0.033
0.035
0.020
0.171
0.005
0.005
0003
0.003
0018
0.005
0.003
0.003
0.005
0.002
preparation.
N03-).
N03- (mg I/1)3
Rhinelander
EMSL-LV
0.003
0.003
0.003
0003
0.003
0.003
0.003
0.003
0.003
0.003
—
—
0.006
0.014
0.006
0.008
0.007
0.013
0.023
0.015
0.186
0.006
0.006
0.003
0.003
0.003
0003
0.003
0.003
0.002
0 003
0.003
0.003
Lexington
EMSL-LV
0001
0.001
0.001
0.001
0.001
0.001
0.001
0.005
0.001
0.005
0.044
0.001
0.042
1.130
0.146
0.046
0.-44
0.088
0.104
0.025
0023
0.008
0.001
0.001
0001
0.001
0.001
0.001
0001
0.001
0.001
0.001
Lexington
WWU
NDb
ND
ND
ND
ND
ND
—
ND
ND
ND
—
0.046
0.046
0.077
1.65
0.128
0.062
0.046
0.126
0.108
0.032
0.030
0.004
ND
ND
ND
ND
ND
ND
ND
ND
ND
ND
Table 18. Nitrate Concentrations in Field Laboratory Blank Samples in Nitrate Contamination Experiment,
Eastern Lake Survey - Phase I.
43
-------
Internal laboratory blanks were not routinely ana-
lyzed at the field laboratories because the limited
number of analyses performed there would not have
provided sufficient QA/QC information to justify the
extra effort and expense. There were few problems
achieving detection limits for measurements made in
the field laboratories. In addition, pH and DIG meas-
urements are highly variable in blank samples
because of the tendency of blanks to absorb CO2 from
the air.
In total, 245 field blanks were collected during the
ELS-I. The data were analyzed to provide an overall
estimate of the normal background contamination
that occurred during sampling and analysis and to
identify and correct any significant contamination
problems as they occurred. A statistical evaluation of
the verified data yielded a system decision limit, a
system detection limit, and an estimated instrumen-
tal detection limit for each variable (Appendix A). The
system decision limit represents the lowest instru-
ment signal that can be distinguished from back-
ground at a = 0.05. The system detection limit is
based on the reproducibility of the field blank analy-
sis; it represents the lowest concentration that can be
measured above the system decision limit.
The estimated instrumental detection limit was cal-
culated from the reported calibration blank data. It
was used to provide a preliminary QA check on the
analytical results provided by the contract laboratory.
The system detection limit should be comparable to
the instrumental detection limit reported by the con-
tract laboratory if there is no variability added as a
result of sample collection, processing, and ship-
ment.
Table 19 provides a summary of field and laboratory
blank data for 20 parameters from the ELS-I. Values
for individual laboratories are listed in Appendix B.
The required instrumental detection limits (IDL's)
were achieved in the contract laboratories for all
parameters except Fe and NO3~. However, for some of
the parameters the system decision limit is consider-
ably higher than the required IDL; this is possibly due
to background sources of contamination.
The evaluation of blank data demonstrated that
detection limit goals were generally achieved. How-
ever, to interpret the data, results from the field blanks
must be taken into consideration. Although
extremely low detection limits can be achieved in the
laboratory, they are of little value in defining usable
data when they are lower than the background from
sample collection and handling. The system decision
limit and system detection limit must therefore be
considered as the real limits for data interpretation.
Duplicate Data
Data for estimating overall within-batch precision
were obtained using field duplicate samples. These
were processed by the field laboratories and were
analyzed with the routine samples and field blanks at
the contract laboratories. The identity of the field
duplicates was not disclosed to the contract labora-
tories. The field and contract laboratories also per-
formed duplicate chemical analyses on one sample
per batch as a QC check on analytical within-batch
precision.
In total, 125 field duplicates were collected during the
ELS-I. The estimates of overall within-batch precision
obtained from field duplicate data include the effects
of sample collection, processing, shipping, and anal-
ysis, but do not include the effects of among-batch
variation that may have been caused by day-to-day
differences such as different calibration curves.
Within-batch precision was calculated as the root
mean square of percent relative standard deviation
(%RSD) to estimate the 'average' %RSD over the
range of values for routine/duplicate pairs. The esti-
mated precision values are directly comparable to the
intralaboratory precision goals which are also
expressed as %RSD.
A quantitation limit was calculated for each parame-
ter to distinguish values which are expected a priori
to have greater and more variable relative errors (i.e.,
values close to the detection limit) from values which
are expected to have smaller and more consistent rel-
ative precision (values much greater than the detec-
tion limit). The quantitation limit should not be con-
fused with instrumental detection limits, system
detection limits, or system decision limits, all of
which serve to constrain the actual size of the usable
data set. All of the verified data, including values
below the quantitation limits, were considered in the
validation process; quantitation limits were calcu-
lated solely for the purpose of estimating the charac-
teristic relative precision. A more detailed discussion
of the statistical approach for evaluating within-
batch and among-batch precision is provided in
Appendix A.
Overall within-batch precision estimates for 23
parameters from the ELS-I are presented in Table 20.
The three types of pH measurements were evaluated
in terms of absolute rather than relative standard
deviation. The estimated overall within-batch preci-
sion for pH in all four laboratories was within a range
of 0.05 to 0.09 pH unit. No single laboratory had uni-
formly high or low precision; however, air-equilibrated
pH data fromVersar and initial ANC and BNCpH data
from USGShad higher variability than the other meas-
urements (see Appendix C).
The measured concentrations of some parameters
(Ca, Mg, Na, SO42~) were far above both the reported
and the required instrumental detection limits. The
results indicate that overall within-batch precision
was better than the intralaboratory precision goal for
each of these parameters. Most or all of these pairs
44
-------
Parameter
Al, mgL~1
total extractable
total
ANC./^eq L~1
Ca, mg ~1
C1"1, mgL"1
Conductance, /*S cm'1
DIG, mg L~1
air-equilibrated
initial
DOC, mg L"1
F~, total dissolved,
mg L~ '
Fe, mg L"1
K, mg L"1
Mn, mg L'1
Mg, mg L"1
Na, mg L"1
NH4+,mgL"1
N03", mg L"1
all batches
aliquot 3
aliquot 5
P, total, mg L"1
Si02, mg L~1
S042", mg L"1
a See Drouse et al. (1986).
Instrumental Detection
Required3 Reported3
0.005 0.003
0.005 0.003
_e e
0.01 0.01
0.01 0.01
--' 0.0
0.05 0.03
e e
' 0.1 0.1
0.005 0.002
0.01 0.02
0.01 0.00
0.01 0.01
0.01 0.00
0.01 0.00
0.01 0.01
0.005 0.006
0.002 0.001
0.05 0.02
0.05 0.02
11 Estimated instrumental detection limit (IOL) = (2{P95-P50)), where P95
(median), of laboratory calibration blank samples (see Appendix A).
Limit
Estimated11
0.013
0.004
e
0.00
0.05
1.6
0.14
e
0.3
0.008
0.02
0.00
0.01
0.00
0.10
0.03
0.037
0.003
0.06
0.09
= 95th percentile, and P5
System
Detection
Limit0
0.014
0.050
11
0.06
0.13
1.4
0.20
0.38
0.4
0.008
0.04
0.03
0.02
0.02
0.06
006
0.760
0.046
1.782
0.014
0.21
0.13
o = 50th percentile
System
Decision
Limit"
0.008
0.030
6.9
0.03
0.08
1.3
0.28
0.42
0.4
0.005
0.02
0.02
0.01
0.01
0.03
0.04
0.389
0.023
0.919
0.008
0.11
0.09
0 System detection limit = (2(P95-P50)), where P95 = 95th percentlle, and P60 = 50th percentlle (median), of field blank
samples (see Appendix A).
d System decision limit = (P95) =
95th percentile of field blank samples
(see Appendix A).
e Measurements of laboratory calibration blanks not required.
1 For conductance, the mean of six
0.9 /uS cm"1.
nonconsecutive laboratory calibration
blank measurements was
required to be less
than
Table 19. Instrumental Detection Limits, System Detection Limits, and System Decision Limits for 20
Parameters, Eastern Lake Survey - Phase I.
45
-------
Pairs With Mean>0
Parameter
A1, total extractable mg L~1
all values
J<= 0 010
Y>0 010
A1, total, mg L~1
all values
x<= 0010
x"> 0 010
ANC,Meq L"1
Ca, mg L"1
C1~, mg L"1
Conductance, /*S cm'1
DIG, mg L"1
air-equilibrated
initial
DOC, mg L"1
all values
7<= 5
x">5
F"
Fe, mg L~1
K, mg L"1
Mg, mg L~1
RDLa nb
0.005 113
69
44
0005 125
9
116
5 121
0.01 125
001 125
— e 125
0.05 124
0.05 125
0 1 125
66
59
0005 125
0.01 118
0.01 125
001 125
Root
Mean
Square
of %RSDC
46
59
21
35
55
33
20
23
20
1.9
12
69
9.8
6.7
12
8.1
97
4 7
2.3
Quanti-
tation
Limit
(10sB)d
0 039
0.277
56.6
0.14
0 38
90
0.56
1 09
2 5
0.034
0.12
0.37
0 04
Pairs With Mean >10sBc
nb
9
0
9
1
0
1
90
125
85
125
94
85
105
46
59
62
32
82
125
Root
Mean
Square
of %RSDC
11
11
24
24
10
2.3
17
1 9
5.0
3 7
10
5.6
12
8 9
10
3 7
2 3
Overall
Intra-
Laboralory
Precision
Goal
(%RSD)C
20
10
20
10
10
5
5
3
10
10
10
5
5
10
10
10
Table 20. Overall Within-Batch Precision Estimated From Field Duplicate Data and Field Blank Data for Measurements of 23 Parameters, Eastern Lake
Survey - Phase I.
-------
Pairs With Mean>3
Root
Mean
Square
Parameter
Mn.mg L"1
Na, mg L"1
NH4 + , mg L"1
N03~, mg L"1
all batches
aliquot 3
aliquot 5
PH
air-equilibrated
initial ANC
initial BNC
P, total, mg L"1
all values
x<= 0.010
x>0.010
Si02, mg L"1
S042~, mg L"1
RDL3
001
0.10
0.01
0.005
::|
0.002
0.05
0.05
n" of%RSD°
82
125
113
113
45
68
125
122
57
65
120
125
57
4.2
34
443
213
544
51
37
44
30
44
6.6
Pairs With Mean>10sB°
Quanti- Root
tation Mean
Limit Square
(lOsu)" nb of%RSDc
0.093 6 11
0.20 121 43
0.56 0
0 894 5 60
0.163 9 45
1.143 0
"I ::f ::|
-t -f -f
0.037 4 9.7
0
4 97
2.34
1.68 115 6.5
Overall
Intra-
Laboratory
Precision
Goal
(%RSD)C
10
5
10
10
u
0.010
0.010
0 01n
20
10
5
aRDL = required detection limit (in applicable units).
bn = number of routine/field duplicate
pairs.
°%RSD = percent relative standard deviation.
dsB = standard deviation of field blank
aliquot 5, n = 145.
measurements
(see Appendix A) with n = 245
except for total
e For conductance, the mean of six nonconsecutive calibration blank measurements was required to be
1 Not applicable.
9 Root mean square of absolute standard deviation (pH
h Absolute precision goal (pH unit).
unit).
A1, n = 244; NH4 + , n = 243; N03~, aliquot 3, n = 99,
less than 0.9/uS CM"1.
and N03~,
Table 20. Overall Within-Batch Precision Estimated From Field Duplicate Data and Field Blank Data for Measurements of 23 Parameters, Eastern Lake
Survey - Phase I (Continued).
-------
had a mean concentration greater than the quantita-
tion limit (see Table 20.
Other chemical parameters (Mn, Fe, total extractable
Al, total Al, Cf, SiO2, and total P) were characterized
by low concentrations (at or below the quantitation
limit) in many of the sample pairs. The relative preci-
sion of the data is biased high due to the increase in
absolute variability of the measurement near the
instrumental detection limit (Appendix A). Even if the
absolute variability in the chemical analysis was uni-
form at all concentrations, an estimate of the true pre-
cision would be difficult to ascertain over the entire
range of values when it Is expressed as a percentage
of the mean concentration.
The estimated precision for a!! duplicate data with
concentrations greater than zero is also provided in
Table 20. For these parameters, all pairs with concen-
trations above the quantitation limit had overall
within-batch precision that was better than the over-
all intralaboratory precision goal, except for_tptal
extractable Al, total Al, Cl~, DOC (for pairs with X >5
mg L~1), Mn, NO3~2, and SO4~2. The data user should
again consider the estimated quantitation limit when
interpreting the precision of data.
Duplicate analyses, designated as trailer duplicates,
were also performed for all measurements made in
the field laboratory. One trailer duplicate was ana-
lyzed per sample batch. This duplicate analysis was a
QC step to assure consistent measurements within
the field laboratory. A QC limit of ±0.1 pH unit was
established for pH measurements, and a QC limit of
±10 percent was established for DIG, true color, and
DIG measurements. If the QC limits were exceeded,
the laboratory was required to analyze a second
duplicate. If the limits were still exceeded after rean-
alysis, the laboratory supervisor made additional
efforts to identify and correct the problem until it was
resolved.
Precision estimates from field duplicate and trailer
duplicate data are presented in Table 21. For trailer
duplicates, the observed analytical within-batch pre-
cision for each parameter was within the analytical
intralaboratory precision goal established for the
ELS-I. The overall within-batch precision for field
duplicates was within the overall intralaboratory pre-
cision goal for pH, but was not for DIG, true color, or
turbidity. There was little bias apparent among field
stations for these analyses, based on gross observa-
tion of the data. Bias is discussed in greater detail in
Appendix A.
The contract laboratories were required to analyze
one sample per batch in duplicate. For each parame-
ter, the analytical within-batch precision was required
to be within the analytical intralaboratory precision
goal. Analytical within-batch precision estimates
from contract analytical laboratory duplicate pH data
are presented in Table 22. The results for pH indicate
that precision was within a range of ±0.04 to 0.08 pH
unit. These data may have been biased because the
laboratory analyst knew which samples were dupli-
cates and that the QC process required that the preci-
sion goal be achieved. However, the results are an
indication of the precision that was achieved within
each laboratory with the method QC requirements.
Precision estimates for the other 20 parameters from
contract laboratory duplicate data are also shown in
Table 22. The estimated analytical within-batch preci-
sion for pairs with mean >10sB was better than the
analytical intralaboratory precision goal for each of
the parameters except total extractable Al>0.010 mg
L~1, CI~,Mn, SO42", and conductance.
In general, the overall and analytical intralaboratory
precision goals were achieved between values for
routine/field duplicate pairs as well as between con-
tract laboratory duplicate pairs. The exceptions to
these results are discussed in Section 5. Thus, esti-
mated overall within-batch precision was considered
to be adequate to meet the DQO's established at the
beginning of the ELS-I.
Audit Sample Data
Audit sample data can be used to obtain estimates of
among-batch precision and interlaboratory bias (see
Table 3). Six types of audit samples (two field natu-
rals, two field synthetics, and two laboratory synthet-
ics) were analyzed for 23 parameters during the ELS-I.
Field naturals and field synthetics were handled in
the same manner as routine samples (see Section 2)
to estimate the overall among-batch precision includ-
ing the effects of sample processing. Laboratory syn-
thetics were prepared as processed aliquots by the
analytical support laboratory; these were relabeled at
the field labratory and incorporated with the sample
batch to estimate analytical among-batch precision.
The natural audit samples were also used to estimate
relative interlaboratory bias by comparing measured
values from the contract laboratories with theoretical
values and with referee laboratory measurements.
The synthetic audits were used to provide informa-
tion on absolute interlaboratory bias which required
that they be prepared from solutions of known com-
position. Examination of both the theoretical values
and the measured values for synthetic audit samples
indicated that the actual sample composition may
have differed on different days, and that measure-
ment imprecision and interlaboratory bias were small
by comparison (see Appendix A).
One type of field natural sample (FN2, from Big
Moose Lake in the Adirondack Mountains, New York)
was representative of lakes which are sensitive to
acidic deposition; the other type (FN3, from Lake
Superior near Duluth, Minnesota) represented sys-
tems with high ANC. In total, 41 FN2 samples and 7
FN3 samples were used during the survey. Two con-
-------
Parameter
PH
DIG, mg I/1
True color, PCU
Turbidity, NTU
Overall and
Analytical
Intralaboratory
Precision
Goal
(%RSD)a
0.1°
10
10
10
Overall
Within-Batch
Precision
(Field Duplicates)
Root Mean
Square of
nb %RSD"
124 0.04d
123 16
125 22
125 19
Analytical
Within-Batch
Precision
(Trailer Duplicates)
Root Mean
Square of
nb %RSDa
93 0.01°
116 4.6
118 1 5
117 8.4
aRSD = relative standard deviation.
bn = number of duplicate pairs.
Absolute standard deviation (pH
unit).
dRoot mean square of absolute standard deviation (pH unit).
Table 21. Overall and Analytical Within-Batch Precision Estimated From Field Duplicate and Trailer Duplicate
Data for Measurements of Four Parameters, Eastern Lake Survey - Phase 1.
centrations of synthetic audit samples were also pre-
pared including 36 field highs, 21 laboratory highs, 42
field lows, and 43 laboratory lows.
Table 23 summarizes the mean sample composition
and the overall amongbatch precision for 23 parame-
ters estimated from FN2 and FN3 data. Mean values
were less than quantitation limits for 7 parameters
measured in FN2 samples and for 7 parameters mea-
sured in FN3 samples. For ANC, air-equilibrated DIG,
and initial DIG, estimated precision was better (i.e.,
%RSD was lower) for FN3 samples which had means
greater than quantitation limits than for FN2 samples
which had means less than quantitation limits. This is
consistent with our expectations. The precision of
total extractable Al, total Al, and DOC measurements
was slightly better with FN2 samples than with FN3
samples; for those parameters, mean FN3 values
were less than quantitation limits. A test of signifi-
cance of the differences was not performed because
of the small number of FN3 samples used.
For both FN2 and FN3 samples, means were less than
quantitation limits for Fe, Mn, NH4+, and total P, and
means were greater than quantitation limits for the
remaining 10 parameters. Quantitation limits were
not applicable for the 3 pH measurements. For the
field natural audit samples with means greater than
quantitation limits, the overall interlaboratory preci-
sion goal (estimated as twice the overall intralabora-
tory precision goal, see Table 20) was achieved for all
parameters except total extractable Al, total Al, and
Cl~ in FN2 samples.
Tables 24 and 25 summarize the mean values for sam-
ple composition and the overall and analytical
among-batch percision estimated from measure-
ments of synthetic audit samples. For the field high
synthetics, the mean value for total Al was less than
the quantitation limit; all other parameters (except
pH) had means above the quantitation limits. The
overall interlaboratory precision goal was not met for
total Al, as was expected, and was also not met for Cl~
, air-equilibrated and initial DIG, DOC, Mn, NH/,
NO3~, and SiO2. Means for laboratory high synthetics
were greater than quantitation limits for all parame-
ters (not applicable for pH). Analytical interlaboratory
precision goals were met for all parameters except
total Al, Ca (Lot 4), Cl- (Lots 5 and 6), conductance, air-
equilibrated and initial DIG, Mn, NH4 + , NO3~, and
SiO2.
For the field low synthetic audit samples, the mean
values for total Al, Cl", DOC, K, Mn, NH4', NO", total P,
and SiO2 were less than the quantitation limit; and the
mean of Ca (Lot 4) values was equal to the quantita-
tion limit. The overall interlaboratory precision goals
were not met for any of these 10 parameters, with the
exception of K and Mn. For the other 13 parameters,
the overall interlaboratory precision goals were met
49
-------
Pairs With Mean<0
Parameter
A1, total extractable mg L"1
all values
x"<= 0 010
x">0 010
A1, total, mg L~1
all values
'x<= 0 010
I> 0.010
ANC, Meq L"1
Ca, mg IT1
Cr, mg L"1
Conductance, /uS cm'1
DIG, mg L"1
air-equilibrated
initial
DOC, mg L"1
all values
7<= 5
>f>5
F~, total dissolved,
mg L"1
Fe, mg L"1
K, mg L~1
Mg, mg L"1
RDLa
0.005
0.005
5
0.01
0.01
-e
0.05
0.05
0.1
0.005
0.01
0.01
0.01
nb
123
22
101
126
8
118
116
123
127
125
126
127
126
85
41
127
120
125
121
Root Quanti-
Mean tation
Square Limit
of%RSD° (10sB)d
13 0.038
14
12
19 0.012
57
12
30 56. 6f
0.88 004
23 0.11
11
7.1 0.60
3.7 0 54
68 10
8.1
2.4
2.5 0015
14 0.05
18 0.04
0.64 0 01
Pairs With
nb
46
0
46
54
0
9
86
123
124
5.4
94
113
114
73
41
123
101
121
121
Mean <10sBd
Root
Mean
Square
of %RSD°
18
—
18
—
5.4
2.1
0 88
7.1
123
2.2
3 5
2.4
2.5
2 4
2.5
4.3
1 5
0 64
Analytical
Intra-
Laboratory
Precision
Goal
(%RSD)C
20
10
20
10
10
5
5
101
10
10
10
5
5
10
5
5
Table 22. Analytical Within-Batch Precision Estimated From Contract Laboratory duplicate Data and Calibration Blank Data for Measurements of 23
Parameters, Eastern Lake Survey - Phase I.
-------
Pairs With Mean>0 Pairs With Mean >10sBc
Parameter RDLa
Mn, mg L"1 0.01
Na, mg L~1 0.10
NH4 + ,mgL"1 0.01
N03~, mg L"1
all batches 0.005
aliquot 3
aliquot 5
pH
air-equilibrated -g
initia ANC -g
initia BNC --g
P, total, mg L~1
all values ' 0.002
7 = 0.010
x 0.010
Si02, mg L"1 0,05
S042", mg L"1 0.05
aRDL = required detection limit (in applicable units).
bn = number of contract laboratory duplicate pairs.
C%RSD = percent relative standard deviation.
dsB = standard deviation of measurements of contract laboratory
N03~, aliquot 5, n = 72.
Analytical
Intra-
Root Quanti- Root Laboratory
Mean tation Mean Precision
Square Limit Square Goal
n" of%RSD° (10sB)° n° of%RSDc (%RSD)°
102
123
121
122
51
71
127
127
127
125
30
95
127
127
calibration blanks
e For conductance, the mean of six nonconsecutive calibration blank measurements
1 Quantitation limit calibrated using field blank data (see text).
9 Not applicable.
h Root mean square of absolute standard deviation (pH unit).
' Absolute standard deviation (pH unit).
21 0.030 73 17 10
13 0.10 121 0.96 5
7.1 0.15 54 2.3 5
13 0.157 86 3.6 10
20 0.110 39 3.7
3.9 0.185 46 3.3
0.08h --g --g --g 0.01J
0.04h "8 -Q --Q 0.05
o.07h --g --§ -g o.os1
13 0.008 104 9.9
20 9 8.9 20
10 95 10 10
2.5 2.37 64 2.2 5
11 1.18 124 11 5
with n = 127 except for DIC, n = 125; Mn, n = 126; N03", aliquot 3, n = 53; and
was required to be less than 0.9/iS cm"1.
Table 22. Analytical Within-Batch Precision Estimated From Contract Laboratory duplicate Data and Calibration Blank Data for Measurements of 23
Parameters, Eastern Lake Survey - Phase I (Continued).
-------
Parameter
A1, mg I/1
total extractable
total
ANC.neq L~1
Ca, mg L"1
C1", mg L"1
Conductance, /iS cm'1
DIG, mg L"1
air-equilibrated
initial
DOC, mg L"1
F", total
dissolved, mg L"'
Fe, mg L"1
K, mg L"1
Mg, mg L"1
Mn, mg L~'
Na, mg L"1
NH4' , mg L"'
N03~, mg L"1
all batches
aliquot 3C
aliquot 5d
PH
air-equilibrated
initial ANC
initial BNC
P, total, mg L"1
Si02, mg L"1
SO/", mg L"1
FN2 (n =
Mean
0 182
0 305
2.4e
1 91
0 61
26.7
0 19e
0.42e
33
0 077
0 02e
0 49
0 35
0.07e
0 67
0 063
1 425
1 473
1.403
5 18
507
5.08
0.0026
4.33
6.95
41)
%RSDa
34
28
258
9.2
51
3.6
GO
27
9 3
4.0
66
4.3
2 7
26
5 1
54
5 5
3.0
5 8
0.271
0041
0.051
144
8.1
7 3
FN3 (n
Mean
0 002"
0 021e
8482
13 28
1.39
96.4
9.68
9.93
1.4e
0 035
OOO3
0 49
2 79
0 00e
1.30
0.01e
1.401
1.422
1.385
8.23
776
7.79
0 001e
2.72
3.26
= 7)
%RSDa
87
123
2 3
55
2 0
1.7
3 8
4 2
9 7
5 2
137
12
3 6
362
6 4
130
58
3 6
8 5
0 081
O1
0'
143
6.0
3.3
Quantitation
Limit
(10sB)b
0 039
0 277
56 6
0 14
0 38
9.0
0 56
1 09
2 5
0 034
0 12
0 37
0 04
0093
0 20
0 56
0984
0 163
1 143
—9
"9
"9
0 037
2 34
1 68
a Percent relative standard deviation.
bsB = standard deviation of measurements of field
0 Aliquot 3 samples after filtration protocol change
d Aliquot 5 samples (FN2, n = 28; FN3, n = 3)
e Mean less than quantitation limit
1 Absolute standard deviation.
g Quantitation limits not applicable
blank samples
(FN2, n = 13;
(see Appendix A).
FN3n = 4).
for pH measurements.
Table 23. Overall Among-Batch Precision Estimated From Field Natural, Lot 2 (FN2, Big Moose Lake) and Field
Natural, Lot 3 (FN3, Lake Superior) Audit Sample Data for Measurements of 23 Parameters, Eastern Lake Survey
- Phase I.
52
-------
Parameter
A1,mgL"1
total extractable10
total
ANC,/ueq L"1
Ca, mg L"1
Lot 4
Lot 5 and 6
Cr, mg L"1
Lot 4
Lot 5 and 6
Conductance, ^S cm'1
DIG, mg L"1
air-equilibrated
initial
DOC. mg L"1
F", total dissolved, mg L"1
Fe, mg L~lc
K, mg L"1
Mg. mg L~1
Mn, mg L~1
Na. mg L'1
NH^, mg L"'
X
0 199
476.7
1.66
1 95
366
4.00
104 5
4.72
594
100
0445
—
3 03
2 37
1 19
11 84
1 26
Overall Among-Batch
(Field Highs)
%RSDa
d 30
17
66
8 2
40
18
4 0
27
33
20
94
—
63
38
13
8 7
13
Precision
Quantitation.
Limit
(10sB)b
0.039
0.277
57.6
0.14
038
90
056
1 09
2 5
0034
0.12
0 37
0 04
0 093
020
0 56
Analytical Among
(Laboratory Highs)
-Batch
Y %RSDa
0.161
0.194
485.9
1 55
1.96
3.39
581
104.3
4.79
5.94
103
0.435
0.18
3 02
236
1.39
11 87
1 34
19
34
20
12
5.1
7.1
92
39
27
32
16
2.8
11
5 0
34
29
8 8
13
Precision
Quantitation
Limit
(10sB)b
0.038
0012
56. 69
0 04
0.11
5.4
0.60
0.54
1 0
0015
0 05
0 04
0 01
0 030
0 10
0 15
Theoretical
Value
-h
0.19
-h
1.54
2.39
2.72
4 22
-h
--h
3.10
10.0
0452
0 15
2.97
2 43
1 50
12.41
1 25
Table 24. Mean Measured Values, Overall and Analytical Among-Batch Precision Estimates, and Theoretical Concentrations of High Synthetic Audit
Samples, Eastern Lake Survey - Phase I.
-------
Parameter
N03", mg L~1
(all batches)
PH
air-equilibrated
initial ANC
initial BNC
P, total, mg I/1
Si02, mg L"1
S042", mg LM
a Percent relative standard deviation.
Overall Among-Batch Precision Analytical Among-Batch Precision
Quantitation Quantitation
(Field Highs) Limit (Laboratory Highs) Limit
Y %RSDa (10sB)b X" %RSDa (10sB)b
1 816 37 0 894 1 431 47 0 157
7.72 0 54e — g 7.87 0 15e -g
7.08 0.25e -g 7.05 0 26e --g
7.12 027e -g -h
0057 23 0037 0061 16 0008
9.23 14 2.34 9.42 15 237
14.39 5.6 1.68 14.47 4.8 0.18
Theoretical
Value
1.707
--h
-h
0.075
10.70
14 09
bsB = standard deviation of measurements of field blank or laboratory blank samples (see Appendix A).
0 Laboratory synthetic audit samples
d Mean less than quantitation limit.
only (see Section 4).
e Absolute standard deviation (pH unit).
1 Quantitation limits not applicable for pH measurements.
9 Quantitation limit calculated using field blank data.
h Theoretical concentration not available.
Table 24. Mean Measured Values, Overall and Analytical Among-Batch Precision Estimates, and Theoretical Concentrations of High Synthetic Audit
Samples, Eastern Lake Survey - Phase I (Continued).
-------
Parameter
A1, mg I/1
total extraclable10
total
ANC,/ieq L"1
Ca, mg L"1
Lot 4
Lot 5 and 6
CT, mg L"1
Lot 4
Lot 5 and 6
Conductance, ^S cm'1
DIG, mg L~1
air-equilibrated
initial
DOC, mg L'1
F~. total dissolved, mg L~1
Fe, mg L"'°
K. mg L"1
Mg, mg L"1
Mn, mg L"1
Ma, mg L"1
NH4 + , mg L"1
X
—
0 023
112
0.14d
1 17
033d
037d
19.0
1.35
1.65
10d
0.042
~
023d
042
1.091
2.71
1.16d
Overall Among-Batch
(Field Highs)
%RSDa
—
d 22
7.4
19
7.1
71
22
68
8.9
13
48
21
—
19
73
10
5.2
27
Precision
Ouantitation
Limit
(10sB)b
0.039
0.277
56.6
0.14
038
9.0
0.56
1.09
2.5
0.034
0.12
0.37
0.04
0093
0.20
0.56
Analytical Among-Batch
(Laboratory
X"
0.018d
0.031
110
0.14
0.16
0.30
0 35
18.7
1.35
1.65
1.2
0.040
0 07
0.23
0.43
0.092
2.69
0.19
Highs)
%RSDa
33
90
9 4
11
8 7
14
31
6.4
11
15
61
50
23
10
4.5
16
9.3
22
Precision
Quantitation
Limit
(10sB)b
0.038
0 012
56 69
004
0.11
5.4
060
0.54
1 0
0 015
005
0.04
001
0.030
0.10
0 15
Theoretical
Value
-f
002
-f
0.13
0 19
0.22
0 34
--f
--f
0.96
1 0
0042
0 06
0 20
0.45
0 10
2 79
0.17
Table 25. Mean Measured Values, Overall and Analytical Among-Batch Precision Estimates, and Theoretical Concentrations of High Synthetic Audit
Samples, Eastern Lake Survey - Phase I.
-------
Parameter
N03", mg I/1
pH
air-equilibrated
initial ANC
initial BNC
P, total, mg L"1
Si02, mg L~1
SO/", mg L"1
Overall Among-Batch Precision Analytical Among-Batch Precision
Quantitation Quantitation
(Field Highs) Limit (Laboratory Highs) Limit
Y %RSDa (10sB)b 7 %RSD" (10sB)b
0 547d 66 0894 0471 37 0.157
7.34 0 14e -f 7.29 0.13e -f
6.87 0.11e --f 686 0.12e -f
6.96 Oe --f 6.93 0.15e -f
0.021" 29 0.037 0022 41 0.008
1.00d 31 2 34 1 02d 20 2.37
227 10 1.68 2.28 4.4 0.18
Theoretical
Value
0.466
-f
--f
-f
0 027
1.07
2 28
3 Percent relative standard deviation.
bsB = standard deviation of
c Laboratory synthetic audit
d Mean equal to or less than
measurements of field blank or laboratory blank samples (see Appendix A).
samples only (see Section 4).
quantitation limit.
e Absolute standard deviation (pH unit).
1 Not applicable
9 Quantitation limit calculated using field blank data.
Table 25. Mean Measured Values, Overall and Analytical Among-Batch Precision Estimates, and Theoretical Concentrations of High Synthetic Audit
Samples, Eastern Lake Survey - Phase I (Continued).
-------
by all parameters except conductance, total dis-
solved F~, air-equilibrated pH, initial ANC pH, and ini-
tial BNC pH. Means for laboratory low synthetics
were greater than quantitation limits for all parame-
ters except total extractable Al and SiO2 (not applica-
ble for pH) The analytical interlaboratory precision
goals were met for ANC, Ca (Lots 5 and 6), air-equili-
brated and initial DIG, total dissolved F", K, Mn, Na,
NH4 , and SO4~2; they were not met by the other 13
parameters.
It is evident from the results for synthetic audit sam-
ples that the levels of among-batch precision
expected for these measurements were not attained
for all parameters, despite the high sample values
The reasons for these discrepancies presumably
include mixing error or sample instability or both;
they are discussed in greater detail in Appendix A.
The results for field and laboratory audit samples
show that quantitation limits were a useful means of
classifying the data, which was necessary to objec-
tively evaluate among-batch precision estimates for
the 23 parameters. It was possible to determine
whether the observed precision estimates for each
parameter were reasonable in relation to the DQO's
established for the survey. High precision was not
expected for measurements close to the detection
limits, and it was not achieved for any parameters
with means less than quantitation limits except DOC.
Conversely, high precision was expected and was
generally achieved for measurements with higher
mean values (i e., for 4 parameters with FN2 samples
and for 1 parameter with FN3 samples). Estimated
overall and analytical within batch precision of DOC
measurements showed a pattern which was opposite
to that shown by all other parameters; measurements
at higher concentrations exhibited greater variability
than those at lower concentrations.
Natural and synthetic audit samples were also used
to judge the performance of contract analytical labo-
ratories on a batch basis (Drouse et al., 1986). Table 26
provides exam pies of the 95 percent confidence inter-
vals (performance windows) used for data evaluation.
Actual laboratory performance was judged according
to 95 percent confidence intervals for each lot of each
type of audit sample. Any audit sample outside the
interval was flagged for further verification. These
flags were used to concentrate verification efforts on
potential analytical problems. Field synthetic audit
samples were found to contain precipitated iron and
aluminum which were removed by filtering at the field
laboratory (see above); those samples were not inclu-
ded in the statistical evaluation.
57
-------
Field and
FN2a
Parameter
A1, mg L"1
total extractable0
total
ANC,Meq L"1
Ca, mg L"1
CT, mg L'1
Conductance's cm"1
DIG, mg L"1
air-equilibrated
initial
DOC, mg L"1
F~ total dissolved, mg L"1
Fe, mg L"1c
K, mg L"1
Mg, mg L"1
Mn, mg L"1
Na, mg L"1
NH4+, mg L"1
N03~, mg L"1
pH
air-equilibrated
initial ANC
initial BNC
P total, mg L"1
Si02, mg L"1
SO/", mg L"1
a FN2 = Field natural, lot
b Ft\!3 = Field natural, lot
0 Laboratory synthetic aud
High
0.308
0.443
14.8
2.03
0.61
28.0
0.21
0 61
3 5
0.083
0.04
0.53
0.36
0.08
0.74
0.12
1.557
520
5.15
5.18
0.006
5.05
7.52
2 (Big Moose Lake)
3 (Lake Superior).
samples only (see Section
Low
0056
0.195
-10.1
1.75
0.45
25.2
0.10
021
29
0.071
-0.00
0.45
0.33
0.06
0.60
0.00
1.316
5.07
4.98
4.98
-0.003
3.62
6.20
4).
FN3b
High
0 007
0.023
8680
1395
1.46
100 5
10.61
10 99
1.7
0 039
-0.01
0.57
3.05
0.00
1.51
0.03
1.607
8 28
8.07
8.08
0 007
3 14
354
Low
-0.003
-0 006
815.1
12.12
1 32
92 3
8.76
888
1 0
0.030
0.21
0.46
2 54
-0.01
1.09
-0.02
1.205
8.19
7.44
7.49
-0.004
231
2.99
Laboratory
High Synthetics
High
0.22
0.27
654 4
2.21
4.35
112.9
7 29
9.88
14 5
0.471
0.15
330
2 54
1.58
13.87
1.58
3.090
8.14
757
7.64
0.078
11.90
15.67
Low
0.10
0.14
305.9
1.30
2.67
96.1
2.14
1.99
6.1
0.404
009
2 83
2.20
0.94
9.76
0.97
0.264
7.58
6.56
6.58
0.038
6.74
13 05
Field and
Laboratory
Low Synthetics
High
0 03
0.01
128 3
0.19
0.38
21.1
1.61
2 07
1.9
0045
0 05
0 27
4.58
0.10
2.98
0 25
0927
7.59
7.09
7 21
0 031
1.43
2 47
Low
0.01
94.4
0.11
0 23
16.5
1.11
1.24
0.1
0 036
0.18
0.39
0 08
2 45
0.10
0.037
7 04
6 63
6.65
0.010
0 57
2 09
Table 26. Contract Analytical Laboratory Performance Windows for Audit Sample Measurements, Eastern Lake Survey - Phase I.
-------
Section 5
Data Variability in the ELS-/
Four types of precision estimates (overall within-
batch, overall amongbatch, analytical within-batch,
and analytical among-batch) identify the amounts of
data variability that can be attributed to sample col-
lection, processing, storage, and analysis. For the
ELS-I, each type of precision estimate was used to
estimate a different aspect of data variability.
Overall within-batch precision (the total amount of
data variability for samples collected and processed
on a given day) was estimated using field duplicate
data. Analytical within-batch precision (the portion of
the total data variability that occurred during chemi-
cal analysis of the samples collected and processed
on a given day) was estimated using laboratory dupli-
cate data. The differences between the amounts of
overall (field duplicate) and analytical (laboratory
duplicate) within-batch precision indicate the
amount of data variability that occurred during sam-
ple collection, processing, and storage.
Overall among-batch precision (total data variability
among all sample batches) was estimated using data
from field natural audit samples and field synthetic
audit samples. Analytical among-batch precision (the
portion of the total among-batch variability that could
be attributed to measurement imprecision including
temporal effects) was estimated using data from lab-
oratory synthetic audit samples. The field natural
audit samples were also used to provide an estimate
of relative interlaboratory bias, and the field synthetic
and laboratory synthetic audit samples were used to
determine absolute interlaboratory bias and accu-
racy.
COMPARISONS OF PRECISION
ESTIMATES
The four types of precision estimates are expected to
relate in numerically consistent ways. By comparing
the observed relationships between precision esti-
mates to the expected relationships, the quality of
the data can be evaluated. Quantitation limits (based
on the variability of field blank and laboratory blank
measurements) were used both for the evaluation of
individual precision estimates (see Section 4) and for
comparing precision estimates.
The quantitation limits provided a means of classify-
ing the data which was necessary to objectively eval-
uate precision. It is important to understand that the
quantitation limit provided only two precision catego-
ries for each precision estimate (pairs with mean val-
ues greater than zero and pairs with mean values
greater than the quantitation limit), whereas the rela-
tive precision is expected to vary along the entire
range of mean sample values (Mericas et al. 1986).
Therefore, if the sample concentrations in two popu-
lations are sufficiently different, the relative preci-
sions of the sample populations will differ even when
both populations have mean values above a quantita-
tion limit. Such differences may be unimportant
when, for example, the precision estimates for both
populations meet predetermined precision goals.
ELS-I data quality was evaluated by comparing the
relative variability of samples which had means
abovethequantitation limitsagainstthe DQOsestab-
lished prior to the survey. When precision is
expressed in absolute terms (as it was for pH in the
ELS-I), the observed precision can be directly com-
pared against the DQOs.
The inherent differences between the methods used
to calculate precision estimates should also be con-
sidered when comparing chemical measurements
from the ELS-I. Precision estimates for many samples
that have the same concentration (e.g., among-batch
precision using audit samples) can be expressed in
terms of the percent relative standard deviation
(%RSD). Among-batch precision estimates were
therefore directly comparable with the interlabora-
tory precision goals which were part of the DQOs.
Precision estimates for measurements that have a
wide range of values (e.g., field duplicate pairs and
laboratory duplicate pairs) can be calculated as the
root mean square (RMS) of %RSD.The RMS of %RSD
estimates the mean of the %RSD over the observed
range of mean sample pair values. Because the RMS
of %RSD is an estimate of the true %RSD, the within-
batch precision values from survey data are directly
comparable to the interlaboratory precision goals as
set forth in the DQOs for the ELS-I.
EXPECTED RELATIONSHIPS BETWEEN
PRECISION ESTIMATES
Overall within-batch precision estimates were
expected to be numerically larger than analytical
within-batch precision estimates by an amount equal
59
-------
to the variability from sample collection, processing,
and storage (see above). Similarly, overall among-
batch precision estimates were expected to be
numerically largerthan analytical among-batch preci-
sion estimates. Analytical and overall among-batch
precision estimates were expected to be numerically
larger than the corresponding analytical and overall
within-batch precision estimates by amounts that
were equal to the temporal variability.
The exceptions to the four types of expected relation-
ships between precision estimates in the ELS-I are
listed below. Unless specifically noted, the compari-
sons which include overall or analytical within-batch
precision estimates are based on duplicate pairs with
means greater than the quantitation limits. Where the
mean of the audit sample values for total extractable
Al or total Al were less than 0.010 mg L~1,the among-
batch precision estimates for those measurements
were compared to the corresponding within-batch
precision estimates for duplicate pairs with means
less than 0.010 mg L""1. Similarly, where the mean of
audit sample DOC values was less than 5.0 mg L"1,
the among-batch precision for those measurements
was compared to the corresponding within-batch pre-
cision estimate for the duplicate pairs which had
means less than 5.0 mg L~! This convention was also
followed for the comparisons between precision esti-
mates for Al and DOC means which were greater than
0.010 and 5.0 mg L~1, respectively.
Overall within-batch precision estimates were numer-
ically larger than analytical within-batch precision
estimates for all parameters except total extractable
Al, conductance, Mn, total P, and SO4~2 (see Tables 20
and 22). Overall among-batch precision estimates
from FN2 audit samples were numerically largerthan
overall within-batch precision estimates for all
parameters except total dissolved F~, NO3~, initial
ANC pH, and initial BNC pH (see Tables 20 and 23).
This relationship was observed for both FN2 samples
with means greater than the quantitation limits and
for FN2 samples with means less than the quantita-
tion limits. Overall among-batch precision estimates
from FN3 audit samples were numerically larger than
overall within-batch precision estimates for 15 of 23
parameters (see Tables 20 and 23). Exceptions to this
relationship include values for ANC, Cl~, conduc-
tance, air-equilibrated DIG, total dissolved F~, NO3~,
air-equilibrated pH, and SO4~2. Overall among-batch
precision estimates from FH audit samples were
numerically larger than overall within-batch precision
estimates for all parameters except NH4', NO3~, and
SO4~2(seeTables20and 24). Overall among-batch pre-
cision estimates from FL audit samples were numeri-
cally larger than overall within-batch precision esti-
mates for all parameters except total Al, ANC, Mn,
and NH4+, (see Tables 20 and 25). Lastly, analytical
among-batch precision estimates from both LH and
LLaudit samples were numerically largerthan analyt-
ical within-batch precision for all parameters except
conductance, Mn, and SO4
25).
-2
(see Tables 22, 24, and
Exceptions to the expected relationships were gener-
ally associated with the presence of one or more
extreme outliers in the verified data set, with values
close to the detection limit, or with a methodological
problem. Many exceptions involved small differences
in estimated precision. In several cases it was neces-
sary to retain confirmed but questionable values in
the verified data set which were later deleted during
data validation. The confirmed, questionable values
influenced the statistical evaluation of the ELS-I data.
For subsequent surveys, a data qualifier (XO) was
added to ensure that such values were retained in the
verified data set but were not included in statistical
calculations.
This brief discussion of the discrepancies in the
expected ELS-I precision relationships is a first step
in understanding the mechanisms that affect data
variability. A more comprehensive analysis of these
discrepancies is needed to adequately describe the
potential contributing factors and their implications.
The complexity of such an analysis is a subject which
calls for detailed evaluation in a separate report.
60
-------
Section 6
Summary
In general, the ELS-I was conducted smoothly and
efficiently with surprisingly few problems given the
magnitude of the survey. This success may be attrib-
uted to the use of a peer-reviewed research plan, a QA
plan, and operations, training, and methods manuals,
as well as to the efforts of all of the individuals
involved. When problems did occur, they were identi-
fied and generally resolved quickly; this indicates
that the checks and balances fundamental to the QA
program operated effectively.
Implementation of pilot studies prior to the ELS-I
proved to be worthwhile in that many field sampling
and analytical issues which could have caused prob-
lems during the full survey were corrected without
loss of survey data. In particular, a nitrate contamina-
tion problem was identified and partially resolved
prior to initiation of the survey.
The operational QA program appeared to be adequate
to ensure that all samples were collected and ana-
lyzed consistently and that the resulting data were of
known and traceable quality. The field QC and calibra-
tion procedures proved sufficient to detect specific
instrument or operator problems; this was evidenced
by detection of both a faulty pH meter and contamina-
tion of field blanks which was caused by the use of
deionized water from a supply not meeting ELS-I
requirements.
Analytical QC procedures were also sufficient as evi-
denced by the detection and resolution of both a sil-
ica calibration error and of aluminum contamination
which resulted from the use of borosilicate glassware
and from floor sanding. After the ELS-I, the base coor-
dinators recommended improvements in the training
program. This recommendation was addressed by
using the same personnel on later surveys for the
NSWS. All returning NSWS personnel received
"refresher" training in addition to training in proto-
cols which were specific to each subsequent survey.
The data base entry and verification procedures
enabled virtually all transcription, transposition, and
typographical errors on the various forms to be
detected and corrected. A problem was identified
with the error correction procedures in that the
method was slow; the alternative method that was
employed was labor-intensive. The correction proce-
dures were revised for later surveys.
Evaluation of the QA and QC sample data indicated
that the data quality objectives for detectibility and
precision were achievable. The importance of using
field blank measurements during data interpretation
was demonstrated. Little bias was detected among
field stations and analytical laboratories. Evidence
was presented to show that the bias that did exist
could be quantified, and that correction factors could
be applied to the values to aid in data interpretation
(see Appendix A).
These results point to the usefulness of the QA pro-
gram in assuring consistency and reliability of the
data collected in the NSWS. To that end, the ELS-I QA
program was successful in producing data of known
and verifiable quality in accordance with the objec-
tives of the NSWS.
61
-------
Section 7
References
American Society for Testing and Materials, 1984. Annual Book of ASTM Standards, v. 11.01, Standard Specifi-
cation for Reagent Water, D1193-77 (reapproved 1983). ASTM, Philadelphia, Pennsylvania.
Barnes, R. B., 1975. The Determination of Specific Forms of Aluminum in Natural Water. Chem. Geol., v. 15, pp.
177-191.
Driscoll, C. T, J. P. Baker, J. J. Bisogni, and C. L. Schofield, 1983. Aluminum Speciation and Equilibria in Dilute
Acidic Surf ace Waters of the Adirondack Region of New York State, pp. 55-75. In: O. P. Bricker(ed.), Geolog-
ical Aspects of Acid Deposition. Butterworth Publishers, Boston, Massachusetts.
Drouse, S. K., and M. D. Best, 1986. Comparability of Chemical Data Obtained Using ICPES and AAS, 1C, or
Colorimetry. (In preparation.)
Drouse, S. K., D. C. Hillman, L. W. Creelman, and S. J. Simon, 1986. Quality Assurance Plan for the National
Surface Water Survey-Eastern Lake Survey (Phase I--Synoptic Chemistry). U.S. EPA, Las Vegas, Nevada.
Eilers, J. M., D. J. Blick, Jr., and M. Dehaan. 1986. National Surface Water Survey, Eastern Lake Survey- Phase I,
Validation of the Eastern Lake Survey - Phase I Data Base. U.S. EPA, Corvallis, Oregon.
Fountain, J., and D. T. Hoff, 1985 (Draft). AQUARIUS User's Guide. Internal Report for Lockheed-EMSCO, Las
Vegas, Nevada.
Hillman, D. C., J. F. Potter, and S. J. Simon, 1986. Analytical Methods Manual for the National Surface Water
Survey - Eastern Lake Survey (Phase I Synoptic Chemistry). U.S. EPA, Las Vegas, Nevada.
Kramer, J. R., 1984. Modified Gran Analysis for Acid and Base Titrations. Environmental Geochemistry Report
No. 1984-2. McMaster University, Hamilton, Ontario, Canada.
Linthurst, R. A., D. H. Landers, J. Eilers, D. F. Brakke, W. S. Overton, E. R Meier, and R. E. Crowe., 1986. Character-
istics of Lakes in the Eastern United States. Volume I: Population Descriptions and Physico Chemical
Relationships. EPA-600/4-86-007A, U.S. EPA, Washington, D.C. 275 pp.
May, H. M., P. A. Helmke, and M. L. Jackson, 1979. Determination of mononuclear dissolved aluminum in near-
neutral waters. Chem. Geol. 24:259-269.
McQuaker, N. R., P. D. Kluckner, and D. Sandberg, 1979. Chemical analysis of acid precipitation: pH and acidity
determinations. Environ. Sci. and Technol. v. 17, n. 7, pp. 431-435.
Mericas, C. E., M. D. Best, and R. D. Schonbrod, 1986. Measurement uncertainty in the National Surface Water
Survey. (In preparation.)
Morris, F. A., D. V. Peck, M. B. Bonoff, and K. J. Cabbie, 1986. National Surface Water Survey - Eastern Lake
Survey (Phase I -- Synoptic Chemistry). Field Operations Report. U.S. EPA, Las Vegas, Nevada.
O'Dell, J. W., J. D. Pfaff, M. E. Gales, and G. D. McKee, 1984. Technical Addition to Methods for the Chemical
Analysis of Water and Wastes, Method 300.0, The Determination of Inorganic Ions in Water by Ion Chro-
matography. EPA-600/4-85-017. U.S. EPA, Cincinnati, Ohio.
Oliver, B. G., E. M. Thurman, and R. L. Malcolm, 1983. The Contribution of Humic Substances to the Acidity of
Colored Natural Waters. Geochim. Cosmochim. Acta, v. 47, pp. 2031-2035.
Overton, W. S., P. Kanciruk, L. A. Hook, J. M. Eilers, D. H. Landers, D. F. Brakke, D. J. Blick, Jr., R. A. Linthurst, and
M. D. DeHann, 1986. Characteristics of Lakes in the Eastern United States. Volume II: Lakes Sampled and
Descriptive Statistics for Physical and Chemical Variables. EPA-600/4-86-007C, U.S. EPA, Washington,
D.C., 374 pp.
62
-------
Permutt, T. J., and A. K. Pollack, 1986. Analysis of Quality Assurance Data for the Eastern Lake Survey. Final
Report for Lockheed-EMSCO, Las Vegas, Nevada. 47 pp.
Quave, S. A., 1980. Nitrite-nitrate nitrogen in drinking water. Holding time and preservation study. Memo to L. B.
Lobring, U.S. EPA, Cincinnati, Ohio, September 25, 1980.
Rosen, A. E., and P. Kanciruk. 1985. A generic data entry quality assurance tool. SAS (Statistical Analysis Sys-
tem, Inc.) SAS User's Group, Interna tional, Proceeding SUGI 10, Reno, Nevada.
SAS Institute, Inc., 1985. SAS User's Guide: Basics, Version 5 Edition. SAS Institute, Inc. Gary, North Carolina
1290pp.
Skougstad, M. W., M. J. Fishman, L. C. Friedman, D. E. Erdman, and S. S. Duncan (eds.), 1979. Methods for
Determination of Inorganic Substances in Water and Fluvial Sediments: Techniques of Water-Resources
Invest! gations of the United States geological Survey, Book 5. U.S. Government Printing Office, Washing-
ton, D.C.
Stapanian, M. A., T. E. Lewis, M. D. Best, and C. E. Mericas, 1986. Water chemistry methods in acid deposition
research: A comparative study of analyses from Canada, Norway, and the United States. (In preparation.)
U.S. Environmental Protection Agency, 1983 (revised). Methods for Chemical Analysis of Water and Wastes
EPA-600/4-79-020. U.S. EPA, Cincinnati, Ohio.
U.S. Environmental Protection Agency, 1984a. National Surface Water Survey, Phase I. U.S. EPA, Office of
Research and Development, Washington, D.C.
U.S. Environmental Protection Agency, 1984b. National Surface Water Survey, Phase I. Research Plan, A Sum-
mary of Contents. U.S. EPA, Corvallis, Oregon.
Williams, T. J., 1979. An Evaluation of the Need for Preserving Potable Water Samples for Nitrate Testing. J. Am.
Waterworks Association. March:157-160.
Yfantis, E. A., M. A. Stapanian, G. T. Flatman.'E. R Meier, and D. L. Montague, 1986. Comparison of Univariate and
Chemometrical Analysis of Multilabora tory data. Anal. Chem. Acta (in press).
63
-------
APPENDIX A
ANALYSIS OF QUALITY ASSURANCE DATA
FOR THE EASTERN LAKE SURVEY
Prepared by
T. J. Permutt and A, K. Pollack
Systems Applications, Inc.
101 Lucas Valley Road
San Rafael, California 94903
The following document is a final report on the statistical evaluation of quality assurance data for the
ELS-I which was prerpared for Lockheed-EMSCO.
-------
Final Report
ANALYSIS OF QUALITY ASSURANCE DATA
FOR THE EASTERN LAKE SURVEY
SYSAPP-86/040
26 February 1986
Prepared for
Lynn Creelman
Locktieed-EMSCO
1050 East Flamingo Road, Suite 246
Las Vegas, Nevada 89119
Purchase Order IEH392
Prepared by
T. J. Permutt
A. K. Pollack
Systems Applications, Inc.
101 Lucas Valley Road
San Rafael, California 94903
87770:'51B5r
66
-------
CONTENTS
1 INTRODUCTION 1
2 DETECTION LIMITS 4
Decision Limits 5
Detection Limits 7
Quant1tat1on Limits 11
3 DUPLICATE SAMPLES 15
Introduction 15
Method 15
Results 18
4 NATURAL AUDIT SAMPLES 23
Int roduct 1 on 23
Method 23
Results 27
5 SYNTHETIC AUDIT SAMPLES 35
Int roduct 1 on 35
Bias and Precision 35
Measured Versus Theoretical Concentrations 37
Field Versus Laboratory Audits 41
6 CONCLUSIONS AND RECOMMENDATIONS 43
What the QA Data Show 43
Synthetic Audits 46
Iron 47
Detection Limits 47
References 49
BSiesr i
67
-------
FIGURES
2-1 Summary of sulfate measurements of 245 field blanks 8
2-2 Calculation of the detection limits (0) 12
4-1 Interlaboratory bias and trend 1n pH of natural audits
from B1g Moose Lake (EMSI and Versar only) 23
4-2 Interlaboratory bias and trend 1n pH of natural audits
from Big Moose Lake (four laboratories) 25
5-1 Concentration of ammonium measured in laboratory high
synthetic audit samples 35
5-2 Concentration of dissolved inorganic carbon (initia1)
measured in laboratory high audit samples , 39
5-3 Concentration of silica measured 1n laboratory high
synthetic audit samples 41
68
-------
TABLES'
2-1 System decision and detection limits 9
3-1 Various summary statistics for the standard deviation of
duplicate pairs 1n a Monte Carlo experiment 16
3-2 H1th1n-batch precision for pH measurements by contract
1 aboratoM es 18
3-3 W1th1n-batch precision of laboratory duplicates 19
3-4 H1th1n-batch precision of field duplicates 20
3-5 Very high RSOs 1n laboratory duplicate pairs with a mean
measurement above the quantltatlon limit 21
4-1 Overall precision estimated from natural audits 27
4-2 Overall precision of pH measurements 1n natural audits 28
4-3 Relative Interlaboratory bias estimated from natural audits
and controlling for measurement trend 29
4-4 Relative Interlaboratory bias estimated from natural audits
and Ignoring measurement trend 30
4-5 Interlaboratory bias estimated from natural audits 32
4-6 Trends (per month) 1n measurements of natural audit
samples (B1g Moose Lake) 33
5-1 Field and laboratory high synthetic audits 37
5-2 Field and laboratory low synthetic audits 38
6-1 Summary of estimates of precision 43
111
69
-------
1 INTRODUCTION
The National Surface Water Survey, designed and begun by the U.S. Environ-
mental Protection Agency 1n cooperation with the National Add Precipita-
tion Program, 1s a three-phase project Intended to document the chemical
and biological condition of lakes and streams considered susceptible to
the effects of acidic deposition. Phase IA of the program 1s a survey of
lake chemistry. It consists of the Eastern Lake Survey, with which this
report 1s concerned, and the Western Lake Survey, which 1s currently
underway. The primary objectives of the phase IA survey are to determine
how many lakes have low pH or alkalinity 1n regions of the United States
potentially sensitive to acidic deposition and to understand the chemical
composition of these lakes. From the lakes sampled 1n phase IA some will
be selected for Intensive study 1n phases II and III of the project (EPA,
1984).
Four regions of the United States were selected for study In phase IA:
the Northeast, the Southeast, the upper Midwest, and the mountainous
West. These regions were chosen because they are known to contain an
abundance of low-alkalinity lakes, which are considered sensitive to
acidic deposition. The Eastern Lake Survey covered the Northeast, South-
east, and upper Midwest regions and was completed 1n late 1984.
The sampling plan 1s a stratified design. Each region 1s divided Into
subreglons, and within a subreglon lakes are divided Into three alkalinity
classes. In addition, there are minor stratification variables: lake
size, elevation, and watershed size. An extensive quality assurance/
quality control (QA/QC) program was designed and carried out to assure
high-quality data and to Identify any problems 1n the collection, pro-
cessing, and analyzing of the lake water samples (Drouse et a!., 1985).
While the primary purpose of the QA/QC data was to Identify and correct
potential data quality problems during the survey, they also serve to
assess the data quality that was achieved, which 1s the subject of this
report.
Two-liter lake samples were collected by helicopter crews and processed by
one of eight field laboratories. At the field laboratories the samples
were split Into seven allquots, labeled (by coded Identification number,
not by lake name), and shipped to one of four contract laboratories for
85185r 2
70
-------
chemical analysis. Each field laboratory was supposed to send all of Its
samples (audit and routine) to one contract laboratory, but this protocol
was not always maintained. Four contract laboratories analyzed the lake
samples: EMSI analyzed samples collected by four of Its field labora-
tories from 8 October to 7 November 1984; Versar analyzed samples collec-
ted by two of Its field laboratories and two of EMSI's field laboratories
from 7 October to 16 November 1984; Global analyzed samples collected by
Its field laboratory and two of Versar's field laboratories from 20 Octo-
ber to 29 November 1984; and one U.S. Geological Survey field laboratory
collected samples after all other field laboratories completed their
sampling, from 2 December to 14 December 1984.
The sample load for a field laboratory was restricted to 24 routine lake
samples per day. A batch Is defined as the set of samples collected by a
single field laboratory on a single day (I.e., 24 routine samples per
batch). Each batch was Identified by a three-digit code. Over the dura-
tion of the Eastern Lakes Survey, a total of 127 batches were analyzed.
One duplicate sample was obtained at one lake (usually the first) sampled
1n each batch.
Each lake was visited only once during the survey and each sample was
processed by one field station and one contract laboratory. Consequently,
some Important aspects of the quality of the data cannot be judged from
the routine data themselves. Systematic differences between laboratories
could result from minor differences 1n Instrumentation or procedure or
from real differences between the sets of lakes measured by the different
laboratories. Similarly, variations among measurements by one laboratory
could reflect real differences between lakes or simply the Imprecision of
the measurements. The routine data do not allow these possibilities to be
distinguished. The quality assurance data, however, make H possible to
estimate the likely magnitudes of various kinds of error. Variations
larger than these 1n the routine measurements may confidently be supposed
to represent actual differences between lakes.
Several kinds of blind QA samples are used to estimate the components of
the error 1n the wnole system from collection to analytical determina-
tion. For one, there are duplicates of the routine samples. These dupli-
cates are of two types, called field and laboratory duplicates. Field
duplicates are second samples collected from one lake 1n each batch and
processed 1n parallel with the routine samples. Laboratory duplicates are
splits made at the analytical laboratories so that the analysis, but not
the routine sample processing, 1s duplicated. Both types of duplicate
measurements Indicate, by the difference between the measurements of a
pair, the repeatability of measurements by a given laboratory on a given
day. The laboratory duplicates Indicate the the repeatability of the
analytical determination alone, the field duplicates Indicate the repeata-
esissr 2
71
-------
b1!1ty of the entire system of measurement from collection to analysis.
Neither type tells us anything about variation across laboratories or
across days, such as would be produced by errors 1n calibration.
For this we have the natural and synthetic audit samples. Natural audit
samples were prepared dally from stored lots of water originally collected
from B1g Moose Lake 1n the Adlrondacks and Lake Superior. Synthetic audit
samples were prepared dally by diluting concentrate lots of known composi-
tion. There are both field and laboratory synthetic audits: field syn-
thetic audits were shipped to the field stations 1n two-liter containers
and were divided Into allquots at the field stations, whereas laboratory
synthetic audits arrived at the field stations ready for reshlpment to the
contract laboratories and thus skipped the processing at the field sta-
tions. The natural audits are field audits. These samples thus furnish
repeated measurements of the same thing, by the same and by different
laboratories, assuming stability of the natural audits 1n storage and
consistency 1n preparation of the synthetic audits. Unlike the dupli-
cates, therefore, they give Information about relative bias between
laboratories and about reproduclblHty of measurements from day to day.
In addition, there 1s some Independent Information about what 1s 1n the
synthetic audit samples, namely the recipes for them. Unfortunately, this
Information turns out to be very unreliable.
The quality assurance plan also provided for blank samples to reveal any
problems of contamination and also study the detectablllty of analytes at
low concentrations. There were laboratory blanks, which are not discussed
1n this report, and field blanks, which were samples of delonlzed water
processed at the field stations like routine samples.
This report discusses methods we developed for estimating measurement
errors from the quality assurance data as well as the results. He
consider first the evaluation of detectablHty from field blank data, then
the analyses of duplicate data and natural and synthetic audits. Our work
1s concerned with 24 parameters measured by the contract laboratories.
Mineral acidity and carbonate alkalinity were not analyzed because most
of the measurements were missing or considered unreliable. Nitrate 1s
Included, but 1t 1s being studied 1n much greater detail by Liggett
(1985).
85185^ 2
72
-------
2 DETECTION LIMITS
For the purpose of evaluating the detectab111ty of various parameters at
low levels, the quality assurance plan for the Eastern Lake Survey pro-
vided for some 245 field blank samples. A Van Dorn sampler was filled
with delonlzed water, and allquots were prepared at the field stations 1n
the same way as routine samples. The blank samples were then shipped to
the contract laboratories along with routine samples. The labels were
coded so that the laboratories were blind as to which samples were blank.
Because the field blank samples were processed as much like the routine
samples as possible, measurements of the blanks are subject to errors from
all the sources that affect the routine measurements. Internal blanks
within the contract laboratories were also measured and used to calculate
Instrument detection limits. These Instrument detection limits are
unlikely to be achieved 1n routine samples because the Internal blanks are
Immune to three kinds of error that affect routine (and field blank)
measurements. First, the Internal duplicates are not analyzed blind, and
analyses of samples known to be blanks may be expected to have rather less
variable results than blind analyses. Second, the field blank samples,
like the routine samples, are exposed to varying degrees of contamination
1n processing before they reach the laboratory, whereas the Internal
blanks are not. Third, errors 1n calibration of an Instrument do not
affect the calculated Instrument detection limit. The Instrument detec-
tion limit may be calculated from the standard deviation of 10 measure-
ments of Internal blanks on the same day. If the calibration 1s wrong
that day, the mean, but not the standard deviation, will be affected. In
contrast, the field blanks are analyzed on many different days, and so
variation 1n field blanks reflects errors In calibration.
Our definitions and method of calculating system detection and decision
limits from measurements of field blanks follow In spirit those of Hubaux
and Vos (1970). The details are different for a number of reasons.
Hubaux and Vos were concerned with allowing for calibration error when the
measurements are all from one calibration; we are not because the measure-
ments are from many calibrations so that their variation Includes calibra-
tion error. They were also concerned with allowing for uncertainty about
the long-run distribution of measurements of blanks when only a few blanks
are measured; we are not because there are 245 field blanks. On the other
•siesr 3
73
-------
hand, by using nonparameteMc techniques, we are able to dispense with
their assumption that measurements have normal distributions. This 1s
fortunate because the field blank data clearly do not follow normal dis-
tributions.
Following Hubaux and Vos, we distinguish two kinds of limits: decision
limits and detection limits. We also calculate a third kind of limit,
sometimes called a quantHatlon limit. The decision limit 1s a measure-
ment that reliably Indicates a level above background. The detection
limit 1s a concentration that would probably be detected 1f that concen-
tration were present. The quantHatlon limit Is a concentration that can
be measured with reasonable precision. These concepts are explained 1n
turn below.
DECISION LIMITS
We define the decision limit, following Hubaux and Vos, as "the lowest
signal that can be distinguished from the background." It 1s thus a mea-
surement so high that 1t rarely occurs 1n blank samples. If this high a
measurement does occur, 1t 1s therefore a more or less reliable Indication
that the sample 1s not a blank; the more rarely such a measurement occurs
1n blanks, the more reliable an Indication 1t 1s. If the decision limit
1s set at the upper a quantlle of the distribution of blanks, a measure-
ment above the decision limit can be said to be significantly different
from background at level a. This means that the probability of erron-
eously calling a measurement significant that 1s really background 1s only
It remains only to choose an error rate o that we are willing to accept
and then to estimate the upper a quantlle of the background distribu-
tion. Provided a 1s not too small (at least several times 1/245), we can
estimate the quantlle simply, by the observed quantlle of the 245 measure-
ments. For example, for o - .05, the decision limit Is taken to be the
95th percentlle of the measurements of blanks. Measurements higher than
tnls occur only 5 percent of the time for blank samples, so such a
measurement 1s a significant Indication (at the .05 level) that the sample
1n question 1s not a blank.
There 1s no consensus on what the probability of error a should be. It
might seem desirable to make o very small, perhaps a fraction of 1 per-
cent. This approach has two drawbacks. First, 1t 1s Impossible to reli-
ably estimate very high quantlles without very many observations. For
example, suppose that a certain kind of error, say an unusual contamina-
tion, affected one sample 1n 100 and that decision limits were based, as
they often are, on sets of 10 samples. Suppose that the decision limit
85185T 3
74
-------
was to be the 99tn percentlle. In nine out of every 10 sets of 10
samples, none of the samples would be affected by the unusual error, so
that the 99th percentlle would be underestimated. In one out of 10 sets,
a measurement would be affected, and then such an error would be assumed
to happen In about one out of 10 measurements; the 99th percentlle would
thus be overestimated. Quite simply, we get a very poor Idea of what
happens 1n one out of 100 measurements (the 99th percentHe) by looking at
10 measurements.
Second, even when there are enough measurements to estimate a very high
quantlle, It may not be desirable to do so. To continue the example, with
245 observations we are likely to see two or three of those unusual
errors. The 99th percentHe of the 245 measurements will therefore
probably be one of those unusual measurements. The question 1s, do we
want the reported decision limit to reflect this rather extreme behavior
of the measuring process at Its worst, or somewhat more typical behavior?
Our answer 1s, the more typical behavior. Accordingly, we use the 95th
percentHe of the measurements of blanks as the decision limit. Any
measurement above our calculated decision limit is thus significantly
different from background at a significance level (or error rate) of
a - .05.
It 1s usual to calculate a decision limit from the mean (7) and standard
deviation of the measurements of blanks. For example, the decision limit
might be taken to be 7 + 3 s and the detection limit, which will be dis-
cussed later, 6 s. More commonly, 3 s 1s used as the detection limit, and
this corresponds to a decision limit of 7 + 1.5 s. So defined, the
decision limit can be calculated from very few measurements, even as few
as two. However, sucn a definition has two serious disadvantages. First,
1t 1s Impossible to know the error rate. The measurements are sometimes
assumed to follow a normal distribution; 1f they did, 7 + 1.5 s would be
about the 93rd percentlle and the error rate would be about 7 percent.
Unfortunately, the validity of this assumption cannot be checked unless
there are so many measurements that the assumption 1s unnecessary. On the
other hand, even 1f the distribution 1s not normal, Long and Wlnefordner
(1983) point out that a mathematical result called Chebyshev's Inequality
limits the error rate. Unfortunately, the limit Is crude: anywhere from
none up to 4/13 of a distribution can be at least 1.5 standard deviations
above the mean.
The second disadvantage of basing the decision limit on the mean and stan-
dard deviation 1s that these statistics are very sensitive to outliers.
If there are a few very high measurements, the standard deviation 1n
particular may be more heavily Influenced by these few than by all the
rest. Here again, the decision limit would reflect the extreme rather
than the typical behavior of the system. Nevertheless, 1n addition to the
BSiesr 3
75
-------
95th percentlle, for purposes of comparison we calculate a decision limit
•s 1.5 standard deviations above the mean of the measurements of blanks.
The case of sulfate Illustrates all these points. Figure 2-1 summarizes
the 245 measurements of sulfate 1n field blanks. The distribution 1s very
skewed to the right; 1t does not look like a normal distribution at all.
The highest value (2.61 mg/1), 1s more than eight times the second highest
(0.31 mg/1), which 1s 1n turn twice the third highest (0.16 mg/1). The
mean 1s 0.043 mg/1 and the standard deviation 1s 0.17 mg/1. Both are
heavily Influenced by the outliers: eliminating just the one largest
measurement would reduce the mean to 0.032 mg/1 and the standard deviation
to 0.034 mg/1. The 95th percentHe 1s 0.093 mg/1. There are several
measurements fairly close to that above and below; the nearest are 0.085
and 0.094 mg/1. On the other hand, 1.5 standard deviations above the mean
is 0.30 mg/1; this would be the 93rd percentile of a normal distribution,
but 1t 1s above the 99th percentile for this distribution. There 1s only
one measurement anywhere near 0.30 mg/1, and only one other that 1s
higher. Thus the 7 + 1.5 s decision limit mainly reflects the two largest
measurements out of 245.
Table 2-1 shows the system decision limits for all parameters (except pH)
analyzed by the contract laboratories, as well as the detection limits,
which we discuss next. Both the parametric version based on mean and
standard deviation and the nonparametric version we recommend are shown.
DETECTION LIMITS
Hubaux and Vos call a detection limit "the limit at which a given analyti-
cal procedure may be relied upon to lead to detection...." "Detection"
here can be understood to mean a measured concentration above the decision
limit, for 1f the measurement 1s above the decision limit it can be
reliably asserted that the analyte 1s present 1n more than background
amounts, i.e., that 1t has been detected. The question thus becomes, how
high does the true concentration have to be before the measured concentra-
tion can be relied on to be above the decision limit? If D 1s the
detection limit, then when the true concentration 1s D, the measured
concentration should be above the decision limit except with some small
probability 6. Thus, two kinds of error are relevant to the definition of
a detection Hm1t. First, false detection occurs when the measurement 1s
above the decision limit but the sample 1s In fact a blank. The decision
limit was chosen to limit the probability of this kind of error to a.
Second, failure to detect occurs when the sample 1s not a blank but the
measurement 1s below the decision limit. The higher the true concentra-
tion, the less probable 1s this kind of error. The true concentration at
which the probability 1s just e is the detection limit.
85185r 3
76
-------
MOMENTS
OUANTILES(DEF'3>
EXTREMES
1 245
WAN •.0428653
SID DEV •.168*75
'KEVNESS 14.7423
USS 7.33883
CV 393.939
T:MEAN-» 3.9733Z
jCN RANK MM*
HIM ~- • 223
2.65»«
2.35*
2.05*
1.75*
CO
1.45*
1.15*
• .»5»
0.55*
•
^ •
».25«
*•
•
SUM WCTS 245 1 MX MAX 2. SI 99X 0.15 LOWEST HICNCST
*""..„ . I».4S3 75X 03 0.046 95X 0.093 -0.023 0.137
VARIANCE •.•282494 S0XMED ».03 90X 0.066 -0.0Z3 0.142
KURTOSIS 225.66 2bXOI 0.01 I»X 0 -0.023 0.IS
«S -r 6.89285 0X MIN -0.023 5X -0.004 -0.02 0.31
"0 MEAN •.0107379 |X -0.023 -0.02 2.61
PROBXTI 0.0001 RANGE 2.633
PROBXSI «.0001 03-01 0.036
MODE 0.03
HISTOGRAM * BOXPLOT NORMAL PROBABILITY PLOT
1 • 2.65* •
2.35
2.05
1.75
1.45
1.15
•.as
• .55
I •
• .25
• t
****
#*#*++ •
******
• MAY REPRESENT UP TO 5 COUNTS -Z -1 • «I *2
FIGURE 2-1. Summary of sulfate measurements of 245 field blanks.
-------
TABLE 2-1. System decision and detection limits.
System Detection Limit
Parametric
Parameter* (3 s)
Calcium
Magnesium
Potassium
Sodium
Manganese
Iron
Aluminum (extractable)
Chloride
Sulfate
Nitrate
Silica
Fluoride (total)
DOC
Ammonium
Acidity (peq/1)
Alkalinity (peq/1)
Conductivity (pS/cm)
DIC (equilibrated)
DIC (Initial)
Phosphorus (total)
Aluminum (total )
0.043
0.012
0.11
0.060
0.028
0.037
0.012
0.11
0.52
1.0
0.70
0.010
0.76
0.17
45
17
2.8
0.17
0.33
0.011
0.084
Nonparametrlc
(2
-------
The detection limit tnus depends on the decision limit and, through 1t, on
the distribution of measurements of blanks. The detection Hm1t also
depends on the distribution of measured concentrations for various small,
positive true concentrations: 1n fact, the detection limit 1s that true
concentration for which the e quantHe of measured concentrations 1s the
decision limit. Unlike the distribution of blanks, the distribution of
measurements at the detection limit has not been observed. To observe 1t
would require audit samples with true concentration equal to the detection
limit, and to get these entails knowing the detection limit 1n advance.
Instead, the distribution at the detection limit must be Inferred from the
distribution of blanks.
Let us therefore look again at the distribution of sulfate measurements 1n
blanks, which was presented 1n Figure 2-1. Some new points are now of
Interest. The observations are mostly positive: 91 percent are more than
zero, the median 1s 0.030 mg/1, and the mean 1s 0.043 mg/1. There are two
reasons for this result. First, while the field blank samples start out
as delonlzed water, they go through the same processing as the lake
samples before they get to the laboratory and thus may be contaminated 1n
the process. It 1s likely that many blanks actually do contain measurable
amounts of sulfate by the time they are analyzed. Second, the measurement
error for a blank sample 1s very likely to be positive. It 1s not
Impossible to measure a concentration as negative because of the calibra-
tion of the Instrument: If an unknown produces a signal less than that of
the calibration blanks, the measured concentration 1s negative. Still,
errors are much more likely to be positive than negative: there are many
ways to overestimate a concentration that 1s nearly zero, and not many
ways to underestimate 1t. Furthermore, even when the calibrated Instru-
ment produces a negative reading, this reading might not appear In the
reported data: 1t might be recorded as zero, or the Instrument might be
recalibrated and this reading might be discarded.
These effects on the distribution of blanks are Important precisely
because the distribution of measurements at the detection limit 1s not
likely to be affected in quite the same way. Suppose that the true con-
centration of sulfate in a lake sample 1s some small positive amount x and
that this sample 1s measured many times. What does the distribution of
these measurements look like? It Is reasonable to assume that contamina-
tion affects this sample the same way 1t affects blanks since contamina-
tion simply adds sulfate to the sample, no matter how much was 1n it to
start with. (This would not be reasonable for parameters that are not
additive, e.g., conductivity.) However, 1t 1s not reasonable to assume
that measurement error affects this sample the same way It affects
blanks. Along with positive errors, negative errors of up to x are now
possible, and 1f most errors are smaller than x, negative and positive
esiesr 3
10
79
-------
errors will be about equally likely. Furthermore, the reporting problems
with negative measurements will have little effect since there are few
negative measurements: only an unusually large (more than x) negative
error produces a negative measurement. On the other hand, the upper part
of the distribution of measurements of this sample might resemble that for
blanks, shifted by x, since the sources of positive errors are the same.
As x moves away from zero, then, the shape of the lower half of the dis-
tribution changes. The lower half 1s likely to look more and more like a
mirror-Image of the upper half. The upper half keeps about the same shape
as for x - 0, a blank.
The problem 1s to find an x for which the lower B quantlle of the distri-
bution 1s the decision limit. With a true concentration of x, the mea-
surements will be above the decision limit except with probability e, and
so the analyte will be "detected" with probability 1 - B. This is the
detection limit.
Figure 2-2 shows a solution. It Is assumed that 1n the distribution of
measurements at true concentration D, the half above the median is the
same as the upper half of the distribution of measurements of blanks, but
shifted by D. The lower half 1s a mirror-Image. The detection limit D is
the sum of A, the distance from the median of measurements of blanks to
the decision limit, and B, the distance from the lower B quantlle to the
median at true concentration D. By symmetry, B equals C, the distance
from the median to the upper B quantlle, and by assumption, this distance
Is the same at D as for blanks. Under these assumptions, the decision
limit can be calculated from the distribution of measurements of blanks as
D - A + B
« A + C
where o^ and q. are the upper a and B quant lies and q 5 1s the median. We
use B • o • .05, giving a detection limit of 2(P95 - ^g), I.e., twice the
difference between the 95th percent He and the median of the measurements
of blanks.
As with decision limits, 1t 1s more usual to calculate detection limits
from the standard deviation of the measurements of blanks. The detection
limit is often reported as 3 s, where s 1s the standard deviation. This
may be taken to correspond to our procedure with a decision limit of
T + 1.5 s, where T 1s the mean, and with the mean replacing the median.
The error rates, assuming a normal distribution, would be o » B * .07. Of
BSiesr 3
11
80
-------
Distribution of measurements
of blanks (true concentration zero)
Distribution of measurements
at true concentration D
Measured
concentration
Decision limit
FIGURE 2-2. Calculation of the detection limit (D).
for discussion.
See the text
65185
12
81
-------
course the same detection limit gives different combinations of o and 6
with different decision limits. For example, with 7 + 3 s as the decision
limit and 3 s as the detection limit, and stm assuming normality, a -
0.003 but B « 0.5, which can hardly be called reliable detection.
Basing detection limits on the standard deviation has the same disadvan-
tages as apply for decision limits. The standard deviation 1s very sensi-
tive to outliers. Since the distributions are not normal, the error rates
cannot be known. In addition, to assume normality 1s to assume the same
distributional shape for blanks and nonblanks; but, as we have argued, the
one 1s skewed and the other may be almost symmetric. Still, we compute a
3 s detection limit for purposes of comparison.
For parameters that are not additive, like conductivity, the Interpreta-
tion of the detection limit 1s a little complicated; Indeed, the concept
of detection limits may not be very useful for conductivity. Conductivity
different from that of a blank will be detected reliably 1f the conduc-
tivity of the sample exceeds that of a blank by more than the detection
limit. For concentrations, on the other hand, the difference between the
true concentration when measured and the background added 1n processing 1s
simply the true concentration 1n the sample to begin with. So the Inter-
pretation for concentrations 1s more straightforward: a concentration
that 1s above the detection limit before processing will be detected
reliably. The detection limits for all chemical parameters (except pH)
analyzed by the contract laboratories are given 1n Table 2-1. Again we
show both the traditional detection limit based on standard deviation and
the nonparametHc version we recommend.
QUANTITATION LIMITS
If the true concentration 1s at the detection limit, the measured concen-
tration will probably be above the decision limit: this 1s how those two
limits were chosen. However, 1f the median background concentration 1s
small, the decision limit 1s only about half the detection limit. So
negative relative errors of as much as 50 percent will occur with proba-
bility B 1n measuring such low true concentrations. Smaller relative
errors are expected at higher true concentrations. If the absolute stan-
dard deviation were the same as for blanks, the relative standard devia-
tion would be more than 10 percent for any true concentration up to 10 s
and less than 10 percent for any above 10 s. This 1s only saying that s
1s more than 10 percent of any number up to 10 s, which has sometimes been
called the limit of quantltatlon (e.g., by Long and Wlnefordner, 1983).
We use 1t 1n the next section 1n computing estimates of precision:
samples believed to have true concentrations below the quantltatlon limit
are excluded from estimates of precision since high precision 1s not
expected at such low concentrations. Quantltatlon limits are shown 1n
Table 2-1.
esiesr 3
13
82
-------
3 DUPLICATE SAMPLES
INTRODUCTION
At one lake each day, usually the first, each helicopter crew collected a
second sample. This duplicate sample was processed by the field station
1n the usual way and shipped to the contract laboratory with a label coded
so that 1t was Indistinguishable from the routine samples 1n the batch.
Differences between measurements of these field duplicate samples reflect
analytical error and also errors at the lake and at the field station.
For example, the sampling procedure might vary slightly from one sample to
the next, or one of the samples might be contaminated 1n preparing the
allquots at the field station.
In addition, duplicate samples 1n each batch were created at the contract
laboratory by splitting one sample. Differences 1n measurements of these
laboratory duplicates reflect only analytical errors at the contract
laboratory, and only variation within a batch and not from day to day.
For example, unless the Instrument 1s recalibrated within the batch, any
error 1n calibration will not be seen as a difference within a duplicate
pair.
Field and laboratory duplicate pairs are thus quite different 1n origin
and provide Information on different aspects of data quality. Neverthe-
less, the form of the data 1s the same—pairs of measurements of the same
thing. Our method of analysis 1s also the same for both, and so we
discuss them together here.
METHOD
Since the analysis 1s simplest for the three measurements of pH by the
contract laboratories, we begin there. Errors 1n pH measurements are
believed to be about the same size for any pH over the relevant range; 1n
any case, since pH 1s already a logarithm, there 1s no point 1n taking
relative standard deviations. The standard deviation of each duplicate
pair 1s
•siesr i,
14
83
-------
which can be written more simply as
s - |R - D|//2
where R and D are the routine and duplicate measurements. Now the ques-
tion arises of how to combine the standard deviations for duplicate pairs
to get an estimate of precision. According to statistical theory, the
right summary statistic 1s the pooled or root-mean-square standard devia-
tion
RMS standard deviation - (••••
where n 1s the number of pairs.
To show the different effects of using this and other summary statistics
for the standard deviation, we did a Monte Carlo experiment. Consider
an Imaginary lake with pH 5.08 and measurements of pH subject to errors
normally distributed with mean zero and standard deviation 0.03.
(Actually, this approximately describes B1g Moose Lake 1n the natural
audit measurements.) We drew 10 sets of 100 pairs of samples from such an
Imaginary lake. We calculated the standard deviation of each pair, and
then we computed four summary statistics for the 100 standard deviations
1n each set: the root mean square (RMS), the mean, the median and the
geometric mean. The results are shown 1n Table 3-1. As we would expect
from statistical theory, the 10 root mean squares are clustered around the
actual measurement error of 0.03, and all three other summary statistics
are systematically too small.
For parameters other than pH there are two complications. First, errors
are expected to be roughly proportional to concentration, at least at hlgti
concentrations. It 1s therefore usual to report precision 1n the form of
relative standard deviation (RSD). We calculate RSD for each duplicate
pair by dividing Us standard deviation by Its mean, and again we use root
mean square as a summary statistic for RSDs. Second, the proportional
relationship breaks down at very low concentrations: as concentration
approaches zero, measurement error does not vanish. Therefore, RSDs may
be very high at low concentrations. What 1s of Interest 1s the RSD at
concentrations that are not extremely low. We therefore compute the RMS
RSD for pairs with mean greater than the quantltatlon limit, defined as 10
times the standard deviation of measurements of field blank samples.
We note that the theoretical argument for using the RMS of RSDs, unlike
standard deviations, 1s only approximate. However, the approximation 1s
esissr n
15
84
-------
TABLE 3-1. Various summary statistics
for the standard deviation of duplicate
pairs 1n a Monte Carlo experiment.
RMS
0.030
0.027
0.029
0.033
0.030
0.031
0.030
0.031
0.030
0.026
Mean
0.024
0.022
0.024
0.026
0.025
0.024
0.024
0.025
0.024
0.022
Median
0.020
0.019
0.021
0.025
0.020
0.020
0.020
0.021
0.019
0.024
Geometric
Mean
0.018
0.014
0.017
0.017
0.017
0.017
0.017
0.018
0.016
0.016
Ten sets of 100 duplicate pairs are
represented. All 2000 measurements
have a normal distribution with mean
5.08 and standard deviation 0.03.
The standard deviation of each pair
was calculated, and then four summary
statistics were calculated for the
100 standard deviations 1n each set.
esiesr s
16
85
-------
standard, and 1t Is reliable whenever RSDs are much less than 1 (100
percent), which 1s the case here. Use of the RMS RSD has been recommended
by the Environmental Protection Agency for Its QA work (EPA, 1983).
RESULTS
Table 3-2 shows the estimates of precision for the three pH parameters.
As expected, the figures for field duplicates are slightly higher than
those for laboratory duplicates, reflecting some variability 1n the field
procedure or perhaps some real difference 1n the pH of water collected 1n
successive samples. Most of the variability 1s already present 1n labora-
tory duplicates, however, and so must be attributed to analytical Impre-
cision.
Tables 3-3 and 3-4 show the estimates of relative precision of measure-
ments other than pH from field and laboratory duplicate pairs. The pre-
cision (RMS RSD) 1s calculated for all pairs with a positive mean measured
concentration and for just those pairs with a mean above the quantltatlon
limit. Again, the estimates from field duplicate pairs are generally a
little higher than those from laboratory duplicate pairs, as expected.
Extractable aluminum, sulfate, conductivity, and phosphorus are excep-
tions.
The estimates of precision from laboratory duplicate pairs for half a
dozen parameters, Including all these exceptions, are each heavily Influ-
enced by a single very high RSD. Table 3-5 Identifies these outliers and
shows the RSD recalculated when they are eliminated.
BSlSSp K
17
86
-------
TABLE 3-2. W1thin-batch precision for pH
measurements by contract laboratories
(RMS standard deviation).
Laboratory Field
Duplicates Duplicates
Parameter (N « 127) (N •» 125)
pH (equilibrated) 0.077 0.085
pH (alkalinity) 0.042 0.052
pH (acidity) 0.065 0.075
85185 5
18
87
-------
TABLE 3-3. Hlthln-batch precision of laboratory duplicates.
Parameter8
Calcium
Magnesium
Potassium
Sodium
Manganese
Iron
Aluminum
(extractable)
Chloride
Sulfate
Nitrate
Silica
Fluoride (total)
DOCC
Mean > 5 mg/1
Mean < 5 mg/1
Ammonl um
Acidity (peq/1)
Alkalinity (yeq/1 )
Conductivity (yS/cm)
DIG (equilibrated)
DIG (Initial)
Pnospnorus (total )
Aluminum (total)
Pairs
No. of
Pairs
123
121
125
123
102
120
123
127
127
122
127
127
126
121
112
116
125
126
127
125
126
with Mean > 0
RMS RSD (%)
0.89
0.63
18
13
21
14
13
23
11
13
2.5
2.5
6.8
7.1
28
30.
11
7.1
3.7
13
19
Quant1tat1on
Unit
(10 sB)b
0.15
0.039
0.37
0.20
0.093
0.12
0.039
0.39
1.7
3.4
2.3
0.034
2.5
0.56
150
57
9.0
0.56
1.1
0.037
0.28
Pairs With
No. of
Pairs
123
121
98
119
36
71
45
85
104
0
64
120
96
41
55
19
5
86
122
99
93
20
9
Mean > 10 SR
RMS RSD (X)
0.89
0.63
1.2
0.97
1.6
3.0
18
8.2
12
--
2.2
2.5
2.3
2.4
2.2
1.4
10.
2.1
11
2.4
2.4
19
5.4
a All parameters are measured 1n mg/1 unless otherwise stated.
b SB 1s standard deviation of measurements of blank samples.
c Precision goals for DOC are different above and below 5 mg/1.
85185P 5
19
88
-------
TABLE 3-4. H1tn1n-batcn precision of field
Parameter8
Calcium
Magnesium
Potassium
Sodium
Manganese
Iron
Al um1 num
(extractable)
Chloride
Sulfate
Nitrate
Silica
Fluoride (total)
DOCC
Mean > 5 mg/1
Mean < 5 mg/1
Ammonium
Acidity (ueq/1)
Alkalinity (yeq/1 )
Conductivity (viS/cm)
DIC (equilibrated)
QIC (Initial)
Phosphorus (total )
Aluminum (total)
Pairs
No. of
Pairs
125
125
125
125
82
118
112
123
125
116
120
125
125
113
118
121
125
124
125
122
125
wltn Mean > 0
RMS RSD (%)
2.3
2.3
4.7
4.2
57
97
39
15.
6.6
140
44
8.2
9.8
34
700
20.
1.9
12
6.9
37
35
duplicates.
Quant1tat1on
Limit
(10 sB)b
0.15
0.039
0.37
0.20
0.093
0.12
0.039
0.39
1.7
3.4
2.3
0.034
2.5
0.56
150
57
9.0
0.56
1.1
0.037
0.28
Pairs
No. of
Pairs
125
125
82
121
6
32
9
85
115
1
50
62
105
59
46
0
3
90
125
94
85
4
1
With Mean > 10 SR
RMS RSD (%}
2.3
2.3
3.7
4.3
11
10.
11
17
6.5
65
2.7
9.0
10.
12.
5.8
__
74
10.
1.9
5.0
3.7
9.7
24
8 All parameters are measured 1n mg/1 unless otherwise stated.
b Sg 1s standard deviation of measurements of blank samples.
c Precision goals for DOC are different above and below 5 mg/1.
20
89
-------
TABLE 3-5. Very high RSDs 1n laboratory duplicate pairs with a mean
•easuement above the quantHatlon limit.
__
Highest Root Mean Square
Highest RSD RSD RSD Eliminating
Parameter (percent) Batch (percent) this Pair (percent)
Aluminum (extractable) 118 518 9.7 3.3
Chloride 73 105 7.1 2.1
Sulfate 117 317 5.9 1.3
Acidity 21 605 1.7 1.3
Conductivity 116 510 3.7 0.72
Phosphorus (total) 84 500 6.8 2.4
85185 5
21
-------
4 NATURAL AUDIT SAMPLES
INTRODUCTION
Large quantities of water were collected from B1g Moose Lake and Lake
Superior some months before the Eastern Lake Survey. Two-liter samples of
this water were shipped dally to the field stations, where they were
handled like the routine samples; the field stations prepared allquots and
shipped them to the contract laboratories with coded labels along with the
day's routine samples.
Measurements of these natural audit samples give Information about two
aspects of the measuring process and consequently about the quality of the
routine data. First, they are measurements by different laboratories of
the same thing, provided the actual composition of the samples 1s con-
stant. If the measurements by these laboratories differ systematically,
we cannot tell which of two laboratories 1s right, 1f either, because we
have no way of knowing exactly wnat 1s 1n the natural audit samples. We
can. however, estimate the relative bias, I.e., the adjustment required to
make measurements from the different laboratories comparable. Second,
random variations 1n repeated measurements of the same thing are an Indi-
cation of the precision of the measurements, again assuming the actual
composition 1s constant.
METHOD
Measurement bias between laboratories, measurement precision, and changes
1n the samples over time are all related. All three are visible 1n Figure
4-1, whlcn shows the measurements of Initial pH from the acidity tltratlon
of samples from B1g Moose Lake (type FN2) performed by EMSI and Versar.
On the whole, 1n Figure 4-1 the E's are above the V's: there 1s an upward
bias of the measurements by EMSI relative to those by Versar, or, equlva-
lently, a downward bias by Versar relative to EMSI. There 1s an upward
trend 1n the measurements over time: later ones are rather higher than
earlier ones on average. There 1s also some scatter, or Imprecision, 1n
measurements by the same laboratory at about the same time.
esissr 7
22
91
-------
o
10
EMSI
Versar
6
9
12
15
18
October
21
24
Batch
27
Date
30
2
5 8 11
November — -
14
FIGURE 4-1. Inter!aboratory bias and trend 1n pH of natural audits from
Big Moose Lake (EMSI and Versar only).
85185
23
92
-------
The upward trend 1n measurements over time 1s somewhat hidden by the
Interlaboratory bias. The latest measurements are by Versar and so
probably a little lower than 1f they had been by EMSI. The trend would
have looked stronger 1f the latest measurements had been higher, and they
probably would have been higher but for the bias. Indeed the trend 1n
each of the laboratories separately 1s stronger than when they are com-
bined.
Conversely, the Increase 1n measured values over time hides some of the
bias. Since the measurements by Versar are on average slightly later than
those by EMSI, they are probably slightly higher than 1f they had been
performed earlier. Thus, the bias shows up a little less 1f the averages
of all measurements by EMSI and by Versar are compared than 1f measure-
ments by the two laboratories around the same time are compared.
It seems reasonable to estimate bias and measurement trend simultaneously
by the technique known as analysis of covariance. Separate but parallel
least-squares lines are fit for the two laboratories. Bias 1s estimated
by the vertical distance between the lines: this 1s the average differ-
ence between contemporaneous measurements. The trend 1s estimated by the
common slope, which 1s the slope of the measurements corrected for bias.
Precision can be estimated two ways. If the estimates of bias and trend
are assumed to be correct, measurements can be corrected for both. The
precision of such corrected measurements 1s estimated by the standard
deviation of the audit measurements around their respective least-squares
lines. On the other hand, the estimated bias and trend are small compared
to the residual scatter. It might, therefore, be just as well not to
correct the measurements. In this case the precision 1s estimated simply
by the standard deviation of the audit measurements.
The picture 1s much less neat when the other two laboratories are added
(Figure 4-2). Versar's and EMSI's measurements are 1n October and early
November, but USGS's are 1n December. The analysis of covariance extrapo-
lates the linear trend from October and November Into December. If that
were true, the USGS measurements should be very high, unless they are
biased. Accordingly, a very large negative bias 1s estimated for USGS's
to account for their being about the same as Versar's on average.
There 1s no evidence 1n the data of a linear upward trend Into December.
On the other hand, there 1s no way to distinguish a change 1n the trend
from a bias at USGS because there are no other measurements contemporane-
ous with USGS's. The best we can do 1s to estimate bias Ignoring trend,
by analysis of variance, as well as controlling for trend by analysis of
covariance. The disparity of the estimates of bias for USGS by the two
•5185 7
24
93
-------
u
10
EMS I
USGS
1 6 11 16 21 26 31 5 10 15 20 25 30 5 10 15
October November -December-
Batch Date
FIGURE 4-2. Inter!aboratory bias and trend 1n pH of natural audits from
B1g Moose Lake (four laboratories).
B5185
25
94
-------
methods 1s a warning of the uncertainty of these estimates. Incidentally,
the analysis of variance provides a third estimate of precision—correct-
ing for bias but Ignoring trend.
RESULTS
Precision
Estimates of precision for parameters other than pH are shown 1n Table 4-1.
Since precision for these parameters 1s believed to be roughly propor-
tional to concentration, 1t 1s reported as a relative standard deviation
(RSD), which 1s the appropriate standard deviation divided by the mean.
(The standard deviation can be the ordinary standard deviation. Ignoring
bias and trend; within-laboratory standard deviation from the analysis of
variance, correcting for bias; or residual standard deviation from the
analysis of covaMance, correcting for bias and trend.) For Lake Superior
there are only seven samples, too few to estimate bias or trend; thus only
the estimated precision Ignoring bias and trend 1s reported. Table 4-1
also shows the means and the quantltatlon limit. If the concentration 1s
below or near this limit, a high RSD 1s expected since large relative
errors may occur at low concentrations. Iron, for example, 1s simply not
found 1n determlnable quantities 1n the natural audit samples. The esti-
mates of precision for Iron should therefore not be taken as representa-
tive of the precision with which Iron 1s measured at higher concentra-
tions.
On the whole, correcting for bias alone or for both bias and trend makes
little difference In the estimates of precision. The biases and trends
are small compared to the random variation even though they are sometimes
large enough and systematic enough to be statistically significant.
For pH, since precision 1s more or less constant over the relevant range,
there 1s no need for RSDs. (Anyway, absolute differences 1n pH already
represent relative differences 1n hydrogen 1on concentration.) Therefore
the estimates of precision for pH are 1n pH units rather than percents.
Also, there 1s no problem of quantltatlon limits for the pH measure-
ments. The estimates of precision for pH are 1n Table 4-2.
Bias
The estimates of bias, both correcting for trend and Ignoring trend, are
shown 1n Tables 4-3 and 4-4. Like precision, bias might be supposed to be
roughly proportional to concentration; accordingly, we show bias as a
percentage of the mean. Again, pH measurements are an exception and are
85185 7
26
95
-------
TABLE 4-1. Overall precision estimated from natural audits.
Parameter*
Calcium
Magnesium
Potassium
Sodium
Manganese
Iron
Aluminum
(extractable)
Chloride
Sulfate
Nitrate
Silica
Fluoride (total)
DOC
Ammonium
Acidity (peq/1)
Alkalinity (ueq/1)
Conductivity (pS/cm)
QIC (equilibrated)
DIC (Initial)
Phosphorus (total )
Aluminum (total)
Ignoring
Bias and
Trend
9.2
2.7
4.3
5.1
26
66
34
51
7.3
25
8.1
4.1
9.3
54
40.
260
3.6
60.
27
140
28
B1g Moose Lake
Rsn(t)
Correcting
for
Bias
6.5
1.9
4.4
4.7
23
65
32
50.
7.4
26
7.3
3.9
8.7
50.
36
246
2.8
53
17
140
22
(N - 41)
Correcting
for Bias
and Trend
6.4
1.9
4.4
4.4
23
65
32
50.
7.2
26
7.4
3.9
8.8
50.
34
249
2.8
53
15
128
21
Mean
1.9
0.35
0.49
0.67
0.070
0.02
0.18
0.61
6.9
1.6
4.3
0.077
3.3
0.059
51
2.4
27
0.19
0.42
0.002
0.31
Lake Superior
RSO(t)
Ignoring
Bias and
Trend
5.5
3.6
12
6.4
360
140
86
2.0
3.3
5.8
6.0
5.1
9.7
130
63
2.3
1.7
3.8
4.2
140
120
(N - 7)
Mean
13
2.8
0.49
1.3
0.003
0.003
0.002
1.4
3.3
1.4
2.7
0.035
1.4
0.007
30.
850
96
9.7
9.9
0.001
0.021
Quantltatlon
Limit
0.15
0.039
0.37
0.20
0.093
0.12
0.039
0.39
1.7
3.4
2.3
0.034
2.5
0.56
150
57
9.0
0.56
1.1
0.037
0.28
* All parameters are measured In mg/1 unless other-wide stated.
85185 5
-------
TABLE 4-2. Overall precision of pH measurements 1n natural audits.
Big Moose Lake (N * 41)Lake Superior
Precision (N * 7)
Correcting Precision
Ignoring Bias Correcting for Bias Ignoring Bias
Parameter Mean and Trend for Bias and Trend Mean and Trend
pH (equilibrated) 5.18 0.27 0.27 0.28 8.23 0.08
pH (alkalinity) 5.07 0.04 0.04 0.03 7.76 0.12
pH (acidity) 5.08 0.05 0.04 0.03 7.79 0.11
85185 5
28
97
-------
TABLE 4-3. Relative Interlaboratory bias (expressed as a percent of the mean) estimated from natural audits and controll-
ing for measurement trend (estimate t standard error of estimate).
Calcium
Magnesium
Potassium
Sodium
Manganese
Iron
Aluminum
(extractable)
Chloride
Sulfate
Nitrate
Silica
Fluoride (total)
DOC
Ammonium
Acidity
Alkalinity
Conductivity
DIC (equilibrated)
OIC (Initial)
Phosphorus (total)
Aluminum (total)
EMS I - Global
0.6 i 4.4
0.1 t 1.3
-0.2 t 3.0
0.7 t 3.0
-40.5 t 15.5
47.6 t 44.5
21.1 t 21.5
14.8 t 33.7
-0.5 t 4.9
-0.4 t 17. -4
-12.9 J 5.1
-4.3 t 2.7
-4.5 ± 6.0
-51.9 t 34.1
77.9 ± 22.9
-27.8 ± 169.7
3.4 J 1.9
-102.5 t 36.0
-79.8 i 10.4
107.6 ± 87.2
47.7 ± 14.6
EKSI - USGS
-32.0 ± 6.3
-5.7 t 1.8
-0.1 t 4.3
4.1 t 4.3
22.0 t 22.3
0.2 t 64.0
-31.3 * 31.0
-3.6 t 48.6
-13.7 t 7.1
-18.5 i 25.1
-0.7 t 7.3
-0.9 t 3.9
-15.6 t 8.6
-30.1 t 49.1
119.1 ± 33.0
385.8 t 244.4
-9.2 t 2.8
-15.1 i 51.9
6.0 t 15.0
161.0 t 125.7
90.0 t 21.1
EHSI - Versar
-1.1 i 2.2
-3.8 t 0.6
2.1 t 1.5
3.2 t 1.5
6.3 t 7.7
12.5 t 22.1
-16.9 t 10.7
-30.2 i 16.8
-1.5 t 2.4
5.1 i 8.7
3.2 ± 2.5
-2.9 t 1.3
2.5 ± 3.0
40.5 t 16.9
4.7 ± 11.4
146.4 t 84.3
0.5 i 1.0
2.0 ± 17.9
-14.0 + 5.2
-1.0 i 43.3
23.6 1 7.3
Global - USGS
-32.6 t 6.1
-5.8 t 1.8
0.1 t 4.2
3.4 t 4.2
62.6 t 21.6
-47.3 t 61.9
-52.4 ± 30.0
-18.4 + 47.0
-13.2 i 6.9
-18.1 i 24.3
12.2 J 7.0
3.4 + 3.7
-11.1 t 8.4
21.8 t 47.5
41.2 t 31.9
413.6 t 236.4
-12.6 t 2.7
87.4 i 50.2
85.7 + 14.5
53.4 t 121.5
42.3 1 20.4
Global - Versar
-1.8 l 4.4
-3.9 i 1.3
Z.3 ± 3.0
2.4 ± 3.0
46.8 i 15.4
-35.1 t 44.3
-38.0 t 21.4
-45.0 t 33.6
-1.0 t 4.9
5.4 i 17.4
16.1 i 5.0
1.4 t 2.7
7.0 t 6.0
92.4 ± 34.0
-73.2 t 22.9
174.2 ± 169.1
-2.9 t 1.9
104.5 t 35.9
65.8 + 10.4
-108.6 t 87.0
-24.1 t 14.6
USGS - Versar
30.8 ± 6.3
1.9 t 1.8
2.2 t 4.3
-0.9 t 4.3
-15.8 t 22.1
12.2 t 63.6
14.4 t 30.8
-26.6 t 48.2
12.2 J 7.0
23.6 t 24.9
3.9 + 7.2
-2.0 ± 3.8
18.1 i 8.6
70.7 t 48.7
-114.4 t 32.8
-239.4 t 242.6
9.7 ± 2.8
17.1 t 51.5
-20.0 l 14.9
-162.0 t 124.7
-66.4 t 20.9
-------
TABLE 4-4. Relative 1nterlaboratory bias (expressed as a percent of the mean) estimated fro* natural audits and Ignoring
measurement trend (estimate t standard error estimate).
Calcium
Magnesium
Potassium
Sodium
Manganese
Iron
Aluminum
(extractable)
Chloride
Sulfate
Nitrate
Silica
Fluoride (total)
DOC
Ammonium
Acidity
Alkalinity
Conductivity
DIC (equilibrated)
DIC (Initial)
Phosphorus (total)
Aluminum (total)
EMSI - Global
3.0 4 4.1
0.8 i 1.2
0.4 ± 2.1
-2.4 i 2.9
-50.0 4 14.4
66.3 t 40.8
34.4 4 20.0
-3.0 t 31.2
2.6 4 4.6
10.4 i 16.2
-11.9 * 4.6
-4.5 ± 2.4
-3.7 4 5.4
-42.1 ± 31.0
55.8 1 22.3
-50.7 t 153.5
4.2 4 1.8
-113.5 t 32.8
-68.2 t 10.4
6.9 1 87.3
37.8 t 13.7
EMSI - USGS
-25.5 4 4.1
-3.8 t 1.2
1.4 t 2.7
-4.7 t 2.9
-4.0 t 14.4
51.9 i 40.8
5.4 ± 20.0
-52.9 1 31.2
-5.4 + 4.6
11.3 t 16.2
2.0 4 4.6
-1.4 t 2.4
-13.4 ±5.4
-3.2 t 31.0
57.9 t 22.3
322.7 t 153.7
-7.1 ± 1.8
-45.4 t 32.8
38.0 t 10.4
-117.3 4 87.3
62.7 ± 13.7
EMSI - Versar
-1.0 + 2.2
-3.8 4 0.6
2.2 t 1.5
3.0 t 1.6
5.9 * 7.8
13.3 t 22.1
-16.3 t 10.9
-30.9 t 16.9
-1.4 f 2.5
5.5 t 8.8
3.3 4 7.5
-3.0 4 1.3
2.5 4 2.9
40.9 4 16.8
3.8 t 12.1
145.4 4 83.2
0.5 4 1.0
1.5 i 17.8
-13.5 4 5.6
-5.3 4 47.4
23.1 ± 7.4
Global - USGS
-28.5 J 5.3
-4.5 4 1.6
1.1 t 3.6
-2.2 4 3.9
46.0 4 18.9
-14.4 4 53.4
-29.0 4 2fi.2
-49.9 4 40.8
-7.9 + 6.0
0.9 4 21.3
13.9 4 6.0
3.0 4 3.2
-9.7 4 7.1
39.0 4 40.6
2.1 i 29.2
373.3 4 200.9
-11.2 4 2.3
68.1 i 42.9
106.2 ± 13.6
-124.2 4 114.3
24.9 ± 17.9
Global - Versar
-4.0 4 4.1
-4.5 4 1.2
1.8 4 2.7
5.5 4 3.0
55.8 4 14.5
-53.0 4 40.9
-50.7 4 20.1
-27.9 4 31.3
-3.9 4 4.6
-4.9 4 16.3
15.1 4 4.6
1.5 4 2.4
6.2 4 5.4
83.1 4 31.1
-52.0 t 22.4
196.1 4 154.1
-3.7 4 1.8
115.0 4 32.9
54.7 4 10.4
-12.2 4 87.7
-14.6 ± 13.8
USGS - Versar
24.5 4 4.1
-0.0 t 1.2
0.7 t 2.7
7.7 4 3.0
9.9 4 14.5
-38.6 + 40.9
-21.7 4 20.1
21.9 4 31.3
4.0 4 4.6
-5.8 4 16.3
1.2 4 4.6
-1.5 4 2.4
15.9 4 5.4
44.1 4 31.1
-54.1 t 22.4
-177.2 4 154.1
7.6 4 1.8
46.9 t 32.9
-51.5 4 10.4
112.1 4 87.7
-39.5 t 13.8
-------
Shown separately 1n Table 4-5. Each estimate of relative bias between two
laboratories 1s accompanied by Its standard error, a measure of the sta-
tistical uncertainty 1n the estimate. Again, we caution that tne uncer-
tainty for estimates Involving USGS may be much greater because bias and
trend cannot be distinguished 1n the data from this laboratory. The
magnitude of this sort of uncertainty 1s suggested by the difference
between corresponding estimates Ignoring and controlling for trend.
There are three different standards against which the biases should be
measured; tne biases are on the whole rather small on all three scales.
First, the biases are small compared to the precision of the Individual
measurements. This 1s why correcting for bias did not Improve precision
much. Correcting for bias Improves the precision of Initial dissolved
Inorganic carbon from 27 to 17 percent, the largest change among the
parameters 1n this study but still only about a third. For most
parameters the effect 1s much smaller. Thus, bias 1s small 1n the sense
that 1t does not contribute much to the Inaccuracy of Individual measure-
ments.
Second, bias 1s a small percentage of the average measured concentration
1n most cases. Most of the exceptions fall Into three categories: (a)
estimates Involving USGS, which are not very reliable; (b) estimates for
parameters for which precision 1s also relatively poor, e.g., chloride;
and (c) estimates for parameters that are not present 1n the samples 1n
amounts large enough to measure accurately, e.g., Iron.
Finally, 1n many but not all cases the estimated bias 1s small compared to
the standard error of the estimate. For example, the relative bias for
calcium measurements by EMSI and Versar 1s about 1 percent, but this esti-
mate 1s subject to an uncertainty of about 2 percent. In such cases the
evidence that there Is any bias at all 1s not statistically significant.
Trend
Table 4-6 shows the estimates of linear trends 1n the measurement of the
parameters over time. Like the biases, the trends are small compared to
(a) the precision of the measurements, (b) the measurements themselves,
and (c) usually, but not always, the uncertainty 1n estimating them.
Statistically significant trends (t-test, p < .05), all upward, are seen
for sodium, acidity, Initial dissolved Inorganic carbon, total phosphorus,
and Initial pH from both acidity and alkalinity tUratlons.
85185T 7
31
100
-------
TABLE 4-5a. Interlaboratory bias (expressed as pH units) estimated from natural audits: controlling for measurement
trend (estimate t standard error of estimate).
pH
pH
pH
(equilibrated)
(alkalinity)
(acidity)
TABLE 4-5b. Ignoring
pH
pH
pH
(equilibrated)
(alkalinity)
(acidity)
EMSI
0.1
-0.02
0.04
- Global EMSI
t o.
40.
± 0.
2 0.02
02 0.13
02 0.16
- USGS
i 0.
i 0.
t 0.
3
03
03
EMSI - Versar
0.1 i 0.1
0.02 ± 0.01
0.06 ± 0.01
Global
0.1
0.15
0.12
- USGS
t 0.3
t 0.03
± 0.03
Global - Versar
0.0 ± 0.2
0.04 t 0.02
0.03 t 0.02
USGS - Versar
-0.1 + 0.3
-0.11 t 0.03
-0.10 ± 0.03
measurement trend.
EMSI
0.1
-0.05
0.00
- Global EMSI •
1 0.2 0.2
t 0.02 0.03
* 0.02 0.05
• USGS
t o.
2
t 0.02
t 0.02
EMSI - Versar
0.1 t 0.1
0.02 i 0.01
0.06 + 0.01
Global
0.0
0,08
0,05
- USGS
f 0.2
i 0.03
1 0.03
Global - Versar
0.0 J 0.2
0.08 + 0.02
0.06 i 0.02
USGS - Versar
0.0 + 0.2
-0.01 i 0.02
0.01 + 0.02
-------
TABLE 4-6. Trends (per month) 1n measurements of
natural audit samples (Big Moose Lake).
85185 5
Parameter*
Calcium
Magnesium
Potassium
Sodium
Manganese
Iron
Aluminum (extractable)
Chloride
Sulfate
Nitrate
Silica
Fluoride (total)
DOC
Ammonium
pH (equilibrated)
pH (alkalinity)
pH (acidity)
Acidity (ueq/1)
Alkalinity (weq/1)
Conductivity (uS/cm)
DIC (equilibrated)
DIC (Initial)
Phosphorus (total )
Aluminum (total)
Estlnated
Trend
-0.08
-0.004
0.00
0.03
0.012
-0.007
-0.04
0.20
-0.4
-0.3
-0.1
0.000
0.0
-0.011
0.0
0.07
0.07
20.
1
-0.4
0.04
-0.09
0.003
0.05
Standard Error
of Estimate
0.06
0.003
0.01
0.01
0.008
0.007
0.03
0.15
0.3
0.2
0.2
0.002
0.1
0.015
0.1
0.01
0.02
9
3
0.4
0.05
0.03
0.001
0.03
* All parameters measued 1n mg/1 unless otherwise noted.
33
102
-------
SYNTHETIC AUDIT SAMPLES
INTRODUCTION
Four kinds of synthetic audit samples were used 1n the Eastern Lakes Sur-
vey. All synthetic audits were mixed and diluted from stock solutions
according to one of two recipes, one giving relatively high concentra-
tions, the other low. "Field synthetic audits," like the natural audits,
were sent dally 1n two-liter containers to the field stations, where they
were processed Into allquots by the usual procedure. "Laboratory syn-
thetic audits" arrived at the field station as allquots; the field sta-
tions, without opening the allquots, simply relabeled them and shipped
them with the routine samples to the contract laboratories.
In addition to their use 1n quality control, the synthetic audits have
three uses 1n quality assessment: estimation of precision and Inter-
laboratory bias; comparison of measured with theoretical concentrations;
and evaluation of performance of field stations by comparison of field and
laboratory audits. Each of these three subjects 1s discussed 1n turn.
BIAS AND PRECISION
In principle, the synthetic audits, like the natural audits, can be used
to estimate precision and Inter!aboratory bias. Differences 1n measure-
ments of the same thing 1n the same laboratory at different times Indicate
how precise those measurements are. Systematic differences 1n measure-
ments of the same thing between different laboratories Indicate a relative
bias between the laboratories. Unfortunately, there 1s ample evidence
that the synthetic audit data are not repeated measurements of the same
thing.
Figure 5-1 shows the concentrations of ammonium measured 1n laboratory
high synthetic audit samples by three contract laboratories. With one
exception (a measurement of about 1.9 mg/1 by Versar on 18 October) the
measurements on Individual days form very tight groups, but the groups are
spread far apart. That 1s, measurements on a given day are 1n close
agreement both within and among laboratories, but on different days they
85185r e
34
103
-------
E - EMSI
V - Versar
G - Global
U - USGS
11 15 19 23 27 31 4 8 12 16 20 24 28 2 6 10
October November December
Batch Date
FIGURE 5-1. Concentration of ammonium measured in laboratory high synthetic
audit samples.
65185 35
104
-------
are very different. Probably the concentrations were very different on
different days, and Imprecision 1n measurement and bias among laboratories
were very small by comparison.
We would like to be able to quantify this comparison. An appropriate
statistical technique 1s analysis of variance with fixed and random
effects. Basically, differences between days are attributed to errors in
preparation of the audit samples, and differences within days are attribu-
ted to errors 1n measurement. But we need to consider many variables, or
effects, as they are called in statistics. There may be bias between
laboratories. Measurements within a laboratory on the same day may be
less variable than those on different days. There may be variation 1n
preparation of the several samples on the same day, though less than
across days. There may be systematic as well as random errors in prepara-
tion across time. With enough data all these effects could be estimated,
but in this study there are not enough data. We did fit some random-
effects models, but we consider the results wholly unreliable. The esti-
mates are extremely sensitive both to single data points (like the
exceptional 18 October ammonium concentration mentioned earlier) and to
the statistical method of estimation chosen from several equally reason-
able alternatives. With such large errors 1n preparation there 1s simply
not enough information 1n the synthetic audits to estimate errors in mea-
surement.
MEASURED VERSUS THEORETICAL CONCENTRATIONS
Comparing the concentration measured 1n the synthetic audit samples with
the theoretical concentrations furnishes very useful confirmation that the
system 1s measuring what 1t was meant to measure. Tables 5-1 and 5-2 show
the theoretical values and the means of the measurements of field and
laboratory, high and low synthetic audit samples (calcium and chloride are
divided by lot because the theoretical concentrations changed between
lots). For most parameters there 1s reasonable 1f not very close agree-
ment between the measured and theoretical values.
The most notable exception 1s Iron, which 1s essentially absent from the
field audit samples but present 1n the laboratory audits. (Although there
1s no theoretical concentration, a similar difference 1s seen between
field and laboratory audits for extractable aluminum; a smaller difference
1s seen for manganese). Apparently Iron 1s removed by the processing 1n
the field. Also, Initial dissolved Inorganic carbon is consistently above
Us theoretical value (Figure 5-2). However, some confusion 1s added by
an apparent change in preparation around 23 October. This change 1s also
85185 8
36
105
-------
TABLE 5-1. Field and laboratory high synthetic audits.
Parameter*
Calcium
Lot 4 (n - 20. 14)
Lots 5, 6 (n - 16. 7)
Magnesium
Potassium
Sodium
Manganese
Iron
Aluminum (extractable)
Chloride
Lot 4 (n - 20, 14)
Lots 5, 6 (n - 16, 7)
Sulfate
Nitrate
Silica
Fluoride (total)
DOC
Ammonium
pH (equilibrated)
pH (alkalinity)
pH (acidity)
Acidity (ueq/1)
Alkalinity (ueq/1)
Conductivity (uS/cm)
DIC (equilibrated)
DIC (Initial)
Phosphorus
Aluminum (total)
Field
Mean
1.66
1.95
2.37
3.03
11.84
1.19
0.001
0.037
3.66
4.00
14.39
1.816
9.23
0.445
10.01
1.256
7.72
7.08
7.12
73
477
104.5
4.72
5.94
0.057
0.199
(n - 36)
Standard
Deviation
0.11
0.16
0.09
0.19
1.03
0.16
0.099
0.027
1.45
0.73
0.80
0.666
1.26
0.042
1.96
0.162
0.54
0.25
0.27
56
82
4.2
1.28
1.98
0.013
0.059
Laboratory (n « 21)
Mean
1.55
1.96
2.36
3.02
11.87
1.39
0.175
0.161
3.39
5.81
14.47
1.431
9.42
0.435
10.25
1.342
7.87
7.05
7.11
85
486
104.3
4.79
5.94
0.061
0.195
Standard
Deviation
0.18
0.10
0.08
0.15
1.04
0.04
0.019
0.030
0.24
5.34
0.69
0.672
1.39
0.012
1.68
0.176
0.15
0.26
0.25
68
95
4.1
1.31
1.90
0.010
0.067
Theoretical
1.54
2.39
2.43
2.97
12.41
1.50
0.15
__
2.72
4.22
14.09
1.707
10.70
0.452
10.0
1.25
—
--
--
--
—
—
—
3.10
0.075
0.19
Measured 1n mg/1 unless other noted.
65185 5
37
106
-------
TABLE 5-2. Field and laboratory low synthetic audits.
Calcium
Lot 4 (n - 20, 20)
Lots 5, 6 (n « 23, 23)
Magnesium
Potassium
Sodium
Manganese
Iron
Aluminum (extractable)
Chloride
Lot 4 (n « 20, 20)
Lots 5, 6 (n - 23, 23)
Sulfate
Nitrate
S111C8
Fluoride (total)
DOC
Ammonium
pH (equilibrated)
pH (alkalinity)
pH (acidity)
Acidity (peq/1)
Alkalinity (yeq/1)
Conductivity (uS/cm)
DIC (equilibrated)
DIC (Initial)
Phosphorus (total )
Aluminum (total)
Field (n - 36)
Standard
Mean Deviation
0.144 0.027
0.169 0.012
0.424 0.031
0.231 0.043
2.71 0.14
0.091 0.009
0.001 0.012
0.005 0.004
0.329 0.234
0.365 0.081
2.27 0.23
0.547 0.362
1.00 0.31
0.042 0.009
0.982 0.476
0.158 0.043
7.34 0.14
6.87 0.11
6.96 0.19
16.3 23.3
112 8.3
19.0 1.3
1.35 0.12
1.65 0.21
0.021 0.006
0.023 0.005
Laboratory (n « 21)
Mean
0.137
0.161
0.426
0.225
2.69
0.092
0.070
0.018
0.300
0.347
2.28
0.471
1.02
0.040
1.190
0.186
7.29
6.86
6.93
16.1
110
18.7
1.35
1.65
0.022
0.031
Standard
Deviation
0.015
0.014
0.019
0.023
0.25
0.014
0.016
0.006
0.042
0.109
0.10
0.172
0.20
0.002
0.724
0.041
0.13
0.12
0.15
27.3
10.3
1.2
0.15
0.24
0.009
0.028
Theoretical
0.13
0.19
0.45
0.20
2.79
0.10
0.06
--
0.22
0.34
2.28
0.466
1.07
0.042
1.0
0.17
—
._
—
—
—
--
--
0.96
0.027
0.02
* Measured 1n mg/1 unless otherwise noted.
85185 5
38
107
-------
•
T
^
Et
0
o»
H4
c
•"*
a
i
K
I
.
t V V
11 15 19 2
October
I . E - EKSI
' V « Versar
G - Global
U - USGS
fl
V
V
V
V
V
theoretical value
3 27 31 4 8 12 16 20 24 28 2 6 10
Batch Date
FIGURE 5-2. Concentration of dissolved Inorganic carbon (Initial)
measured 1n laboratory high audit samples. Measurements before and
after 23 October (vertical line) seem to be different.
851B5
39
108
-------
noticed 1n several other parameters, especially sodium, silica (Figure
5-3), acidity, alkalinity, and pH, and may have been caused by a change 1n
the concentration of the stock solution of sodium silicate.
We do not think any detailed statistical analysis of the differences
between measured and theoretical concentrations 1s called for. The like-
liest reason for any difference 1s not a problem 1n processing or analysis
but rather that the theoretical concentrations are not what was actually
1n the samples.
FIELD VERSUS LABORATORY AUDITS
It was hoped that by comparing the variability of the field audits, which
were processed by field stations, with that of the laboratory audits,
which were not, one could distinguish the amount of variability Introduced
at the field stations from the Imprecision of the analytical labora-
tories. As 1t happens, both of these sources of variability are swamped
by variability 1n the preparation of synthetic audit samples. The
standard deviations of measurements of the four types of synthetic audit
samples (field and laboratory, high and low) are In Tables 5-1 and 5-2.
On the whole the field audits are not any more variable than the labora-
tory audits. That 1s, preparation of the allquots 1n the field Introduces
no more variability than preparation 1n the laboratory. Again, finer
analysis 1s made Impossible by the magnitude of the errors 1n preparation
of synthetic audit samples.
8S185 8
40
109
-------
r— *• *
E - EMSI
V «= Versar
G - Global
U - USGS
11 15 19 23 27
October
31
8 12 16 20 24
November
Batch Date
28
2 6 10
—December
FIGURE 5-3. Concentration of silica measured 1n laboratory high
synthetic audit samples. Measurements before and after 23 October
(vertical line) seem to be different.
85185
41
no
-------
6 CONCLUSIONS AND RECOMMENDATIONS
WHAT THE QA DATA SHOW
One of the purposes of the QA data was to allow us to estimate the
Imprecision of routine lake sample measurements. In this report we have
provided many estimates of various kinds of precision. In this section we
relate all these estimates. Table 6-1 shows our estimates of w1th1n-batch
laboratory precision, w1th1n-batch precision (Including errors made both
at the field stations and at the analytical laboratories), and overall
precision. The estimates of precision are for measurements above the
quantltatlon limit for each parameter, as we know that there are large
relative errors below the quantltatlon limits.
The findings of our analysis may be Illustrated by considering together
the Information given by all the quality assurance data about one para-
meter. We use calcium as the example. The estimate of measurement
precision from laboratory duplicate pairs for calcium 1s 0.89 percent.
Thus, the reprodudbUKy of determinations of calcium within any single
laboratory and single batch 1s excellent. When precision 1s estimated
from field duplicate pairs, the estimate rises to 2.3 percent. So, when
the entire system of collection, field processing, and laboratory analysis
1s considered (and when the contract laboratory 1s blind to the dupli-
cates) reprodudblHty 1s much less but still very good. This reprodud-
bH1ty 1s still for measurements within the same laboratory and within the
same batch. These are likely to be done by a single analyst, possibly off
a single calibration, certainly on a single day, and so under broadly
similar laboratory conditions.
Information about variation from day to day and from laboratory to labora-
tory comes from the natural audits. Within the same laboratory the esti-
mated precision for the determination of calcium 1n B1g Moose Lake samples
1s 6.5 percent. The measurements by the USGS laboratory are some 25 per-
cent higher than the others, either because of bias or because of a change
1n the samples over time; consequently 1f all B1g Moose Lake measurements
are lumped together without regard to laboratory or date, the estimate of
precision rises to 9.2 percent.
Bsiesr 9
42
-------
TABLE 6-1. Summary of estimates of precision (percent RSD, except for pH,
which 1s expressed as SD 1n pH units).
Natural Audits
B1g Moose Lake
Parameter*
Calcium
Magnesium
Potassium
Sodium
Manganese
Iron
Aluminum
(extractable)
Cnlorlde
Sulfate
Nitrate
S111ca
Florlde (total)
DOC
Ammo n1 urn
Acidity
Alkalinity
Conductivity
DIC (equilibrated)
DIC (Initial)
Phosphorus (total)
Aluminum (total)
pH (equilibrated)
pH (alkalinity)
pH (acidity)
Correcting
for Bias
and Trend
6.4
1.9
4.4
4.4
23
65
32
50.
7.2
26
7.4
3.9
8.8
50.
34
249
2.8
53
15
128
21
0.28
0.03
0.03
Ignoring
Bias and
Trend
9.2
2.7
4.3
5.1
26
66
34
51
7.3
25 S.S
8.1
4.1
9.3
54
40.
260
3.6
60.
27
140
28
0.27
0.04
0.05
Lake
Superior
Ignoring
Bias and
Trend
5.5
3.6
12
6.4
360
140
86
2.0
3.3
5.8
6.0
5.1
9.7
130
63
2.3
1.7
3.8
4.2
140
120
0.08
0.12
0.11
Field
Duplicates
2.3
2.3
3.7
4.3
11
10.
11
23
6.5
65
2.7
9.0
10.
._
74
10.
1.9
5.0
3.7
9.7
24
0.085
0.052
0.075
Laboratory
Duplicates
0.89
0.63
1.2
0.97
1.6
3.0
3.3
2.1
1.3
__
2.2
2.5
2.3
1.4
1.3
2.1
0.72
2.4
2.4
2.4
5.4
0.077
0.042
0.065
85185 5
43
112
-------
The rise from 0.89 to 2.3 to 6.5 to 9.2 percent as we move from the
laboratory duplicates through the field duplicates to the natural audits
1s not surprising, and 1t demonstrates the Importance of the quality
assurance program. As 1t happened, the Imprecision of a perfectly cali-
brated Instrument under constant conditions made only a small contribution
to overall uncertainty, compared to errors 1n calibration and variation
over time 1n conditions or procedures 1n the laboratory and 1n the
field. These other effects can only be judged by audits treated as much
as possible like routine samples under field conditions. The blind field
natural audits serve this purpose well. The estimates of precision from
field natural audits are therefore probably the best Indicators of uncer-
tainty in the routine samples.
Along with calcium, the Big Moose Lake natural audits probably give
reasonable estimates of overall precision 1n routine samples for mag-
nesium, potassium, sodium, aluminum (extractable and total), chloride,
sulfate, silica, fluoride, dissolved organic carbon, and all three mea-
surements of pH by contract laboratories. The levels of other parameters
1n 81g Noose Lake samples are below the quantltatlon limit, and the rela-
tive standard deviations at such low concentrations should not be con-
sidered representative. These parameters fall Into three classes.
Dissolved Inorganic carbon (equilibrated and initial). There Is
quantifiable DIG 1n natural audit samples from Lake Superior. The
RSD for these samples can be used as an estimate of overall precision
in place of that for Big Noose Lake.
Alkalinity. The alkalinity of samples from Lake Superior 1s well
above the quantltatlon limit; Indeed, at 850 ueq/1 it 1s probably
much higher than that of most lakes in the survey. The precision of
Lake Superior measurements is therefore not representative of routine
measurements. We therefore have no good estimate of overall pre-
cision for alkalinity. The closest we can come 1s the within-batcn
estimate of 10 percent from field duplicates, recognizing that w1tn-
1n-batch calculations usually underestimate overall imprecision. On
the other hand, the low synthetic audits, 1n spite of preparation
error, also have RSDs around 10 percent, so that overall Imprecision
cannot be much higher. The figure of 10 percent 1s therefore a rea-
sonable estimate of precision of measurements of alkalinity 1n rou-
tine samples.
Nitrate, ammonium, acidity, phosphorus, Iron, and manganese. The
levels of these parameters in the natural audits are below the
quantltatlon limits. So are the levels 1n almost all the lakes for
which there are field or laboratory duplicates, and so presumably in
almost all the routine samples. Thus these things were not measured
esissr 9
44
113
-------
very precisely simply because there was not enough of them to
measure. One could say that the lakes contained no measurable levels
of these parameters, or that the system was Insufficiently sensitive
to measure what was 1n the lakes; the matter will be judged according
to the quantltatlon limits.
SYNTHETIC AUDITS
The synthetic audit samples turned out not to be useful for estimating
errors 1n measurement because these were much smaller than errors 1n
preparation. In retrospect this 1s not surprising: 1t 1s hard to prepare
something more accurately than one can measure 1t.
The synthetic audits, however, are the only samples that furnish one
Important kind of data, but not enough of them--measurements of split
samples by different laboratories. Such measurements are very useful 1n
estimating Inter!aboratory bias. For most parameters the stability of the
natural audits 1s sufficient for this purpose, but 1t would be preferable
to be able to compare measurements on the same day. Also, the natural
audits are field audits and so Include effects of processing by field
stations, while laboratory synthetic audits could give Information about
bias between laboratories without Involving field station effects.
We recommend that 1n the future synthetic audits always be treated as
split samples. Dally lots should be prepared and divided Into allquots,
and the allquots should be assigned to different contract laboratories at
random. The QA data base should Identify the dally lot for each sample so
that comparisons between laboratories can be made using data from the same
lot. Variation among lots 1s then of little Importance. Indeed 1t might
even be desirable to vary the composition of samples between lots, both to
prevent recognition of blind audit samples and to see precision and
accuracy at various concentrations.
DETECTION LIMITS
To calculate detection limits from measurements of blank samples, we had
to make certain unverlflable assumptions about the response of the measur-
ing system at low concentrations. In particular, we assumed that the
calibration of the system was linear (with slope 1) although the Intercept
need not be zero. Thus, some background might b« added to samples 1n
processing, but the amount of background should be Independent of the
concentration.
85185T 9
45
114
-------
He believe that 1f detection limits are Important, they should be deter-
mined empirically. That 1s, audit samples with a concentration near the
putative detection limit for an analyte ought to be measured, and 1t
should be seen whether the analyte 1s reliably detected or not. Decision
limits, on the other hand, can be straightforwardly defined and accurately
estimated from blank data alone.
I5185r 9
46
115
-------
REFERENCES
Drouse, S. K., P. C. Hlllman, L. W. Creelman, J. F. Potter, and S. J.
Simon. 1985. [Draft] "Quality Assurance Plan for the National
Surface Water Survey Project." Lockheed-EMSCO. Las Vegas, NV.
EPA. 1983. "Calculation of Data Quality Indicators." U.S. Environmental
Protection Agency, Office of Research and Development.
EPA. 1984. National Surface Water Survey; National Lake Survey-
Phase I. U.S. Environmental Protection Agency, Office of Research
and Development, Washington, D.C. (NAPAP Project Reference Number
EI-23).
Hubaux, A., and G. Vos. 1970. Decision and detection limits for linear
calibration curves. Analytical Chemistry. 42:849.
Liggett, W. 1985. [Draft] "The Instability of Nitrate Concentrations 1n
Lake Samples." National Bureau of Standards, Galthersburg, MD.
Long, 6. L., and J. D. Wlnefordner. 1983. Limit of detection: a closer
look at the IUPAC definition. Analytical Chemistry. 55:712A.
esiesr
47
116
-------
APPENDIX B
INSTRUMENTAL DETECTION LIMITS, SYSTEM DETECTION LIMITS.
AND SYSTEM DECISION LIMITS BY LABORATORY,
EASTERN LAKE SURVEY — PHASE I
Values were reported by the contract analytical laboratories or were estimated from field blank data
and calibration blank data.
117
-------
Parameter
Al , extractable, mg L"1
Al , total, mg L"1
Ca, mg L-1
Cl~, mg L"1
Conductance, uS cm~l
DIG, air-equilibrated, mg L~*
DOC, mg L-1
F", total dissolved, mg L"1
Fe, mg L"1
K, mg L-l
Mg, mg L~l
Mn, mg L-1
Na, mg L~l
NH4+, mg L'1
N03", mg L"1
P, total , mg L~l
Si02, mg L"1
S042", mg L"1
TABLE B-
Reported
EMSI
0.002
0.002
0.01
0.00
-0.4C
0.02
0.1
0.003
0.03
0.00
0.00
0.01
0.00
0.01
0.006
0.001
0.03
0.01
1.
Instrumental
Global
0.002
0.002
0.00
0.00
-0.2C
0.04
0.1
0.001
0.02
0.00
0.00
0.01
0.00
0.00
0.003
0.001
0.02
0.01
Detection
USGS
0.003
0.002
0.00
0.00
0.5C
0.03
0.1
0.002
0.01
0.01
0.00
0.00
0.00
0.01
0.006
0.001
0.02
0.03
Limits
Versar
0.003
0.004
0.01
0.01
0.5C
0.003
0.1
0.002
0.01
0.01
0.00
0.00
0.00
0.01
0.006
0.001
0.02
0.03
aThree times the standard deviation of ten^nonconsecutive laboratory
calibration blank measurements (see Drouse et al., 1986).
DA11 batches (aliquots 3 and 5).
cFor conductance, the mean of six nonconsecutive blank measurements
was required to be less than 0.9 uS cm~l.
118
-------
611
*(9 PUB e sq.onb.LLB) sau.Dq.Bq LLVq
•sq.uauiajnsBaui
>)UBLq uo.Lq.Bjqi.LBD A~joq.BJoqBL i° 'aj^puaojad u.q.09 = OS
-------
Parameter
Al, extractable, mg L-1
Al, total, mg L'l
ANC, ueq L'1
Ca, mg L"1
C1-, mg L'1
Conductance, uS cm~l
DIC, air-equilibrated, mg L~l
DIC, initial, mg L"1
DOC, mg L'1
F~, total dissolved, mg L~l
Fe, mg L~l
K, mg L'1
Mg, mg L~l
Mn, mg L~l
Na, mg L-1
NH4+, mg L'1
N03~, mg L~lb
N03~, mg L~lc
NO-f, mg L~ld
P, total, mg L'1
Si02, mg L'1
S042-, mg L'1
TABLE B-3.
EMS I
0.011
0.59
12
0.08
0.24
1.8
0.20
0.34
0.6
0.072
0.02
0.04
0.00
0.02
0.20
0.46
0.096
0.049
0.118
0.018
0.14
0.14
System Detection
Global
0.00
0.088
6.0
0.10
0.04
1.0
0.34
0.84
1.0
0.060
0.00
0.06
0.02
0.14
0.02
0.10
0.082
0.091
0.066
0.022
0.34
0.06
Limit*
USGS
0.004
0.708
3.4
0.02
0.06
1.0
0.12
0.08
0.6
0.002
0.00
0.02
0.02
0.00
0.04
0.05
0.048
0.048
—
0.002
0.02
0.10
Versar
0.006
0.036
6.2
0.06
1.4
0.8
0.16
0.28
0.4
0.000
0.00
0.02
0.00
0.00
0.20
0.08
0.270
0.045
0.341
0.010
0.40
0.10
aSystem detection limit = 2(Pgg - PCQ). where Pgg = 95th percentile,
and PSQ = 50th percentile, of field blank measurements.
bAll batches (aliquots 3 and 5).
cAliquot 3 after filtration protocol change.
^Aliquot 5.
120
-------
Parameter
AT, extractable, mg L"1
AT, total, mg L"1
ANC, ueq L'1
Ca, mg L"1
C1-, mg L'1
Conductance, uS cm~l
DIG, air equilibrated, mg L"1
DIG, initial, mg L"1
DOC, mg L~l
F~, total dissolved, mg L~l
Fe, mg L~l
K, mg L'1
Mg, mg L~l
Mn, mg L"1
Na, mg L'1
NH4+, mg L"1
N03", mg L~lb
N03", mg L"lc
N03~, mg L"ld
P, total, mg L"1
Si02, mg L"1
S042-, mg L"1
aSystem decision limit = Pg5,
measurements.
TABLE
EMSI
0.008
0.030
10.7
0.04
0.13
0.6
0.28
0.46
0.5
0.038
0.01
0.02
0.00
0.01
0.10
0.24
0.056
0.019
0.068
0.010
0.08
0.10
where Pgg
B-4.
System Dec
Global
0.004
0.045
3.9
0.05
0.02
0.6
0.33
0.69
0.6
0.020
0.00
0.03
0.01
0.07
0.01
0.05
0.041
0.046
0.033
0.014
0.17
0.04
= 95th percenti
ision Limit3
uses
-0.002
0.365
-8.9
0.01
0.04
1.4
0.24
0.27
0.5
0.001
0.00
0.01
0.01
0.00
0.00
0.02
0.024
0.024
—
0.001
0.01
0.10
le of field
Versar
0.004
0.023
3.1
0.03
0.10
1.4
0.26
0.30
0.4
0.000
0.00
0.01
0.00
0.00
0.10
0.04
0.131
0.029
0.167
0.005
0.15
0.08
blank
bAll batches (aliquots 3 and 5).
cAliquot 3 after filtration protocol change.
^Aliquot 5.
727
-------
APPENDIX C
OVERALL WITHIN-BATCH PRECISION
BY LABORA TORY FOR 23 PARAMETERS,
EASTERN LAKE SURVEY — PHASE I
Values were estimated from field duplicate and field blank data
123
-------
TABLE C-l.
Overall Wi thin-Batch
Pairs with
Parameter
Al , extractable, mg L"1
all values
x <= 0.010
x > 0.010
Al , total , mg L-1
all values
7 <= 0.010
x > 0.010
ANC, ueq L"l
Ca, tng L"l
CT, mg L'1
Conductance, pS cm"1
DIC, mg L-1
air-equilibrated
initial
DOC, mg L'1
all values
x <= 5
7 > 5
Fe, mg L"1
F~, total dissolved, mg L~l
K, mg L-1
Mg, mg L"j
Mn, mg L"1
Na, mg L'1
NH^, mg L"|
N03~, mg L'1
all batches
aliquot 3
aliquot 5
PH
air-equilibrated
initial ANC
initial BNC
P, total, mg L'l
all values
7 <= 0.010
7 > 0.010
Si02, mg L"1
SO/', mg L'1
n
47
26
21
51
0
51
50
51
51
51
51
51
51
29
22
49
51
51
51
46
51
51
50
12
38
51
51
50
48
33
15
51
51
Mean > 0
RMS of SRSDa
47
61
16
39
—
39
19
2.6
21
1.8
3.9
7.6
14
8.0
20
141
9.4
6.5
3.3
69
1.9
18
308
6.1
354
0.07
0.04
0.06
40
45
27
20
2.9
EMSI
Precision
Pairs with
Mean > lOse
n
1
0
1
3
0
3
47
51
37
51
44
44
40
18
22
4
34
51
51
4
49
0
7
4
5
—
—
—
0
0
0
42
51
RMS of ZRSDa
24
—
24
36
—
36
10
2.6
24
1.8
2.2
5.3
15
6.5
20
14
10
6.5
3.3
13
1.9
—
0.85
1.7
0.83
--
—
—
—
—
—
20
2.9
Quantisation
Limit
n
99
99
99
99
99
99
99
99
99
99
99
99
99
99
99
97
98
21
77
--
--
—
99
99
99
10sR
0.046
0.187
29.1
0.13
0.45
5.9
0.52
0.91
2.7
0.18
0.029
0.10
0.03
0.06
0.28
0.80
0.182
0.009
0.184
— —
—
—
0.005
0.04
0.31
aRoot mean square of percent relative standard
pH).
deviation (RMS of absolute SD for
124
-------
TABLE C-2.
GLOBAL
Overall Within-Batch
Pairs with
Mean > 0
Parameter
Al , extractable, mg L"1
all values
7 <= 0.010
x > 0.010
Al, total, mg L'1
all values
7 <= 0.010
x > 0.010
ANC, ueq L'l
Ca, mg L"1
CT, mg L'1
Conductance, uS cm"1
DIC, mg L"1
air-equilibrated
initial
DOC, mg L"1
all values
x <= 5
7 > 5
Fe, mg L"1
F~, total dissolved, mg L'1
K, mg L"1
Mg, mg L"1
Mn, mg L"1
Na, mg L"1
NH4+, mg L'1
N03~, mg L"1
all batches
aliquot 3
aliquot 5
PH
air-equilibrated
initial ANC
initial BNC
P, total , mg L"1
all values
x <= 0.010
7 > 0.010
Si02, mg L"1
S042', mg L'1
n
17
15
2
17
3
14
16
17
17
17
16
17
17
9
8
16
17
17
17
11
17
17
15
10
5
17
17
17
17
10
7
17
17
RMS of %RSDa
36
38
9.8
21
30
19
3.1
4.1
4.4
4.1
11
6.7
5.9
7.8
2.6
21
3.4
2.0
2.0
46
3.0
21
38
27
53
0.03
0.06
0.07
44
38
51
3.7
1.3
Precision
Pairs with
Mean > IOSR
n
2
0
2
5
0
5
14
17
17
17
14
12
12
4
8
4
17
17
17
0
17
0
7
3
4
—
—
__
0
0
0
16
17
RMS of 2RSDa
9.8
—
9.8
23
—
23
1.5
4.1
4.4
4.1
11
3.7
5.4
8.6
2.6
16
3.4
2.0
2.0
—
3.0
2.6
2.0
3.0
—
—
__
—
—
__
3.8
1.3
Quantisation
Limit
n
31
30
31
31
31
31
31
31
31
31
31
31
31
31
31
31
31
23
8
—
—
__
31
31
31
10SR
0.016
0.128
47.9
0.20
0.08
2.6
0.97
1.64
1.9
0.13
0.007
0.09
0.05
0.23
0.06
0.15
0.114
0.114
0.119
—
—
0.044
0.54
0.12
aRoot mean square of percent relative standard deviation (RMS of absolute SD
for pH).
725
-------
TABLE C-3.
Overall Wi thin-Batch
Pairs with
Parameter
Al , extractable, mg L"l
all values
x <= 0.010
x > 0.010
Al, total, mg L"1
all values
x <= 0.010
x > 0.010
ANC, ueq L'l
Ca, mg L"1
CT, mg L~!
Conductance, uS cm"1
DIC, mg L"1
air-equilibrated
initial
DOC, mg L"1
all values
"x <= 5
x > 5
Fe, mg L"l
F~, total dissolved, mg L"1
K, mg L"l
Mg, mg L~J
Mn, mg L"1
Na, mg L"1
NH4+, mg L'1
N03", mg L L
all batches
aliquot 3
aliquot 5
pH
air-equilibrated
initial ANC
initial BNC
P, total , mg L"1
all values
7 <= 0.010
x > 0.010
Si02, mg L"1
S042", mg L'1
n
6
1
5
10
0
10
8
10
10
10
10
10
10
5
5
9
10
10
10
7
10
10
9
9
0
10
10
9
10
6
4
10
10
Mean > 0
RMS of ZRSDa
6.6
0.00
7.2
34
—
34
28
0.47
2.3
0.65
10
11
3.7
3.4
4.0
67
12
1.4
0.49
58
1.0
23
68
68
—
0.06
0.10
0.18
18
22
11
3.7
5.3
USGS
Precision
Pairs with
Mean > lOsg
n
0
0
0
0
0
0
6
10
10
10
8
9
9
4
5
6
10
10
10
2
10
2
3
3
0
—
—
__
9
5
4
8
10
RMS of ZRSDa
—
—
—
—
—
—
1.9
0.47
2.3
0.65
6.9
9.6
3.9
3.8
4.0
7.1
12
1.4
0.49
18
1.0
3.2
6.6
6.6
—
--
—
--
19
24
11
2.3
5.3
Quantitation
Limit
n
18
18
18
18
18
18
18
18
18
18
18
18
18
18
18
18
18
18
0
--
—
--
18
18
18
= = = = =
10sp
0.024
0.850
17.3
0.03
0.10
1.9
0.38
0.34
1.1
0.00
0.005
0.03
0.03
0.01
0.02
0.05
0.060
0.060
—
--
—
—
0.004
0.14
0.38
aRoot mean square of percent relative standard deviation (RMS of absolute SD
for pH).
126
-------
ill
•(Hd
OS svuosqe ^o SWH) uo.ueuap pjepue^s 3A.nei.aJ auao
fr9 *Z
Z9'£
220'0
—
—
—
S9£'T
T£2'0
90fr *T
T£'0
2T'0
OO'O
KTO
89'0
TWO
20'0
L'Z
99'0
£fr'0
fr'8
LZ'Q
9T'0
6'Efr
860'0
020'0
JSOT
q.LUi|
L6
L6
L6
—
—
—
09
L£
L6
L6
L6
L6
L6
L6
L6
L6
L6
L6
L6
L6
1.6
L6
L6
L6
L6
u
n
uo.UBaiq.UBno
TT 82 6'6 Lt
6'T
Ofr
—
Ofr
—
—
—
—
fr6
--
9'L
K9
TT
SZ'O
6'£
2'£
£T
E'fr
8'fr
9'fr
L'V
9'9
£t»'0
0£
T£'0
£T
L2
—
Li
L£
—
L£
Easbs jo swa
HSOT < UB8W
Ljq.LM SJLBd
UOj.Sj.39Jd I
HVSH3A
2T
9
0
S
—
—
—
0
2
0
T
Sfr
8T
Z.fr
6T
frT
2£
fr2
T2
9fr
9t»
fr£
Zfr
82
Zfr
2£
OT
—
OT
OT
0
OT
u
me
TZ 2fr
92 6£
69 8
t?£ ^f
SO'O Zfr
SO'O Zfr
TT'O £t>
08Z S2
ZZ£ frT
L99 6£
£9 S£
f 9 IM
TT 8T
9Z'0 Zfr
£'£ Lt
9'9 Zfr i-l
9fr frfr
£*fr fr2
9'fr £2
9't- LV
8'fr Zfr
L\ Uv
£fr'0 it
t2 LV
U'O Zt>
T2 Zfr
62 Tfr
t?9 9
S£ Lv
62 9T
29 Z2
29 £fr
Basaz jo swa u
0 < UB3W
U,q.LM S-UEd
g-u.Lnq.LM LLBU9AQ
ijad j.o guBnbs uB9iu q.
ooyE
T.l 6ui './OS
T-l ^^ '
OTO'O <
OTO'O =>
sen LEA L
j_l 6ui • |.Bq.o
ON8 LB.U.L
ONV LBH.I
p9q.Bjqi |.inb9 — i
S q.onb.L
E qonbi.
saL)oq.eq [
T.l Bui '
T 1 6ui •
T.-1 6"J
•[-1 Bui
T-l Bui
2OIS
x
x
L"B
q 'd
UL
U.L
IB
' Hd
LB
LB
_£ON
.'HN
'BN
'uw
*BW
I-l 6111 «x
Bui 'pgALOssip [Bq,oq.
T-l Bui
9 <
S =>
S9n[BA L
T-l Bui
p9q.BjqiLLnb9-j
T-l Bui
T-Uio sri 'aouBion
T-l Bui
T-l Bui
T-l barl
OTO'O <
OTO'O =>
S9H[BA L
T_l 6ui ' [.eqoq.
OTO'O <
OTO'O =>
S9n[EA L
Y_l 6ui '9[qBq.DBJqx9
ueiauiwBd
•_j
'9J
X
X
IB
Ul
LB
'310
puoo
'-13
'BO
'ONV
x
x
LB
'LV
x
X
LB
•LV
•fr-0 318V1
-------
APPENDIX D
ANALYTICAL W/TH/N-BATCH PRECISION
BY LABORATORY FOR 23 PARAMETERS,
EASTERN LAKE SURVEY — PHASE I
Values were estimated from contract analytical laboratory duplicate and calibration blank data
129
-------
TABLE D-l.
EMS I
Analytical Within-Batch Precision
Pairs with
Parameter
Al, extractable, mg L~l
all values
x <= 0.010
x > 0.010
Al, total, mg L'l
all values
x <= 0.010
7 > 0.010
ANC, ueq L'l
Ca, mg L'l
CT, mg L-l
Conductance, uS cm"l
DIC, mg L"l
air-equilibrated
initial
DOC, mg L'l
all values
x <= 5
x" > 5
Fe, mg L"l
F", total dissolved, mg
K, mg L-l
Mg, mg L"l
Mn, mg L~l
Na, mg L~l
NH4+, mg L'J
N03~, mg L"1
all batches
aliquot 3
aliquot 5
PH
air-equilibrated
initial ANC
initial BNC
P, total , mg L"1
all values
7 <= 0.010
x > 0.010
Si02, mg L"1
S042-, mg L'1
n
49
6
43
49
1
48
47
49
49
49
49
49
49
25
24
47
L-l 49
49
49
44
49
49
47
10
37
49
49
49
49
24
25
49
49
Mean > 0
RMS of %RSDa
3.2
3.3
3.2
17
7.6
17
14
1.1
10
0.81
3.5
3.7
1.6
1.7
1.4
8.5
1.7
1.7
0.74
11
0.90
4.1
3.2
3.6
3.1
0.02
0.03
0.07
16
22
5.5
3.1
0.96
n
43
0
43
48
0
48
49
49
49
27
43
49
25
24
36
49
49
49
20
49
26
29
4
24
—
—
_ _
23
0
23
49
49
Pairs with
Mean > lOsg
RMS of ZRSDa
3.2
—
3.2
17
—
17
1.1
10
0.81
1.7
3.3
1.6
1.7
1.4
4.1
1.7
1.7
0.74
2.7
0.90
2.0
2.7
0.62
2.6
—
—
5.4
—
5.4
3.1
0.96
Quantisation
Limit
n
49
49
—
49
49
49
48
48
49
49
49
49
49
48
49
49
49
11
38
—
—
__
49
49
49
10sB
0.010
0.011
_.
0.03
0.03
2.4
0.73
0.58
0.4
0.07
0.013
0.02
0.01
0.04
0.02
0.06
0.076
0.072
0.078
—
—
0.011
0.12
0.19
aRoot mean square of percent relative standard deviation (RMS of absolute SD
for pH).
130
-------
TABLE D-;
Parameter
Al, extractable, mg L"1
all values
x <= 0.010
x > 0.010
Al, total, mg L'1
all values
x <= 0.010
7 > 0.010
ANC, ueq L~l
Ca, mg L'l
CT, mg L~l
Conductance, uS cm~l
DIC, mg L~l
air-equilibrated
initial
DOC, mg L"1
all values
7 <= 5
7 > 5
Fe, mg L'l
F", total dissolved, mg L~l
X, mg L'l
Mg, mg L'~l
Mn, mg L"l
Na, mg L~l
NH4+, mg L'J
N03", mg L"1
all batches
aliquot 3
aliquot 5
pH
air-equilibrated
initial ANC
initial BNC
P, total , mg L~l
all values
x <= 0.010
x > 0.010
Si02, mg L'1
S042', mg L'1
Analytical Wii
Pairs with
n
19
11
8
18
7
11
18
15
19
17
18
19
18
11
7
14
19
17
13
13
15
15
16
12
4
19
19
19
18
5
13
19
19
Mean > 0
RMS of SR!
31
19
42
39
GLOBAL
hin-Batch Precision
Pairs with
Mean > lOsg
Da n
8
0
8
11
61 0
10 11
67
1.0 15
56 16
11 16
17 16
6.7 19
16 18
20 11
4.5 7
38 11
3.3 16
49 14
0.78 13
39 13
37 13
13 13
36 15
41 12
4.5 4
0.03 0
0.06 0
0.08 0
4.7 14
6.8 1
3.6 13
3.6 19
11 16
RMS of 2RSDa
42
—
42
10
—
10
1.0
4.4
1.2
4.0
6.7
16
20
4.5
5.3
3.6
1.2
0.78
39
0.41
14
3.7
41
4.5
—
—
_.
4.1
8.3
3.6
3.6
2.2
Quantisation
Limit
n
19
19
—
19
19
19
19
19
19
19
19
19
19
19
19
19
19
15
4
—
—
__
19
19
19
10SR
0.013
0.011
—
0.06
0.06
0.8
0.37
0.37
0.0
0.06
0.002
0.02
0.00
0.00
0.04
0.02
0.080
0.000
0.035
—
—
__
0.006
0.02
0.05
aRoot mean square of percent relative standard deviation (RMS of absolute SO
for pH).
131
-------
TABLE D-3.
USGS
Analytical Wi thin-Batch Precision
Parameter
Al , extractable, mg L"l
all values
x <= 0.010
x > 0.010
Al , total , mg L"l
all values
x <= 0.010
x > 0.010
ANC, ueq L'l
Ca, mg L~l
C1-, mg L-l
Conductance, uS cm~l
DIG, mg L"l
air-equil ibrated
initial
DOC, mg L'1
all values
x <= 5
x > 5
Fe, mg L"l
F~, total dissolved, mg L"1
K, mg L-1
Mg, mg L"l
Mn, mg L~l
Na, mg L~l
NH4+, mg L-l
N03", mg L"1
all batches
aliquot 3
aliquot 5
pH
air-equilibrated
initial ANC
initial BNC
P, total , mg L"l
all values
x <= 0.010
x > 0.010
Si02, mg L"1
SO/', mg L~!
n
7
1
6
10
0
10
3
10
10
10
10
10
10
6
4
10
10
10
10
10
10
10
10
10
0
10
10
10
9
0
9
10
10
Pairs with
Mean > 0
RMS of ?RSDa
7.5
16
4.9
4.5
—
4.5
5.9
0.66
2.6
0.51
3.6
2.1
10
12
0.89
2.1
2.9
0.91
0.33
0.73
1.6
3.7
1.4
1.4
—
0.26
0.10
0.13
3.0
—
3.0
0.87
1.3
n
4
0
4
10
0
10
10
10
10
7
10
8
4
4
10
10
10
10
10
10
10
9
9
0
—
—
9
0
9
10
10
Pairs with
Mean > 10sR
RMS of JRSDa
3.9
—
3.9
4.5
—
4.5
0.66
2.6
0.51
3.9
2.1
1.7
1.9
0.89
2.1
2.9
0.91
0.33
0.73
1.6
3.7
1.5
1.5
—
—
—
3.0
—
3.0
0.87
1.3
Quantisation
Limit
n
10
10
__
10
10
10
10
10
10
10
10
10
10
10
10
10
10
10
0
--
—
10
10
10
lOSp
0.019
0.004
_ _
0.00
0.11
1.1
0.41
0.26
1.4
0.02
0.012
0.00
0.00
0.02
0.00
0.03
0.013
0.013
—
—
--
0.000
0.04
0.27
aRoot mean square of percent
for pH).
relative standard deviation (RMS of absolute SD
132
-------
TABLE D-4.
VERSAR
Analytical
Within-Batch Precision
Pairs with Pairs with
Mean > 0 Mean > lOsp
Parameter
Al, extractable, mg L'l
all values
x <= 0.010
x > 0.010
Al, total, mg L-1
all values
7 <= 0.010
7 > 0.010
ANC, ueq L~l
Ca, mg L"1
CT, mg L-l
Conductance, uS cm~l
DIC, mg L"1
air-equilibrated
initial
DOC, mg L"l
all values
x <= 5
x > 5
Fe, mg L'1
F", total dissolved, mg L~l
'K, mg L-l
Mg, mg L"l
Mn, mg L~l
Na, mg L~l
NH/, mg I'}
N03~, mg L'1
all batches
aliquot 3
aliquot 5
PH
air-equilibrated
initial ANC
initial BNC
P, total , mg L"1
all values
x <= 0.010
x > 0.010
Si02, mg L"1
SO/', mg L'1
n
48
4
44
49
0
48
49
49
49
49
49
49
43
6
49
49
49
49
35
49
47
49
19
30
49
49
49
49
1
48
49
49
RMS Of
4.2
3.7
4.2
6.6
—
6.6
16
0.55
3.1
17
1.6
1.7
2.5
2.5
2.5
2.9
2.8
1.4
0.52
24
0.94
7.3
4.8
5.0
2.4
0.02
0.01
0.02
13
0.00
13
1.4
17
ZRSDa n
7
0
7
49
0
49
49
48
48
49
49
44
38
6
49
49
49
49
35
49
18
39
17
23
—
—
__
49
1
48
21
49
RMS of ZRSDa
2.9
—
2.9
6.6
—
6.6
0.55
3.1
17
1.6
1.7
2.6
2.6
2.5
2.9
2.8
1.4
0.52
24
0.94
1.9
4.2
4.8
4.0
—
—
13
0.00
13
1.5
17
Quantitation
Limit
n
49
49
—
49
49
49
48
48
49
49
49
49
49
49
49
49
47
17
30
—
—
__
49
49
49
10sp
0.053
0.013
__
0.02
0.16
1.3
0.15
0.15
1.0
0.01
0.012
0.06
0.01
0.01
0.15
0.21
0.233
0.177
0.261
—
—
__
0.005
3.56
0.18
aRoot mean square of percent relative standard deviation (RMS of absolute SD
for pH).
133
-------
APPENDIX E
OVERALL AND ANALYTICAL AMONG-BATCH PRECISION
BY LABORA TORY FOR 23 PARAMETERS,
EASTERN LAKE SURVEY - PHASE I
Values were estimated from field natural. Lot 2 (FN2) and field natural. Lot 3 (FN3) audit sampled data,
field and laboratory high synthetic (FH, LH) audit sample data, and field and laboratory low synthetic
(FL and LL) audit sample data.
135
-------
TABLE E-l.
EMSI
Overall Among-Batch Precision
FN2 Samples (n = 18)
Parameter
Mean
2RSD*
FN3 Samples (n = 3)
Mean
SRSD*
Al, mg L-l
total extractable
total
ANC, ueq L'1
Ca, mg L"l
CT. mg L-l
Conductance, uS cm"l
DIC, mg L-l
air-equilibrated
initial
DOC, mg L'1
Fe, mg L~l
F", total dissolved,
K, mg L~l
Mg, mg L"1
Mn, mg L~l
Na, mg L~l
NH4+, mg L'1
N03~, mg L"l
PH
air-equil ibrated
initial ANC
initial BNC
P, total , mg L'l
Si02, mg L"1
S042-, mg L'1
====================
0.175
0.357
4.3b
1.87
0.51
26.7
0.17b
0.39b
3.2
0.02b
mg L'l 0.076
0.49
0.34
0.07
0.68
0.07b
1.449
5.25
5.07
5.11
0.001
4.36
6.89
====================
25
16
74
7.9
2.0
3.1
52
16
4.3
75
2.6
4.6
1.9
6.4
2.6
12
2.6
0.40
0.03
0.05
158
4.0
2.1
===========
0.002b
0.032b
841.0
57.84C
1.40
97.4
9.61
9.86
1.5b
0.00b
0.035
0.52
2.74
O.OO15
1.34
0.01b
1.46
8.18
7.86
7.88
0.001
2.66
3.32
===============
143
125
1.0
135C
1.4
1.1
4.4
4.2
10
2.0
3.5
1.5
__
3.2
201
3.3
0.10
0.08
0.06
187
3.7
0.83
aPercent relative standard deviation (absolute SD for pH). Not applicable for
pairs with Y = 0.
bMean less than quantisation limit (see Table C-l).
cWhen one confirmed but questionable value of 147.7 mg L~l is omitted, mean =
12.91 mg L'1 and RSD = 5.7%.
136
-------
TABLE E-2.
GLOBAL
Overall Among-Batch Precision3
FN2 Samples (n = 3)
Parameter
Mean
ZRSD&
Al, mg L'1
total extractable
total
ANC, ueq L"1
Ca, mg L"1
CT, mg L'1
Conductance, uS cm~l
DIC, mg L'1
air-equilibrated
initial
DOC, mg L"1
Fe, mg L~l
F~, total dissolved, mg L-1
K, mg L~!
Mg, mg L"1
Mn, mg L"1
Na, mg L~l
NH4+, mg L'1
N03~, mg L'1
PH
0.112
0.242
5.5C
1.81
0.52
25.6
0.39C
0.68C
3.4
o.oic
0.079
0.49
0.34
0.10C
0.69
0.09C
1.471
59
7.9
85
3.1
2.2
3.5
80
19
11
173
5.5
2.0
1.7
64
0.83
43
2.5
air-equilibrated
initial ANC
initial BNC
P, total, mg L~l
Si02, mg L'1
S042', mg L"1
3———»————3——S™3SS
aNo FN3 samples were analyzed by Global.
DRelative standard deviation (absolute SD for pH).
cMean less than quantisation limit (see Table C-2),
5.14
5.13
5.11
0.001C
4.88
6.72
0.03
0.02
0.02
284
5.3
1.8
737
-------
TABLE E-3.
USGS
Overall Among-Batch Precision3
FN2 Samples (n = 3)
Parameter
Mean
2RSDb
Al, mg L'1
total extractable
total
ANC, ueq L~l
Ca, mg L"1
Cl~, mg L~l
Conductance, uS cm~l
DIC, mg L-1
air-equilibrated
initial
DOC, mg L-1
Fe, mg L~l
F~, total dissolved, mg L"1
K, mg L-l
Mg, mg L"1
Mn, mg L~l
Ha, mg L'l
NH,
N0
pH
mg L"1
mg L'1
air-equilibrated
initial ANC
initial BNC
P, total , ma L~l
SiO?, mg L-1
SO
'2-
2-
mg L
_i
0.165
0.166C
-3.4C
2.36
0.83
28.6
0.26C
0.23C
3.7
0.01
0.077
0.49
0.35
0.07
0.71
0.07
1.457
5.10
5.04
5.06
0.003C
4.27
7.27
11
70
395
13
51
6.8
38
37
25
7.7
6.6
1.1
2.1
0.81
4.7
50
3.1
0.05
0.05
0.01
96
0.66
9.7
3No FN3 samples were analyzed by USGS.
^Percent relative standard deviation (absolute SD for pH)
cMean less than quantitation limit (see Table C-3).
138
-------
TABLE E-4.
VERSAR
Overall Among-Batch Precision
FN2 Samples (n = 17)
FN3 Samples (n = 4)
Parameter
Mean
ZRSDa
air-equilibrated
initial ANC
initial BNC
total, mg L"1
mg L~*
5.13
5.05
5.05
0.002b
4.22
0.04
0.05
0.03
127
10
S042", mg L
-1
Mean
8.26
7.68
7.72
0.002b
2.78b
Al. mg L-l
total extractable
total
ANC, ueq L'l
Ca, mg L~l
CT, mg L-l
Conductance, uS cm"l
DIG, mg L"l
air-equilibrated
initial
DOC, mg L~l
Fe, mg L"l
F", total dissolved, mg L"l
K, mg L-l
Mg, mg L'l
Mn, mg L"1
Na, mg L"l
NH4+, mg L'1
N03", mg L'1
PH
0.205
0.286
0.8b
1.89
0.69
26.5
0.17b
0.45b
3.16
0.02b
0.078
0.48 b
0.35
0.06
0.66
0.04b
1.386b
35
25
797
1.2
6.3
0.74
24
14
6.4
33
4.2
4.6
1.9
7.7
6.6
93
7.4
0.002b
0.014b
853.7
13.10
1.38
95.7
9.74
9.99
1.3b
O.OQb
0.035b
0.47b
2.84
0.01
1.28
0.01b
1.363b
43
86
3.0
3.2
2.3
1.8
3.8
4.7
6.3
6.9
16
4.1
200
7.9
68
5.3
0.05
0.09
0.10
122
7.2
6.99 10 3.22 4.0
apercent relative standard deviation (absolute SD for pH}. Not applicable for
pairs with Y = 0.
bMean equal to or less than quantisation limit (see Table C-4).
139
-------
TABLE E-5.
EMS I
Overall Among-
Batch Precision
FH Samples (n = 14)
Analytical Among-
Batch Precision
LH Samples (n = 9)
Parameter
Mean
2RSDa
pH
air-equilibrated
initial ANC
initial BNC
P, total, ma L~l
SiO,, mg L"1
SO,
2-
mg L
-1
7.62
6.96
7.00
0.054
9.01
14.6
0.58
0.25
0.26
16
14
2.1
Mean
81
94
06
062
08
14.73
%RSDa
Al , mg L"l
total extractableb
total
ANC, ueq L'1
Ca, mg L"1
Lot 4 (n = 11, 9)
Lots 5 and 6 (n = 3, 0)
Cl", mg L'l
Lot 4 (n = 6, 9)
Lots 5 and 6 (n = 3, 0)
Conductance, uS cm"!
DIC, mg L-l
air-equilibrated
initial
DOC, mg L~l
Fe, mg L~lb
F~, total dissolved, mg L"1
K, mg L-l
Mg, mg L~l
Mn, mg L"l
Na, mg L"l
NH.+, mg L"1
N03~, mg L"1
—
0.233
465.8
1.65
1.73
3.33
4.03
103.3
4.79
7.43
9.8
—
0.438
3.07
2.32
1.13
11.83
1.25
1.661
—
23
23
9.3
3.3
3.0
1.1
1.4
32
28
24
—
5.5
2.5
1.8
16
8.6
16
29
0.153
0.219
483.9
1.50
—
3.36
—
102.3
4.99
7.26
10.4
0.18
0.439
3.08
2.33
1.38
11.67
1.35
1.353
20
35
19
14
—
1.5
—
2.1
33
27
20
11
2.9
2.4
2.1
3.2
9.9
9.1
45
0.14
0.30
0.31
4.3
15
2.4
^Percent relative standard deviation (absolute SD for pH). Not applicable for
pairs with X = 0.
t>Laboratory synthetic audit samples only (see Section 4).
140
-------
TABLE E-6.
GLOBAL
Overall Among-
Batch Precision
FH Samples (n = 5)
Analytical Among-
Batch Precision
LH Samples (n = 2)
Parameter
Mean
%RSDa
Mean
2RSDa
Al, mg L"1
total extractable0
total
ANC, peq L"1
Ca, mg L"1
Lot 4 (n = 3, 1)
Lots 5 and 6 (n = 2,
CT, mg L"1
Lot 4 (n = 3, 1)
Lots 5 and 6 (n = 2,
Conductance, uS cm~l
DIC, mg L"1
air-equilibrated
initial
DOC, mg L'l
Fe, mg L'lb
F", total dissolved, mg
K, mg L"1
Mg, mg L"l
.Mn, mg L'1
Na, mg L"l
NH4+, mg L"1
N03~, mg L'1
PH
air-equilibrated
initial ANC
initial BNC
P, total, mg L"1
Si02, mg L"1
S042-, mg L'1
—
0.175
482.7
1.69
1) 1.90
3.15
1) 4.10
100.5
4.94
5.26
10.8
—
L-l 0.440
3.08
2.27
1.17
12.04
1.23
1.313
7.89
7.15
7.10
0.048
10.13
13.96
—
6.3
15
2.4
4.1
1.8
5.7
4.5
26
18
20
—
3.2
4.5
2.9
18
14
10
60
0.08
0.22
0.25
19
13
2.8
0.129
0.177
392.1
1.65
1.82
3.27
3.95
103.1
5.97
6.46
9.9
0.18
0.430
2.97
2.24
1.44
13.07
1.27
1.464
7.94
7.00
7.04
0.054
11.55
14.65
21
0.80
49
—
—
—
—
3.9
9.4
3.7
4.3
7.9
2.3
3.6
3.8
2.0
2.8
0.17
15
0.05
0.01
0.05
9.3
4.0
0.77
aPercent relative standard deviation (absolute SD for pH).
pairs with X = 0 or where n = 1.
bLaboratory synthetic audit samples only (see Section 4).
Not applicable for
141
-------
TABLE E-7.
Overall Among-
Batch Precision
FH Samples (n = 3)
USGS
Analytical Among-
Batch Precision
LH Samples (n = 1)
Parameter
Mean
ZRSDa
so.
2'
mg L
"1
14.53
3.2
Mean
14.10
2RSDa
Al, mg L'1
total extractable'3
total
ANC, peq L"1
Ca, mg L'1
Lot 4 (n = 0, 0)
Lots 5 and 6 (n = 3, 1)
CT, mg L'1
Lot 4 (n = 0, 0)
Lots 5 and 6 (n = 3, 1)
Conductance, uS cm'1
DIC, mg L'1
air-equil ibrated
initial
DOC, mg L"1
Fe, mg L~lb
F~, total dissolved, mg L"1
K, mg L""1
Mg, mg L"1
Mn, mg L'1
Na, mg L"1
NH4+, mg L-1
N03", mg L"1
pH
air-equilibrated
initial ANC
initial BNC
P, total, mo L'1
SiOo, mg L
—
0.175C
542.7
—
2.00
—
3.50
110.6
6.33
6.55
10.0
—
0.455
3.11
2.33
1.32
12.47
1.32
1.700
7.71
7.29
7.53
0.061
10.07
—
3.2
2.8
—
13
—
17
1.4
13
26
18
—
0.46
4.9
2.8
6.4
0.90
10
29
0.06
0.35
0.24
23
1.6
0.159
0.175
554.7
__
1.81
__
2.96
111.1
6.01
6.03
10.1
0.12
0.456
3.10
2.28
1.47
12.34
1.42
2.02
7.65
7.38
7.38
0.075
10.24
__
__
__
__
__
__
__
__
__
__
__
__
—
__
__
__
—
—
—
__
—
__
—
—
aPercent relative standard deviation (absolute SD for pH). Not applicable for
pairs with Y = 0 or where n = 1.
^Laboratory synthetic audit samples only (see Section 4).
cMean less than quantisation limit (see Table C-3).
142
-------
TABLE E-8.
VERSAR
Overall Among-
Batch Precision
PH
air-equilibrated
initial ANC
initial BNC
P, total, mg L~l
Si02,
V
mg
, mg
mg
L'1
7.75
7.14
7.16
0.062
8.94
14.32
0.64
0.20
0.21
26
14
8.4
Analytical Among-
Batch Precision
FH Samples (n = 14}
Parameter
Mean
2RSD3
LH Samples (n = 9)
Mean 2RSDa
AT, mg L~l
total extractableb
total
ANC, ueq L~l
Ca, mg L~l
Lot 4 (n = 6, 4)
Lots 5 and 6 (n = 8, 5)
Cl", mg L"l
Lot 4 (n = 6, 4)
Lots 5 and 6 (n = 8, 5)
Conductance, uS cm"l
DIC, mg L"l
air-equilibrated
initial
DOC, mg L"l
Fe, mg L~lb
F", total dissolved, mg L"l
K, mg L-l
Mg, mg L-l
Mn, mg L"1
Na, mg L"l
NH4+, mg L'1
NO.', mg L'1
—
0.179
471.4
1.68
2.03
4.52
4.15
105.9
4.22
4.57
9.9
—
0.451
2.96
2.45
1.23
11.64
1.26
1.800
—
37
13
2.2
2.1
57
23
4.2
18
21
16
.-
14
9.2
2.2
6.0
7.5
12
21
0.177
0.176
502.5
1.65
2.02
3.49
6.75
105.7
4.19
4.50
10.2
0.18
0.430
2.96
2.43
1.38
11.76
1.35
1.57
12
37
16
2.0
1.0
13
92
4.5
17
20
16
6.3
2.4
6.4
1.9
1.9
8.1
18
44
95
14
14
060
9.20
0.13
0.20
0.21
22
14
^Percent relative standard deviation (absolute SD for pH),
pairs with T = 0.
bLaboratory synthetic audit samples only (see Section 4).
14.20 6.5
Not applicable for
143
-------
TABLE E-9.
EMS I
Overall Among-
Batch Precision
FL Samples (n = 16)
Analytical Among-
Batch Precision
LL Samples (n = 16)
Parameter
Mean
2RSDa
S042-, mg L'1
2.38
9.8
Mean
%RSDa
Al , mg L"l
total extractableb
total
ANC, peq L'1
Ca, mg L~l
Lot 4 (n = 11, 10)
Lots 5 and 6 (n = 5, 6)
Cl", mg L"l
Lot 4 (n = 11, 10)
Lots 5 and 6 (n = 5, 6)
Conductance, uS cm"!
DIC, mg L"l
air-equilibrated
initial
DOC, mg L'l
Fe, mg L'lb
F-, total dissolved, mg L"l
K, mg L-l
Mg, mg L"l
Mn, mg L"l
Na, mg L'l
NH4+, mg L'1
N03", mg L'1
pH
air-equilibrated
initial ANC
initial BNC
P, total, ma L~l
Si 0?, mg L'1
—
0.025b
112.8
0.14
0.17
0.39C
0.32C
18.9
1.34
1.83
0.9C
—
0.041
0.22
0.41
0.09
2.75
0.18C
0.547
7.24
6.83
6.93
0.022
1.00
__
16
6.1
8.4
3.6
87
2.6
7.9
8.2
7.2
45
—
24
11
8.8
7.0
3.9
23
8.3
0.05
0.07
0.13
35
43
0.017
0.033
115
0.13
0.17
0.29
0.31
18.5
1.32
1.75
1.28
0.08
0.040
0.22
0.42
0.10
2.77
0.20
0.609
7.23
6.82
6.90
0.021
0.97
34
54
6.8
9.9
9.5
7.3
1.7
5.4
8.5
15
78
21
6.8
10
2.9
6.3
3.4
17
60
0.05
0.09
0.12
24
16
aPercent relative standard deviation (absolute SD for pH)
pairs with Y = 0.
DLaboratory synthetic audit samples only (see Section 4).
°Mean less than quantisation limit (see Table C-l).
2.30 1.8
Not applicable for
144
-------
TABLE E-10.
GLOBAL
Overall Among-
Batch Precision
FL Samples (n = 7)
Analytical Among-
Batch Precision
LL Samples (n = 8)
Parameter
Mean
%RSDa
SO,
2-
mg L
-1
2.21
3.3
Mean
2.21
Al, mg L'1
total extractableb
total
ANC, ueq L-1
Ca, mg L"1
Lot 4 (n = 1, 2)
Lots 5 and 6 (n = 6, 6)
Cl", mg L"1
Lot 4 (n = 1, 2)
Lots 5 and 6 (n = 6, 6)
Conductance, uS cm~l
DIC, mg L"1
air-equil ibrated
initial
DOC, mg L"1
Fe, mg L~lb
F", total dissolved, mg L"1
K, mg L'1
Mg, mg L~J
Mn, mg L~l
Na, mg L"1
NH4+, mg L"1
N03", mg L"1
pH
air-equilibrated
initial ANC
initial BNC
P, total, mg L~l
SiOo, mg L"1
—
0.020C
115.13
0.16C
0.17C
0.26
0.33
18.0
1.37
1.66
1.3C
—
0.041
0.23
0.42
0.09
2.66
0.15
0.245
7.40
6.96
7.00
0.019C
1.29
—
38
6.2
—
4.4
—
8.4
7.1
13
12
57
—
5.9
12
10
19
1.7
20
105
0.02
0.10
0.13
35
9.1
0.015
0.019
115.15
0.17
0.16
0.30
0.33
17.8
1.35
1.75
1.5
0.06d
0.040
0.24
0.42
0.10
2.51
0.16
0.29
7.40
6.99
7.00
0.024
1.26
33
38
6.4
4.3
9.9
2.4
7.9
8.0
19
15
50
40
5.4
10
8.2
6.3
19
11
92
0.02
0.11
0.12
82
11
2.5
aPercent relative standard deviation (absolute SD for pH). Not applicable for
pairs with Y = 0 or where n = 1.
^Laboratory synthetic audit samples only (see Section 4).
cMean less than quantisation limit (see Table C-2).
dMean equal to or less than quantisation limit (see Table D-2).
145
-------
TABLE E-ll.
USGS
Overall Among-
Batch Precision
FL Samples (n = 4)
Analytical Among-
Batch Precision
LL Samples (n = 3)
Parameter
Mean
2RSDa
Ca, mg L"1
Lot 4 (n = 0, 0)
Lots 5 and 6 (n = 4, 3)
Cl~, mg L"1
Lot 4 (n = 0, 0)
Lots 5 and 6 (n = 4, 3)
Conductance, uS cm"1
DIC, mg L"1
air-equilibrated
initial
DOC, mg L'1
Fe, mg L"lb
F~, total dissolved, mg L"
K, mg L"1
Mg, mg L'1
Mn, mg L"1
Na, mg L"1
NH,
NO-
PH
mg L
mg L
'1
"1
air-equilibrated
initial ANC
initial BNC
P, total, mg L'1
Si02, mg
mg
L"1
SO,
2-
mg L
'1
0.17
0.40
21.1
1.41
1.44
1.2
0.040
0.32
0.43
0.09
2.76
0.16
0.267
7.25
6.93
7.22
0.020
1.03
2.23
10
24
1.6
7.7
1.9
49
1.3
27
3.3
1.6
2.1
8.9
108
0.15
0.12
0.50
19
1.1
4.8
Mean
0.17
0.33
20.8
1.44
1.43
1.3C
0.05
0.040
0.23
0.43
0.10
2.75
0.17
0.355
08
80
11
0.022
1.04
2.23
2RSDa
Al, mg L"1
total extractableb
total
ANC, ueq L"1
—
0.018C
112.6
—
12
14
0.014d
0.080
106.5
28
120
2.9
8.6
12
1.2
8.3
14
7.1
12
1.4
13
4.0
2.8
2.7
8.5
81
0.07
0.14
0.33
24
2.4
4.1
aPercent relative standard deviation (absolute SD for pH). Not applicable for
pairs with J = 0.
bLaboratory synthetic audit samples only (see Section 4).
cMean equal to or less than quantitation limit (see Table C-3).
^Mean less than quantitation limit (see Table D-3).
146
-------
TABLE E-12.
VERSAR
Overall Among-
Batch Precision
Analytical Among-
Batch Precision
FL Samples (n = 16)
Parameter
Al, mg L~l
total extractableb
total
ANC, ueq L'1
Ca, mg L"1
Lot 4 (n = 11, 11)
Lots 5 and 6 (n = 5, 5)
CT, mg L'1
Lot 4 (n = 11, 11)
Lots 5 and 6 (n = 5, 5)
Conductance, uS cm"!
DIG, mg L'1
air-equilibrated
initial
DOC, mg L~l
Fe, mg L~lb
F~, total dissolved, mg L~l
K, mg L"1
Mg, mg L~l
Mn, mg L~l
Na, mg L"1
NH4+, mg L'1
N03', mg L'1
pH
air-equilibrated
initial ANC
initial BNC
P, total , mg L~l
Si02, mg L"1
S042-, mg L'1
Mean
—
0.025C
108.9
0.15C
0.16C
0.28C
0.43
19.0
1.35
1.52
0.9C
—
0.044
0.22
0.44
0.09
2.68
0.14°
0.458C
7.43
6.85
6.91
0.02QC
0.87C
2.21C
%RSDa
—
20
7.0
24
5.7
11
26
3.8
9.3
8.9
41
__
24
9.3
2.3
5.8
7.1
36
32
0.16
0.12
0.09
25
21
12
LL Samples
Mean
0.022d
0.027
103.2
0.14
0.15
0.31
0.42
18.9
1.36
1.54
0.9
0.07
0.04
0.22
0.44
0.08
2.68
0.19d
0.481
7.33
6.85
6.89
0.022
0.94
2.32
(n = 16)
»SDa
17
32
11
5.3
2.9
18
51
3.8
9.0
8.7
30
10
5.0
9.1
1.8
24
6.2
28
18
0.15
0.12
0.11
22
22
5.9
aPercent relative standard deviation (absolute SD for pH). Not applicable for
pairs with Y = 0.
^Laboratory synthetic audit samples only (see Section 4).
cMean equal to or less than quantisation limit (see Table C-4).
dMean less than quantitation limit (see Table D-4).
147
------- |