United States Office of Water (4607) EPA-815-R-00-023
Environmental Protection Washington, D.C. 20460 December 2000
Agency
ARSENIC OCCURRENCE IN PUBLIC
DRINKING WATER SUPPLIES
Prepared for:
Office of Ground Water and Drinking Water
1200 Pennsylvania Avenue, N.W., MS 4607
Washington, D.C. 20460
MS 4607
Andrew E. Schulman
USEPA Work Assignment Manager
Prepared by:
ISSI Consulting Group
8455 Colesville Road, Suite 915
Silver Spring, MD 20910
Under Contract No: 68-C7-0005
The Cadmus Group, Inc.
1901 North Fort Myer Drive, Suite 900
Arlington, VA 22209
Under Contract No.: 68-C-99-206
And:
ICF Consulting
101 Lucas Valley Road
San Rafael, CA 94903
-------
This page intentionally left blank
-------
Table of Contents
List of Acronyms iii
Acknowledgments v
Executive Summary vi
1. Introduction 1
1.1 Purpose of this Document 2
1.2 Organization of the Document 2
2. Sources of Arsenic 5
2.1 Physical and Chemical Properties of Arsenic 5
2.1.1 Environmentally Relevant Arsenic Species 5
2.2 Natural Sources of Arsenic 8
2.3 Anthropogenic Sources of Arsenic 14
3. Fate and Transport of Arsenic 21
3.1 Relationship of Fate and Transport Properties to Source Intake 21
3.2 Relationship of Fate and Transport Properties to Treatment and Distribution . . 22
4. Sources of Data on Arsenic Occurrence in Drinking Water Supplies 25
4.1 Arsenic Occurrence and Exposure Database (AOED) 25
4.1.1 Safe Drinking Water Information System (SDWIS) 27
4.1.2 State Compliance Monitoring Databases 28
4.1.3 Building the AOED from SDWIS and State Compliance Monitoring
Databases 47
4.2 Comparison Databases 50
4.2.1 National Arsenic Occurrence Survey (NAOS) Database 50
4.2.2 USGS Arsenic Databases 52
4.2.4 Metropolitan Water District of Southern California (Metro) Database . . 54
4.3 Other Databases 55
4.3.1 1969 Community Water Supply Survey 55
4.3.2 1978 Community Water Supply Survey 55
4.3.3 Rural Water Survey 56
4.3.4 National Organics Monitoring Survey 56
4.3.5 Western Coalition of Arid States Research Committee Arsenic
Occurrence Study 56
4.3.6 Association of California Water Agencies Database (ACWA) 57
5. Arsenic Occurrence Patterns in the United States 59
5.1 Stratification by Source Water Type 59
5.2 Stratification by System Size 63
5.3 Stratification by System Type 70
-------
5.4 Regional Stratification 73
5.5 Arsenic Distributions at the State Level 77
5.6 Summary of Patterns of Arsenic Occurrence 85
6. National Occurrence Estimates 87
6.1 Arsenic National Occurrence Project!on Methodology 87
6.1.1 System Means 88
6.1.2 State Exceedance Probability Distributions 89
6.1.3 Regional Exceedance Probability Distributions 92
6.1.4 National Exceedance Probability Distributions 96
6.1.5 Number of Systems Exceeding Alternative MCLs 97
6.2 Arsenic National Occurrence Estimates Results 97
6.2.1 Community Water Supply Systems 100
6.2.2 Non-Transient, Non-Community Water Supply Systems 100
6.3 Comparisons of Occurrence Estimates 105
6.3.1 Comparison of AOED, NAOS, USGS, MWDSC, and Wade Miller
Occurrence Estimates at the National and Regional Level 105
6.3.2 Comparison of AOED, Kennedy-Jenks, and Saracino-Kirby
Occurrence Estimates for California 110
6.4 Uncertainty Analysis 113
6.4.1 Purpose of Uncertainty Analysis 113
6.4.2 Uncertainty Analysis Methodology 116
7. Intra-system Variability 123
7.1 Purpose of Analyses 123
7.2 Available Data 123
7.3 Analytical Methods and Results 124
7.3.1 Estimation of Concentrations for Non-Detects 126
7.3.2 Log-Normal Mixed Model 126
7.3.3 Results 127
7.4 Summary of Intra-system Analyses 128
8. Temporal Variability 129
8.1 Purpose of Analysis 129
8.2 Available Data and Results 129
9. References 133
-------
Appendices
Appendix A: Adapted Regression on Order Statistics Methodology
Appendix B: Analysis Results
Appendix B-l: State Exceedance Probability Distributions
Appendix B-2: Box Plots
Appendix B-3: Lognormal Probability Plots of System Means
Appendix C: Summaries of Pre-1980 Data Sets
Appendix D: Database Specifications and Data Conditioning
Appendix D-l: Individual State Database Specifications for Preliminary
Database
Appendix D-2: AOED Database Specifications
Appendix D-3: Initital Datat Conditioning Process
List of Acronyms
AA
ACWA
AES
ANOVA
AOED
ASDTR
AWWA
CCA
CV
CWS
CWSS
DMAA
DSMA
EPA
FIFRA
FRDS
GW
IAOED
ICP
MCL
Metro
MMAA
MS
MSMA
NAS
NAOS
NIPDWR
MRS
NOF
Atomic Adsorption
Association of California Water Agencies
Atomic Emission Spectrometry
Analysis of Variance
Arsenic Occurrence and Exposure Database
Agency for Toxic Substances and Disease Registry
American Water Works Association
Chromated Copper Arsenate
Coefficient of Variation
Community Water Supply
Community Water Supply Surveys
Dimethylarsinic Acid
Di sodium Methanearsonate
Environmental Protection Agency
Federal Insecticides, Fungicides, and Rodenticides Act
Federal Reporting Data System
Ground Water
Intermediate Arsenic Occurrence and Exposure Database
Inductively Coupled Plasma
Maximum Contaminant Level
Metropolitan Water District of Southern California
Monomethyl-Arsionic Acid
Mass Spectrometry
Monosodium Methanearsonate
National Academy of Sciences
National Arsenic Occurrence Survey
National Interim Primary Drinking Water Regulations
National Inorganics and Radionuclides Survey
Natural Occurance Factor
in
-------
NOMS
NPDWR
NPL
NRC
NTNCWS
NWIS
OGWDW
POE
PWS
PWSID
RDS
RIA
ROS
RWS
SDWA
SDWIS
SW
TMA
TNCWS
TOC
TRI
USEPA
USGS
VOC
WATSTORE
WESCAS
WITAF
National Organics Monitoring Survey
National Primary Drinking Water Regulations
National Priorities List
National Research Council
Non-Transient Non-Community Water Supply
National Water Information System
Office of Ground Water and Drinking Water
Points of Entry
Public Water Supply or Supplies
Public Water Supplies Identification Number
Raw Data Sets
Regulatory Impact Analysis
Regression on Order Statistics
Rural Water Survey
Safe Drinking Water Act
Safe Drinking Water Information System
Surface Water
Trimethylarsine
Transient Non-Community Water Supplies
Total Organic Carbon
Toxics Release Inventory
United States Environmental Protection Agency
United States Geological Society
Volatile Organic Compounds
USGS's Water Quality Database
Western Coalition of Arid States
Water Industry Technical Action Fund
IV
-------
Acknowledgments
Primary Authors and Contributors:
Kathleen Bell, IS SI Consulting Group
Jonathan Cohen, ICF Consulting
Stiven Foster, IS SI Consulting Group
Eric Hack, ICF Consulting
Robert Iwamiya, ICF Consulting
David Kaczka, IS SI Consulting Group
Jonathan Koplos, The Cadmus Group
Frank Letkiewicz, The Cadmus Group
Andrew Schulman, U.S. EPA OGWDW
Ben Smith, U.S. EPA OGWDW
Jennifer Wu, U.S. EPA OGWDW
Internal USEPA Consultants:
Timothy Barry, U.S. EPA OPPE
Irene S. Dooley, U.S. EPA OGWDW
Henry Kahn, U.S. EPA OST
Jade Lee, U.S. EPA OST
Elizabeth Margosches, U.S. EPA OPPTS
External Peer Reviewers:
Andrew Eaton, Montgomery Watson Laboratories, Pasadena, California
Chuck Kroll, SUNY College of Environmental Science and Forestry, Syracuse, New York
David Ruppert, Cornell University, Ithaca, New York
-------
Executive Summary
The Safe Drinking Water Act (SOWA), 42 U.S.C. §§300f-300j, originally enacted in
1974, directs the U. S. Environmental Protection Agency (EPA) to identify and regulate
contaminants in public drinking water. Section 1412(b)(12)(A) of SDWA, as amended by the
1996 Amendments, required EPA to propose a National Primary Drinking Water Regulation for
arsenic by January 1, 2000, and to issue a final regulation by January 1, 2001. One of the
elements that supports the development of the proposed regulation is information on arsenic
occurrence in drinking water, specifically estimates of the size of populations and number of
systems that are affected by different levels of arsenic in drinking water. This report presents the
arsenic occurrence analysis, which was prepared for the EPA Office of Ground Water and
Drinking Water (OGWDW) by ISSI under Contract 68-C7-0005, and by The Cadmus Group
under Contract 68-C-99206, both with subcontract support by ICF Consulting.
Sources of Arsenic in the Environment
Arsenic is released to the environment from a variety of natural and anthropogenic
sources. In the environment, arsenic occurs in rocks, soil, water, air, and in biota. Average
concentrations in the earth's crust reportedly range from 1.5 to 5 mg/kg (Cullen and Reimer,
1989). Higher concentrations are found in some igneous and sedimentary rocks, particularly in
iron and manganese ores (Welch et al, 1988). In addition, a variety of common minerals contain
arsenic, of which the most important are arsenopyrite (FeAsS), realgar (AsS), and orpiment
(As2S3). Natural concentrations of arsenic in soil typically range from 0.1 to 40 mg/kg, with an
average concentration of 5 to 6 mg/kg (National Academy of Sciences (NAS), 1977). Through
erosion, dissolution, and weathering, arsenic can be released to ground water or surface water.
Geothermal waters can be sources of arsenic in ground water, particularly in the Western United
States (Nimick et al., 1998, Welch et al., 1988). Other natural sources include volcanism and
forest fires.
Anthropogenic sources of arsenic relate to its use in the lumber, agriculture, livestock,
and general industries. Most agricultural uses of arsenic are banned in the United States.
However, organic arsenic is a constituent of the organic herbicides monosodium
methanearsonate (MSMA) and di sodium methanearsonate (DSMA), which are currently applied
to cotton fields as herbicides (Jordan etal., 1997). Organic arsenic is also a constituent of feed
additives for poultry and swine, and appears to concentrate in the resultant animal wastes (NAS,
1977). The potential impact of arsenic in animal wastes used to fertilize crops is uncertain.
Most of the arsenic used in the United States is for the production of chromated copper
arsenate (CCA), the wood preservative (Reese, 1998). CCA is used to pressure treat lumber and
is classified as a restricted use pesticide by the USEPA. A significant industrial use of arsenic is
the production of lead-acid batteries, while small amounts of very pure arsenic metal are used to
produce the semiconductor crystalline gallium arsenide, which is used in computers and other
electronic applications.
VI
-------
Arsenic is also released from industrial processes, including the burning of fuels and
wastes, mining and smelting, pulp and paper production, glass manufacturing, and cement
manufacturing (USEPA, 1998b). In addition, past waste disposal sites may be contaminated with
arsenic. Arsenic is a contaminant of concern at 916 of the 1,467 sites on the National Priorities
List (NPL) (Agency for Toxic Substances and Disease Registry (ATSDR), 1998). Sites included
on the NPL have the potential to release contaminants to ground water or surface water in the
vicinity of the site.
Anthropogenic releases of arsenic to the environment can be estimated from Toxics
Release Inventory (TRI) data. These data indicate that 7,947,012 pounds of arsenic and arsenic-
containing compounds were released to the environment in 1997, a significant increase from
3,536,467 pounds in 1995 (USEPA, 1999a). The increase primarily occurred at one facility,
where arsenic on-site land releases increased by 3.58 million pounds from 1995 to 1997 because
of a change in the facility's smelting process that was implemented to reduce sulfur dioxide
emissions. The TRI data do omit some potentially significant arsenic sources, including arsenic
associated with the application of herbicides and fertilizers and arsenic released from mining
facilities and electric utilities.
Arsenic Fate and Transport
Once arsenic released from natural or anthropogenic sources enters ground water or
surface water, a variety of processes affect its fate and transport. These include oxidation-
reduction reactions, transformations, ligand exchange, and biotransformations. The factors that
affect these reactions include the oxidation state of the arsenic, oxidation-reduction potential
(Eh), pH, concentrations of iron, metal sulfides, and sulfides, temperature, salinity, and the
distribution and composition of biota (ATSDR, 1998; Robertson, 1989; Welch et al, 1988). The
predominant forms of arsenic in ground water and surface water are arsenate (+5) and arsenite
(+3). Arsenite is generally associated with anaerobic conditions. Oxidation state appears to be
the most important factor that determines the fate and transport of arsenic through drinking water
treatment systems. Arsenate is more easily removed because of its ionic charge, and activated
alumina, ion exchange, and reverse osmosis technologies can achieve relatively high arsenic
removal rates. These technologies do not achieve comparable removal rates for arsenite.
Oxidization of arsenite to arsenate can improve removal efficiencies. Treatment efficiencies may
also be affected by water pH, depending on the technology applied, and competing ions. Higher
pH tends to decrease removal rates (Rubel and Hathaway, 1987); high sulfate, fluoride, and
phosphate concentrations also tend to decrease removal rates (Jekel, 1994).
Data Sources on Arsenic Occurrence in Drinking Water Systems
There are a variety of sources of information on arsenic in drinking water. This study is
based largely on arsenic data from 25 State compliance monitoring data sets, and information on
individual system characteristics that are provided in the Safe Drinking Water Information
System (SDWIS). Figure ES-1 presents the States for which compliance monitoring data were
available. As this figure shows, the Midwestern, South Central, North Central, and Western
Vll
-------
Figure ES-1: States with Suitable Arsenic Compliance Monitoring Data
NORTH
DAKOTA (MINNESOTA
NEW HAMPSHIRE
MASSACHUSETTS
k
RHODE ISLAND
CONNECTICUT
NEW JERSEY
States with compliance data
States without compliance data
Vlll
-------
regions of the United States are well represented, but fewer compliance monitoring data sets are
available for the States in the New England, Mid-Atlantic, and Southeastern regions.
These compliance monitoring data sets offer several benefits. For many States, they
represent almost every ground water and surface water community water supply (CWS) system
in the State. In total, the compliance monitoring data in these databases represent over 18,000 of
the approximately 24,000 CWS systems that do not purchase water in the United States; there are
a total of about 54,000 CWS systems in the United States. In addition, a smaller number of non-
transient, non-community water supply (NTNCWS) systems are also represented in the State
compliance monitoring data sets. These data sets contain multiple samples from individual
systems, which can facilitate analysis of the variability in arsenic levels over time, or from
location to location, or point-of-entry to point-of-entry, within individual systems. However,
several of these data sets include samples that are censored1 within, rather than below, the
regulatory range of interest. To manage these multiple reporting levels and to calculate system
mean arsenic levels, regression on order statistics was used in the data analyses presented in
Chapters 5, 6, and 7 of this report.
Other arsenic databases are available for information on arsenic. Some are suitable for
the development of national arsenic occurrence estimates, and others are less suitable for this
purpose. The databases which may be suitable for the development of arsenic occurrence
estimates include the National Arsenic Occurrence Survey (NAOS), the United States Geological
Society (USGS) ambient ground water arsenic databases, the National Inorganics and
Radionuclides Survey (NIRS), and the Metropolitan Water District of Southern California
Survey (Metro). NAOS is based on a representative proportional stratified sampling design. It
includes 517 raw water samples from ground water and surface water systems in the United
States. The analytical method used had a detection level of 0.5 |ig/L, below the regulatory range
of interest (2.0 to 50 |ig/L). Arsenic removal efficiencies associated with the treatment
technologies in place in each system were used to calculate expected finished water arsenic
concentrations. The USGS database is another relevant source of information on arsenic
occurrence. It contains approximately 20,000 ambient ground water samples collected
throughout the United States. These samples were analyzed with a consistent method which has
a detection level of 1 |ig/L, and according to consistent quality control and quality assurance
protocols. Metro contains 144 samples which were primarily collected from ground water and
surface water systems in the United States that serve populations of at least 50,000 people.
These samples have a low detection limit, but are not associated with an individual public water
supply identification number (PWS ID). The NAOS, USGS, NIRS, and Metro databases are used
as comparison tools in Chapter 6 of this report.
1 Censored data are samples with contaminant concentrations reported as less than the analytical detection limit.
Actual contaminant concentrations in these samples may be positive, and may range from zero to the detection limit.
In the case of a naturally occurring contaminant, such as arsenic, contaminant concentrations may be exceedingly
low, but are rarely zero.
ix
-------
Other databases offer data that could be used to estimate arsenic occurrence, but were not
used in this report. The MRS includes samples from approximately 1,000 ground water systems
in the United States. Most of the MRS samples were collected from ground water systems that
serve fewer than 3,300 people, and most of the samples (95 percent) are censored at 5 |ig/L. For
these reasons, the MRS database was not used in the development of arsenic occurrence
estimates presented in this report.
Because several other databases contain very old arsenic sample results and a high
proportion of the results were censored, they were not used in this occurrence analysis. These
databases include the 1969 and 1978 Community Water Supply Surveys (CWSS), the Rural
Water Survey (RWS), and the National Organics Monitoring Survey (NOMS). Because these
results were quite old and highly censored, they were not used in this occurrence estimation.
Another database, the Western Coalition of Arid States (WESCAS) database, also was not used
in this analysis because the data did not necessarily represent arsenic levels at individual PWS,
and because the data conventions used appeared to have been inconsistent from State to State.
Patterns of Arsenic Occurrence in Drinking Water
The data were analyzed with respect to a variety of potential stratification variables,
including source water type, system size and type, and regional stratification. Distributions of
arsenic in ground water and surface water systems were clearly different; therefore, the
occurrence analyses were stratified on the basis of source water type. There were some
differences between arsenic levels for CWS and NTNCWS systems, although there was limited
data for surface water NTNCWS systems. Therefore, the occurrence analyses for ground water
were stratified by system type and the occurrence analyses for surface water were based only on
the CWS data. Other authors, who have evaluated regional differences in arsenic occurrence,
concluded that arsenic levels may differ from region to region. Regional stratification was
applied in these occurrence analyses. State compliance monitoring data sets were stratified into
the 7 regions that were identified by Frey and Edwards (1997). This stratification scheme is
convenient because the State compliance monitoring data can be easily sorted and evaluated by
State. Regional stratification would be unnecessary if data were available for all 50 States.
However, because average arsenic concentrations may differ from region to region, and because
the representation of States within each region differs from region to region, regional
stratification was applied to control these differences and to yield more accurate occurrence
estimates.
Predicted Number of Ground Water and Surface Water Systems Exceeding Potential
Regulatory Levels
Using the State compliance monitoring data, estimates of the proportions and numbers of
systems that may exceed specific maximum contaminant level (MCL) alternatives were
developed. Separate estimates were developed for ground water and surface water systems,
although both estimates were developed through a similar five-step process. In the first step,
system mean concentrations are estimated for each system in the compliance monitoring
database. Second, estimates of exceedance probabilities were developed for each State. Third,
State estimates were grouped and weighted, in order to develop regional arsenic occurrence
-------
estimates. Fourth, regional estimates were weighted to develop national estimates of the
proportion of systems which are likely to have mean system arsenic levels above specific
concentrations of interest. Fifth, estimated exceedance probabilities were multiplied by the total
number of ground water or surface water systems in the United States to estimate the total
number of systems with various system mean arsenic concentrations. Due to the limited amount
of available data, the exceedance probability estimates for surface water CWS systems were used
to estimate the exceedance probabilities for surface water NTNCWS systems. As shown in
Tables ES-1, ES-2, ES-3, and ES-4, these estimates are based on the number of systems in
specific size categories.
Under these estimates, 11,873 ground water CWS systems are estimated to have mean
arsenic levels that exceed 2 |ig/L. The number of systems with mean arsenic levels above the
potential MCL alternatives decreases rapidly as the potential MCL alternative concentration
increases, but 5,252 systems are predicted to have mean arsenic levels greater than 5 |ig/L, and
2,302 systems are predicted to have mean arsenic concentrations greater than 10 |ig/L. Arsenic
occurrence is projected to be significantly lower in surface water systems. For example, 1,052
surface water CWS systems are predicted to have arsenic levels above 2 |ig/L, 325 surface water
CWS systems are predicted to be out of compliance with an arsenic MCL of 5 |ig/L, and 86 are
predicted to exceed an alternative MCL of 10 |ig/L. For NTNCWS ground water systems, 6,306,
3,064, and 1,050 systems are predicted to have mean arsenic concentrations greater than 2, 5,
and 10 |ig/L, respectively. For NTNCWS surface water systems, 80, 25, and 7 systems are
predicted to have mean arsenic concentrations greater than 2, 5, and 10 |ig/L, respectively.
These arsenic estimates resemble other recently generated arsenic occurrence estimates.
At concentrations of 5 and 10 |ig/L, these exceedance estimates are quite similar to estimates
developed by Frey and Edwards (1997). These estimates are also similar to, although slightly
lower than, the number of ground water systems that are estimated to be impacted at
concentrations of 5 and 10 jig/L based on the USGS database. However, the arsenic occurrence
estimates presented in this report are higher than estimates that were developed in 1992 for the
USEPA (Wade Miller, 1992).
An uncertainty analyses was conducted to determine the potential amount of error in the
exceedance probability estimates. To determine 95 percent confidence intervals, it was necessary
to perform a statistical simulation to quantify the potential sources of uncertainty. Three sources
of uncertainty were identified and simulated: 1) sampling variability, within and between
systems; 2) the fill-in of censored observations in the estimation of system means; and 3) fitting
of lognormal distribution to populations of system means within each State.
Intra-system Variability
The purpose of the intra-system analysis is to facilitate prediction of the number of
points-of-entry or POE that will be affected by various MCL alternatives. Compliance with the
arsenic standard is measured at the point-of-entry to the distribution system, and individual
systems can have multiple points-of-entry. Arsenic levels in POE drive compliance costs and
risk reduction benefits more directly than do system mean arsenic levels. Data are not available
XI
-------
Table ES-1
Estimated Arsenic Occurrence in U. S. Ground Water CWS
System Size
(Population Served)
<25
25-100
101-500
501-1,000
1,001-3,300
3,301-10,000
10,001-50,000
50,001-100,000
100,001-1,000,000
> 1,000,000
Total Systems
Lower 95% CI:
Upper 95% CI:
Total
Number of
Systems1
178
14025
14991
4671
5710
2459
1215
131
61
2
43443
Number of Systems with Arsenic Concentrations2 (|J.g/L) of:
>2
49
3833
4097
1277
1561
672
332
36
17
1
11873
11543
13007
>3
35
2788
2980
929
1135
489
242
26
12
0
8636
8363
9501
> 5
22
1696
1812
565
690
297
147
16
7
0
5252
5100
5665
>10
9
743
795
248
303
130
64
7
3
0
2302
2250
2567
>15
5
429
459
143
175
75
37
4
2
0
1329
1269
1499
>20
4
281
300
93
114
49
24
3
1
0
869
821
995
>25
3
199
213
66
81
35
17
2
1
0
617
573
712
>30
2
147
157
49
60
26
13
1
1
0
456
421
534
>40
1
90
96
30
37
16
8
1
0
0
278
252
335
>50
1
60
64
20
25
11
5
1
0
0
187
165
226
Notes:
CI: confidence interval
1 Based on 1998 Baseline SDWIS data for purchased and non-purchased systems. Systems characterized as GW under the influence of SW are considered to
be surface water systems.
2 Based on national weighted point estimates presented in Table 6-3a.
3 Totals may not add up due to rounding of the number of systems to the nearest whole number.
Xll
-------
Table ES-2
Estimated Arsenic Occurrence in U. S. Surface Water CWS
System Size
(Population Served)
<25
25-100
101-500
501-1,000
1,001-3,300
3,301-10,000
10,001-50,000
50,001-100,000
100,001-1,000,000
> 1,000,000
Total Systems
Lower 95% CI:
Upper 95% CI:
Total Number
of Systems1
74
1001
1983
1219
2420
1844
1606
300
261
13
10721
Number of Systems with Arsenic Concentrations2 (|J.g/L) of:
>2
7
98
195
120
238
181
158
29
26
1
1052
973
2730
>3
4
56
110
68
135
103
89
17
15
1
597
514
2212
> 5
2
30
60
37
73
56
49
9
8
0
325
193
1036
>10
1
8
16
10
19
15
13
2
2
0
86
56
167
>15
0
5
9
6
11
9
7
1
1
0
50
25
107
>20
0
3
6
4
8
6
5
1
1
0
34
14
88
>25
0
2
5
o
5
6
4
4
1
1
0
26
9
77
>30
0
2
4
2
5
3
3
1
0
0
20
6
71
>40
0
1
o
3
2
3
2
2
0
0
0
14
3
65
>50
0
1
2
1
2
2
2
0
0
0
10
2
63
Notes:
CI: confidence interval
1 Based on 1998 Baseline SDWIS data for purchased and non-purchased systems. Systems characterized as GW under the influence of SW are considered to
be surface water systems.
2 Based on national weighted point estimates presented in Table 6-3a.
3 Totals may not add up due to rounding of the number of systems to the nearest whole number.
Xlll
-------
Table ES-3
Estimated Arsenic Occurrence in U. S. Ground Water NTNCWS
System Size (Population
Served)
<25
25-100
101-500
501-1,000
1,001-3,300
3,301-10,000
10,001-50,000
50,001-100,000
100,001-1,000,000
> 1,000,000
Total Systems
Total
Number of
Systems1
31
9732
7103
1996
696
62
15
0
0
0
19635
Number of Systems with Arsenic Concentrations2 (|J.g/L) of:
>2
10
3126
2281
641
224
20
5
0
0
0
6306
>3
8
2358
1721
484
169
15
4
0
0
0
4758
>5
5
1519
1109
312
109
10
2
0
0
0
3064
>10
2
520
380
107
37
3
1
0
0
0
1050
>15
1
306
223
63
22
2
0
0
0
0
617
>20
1
203
148
42
14
1
0
0
0
0
409
>25
0
145
105
30
10
1
0
0
0
0
292
>30
0
108
79
22
8
1
0
0
0
0
219
>40
0
67
49
14
5
0
0
0
0
0
136
>50
0
46
34
9
3
0
0
0
0
0
93
Notes:
1 Based on 1998 Baseline SDWIS data for purchased and non-purchased systems. Systems characterized as GW under the influence of SW are considered to
be surface water systems.
2 Based on national weighted point estimates presented in Table 6-3b.
3 Totals may not add up due to rounding of the number of the systems to the nearest whole number.
xiv
-------
Table ES-4
Estimated Arsenic Occurrence in U. S. Surface Water NTNCWS
System Size (Population
Served)
<25
25-100
101-500
501-1,000
1,001-3,300
3,301-10,000
10,001-50,000
50,001-100,000
100,001-1,000,000
> 1,000,000
Total Systems
Lower 95% CI:
Upper 95% CI:
Total
Number of
Systems1
5
280
314
107
80
23
5
1
1
0
816
Number of Systems with Arsenic Concentrations2 (|J.g/L) of:
>2
0
27
31
11
8
2
0
0
0
0
80
74
208
>3
0
16
17
6
4
1
0
0
0
0
45
39
168
>5
0
9
10
o
3
2
1
0
0
0
0
25
15
79
>10
0
2
3
1
1
0
0
0
0
0
7
4
13
>15
0
1
1
0
0
0
0
0
0
0
4
2
8
>20
0
1
1
0
0
0
0
0
0
0
3
1
7
>25
0
1
1
0
0
0
0
0
0
0
2
1
6
>30
0
1
1
0
0
0
0
0
0
0
2
0
5
>40
0
0
0
0
0
0
0
0
0
0
1
0
5
>50
0
0
0
0
0
0
0
0
0
0
1
0
5
Notes:
1 Based on 1998 Baseline SDWIS data for purchased and non-purchased systems. Systems characterized as GW under the influence of SW are considered to
be surface water systems.
2 Based on national weighted point estimates presented in Table 6-3b. These estimates were derived from CWS SW data.
3 Totals may not add up due to rounding of the number of systems to the nearest whole number.
xv
-------
that would allow the development of directly representative estimates of arsenic occurrence.
Instead, this report uses compliance monitoring data with POE identifiers to quantify a
relationship between POE means and system means, so that the number of POE means in the
United States that are likely to exceed specific regulatory alternatives can be calculated from a
distribution of system means. This relationship was summarized by the estimated coefficient of
variation (CV), or relative standard deviation, for the intra-system variability. The approach used
was based on fitting a statistical model to account both for the intra-system variability between
different POE and for temporal and analytic variability between multiple measurements at the
same POE. The estimated intra-system variability CVs were 37 % for ground water CWS
systems, 53 % for surface water CWS and NTNCWS systems, and 25 % for ground water
NTNCWS systems. The CV values that were calculated under these analyses were used in a
regulatory impact analyses (RIA) conducted under a separate work assignment to estimate the
number of POE that may exceed regulatory alternatives.
XVI
-------
1. Introduction
Under the Safe Drinking Water Act (SOWA), 42 U.S.C. §§300f-300j, originally enacted
in 1974, arsenic is a regulated drinking water contaminant. In 1975, U.S. Environmental
Protection Agency (USEPA) issued a National Interim Primary Drinking Water Regulation
(NIPDWR) for arsenic at 50 |_ig/L (USEPA, 1975). This value is the current Maximum
Contaminant Level (MCL) for arsenic in drinking water. In 1986, Congress converted the
NIPDWR for arsenic to a National Primary Drinking Water Regulation (NPDWR), and directed
USEPA to revise National Primary Drinking Water Regulations. In 1994, following a consent
decree in a suit2 between USEPA and Citizens Concerned about Bull Run, Inc., a citizens' group,
USEPA organized an internal workgroup for the purpose of addressing risk assessment,
treatment, analytical methods, arsenic occurrence, exposure, costs, implementation issues, and
regulatory options for arsenic. In 1995, USEPA deferred the proposal of the regulation in order
to better characterize the human health effects associated with chronic low-level exposure to
arsenic and treatment costs. In accordance with Safe Drinking Water Act (SOWA), as amended
in 1996, Section 1412(b)(12)(A) directs USEPA to propose a National Primary Drinking Water
Regulation for arsenic by January 1, 2000, and to issue a final regulation by January 1, 2001.
Arsenic (As) is a metallic element that occurs at low concentrations in most rocks and
soils (Yan-Chu, 1994). To a small extent, it occurs in the elemental state; however, higher
concentrations of arsenic principally occur in mineral complexes with metals and other elements
(Welch et al, 1988). For example, arsenic is a common impurity in the sulfide ores of lead,
copper, and zinc. Arsenic is released into the environment from natural processes such as the
weathering and dissolution of arsenic-containing minerals and ores (Yan-Chu, 1994). In addition
to its release from natural sources, arsenic is released from a variety of anthropogenic sources
(USEPA, 1998b), including:
Manufacturing of metals and alloys
Petroleum refining
Pharmaceutical manufacturing
Pesticide manufacturing and application
Chemicals manufacturing
Burning of fossil fuels
Waste incineration
These anthropogenic releases of arsenic can elevate environmental arsenic concentrations.
Human exposure to arsenic can result in a variety of chronic and acute effects. In
particular, there is evidence that associates chronic arsenic ingestion at low concentrations with
increased risk of skin cancer, and that arsenic may cause cancers of the lung, liver, bladder,
kidney, and colon (ATSDR, 1998). Because of the human health risks associated with arsenic,
USEPA regulates the level of arsenic in drinking water.
2 Miller v. USEPA, No. 89-CV-6328 (D. Ore., filed August 31, 1989).
1
-------
The objective of this study for the USEPA's Office of Ground Water and Drinking Water
(OGWDW) is to estimate arsenic occurrence in public water supplies (PWS) in relation to
various MCL alternatives. Estimates of the number of people exposed to various concentrations
of arsenic in drinking water will be presented separately, and are not included in this report. The
arsenic occurrence estimates are also bounded by 95 percent confidence intervals. These
estimates will be significant in the development of the proposed arsenic regulation.
1.1 Purpose of this Document
This report summarizes the results of the arsenic occurrence analysis conducted by ISSI
Consulting Group and The Cadmus Group, with their subcontractor, ICF Consulting for the
USEPA's OGWDW. The estimates of arsenic occurrence presented in this report differ from
those presented in other studies because of their strong reliance on existing compliance
monitoring data that were voluntarily provided to USEPA and its contractors. In addition, new
techniques have been applied to evaluate the statistical distributions of arsenic in drinking water
to estimate percentages of regulatory exceedances, to estimate the variability in arsenic levels
within systems, and to estimate the relative uncertainty associated with these predictions.
1.2 Organization of the Document
This report is organized in seven sections that are relevant to the estimation of arsenic
occurrence in drinking water in the United States. The remaining sections of this document are
organized as follows:
Chapter 2: Sources of Arsenic identifies naturally occurring and anthropogenic sources of
arsenic in the environment, with a particular focus on sources of arsenic to
drinking water.
Chapter 3: Fate and Transport of Arsenic presents information on the physical and chemical
characteristics of arsenic and the relation between those properties and the
presence of arsenic in source waters. In addition, this section presents an
overview of the potential fate and transport of arsenic within treatment and
distribution systems.
Chapter 4: Sources of Data on Arsenic Occurrence in Drinking Water Supplies presents a
summary of the approaches used to identify and select data on arsenic occurrence
for use in this occurrence assessment. In addition, this section presents summary
information on the data sources that were used in the occurrence assessment.
Chapter 5: Arsenic Occurrence Patterns in the United States discusses the analyses that were
applied to identify patterns in the data and the conclusions developed as a result of
those analyses.
-------
Chapter 6: National Occurrence Estimates presents estimates of the number of systems
projected to exceed specific arsenic levels, describes the method used to develop
these estimates, and discusses the uncertainty in these estimates.
Chapter 7: Intra-system Variability Assessment provides an overview of the variations in
arsenic levels from location to location within public water supply systems.
Chapter 8: Temporal Variability Analysis examines the variability of arsenic concentrations
over time in a source.
-------
This page intentionally left blank
-------
2. Sources of Arsenic
This section discusses the physical and chemical properties of arsenic, and the natural and
anthropogenic sources of arsenic to the environment, particularly sources that may affect
drinking water in the United States. The primary natural sources include the earth's crust, soil
and sediment, geothermal activity, and volcanic activity. The most significant anthropogenic
sources are agricultural, industrial, and mining activities.
2.1 Physical and Chemical Properties of Arsenic
Arsenic (As) is a silver-gray brittle crystalline solid (Budavari, et al. 1989). It also exists
in black and yellow amorphous forms. Arsenic appears in Group 15 on the periodic table in the
first long period with a d-shell just below the valence shell. Arsenic has an atomic weight of
74.9216 and an atomic number of 33. Silver-gray arsenic has a specific gravity of 5.73; a melting
point of 817 °C (28 atm) and sublimes at 613°C. The yellow amorphous form of arsenic has a
specific gravity of 1.97. Elemental arsenic can be present as a metalloid, although arsenic has an
elemental structure similar to non-metals. In the vapor state, arsenic occurs as a tetrameric
molecule (As4). In high oxidation states arsenic displays covalent tendencies, while in low
oxidation states it shows ionic tendencies (Ferguson, 1990). The physical and chemical
properties of arsenic are summarized in Table 2-1.
The valence states of As are: -3, 0, +1, +3, and +5 (Welch et al., 1988). Elemental
arsenic (valence 0) is rarely found under natural conditions. The +3 and +5 states are found in a
variety of minerals and in natural waters. Many of the chemical behaviors of arsenic are linked
to the ease of conversion between +3 and +5 valence states (National Research Council (NRC),
1999). The valence state affects the toxicity of arsenic compounds. While arsine (-3) is the most
toxic, the following are successively less toxic: organo-arsines, arsenites (+3), arsenates (+5),
arsonium metals (+1), and elemental arsenic (0).
2.1.1 Environmentally Relevant Arsenic Species
Arsenic occurs naturally as a constituent of a number of different compounds in both
marine and terrestrial environments. Arsenic species are classified as either organic or inorganic.
If carbon is present within the compound it is considered to be an organic arsenic species. Table
2-2 includes a summary of both organic and inorganic species of arsenic which may be found in
food and water.
Inorganic Arsenic
Inorganic arsenic, with +5 (arsenate) and +3 (arsenite) oxidation states, is more prevalent
in water than organic arsenic (Irgolic, 1994; Clifford and Zhang, 1994). The dominant arsenic
species depends on pH and redox conditions. In general +5 predominates under oxidizing
conditions and +3 predominates under reducing conditions (ATSDR, 1998; Clifford and Zhang,
1994).
-------
Table 2-1
Physical and Chemical Properties of Arsenic
CAS Number
Atomic Number
Atomic Weight
Melting Point at 28 atm
Boiling Point
Critical Temperature
Heat of Vaporization
Critical Pressure
Density (at 14°C)
Most Stable Isotope
Covalent Radius
Atomic Radius
Ionic Radius
Vapor Pressure
7440-38-2
33
74.92
817°C
613°C
1,400°C
11.2 kcal/g-atom
22.3 MPa
5.727g/cm3
75 As
1.19 angstroms
1.39 angstroms
2.22 angstroms
1 mm (375°C)
10mm(437°C)
100mm(518°C)
Excerpted from Budavarie/ ai, 1989
-------
Table 2-2
Inorganic and Organic Arsenic Compounds
Name
Arsanilic acid
Arsenous acid
Arsenic acid
Monomethylarsonic acid
Methylarsonous acid
Dimethylarsinic acid
Dimethylarsinous acid
Roxarsone
Trimethylarsine
Trimethylarsine oxide
Tetramethlarsonium ion
Arsenocholine
Arsenobetaine
Arsenic-containing ribo-sides
Abbreviation
As(ni)
As(V)
MMAA
MMAA(HI)
DMAA
DMAA(IH)
TMA
TMAO
Me4As+
AsC
AsB
Arsenosugar X-XVa
Arsenolipidb
Chemical Formula
C6H8AsNO3
H3AsO3
H3AsO4
CH3AsO(OH)2
CH3As(OH)2[CH3AsO]n
(CH3)2AsO(OH)
(CH3)2AsOH [((CH3)2As)2O]
C6H6AsN06
(CH3)3As
(CH3)3AsO
(CH3)4As +
(CH3)3As+CH2CH2OH
(CH3)3As+CH2COQ-
Excerpted from NRC, 1999
-------
Examples of inorganic arsenic compounds found in the environment include oxides (i.e.
As2O3, As2O5, R3AsO)n, R^sOtOH)^ (n=l,2)) and sulfides (As2S3, AsS, HAsS2, HAsS/')
(Cullen and Reimer, 1989). Inorganic arsenic species which are stable in oxygenated waters
include arsenic acid (As(V)) species (i.e. H3AsO4, H2AsO4", HAsO42" and AsO43"). Arsenous acid
(As(ni)) is also stable as H3AsO3 and H2AsO3" under slightly reducing aqueous conditions.
In addition to geochemical factors, microbial agents can influence the oxidation state of
arsenic in water, and can mediate the methylation of inorganic arsenic to form organic arsenic
compounds. Microorganisms can oxidize arsenite to arsenate, reduce arsenate to arsenite, or
reduce arsenate to arsine (Cullen and Reimer, 1989). Bacterial action also oxidizes minerals
such as orpiment (As2S3), arsenopyrite (FeAsS), and enargite (Cu3AsS4) releasing arsenate.
Under aerobic conditions, the common aquatic bacterium Pseudomonasfluorescens reduces
arsenate to arsenite. In a river in New Zealand, investigators found the predominant oxidation
state of arsenic varied seasonally because of (at least in part) the bacterium Anabaena
oscillaroides which reduces arsenate to arsenite. Arsenite was found to predominate in spring
and summer months, while arsenate was prevalent at other times of the year.
Organic Arsenic
Organic arsenic compounds such as Monomethylarsonic acid (MMAA), Dimethylarsinic
acid (DMAA), Trimethylarsine (TMA), and Trimethylarsine oxide (TMAO) are generally
associated with terrestrial settings, however, some are found in water (NRC, 1999). Organic
arsenic is produced naturally in the environment in natural gas (ethylmethylarsines), shale oil, in
water when microorganisms metabolize inorganic arsenic, and in the human body, as a result of
enzyme activity in the liver (USEPA, 1993; Berger and Fairlamb, 1994).
In studies of arsenic speciation in natural waters reviewed by the National Research
Council (1999), organic arsenical compounds were reported to have been detected in surface
water more often than in ground water. Surface water samples reportedly contain low but
detectable concentrations of arsenic species such as MMAA, DMAA, Arsenocholine (AsC),
TMAs, and species similar to Arsenobetaine (AsB). Methylarsenicals have been reported to
comprise as much as 59% of total arsenic in lake water. In some lakes, DMAA has been reported
as the dominant species, and concentrations appear to vary seasonally as a result of biological
activity within waters, with the highest concentrations observed in May and June.
2.2 Natural Sources of Arsenic
Arsenic occurs in the environment in rocks, soil, water, air, and in biota; and
concentrations of arsenic in a variety of environmental media are presented in Table 2-3. The
following sections discuss important natural sources of arsenic in the environment. Most arsenic
in the environment exists in rock or soil (ATSDR, 1998). Because arsenic occurs naturally in
rock, soil and sediment, these sources are particularly important determinants of regional levels
of arsenic in ground water and surface water. Natural sources of arsenic are discussed below,
' R=H, Me, Cl, etc.
-------
Table 2-3
Arsenic Concentrations in Environmental Media
Environmental Media
Air
Rain from unpolluted ocean air
Rain from terrestrial air
Rivers
Lakes
Ground water
Sea water
Soil
Stream/river sediment
Lake sediment
Igneous rock
Metamorphic rock
Sedimentary rock
Biota - Green Algae
Biota - Brown Algae
Arsenic Concentration
Range
1.5-53
0.019
0.46
0.20-264
0.38-1,000
-< 1.0- >- 1,000
0.15-6.0
0.1-1,000
5.0-4,000
2.0-300
0.3-113
0.0-143
0.1-490
0.5-5.0
30
Units
ng/m"3
Hg/L
Hg/L
Hg/L
Hg/L
Hg/L
Hg/L
mg/kg
mg/kg
mg/kg
mg/kg
mg/kg
mg/kg
mg/kg
mg/kg
Excerpted from NAS, 1977.
-------
and anthropogenic sources are discussed subsequently in Section 2.3, and the cycling of arsenic
in the environment is depicted in Figure 2-1.
Man-made Arsenic Sources
^?v
Fertilizers and
Pesticides
^N^N ^N^N
Magma Arsenic
(Water-Soluble)
Ground Water
Natural Arsenic
Unavailable Arsenic Coal
Insoluble Salts Oil
Surface-Adsorbed Minerals
Organically Bound
FIGURE 2-1. Environmental Transfer of Arsenic.
Earth's Crust
Arsenic is the twentieth most abundant element in the earth's crust (ATSDR, 1998; NAS,
1977). Concentrations of arsenic in the earth's crust vary, but average concentrations are
generally reported to range from 1.5 to 5 mg/kg (ATSDR, 1998; Cullen and Reimer, 1989; NAS,
1977). Arsenic is a major constituent of many mineral species in igneous and sedimentary rocks;
Table 2-4 presents concentrations of arsenic in igneous and sedimentary rocks. Among igneous
rock types, the highest arsenic concentrations are found in basalts. Sedimentary rocks,
particularly iron and manganese ores, often contain higher average arsenic concentrations than
igneous rocks (Welch et al, 1988).
Table 2-5, above, lists common minerals that contain arsenic. Arsenopyrite (FeAsS),
realgar (AsS), and orpiment (As2S3) are the most important of these minerals, and they are
commonly present in the sulfide ores of other metals including copper, lead, silver and gold
(Yan-Chu, 1994). Arsenic may be released from these ores to soil (Yan-Chu, 1994), surface
water (Mok and Wai, 1989), ground water (Welch et al., 1988), and the atmosphere (ATSDR,
1998).
10
-------
Table 2-4
Arsenic in Igneous and Sedimentary Rocks
Rocks
Igneous Rocks:
Ultrabasic
Basalts, gabbros
Andesites, dacites
Granitic
Silicic volcanic
Sedimentary Rocks:
Limestones
Sandstones
Shales and clays
Phosphorites
Sedimentary iron ores
Sedimentary manganese ores
Coal
No.
Analyses
37
146
41
73
52
37
11
324
282
110
1,150
Arsenic Concentration
Range Usually
Reported
0.3-16
0.06-113
0.5-5.8
0.2-13.8
0.2-12.2
0.1-20
0.6-120
0.3-490
0.4-188
1-2,900
(up to 1.5%)
0-2,000
(mg/kg)
Average
3.0
2.0
2.0
1.5
3.0
1.7
2.0
14.53
22.6
400
4-13
Adapted from NAS, 1977
"Excludes one sample containing As at a concentration of 490 mg/kg.
Table 2-5
Common Minerals of Arsenic
Arsenopyrite, FeAsS
Lollingite, FeAs2
Orpiment, As2S3
Realgar, As4S4
Chloanthite, NiAs2
Niciolite, NiAs
Smalite, CoAs2
Cobaltite, CoAsS
Gersdorffite, NiAsS
Tennantite, 4Cu2SAs2SS3
Proustite, 3Ag2SAs2S3
Enargite, 3Cu2SAs2S5
Adapted from Ferguson, 1990.
11
-------
In their evaluation of the regional distribution of arsenic in ground water in the Western
United States, Welch et al, (1988) evaluated the association between aquifer geology and arsenic
concentrations in ground water. Higher arsenic concentrations (ground water concentrations
greater than 50 |ig/L) were associated with sedimentary deposits derived from volcanic rocks.
These geological conditions occurred at a few locations in the Western Mountain Ranges
(notably near Reno, Nevada, and Eugene, Oregon), and in the Alluvial Basins. Within the
Alluvial Basins, elevated arsenic concentrations were associated with sediments derived from
volcanic rocks rather than non-sedimentary and unmineralized volcanic rock. Weathering of the
volcanic rocks may result in the concentration of arsenic onto ferric oxyhydroxide that are
deposited with sediments (Welch etal, 1988). Lower arsenic concentrations are associated with
regions underlain by carbonate rocks and volcanic basalts. The regions with moderate arsenic
levels include parts of the Alluvial Basins in western Utah and eastern Nevada and the Columbia
Lava Plateau.
In other parts of the United States, high (greater than 50 |ig/L in ground water)
concentrations of arsenic are recognized to be associated with other geological formations. In
Eastern Michigan and Northeastern Wisconsin, high arsenic concentrations are associated with
sulfide mineral deposits in sedimentary rocks (Westjohn et al., 1998; Simo et al., 1996), and in
the upper Midwest, higher arsenic concentrations are associated with iron oxide rich sedimentary
deposits. Moderate to high arsenic concentrations (10 to 50 |ig/L in ground water) in portions of
the Northeastern United States (Massachusetts to Maine) appear to be related to sulfide minerals
in the bedrock aquifers (Marvinney et al., 1994).
Soil and Sediment
Arsenic concentrations in soils depend in part on the parent materials from which the
soils were derived, although they may be enriched by other sources, including anthropogenic
sources. Typical natural concentration ranges are 0.1 to 40 mg/kg, with an average concentration
of 5-6 mg/kg (NAS, 1977). The level of arsenic in soil derived from basalts tends to be higher
than in soils of granitic origin, and concentrations of 20 to 30 mg/kg may be found in soils
derived from sedimentary rocks (Yan-Chu, 1994). In areas of recent volcanism, soils average
arsenic concentrations are approximately 20 mg/kg. Very high natural concentrations of arsenic
(up to 8,000 mg/kg) may occur in soils that overlay deposits of sulfuric ores (NAS, 1977).
Arsenic can be found in soil in the inorganic state bound to cations, and it can also be found
bound to organic matter. Arsenic may be transferred to surface water and ground water through
erosion and dissolution; plants may also uptake arsenic. Because arsenic can be fixed in
inorganic and organic compounds in soil, soil may also be a sink for arsenic.
In bottom sediments of rivers and lakes, concentrations of arsenic in surface sediments
tend to exceed those found in deeper sediments due to diagenetic cycling (Nimick et al., 1998).
Recent anthropogenic arsenic releases may also result in the elevation of arsenic concentrations
in surface sediments. This phenomenon has been observed in Lake Michigan (NAS, 1977). In
the Madison and Missouri River Basins, Nimick et al., documented average arsenic
concentrations in bottom sediments that ranged from 7 to 102 mg/kg. These concentrations
differed substantially over the length of the river system. In Lake Michigan, average
concentrations of arsenic in surficial sediment were reported to be 12.4 mg/kg (NAS, 1977). In
12
-------
sediments contaminated by mining and industrial activities, arsenic concentrations can be greatly
enriched. In creeks affected by mining activities in Idaho, Mok and Wai (1989) found arsenic
concentrations that ranged from 42.1 to 2550.4 mg/kg. In shallow waters, wave action and
seasonal high flow scouring can result in resuspension of arsenic rich surface sediments, whereas
in deeper lakes, arsenic may be permanently sequestered in sediment (Nimick et al, 1998).
Arsenic may also be released from bottom sediments as a result of microbial action.
Arsenic in soil and sediment may undergo microbial degradation or transformation. In
soil, arsenic in the form of arsenates, arsenites, monomethyl-arsionic acid (MMAA) or
dimethylarsinic acid (DMAA) may be biotransformed to arsine gases (Yan-Chu, 1994). These
arsine gases are subsequently volatilized to the environment. In sediment, biologically mediated
methylation of arsenates increases the solubility of arsenic, and may increase arsenic
concentrations in water (Mok and Wai, 1994). Conversely, the biologically mediated
demethylation of DMAA and MMAA can result in the formation of arsenates, which are
strongly adsorbed onto sediments.
Geothermal Waters
Geothermal water can be sources of arsenic in surface water and ground water. Welch et
al., (1988) identified 14 areas in the Western United States where arsenic conditions in water
exceed 50 jig/L because of known or suspected geothermal sources. In these areas, dissolved
arsenic concentrations ranged from 80 to 15,000 |ig/L. Welch et al.,found that mean dissolved
arsenic concentrations in geothermal ground waters are higher than mean arsenic concentrations
in non-thermal ground waters in any of the physiographic provinces in the United States. Flow
of arsenic-enriched geothermal water from hot springs may result in high concentrations of
arsenic in surface water systems. In Yellowstone National Park, the arsenic concentrations in
geysers and hot springs range from 900 to 3,560 |ig/L (Stauffer and Thompson, 1984). Waters
from these sources cause elevated arsenic levels in the Madison and Missouri Rivers far
downstream of the park boundaries (Nimick et al, 1998). As a result, cities that use water from
portions of the Missouri River for municipal supply must treat it to reduce arsenic
concentrations. Geothermal sources of arsenic are primarily located in the Western United
States.
Other Sources
Natural emissions of arsenic associated with volcanic activity and forest and grass fires
are recognized to be significant. Indeed, volcanic activity appears to be the largest natural source
of arsenic emissions to the atmosphere (ATSDR, 1998). Estimates of natural releases (of which
volcanic arsenic emissions are the primary source) show significant range, and are summarized in
Table 2-6.
13
-------
Table 2-6
Estimated Natural Average Arsenic Releases to the Atmosphere
Study
Tamaki and Frankenburger, 1992
Pacynaetal., 1995
Loebenstein, 1994
Estimated annual natural releases (metric tons)*
44,100
1,100-23,500
2,800 - 8,000
While at least one study suggests that natural arsenic emissions slightly exceed industrial
emissions (Tamaki and Frankenburger, 1992), other studies suggest that industrial emissions of
arsenic are significantly greater than natural emissions (Pacyna et al, 1995; Loebenstein 1994).
Thus, the relative contributions of volcanic sources, other natural sources, and anthropogenic
sources to the atmosphere have not been definitively established.
In summary, the primary natural sources of arsenic in the United States include arsenic in
geological formations (rock as well as soil and sedimentary deposits) and arsenic associated with
geothermal waters. In geological formations, higher arsenic concentrations tend to be associated
with sedimentary rocks derived from acidic to intermediate volcanics, which occur primarily in
parts of the Western United States, and from sulfide minerals and iron and manganese oxides
associated with sedimentary rocks, which occur in the parts of the Northeastern and Upper
Midwestern United States. Geothermal activity may affect arsenic levels in ground water and
surface water in some regions of the United States, particularly in the Western United States. In
addition to these recurrent sources, volcanos and forest fires may also result in the sporadic
release of large amounts arsenic to the environment.
2.3 Anthropogenic Sources of Arsenic
From man-made sources, arsenic is released to terrestrial and aquatic environments and to
the atmosphere. The anthropogenic impact on arsenic levels in these media depends on the level
of human activity, the distance from the pollution sources, and the dispersion and fate of the
arsenic that is released. This section discusses the major current and past anthropogenic sources
of arsenic, which are wood preservatives, agricultural uses, industry, and mining and smelting. It
also provides an overview of other sources of arsenic in the environment. Table 2-7 provides an
overview of the use of arsenic in specific economic sectors. It is important to note that some of
these uses are banned in the United States. After these sources are discussed, an overview of
anthropogenic arsenic releases to the environment, based on Toxics Release Inventory (TRI)
data, is provided.
14
-------
Table 2-7
Summary of Current and Past Uses of Arsenic
Sector
Lumber
Agriculture
Livestock
Medicine
Industry
Uses
Wood preservatives
Pesticides, insecticides, defoliants, debarking agents, soil sterilant
Feed additives, disease preventatives, animal dips, algaecides
Antisyphilitic drugs, treatment of trypanosomiasis, amebiasis, sleeping sickness
Glassware, electrophotography, catalysts, pyrotechnics, antifouling paints, dye and soaps, ceramics,
pharmaceutical substances, alloys (automotive solder and radiators), battery plates, solar cells,
optoelectronic devices, semiconductor applications, light emitting diodes in digital watches
Source: Azcue andNriagu, 1994.
Wood Preservatives
About 90% of the arsenic that is consumed in the United States on an annual basis is used
to manufacture the wood preservative chromated copper arsenate (CCA) (Reese, 1999; Reese,
1998). CCA is an inorganic arsenic compound and consists of arsenic, chromium and copper.
There are a number of CCA type products in the market, and the precise molecular composition
may vary from formulation to formulation. Different arsenic compounds are used as active
ingredients in CCA, including arsenic acid (H3AsO4), arsenic pentoxide (As2O5), and sodium
arsenate (Na2HAsO4). These arsenical ingredients in CCA are primarily produced from arsenic
trioxide (As2O3), although arsenic trioxide itself is not used as a wood preservative in CCA
treatment solutions. EPA classifies CCA as a restricted use pesticide (USEPA, 1997a, USEPA,
1984). CCA is used to pressure treat lumber, which is typically used for construction of decks,
fences, and other outdoor applications. As a result, CCA consumption is related to the
performance of the housing industry. Currently, there are three primary manufacturers of CCA,
and these firms are located in the Southeastern United States (Reese, 1998).
The current releases of arsenic to the environment from the manufacture and use of CCA
are included in the Toxics Release Inventory, which is discussed below. Earlier releases of
arsenic to the environment at some wood preserving sites have resulted in localized
contamination of environmental media, and several wood preserving sites are included on the
NPL list. In addition, there is some evidence that some arsenic may be released to soil from
CCA treated lumber (USEPA, 1997a). EPA is evaluating the release of arsenic from treated
wood. However, no data are currently available that indicates releases of arsenic from treated
lumber could affect drinking water quality.
15
-------
Agricultural Uses
Past and current agricultural uses of arsenic and arsenic compounds include pesticides,
herbicides, insecticides, defoliants, and soil sterilants (Azcue andNriagu, 1994). Arsenic
compounds were also used in animal dips and are currently used in raising livestock as feed
additives and for disease prevention (Azcue and Nriagu, 1994). These uses of arsenic
compounds are discussed below.
Arsenic is a constituent of organic agricultural pesticides that are currently used in the
United States. The most widely applied organoarsenical pesticide is monosodium
methanearsonate (MSMA), which is used to control broadleaf weeds (Jordan et al., 1997).
MSMA was the twenty-second most commonly applied conventional pesticide in the United
States in 1995 (USEPA, 1997b). In 1995, approximately 4 to 8 million pounds of MSMA were
applied. It is primarily applied to cotton; small amounts of disodium methanearsonate (DSMA)
and cacodylic acid are also applied to cotton fields as herbicides. MSMA and DSMA have also
been used for the postemergence control of crabgrass, Dallisgrass, and other weeds in turf (NAS,
1977), and cacodylic acid is used for weed control and monocotyledonous weeds.
Organoarsenicals in soil are metabolized to alkylarsines and arsenate by soil bacteria (USEPA,
1998b). No published data are currently available on the leaching or transport of MSMA or
DSMA from soil.
Inorganic arsenic was also a constituent of a variety of pesticides, herbicides, and
fungicides. The last agricultural use of inorganic arsenic, which was the use of arsenic acid
pesticide on cotton, was voluntarily canceled in 1993 (USEPA, 1998a, 58 FR 64579). Other
inorganic arsenic-containing pesticides included sodium arsenate, calcium arsenate, copper
acetoarsenite (Paris Green), copper arsenate, and magnesium arsenate. These were used to
control potato beetles, boll weevils, grasshoppers, moths, bud worms, and other insects in the
United States (Thompson, 1973). Inorganic arsenic compounds (arsenic acid, arsenic trioxide,
and sodium arsenate) are currently only used in sealed ant bait and wood preservatives (USEPA,
1995). Formerly registered uses of inorganic arsenic under the Federal Insecticides, Fungicides,
and Rodenticides Act (FIFRA) were pesticides, insecticides, rodenticides, cotton dessicants, anti-
fouling paints for boat hulls, soil sterilants, and herbicides for crab grass and other weeds4.
These uses were voluntarily canceled by the product manufacturers by 1993. Sodium arsenite
was used as cattle and sheep dips (Azcue and Nriagu, 1994), but its is now banned in the United
States (USEPA, 1999b). Other inorganic arsenic pesticides that are banned in the United States
include calcium arsenate, copper arsenate, and lead arsenate. The use of arsenic trioxide and
sodium arsenate is severely restricted.
Arsenic-based inorganic pesticides were applied to various agricultural crop lands, and
the historical application of these pesticides has contaminated soil with arsenic residues. The use
of lead arsenate as an insecticide for worms and moths resulted in contamination of soils in apple
orchards (Davenport and Peryea 1991; Maclean and Langille, 1981; Hess and Blanchar, 1977).
4 Based on queries of the EPA Office of Pesticide Programs Pesticide Product Information System databases
conducted by ISSI Consulting Group in July, 1999. This database is searchable online at
http ://www. cdpr. ca.gov/docs/epa/epamenu. htm.
16
-------
Steevens, et al, (1972) showed that the use of sodium arsenate as a potato defoliant raised total
arsenic concentrations in Wisconsin potato field soils. The leaching of arsenic from agricultural
topsoil to subsoil has been reported (Peryea, 1991; NAS, 1977). Leaching is more likely to occur
in sandy soils than in organic soils, and sodium or calcium arsenates tend to leach faster than
aluminum or iron arsenates (NAS, 1977). Fertilization of fields with phosphates may increase
the rate of leaching of arsenic from soils (Davenport and Peryea, 1991; Woolson et al., 1973),
and can increase arsenic concentrations in the subsoil and shallow ground water (Peryea and
Kammereck, 1997; Davenport and Peryea, 1991; Peryea, 1991). This historical pesticide use is
not expected to have resulted in widespread ground water contamination. Peters et al., (1999)
found that arsenic levels in drinking water did not correlate with agricultural activities in the
State of New Hampshire. Fuhrer et al., (1996), however, reported that filtered water arsenic
concentrations in the Yakima River Basin were higher in agriculturally affected areas than in
other areas. These authors suggested high arsenic concentrations in orchard soils from past lead-
arsenate applications were the likely source of the arsenic in surface water.
Organic arsenic (e.g., roxarsone and arsanilic acid) is a constituent of feed additives for
poultry and swine for increased rate of weight gain, improved feed efficiencies, improved
pigmentation, and disease treatment and prevention (e.g., swine dysentary, chronic respiratory
disease, specific infections) (21 CFR 558.530; 21 CFR 558.62). These additives undergo little or
no degradation before excretion (NAS, 1977; Moody and Williams, 1964; Aschbacher and Feil,
1991). Arsenic concentrations in animal wastes are reported to range from 4 to 40 mg/kg (Isaac
et al, 1978). Application of arsenic containing wastes as fertilizer to field plots is reported to
elevate arsenic concentrations in soil water. Two other studies, however, reported that the
fertilization of crops with poultry litter did not change the arsenic content of soils or crops (Smith
et al., 1992; Morrison, 1975). Therefore, potential effect of arsenic from animal wastes upon
ground water and surface water is not well understood.
Industrial Uses and Releases
Arsenic and arsenic compounds are used in a variety of industrial applications. Arsenic
metal is used in the production of posts and grids for lead-acid storage batteries, and is used in
the formulation of some copper alloys (Reese, 1998). As maintenance-free automotive batteries,
which contain little or no arsenic, replace lead-acid storage batteries, the demand for arsenic
metal is expected to decrease. An arsenic containing compound, crystalline gallium arsenide, is
produced from very pure arsenic metal. Crystalline gallium arsenide is a semiconducting
material used in computers, optoelectronic devices and circuits, and other electronic applications.
Industrial processes including the burning of fossil fuels, combustion of wastes
(hazardous and non-hazardous), mining and smelting (discussed separately below), pulp and
paper production, glass manufacturing, and cement manufacturing can result in emissions of
arsenic to the environment (USEPA, 1998b). Coal-burning power plants may emit aerosols and
fly ash that contain arsenic (Yan-Chu, 1994). These arsenic-containing particles may be
deposited from the atmosphere onto soil and surface waters. The amount of arsenic that is
emitted by power plants and fossil fuel combustion is not included in current Toxics Release
17
-------
Inventory data.5 However, utility emissions of arsenic and several other metals will be added to
the TRI in 1999 (USEPA, 1999a).
Past waste disposal practices have impacted arsenic concentrations in ground water and
surface water at waste disposal sites. Arsenic has been identified as a contaminant of concern at
916 of the 1,467 Superfund National Priority List (NPL) hazardous waste sites (ATSDR, 1998),
and arsenic is listed as the highest priority contaminant on the ATSDR/EPA list of the hazardous
substances at NPL Sites (ATSDR, 1997). There is a potential for releases of arsenic from waste
sites to affect ground water or surface water in the vicinity of the waste sites.
Mining and Smelting
Arsenic can be obtained from two of its ores, arsenopyrite and lollingite, by smelting in
the presence of air around 650-700 °C (Kirk-Othmer, 1992), or arsenic trioxide (As2O3) in flue
dust from the extraction of lead and copper can be captured (Ferguson, 1990). Subsequently,
arsenic trioxide can be used to produce other arsenic compounds or purified to elemental arsenic.
Arsenic trioxide was produced for commercial use in the United State at the ASARCO
smelter in Tacoma, Washington, until 1985, at which time the smelter ceased operations
(ATSDR, 1998). The USEPA Office of Air Quality Planning and Standards indicates that
primary and secondary6 lead smelters, primary copper smelters, and secondary aluminum
operations are potential sources of arsenic (USEPA, 1998b). There are presently three active
primary lead smelters, two of which are located in Missouri and one is in Montana. In addition,
there are 19 active secondary lead smelters located throughout the United States. The seven
primary copper smelters are located in Arizona (3), New Mexico (2), Texas (1), and Utah (1).
Secondary aluminum operations locations were not identified in USEPA 1998b.
Arsenic may be emitted to the atmosphere from metals smelters, and deposited with
precipitation downwind of the smelter. In Washington, Crecelius (1975) found higher
concentrations of arsenic in rain and snow downwind of a smelter (17 |ig/L) than in rain in
unpolluted areas (-< 1 |ig/L). Such atmospheric deposition could affect arsenic concentrations in
soil and other environmental media. Miesch and Huffman (1972) detected increased
concentrations in soil downwind of a smelter in Helena, Montana.
High concentrations of arsenic may occur in areas that are near or affected by current or
historical mining activities. Sulfide-bearing rocks are often mined for gold, lead, zinc, and
copper, and arsenic is frequently found as an impurity in the sulfide ores of these metals. The
drainage from abandoned mines and mine wastes is typically acidic, and dissolved arsenic
concentrations can be as high as 48,000 |ig/L in mine drainage (Welch et al, 1988). In mining
areas, the arsenopyrite (FeAsS) that occurs in association with ores and arsenic bearing pyrite is a
5 The Toxic Release Inventory is a database of toxic releases in the United States compiled annually from SARA
Title III Section 313 reports.
6 Primary smelters produce metals from raw ores, whereas secondary smelters reclaim metals from used and recycled
materials, such as scrap metal and used batteries.
18
-------
common source of dissolved arsenic. The mineral orpiment, realgar, and arsenic-rich iron oxides
are other sources of dissolved arsenic (Welch et al, 1988) at mining sites.
Other Uses and Sources
There is some evidence that some volatile organic compounds (VOCs) in ground water
may facilitate the release of arsenic from aquifer materials to ground water. Ground water that is
affected by VOCs like petroleum products and other landfill wastes may be sufficiently reduced
to result in elevated dissolved iron-oxide concentrations. Under these reducing conditions,
aquifer materials may be a source of dissolved arsenic in ground water (Ogden, 1990).
From the Civil War until approximately 1910, arsenic was used as an embalming fluid,
and elevated concentrations of arsenic were detected in ground water at cemeteries in Iowa and
New York (Konefes and McGee, 1996). Therefore, it appears that cemeteries may be sources of
localized arsenic contamination in ground water. However, the extent of or potential for arsenic
contamination associated with ground water in cemeteries has not been broadly evaluated.
Historically, inorganic and organic arsenic compounds were used as therapeutic agents.
The first recognized use of inorganic arsenic as a therapeutic agent was in 1786; Fowler's
solution, which contained approximately 1 percent arsenic trioxide, was a common arsenic
containing medicinal (NRC, 1999). Inorganic arsenic was used to treat symptoms of skin
diseases such as eczema and psoriasis, malarial and rheumatic fevers, asthma, pernicious anemia,
leukemia, Hodgkin's disease, and pain. Because of concern over inorganic arsenic's toxic
effects, therapeutic use of inorganic arsenic ceased in the 1970s. Organic arsenicals were used
for the treatment of spirochetal and protozoal diseases through the first half of the twentieth
century. Salvarsan (arspheramine) was a common anti-syphilitic from 1907 until its use was
supplanted by penicillin in the 1940s and 1950s. Organic arsenic compounds were used for the
treatment of amebiasis and trypanosomaisis. Melarsoprol (organic arsenic compound) continues
to be used for treatment of trypanosomaisis. In addition, arsenic trioxide is being researched as
therapy for acute promyelocytic leukemia.
Releases to the Environment Reported in the TRI1997
The Toxics Release Inventory (TRI) is a national database that identifies facilities, and
chemicals manufactured and used at identified industrial facilities. The TRI database also
identifies the amounts of these chemicals that are released to the environment annually as a result
of routine operations, accidents, and other one-time events, and the amounts of chemicals that are
managed in on- and off-site waste for each year since 1987. Annually, certain facilities must
report basic information about their facilities and operations, and about the amounts of listed
toxic chemicals used, released, recycled, or otherwise managed at the facility. Facilities must
report on arsenic and arsenic compound use, release, management, and disposal. Therefore, the
TRI data provides a significant amount of information regarding industrial releases of arsenic to
the environment, and trends in those releases since 1987. Information on arsenic releases to the
environment, based on recent TRI data, is summarized below.
19
-------
TRI facilities reported managing a total of 14,898,807 pounds of arsenic and arsenic
compound containing wastes that were either production or non-production related in 1997
(USEPA, 1999a). Total on- and off-site releases in 1997 were reported to be 7,947,012 Ibs. The
majority of these releases (6,046,473 Ibs) were on-site, and 95 percent of on-site releases were to
land (5,766,252 Ibs). The majority of the arsenic was released to the land via surface
impoundments or non-Resource Conservation and Recovery Act (RCRA) Part C landfills; less
than 1 percent of the arsenic and arsenic compounds waste is disposed of in RCRA Subtitle C
landfills. After releases to land, releases to air are the second largest component of on-site
releases, with a total of 199,918 Ibs of arsenic and arsenic compounds released from stacks, point
sources, and fugitive or non-point sources. In addition, 76,170 Ibs were reported to have been
injected underground and 4,133 Ibs were reported to have been released to surface water in 1997.
Of the 1,900,539 Ibs that were disposed of off-site, most (1,460,728 Ibs) were disposed of to
landfills and surface impoundments, and significant amounts were released by underground
injection (209,716 Ibs) or were solidified or otherwise stabilized (149,416 Ibs). Small amounts
were stored or transferred to wastewater treatment plants.
Data contained in the TRI indicate that releases of arsenic and arsenic compounds from
TRI reporting facilities to the environment have increased in recent years. From 1995 to 1997,
TRI data indicates that total on-site and off-site releases of arsenic have risen from 3,536,467 Ibs
to 7,947,012 Ibs. The increase primarily occurred at one facility, where arsenic on-site land
releases increased by 3.58 million pounds from 1995 to 1997 because of a change in the facilities
smelting process that was implemented to reduce sulfur dioxide emissions. However, from 1995
to 1997 the quantity of arsenic in air emissions, underground injection, releases to land, and
transfers to off-site disposal have all risen, while only the quantity of arsenic discharged to
surface water is reported to have decreased (USEPA, 1999a). Total releases of arsenic to the
environment in 1997 also exceed those for the base year (1988) for TRI data, when total on-site
and off-site emissions of arsenic were 6,911,043 Ibs.
Although the TRI data include estimated emissions from many sources of anthropogenic
sources of arsenic to the environment, the data do not include several potentially significant
sources of arsenic emissions, and therefore the data it contains should be interpreted with some
care. For example, TRI release data do not include arsenic in organoarsenical herbicides that are
applied to cotton fields. Several other potentially significant sources of arsenic emissions will
begin reporting to the TRI in 1999 for the year 1998, including coal and oil burning electrical
utilities, coal mining, and metals mining (USEPA, 1999). The addition of these industrial
sectors should improve the accuracy of reporting of emissions of arsenic and arsenic compounds.
20
-------
3. Fate and Transport of Arsenic
3.1 Relationship of Fate and Transport Properties to Source Intake
Arsenic concentration in fresh waters shows considerable variation with geological
composition of the drainage area and the level of anthropogenic input. The fate and transport of
arsenic in ground water and surface water are discussed separately below:
Ground water
Arsenic may be released to ground water in a variety of ways, which include, but are not
limited to, weathering of earth's crust and soil materials, discharge from industrial processes, and
overland runoff from agricultural and urban areas. In water, arsenic can undergo a series of
transformations, including oxidation-reduction reactions, ligand exchange, and
biotransformations (ATSDR, 1998; Welch et al, 1988). Several factors have been identified
which effect the fate and transport processes in ground water. These include the oxidation state
of the arsenic, oxidation-reduction potential (Eh), pH, iron concentrations, metal sulfide and
sulfide concentrations, temperature, salinity, and distribution and composition of the biota
(ATSDR, 1998; Roberston, 1989; Welch et al., 1988). The predominant form of arsenic is
usually arsenate (As+5), although arsenite (As+3) may be present under some conditions (Irgolic,
1994; Welch et al., 1988). However, the NRC (1999) noted that arsenite might be more
prevalent than anticipated.
Surface water
The processes that affect arsenic fate and transport in surface water are analogous to those
that are operative in ground water systems. Thus, the factors that affect arsenic transformations
and transport include the oxidation state of the arsenic, oxidation-reduction potential (Eh), pH,
iron concentrations, metal sulfide and sulfide concentrations, temperature, salinity, and
distribution and composition of the biota (ATSDR, 1998). However, there are additional factors
that affect arsenic fate and transport in surface water systems. These include total suspended
sediment (Nimick et al., 1998; Waslenchuk, 1979), seasonal water flow volumes and rates
(Nimick et al, 1998; Waslenchuk 1979), and time of day (Nimick et al, 1998).
Sorption of arsenic to suspended sediment may strongly affect the fate and transport of
arsenic in surface water systems. Where pH and arsenic concentrations are relatively high, and
total suspended sediment levels are relatively low, sorption processes may be less important
(Nimick et al, 1998). However, where suspended sediment loads are higher, arsenic
concentrations are lower, and pH levels are lower, arsenic is more likely to be present in the
suspended paniculate phase rather than the dissolved phase. Paniculate phase arsenic may settle
to bottom sediment in reservoirs and areas with low flow levels. The sorption of arsenic onto
suspended sediment is a mechanism for the removal of dissolved arsenic from surface water.
Sorption is greater when the amount of suspended sediment is greater. In surface water, lakes
may interrupt the downstream transport of particulate sorbed arsenic. In deeper lakes,
remobilization of arsenic from the sediment may be minimal, whereas in shallower lakes, arsenic
21
-------
may be remobilized faster from wind induced wave action and high-flow scouring. Large and
deep reservoirs are more likely to be long-term sinks for arsenic.
Seasonal variations of arsenic concentrations have been observed in surface water
systems, and these variations appear to be related to the source of the arsenic and the flow of the
river. In the Madison River, where relatively constant inputs of arsenic originate from
geothermal sources, arsenic concentrations at one point ranged from 110 to 370 |ig/L in samples
collected between 1986 and 1995 (Nimick et al, 1998). The highest concentrations occurred
during periods of low flow, and the lowest concentrations occurred during periods of high flow.
Waslenchuk (1979) measured arsenic concentrations in rivers in the Southeastern United States,
where average arsenic levels are far lower than in the Madison River. In the rivers that
Waslenchuk studied, seasonal variations of as great as 0.2 jig/L were observed around average
concentrations that ranged from 0.15 to 0.45 |ig/L. These variations were related to seasonal
precipitation levels, which flushed higher concentrations of arsenic into the rivers in the spring.
Arsenic concentrations in surface water may also change during one day, as a result of
changes in water pH that are attributable to incoming solar radiation and photosynthesis (Nimick
et al, 1998). These consistent daily changes are called diurnal variability. Because of
photosynthesis, the water pH tends to increase later in the day, and dissolved arsenic
concentrations also tend to increase. Diurnal variations in dissolved arsenic concentrations of as
much as 21-percent were observed at three of the five sites on the Madison and Missouri Rivers,
but diurnal variability was not seen at the other two sites. Thus, diurnal variability may affect
arsenic concentrations in some surface water sources.
3.2 Relationship of Fate and Transport Properties to Treatment and Distribution
A variety of processes and factors affect the fate and transport of arsenic within public
water supply treatment systems. The most important factor appears to be the oxidation state
(arsenate or arsenite). The presence of competing ions, especially sulfate and fluoride, are also
important, as is pH. This section discusses the effects of these factors upon arsenic
concentrations and removal from water in treatment systems.
Arsenic in water commonly occurs as arsenate, As (V), or arsenite, As (ID). The
chemical species that are formed depends upon the oxidation-reduction conditions and the pH of
the source water. The common soluble species of arsenate are H3AsO4, H^AsCV", HAsO42", and
AsO43"; whereas the common soluble species of arsenite are H3AsO3 and t^AsCV". At typical
pHs, the predominant arsenite form is the neutral species (H3AsO3), while the predominant
arsenate species are the anions I^AsCV" and HAsO42". Because of its ionic charge, arsenate is
more easily removed from source waters than arsenite. In particular, activated alumina, ion
exchange, and reverse osmosis may achieve relatively high arsenate removal rates, but they show
lower treatment efficiencies for arsenite.
Arsenite can be oxidized to arsenate, and this can improve arsenic removal efficiencies.
In water that contains no ammonia or total organic carbon (TOC), chlorine rapidly (in less than 5
22
-------
seconds at chlorine concentrations of 1.0 mg/L) oxidizes approximately 95 percent of arsenite to
arsenate (Clifford, 1986). The reaction through which chlorine oxidizes arsenite is:
As3+ + C12 ^ As5+ + 2Cr [2]
The presence of ammonia and TOC slows this oxidation process. Monochloramine at the
concentration of 1.0 mg/L oxidized 45 percent of arsenite to arsenate. Oxygen may slowly
oxidize arsenite to arsenate, although this reaction occurs slowly in laboratory studies. Shen
(1973) indicated that potassium permanganate can oxidize arsenite. Potassium permanganate
oxidizes arsenite according to the following reaction:
3As3+ + 2KMnO4 ^ 2MnO4 + 3As5+ [3]
Therefore, it appears that chlorine and potassium permanganate are the most effective
processes to oxidize arsenite to arsenate. The oxidation of arsenic from its trivalent state to its
pentavalent state can allow treatment plants to increase the removal efficiencies of treatment
technologies.
The water pH also affects the removal efficiencies of treatment technologies for arsenic,
and therefore, the level of and persistence of arsenic it drinking water. For example, activated
alumina removes arsenic most efficiently at pH 6, but yields lower treatment efficiencies at pH 9
(Rubel and Hathaway, 1987). Removal efficiencies for alum coagulation tend to decrease at pHs
greater than 7.0 (GSulledge and O'Connor, 1973).
Competition for adsorption sites with other ions may also affect the persistence and
removal of arsenic from source water in treatment plants. In particular, sulfate in source waters
may reduce the efficiency of arsenic removal. Sorg (1990) showed that waters with high sulfate
levels correlate with lower arsenic removal by ion exchange technologies. Anion exchange
resins preferentially adsorb sulfate over arsenic. Because of this preference for sulfate over
arsenic, under some conditions displacement of arsenic can result in peaks of arsenic in effluent
which exceed the concentrations of arsenic in source water influent. Jekel (1994) reported that
the competitive effects of sulfate, fluoride, and phosphate may reduce the effectiveness of
activated alumina treatments, particularly when the competitors are in the range of 0.1 to 2
mg/kg, and arsenic removal to the ppb level is required. If the competing ions are present in
small concentrations, activated alumina can be applied successfully at slightly acidic pH ranges
(pH 5. 5 to 6.0).
In summary, three factors are particularly important to the fate of arsenic in treatment and
distribution systems. These include the oxidation state, the pH of the source water, and the
presence of other ions which may compete for adsorption sites in treatment technologies. The
most significant factor that affects the fate and transport of arsenic in treatment and distribution
systems appears to be the arsenic oxidation state. Arsenate is removed more efficiently than
arsenite.
23
-------
This page intentionally left blank.
24
-------
4. Sources of Data on Arsenic Occurrence in
Drinking Water Supplies
The potential health concerns associated with arsenic in drinking water have long been
recognized, and therefore a significant amount of data is available on source water and treated
drinking water arsenic concentrations. These data sources, which include national, regional, and
State databases, differ substantially in size, content, and quality. Table 4-1 lists the national data
sources which were available for development of arsenic occurrence estimates, and provides
general information about the characteristics of these data sources.
An important source of information for estimating arsenic occurrence on a national basis
is compliance monitoring data collected in accordance with the Safe Drinking Water Act
(SDWA). The estimates of arsenic occurrence and intra-system variability presented in Chapters
6 and 7 of this report were developed using compliance monitoring data submitted by several
states. In this Chapter, Section 4.1 discusses the development of this database from sets of
compliance monitoring data submitted to the USEPA, and from information contained in the
Safe Drinking Water Information System (SDWIS). In this report, this database is referred to as
the Arsenic Occurrence and Exposure Database (AOED).
Other arsenic occurrence surveys also provide potentially important sources of arsenic
occurrence information. Recently developed databases include the National Arsenic Occurrence
Survey (NAOS) the United States Geological Survey Arsenic Database (USGS), and the
Metropolitan Water District of Southern California database (Metro). These databases are
described in Section 4.2, and they are used in Chapter 6 to provide comparisons with the
occurrence projections developed using AOED. Another database that could be used as a
comparison tool is the National Inorganics and Radionuclides Survey (NIRS). This database is
also described in Section 4.2.
In addition to the compliance monitoring data and the databases that have been used as
comparison tools in this report, a variety of other data sources are available that provide arsenic
occurrence information. However, these data sources were not used in this report for various
reasons. These databases, which include the Rural Water Survey (RWS), the 1969 and 1978
Community Water Supply Surveys (CWSS), the National Organics Monitoring Survey (NOMS),
the occurrence data gathered by the Western Coalition of Arid States (WESCAS) and the
Association of California Water Agencies (ACWA) database, are briefly described in Section
4.3, and the reasons why these databases were deemed unsuitable for use in this occurrence
estimation are presented.
4.1 Arsenic Occurrence and Exposure Database (AOED)
The Arsenic Occurrence and Exposure Database was developed to support the USEPA's
effort to estimate arsenic occurrence in the United States. The database includes information
from SDWIS and from State compliance monitoring data sets that were provided to IS SI.
Section 4.1.1 describes SDWIS. Section 4.1.2 describes the State compliance monitoring
25
-------
Table 4-1
Summary of National Arsenic Data Sources
Database
SDWIS
State Compliance
Databases
MRS
NAOS
Metro
USGS
1969 CWSS
1978 CWSS
RWS
NOMS
Number of Systems
~ 55,000
24,247
982
<517
112
~ 20,000
969
<350
92
113
Media (GW or SW)
GW and SW
GW and SW
GW
GW and SW
GW and SW
GW
GW and SW
GW and SW
GW and SW
GW and SW
Reporting Limit
(Mg/L)
50
0.001 - 10
5
0.5
0.5
1
5
2.5
2
??
Source Data
-
/
-
-
~
-
-
-
-
Drawbacks
Reports only MCL
violations
Not all States are
covered
95% Censored
Untreated and
predicted drinking
water arsenic
concentrations
Large systems
Untreated and non-
public water supply
water
Pre 1980 data
Pre 1980 data
Pre 1980 data
Pre 1980 data
Notes:
GW - Ground water
SW - Surface water
26
-------
databases. Section 4.1.3 explains how SDWIS and State compliance monitoring data sets were
used to build the AOED.
4.1.1 Safe Drinking Water Information System (SDWIS)
SDWIS provides a complete inventory of information on all public and private water
supply systems in the United States. The SDWIS database is used for compliance tracking and to
allocate SDWA grant monies. Violations of the current arsenic MCL of 50 |ig/L are recorded in
the SDWIS database, but measured concentrations below the MCL are not recorded. While this
information is useful for determining the number of systems that have violated current Federal
arsenic drinking water limits, it is not suitable for estimating arsenic concentrations in the range
of2to50|ig/L.
SDWIS is a reliable source of information on the characteristics of individual public
water systems (PWS) in the United States. Since the set of PWS in the United States changes
over time, the SDWIS information is checked for accuracy and updated on an annual basis. Near
the end of the calendar year, the SDWIS inventory is "frozen", such that no more additions are
made to the database that year. This inventory is then distributed to State drinking water
programs for verification of the numbers and types of systems (Science Applications
International Corporation (SAIC), 1999). The core verified data in the SDWIS inventory
include:
System name and address;
Federal identification number (SDWIS ID number or PWSID);
Source water type;
Ownership category;
Population served; and
Regulatory classification (system type).
In regard to the category of source water type, a system may include more than one type
of source. Small systems typically have only one source, but larger systems may have multiple
sources. Although most large systems are served by either a ground water or a surface water
source, some systems do receive water from a mix of ground water and surface water sources. In
SDWIS, any water system with a continuous source of surface water is defined as a surface water
system, even if 75 percent, or 99 percent, of the surface water is from a ground water source.
Therefore, systems that rely entirely on surface water sources and blended water systems are
coded as surface water systems in SDWIS, and systems that rely entirely on ground water sources
are coded as ground water systems in SDWIS. SDWIS also includes the source water type
category "ground water under the influence of surface water." The regulatory definition
essentially defines this category as systems where, for various reasons, the ground water is
expected to be contaminated similarly to nearby surface water sources. For the occurrence
analyses described in this report, ground water systems under the influence of surface water were
included in the surface water category.
27
-------
SDWIS characterizes the population served on the basis of retail customers only. The
SDWIS inventory populations do not include the number of people served with water that is
wholesaled by any individual utility (SAIC, 1999).
Ownership categories are limited to public and private, while regulatory classification
includes community water supplies (CWS), non-transient, non-community water supply systems
(NTNCWS) and transient non-community water supplies (TNCWS). The arsenic occurrence
estimates presented in this report are for CWS and NTNCWS.
In summary, the SDWIS database is a very useful source of information on the
characteristics of individual systems. In fact, as discussed in Section 4.1.3, SDWIS was a key
resource for information on the characteristics of individual systems for the development of
AOED. However, it does not contain information on the levels of arsenic in individual systems
that can be used to estimate arsenic occurrence within the range of interest.
4.1.2 State Compliance Monitoring Databases
An important source of data for estimating arsenic levels in drinking water is compliance
monitoring data that is collected from the drinking water utilities in accordance with the SDWA,
to comply with the current arsenic MCL of 50 ug/L. The compliance monitoring data sets were
submitted voluntarily to the USEPA by State drinking water agencies, either directly or through
other organizations (e.g., AWWA, EPA Office of Research and Development, EPA Regional
Office, Association of Public Health Laboratories), during this study and during earlier studies.
Such data was available from 32 States, although data from only 25 States was found to be
suitable for inclusion in this occurrence study. Table 4-2 presents an overview of the suitable
compliance monitoring data for each State. Figure 4-1 shows the geographical distribution
represented in the 25 State's compliance monitoring databases. The characteristics of the 25
suitable individual state databases are documented in Appendix D-l. While the States for which
compliance monitoring data are available are distributed throughout the United States, this figure
shows that the States are not evenly distributed. In particular, few data sets are available for
States in the New England, Mid-Atlantic, and Southeastern United States. In contrast, the
Midwestern, North Central, South Central, and Western Regions appear to be fairly well
represented. The States in these regions are described on Section 5.5; see Figure 5-2 for a map of
the United States showing the seven regions.
A number of data sets were considered for inclusion in the AOED database, but were not
included for various reasons. Table 4-3 lists the States for which at least one data set was
excluded from the AOED database, and the specific reason why the data set was excluded. Data
sets from 19 States were excluded from the database. However, 12 of these States submitted
multiple data sets; the single most representative data set from each of these States was chosen
and included in AOED. In most cases, these States submitted multiple compliance monitoring
data sets, and the most recent data sets represented the largest number of systems. The later
compliance monitoring data sets also tended to have the lowest detection levels, although this
was not true in all cases. In one case, Maine, the older data set was included because the newer
data set only reported arsenic levels above 20 pg/L. Minnesota had two types of data set, one
28
-------
Table 4-2
Overview of State Compliance Monitoring Data
State
Alaska
Alabama
Arkansas
Arizona
California
Illinois
Indiana
Kansas
Kentucky
Maine
Michigan
Minnesota
Missouri
Montana
North Carolina
North Dakota
New Hampshire
New Jersey
New Mexico
Nevada
Ohio
Oklahoma
Oregon
Texas
Utah
Number of
CWSGW
326
263
371
668
1369
1082
648
506
88
109
644
863
773
484
1735
197
504
438
573
221
875
446
583
3105
327
Number of
cwssw
109
68
76
46
222
103
51
101
150
29
33
23
89
47
169
19
37
29
29
31
139
210
134
326
38
Number of
NTNCWS
GW
140
31
0
190
376
0
538
62
0
0
230
634
190
2
562
20
0
758
140
0
0
0
112
580
50
Number of
NTNCWS
SW
26
5
0
4
11
0
4
2
0
0
0
6
4
0
8
6
0
o
6
8
0
0
0
9
27
3
Reporting
Limits
1-5
1
5
0.4 - 10
0.001 - 10
0.005 - 10
0.5-8
1
1-5
1
0.3-2
1-5
1
1
0.2 - 10
0.2-1
5
0.1 -10
0.3 - 10
o
5
1-10
2
0.5 - 10
1 -10
0.1-10
Mode of Reporting
Limits
1
1
5
10 (GW) and 5 (SW)
10
2 (GW) and 1 (SW)
1
1
1
1
0.3 (GW) and 2 (SW)
1
1
1
10 (GW) and 5 (SW)
1
5
5 (GW) and 2 (SW)
5
o
3
10
2
5
2
5
Date Ranges
1991 - 1997
1985 - 2000
1996 - 1998
1988 - 1998
1981-2000
1993 - 2000
1996 - 1999
1992 - 1997
Not reported
1991 - 1994
1993 - 1997
1992 - 1997
1995 - 1997
1980 - 1992
1980 - 2000
1993 - 1995
1990 - 1994
1993 - 1997
1983 - 2000
1991 - 1997
1981 - 1994
1995 - 1998
1990 - 1998
1994 - 1999
1980 - 1999
Intra-system
Data
No
Yes
Yes
No
Yes
Yes
Yes
No
No
No
No
No
No
No
Yes
No
No
No
Yes
No
No
Yes
No
Yes
Yes
29
-------
Figure 4-1: States with Suitable Arsenic Compliance Monitoring Data
NORTH
DAKOTA (MINNESOTA
NEW HAMPSHIRE
MASSACHUSETTS
RHODE ISLAND
CONNECTICUT
NEW JERSEY
DELAWARE
MARYLAND
States with compliance data
States without compliance data
30
-------
Table 4-3
State Data Sets Excluded from AOED
State
Alaska
Arizona
California
Florida
Idaho
Illinois
Iowa
Louisiana
Maine
Michigan
Minnesota
New Mexico
Oregon
Pennsylvania
North Dakota
South Dakota
Texas
Utah
West Virginia
Reason Data Set Excluded
Multiple data sets submitted; only the most recent data set selected.
Multiple data sets submitted; only the most recent data set selected.
Multiple data sets submitted; only the most recent data set selected.
Data set unsuitable; all results censored, no detection limit reported.
Data set unsuitable; no PWS ID numbers were provided.
Multiple data sets submitted; only the most recent data set selected.
Data set unsuitable; no reporting limits; results rounded to nearest 10 ug/L.
Data set unsuitable; no reporting limits; only detected values provided, results rounded to nearest
10 ug/L.
Selected older data set; recent data set censored at 20 ug/L.
Multiple data sets submitted; only the most recent data set selected.
Selected data set associated with PWSID numbers.
Multiple data sets submitted; only the most recent data set selected.
Multiple data sets submitted; only the most recent data set selected.
Data set unsuitable; all results censored, reporting limit of 50 ug/L.
Multiple data sets submitted; only the most recent data set selected.
Data set unsuitable; no result code flag was provided.
Multiple data sets submitted; only the most recent data set selected.
Multiple data sets submitted; only the most recent data set selected.
Data set unsuitable. Not made available electronically. Most results censored
containing compliance monitoring data, and one containing data that was not from PWS systems
(data from private domestic wells or non-public water supply wells). In this case, only the most
recently submitted compliance monitoring data was included. For seven States, the available
data sets were unsuitable and were not included in the State compliance monitoring database.
Only one set of data was available for each of these States, thus, these States are not represented
in the compliance monitoring database. The reasons for exclusion of these data sets are listed in
Table 4-3. ISSI contacted representatives of the State agencies that provided these data sets, but
was unable to resolve the problems with these data sets, or obtain newer, suitable data sets.
31
-------
The compliance monitoring data sets were submitted voluntarily to the USEPA by State
drinking water agencies, either directly or through other organizations (e.g., AWWA, EPA Office
of Research and Development, EPA Regional Office, Association of Public Health Laboratories),
during this study and during earlier studies. These data sets generally included the following
information:
System name and address;
PWS ID number;
Sample collection date;
Result; and
Detection limit, if arsenic was not detected in the sample.
Because these data were collected for compliance purposes, the samples were assumed to
have been collected from the points-of-entry (POE) into the distribution system, which is
representative of each well or source after treatment. Thus, the arsenic values represent finished
water, and should be representative of the arsenic levels to which consumers are exposed.7
Compliance with the arsenic standard is measured at the POE. Therefore, these data are directly
relevant to the estimation of regulatory costs and benefits for the RIA.
Representation of Systems in Each State
As shown in Table 4-2, the characteristics of the data differ from State to State. Tables
4-4a and 4-4b present the numbers of CWS and NTNCWS systems contained in the State data
sets as a percentage of the total numbers of non-purchased CWS and NTNCWS systems in the
State based on SDWIS. Purchased water systems were excluded from the occurrence data base
used for the analyses. (The SDWIS total numbers of purchased and non-purchased systems in
each state were used in the development of regional and national occurrence estimates, as
described in Chapter 6.) For purchased water systems, EPA allows the State to decide how each
system will conduct monitoring. In some cases, monitoring may be done by only the wholesaling
system. The SDWIS database, however, only contains data on retail populations and does not
indicate to which systems a wholesaler provides its water. As a result, we do not know to what
extent purchased water systems are representative of other water systems in the State's database.
Using the purchased water system data to develop the State probability estimates could bias the
estimates by double counting individual system results. Developing results based on non-
purchased water system results ensures that independent data form the basis for the State
estimates.
Tables 4-4a and 4-4b show that both ground and surface water systems are well
represented in the compliance monitoring databases, particularly for CWS systems. Overall, the
compliance monitoring data represent 82.8 percent of the CWS ground water systems in each
State and 82.4 percent of the CWS surface water systems in each State. For NTNCWS, the
compliance monitoring data represent 57.6 percent of the ground water systems in each State and
63.0 percent of the surface water systems in each State. For 13 States, at least 90 percent of the
7 See convention number 3 in Section 4.1.3 below for details about how untreated samples and samples before
treatment were used in the occurrence analyses.
32
-------
Table 4-4a
CWS Systems in SDWIS and State Compliance Monitoring Data
State
Alaska
Alaska
Alabama
Alabama
Arkansas
Arkansas
Arizona
Arizona
California
California
Illinois
Illinois
Indiana
Indiana
Kansas
Kansas
Kentucky
Kentucky
Maine
Maine
Michigan
Michigan
Minnesota
Minnesota
Missouri
Missouri
Montana
Montana
Source Type
GW
SW
GW
SW
GW
SW
GW
SW
GW
SW
GW
SW
GW
SW
GW
SW
GW
SW
GW
SW
GW
SW
GW
SW
GW
SW
GW
SW
Systems in
SDWIS3
337
111
270
69
373
78
721
46
2652
522
1129
108
794
54
516
101
101
173
315
79
1141
76
902
24
1097
96
544
50
Systems in State
Data Set3
326
109
263
68
371
76
668
46
1369
222
1082
103
648
51
506
101
88
150
109
29
644
33
863
23
773
89
484
47
Percent Coverage
96.7
98.2
97.4
98.6
99.5
97.4
92.6
100.0
51.6
42.5
95.8
95.4
81.6
94.4
98.1
100.0
87.1
86.7
34.6
36.7
56.4
43.4
95.7
95.8
70.5
92.7
89.0
94.0
33
-------
Table 4-4a (continued)
CWS Systems in SDWIS and State Compliance Monitoring Data
State
North Carolina
North Carolina
North Dakota
North Dakota
New Hampshire
New Hampshire
New Jersey
New Jersey
New Mexico
New Mexico
Nevada
Nevada
Ohio
Ohio
Oklahoma
Oklahoma
Oregon
Oregon
Texas
Texas
Utah
Utah
Total Systems
Total Systems
Source Type
GW
SW
GW
SW
GW
SW
GW
SW
GW
SW
GW
SW
GW
SW
GW
SW
GW
SW
GW
SW
GW
SW
GW
SW
Systems in
SDWIS3
1848
171
202
19
618
38
494
30
582
29
249
34
1016
165
458
211
665
138
3421
341
334
38
20779
2801
Systems in State
Data Set3
1735
169
197
19
504
37
438
29
573
29
221
31
875
139
446
210
583
134
3105
326
327
38
17198
2308
Percent Coverage
93.9
98.8
97.5
100.0
81.6
97.4
88.7
96.7
98.5
100.0
88.8
91.2
86.1
84.2
97.4
99.5
87.7
97.1
90.8
95.6
97.9
100.0
82.8
82.4
Note:
a The total number of non-purchased public and private systems.
water systems are included as surface water systems.
Ground water under the influence of surface
34
-------
Table 4-4b
NTNCWS Systems in SDWIS and State Compliance Monitoring Data
State
Alaska
Alaska
Alabama
Alabama
Arizona
Arizona
California
California
Indiana
Indiana
Kansas
Kansas
Michigan
Minnesota
Minnesota
Missouri
Missouri
Montana
Montana
North Carolina
North Carolina
North Dakota
North Dakota
New Jersey
New Jersey
New Mexico
New Mexico
Oregon
Oregon
Texas
Texas
Utah
Utah
Total Systems
Total Systems
Source Type
GW
SW
GW
SW
GW
SW
GW
SW
GW
SW
GW
SW
GW
GW
SW
GW
SW
GW
SW
GW
SW
GW
SW
GW
SW
GW
SW
GW
SW
GW
SW
GW
SW
GW
SW
Systems in
SDWIS3
152
28
43
7
209
5
1068
53
703
6
64
2
1783
653
6
228
4
213
3
644
8
22
7
975
3
153
8
325
9
727
48
52
3
8014
200
Systems in State
Data Set3
140
26
31
5
190
4
376
11
538
4
62
2
230
634
6
190
4
2
0
562
8
20
6
758
3
140
8
112
9
580
27
50
3
4615
126
Percent
Coverage
92.1
92.9
72.1
71.4
90.9
80.0
35.2
20.8
76.5
66.7
96.9
100.0
12.9
97.1
100.0
83.3
100.0
0.9
0.0
87.3
100.0
90.9
85.7
77.7
100.0
91.5
100.0
34.5
100.0
79.8
56.3
96.2
100.0
57.6
63.0
Note:
a The total number of non-purchased public and private systems. Ground water under the
influence of surface water systems are included as surface water systems.
-------
CWS ground water systems are represented, and another six states have between 85 and 90 %
coverage. California, Maine, and Michigan had the lowest coverage of CWS ground water
systems, on a percentage basis; yet, in each of these States, more than 100 systems are
represented, and more than 1,000 systems are represented in California. Among CWS surface
water systems, 20 States data sets include at least 90 percent of PWS, and the lowest percentage
of coverage is provided by the data sets for California, Maine, and Michigan. For NTNCWS
systems, 7 of the 17 states with ground water systems had 90 percent or more of their systems
represented, and 8 of the 16 states with surface water systems had 90 percent or more of their
systems represented. The lowest percentage coverages for NTNCWS systems were for Montana
(0 out of 3 surface water systems and 1 % of ground water systems), Michigan (12.9 % of
ground water systems), and California (35.2 % of ground water systems and 20.8 % of surface
water systems).
The total number of ground water and surface water systems represented is far higher
than the number available from other national data sources: There were 17,198 CWS ground
water systems, 2,308 CWS surface water systems, 4,615 NTNCWS ground water systems, and
126 NTNCWS surface water systems. (Note that these counts include all systems in the
database. As discussed below, 1,588 of these 24,247 systems were for years with higher reporting
limits and were not used for the analyses in Chapters 5, 6, and 7).
Analytical Methods and Reporting Limits
Data for 11 of the States are censored8 at a single reporting limit, but for 14 States, data
are censored at multiple reporting limits that usually range from 1-10 |ig/L. Where there are a
range of reporting limits, it suggests that arsenic samples may have been analyzed with more than
one sample method. Several of the State data sets include method numbers associated with
individual samples. Based on these method numbers, the following analytical techniques, which
are approved for compliance monitoring at the arsenic MCL of 50 |ig/L9, are represented among
the compliance monitoring data sets (systems must use one of these approved methods for
compliance monitoring at a certified laboratory):
Inductively coupled plasma (ICP) - Atomic Emission Spectrometry (AES)
ICP-Mass Spectrometry (MS)
Platform Graphite Furnace - Atomic Adsorption (AA)
Graphite Furnace AA; and
Hydride Generation AA.
Because these samples were collected for compliance monitoring, it is assumed that
appropriate sample collection and laboratory quality assurance and quality control protocols were
followed.
8 Censored data are samples with contaminant concentrations reported as less than the analytical detection limit.
Actual contaminant concentrations in these samples may be positive, and may range from zero to the detection limit.
In the case of a naturally occurring contaminant, such as arsenic, contaminant concentrations may be exceedingly
low, but are rarely zero.
9 40 CFR Section 141.23.
36
-------
Time trends were found in the reporting limits for 8 of the 14 States with multiple
reporting limits when the data were sorted by date. These trends were found by simple
inspection, rather than a formal statistical trend test. These data are summarized in Table 4-5,
combining data for CWS and NTNCWS systems. This summary table shows the distributions of
the reporting limits for two groups of years in each state, one group having generally lower
reporting limits. These groupings were chosen separately for each state based on a detailed
review of the reporting limit distributions by state and year (not shown). Thus, the selected break
year varies across the eight states.
Table 4-5
Summary of Reporting Limits (RLs) for Eight States
State
Alaska
Alaska
Arizona
Arizona
California
California
Illinois
Illinois
Minnesota
Minnesota
New Mexico
New Mexico
Oregon
Oregon
Utah
Utah
Year Range
1991 to 1994
1995 to 1997
1988 to 1995
1996 to 1998
1980 to 1994
1995 to 2000
1992 to 1995
1996 to 2000
1992 to 1993
1994 to 1997
1980 to 1994
1995 to 2000
1990 to 1992
1993 to 1998
1980 to 1988
1989 to 1999
Total
Number
of
Samples
1152
367
5516
1459
13481
11425
1321
3019
621
3085
2765
2855
1211
1222
2030
2872
Total ND
Samples,
RL
Known
767
307
3595
598
9574
3414
1037
2090
442
1730
1604
816
5
972
1228
2364
Total
ND
Samples,
RL Estimated
0
0
2
1
1297
3188
1
2
0
0
109
4
1032
59
1
0
Total ND
Samples,
RL
Known,
<=2
767
0
133
19
1333
3314
1020
1891
266
1730
114
740
0
284
1156
388
Total ND
Samples,
RL
Known,
2.01 to 5
0
307
1401
553
3250
92
16
187
176
0
1334
72
3
521
22
1879
Total ND
Samples,
RL
Known,
5.01 to 10
0
0
2061
26
4991
8
1
12
0
0
156
4
2
167
50
97
The total numbers of non-detect samples given in these tables include non-detect samples
where a known reporting limit was supplied with the data and non-detect samples where the
reporting limit was unknown. Samples with zero concentration values were also treated as non-
detect samples with an unknown reporting limit. The reporting limit distributions shown in these
tables are for the non-detect samples with known reporting limits only. For the remaining non-
detect samples with unknown reporting limits, the reporting limit was estimated as the most
commonly occurring reporting limit for the state and year range, as described in section 4.1.3
37
-------
below. To avoid a biased comparison, those non-detect samples with unknown, estimated
reporting limits are not included in these counts of non-detect samples with reporting limits
below 2, from 2 to 5, or from 5 to 10 |ig/L.
These data show that reporting limits have decreased in four of the seven States other
than Oregon, and increased in the other three States. For Oregon, almost all of the non-detects
between 1990 and 1992 had an unknown reporting limit, and so use of the newer data is
preferred. In Arizona, the majority of samples collected between 1988 and 1995 had reporting
limits of 10 |ig/L, but between 1996 and 1998, the primary reporting limit declined to 5 |ig/L. In
California, Minnesota, and New Mexico, reporting limits were sharply lower in the later time
periods indicated in Table 4-5. In these three States, the majority of non-detect samples had a
reporting limit of 2 |ig/L in the later time periods. In contrast, reporting limits in Alaska rose
from 2 |ig/L to 5 |ig/L after 1994 and the reporting limits in Utah rose from values mainly <2
|ig/L in 1980 to 1988 to values in the range 2-5 |ig/L from 1989 on. In Illinois, moderate
increases in reporting limits were seen in the later time period. No time trends were observed in
the data for the remaining 6 States - Indiana, Kentucky, Michigan, New Jersey, North Carolina,
and Ohio - with multiple reporting limits.
Based on these analyses, a decision was made that for those eight states with strong time
trends in the reporting limits, the data subsets for the years with the lower reporting limits would
be used for most of the analyses, including the occurrence computations in chapter 6 and the
intra-system variability analyses in chapter 7. The main reason for using these data subsets is
that there can be substantial bias introduced by the substitution method of half the detection limit
for non-detects when detection limits are higher. This substitution method was used for systems
with four or fewer detected values (i.e., values above the reporting limit), or with five or more
equal detected values, which was the case for the majority of systems with some non-detects.
Thus it is reasonable to use the subset with lower reporting limits even if that meant that the
older data was used instead of newer data with higher reporting limits. We believe that any
change in background arsenic levels occurs very slowly, so that it is reasonable to use data as old
as 1980 if that data is of higher quality. For the other 17 states, all the data would be used. The
AOED includes the original data set with all years of data and includes the data subsets used for
the analyses.
As discussed in the last paragraph, for eight states the state subset of years with lower
reporting limits was used to develop the occurrence estimates. Tables 4-6a (CWS systems) and
4-6b (NTNCWS systems) show the years used to define those state subsets and compare the
arsenic occurrence data contained in the complete State data sets with the occurrence data
contained in the subsets for ground water and surface water systems. The tables show the
numbers of samples, systems and percentage of non-detect samples. Also shown are the
minimum, mean, and maximum of the system means; the calculation of estimated system means
for systems with one or more non-detect samples using a regression or order statistics approach
is detailed in Section 6.1.1 and appendix A below. Because the subsets have lower detection
levels, it was expected that the subsets would have lower censoring levels and more accurate
system mean arsenic concentrations. For CWS systems in seven of the eight States, subsetting
reduces the percent of samples censored for ground water systems. The largest reductions
occurred in the States of Arizona (16.3 percentage points), California (13.3 percentage points),
38
-------
Table 4-6a
Overview of Complete Data Sets versus Data Subsets of CWS Systems
State &
Source Type
AKGW
AKGW
AZGW
AZGW
CAGW
CAGW
ILGW
ILGW
MNGW
MNGW
NMGW
NMGW
ORGW
ORGW
UTGW
UTGW
AKSW
AKSW
AZSW
AZSW
CASW
CASW
ILSW
ILSW
MNSW
MNSW
NMSW
NMSW
ORSW
ORSW
UTSW
UTSW
Data Set
All
1991 to 1994
All
1996 to 1998
All
1995 to 2000
All
1992 to 1995
All
1994 to 1997
All
1995 to 2000
All
1993 to 1998
All
1980 to 1988
All
1991 to 1994
All
1996 to 1998
All
1995 to 2000
All
1992 to 1995
All
1994 to 1997
All
1995 to 2000
All
1993 to 1998
All
1980 to 1988
Number
of
Systems
326
304
668
279
1369
1224
1082
750
863
829
573
559
583
316
327
263
109
106
46
33
222
176
103
93
23
23
29
29
134
129
38
35
Number
of
Samples
628
476
4126
805
21143
9494
3576
1131
2423
2021
4775
2334
1304
576
3308
1284
532
427
2096
518
2769
1280
764
190
63
54
430
141
944
514
1376
698
Average
System
Mean
[As]*
4.23
4.11
7.82
9.53
4.92
4.20
2.36
2.11
2.76
2.77
4.22
3.81
3.41
2.77
3.80
2.89
1.28
1.29
7.21
4.68
3.33
2.38
0.92
0.75
0.87
0.88
2.01
1.03
2.30
1.45
2.41
1.94
Min.
System
Mean
[As]*
0.03
0.01
0.42
0.28
0.00
0.00
0.04
0.01
0.05
0.04
0.20
0.06
0.30
0.04
0.41
0.05
0.07
0.13
2.00
1.90
0.26
0.24
0.50
0.30
0.58
0.63
0.66
0.18
0.45
0.22
0.43
0.18
Max.
System
Mean
[As]*
50.00
61.88
118.19
101.60
96.71
99.00
61.40
59.10
65.82
65.82
42.56
57.72
50.50
56.00
51.13
55.49
13.10
14.33
63.45
15.25
68.33
39.25
4.42
3.18
1.60
1.25
3.74
4.32
16.03
10.66
15.41
20.56
Percent
Samples ND
64.3
60.3
64.0
47.7
68.4
55.1
69.9
75.8
57.9
55.6
43.9
27.5
78.2
75.3
71.4
57.8
82.1
78.0
51.9
29.3
84.2
76.4
82.7
95.3
82.5
79.6
71.9
53.2
94.0
94.2
79.0
65.6
* Average, minimum, and maximum CWS system mean arsenic concentrations in ug/L .
39
-------
Table 4-6b
Overview of Complete Data Sets versus Data Subsets of NTNCWS Systems
State &
Source
Type
AKGW
AKGW
AZGW
AZGW
CAGW
CAGW
MNGW
MNGW
NMGW
NMGW
ORGW
ORGW
UTGW
UTGW
AKSW
AKSW
AZSW
AZSW
CASW
CASW
MNSW
MNSW
NMSW
NMSW
ORSW
ORSW
UTSW
UTSW
Data Set
All
1991 to 1994
All
1996 to 1998
All
1995 to 2000
All
1994 to 1997
All
1995 to 2000
All
1993 to 1998
All
1980 to 1988
All
1991 to 1994
All
1996 to 1998
All
1995 to 2000
All
1994 to 1997
All
1995 to 2000
All
1993 to 1998
All
1980 to 1988
Number
of
Systems
140
131
190
77
376
330
634
630
140
140
112
84
50
17
26
24
4
2
11
10
6
6
8
8
9
8
3
1
Number
of
Samples
275
191
726
128
948
620
1204
993
388
358
155
107
193
42
84
58
27
8
46
31
18
17
27
22
30
25
25
6
Average
System
Mean
[As]*
5.99
5.39
9.62
6.73
4.52
4.13
2.66
2.65
5.62
5.66
2.25
2.25
4.45
3.38
1.48
1.13
3.44
4.25
3.46
1.41
0.66
0.70
0.97
0.87
NA
NA
1.72
NA
Min.
System
Mean
[As]*
0.04
0.02
0.60
0.19
0.00
0.00
0.03
0.03
0.12
0.10
0.12
0.10
0.28
0.11
0.64
0.36
2.50
2.50
0.25
1.00
0.50
0.50
0.50
0.50
NA
NA
0.50
NA
Max.
System
Mean
[As]*
54.00
54.00
195.00
53.92
110.00
110.00
51.86
56.00
47.40
47.40
20.00
20.00
36.33
16.00
2.67
2.75
6.27
6.00
5.00
2.75
1.00
1.25
2.05
2.05
NA
NA
2.50
NA
Percent
Samples
ND
60.0
55.0
62.0
48.4
66.6
59.5
58.5
55.4
28.9
24.9
84.5
82.2
62.2
54.8
81.0
72.4
66.7
12.5
84.8
77.4
83.3
82.4
66.7
59.1
100.0
100.0
96.0
100.0
* Average, minimum, and maximum NTNCWS system mean arsenic concentrations in ug/L . The system
means are undefined if all samples in all systems are censored, and then the average minimum, and
maximum system means are not available (NA).
40
-------
New Mexico (16.4 percentage points), and Utah (13.6 percentage points). For each of these
States, subsetting the data excludes an adequate number of samples with reporting limits of 10.
For Illinois, the censoring percentage rose slightly from 69.9 % to 75.8 %. For the other three
States, reporting levels in the original data sets are generally lower than those in the original data
sets for Arizona, California, New Mexico, and Utah. This explains the smaller reductions in
percentage of samples that are censored for the Alaska, Minnesota, and Oregon. For CWS
surface water systems, subsetting reduces the percent of samples that are censored for six of the
eight States, with the most significant decreases also occurring in the States of Arizona (22.6
percentage points), California (7.8 percentage points), New Mexico (18.7 percentage points), and
Utah (13.6 percentage points). The percentage of censored CWS surface water samples rose for
the States of Illinois (13.4 percentage points) and Oregon (0.2 percentage points); these two
States had the highest levels of censoring among the eight States. For NTCNWS ground water
and surface water systems, subsetting reduced the percentages of censored samples in eight and
seven of the eight states, respectively. The exception was a slight rise for SW systems in Utah.
The largest reductions were for Arizona, where the percentage censored was reduced by 13.6
percentage points for ground water and by 54.2 percentage points for surface water (but from
only 27 samples).
The minimum and average system mean arsenic concentrations for both ground water and
surface water systems are generally a little lower for the data subsets than for the complete State
data sets. This difference is probably a result of the conventions that were applied to estimate
system mean arsenic concentrations. These conventions are presented in Section 6.1.1 of this
report. In cases where there are five or more detects that are not equal, a regression on order
statistics method (detailed in appendix A) was used to estimate the concentrations for the non-
detects by fitting a lognormal distribution to the detects. However, the convention that may have
the greatest impact is for the calculation of the system mean arsenic concentration when there are
four or fewer detects ( samples that were above the reporting limit) or when there are five or
more detects that are all equal. In this case, non-detects are treated as having an estimated
concentration of half the reporting limit. For example, consider a hypothetical system in
California with 12 samples, including 2 positive detects at concentrations of 2.5 and 2.8 |ig/L,
five non-detects at 2 |ig/L and five non-detects at 10 |ig/L. The samples with reporting limits of
10 were collected prior to 1995, while the positive detects and the samples with a reporting limit
of 2 |ig/L were collected after 1995. In accordance with the data convention described above, the
mean arsenic concentration for this system is 2.94 |ig/L when the entire data set is considered,
and is 1.47 |ig/L when the recent data subset is considered. Thus, as this example indicates, the
higher reporting levels associated with the complete data sets can result in higher individual
system mean concentrations. In turn, higher system mean concentrations result in higher State
mean arsenic concentrations. Because a sample with a lower reporting limit provides a more
accurate indication of the arsenic concentration in drinking water, system means calculated from
such samples are believed to be more accurate than system means calculated from samples with
higher reporting limits.
41
-------
Tables 4-7a (CWS systems) and 4-7b (NTNCWS systems) summarize the differences in
the number of systems that are represented in the complete State data sets relative to the number
of systems that are represented in the subsets of data. The highest reductions in coverage for
ground water CWS systems occurred for the States of Arizona, Illinois, and Oregon. However,
as shown in Table 4-6a, each of the data subsets for these States contains a large number of
systems. For surface water systems, the greatest reductions in coverage occurred in Arizona and
California. Again, even with the reduction in coverage, these data sets provide an adequate
amount of data for arsenic occurrence estimation. For NTNCWS systems the greatest reductions
in coverage were for Utah, which originally had the smallest number of NTNCWS systems in the
database, and for Arizona, which again provided an adequate amount of data after subsetting.
Table 4-7a
Relative Change in CWS System Coverage
State
AK
AZ
CA
IL
MN
NM
OR
UT
Groundwater
-6.75%
-58.23%
-10.59%
-30.68%
-3.94%
-2.44%
-45.80%
-19.57%
Surface Water
-2.75%
-28.26%
-20.72%
-9.71%
0%
0%
-3.73%
-7.89%
Table 4-7b
Relative Change in NTNCWS System Coverage
State
AK
AZ
CA
MN
NM
OR
UT
Groundwater
-6.43%
-59.47%
-12.23%
-0.63%
0%
-25%
-66%
Surface Water
-7.69%
-50%
-9.09%
0%
0%
-11.11%
-66.67%
42
-------
As a result of the association between reporting limits and time in the 8 States, most of
the following statistical analyses in Chapters 5, 6, and 7 were conducted using the subsets of data
for each State that cover the period when reporting limits were lower.
Multiple Samples for Individual PWS (Over Time and Space)
One of the unique aspects of the compliance monitoring data sets is that many of them
provide multiple samples for many PWS. There are multiple measurements at the same location
(POE) but different dates and times (temporal variability). There are sometimes measurements at
multiple POE locations (spatial variability). For the systems in the INTRA database, the POE is
identified and so the temporal and spatial variability can be separated (see Chapter 7). For other
systems without POE identifiers, it is not possible to distinguish the two types of variability.
Tables 4-8a (CWS) and 4-8b (NTNCWS) summarize the average, minimum, and
maximum number of samples per PWS in each State data set, by source water type. For the eight
states with data subsets having lower reporting limits, the data subset was used. In general, there
are more samples per system for surface water systems than for ground water systems. This is
consistent with expectations, because surface water systems are required to monitor more
frequently than ground water systems. In several States, some systems are represented by more
than a hundred samples. Within individual systems, some of these samples have been collected
on the same day, while others have been collected over many years. In addition, samples can
come from multiple POE in a system. In some States, these samples have been censored at
multiple reporting limits, while in other States, all samples are censored at the same reporting
level. Having multiple samples for individual PWS offers several benefits, but also presents
some challenges.
With multiple samples over time, it is possible to average samples in order to develop a
better estimate of system arsenic levels. In addition, having multiple samples for individual PWS
can support analyses of variability in arsenic concentrations over time and from location to
location within individual PWS. For example, the assessment of intra-system variability
presented in Chapter 7 is based on the analyses of arsenic occurrence distributions from location
to location within individual PWS. However, when systems are represented by different
numbers of samples, it may be necessary to estimate a single system level statistic for each
system so that arsenic levels are comparable from system to system. This aggregation ensures
that a system with 50 samples can be represented and compared to a system with only one
sample. In this report, samples for individual systems have been averaged using the procedure
discussed in Section 6.1.1.
43
-------
Table 4-8a
Summary of Numbers of Samples per CWS System for State Compliance Monitoring Data
State
AK
AL
AR
AZ
CA
IL
IN
KS
KY
ME
MI
MN
MO
MT
NC
ND
NH
NJ
NM
NV
OH
OK
OR
TX
UT
Ground Water Systems
Min. (N)
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
Mean (N)
1.6
7
1.8
2.9
7.8
1.5
1.4
4.6
4.6
2.5
2
2.4
1.1
3.6
6
1.1
2.4
3.7
4.2
1
4.2
2.3
1.8
2.1
4.9
Max. (N)
14
81
7
51
458
21
12
39
14
77
253
20
4
15
112
5
45
65
166
2
71
78
28
42
67
Surface Water Systems
Min. (N)
1
4
1
1
1
1
1
1
2
2
1
2
1
2
2
1
1
o
3
i
i
2
1
1
1
1
Mean (N)
4
15.1
4.1
15.7
7.3
2
4.4
6.7
13.2
3.6
1.8
2.3
1
9.4
17
1
5.6
11.1
4.9
1.1
11.5
1.6
4
6.4
19.9
Max. (N)
102
95
20
121
77
6
19
11
28
9
7
o
3
2
24
56
1
15
108
18
2
80
64
15
52
256
44
-------
Table 4-8b
Summary of Numbers of Samples per NTNCWS System
for State Compliance Monitoring Data
State
AK
AL
AZ
CA
IN
KS
MI
MN
MO
MT
NC
ND
NJ
NM
OR
TX
UT
Ground Water Systems
Min. (N)
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
Mean (N)
1.5
3.9
1.7
1.9
1.2
2.2
1.2
1.6
1.1
2
2.3
1
1.2
2.6
1.3
1.7
2.5
Max. (N)
6
11
13
28
8
9
11
6
12
3
9
1
7
27
3
5
6
Surface Water Systems
Min. (N)
1
8
1
1
1
3
-
1
1
-
4
1
2
1
1
1
6
Mean (N)
2.4
9.2
4
3.1
2.8
3
-
2.8
1
-
5.5
1
2.3
2.8
3.1
4.4
6
Max. (N)
6
10
7
6
5
3
-
5
1
-
7
1
o
3
6
5
8
6
45
-------
Water System Type
Many of the compliance monitoring data sets provide data on arsenic occurrence in
NTNCWS systems as well as CWS systems. In fact, as shown on Table 4-2, data sets for 17
States contain some data from NTNCWS systems, although one of these State data sets (for
Montana) contains information for only two systems. For 12 States, more than 100 ground water
NTNCWS systems are represented. Far fewer NTNCWS surface water systems are represented,
but the total population of NTNCWS surface water systems in the United States is quite small
(442 non-purchased NTNCWS systems). Information about NTNCWS systems is not available
in the other arsenic occurrence databases discussed in Section 4.2 and 4.3.
Point-of-Entry (POE) Data
The compliance monitoring data sets are unique from other data sets in that most of them
contain multiple data points for individual PWS systems. As discussed above, there can be
multiple measurements at the same POE but different dates or times, and/or measurements at
different POE for the same system.
For individual systems, compliance with the arsenic standard is measured at the POE into
the distribution system. A POE represents a well or entry point to the distribution system,
therefore a location where treatment may need to be installed if they have arsenic levels above
the MCL. Systems may have more than one POE. Generally larger systems have more POE;
also, individual POE or treatment plants may be supplied by a network of wells. A system with
multiple POE may need to install multiple treatment systems, depending on the contaminant
concentrations at each POE. Therefore, to estimate compliance costs, it would be ideal to know
the number of POE in each system, and link each sample result to a specific POE ID. However, a
complete inventory of individual POE is not available, and many of the compliance monitoring
data sets do not include POE ID.
POE ID are available for the following ten States:
Alabama
Arkansas
California
Illinois
Indiana
New Mexico
North Carolina
Oklahoma
Texas
Utah.
These data were used to assess intra-system variability in arsenic concentrations. The
intra-system variability estimation and results are discussed in Chapter 7 of this report.
46
-------
4.1.3 Building the AOED from SDWIS and State Compliance Monitoring Databases
Two separate databases comprise AOED, and these databases were developed from the
information contained in the individual State compliance data sets and SDWIS. These databases
are named GRAND and INTRA, and they were created in SAS. GRAND was designed to
support statistical evaluations and the development of national occurrence projections presented
in Chapters 5 and 6, and it contains the data sets of all 25 States. INTRA was developed to
support the assessment of intra-system variability presented in Chapter 7, and it includes the ten
data sets which include POE identifiers. The GRAND database has 131,383 records, and INTRA
has 88,855 records. For the eight states where subsets of data for years with lower reporting
limits were used, the entire original database (all years) is included and an additional copy of the
subset data is also included. For example, for California, the original database is the set of
records with ST_GROUP = "CA," and the subset database is the set of records with ST_GROUP
= "CA2." All the records in the data subset appear twice, first with ST_GROUP = "CA" and
second with ST_GROUP = "CA2." After excluding 5 outlier values with concentrations above
1000 |_ig/L and excluding the unsubsetted original data for the eight states, this leaves 76,973
records in GRAND and 49,081 records in INTRA that were used for the analyses in the
remaining chapters of this report, including the regional and national occurrence analyses and the
intra-system variability analyses. Each record in these databases correlates with a finished
drinking water sample from a PWS. Appendix D-2 provides a list of the variables in these
databases, and the characteristics of those variables. All of the records in INTRA are included in
GRAND, although INTRA include two additional variables, which are the POE identifier and the
POE type. INTRA was developed to improve computational efficiency for the intra-system
analyses; it includes only information that is relevant to those analyses, and was developed with
minimal additional effort.
Initially, IS SI compiled the raw State compliance monitoring data sets in an Access
database. The initial data conditioning processes are described in Appendix D-3. Subsequently,
the Access database was converted to a SAS format, and the GRAND and INTRA databases
were created. In addition, raw compliance monitoring data sets were received for the States of
Alabama, Arkansas, Illinois, Indiana, New Mexico, Oklahoma, and Oregon after the Access
database had been converted to SAS, and these compliance monitoring data sets were loaded
directly into SAS. In August and September, 2000, updated and revised compliance monitoring
data was received for the States of Alabama, California, Illinois, New Mexico, North Carolina,
and Texas, and used to replace the previous data for those states. Also in September, 2000, a
previously submitted, but erroneously omitted updated database from Utah was used to replace
the previous data for Utah.
Simply stated, the data conditioning process standardized the coding of variables from
State to State. For example, different States reported sample collection dates in different
formats. Through the data conditioning process, dates were standardized to a consistent format.
Similarly, result flags were initially inconsistent. Some States reported ND (not-detect), others
LT (less than), some N and D, and some marked non-detects with the "<" symbol. After
confirming the meaning of each flag with the data contact in the State, these flags were
standardized to N and D for non-detects and detects. The result and the reporting limit variables
were standardized to |ig/L, because some States provided data in jig/L and others reported data in
47
-------
mg/L. PWS source type was standardized to GW and SW (systems coded as ground water under
the influence of surface water were coded to surface water), and type of water system was
standardized to CWS and NTNCWS. Data for other types of water system were excluded from
the data base.
SDWIS was an important tool for the development of the arsenic databases, and it was
used in a variety of ways. Each of the suitable State compliance monitoring databases provided
SDWIS PWSID numbers. By linking the State compliance monitoring data to SDWIS by PWSID
number, we were able to complete missing data where necessary, update variables which could
change over time (such as population served or if a PWS was active or inactive), and check for
discrepancies between the compliance monitoring data and the SDWIS data. The SDWIS
Baseline for 1998 was used for the analyses in this report. When discrepancies in the type of
water system (CWS or NTNCWS), source type (GW or SW), or population served were
identified between State data and SDWIS data, the State data were corrected to reflect the
SDWIS data. By doing so, the databases were corrected in a manner that was consistent from
State to State. The SDWIS database was used to find and remove from the database records for
suppliers of purchased water. Inactive systems were also removed. In relation to the total number
of records in the database (131,383), about 10 % required correction. For 520 systems, SDWIS
identified these as having purchased water, so these systems were excluded from the database.
Also excluded were a total of 1,798 systems that were not in the SDWIS database (this includes
relatively new systems not included in the 1998 SDWIS update). A total of 18,280 records were
deleted as either being for purchased water or for a system not in SDWIS. For 172 systems,
SDWIS and State data sets disagreed about the type of water system (CWS versus NTNCWS);
and for 119 systems, SDWIS and State data sets disagreed about the system source water type
(GW or SW).
Several other conventions were applied during the development of the GRAND and
INTRA databases, and these conventions are discussed below:
1. Sample analyses conducted prior to 1980 were deleted. This is roughly the year
that new analytical techniques had become widely available and less precise
colorimetric analysis was phased out.
2. Analyses with no PWSID number, or no analytical result, or from systems
identified in SDWIS or in the supplied database as either inactive, abandoned, or
as suppliers of purchased water,10 were deleted.
3. Analyses for water samples identified as either being finished, or from sources not
subject to treatment, or for samples measured after treatment were included in the
database as finished water. For ground water, samples identified as raw samples
or as being measured before treatment were included in the database, and were
included in the data analyses. For surface water, samples identified as raw
samples or as being measured before treatment were excluded from the database,
10 Purchased water systems do not withdraw the water that they supply to their customers directly from a ground
water or surface water source, but instead they purchase it from another water system.
48
-------
and, hence, from the data analyses. Briefly, the "raw" ground water samples were
included since the treatments used are not expected to impact the arsenic
concentration levels in ground water, but may have an impact for surface water.
4. Analyses with results greater than 1000 |ig/L were left in the database but
excluded from the analysis, because these values were believed to be the result of
data entry or data conversions errors, rather than representative of actual arsenic
concentrations in drinking water. As a result of this convention, six arsenic values
were deleted from the analyses, but included in the database. These were all for
CWS ground water systems. There was one sample from a public water supply
system in New Mexico with a value of 189,000 |ig/L; this record was in both the
all year set and the lower reporting limit subset. There were five samples from
Texas with values of 18,500, 18,500, 14,500, 4,800, and 3,400 |ig/L.
5. Arsenic values reported as "zero" or "ND" were considered to represent an
analytical result below the reporting limit. If the State provided reporting limits
for some samples, but did not provide reporting limits for other non-detects, such
as the zero values, then the reporting limit for those records was estimated as the
modal (most commonly occurring) reporting limit for the state and associated
subset of years. For the 17 states where a subset with lower reporting limits was
not defined, the modal reporting limit for all the non-detects was used. For the
other eight states, the estimated reporting limit for the all years subset was the
modal reporting limit for all the non-detects, but the estimated reporting limit for
the lower reporting limit subset of years was the modal reporting limit for all the
non-detects reported in those years. For example, for California the estimated
reporting limits used to fill in unknown reporting limits were 10 |ig/L for the
original 1980 - 2000 subset (ST_GROUP = "CA") but was 2 |ig/L for the 1995 -
2000 subset (ST_GROUP = "CA2"). In California, most of these filled in
reporting limits were for arsenic values reported as "zero."
6. If the State did not disclose the reporting limits for any samples, and reporting
limits could not be determined based on conversations with appropriate State
representatives, reporting limits were assigned based on where the majority of the
lowest measured results clustered. The later convention was applied to all non-
detect data from the States of Alabama and Oregon. For the State of Indiana,
reporting limits were assigned that correlated with laboratory analysis methods.
Appropriate reporting limits were provided by the State representative.
7. Sample results that were non-detect with reporting limits greater than 10 ppb were
deleted. This convention only affected Michigan, in which arsenic results were
associated with four report limits (0.3, 1.0, 2.0, and 50 ppb). Removing arsenic
results associated with a reporting limit of 50 ppb excluded 21% of Michigan
data.
49
-------
8. The State of Missouri reported only positive results. ISSI contacted the State, and
the Missouri Department of Health provided a list of the PWSIDs for all systems
with non-detect measurements for arsenic during the same three year monitoring
time period as the positive results (i.e., detects). Furthermore, the State
representative indicated that these systems were non-detect at the reporting limit
of 1 ng/L. These data were then combined with the positive results.
9. The State of Alaska reported that measurements made in 1990 or earlier are now
considered to be unreliable, and so these records were deleted from the database.
The State of Alaska also reported that until 1999, all their systems were reported
to SDWIS as CWS systems, even though the systems in Alaska included CWS,
NTNCWS and transient non-community water supply systems. The SDWIS 1998
baseline reports no NTNCWS systems for Alaska. The state provided a database
giving the type of water system for each PWSID. This database was used to adjust
the type of water system variable in GRAND and INTRA and also in the SDWIS
database used for the all analyses described in this report.
4.2 Comparison Databases
Four databases are available for use as comparison tools on a national basis: the NAOS,
USGS, NIRS, and Metro databases. All of these databases are national in scope, but the
characteristics of these databases differ substantially. Likewise, the survey methods that were
used to collect samples for NAOS and NIRS were quite different, and samples that are
represented in USGS and Metro databases were not collected in accordance with a specific
survey method. Each of these databases is described below, and comparisons of national arsenic
occurrence estimates based on the AOED, NAOS, USGS, NIRS, and Metro databases are
presented in Chapter 6.
For California, a comparison can be made between the California systems in the AOED
and the ACWA databases. The results of a comparison of the California arsenic occurrence
estimates based on the AOED and estimates based on the ACWA databases is also presented in
Chapter 6.
4.2.1 National Arsenic Occurrence Survey (NAOS) Database
The Water Industry Technical Action Fund (WITAF)11 sponsored the National Arsenic
Occurrence Survey (NAOS), which was a nationwide arsenic survey (Frey and Edwards, 1997).
The NAOS was designed to be quantitatively representative of different source water types,
system sizes, and natural occurrence patterns (based on a derived natural occurrence factor -
discussed below), and to have a low detection limit (0.5 |ig/L).
11 WITAF includes the following organizations: American Water Works Association, National Association of Water
Companies, Association of Metropolitan Water Agencies, National Rural Water Association, and National Water
Resources Association..
50
-------
NAOS was based on a representational survey design. PWS were selected from within
three representational groups, including:
source type (ground water and surface water);
system size (small: 1,000 to 10,000 people served; and large: >10,000 people served);
and,
natural occurrence factor (NOF) level (which includes consideration of geographic
region).
The NAOS survey design used the NOF to qualitatively represent the relative probability
of arsenic occurrence in water sources. For each State, separate ground water and surface water
NOF levels were estimated. USGS's water quality database (WATSTORE) was used to derive
surface water NOF levels for all States except Indiana, and ground water NOFs for 35 States. The
remaining State source types were not represented in WATSTORE. Metro data was used to
confirm the NOFs based on WATSTORE and to derive the missing NOF designations.
NOF levels were designated as low, medium, or high, and were based on total scoring in
four criteria. These elements were:
1) the probability of left censored (below detection limit) data;
2) the probability of measured observations below 5 |ig/L;
3) the probability of measured observations above 20 |ig/L; and
4) local interest in arsenic in source water, as indicated by the total number of samples.
For each criterion, potential scores were 5, 15, or 25 points. Final NOF assignments for
ground water and surface water sources were based on the sum of the scores for these criteria.
These NOF levels served as a basis for identifying distinct regional arsenic occurrence patterns.
The seven regions that were identified, and the States included in each region, are shown below:
Table 4-9
States in the Seven NAOS Regions
Region
New England
Mid-Atlantic
Southeast
Midwest Central
South Central
North Central
Western
States in Region
CT, NH, NJ, NY, MA, ME, RI
DE, KY, PA, MD, NC, SC, VA, WV
AL, FL, GA, MS, TN
OH, IA, IL, IN, MI, MN, WI
AR, CO, KS, LA, MO, ME, MM, OK,
TX
MT, ND, SD, WY
AK, AZ, CA, HI, ID, NV, OR, UT, WA
51
-------
NAOS was designed to include approximately 800 samples that were selected to
proportionally represent the survey design criteria (system sizes, source type, and geographic
region). A number of samples came from the same PWS, but from different facilities or wells
within the PWS. Systems were mailed a survey that requested their participation. Large systems
were selected from the Water Industry Database, and small systems were selected from the
Federal Reporting Data System(FRDS). Approximately the same number of small and large
systems were sampled. Based on source water type distributions contained in FRDS, the ratio of
surface water to ground water systems was 55:45 for large systems, and 30:70 for small systems.
To achieve geographic representativeness, sample sites were selected from within each region
(based on availability of sites within each region).
Sample kits with instructions were mailed to the selected utilities with a questionnaire on
specific details regarding the sample. While 809 sample kits were mailed out, 517 samples were
submitted to the investigators by water utilities (a 63.9 percent response rate). The response rate
for large utilities (70 percent) was slightly higher than for small ones (58 percent). The
investigators found that comparable portions of large and small systems responded within each
survey stratum. Samples were collected from the utilities' raw water sources. During data
handling, raw water results were multiplied by a removal efficiency factor associated with the
treatment train in place at the utility to calculate the likely finished water arsenic concentration.
There were a total of 435 predicted finished water arsenic level samples in the NAOS database
(161 surface water samples and 274 ground water samples). There were an additional 54 surface
water samples that did not have predicted finished water arsenic levels because there was no
treatment information available. Some of the PWS in the survey had multiple water sources.
Based on responses to the questionnaire, the investigators concluded that the participating
utilities did not introduce bias into the survey results by selecting sources with known arsenic
concentrations.
The analytical method used for the NAOS analyses had a detection limit of 0.5 |ig/L, and
therefore, the data set has a lower percentage of non-detects than NIRS and many of the early
data sets. In ground water, arsenic was detected in 58% of samples, and in surface water,
detectable arsenic levels were reported in 73% of samples.
One advantage of the NAOS data set is that all of the data samples were processed by a
single laboratory, which followed strict quality assurance protocols. However, the data set has
two potential drawbacks, one major and one minor. The minor drawback is that the NAOS data
set does not provide PWS identification numbers, so it is difficult to identify the facilities which
may be represented by more than one sample. The major drawback is that the estimated finished
concentrations are not direct measurements of finished arsenic concentrations. There is a
potential for true finished concentrations to differ from the estimated finished concentrations.
The NAOS database was used as a comparison tool to check arsenic occurrence projections
developed from the AOED.
4.2.2 USGS Arsenic Databases
The USGS has collected data on arsenic concentrations in ground water from locations
throughout the United States. These data were collected from a variety of wells, some of which
52
-------
are used for supply of drinking water (about 10% of samples), and others for research,
agriculture, industry and domestic supply, as a part of a variety of Federal, State, and local
projects (Focazio et a/., 2000). These data were not collected specifically to develop national
estimates of arsenic in drinking water; however, the USGS databases provide approximately
20,000 samples (from approximately 20,000 locations throughout the country; 1 sample per well
or spring) that are potentially representative of ground water systems. Analyses were performed
by hydride-generation and atomic-adsorption spectrometry and have a consistent reporting limit
of 1 i-ig/L. Thus, the USGS data provide a significant amount of information about arsenic
concentrations in ground water.
The USGS data is included in four distinct databases, which were derived from existing
databases. Under an interagency agreement with the USEPA, USGS used some of these
databases to project national estimates of arsenic occurrence in community water supplies with
ground water sources. These estimates are discussed in Chapter 6 of this report, and the four
databases which USGS developed are described below.
The Public Supply Database includes data derived from SDWIS on all ground, surface,
and purchased water CWS. The sources of the CWS were reviewed and characterized to identify
systems that are served partially or entirely with ground water. Systems that were at least
partially served by ground water sources were retained, and those that were totally dependent on
surface water (including purchased surface water) were excluded from the database.
The USGS Arsenic Point Database includes arsenic data from the USGS National
Water Information System (NWIS). It includes the results of analyses of filtered water samples
from 20,000 ground water wells and springs throughout the United States from 1973 to 1998.
Codes are provided which allow separation of potable and non-potable water sources. Where
NWIS contained multiple samples from individual wells, the most recent sample was included in
Arsenic Point Database. This database also provides information on water use, well
construction, and basic water quality parameters.
The USGS Arsenic Database of Selected Counties includes counties in which five or
more arsenic samples were present in the Arsenic Point Database, and counties that include five
or more arsenic values through a process of radial extrapolation. For counties with less than five
arsenic values, the radial extrapolation procedure ascribed all samples within a 50 kilometer (31
mile) radius of each county center to that county. Counties with five or more arsenic analyses
within the search radius were included in the database, and the arsenic samples identified by the
radial extrapolation were assumed to be representative of the county. There were a total of
17,496 samples for 1,528 counties in this database. When these data were associated with the
Public Supply Database, this database represented 76% of all large systems and 61% of all small
systems.
The fourth USGS database is the USGS Arsenic Database of All Counties. This
database draws upon information in the Selected Counties database, and the Ground Water Atlas
of the United States. USGS used the Ground Water Atlas to identify the major aquifers in each
county, and then identified the type of aquifer that ground water suppliers in each county
withdraw water from. When median arsenic concentrations were calculated for counties in
53
-------
specific aquifers, they were extrapolated to other counties without arsenic data that use these
aquifers. Thus, arsenic values were extrapolated for the remaining counties in the United States.
From the USGS Arsenic Database of Selected Counties, the USGS developed the arsenic
occurrence estimates for ground water that are presented in Chapter 6. USGS estimated the
percent exceedances for each county by calculating the percentage of data points in each county
(with 5 or more data points) exceeding specific arsenic concentrations, from 1 |_ig/L to 50 |_ig/L.
Then USGS associated the percentages for each county with the number of systems in these
counties (from the USGS public supply database). This information was aggregated for all of the
appropriate counties to derive the national estimates for ground water systems.
4.2.3 National Inorganics and Radionuclides (NIRS) Database
The USEPA's National Inorganics and Radionuclides Survey (NIRS) is a national
database of 983 samples of finished drinking water from community ground water systems. The
survey was designed to be a nationally stratified proportional probability sample, and it was
stratified by system size to be nationally representative (Longtin, 1988). Two percent of the
PWS in the United States in each size stratum were sampled. Local utilities sent field-preserved
samples collected between 1984 and 1986 from 48 States and Puerto Rico to USEPA
laboratories for analysis. Because most of the ground water systems in the United States are
small, the majority of systems represented in NIRS are very small.
NIRS contains sufficient information to support arsenic occurrence projections, and the
database contains all of the critical data elements. The NIRS data, however, are censored at 5.0
|ig/L, so the database does not provide information through the full regulatory range of interest.
In addition, because of the relatively high reporting level, 95% of the arsenic results are censored,
and therefore, projections of arsenic occurrence at levels below 5.0 |ig/L based on NIRS data
includes a high level of uncertainty. For large and very large strata, 100% of the results are
censored.
4.2.4 Metropolitan Water District of Southern California (Metro) Database
The Metro database contains 144 arsenic samples from PWS selected from American
Water Works Association (AWWA) Water Industry Database. The database generally represents
large PWS (> 10,000), and the utilities represented serve more than a third (36%) of the U.S.
population. This database contains 57 ground water and 87 surface water sample results. The
detection limit was 0.5 |_ig/L and detected concentrations ranged from 1 |ig/L to 39 |ig/L in
ground water and 1 |ig/L to 5 |ig/L in surface water. Most of the 144 samples contain some
detectable arsenic; only 33% of the Metro sample results are censored.
The Metro data is a potentially important source of arsenic occurrence information,
although it, like the NAOS database, also has limitations. Metro data represents a subset of
facilities for which there are few samples in NIRS, but it provides little information about arsenic
occurrence in small water systems. Metro data also provides arsenic occurrence information
throughout the regulatory range of interest, because of the low detection level. However, this
data set does not include PWS identification numbers for the sampled facilities. Therefore, the
54
-------
Metro database cannot be used to estimate arsenic occurrence together with other databases,
where the information in the two databases may overlap.
4.3 Other Databases
Several other databases that provide arsenic occurrence information are available;
however, these databases are not used in this report. These databases include the 1969 and 1978
Community Water Supply Surveys (CWSS), the Rural Water Survey (RWS), the National
Organics Monitoring Survey (NOMS), the Western Coalition of Arid States (WESCAS)
database, and the Association of California Water Agencies (ACWA) databases. These
databases, with the exception of WESCAS and ACWA, were not used in this report because the
data that they contain are considered to be too old to accurately represent current conditions for
arsenic occurrence. Older arsenic data may not represent current conditions for several reasons.
The samples may have been analyzed with a less accurate laboratory method, the raw water
sources may have changed, and treatment systems may have been installed. For example,
filtration treatment added to surface water systems to comply with the Surface Water Treatment
Rule would tend to decrease arsenic concentrations through incidental removal of arsenic.
WESCAS data was not used because the data that are contained do not necessarily represent
arsenic levels at individual PWS, and because the data conventions and handling appear to have
been inconsistent from State to State. ACWA data were not used for these data analyses
because compliance monitoring data was available for the State of California. Section 6.3
presents a comparison between the AOED occurrence estimates and the results of two studies of
the ACWA data (Kennedy/Jenks Consultants, 1996, and Saracino-Kirby Inc., 2000). These
databases are discussed below.
4.3.1 1969 Community Water Supply Survey
The U.S. Public Health Service conducted the 1969 CWSS to assess water supply
facilities and drinking water quality in the nation. Samples were collected from randomly
selected sites in the distribution systems of the participating PWS. A total of 969 finished water
samples were collected from 678 ground water supplies, 109 surface water supplies, and
182 mixed sources (purchased water or unspecified source). Of these water samples, analytical
results for arsenic were available for 673 ground water samples, and 106 surface water samples.
Ninety-five percent of the ground water samples and 92 percent of the surface water samples
were censored. Only 33 ground water detections and 9 surface water detections occurred. The
results of the 1969 CWSS survey are summarized in Appendix C. Because this data set was
collected prior to 1980, and the analytical results it contains may be less accurate than more
recent data sets, these data will not be used to project arsenic occurrence and exposure estimates.
4.3.2 1978 Community Water Supply Survey
USEPA conducted a second CWSS in 1978 to determine the occurrence of organic and
inorganic compounds in public water supplies. Approximately 500 water supplies provided
drinking water samples. As a result of analytical anomalies and difficulties, the 1978 CWSS
only contains 259 ground water and 94 surface water analytical results for arsenic. From each
PWS, one to five samples of raw, finished, and/or distribution water were collected from each
55
-------
supply sampled. Due to reporting inconsistencies, distributional and finished sample results were
averaged together and the raw water data were not used. A total of 49 non-censored arsenic
results and 3 non-censored surface water results were observed. The 1978 CWSS arsenic
occurrence results are summarized in Appendix C. Like the 1969 CWSS, this data set is
composed primarily of censored data: 82 percent of ground water sample data and 92 percent
surface water sample results are censored. This data set is also considered to be less accurate
than data collected after 1980, and will not be used to estimate arsenic occurrence and exposure
for this study.
4.3.3 Rural Water Survey
Between 1978 and 1980 the RWS was conducted to evaluate the status of drinking water
in rural America. A total of 71 ground water and 21 surface water samples were analyzed for
arsenic from the 648 PWS surveyed. A total of 23 non-censored ground water and 2 non-
censored surface water arsenic results were observed. The arsenic occurrence data contained in
the RWS are summarized in Appendix C. Sixty-eight percent of the ground water analytical
results and 92 percent of surface water data were censored. The RWS data are deemed
insufficient to project arsenic occurrence and exposure estimates because they were collected
prior to 1980.
4.3.4 National Organics Monitoring Survey
In 1976 and 1977, USEPA conducted NOMS to provide data to support the development
of MCLs for organic compounds in drinking water. Trace elements were analyzed in finished
water samples from 113 PWS; of this data arsenic analytical results were provided for 15 ground
water and 86 surface water samples. A summary of arsenic data from NOMS is presented in
Appendix C. A total of 6 non-censored ground water and 19 non-censored surface water arsenic
results were observed. Sixty percent of ground water analytical results and 78 percent of the
surface water analytical samples were censored. These data are deemed insufficient to project
arsenic occurrence and exposure estimates because they were collected prior to 1980.
4.3.5 Western Coalition of Arid States Research Committee Arsenic Occurrence Study
In 1997, the Western Coalition of Arid States (WESCAS) Research Committee
conducted an arsenic occurrence study. The study consisted of arsenic data obtained from
Arizona, New Mexico, and Nevada. The primary purpose of this study was to collect data on
low levels of arsenic in ground water, particularly in small systems. California data were also
provided to WESCAS by ACWA for the study, but were not included in this database. The
WESCAS data have been manipulated and aggregated, and the data points that are contained do
not necessarily represent arsenic occurrence at individual PWS. For some States, data were re-
defined to median values by area/or provider. However, the Arizona data were aggregated by
county. In addition, PWS identifications were not provided for data from Arizona and Nevada.
Because these data have been manipulated and are not comparable to other data sources, and
because relatively comprehensive compliance monitoring data are available for Arizona, New
Mexico, and Nevada, the WESCAS data were not used to project arsenic occurrence estimates
presented in Chapter 6.
56
-------
4.3.6 Association of California Water Agencies Database (ACWA)
Association of California Water Agencies (ACWA) (Kennedy Jenks Consultants, 1996)
conducted a survey of low level arsenic occurrence in the State of California to determine the
potential impact of a revised arsenic standard upon California water supply systems. More than
1500 samples (1378 ground water and 166 surface water samples) were collected between 1992
and 1994, and these analyses had detection levels of 0.1 to 1 |ig/L. Arsenic was present in a
greater percentage of ground water samples than surface water samples; 28 percent of the ground
water samples were censored, and 52 percent of the surface water samples were censored.
Arsenic concentrations in ground water were slightly higher than in surface water. The
maximum concentrations in these samples were 52 and 30 |ig/L, respectively, for ground water
and surface water samples. Most of the systems represented in this database are medium, large,
or very large PWS systems located in the southern part of the State. The survey also included
information from selected ACWA members and also from Central and West Basin Municipal
Water Districts and Southern California Water Company. This database was not used to develop
occurrence estimates because compliance monitoring data were available for the State of
California. The ACWA database does not provide PWSID numbers for each of the systems that
it contains, and therefore these data cannot be linked to SDWIS. Table 4-10 identifies the
percentage of ground water and surface water sources in California that will require treatment to
satisfy the listed MCLs.
Table 4-10
Estimated Percentage of California Water Supplies Impacted by MCLs
Arsenic MCL (mg/L) Surface Water Plants Ground Water Sources
1 58% 84%
2 15% 56%
5 1% 19%
10 <1% 6%
20 <1% 3%
Adapted from Kennedy Jenks Consultants, 1996.
An updated study of California data from ACWA was carried out in 2000 by Saracino
and Kirby, Inc. (ACWA, 2000).
57
-------
This page intentionally left blank
58
-------
5. Arsenic Occurrence Patterns in the United States
This section provides a discussion of the distribution of arsenic in drinking water in the
United States. Patterns of arsenic occurrence are discussed with respect to system water source
type, system size based on population served, system classification, and from region to region in
the United States. The data discussed in this section come from the AOED. The statistical
techniques applied to identify patterns in the available data are described, as are the results of
these analyses. In addition, the results of other relevant analyses that address patterns in arsenic
occurrence data are discussed and used to verify these analyses. Finally, these sections explain
the basis for decisions that were made with regard to data stratification for the development of
national arsenic occurrence estimates.
5.1 Stratification by Source Water Type
Previous arsenic occurrence studies (Frey and Edwards, 1997; Wade Miller, 1992 and
1989) have stratified drinking water systems on the basis of source water type into ground water
and surface water systems. This stratification is appropriate because these studies indicated that
arsenic concentrations in ground water and surface water differ. Therefore, source water type is a
source of data heterogeneity. Stratification on the basis of source water type may also facilitate
analysis of regulatory impacts, because ground water and surface water systems are regulated
separately. Typical ground water and surface water systems differ in entry point configuration
and treatment train (SAIC, 1999). Therefore, the occurrence analyses presented in Chapter 6
have been stratified on the basis of source water type.
As Tables 5-la (CWS systems) and 5-lb (NTNCWS systems) show, distributions of
system mean arsenic levels in ground water and surface water systems differ in each State in the
AOED database. The system means were calculated using the regression on order statistics
method detailed below in section 6.1.1 and appendix A. In most of the 25 States for which
compliance monitoring data are available, mean arsenic concentrations are higher in ground
water systems than in surface water systems. For example, in Alaska ground water CWS
systems, the average system mean arsenic level was 4.11 |ig/L, while in surface water CWS
systems, the average system mean arsenic level was 1.29 |ig/L. Similarly, in Alaska ground
water NTNCWS systems, the average system mean arsenic level was 5.39 |ig/L, while in surface
water NTNCWS systems, the average system arsenic mean level was 1.13 |ig/L. Arizona is
another example where for ground water CWS systems, the average system mean arsenic level
was 9.53 |ig/L, while for surface water CWS systems, the average system mean arsenic level was
4.68 |ig/L. In addition, while distributions of arsenic in ground water and surface water are
relatively similar at the 25th percentile in many States, the 75th percentiles of the system mean
arsenic levels are higher in ground water than surface water systems in 20 of 25 States for CWS
systems. For NTNCWS systems, the 75th percentiles of the system mean arsenic levels are
higher in ground water than surface water systems in 8 of the 10 states with both ground and
surface water systems.
59
-------
Table 5-la
Distributions of System Means for Community Water Systems
State
AK
AK
AL
AL
AR
AR
AZ
AZ
CA
CA
IL
IL
IN
IN
KS
KS
KY
KY
ME
ME
MI
MI
MN
MN
MO
MO
MT
MT
NC
NC
Source
Type
GW
SW
GW
SW
GW
SW
GW
SW
GW
SW
GW
SW
GW
SW
GW
SW
GW
SW
GW
SW
GW
SW
GW
SW
GW
SW
GW
SW
GW
SW
Systems
(N)
304
106
263
68
371
76
279
33
1224
176
750
93
648
51
506
101
88
150
109
29
644
33
829
23
773
89
484
47
1735
169
Minimum
0.01
0.13
0.24
0.43
2.50
2.50
0.28
1.90
0.00
0.24
0.01
0.30
0.00
0.25
0.19
0.50
0.67
0.79
0.09
0.24
0.01
0.04
0.04
0.63
0.03
0.01
0.08
0.32
0.19
1.22
25th
Percentile
0.35
0.48
0.53
0.60
2.50
2.50
2.20
2.86
1.08
0.96
0.19
0.57
0.03
0.50
0.90
0.75
0.88
1.44
0.60
0.53
0.47
0.18
0.55
0.77
0.24
0.07
0.56
0.60
2.55
2.20
Median
0.90
0.73
0.68
0.68
2.50
2.50
5.00
3.40
1.80
1.30
0.48
0.70
0.09
0.50
1.57
0.90
1.25
1.70
0.83
0.67
1.55
0.33
0.94
0.85
0.43
0.17
0.75
0.89
3.26
2.85
Mean
4.11
1.29
0.73
0.70
2.52
2.52
9.53
4.68
4.20
2.38
2.11
0.75
0.26
0.68
2.64
1.11
1.45
1.78
3.06
1.03
5.31
0.97
2.77
0.88
0.77
0.38
1.81
1.60
3.52
3.01
75th
Percentile
4.40
1.00
0.83
0.79
2.50
2.50
11.00
4.93
4.00
1.77
1.00
0.85
0.24
0.50
3.27
1.25
1.70
2.06
2.25
0.85
6.73
0.60
2.30
0.94
0.71
0.41
1.25
1.97
4.16
3.60
Maximum
61.88
14.33
7.00
1.45
7.00
3.67
101.60
15.25
99.00
39.25
59.10
3.18
7.10
4.00
65.14
3.64
4.50
4.86
53.00
6.70
89.00
8.99
65.82
1.25
34.78
3.60
45.75
7.82
30.17
6.50
60
-------
Table 5-la (continued)
Distributions of System Means for Community Water Systems
State
ND
ND
NH
NH
NJ
NJ
NM
NM
NV
NV
OH
OH
OK
OK
OR
OR
TX
TX
UT
UT
Source
Type
GW
SW
GW
SW
GW
SW
GW
SW
GW
SW
GW
SW
GW
SW
GW
SW
GW
SW
GW
SW
Systems
(N)
197
19
504
37
438
29
559
29
221
31
875
139
446
210
316
129
3105
326
263
35
Minimum
0.03
0.20
0.44
1.89
0.03
0.37
0.06
0.18
0.17
0.07
0.59
1.28
0.11
0.14
0.04
0.22
0.10
0.66
0.05
0.18
25th
Percentile
0.57
0.70
2.31
2.95
0.23
0.67
0.67
0.38
2.06
0.46
2.33
2.24
0.88
0.53
0.45
0.62
0.91
1.27
0.50
0.47
Median
1.65
1.00
3.47
3.63
0.49
1.01
1.75
0.60
5.00
1.44
3.54
2.75
1.52
0.83
1.09
1.07
1.47
1.50
1.00
0.84
Mean
4.85
1.11
6.06
3.85
0.92
1.13
3.81
1.03
11.86
3.38
4.32
3.07
3.01
1.14
2.77
1.45
2.58
1.69
2.89
1.94
75th
Percentile
4.20
1.40
4.76
4.28
1.03
1.30
4.37
1.00
13.00
3.80
5.21
3.50
3.00
1.26
2.73
1.81
2.20
1.85
2.63
1.63
Maximum
51.40
2.40
107.90
10.00
14.00
2.81
57.72
4.32
150.00
39.00
42.97
14.50
78.45
36.35
56.00
10.66
86.85
6.32
55.49
20.56
61
-------
Table 5-lb
Distributions of System Means for Non-Transient Non-Community Water Systems
State
AK
AK
AL
AZ
AZ
CA
CA
IN
KS
KS
MI
MN
MN
MO
MT
NC
NC
ND
ND
NJ
NJ
NM
NM
OR
TX
TX
UT
Source
Type
GW
SW
GW
GW
SW
GW
SW
GW
GW
SW
GW
GW
SW
GW
GW
GW
SW
GW
SW
GW
SW
GW
SW
GW
GW
SW
GW
Systems
(N)
131
24
31
77
2
330
10
538
62
2
230
630
6
190
2
562
8
20
6
758
o
6
140
8
84
580
27
17
Minimum
0.02
0.36
0.50
0.19
2.50
0.00
1.00
0.00
0.33
1.03
0.02
0.03
0.50
0.50
11.00
0.12
1.50
0.03
0.20
0.01
0.50
0.10
0.50
0.10
0.16
0.66
0.11
25th
Percentile
0.32
0.65
0.50
1.18
2.50
0.41
1.00
0.00
0.85
1.03
0.33
0.45
0.50
0.50
11.00
0.73
2.00
0.20
0.40
0.19
0.50
0.72
0.50
0.51
0.95
1.17
0.37
Median
0.87
0.85
0.50
3.00
4.25
1.50
1.00
0.01
1.53
1.40
1.80
0.90
0.50
0.50
21.83
1.35
2.50
0.89
0.70
0.48
0.50
3.38
0.56
1.16
1.53
1.58
1.00
Mean
5.39
1.13
0.50
6.73
4.25
4.13
1.41
0.24
2.12
1.40
4.66
2.65
0.70
0.55
21.83
2.02
2.35
5.36
0.68
1.41
0.80
5.66
0.87
2.25
2.69
1.84
3.38
75th
Percentile
7.00
1.64
0.50
7.00
6.00
4.20
1.58
0.07
2.40
1.77
6.00
2.50
0.96
0.50
32.67
2.43
2.50
4.38
0.90
1.31
1.40
7.50
1.13
2.51
2.28
2.18
4.50
Maximum
54.00
2.75
0.63
53.92
6.00
110.00
2.75
23.00
14.10
1.77
76.00
56.00
1.25
5.20
32.67
39.53
3.29
47.60
1.20
22.00
1.40
47.40
2.05
20.00
70.00
6.84
16.00
62
-------
5.2 Stratification by System Size
An important question is whether or not public water system size, based on population
served, is a determinant of average system arsenic concentrations. If arsenic concentrations are
associated with system size, then it may be appropriate to stratify the data on this basis when
developing occurrence estimates. However, if arsenic concentrations are not associated with
system size, incorrectly stratifying the data by system size could reduce the accuracy of the
arsenic occurrence estimates.
The earlier USEPA occurrence estimate (Wade Miller, 1992) considered this question,
but did not stratify based on system size because there were too few detects in the NIRS, CWSS,
NOMS, and RWS data that were used for this estimate to stratify on this variable. However,
Frey and Edwards (1997) stratified their data into small systems (those serving 1,000 to 10,000
people) and large systems (those serving more than 10,000 people).12 Thus, additional analyses
were conducted to evaluate if arsenic concentrations are associated with system size. The
following section describes the analyses that were conducted on arsenic occurrence and system
size using the AOED data. The system size categories of interest, based on population served,
are:
25-100;
101-500;
501-1,000;
1,001-3,300;
3,301-10,000;
10,001-50,000; and
50,001 or greater.
First, we visually compared the distributions of arsenic in ground water and surface water
systems in the seven system size categories, for each State, and nationally. Separate box plots
were prepared for ground water and surface water and for CWS and NTNCWS systems in each
State. For example, Figure 5-1 shows the boxplot for the California CWS ground water systems.
Plots for each State in the AOED are included in Appendix B-2.
In each plot, system size categories are plotted on the horizontal axis, and arsenic
concentrations are plotted on the log scale on the vertical axis. Each distribution is represented
graphically, and is composed of system mean arsenic concentrations. The procedure for
calculating system mean arsenic concentrations using regression on order statistics is presented in
Section 6.1. The lower and upper ends of each box represent the 25th and 75th percentiles of
each distribution. The line near the middle of each box is the median, and the darkened circle is
the mean. The "whiskers" above and below each box illustrate the pattern of system means in
the tails of the distribution, and are composed of straight lines and plus signs. Lines represent
system means from the quartiles to the 5th and 95th percentiles. Points in the upper and lower
12 Frey and Edwards (1997) used a stratified survey design, of which system size was a stratification variable. They
did not specifically evaluate whether or not arsenic distributions are different for systems of different sizes.
63
-------
100.000
10.000
1.000
0.100
0.010^
0.001
Ground Water (GWJ SysterJis
363 309 119 156 123 111
System Size
System Size
FIGURE 5-1. Boxplnts of System Means By System Size for
Community Water Systems in the State of California
Number of Systems Indicated Below Boxplot
35
25-100 101-500 501-1000 1001-3300 3301-10000 10001-50000 50001+
100.000
I
" 10.000
n
1
| 1.000
3
*3 0.100
0 0.01D
i
en
0.001
Surface Water (SW) Systems *
->
i 1 1
| | 1 f 1 ?
6 22 17 27 29 29 39
25-1DO 101-5OD 5O1-1DQO 1001-33DQ 33D1-1DOQD 10O01-5OOOO 50DO1+
64
-------
5 percent tails are shown as plus signs. Below each distribution is the number of systems in the
sample.
The box plots provide qualitative comparison of the distributions of mean arsenic levels
across the size categories. For most States, these plots demonstrate that the means, medians, and
quartile ranges of the distributions are similar across the size categories for both ground water
and surface water. This is particularly true when there are relatively large numbers of systems
represented in each of the distributions, as is the case in California. The boxplots based on the
smaller sample sizes tend to appear jagged and may not be representative of the underlying,
(presumably) continuous distribution. Otherwise, because the distributions are similar between
size categories, this qualitative analysis suggests that system mean arsenic concentrations are
probably not associated with system size in either ground water or surface water systems.
In addition to the qualitative analysis of arsenic distributions for systems in different size
strata, we applied an analysis of variance (ANOVA) procedure to quantitatively test if arsenic
concentration is dependent on system size. For this analysis, the mean of the natural logarithm of
the system mean arsenic ("log-mean") was calculated for each state and size stratum. For each of
the seven NAOS regions, and for the entire US, these state and size stratum log-means were
averaged across the region, weighting the mean for each state by the number of systems of the
given size stratum in that state (as defined by the 1998 SDWIS baseline, including purchased and
non-purchased systems). These regional mean values of the logarithm of the system mean arsenic
are presented in Tables 5-2a (ground water CWS systems), 5-2b (surface water CWS systems), 5-
3a (ground water NTNCWS systems), and 5-3b (surface water NTNCWS systems). The means
are presented in the top half of each table, and the relative rankings for the size strata are
presented in the bottom half of each table. ANOVA was used to test the statistical significance
of the differences among the size strata weighted log-means in each Region. For each Region, p
values < 0.05 imply that the means of the size strata are statistically significantly different (at the
five percent significance level). For some combinations of type of water system, source type,
NAOS region, and size stratum, no data were available, so the corresponding log-mean and
stratum rank is missing. This occurs in only two cases for CWS systems but quite frequently for
the NTNCWS systems, which tend to serve fewer people.
These results suggest that mean arsenic concentrations may differ significantly from
stratum to stratum. For ground water CWS systems, these differences were statistically
significant in five of the seven regions. For surface water CWS systems, these differences were
statistically significant in three regions. For ground water NTNCWS systems, these differences
were statistically significant in only two of the seven regions, which is partly attributable to the
relative lack of NTNCWS data. For surface water NTNCWS systems, these differences were
statistically significant in two of the five regions which had sufficient data for this calculation.
However, the stratum ranks in these tables show that mean arsenic concentrations do not vary in
a consistent pattern from region to region. For example, for ground water CWS systems, arsenic
concentrations in Region 1 and 3 appear to generally decrease as system size increases, while in
Regions 6 and 7, arsenic concentrations appear to generally increase as system size increases
(except for the large system 50000 + stratum). In the three remaining regions, no systematic
65
-------
Table 5-2a
Log-Means by System Size Category, CWS Ground Water Systems
SIZE STRATUM
25-100
101-500
501-1000
1001-3300
3301-10000
10001-50000
50000 +
p - Value
25-100
101-500
501-1000
1001-3300
3301-10000
10001-50000
50000 +
LOG-MEANS FOR EACH NAOS REGION
1
0.6312
0.5592
-0.0728
-0.0071
-0.4255
-0.8486
-0.9796
0
2
1.1058
1.1150
1.2639
1.1376
1.2612
1.1723
1.1690
0
3
-0.2185
-0.2554
-0.4796
-0.3914
-0.4294
-0.4572
-0.6332
0.32
4
-0.0362
0.0472
-0.1937
-0.2791
-0.4201
-0.3264
0.3083
0
5
-0.1205
0.1269
0.1221
0.2012
0.3680
0.5226
0.0849
0
6
0.0004
0.0575
0.3096
0.3260
0.6417
0.0788
-0.2877
0.13
7
0.6627
0.6715
0.8684
0.8258
0.9186
1.0712
0.5801
0
All
0.4415
0.3968
0.2065
0.1400
0.1238
0.1923
0.3626
0
STRATUM RANK FOR EACH NAOS REGION
1
2
4
3
5
6
7
7
6
1
5
2
3
4
1
2
6
3
4
5
7
o
6
2
4
5
7
6
1
7
4
5
3
2
1
6
6
5
o
5
2
1
4
7
6
5
3
4
2
1
7
1
2
4
6
7
5
3
AOED States in NAOS Regions
Regionl. ME, NH, NJ
Region 2. KY, NE
Region 3 AL
Region 4 IL, IN, OH, MI, MN
Region 5 AR, KS, MO, NM, OK, TX
Region 6 MT, ND
Region 7 AK, AZ, CA, NV, or UT
66
-------
Table 5-2b
Log-Means by System Size Category, CWS Surface Water Systems
SIZE STRATUM
25-100
101-500
501-1000
1001-3300
3301-10000
10001-50000
50000 +
p - Value
25-100
101-500
501-1000
1001-3300
3301-10000
10001-50000
50000 +
LOG-MEANS FOR EACH NAOS REGION
1
-
0.2700
-0.1448
0.2069
0.0319
0.1886
0.1527
0.89
2
0.9413
0.8689
0.9593
0.7680
0.6825
0.7642
0.9756
0
3
-0.2439
-0.6031
-
-0.3303
-0.4577
-0.3347
-0.4317
0.22
4
0.3329
0.2874
-0.1610
-0.2345
-0.0706
-0.3018
-0.1112
0
5
-0.0548
-0.2778
-0.2545
-0.1993
-0.2819
-0.0711
0.4329
0
6
0.4409
0.4457
-0.1719
0.0425
0.0447
-0.2539
0.6247
0.2
7
0.3571
0.2616
0.1206
0.2116
0.3764
0.3384
0.3544
0.07
All
0.2956
0.0887
-0.0099
0.0338
0.1074
0.0859
0.2386
0
STRATUM RANK FOR EACH NAOS REGION
-
1
6
2
5
3
4
o
6
4
2
5
7
6
1
1
6
-
2
5
3
4
1
2
5
6
3
7
4
2
6
5
4
7
3
1
3
2
6
5
4
7
1
2
5
7
6
1
4
o
J
1
4
7
6
3
5
2
AOED States in NAOS Regions
Regionl. ME, NH, NJ
Region 2. KY, NE
Region 3 AL
Region 4 IL, IN, OH, MI, MN
Region 5 AR, KS, MO, NM, OK, TX
Region 6 MT, ND
Region 7 AK, AZ, CA, NV, or UT
67
-------
Table 5-3a
Log-Means by System Size Category, NTNCWS Ground Water Systems
SIZE STRATUM
25-100
101-500
501-1000
1001-3300
3301-10000
10001-50000
50000 +
p - Value
25-100
101-500
501-1000
1001-3300
3301-10000
10001-50000
50000 +
LOG-MEANS FOR EACH NAOS REGION
1
-0.6399
-0.7615
-0.6124
-0.7024
-0.7569
~
~
0.87
2
0.2806
0.2725
0.2675
0.6338
0.5596
~
~
0.46
3
-0.6931
-0.6931
-0.6684
-0.6931
~
~
~
0.51
4
-0.5861
-0.9659
-0.9927
-0.2996
0.2168
~
~
0.03
5
0.1180
-0.0706
0.0702
0.1058
-0.6931
~
~
0.51
6
2.0479
3.2812
-
~
~
~
~
0.65
7
0.2561
0.2362
0.6937
0.2290
0.4953
1.0669
~
0.03
All
-0.2214
-0.2455
-0.2684
-0.0594
0.0956
1.0669
~
0.31
STRATUM RANK FOR EACH NAOS REGION
2
5
1
3
4
~
~
o
5
4
5
1
2
~
~
o
J
3
1
3
~
~
~
o
6
4
5
2
1
~
~
1
4
o
5
2
5
~
~
2
1
-
~
~
~
~
4
5
2
6
o
5
i
-
4
5
6
3
2
1
-
AOED States in NAOS Regions
Regionl. ME, NH, NJ
Region 2. KY, NE
Region 3 AL
Region 4 IL, IN, OH, MI, MN
Region 5 AR, KS, MO, NM, OK, TX
Region 6 MT, ND
Region 7 AK, AZ, CA, NV, or UT
68
-------
Table 5-3b
Log-Means by System Size Category, NTNCWS Surface Water Systems
SIZE STRATUM
25-100
101-500
501-1000
1001-3300
3301-10000
10001-50000
50000 +
p - Value
25-100
101-500
501-1000
1001-3300
3301-10000
10001-50000
50000 +
LOG-MEANS FOR EACH NAOS REGION
1
~
-0.6931
-0.6931
0.3365
~
~
~
~
2
0.6609
0.9163
0.7975
0.9163
~
~
~
0.84
3
~
~
~
~
~
~
~
~
4
-0.6931
0.2231
-0.6931
-0.3670
~
~
~
0
5
-0.4377
-0.2430
0.7178
~
~
~
~
0
6
-0.2798
-0.7811
~
~
~
~
~
0.4
7
0.4496
0.2584
0.0000
~
~
~
~
0.32
All
0.3039
0.1540
0.1030
0.5230
~
~
~
0.2
STRATUM RANK FOR EACH NAOS REGION
~
2.5
2.5
1
~
~
~
4
1.5
3
1.5
~
~
~
~
-
~
~
~
~
~
3.5
1
3.5
2
~
~
~
o
3
2
1
~
~
~
~
1
2
-
~
~
~
~
1
2
o
J
~
~
~
~
2
3
4
1
~
~
-
AOED States in NAOS Regions
Regionl. ME, NH, NJ
Region 2. KY, NE
Region 3 AL
Region 4 IL, IN, OH, MI, MN
Region 5 AR, KS, MO, NM, OK, TX
Region 6 MT, ND
Region 7 AK, AZ, CA, NV, or UT
69
-------
patterns are evident. Similarly, no systematic patterns are evident for surface water CWS
systems in any region or for the NTNCWS systems in any region.
The ANOVA methods do not take into account the potential for different amounts of
uncertainty in the log means, which are attributable to different sample sizes with different
censoring levels and rates. Furthermore, the results for cases with relatively large numbers of
samples often showed statistically significant differences that were not numerically very large.
Thus, the ANOVA results may not clearly demonstrate that arsenic mean levels are associated
with stratum size. An additional consideration is that for the occurrence analysis to take into
account both the arsenic variation by system size within each region, and the variation by States
within each region, would require stratification by State and system size. Such a detailed
stratification would lead to several State and system size combinations being represented by
small data subsets.
5.3 Stratification by System Type
Another potential stratification variable is water system type. Systems considered for this
analysis could be either CWS or NTNCWS. CWS are public water systems that serve at least 15
service connections used by year-round residents or regularly serve at least 25 year-round
residents.13 NTNCWS are public water systems that are not CWS and that regularly serve at
least 25 of the same persons more than 6 months of the year.14 The majority of NTNCWS
systems serve less than 3,300 people. The AOED database contains data for NTNCWS systems
in 17 States, although only two systems are included for the State of Montana. Basic statistics
were calculated using the system means of the ground water CWS and NTNCWS systems in
each of these States, and these statistics are presented in Tables 5-4a (ground water systems) and
5-4b (surface water systems). For each State and source type, the means and standard deviations
for the CWS and NTNCWS systems were computed and compared using a standard t test to
compare the means and an F test (of the variances) to compare the standard deviations;
significant differences at the five percent level are indicated by asterisks. (The Smith-
Satterthwaite version of the t test was used instead of the usual pooled variance t test if the F test
showed that the standard deviations were significantly different).
These data indicate that for ground water, arsenic distributions in NTNCWS are often
quite similar to arsenic distributions in CWS. In general, the means and the level of censoring
for CWS in a particular State are very close to the levels observed in NTNCWS in that State. In
some States, mean levels are slightly higher in CWS systems, whereas in others, mean levels are
slightly higher in NTNCWS systems. Differences in the means were statistically significant (at
the five percent level) in 6 of the 17 states for ground water systems. The standard deviations
tended to be statistically significantly different, although the numerical differences were not very
large and there was no consistent pattern for whether the CWS or NTNCWS systems had the
higher standard deviation. For surface water, the means for the NTNCWS systems were
13 40 CFR Section 141.2.
14 ibid.
70
-------
Table 5-4a
Arsenic Occurrence in Ground Water CWS and NTNCWS Systems
State 1
AK
AK
AL
AL
AZ
AZ
CA
CA
IN
IN
KS
KS
MI
MI
MN
MN
MO
MO
MT
MT
NC
NC
ND
ND
NJ
NJ
NM
NM
OR
OR
TX
TX
UT
UT
System
Type
CWS
NTNCWS
CWS
NTNCWS
CWS
NTNCWS
CWS
NTNCWS
CWS
NTNCWS
CWS
NTNCWS
CWS
NTNCWS
CWS
NTNCWS
CWS
NTNCWS
CWS
NTNCWS
CWS
NTNCWS
CWS
NTNCWS
CWS
NTNCWS
CWS
NTNCWS
CWS
NTNCWS
CWS
NTNCWS
CWS
NTNCWS
Number of
Systems
304
131
263
31
279
77
1224
330
648
538
506
62
644
230
829
630
773
190
484
2
1735
562
197
20
438
758
559
140
316
84
3105
580
263
17
Mean3
4.11
5.39
0.73 *
0.50*
9.53 *
6.73 *
4.2
4.13
0.26
0.24
2.64
2.12
5.31
4.66
2.77
2.65
0.77*
0.55*
1.81
21.8
3.52*
2.02*
4.85
5.36
0.92*
1.41 *
3.81 *
5.66*
2.77
2.25
2.58
2.69
2.89
3.38
Std. Dev.3
7.67*
9.34*
0.49*
0.022 *
13.8*
9.94*
7.57*
8.58*
0.66*
1.34*
3.83 *
2.34*
9
8.27
5.52
5.18
1.93 *
0.48*
3.53 *
15.3*
1.63 *
2.82*
8.13 *
11.4*
1.38*
2.66*
6.00*
7.15*
5.49*
3.17*
4.26*
5.49*
5.59
4.84
Censoring
(%) 2
51.3
52.7
91.3
96.8
40.5
57.1
43.3
52.4
97.5
97.2
13.6
19.4
23.8
23.5
44.8
51.7
90.6
98.4
49.2
0
80.1
92.2
22.3
35
88.4
86.4
25.9
27.9
75
79.8
67
63.8
31.9
41.2
Notes:
'States without ground water NTNCWS systems (AR, IL, KY, ME, NH, NV, OH, OK) are not listed in
Table 5-4a.
2Percent censoring is defined as the percentage of systems with all datapoints censored. Under this definition,
a system with 10 samples, including one detect and 9 non-detects, would not be considered censored.
Statistically significant differences at the five percent significance level are indicated by asterisks.
71
-------
Table 5-4b
Arsenic Occurrence in Surface Water CWS and NTNCWS Systems
State 1
AK
AK
AZ
AZ
CA
CA
KS
KS
MN
MN
NC
NC
ND
ND
NJ
NJ
NM
NM
TX
TX
UT
UT
System
Type
CWS
NTNCWS
CWS
NTNCWS
CWS
NTNCWS
CWS
NTNCWS
CWS
NTNCWS
CWS
NTNCWS
CWS
NTNCWS
CWS
NTNCWS
CWS
NTNCWS
CWS
NTNCWS
CWS
NTNCWS
Number of
Systems
321
74
112
8
574
31
202
4
69
18
338
16
38
12
58
6
87
24
652
54
108
5
Mean3
1.29
1.25
5.72
3.85
2.75
2.14
1.11
1.4
0.88*
0.69*
3.01*
2.35 *
1.11*
0.68*
1.13
0.8
1.35*
0.90*
1.69
1.84
2.1
1.72
Std. Dev.3
2.08*
0.67*
7.05*
1.86*
4.97*
1.48*
0.56
0.42
0.19*
0.28*
1.10*
0.57*
0.53
0.34
0.62
0.46
1.08*
0.58*
0.74*
1.17*
3.3
1.07
Censoring
(%) 2
65.4
59.5
35.7
62.5
61.3
61.3
10.9
0
56.5
66.7
74
87.5
5.3
0
75.9
66.7
25.3
50
57.1
63
27.8
80
Notes:
'States without surface water NTNCWS systems (AL, AR, IL, IN, KY, ME, MI, MO, MT, NH, NV, OH, OK,
OR) are not listed in Table 5-4b.
2Percent censoring is defined as the percentage of systems with all datapoints censored. Under this definition,
a system with 10 samples, including one detect and 9 non-detects, would not be considered censored.
Statistically significant differences at the five percent significance level are indicated by asterisks.
72
-------
consistently lower than the means for the CWS systems (Kansas was the only exception), but
these differences were statistically significant (at the five percent level) in only 4 of the 11 States.
A graphical comparison between the CWS and NTNCWS arsenic distributions by state
and water system type is discussed and presented in Section 5.5 below.
For ground water, because the CWS and NTNCWS arsenic distributions were sometimes
significantly different, and because there was sufficient CWS and NTNCWS data to separately
estimate the arsenic occurrence for each water system type, the arsenic occurrence estimates
derived in chapter 6 were computed separately for CWS and NTNCWS systems. For surface
water systems, although some significant differences were found, it was determined that there
was insufficient data to recommend developing a separate arsenic occurrence estimate for
NTNCWS surface water systems: Only 11 States have NTNCWS surface water systems in the
AOED, and for four of those States (Arizona, Kansas, New Jersey, and Utah), there were less
than 10 surface water NTNCWS systems. Therefore, for surface water systems, the arsenic
occurrence estimates were based only on the CWS system data in the AOED, and the
exceedance distributions for the CWS systems were used to estimate the exceedance distributions
for the NTNCWS systems.
5.4 Regional Stratification
Natural arsenic sources, such as soil and rock, may be significant sources of arsenic in
drinking water. Therefore, regional differences in geology, hydrology, and hydrogeology may
result in significant differences in arsenic occurrence from region to region. Frey and Edwards
(1997) studied regional differences in arsenic occurrence using USGS WATSTORE data and the
Metro database. As discussed in Chapter 4, they calculated arsenic natural occurrence factors
(NOFs), and identified seven different regions that appeared to have distinct arsenic occurrence
characteristics based on these NOFs. USGS used these regions in its estimation of arsenic
occurrence, and presented comparisons of its findings with Frey and Edwards' findings.
The NAOS regions are displayed in Figure 5-2. The States in each region for which
compliance data are available are shaded. This figure illustrates that there are significant
differences in the coverage that AOED provides for these regions. While almost every state is
represented in the Western Region, South Central Region, and the Midwest Central Region,
fewer States are covered in the Southeast Region, the Mid-Atlantic Region, and the New England
Region. Each of these Regions is represented by no more than three states, and the Southern
Region is only represented by Alabama. Frey and Edwards (1997) found that arsenic levels were
generally higher in the Western and Southern Central Regions, and were lower in the Southeast,
the Mid-Atlantic, and the New England Regions.
Where there are regional differences in contaminant occurrence patterns, and differences
in the level of coverage among those regions, regional stratification can improve the accuracy of
the national estimates. The NAOS Regions, which are delineated according to political
boundaries rather than physiographic provinces, are convenient for use with the AOED database,
which can easily be used to analyze arsenic occurrence on a State or regional basis. Use of the
NAOS Regions is also convenient because it facilitates comparisons of arsenic occurrence
73
-------
Figure 5-2: States with Arsenic Compliance Data in the 7 Regions
NEW HAMPSHIRE
MASSACHUSETTS
RHODE ISLAND
'CONNECTICUT
NEW JERSEY
Reg. 1: New
England
Reg. 6: North
Central
Reg. 5: South
Central
States with compliance data
States without compliance data
Regions adapted from Frey and Edwards, 1997
74
-------
presented in this report with arsenic occurrence levels that were reported by Frey and Edwards
(1997) and by USGS (Focazio etal., 2000). Therefore, the regional stratification scheme of the
NAOS survey was applied to AOED during the development of the arsenic occurrence estimates
that are presented in Chapter 6.
The following States in the AOED database represent each of the NAOS Regions:
New England: Maine, New Hampshire, and New Jersey;
Mid-Atlantic: Kentucky and North Carolina;
Southeast: Alabama;
Midwest: Illinois, Indiana, Ohio, Michigan, and Minnesota;
South Central: Arkansas, Kansas, Missouri, New Mexico, Oklahoma, and Texas;
North Central: Montana and North Dakota; and
Western: Alaska, Arizona, California, Nevada, Oregon, and Utah.
An analysis was conducted to evaluate potential bias introduced into the occurrence
estimates (presented in Chapter 6) by the use of the 25 States data from AOED to represent these
regions. This analysis was based on ground water data from the USGS database, and was
designed to semi-quantitatively assess the extent to which an individual State appears to reflect
arsenic occurrence in other States in the region, and in the region as a whole.
In this analysis, percent exceedances were estimated at concentrations of 2, 5, 10, 20 and
50 |ig/L for each region, using USGS data for the States which are represented in AOED in that
particular region. Two databases developed by the USGS were used to support this analysis.
The first database was derived from the USGS Arsenic Database of Selected Counties (see
Section 4.2.2). From this database, exceedance probabilities were estimated by calculating the
percentage of data points in each county exceeding specific arsenic concentrations.
USGS created the second database from information contained in SDWIS, and this
database provides, for each PWS in the United States, State and county Federal Information
Processing Standard (FIPS) codes15 correlated to the PWS location. The products of the county
exceedance probabilities contained in the first database and the number of PWS in each county,
determined from the second database, are the numbers of PWS in each county that may exceed
specific arsenic concentrations. Summing across counties yields the total number of systems in
the State, and the number of systems that are likely to exceed specific arsenic concentrations.
The occurrence estimates were weighted by the number of systems in each State, based on
SDWIS.
Two sets of regional exceedance probabilities were estimated from the USGS State
exceedance probabilities. The first regional exceedance probability was based on the USGS data
for the State in the region represented in AOED. The second regional exceedance probability
was based on USGS data for all of the States in the Region. For example, in the New England
Region, the first estimate was based on USGS data for the States of Maine, New Hampshire, and
15 FIPS codes, which are unique identifiers for each State (two letter abbreviations or two digit numbers) and county
(five digit identification number) in the United States, are assigned by the U.S. Postal Service.
75
-------
New Jersey. Then, the second exceedance estimates were based on the USGS data for all of the
States in the New England Region.
As a result, two sets of occurrence estimates were developed for ground water in each
region based on USGS data, and these two regional estimates were compared. Table 5-5 presents
these regional comparisons, together with national estimates that were developed as the weighted
sum of the regional estimates. We hoped that analyzing these data would provide information
about the use of some States to represent a region. However, it should be noted that the USGS
data is qualitatively different from the AOED data. Therefore, the power of the USGS data to
predict the potential accuracy of the AOED regional arsenic occurrence estimates may be limited.
The USGS data may have different spatial coverage from the AOED data. For example, the
USGS data may contain data from investigations that focused on specific areas in some States.
Some States may have data from a small number of counties. In addition, the USGS data are not
finished water samples collected from PWS facilities.
Table 5-5
Comparison of Regional Ground Water Arsenic Occurrence
Estimates Based on USGS Data
All States in Region vs. States Represented in AOED
Region
1
2
3
4
5
6
7
National
Data Set
All States
AOED States
All States
AOED States
All States
AOED States
All States
AOED States
All States
AOED States
All States
AOED States
All States
AOED States
All States
AOED States
>2u£/L
13.97%
12.61%
15.99%
8.47%
7.78%
10.96%
26.20%
29.11%
19.89%
17.83%
47.12%
57.98%
41.78%
47.33%
23.32%
23.88%
>5fig/L
6.83%
6.83%
5.02%
2.98%
2.82%
0.00%
14.09%
17.36%
9.56%
8.44%
32.00%
43.08%
25.25%
29.64%
12.25%
13.15%
>10jig/L
3.41%
1.96%
2.18%
0.96%
1.53%
0.00%
7.37%
9.58%
4.78%
4.55%
25.12%
35.90%
15.41%
19.05%
6.97%
7.82%
>20jig/L
0.63%
0.43%
0.62%
0.16%
0.89%
0.00%
2.49%
3.38%
1.74%
1.73%
16.37%
24.75%
6.87%
9.03%
2.92%
3.56%
>50jig/L
0.30%
0.00%
0.04%
0.00%
0.17%
0.00%
0.48%
0.67%
0.35%
0.21%
8.74%
13.41%
2.22%
3.04%
0.92%
1.17%
76
-------
For Regions 1, 4, and 5, the USGS data indicate that the States which are represented in
AOED may be reasonably representative of arsenic occurrence in the entire region. In Region 2,
these data suggest that the States which are contained in AOED may have slightly lower arsenic
occurrence concentrations than regional average concentrations when all States are considered.
The States in Regions 6 and 7 that are included in AOED may overestimate regional average
arsenic occurrence concentrations. In Region 3, data are inconsistent, and suggest that Alabama
data will overestimate regional arsenic occurrence at 2 |ig/L, and will underestimate regional
arsenic occurrence at concentrations at or above 5 |ig/L.
National arsenic occurrence estimates, based on all of the States for which USGS
contains data (USGS does not include any samples from Vermont or Hawaii, but does represent
the remaining 48 States) are quite similar to the national arsenic occurrence estimates, based on
the USGS data in the 25 States that are represented in AOED. Both of these estimates were
derived from the respective regional estimates, weighted by the number of ground water systems
in each of the regions. At concentrations of 2 to 50 |ig/L, the estimates based on USGS data for
the 25 States that are represented in AOED are higher than those based on all of the States in
USGS; however, at each concentration of interest, the occurrence estimates differ by less than
one percent. The greatest differences occur at concentrations of 5 and 10 |ig/L, where the
estimates based on the data for the 25 States in AOED are 0.90 and 0.84 percent higher,
respectively, than those based on data for all of the States. These results suggest that the
estimates based on the AOED data, that are presented in Chapter 6 of this report, are probably
not significantly biased at the national level by the lack of complete representation within each
region. The potential overestimation based on data for some regions is balanced by the potential
underestimation of data for other regions. The USGS data also suggest that the data for the 25
States in AOED may slightly, although not significantly, overestimate national occurrence of
arsenic in drinking water. Therefore, this semi-quantitative analysis of the USGS data suggests
that the arsenic occurrence estimates presented in Chapter 6 may be slightly conservative.
5.5 Arsenic Distributions at the State Level
We further investigated the nature of arsenic occurrence in ground water and surface
water in the individual States through graphical analyses of the distributions of system means.
Figures 5-3a, 5-3b, 5-3c, and 5-3d present boxplots of the distributions of system means for
ground water and surface water systems in each of the 25 states with compliance monitoring
data. The procedure for calculating system mean arsenic concentrations is presented in Section
6.1, and the summary statistics that are displayed by the symbols in each boxplot distribution are
described in Section 5.2 of this report. The number of systems represented in each distribution is
noted below each boxplot. Figures 5-3a and 5-3b show the distributions of all system means for
ground water and surface water, respectively. Figures 5-3c and 5-3d show the distributions of the
system means that are not completely censored for ground water and surface water, respectively.
As discussed in chapter 6, a system mean is completely censored if all the arsenic measurements
for that system were censored, i.e., non-detects. In this case, the system mean was estimated by
applying the regression on order statistics method to all the system means for that State. The
states have been ordered by NAOS region, and then alphabetically within each NAOS region
(shown as the number after each state abbreviation).
77
-------
a
a
8
3
100.00
10.00 -
1.00 -
0.10
0.01
100.00
10.00
+
f
t
[
Ground Water (GW) Systems
* : i
, i * ;
t
L
109 504 4i
IS
7
j
rL
88 1735
f
}
,
I
263 7
18 562 31
t
i
±
io e
B
e
6
B
*
I
j
4
8.
19
( +
i +
; * +
<
:
1
875 371 506
230 630 62
MNNKNAI IMMOAK
EHJYCLLNINHRS
1112234444455
S,
0.10
0.01
States / Region
f i *
-. ^ t Ground Water JGW) Systems
*
t
*
7
*
3 5!
i
I
>9 446 31
; ;
I ;
I
1
S
i
35 484 1!
7
3
4
:
279 1224
^
^
221 a
6 2
53
190 140 580 2 * 20 * 131 77 3?0 84 17
MNOTMNAACNOU
OMKXTDKZAVRT
555566777777
States / Region
FIGURE 5-3a. Boxplots of System Means for
GW CWS and NTNCWS by State
Number of Systems radicated Below Boxplot (CWS on left)
All Systems
78
-------
100.00 -
J 10.00 -.
1 :
1
.a 1.00 -
* 0.10:
a :
0.01
Surface Water (SW) Systems
+
(
I . , _ . ,
i i n - ! ! * i R
" I " D
29 37 29 150 169 68 93 51 33 23 139 76 101
38 62
MNNKNAI IMMOAK
EHJYCLLNINHRS
1112234444455
100.00 -
J 10.00 -
1 :
o 1.00 -
^ 0.10^
1 ;
0.01-
States / Region
+ Surface Water (SW) Systems ;
t * t
: f ; i . 1 1 . 1 i
a
IM 'M |(|r l l
9 29 210 326 47 19 106 33 176 31 129 35
8 27 6 24 2 10
MNOTMNAACNOU
OMKXTDKZAVRT
555566777777
States / Region
FIGURE 5-3b. Boxplots of System Means for
SW CWS and NTNCWS by State
Number of Systems radicated Below Boxplot (CWS on left)
All Systems
79
-------
100.00 -
J 10.00 -
1.00-
Ground Water (GW) Systems
, : I
'
0.10.
o.oi-
100.00 -
.J 10.00 -
I :
s
a
* 0.10 -.
0.01
147 51 87 346 23 200 16 491 458 219 3 437
103 44 1 15 176 304 50
MNNKNAI IMMOAK
EHJYCLLNINHRS
1112234444455
States / Region
Ground Water JGW) Systems
"
73 414 184 1026 246 153 148 166 694 143
3 101 210 2 13 62 33 157
79 179
17 10
M N O
O M K
555
T
X
5
M
T
6
N
D
6
A
K
7
A
Z
7
C
A
7
N
V
7
O
R
7
U
T
7
States / Region
FIGURE 5-3c. Boxplots of System Means for
GW CWS and NTNCWS by State
Number of Systems radicated Below Boxplot (CWS on left)
Systems that are Not Completely Censored
80
-------
100.00 -
Surface Water (SW) Systems
J 10.00 -
I :
I
.y 1.00 -
0.1D-
150
44
18
10
42
90
0.01
100.00 -
M
E
1
N
H
1
N
J
1
K
Y
2
N
C
2
A
L
3
I
N
4
States / Region
M
I
4
M O
N H
4 4
A
R
5
K
S
5
J 10.00 -
I :
o 1.00 -
Surface Water (SW) Systems
0.10-
o.oi-
10
M
O
5
N
M
5
O
K
5
T
X
5
M
T
6
N
D
6
A
K
7
A
Z
7
C
A
7
N
V
7
States / Region
FIGURE 5-3d. Boxplots of System Means for
SW CWS and NTNCWS by State
Number of Systems radicated Below Boxplot (CWS on left)
Systems that are Not Completely Censored
O
R
7
U
T
7
81
-------
These boxplots suggest that arsenic concentrations in drinking water may vary
significantly from State to State. Among ground water systems, the State of Indiana has the
lowest arsenic concentration levels, both for CWS and NTNCWS systems, and the States of
Arizona and Nevada appear to have the highest concentrations. Comparing Figures 5-3a and 5-
3c for Indiana shows that the estimated system means for that state are very sensitive to the
conventions used to estimate system means for completely censored systems; the distribution of
uncensored system means for Indiana is comparable to the distributions of uncensored system
means for the other states in the same NAOS region: Illinois, Michigan, Minnesota, and Ohio.
The side by side graphical comparison of the CWS and NTNCWS arsenic distributions by State
for ground water conform the similarity of these distributions noted earlier (in section 5.3).
Among surface water systems, using all estimated system means, Figure 5-3b shows that
the arsenic concentrations were lowest in Missouri, and were highest in the States of Arizona,
North Carolina, New Hampshire, Nevada, and Ohio. It should be noted that the distributions for
North Carolina, New Hampshire, and Ohio may be influenced by the combination of relatively
high detection limits (typically 5 to 10 |ig/L) and high levels of censoring (arsenic was not
detected in any samples from 70 to 78 percent of the surface water systems in these three States).
The influence of censoring is seen by examining Figure 5-3d, which shows that for the systems
that were not completely censored: the lowest values were for North Dakota and Alabama; North
Carolina, New Hampshire, and Ohio have high concentration levels but are not the States with
the highest concentrations; The very few remaining systems in Missouri are roughly at the
median level for the 25 States.
Figure 5-4 shows a lognormal probability plot for ground water systems in the State of
New Jersey, and Figure 5-5 shows a lognormal probability plot for ground water systems in the
State of New Hampshire. Similar plots for the remaining States in the AOED database are
included in Appendix B-3.
In these plots, means for uncensored PWS ID are represented with diamonds. (The
PWSID is uncensored if there is at least one measured detected value for that system.) The
straight line (in logarithmic space) shows the fitted values from the lognormal model. When the
set of uncensored system means falls close to the line of the fitted values from the lognormal
model, this is a qualitative indication that the distribution of system means does not strongly
depart from the lognormal distribution. These plots were not created for States, water system
types, and source types that had four or fewer uncensored system means, since they would not be
expected to be reasonable approximations of the arsenic distributions in that case. With four or
fewer detects, substitution of non-detects by half the reporting limit was used instead of the
regression on order statistics method based on the lognormal model.
In many cases, the distributions of system means are fairly linear, and are consistent with
lognormal distributions. Figure 5-4, for ground water CWS systems in New Jersey, is an
example of a very good fit to the lognormal model. In some cases, the fit of the data is least
strong in the tails of the distributions. The lognormal probability plot for ground water CWS
systems in New Hampshire, shown in Figure 5-5, appeared to depart most strongly from
lognormality. The sharp angle that appears on the probability plot near the first quartile of the
distribution may be related to differences between measured values and censored system values.
82
-------
q
d
9
q
d
|z; « u
83
-------
i
- O
CM
I
q
d
o
9
o
o
q
Q
« u
84
-------
Several States have a similar pattern, in that the distribution provides a reasonably good fit to
lognormality in the upper part of the distribution but may appear to fit less well at lower levels,
which is perhaps due to the model used to fill-in the censored values. In cases like New
Hampshire ground water CWS systems, a different lognormal distribution seems to fit the
concentrations in the lower tail. Down to levels below the regulatory level of interest, the data
are consistent with a lognormal model.
5.6 Summary of Patterns of Arsenic Occurrence
The analyses that are presented in this Chapter were designed to support decisions related
to the selection of an appropriate method for estimating arsenic national occurrence. These
analyses show that, in order to develop more representative estimates of arsenic occurrence, data
should be stratified by source water type, and by region, and, if the data are sufficiently
representative, by system type. The data for surface water NTNCWS systems are not sufficient
to provide separate occurrence estimates for this type of system.. Data should not be stratified by
system size. In addition, these data suggest that arsenic occurrence at the State level is relatively
lognormally distributed above a certain cut-off level. As a result of these findings, in Chapter 6,
arsenic data are stratified accordingly, and are assumed to be lognormally distributed above a
certain cut-off level within States.
85
-------
This page intentionally left blank
86
-------
6. National Occurrence Estimates
This Chapter presents estimates of the number of systems that have mean arsenic levels
equal to or greater than specific MCL16 alternatives or concentrations of interest. First, the model
used to estimate exceedance probabilities is defined, and the assumptions that were made and the
data conventions that were applied are defined. Next, the specific number of systems that are
predicted to exceed potential MCL alternatives are presented for community water supply
systems and non-transient, non-community water supply systems. Then, the exceedance
probabilities calculated using AOED are compared to those developed using the NAOS, USGS,
MRS, Metro, and ACWA databases. The last section of this Chapter discusses an uncertainty
analysis that was conducted on the AOED-based arsenic occurrence estimates, and presents the
results of that analysis. Confidence intervals generated as a result of the uncertainty analysis
were applied to the exceedance estimates to provide a measure of the variability associated with
these estimates.17
6.1 Arsenic National Occurrence Projection Methodology
The methodology applied to develop estimates of arsenic occurrence in ground water and
surface water systems has five steps. These steps result in the derivation of probability
distributions of mean arsenic concentrations and the estimates of numbers of systems that exceed
specific mean concentrations of interest, and the estimation of the numbers of systems with
system means exceeding specific concentrations of interest. This process fits the unique data
structure of the compliance monitoring data in the AOED database. As a slight misuse of
terminology we shall sometimes say that a system exceeds a certain level, instead of a more
precise statement that the system mean for that system exceeds the level. Similarly, the
exceedance probability distributions for the state, region, or nation, are defined as the actual or
estimated percentage of systems in the geographical area with system means above various
levels. The five steps include:
Calculate system arithmetic means;
Calculate State exceedance probability distributions for ground water and surface water;
Apply weighting and develop regional exceedance probability distributions for ground
water and surface water;
Apply weighting and develop national exceedance probability distributions for ground
water and surface water; and
Estimate numbers of systems exceeding levels of interest as the product of the national
probability distributions and the total number of ground water or surface water systems.
16 An MCL is the maximum level of a contaminant that is allowable in public drinking water supplies. When EPA
sets an MCL for a contaminant, the PWS must ensure that the level of this contaminant is maintained at or below its
MCL.
17 Strictly speaking, an estimate is a number that does not have uncertainty or variability. The proper statistical term
is "estimator." Each estimate is a realization (observed value) of the estimator (a random variable, that has a
statistical distribution).
87
-------
In addition, it should be noted that confidence intervals for the system exceedance
estimates are developed through the uncertainty analysis that is presented in Section 6.4. These
intervals provide a measure of the variability associated with the national arsenic occurrence
estimators, and they are applied in Section 6.2 of this report.
6.1.1 System Means
The first step in our analysis was to estimate an average arsenic concentration over time
for each system in the AOED. The number of observations in the AOED for a single system
varies by system and by State; larger systems tend to have more observations, and some States
have many more observations per system. The set of system means is therefore more
representative of arsenic occurrence in water systems than just the set of samples in the AOED
would be.
In order to estimate the system means, we had to account for "censored" or "non-
detected" concentrations. Non-detected concentrations are reported as "non-detect at X jug/L,"
for some detection or reporting limit X. This means that the concentration was somewhere
between 0 and X jug/L; the analytical method used to estimate the concentration is incapable of
reliably measuring concentrations less than X jug/L. Non-detected concentrations should not be
ignored when estimating a mean; they are the lowest observations, so to discard them would
introduce a positive bias in the mean estimate. Instead, we accounted for non-detected
concentrations by "filling them in" statistically in one of two ways, and then averaging, according
to the following plan:
If all observations from the system were detects, we calculated the arithmetic mean of the
observed concentrations for the system.
If the system included at least five detected concentrations that were not all equal, and
some non-detects, we filled in the non-detects by regression on order statistics (ROS), as
described in Appendix A, and then averaged the detected and filled-in concentrations.
If the system included four or fewer detected concentrations, or five or more detects that
were all equal, then we substituted half of the detection limit for each non-detect, and
then averaged the detected and substituted concentrations.
If all samples from a system were censored, the system was labeled "non-detect" at the
mode of the detection limits.
The ROS method uses observations above the detection limit to extrapolate to
observations below the detection limit. The method is described in detail in Appendix A. The
idea is to assume that the arsenic concentrations in a system follow a lognormal distribution;
estimate the distribution's parameters by fitting a line to a lognormal probability plot of the
detected concentrations; then extrapolate the fitted line below the detection limit, in order to
obtain the expected locations of the non-detected concentrations. Appendix B-3 shows plots of
this type for State distributions; although these plots are for States instead of systems, the method
of plotting and filling in is the same. ROS is simple to implement, and it has been shown to
perform nearly as well as traditional methods, such as maximum likelihood, when the lognormal
assumption is correct (Helsel and Cohn, 1988). Perhaps more important is that even when
lognormality does not hold, ROS still performs well while other methods perform less well.
-------
ROS is one of a class of "fill-in" methods, in which non-detects are filled in in some way,
and then parameters are estimated. The simple ^-detection-limit substitution is another, simple
fill-in method, equivalent to assuming that non-detects follow a uniform distribution from 0 to
the detection limit. Kroll and Stedinger (1996) compared three lognormal fill-in type methods,
including ROS, for estimating means from censored data, over a range of water quality and flow
distributions. All three methods follow the fill-in scheme described above, but where ROS uses
regression to estimate the lognormal parameters, the other two methods use maximum likelihood
(ML) and probability-weighted moments. Kroll and Stedinger found that the ROS method
generally performs as well as or better than probability-weighted moments. ROS and ML
perform about equally well under low to moderate censoring, while ML has some advantage
when the censoring fraction is high (around 80%). On the other hand, ROS is simpler to
understand and to implement than ML, which requires an iterative computation for each system.
We decided that the potential benefit of the ML fill-in did not justify its extra computational
complexity.
ROS requires a reasonable number of detected observations in order to give a reliable fit
of the line to the probability plot. In our procedure above, we required at least 5 detected
observations. Where there were fewer than 5 detects for a system, we opted to use the simple
substitution of /^ the detection limit. This method is widely used, and has been shown to
perform better than either of the other two common substitution methods, namely of 0 or 1 times
the detection limit (Helsel and Cohn, 1988).
Following the decision criteria described above, we used ROS and the /^-detection limit
substitution in 2.3% and 13.2%, respectively, of the systems in the AOED. In 20.8% of systems
there were only detected concentrations. Thus we were able to estimate mean arsenic
concentrations for all of these systems. The remaining 63.7% of systems contained only
censored observations, and so were labeled non-detect. Although we did not estimate mean
concentrations for these systems, they still played a role in the estimation of State exceedance
probability distributions, as described in the next section.
6.1.2 State Exceedance Probability Distributions
State exceedance probability distributions indicate the probability that a randomly chosen
PWS from any specific State will have a mean arsenic concentration greater than a particular
concentration of interest. Using the sample set of system means that were derived for each State
from the compliance data in AOED, exceedance probability distributions were developed
separately for ground water CWS systems, surface water CWS systems, and ground water
NTNCWS systems in each State. Exceedance probability distributions were not separately
developed at the State, Regional, or national levels for surface water NTNCWS systems, due to
the limited number of such systems in the AOED database and nationwide. There are only 816
such systems in the nation according to the 1998 Baseline SDWIS database. Instead, the national
arsenic occurrence estimates for surface water NTNCWS systems were developed by applying
the national arsenic occurrence estimated exceedance probabilities for surface water CWS
systems to the numbers of surface water NTNCWS systems.
89
-------
Several methods could be used to estimate exceedance probability distributions in each
state. These include empirical and parametric estimators, using all or only a subset of the
estimated means to estimate the parameters. Empirical distributions estimate the probability of
exceeding any threshold as the observed fraction of the estimated system means that exceed that
threshold. Empirical estimates are simple to compute, and they do not require any assumptions
about the form of the distribution being estimated. A disadvantage of this type of estimate is that
it "jumps" by a discrete amount at each datum, and does not predict well either above or below
the range of the observed data. For example, it estimates zero probability (or in some versions, a
small but fixed probability) of ever seeing a system mean of any size larger than the largest
system mean observed so far.
Parametric estimators assume that the distribution follows a particular form, such as
lognormal. The distribution has parameters which are estimated from the data, by, for example,
maximum likelihood (Cohen, 1991) or an adapted ROS (Appendix A). At the cost of assuming a
particular distributional form, the parametric estimator produces smoothly changing probability
estimates even outside the range of the observed data. It also yields a simple computational form
for the estimates, which is useful as an input to further analyses, such as the cost models used in
the regulatory impact analyses (RIA) (although it should be noted that the RIA relies on National
exceedance probability distributions rather than State exceedance probability distributions).
Another reason to use a parametric fit is that, in the State distributions, the data do not consist of
true system means; rather, they are only estimated means, as described in the previous section.
An empirical estimator of exceedance probability preserves the errors in the estimators of the
mean concentration, while a parametric estimator tends to smooth them.
Because of these advantages, parametric distributions were fit to the distribution of
estimated system means in each combination of water system type, source type, and State. In
particular, lognormal distributions were used, and these provide a reasonably good fit in most
cases. Appendix B-3 shows lognormal probability plots for each State and source water type, in
which log-system means are plotted against their corresponding normal quantiles. Each plot also
shows a regression line fitted to the data in the plot. If the data in a plot are truly lognormally
distributed, they should lie close to the fitted line.
Examination of the plots in Appendix B-3 suggests that in some States and source types,
two distinct populations are present: the plotted system means form a broken line, instead of a
straight line that would indicate a single lognormally distributed population. This effect is most
apparent in the plot for New Hampshire ground water (see Figure 5-5), where the plotted system
means form two lines of different slopes, with the breakpoint at 5 jug/L on the vertical axis. In
New Hampshire, there are no detected concentrations below 5 jug/L, and the detection limit for
all non-detected concentrations is 5 jug/L. Therefore, any system mean that is estimated to be
less than 5 jug/L must have been estimated from a large portion of data that was "filled in" below
5 jug/L from non-detects, as described in the previous section. The probability plot implies that
these systems may form a different distribution than the systems above 5 jug/L, which
presumably have fewer filled-in observations. The implication is that when a large fraction of
the observations from a system are non-detects, filling in the missing observations may not
reproduce the State's arsenic distribution accurately.
90
-------
In light of this evidence, a cutoff point was established for each State and source water
combination, and the values of the system means that were less than this cutoff point were not
used in fitting the lognormal distributions. The numbers of system means below the cutoff point
were used to compute the plotting positions for the system means above the cutoff point,
according to the ROS method. The plotting positions are the censored probability plotting
positions defined in Appendix A. The cutoff points were mostly set equal to the most common
detection limit for the State and source water type. In a few cases cutoff points were set lower
than the modal detection limit, where there were not enough system means above the detection
limit to give reasonably stable parameter estimates. Somewhat arbitrarily, we decided to require
10 or more system means above the selected cutoff point. The cutoff points for State and source
water types are presented in Table 6-1.
Table 6-1
Cutoff Points for Ground Water and Surface Water, in ug/L
State
AL
AK
AR
AZ
CA
IL
IN
KS
KY
ME
MI
MN
MO
Ground Water
1
2
0*
5
2
2
2
2
>2
2
2
2
2
Surface Water
1
2
0*
5
2
1
1
2
>2
1
1
0*
1
State
MT
NC
ND
NH
NJ
NM
NV
OH
OK
OR
TX
UT
Ground Water
>2
5*
2
5
1
2
5
10
2
2*
2
2
Surface Water
>2
5*
1
2*
1
2
5
2*
2
2*
2
1
* Cutoff point set below modal reporting limit. Modal reporting limits for these States are (GW and SW modal
RL are the same except where noted): AR (5); MN (1); NC (10); NH (5); NJ (GW: 5; SW: 2); OH (10); and OR
(5).
Using a parametric approach that we have named right-tailed ROS, lognormal
distributions were fit to the remaining system means in each State, type of water system (CWS or
NTNCWS), and source water type, by fitting a regression line to the data in the plots of
Appendix B-3, that is, to log-system means greater than the cutoff point plotted against their
normal quantiles. The plotting positions are the censored probability plotting positions defined
in Appendix A. The exceedance probability for any given log-system mean arsenic concentration
was then estimated by using the fitted regression line to find the normal quantile corresponding
91
-------
to that concentration, and computing the standard normal probability associated with the quantile
(see A-3 in Appendix A). The estimated probability distributions for each State, type of water
system, and source water type are presented in Appendix B-l.
In a few cases this right-tailed ROS method could not be meaningfully applied because
the number of detected system means was insufficient for analysis. If there were four or fewer
detected system means, then the empirical method was employed instead of right-tailed ROS: If
a system is completely censored, i.e., all the measurements for that system are below the
reporting limit, then the system mean is estimated as one half of the mode of the reporting limits
for that system. Otherwise the system mean is estimated as the arithmetic mean of the measured
and filled-in concentration values, as described in Section 6.1.1. Then, the estimated exceedance
probabilities from the empirical method are the percentages of all system means above the
selected concentration levels. The cases where this empirical method was applied instead of
right-tailed ROS are identified by Use ROS? = "N" in the Appendix B-l tables.
6.1.3 Regional Exceedance Probability Distributions
The third step to developing national occurrence projections was to develop regional
exceedance probability distributions. Separate probability distributions were developed for
arsenic occurrence in ground water CWS systems, surface water CWS systems, and ground water
NTNCWS systems, in each of seven NAOS regions. The seven NAOS regions are based on
those identified by Frey and Edwards (1997), and the States in each region are discussed in
Section 5.4.18 Regional exceedance probability distributions were developed as the weighted
sum of the exceedance probability distributions derived for each State with compliance
monitoring data in the region.19 As such, for Region Y represented by data from t States, where t
is any integer, the Regional distribution for each type of water system and source type was
calculated as:
P>x,R(y) = (n>x,si + n>x,s2+...+n>x,st) / (Nsl + Ns2+...+Nst)
where:
P>x,R(y) is the probability that a PWS system in Region Y will have a mean arsenic
concentration that exceeds the arsenic concentration x;
n>xsl, n>xs2 to n>xst are the number of purchased and non-purchased water systems in
States 1 to t that are predicted to have mean arsenic levels greater than x;
Nsl, Ns2 to Nst are the total number of purchased and non-purchased water systems in
States 1 to t; and
n>x,Ri was estimated by multiplying the regional probability distribution developed in
Section 6.1.3 by the total number of purchased and non-purchased systems in the region
(from SDWIS).
18 Note that any similarities between the boundaries of the NAOS Regions and the boundaries of EPA's Regional
Offices is purely coincidental.
19 The number of systems exceeding specific arsenic concentrations in each state are self-weighted quantities.
92
-------
For each potential MCL alternative or concentration of interest, a separate exceedance
probability was calculated based on available data. States within a region were only used to
estimate exceedance probabilities for arsenic concentrations higher than their censoring level.
For example, in the Western Region, only the States of Alaska, California, and Utah have
detection limits that are equal to or less than 2 |ig/L, so only these States were used to estimate
the number of systems in the region that are likely to have mean arsenic levels greater than 2
|ig/L. They also contribute to the estimates for MCL alternatives of 3, 5, 10, 15, 20, and up to 50
Hg/L. Arizona, Nevada, and Oregon are included for all estimates at 5 jig/L and above. The
remaining States in the Western Region, Idaho and Washington, did not have compliance
monitoring data in AOED. The regional exceedance probability distributions based on the
compliance monitoring data are presented in Table 6-2a for CWS systems and Table 6-2b for
NTNCWS systems.
For NTNCWS systems, some modifications were made to the above approach. For
surface water NTNCWS systems, the Regional exceedance probability estimates were copied
from the surface water CWS Regional exceedance probability estimates. For ground water
NTNCWS systems, NTNCWS data from 13 States and CWS data from Alabama were used to
develop the Regional exceedance probability estimates. Ground water NTNCWS data was used
from the following States: Alaska, Arizona, California, Indiana, Kansas, Michigan, Minnesota,
North Carolina, North Dakota, New Jersey, New Mexico, Oregon, and Texas. The other four
States with some ground water NTNCWS data in the AOED are Alabama, Missouri, Montana,
and Utah. For those four States, there were fewer than ten uncensored systems above the cut-off
and so those data were not used in the development of the Regional probability estimates. Since
Alabama is the only State with compliance data in Region 3, Southeast, the ground water CWS
data from Alabama was substituted for the ground water NTNCWS data so that Regional
estimates could be obtained for that Region.
The convention of using only data from States with a detection limit below or equal to the
concentration of interest to estimate regional occurrence probabilities generally yields regional
exceedance percentages that decline as concentrations increase. However, this convention does
result in two anomalies in the regional probability distributions, which appear as increases in
probabilities as arsenic concentrations rise. Specifically, in Mid-Atlantic ground water CWS
systems, the probability of exceeding 5 |ig/L (0.39 percent) is based on Kentucky data, whereas
the probability of exceeding 10 |ig/L (0.75 percent) is based on data from both Kentucky and
North Carolina. The second anomaly was observed in surface water CWS systems in the New
England Region, where exceedance probabilities rose from 6.2 percent at 3 jig/L to 11.7 percent
at 5 |ig/L. At the lower concentration, the probability is based on data from Maine, while the
higher concentration is based on data from both Maine and New Hampshire.
The data presented in Table 6-2a suggest that arsenic occurrence in ground water CWS
systems is lowest in the Mid Atlantic and South East Regions. Also, these data indicate that
intermediate ground water arsenic mean levels are found in New England, Midwest, North
Central, and South Central Regions, and that the West Region tends to have higher arsenic mean
levels than the other Regions. These regional patterns of arsenic occurrence in ground water are
generally similar to those reported by other studies (Focazio et a/., 2000; Frey and Edwards,
93
-------
Table 6-2a
Regional Exceedance Probability Distribution Estimates for Community Water Systems
Region
Percent of Systems Exceeding Arsenic Concentrations (|J.g/L) of:
2
3
5
10
15
20
25
30
40
50
Ground Water Systems
New England
Mid Atlantic
South East
Midwest
South Central
North Central
West
28.876
-
1.511
28.286
27.162
29.632
42.380
21.668
-
0.873
21.170
18.564
21.265
31.512
20.806
0.393
0.413
13.776
9.696
13.083
25.240
6.951
0.748*
0.135
6.221
3.603
6.011
12.464
4.288
0.156
0.066
3.614
1.830
3.586
7.499
2.948
0.044
0.039
2.387
1.083
2.419
4.996
2.163
0.015
0.026
1.701
0.703
1.755
3.549
1.658
0.006
0.018
1.276
0.486
1.337
2.636
1.065
0.001
0.010
0.794
0.264
0.854
1.596
0.741
0.0
0.006
0.539
0.160
0.594
1.051
Surface Water Systems
New England
Mid Atlantic
South East
Midwest
South Central
North Central
West
8.441
-
0.516
4.371
10.424
19.081
17.758
6.231
-
0.212
3.047
3.826
9.104
12.675
11.736*
0.109
0.062
1.648
0.856
3.247
8.171
0.973
0.014
0.009
0.660
0.184
0.580
3.421
0.578
0.0
0.003
0.431
0.092
0.170
2.006
0.430
0.0
0.001
0.321
0.057
0.065
1.364
0.340
0.0
0.001
0.254
0.039
0.029
1.007
0.279
0.0
0.0
0.209
0.029
0.014
0.782
0.203
0.0
0.0
0.153
0.018
0.004
0.520
0.157
0.0
0.0
0.119
0.012
0.002
0.376
*Regional exceedance probabilities for New England SW are based on Maine data at 2 and 3 |ig/L, and Maine and New Hampshire data at 5 ng/L. For Mid
Atlantic GW, exceedance probabilities are based on Kentucky data at 5 |ig/L and Kentucky and North Carolina data at 10 i-ig/L. The differences in arsenic
occurrence between these States result in Regional exceedance probabilities that increase in these concentration ranges.
94
-------
Table 6-2b
Regional Exceedance Probability Distribution Estimates for Non-Transient Non-Community Water Systems
Region
Percent of Systems Exceeding Arsenic Concentrations (|J.g/L) of:
2
3
5
10
15
20
25
30
40
50
Ground Water Systems
New England
Mid Atlantic
South East1
Midwest
South Central
North Central
West
-
-
1.511
34.505
33.280
35.863
45.017
-
-
0.873
26.169
24.008
29.753
34.294
-
-
0.413
17.085
14.449
22.815
21.923
2.146
1.438
0.135
8.234
5.947
15.044
10.514
1.055
0.774
0.066
4.929
3.145
11.431
6.289
0.610
0.483
0.039
3.290
1.893
9.273
4.200
0.389
0.330
0.026
2.348
1.236
7.819
3.002
0.265
0.239
0.018
1.756
0.855
6.765
2.248
0.140
0.140
0.010
1.078
0.460
5.329
1.385
0.083
0.091
0.006
0.721
0.276
4.391
0.929
Surface Water Systems2
New England
Mid Atlantic
South East
Midwest
South Central
North Central
West
8.441
-
0.516
4.371
10.424
19.081
17.758
6.231
-
0.212
3.047
3.826
9.104
12.675
11.736*
0.109
0.062
1.648
0.856
3.247
8.171
0.973
0.014
0.009
0.660
0.184
0.580
3.421
0.578
0.0
0.003
0.431
0.092
0.170
2.006
0.430
0.0
0.001
0.321
0.057
0.065
1.364
0.340
0.0
0.001
0.254
0.039
0.029
1.007
0.279
0.0
0.0
0.209
0.029
0.014
0.782
0.203
0.0
0.0
0.153
0.018
0.004
0.520
0.157
0.0
0.0
0.119
0.012
0.002
0.376
'Regional exceedance probabilities for NTNCWS SW in the Southeast Region are based on CWS data from Alabama and so are identical to the Southeast
regional exceedance probabilities for CWS SW.
2Regional exceedance probabilities for NTNCWS SW are copied from the regional exceedance probabilities for CWS SW.
*Regional exceedance probabilities for New England NTNCWS SW are based on Maine CWS data at 2 and 3 |ig/L, and Maine and New Hampshire
CWS data at 5 |ig/L. The differences in arsenic occurrence between these States result in Regional exceedance probabilities that increase in these
concentration ranges.
95
-------
1997), except the other studies found that arsenic mean levels in New England were relatively
low and comparable to those in the Mid Atlantic and South East. Therefore, it is possible that
the compliance monitoring data for the States of Maine, New Hampshire and New Jersey may
over estimate arsenic occurrence in the New England Region. In surface water CWS systems,
the lowest arsenic mean levels again occurred in the Mid-Atlantic and South East. Mean levels
generally appeared to increase in the other Regions, from Midwest, to South Central, to the New
England, West, and North Central Regions, although the ordering of these Regions changes with
the concentration level.
6.1.4 National Exceedance Probability Distributions
The fourth step that was used to develop national occurrence estimates was to develop
estimates of national exceedance probability distributions from the regional exceedance
probability distributions. Separate exceedance probability distributions were developed for
arsenic occurrence in ground water CWS systems, surface water CWS systems, and ground water
NTNCWS systems of the United States. National exceedance probability distributions were
developed as the weighted sum of the exceedance probability distributions derived for each
Region, which were presented in Tables 6-2a and 6-2b. As such, when the United States is
represented by data from 7 Regions, the National distribution was calculated separately for each
water system type and source type as:
P>x,us = (n>x,Ri + n>XiR2+...+n>XjR7) / (NR1 + NR2+...+NR7)
where:
P>x,us is the probability that a PWS system in the United States will exceed arsenic
concentration x;
n>x,Ri 5 n>x,R2 to n>xR7 are the number of purchased and non-purchased water systems in
Regions 1 to 7 that are predicted to have mean arsenic levels greater than x; and
NR1, NR2 to NR7 are the total number of purchased and non-purchased water systems in
Regions 1 to 7.
For each potential MCL alternative or concentration of interest, a separate exceedance
probability was calculated based on available data. Exceedance probabilities from all Regions
contributed to the estimation of each of these National exceedance probability distribution
estimates at each concentration, with the exception of the Mid Atlantic Region for CWS systems
and with the exceptions of the Mid Atlantic and New England Regions for NTNCWS systems.
Because regional exceedance probability distributions for CWS systems were not estimated for
the Mid Atlantic Region at concentrations of 2 or 3 |ig/L, no data was available for this Region to
support the development of National estimates at these concentrations. Because regional
exceedance probability distributions for NTNCWS systems were not estimated for the Mid
Atlantic and New England Regions at concentrations of 2, 3, or 5 |ig/L, no data was available for
these Regions to support the development of National estimates at these concentrations.
96
-------
Table 6-3a shows the estimated national arsenic occurrence exceedance probability
distributions for ground water and surface water CWS, and Table 6-3b shows the estimated
national arsenic occurrence exceedance probability distributions for ground water and surface
water NTNCWS. Since the Regional estimates for surface water NTNCWS systems are copied
from the Regional estimates for surface water CWS systems, the national estimates for surface
water NTNCWS systems are the same as the national estimates for surface water CWS systems.
Tables 6-3a and 6-3b also show the exceedance probabilities of lognormal distributions that were
fit, using ROS, to each of the national distributions. The fitted lognormal distributions are
mostly quite close to the original distributions. The lognormal distributions have two
advantages: they can predict the percent exceedances at low concentrations, and they have a
simple functional form that is easy to use in further analyses. In particular, OGWDW used the
fitted lognormal distributions in Tables 6-3a and 6-3b for the cost simulations in its regulatory
impact analysis of the arsenic MCL.
6.1.5 Number of Systems Exceeding Alternative MCLs
The estimated number of systems exceeding alternative MCL levels was calculated by
multiplying the total number of systems in the United States with the probability that a system
would exceed a specific MCL alternative. Separate estimates were developed for ground water
and surface water CWS, and for ground water and surface water NTNCWS. The total number of
systems in each category was derived from 1998 Baseline SDWIS data. It should be reiterated
that the national right-tailed ROS exceedance probability distributions for surface water systems
were derived from analysis of CWS systems, and these probability distributions were applied to
estimate arsenic occurrence in both CWS and NTNCWS surface water systems.
The following section presents estimates of numbers of ground water and surface water
systems within specific size categories that may have mean arsenic levels in excess of specified
MCL alternatives. These estimates are based on the national right-tailed ROS exceedance
probability distributions for ground water and surface water, multiplied by the total number of
systems in the nation in each size category. Analyses presented in Chapter 5 indicated that there
are not meaningful or consistent differences in arsenic occurrence from size stratum to size
stratum. Therefore, for each source water type, the same national right-tailed ROS exceedance
probability distribution was applied for each size stratum.
6.2 Arsenic National Occurrence Estimates Results
The techniques described above were applied to develop estimates of the proportions of
ground water and surface water systems with system mean concentrations above potential
regulatory levels. These estimates are presented separately for community water supply systems
and for non-transient, non-community water supply systems. The calculation of the confidence
intervals is presented in Section 6.4 below.
97
-------
Table 6-3a
Estimated National Exceedance Probabilities for Ground Water and Surface Water Community Water Systems
GW, Weighted Point
Estimate
95 % Confidence Interval
GW, Lognormal fit
SW, Weighted Point
Estimate
95% Confidence Interval
SW, Lognormal fit
GW & SW, Weighted Point
Estimate
Percent of Systems with Mean Arsenic Concentrations Exceeding (|j,g/L) of:
0.5
-
61.01
-
28.596
-
1
-
43.71
-
16.777
-
2
27.33
[26.57,
29.94)
27.55
9.817
[9.08,
25.46]
8.679
23.86
3
19.88
[19.25,
21.87]
19.69
5.566
[4.79,
20.63
5.552
17.05
5
12.09
[11.74,
13.04]
11.99
3.036
[1.8,
9.66]
2.960
10.30
10
5.30
[5.18,
5.91]
5.33
0.799
[0.52,
1.56]
1.117
4.41
15
3.06
[2.92,
3.45]
3.08
0.464
[0.23,
1.00]
0.592
2.55
20
2.00
[1.89,
2.29]
2.01
0.320
[0.13,
0.82]
0.366
1.67
25
1.42
[1.32,
1.64]
1.42
0.239
[0.08,
0.72]
0.247
1.19
30
1.05
[0.97,
1.23]
1.05
0.188
[0.06,
0.66]
0.178
0.88
40
0.64
[0.58,
0.77]
0.64
0.128
[0.03,
0.61]
0.103
0.54
50
0.43
[0.38,
0.52]
0.43
0.095
[0.02,
0.59]
0.067
0.36
Notes: The occurrence estimates presented in Tables 6-4 and 6-5 are based on the right-tailed ROS exceedance probability distributions denoted as
"Weighted Point Estimate".
GW - Ground water, lognormal regression yielded a mean and standard deviation in natural log coordinates of -0.251 and 1.583.
SW - Surface water, lognormal regression yielded a mean and standard deviation in natural log coordinates of-1.678 and 1.743.
The weighted point estimates were developed as described in Sections 6.1.1 to 6.1.4.
Neither a lognormal regression nor 95% confidence intervals were calculated for the combined GW and SW exceedance estimates.
The 95 % Confidence Intervals were derived as described in Section 6.5 using an earlier version of the database, and so may not be fully consistent with the
weighted point estimates and lognormal fit estimates.
98
-------
Table 6-3b
Estimated National Exceedance Probabilities for Ground Water and Surface Water
Non-Transient Non-Community Water Systems
GW, Weighted Point
Estimate
GW, Lognormal fit
SW, Weighted Point
Estimate
SW, Lognormal fit
Percent of Systems with Mean Arsenic Concentrations Exceeding (|j,g/L) of:
0.5
-
68.947
-
28.596
1
-
50.916
-
16.777
2
32.117
32.693
9.817
8.679
3
24.231
23.449
5.566
5.552
5
15.607
14.196
3.036
2.960
10
5.346
6.143
0.799
1.117
15
3.142
3.448
0.464
0.592
20
2.082
2.199
0.320
0.366
25
1.485
1.515
0.239
0.247
30
1.114
1.101
0.188
0.178
40
0.693
0.647
0.128
0.103
50
0.473
0.418
0.095
0.067
Notes: The occurrence estimates presented in Tables 6-6 and 6-7 are based on the right-tailed ROS exceedance probability distributions denoted as
"Weighted Point Estimate".
GW - Ground water, lognormal regression yielded a mean and standard deviation in natural log coordinates of 0.034 and 1.470.
SW - Surface water, lognormal regression yielded a mean and standard deviation in natural log coordinates of-1.678 and 1.743.
The weighted point estimates were developed as described in Sections 6.1.1 to 6.1.4.
The estimates for NTNCWS SW are copied from the estimates for CWS SW in Table 6-3a.
Confidence intervals were not estimated for the NTNCWS data.
99
-------
6.2.1 Community Water Supply Systems
Tables 6-4 and 6-5 present the projected arsenic occurrence levels in community water
supply systems with ground water and surface water sources, respectively. The data for the
probability distributions are derived exclusively from the AOED compliance monitoring
database, and the total number of systems is based on SDWIS data.
Under these estimates, 11,873 (CI: 11,543-13,007) ground water CWS systems are
estimated to have mean arsenic levels that exceed 2 |ig/L. The estimated number of exceeding
systems decreases rapidly at higher potential MCL alternatives. For example, 5,252 (CI: 5,100-
5,665) systems are predicted to have mean arsenic levels greater than 5 |ig/L, whereas 2,302 (CI:
2,250-2,567) systems are predicted to have mean arsenic levels greater than 10 |ig/L. While 869
(CI: 821-995) systems are predicted to have mean arsenic concentrations above 20 |ig/L, 187 (CI:
165-226) systems are estimated to have average arsenic levels greater than 50 |ig/L. It is worth
noting that the number of systems which are predicted to exceed the current MCL of 50 |ig/L is
significantly higher than number of systems which SDWIS indicates actually violate the MCL
(from January, 1996 to March, 1999, 15 ground water CWSs have violated the MCL).
In the United States, there are fewer surface water CWS than ground water CWS, and the
exceedance probabilities for surface water systems decrease more quickly as arsenic
concentrations rise than do the exceedance probabilities for ground water systems. As a result,
fewer surface water systems have mean arsenic levels above specific concentrations of interest
than ground water systems at corresponding arsenic concentrations. Under these estimates, 1,052
(CI: 973-2,730) surface water CWS systems are predicted to have arsenic concentrations above
2 |ig/L, and 325 (CI: 193-1,036) surface water CWS systems are predicted to have mean arsenic
concentrations greater than 5 |ig/L. A total of 86 (CI: 56-167) surface water CWS systems are
predicted to have mean arsenic concentrations that exceed 10 |ig/L, and 34 (CI: 14-88) are
predicted to have mean arsenic concentrations that exceed 20 |ig/L. Ten (CI: 2-63) surface water
CWS systems are predicted to have arsenic concentrations above 50 |ig/L.
6.2.2 Non-Transient, Non-Community Water Supply Systems
Tables 6-6 and 6-7 present the projected arsenic occurrence levels in non-transient, non-
community water supply systems with ground water and surface water sources, respectively. The
data for the ground water probability distributions are derived mainly from the non-transient non-
community water systems in the AOED compliance monitoring database, although for Alabama
and the Southeast region, AOED data from ground water community water supply systems was
used. The data for the surface water probability distributions are derived exclusively from the
community water systems in the AOED compliance monitoring database, as discussed in Section
6.1.5. The total number of systems is based on 1998 Baseline SDWIS data. Confidence
intervals were not developed for the ground water NTNCWS occurrence estimates in Table 6-6.
100
-------
Table 6-4
Estimated Arsenic Occurrence in U. S. Ground Water CWS
System Size (Population
Served)
<25
25-100
101-500
501-1,000
1,001-3,300
3,301-10,000
10,001-50,000
50,001-100,000
100,001-1,000,000
> 1,000,000
Total Systems
Lower 95% CI:
Upper 95% CI:
Total
Number of
Systems1
178
14025
14991
4671
5710
2459
1215
131
61
2
43443
Number of Systems with Mean Arsenic Concentrations2 (|J.g/L) of:
>2
49
3833
4097
1277
1561
672
332
36
17
1
11873
11543
13007
>3
35
2788
2980
929
1135
489
242
26
12
0
8636
8363
9501
> 5
22
1696
1812
565
690
297
147
16
7
0
5252
5100
5665
>10
9
743
795
248
303
130
64
7
3
0
2302
2250
2567
>15
5
429
459
143
175
75
37
4
2
0
1329
1269
1499
>20
4
281
300
93
114
49
24
3
1
0
869
821
995
>25
3
199
213
66
81
35
17
2
1
0
617
573
712
>30
2
147
157
49
60
26
13
1
1
0
456
421
534
>40
1
90
96
30
37
16
8
1
0
0
278
252
335
>50
1
60
64
20
25
11
5
1
0
0
187
165
226
Notes:
CI: confidence interval
1 Based on 1998 Baseline SDWIS data for purchased and non-purchased systems. Systems characterized as GW under the influence of SW are considered to
be surface water systems.
2 Based on national weighted point estimates presented in Table 6-3a.
3 Totals may not add up due to rounding of the number of systems to the nearest whole number.
101
-------
Table 6-5
Estimated Arsenic Occurrence in U. S. Surface Water CWS
System Size
(Population Served)
<25
25-100
101-500
501-1,000
1,001-3,300
3,301-10,000
10,001-50,000
50,001-100,000
100,001-1,000,000
> 1,000,000
Total Systems
Lower 95% CI:
Upper 95% CI:
Total Number
of Systems1
74
1001
1983
1219
2420
1844
1606
300
261
13
10721
Number of Systems with Mean Arsenic Concentrations2 (|J.g/L) of:
>2
7
98
195
120
238
181
158
29
26
1
1052
973
2730
>3
4
56
110
68
135
103
89
17
15
1
597
514
2212
> 5
2
30
60
37
73
56
49
9
8
0
325
193
1036
>10
1
8
16
10
19
15
13
2
2
0
86
56
167
>15
0
5
9
6
11
9
7
1
1
0
50
25
107
>20
0
3
6
4
8
6
5
1
1
0
34
14
88
>25
0
2
5
o
3
6
4
4
1
1
0
26
9
77
>30
0
2
4
2
5
3
3
1
0
0
20
6
71
>40
0
1
3
2
o
5
2
2
0
0
0
14
3
65
>50
0
1
2
1
2
2
2
0
0
0
10
2
63
Notes:
CI: confidence interval
1 Based on 1998 Baseline SDWIS data for purchased and non-purchased systems. Systems characterized as GW under the influence of SW are considered to
be surface water systems.
2 Based on national weighted point estimates presented in Table 6-3a.
3 Totals may not add up due to rounding of the number of systems to the nearest whole number.
102
-------
Table 6-6
Estimated Arsenic Occurrence in U. S. Ground Water NTNCWS
System Size (Population
Served)
<25
25-100
101-500
501-1,000
1,001-3,300
3,301-10,000
10,001-50,000
50,001-100,000
100,001-1,000,000
> 1,000,000
Total Systems
Total
Number of
Systems1
31
9732
7103
1996
696
62
15
0
0
0
19635
Number of Systems with Arsenic Concentrations2 (|J.g/L) of:
>2
10
3126
2281
641
224
20
5
0
0
0
6306
>3
8
2358
1721
484
169
15
4
4758
>5
5
1519
1109
312
109
10
2
0
0
0
3064
>10
2
520
380
107
37
3
1
0
0
0
1050
>15
1
306
223
63
22
2
0
0
0
0
617
>20
1
203
148
42
14
1
0
0
0
0
409
>25
0
145
105
30
10
1
0
0
0
0
292
>30
0
108
79
22
8
1
0
0
0
0
219
>40
0
67
49
14
5
0
0
0
0
0
136
>50
0
46
34
9
o
6
0
0
0
0
0
93
Notes:
1 Based on 1998 Baseline SDWIS data for purchased and non-purchased systems. Systems characterized as GW under the influence of SW are considered to
be surface water systems.
2 Based on national weighted point estimates presented in Table 6-3b.
3 Totals may not add up due to rounding of the number of the systems to the nearest whole number.
103
-------
Table 6-7
Estimated Arsenic Occurrence in U. S. Surface Water NTNCWS
System Size (Population
Served)
<25
25-100
101-500
501-1,000
1,001-3,300
3,301-10,000
10,001-50,000
50,001-100,000
100,001-1,000,000
> 1,000,000
Total Systems
Lower 95% CI:
Upper 95% CI:
Total
Number of
Systems1
5
280
314
107
80
23
5
1
1
0
816
Number of Systems with Arsenic Concentrations2 (|J.g/L) of:
>2
0
27
31
11
8
2
0
0
0
0
80
74
208
>3
0
16
17
6
4
1
0
0
0
0
45
39
168
>5
0
9
10
o
J
2
1
0
0
0
0
25
15
79
>10
0
2
3
1
1
0
0
0
0
0
7
4
13
>15
0
1
1
0
0
0
0
0
0
0
4
2
8
>20
0
1
1
0
0
0
0
0
0
0
3
1
7
>25
0
1
1
0
0
0
0
0
0
0
2
1
6
>30
0
1
1
0
0
0
0
0
0
0
2
0
5
>40
0
0
0
0
0
0
0
0
0
0
1
0
5
>50
0
0
0
0
0
0
0
0
0
0
1
0
5
Notes:
1 Based on 1998 Baseline SDWIS data for purchased and non-purchased systems. Systems characterized as GW under the influence of SW are considered to
be surface water systems.
2 Based on national weighted point estimates presented in Table 6-3b. These estimates were derived from CWS SW data.
3 Totals may not add up due to rounding of the number of systems to the nearest whole number.
104
-------
There are fewer NTNCWS in the United States than CWS, and therefore the numbers of
NTNCWS systems predicted to exceed specific levels are lower than the numbers of CWS that
would exceed similar levels. These projections indicate that 6,306 ground water NTNCWS have
system mean levels above 2 ng/L; 3,064 have system mean levels above 5 ng/L; and 1,050 have
system mean levels above 10 |ig/L. While 409 GW NTNCWS are predicted to have mean
arsenic levels above 20 |ig/L, 93 are expected to have arsenic levels in excess of the current
standard, 50 |ig/L. As there are few surface water NTNCWS in the United States, 80 systems are
predicted to exceed 2 |ig/L; 7 are predicted to exceed 10 ng/L; and 3 are predicted to exceed 20
|ig/L. One surface water NTNCWS facility is predicted to have system mean levels above 50
6.3 Comparisons of Occurrence Estimates
6.3.1 Comparison of AOED, NAOS, USGS, MWDSC, and Wade Miller Occurrence
Estimates at the National and Regional Level
In addition to the AOED occurrence results presented above, four additional studies have
developed national occurrence estimates for arsenic in drinking water. The additional studies
include the NAOS study, a national stratified random sampling of systems (Frey and Edwards,
1997), the USGS study of arsenic occurrence in ground water (Focazio et a/., 2000), the
Metropolitan Water District of Southern California national survey of arsenic occurrence in
surface water and ground water (MWDSC, 1993), and Wade Miller (1992), which is based in
part on the National Inorganic and Radionuclides Survey (NIRS) data for arsenic occurrence in
ground water. The databases and survey methodologies used in these studies are described in
Section 4.2 of this report. It is important to note that each of these occurrence estimates was
developed in a slightly different manner.
The AOED arsenic occurrence estimates (with details presented above) are based on
compliance monitoring data from more than 18,000 systems in 25 states. The NAOS occurrence
estimates are based on a stratified random sampling from representative groups defined by source
type, system size, and geographic location from 517 samples from approximately 500 systems.
The USGS analysis is based on ground water arsenic exceedance estimates for each county. The
MWDSC study is based on survey results from 140 large public water systems across 10 USEPA
regions. The Wade Miller study occurrence results are based on NIRS data, which are
occurrence results from a stratified, nationally-representative, random survey (defined by system
size) of approximately 980 public water systems served by ground water.
Figure 6- la presents arsenic occurrence estimates in surface water systems derived in this
study (for CWS systems) compared to results from the MWDSC and NAOS studies at
concentrations of 2, 5, 10, and 20 |ig/L. Figure 6-lb presents a comparison of arsenic occurrence
estimates from this study for ground water systems with the study findings of MWDSC, NAOS,
USGS, and Wade Miller at concentrations of 2, 5, 10, and 20 |ig/L.
105
-------
re
c
-------
Figure 6-1 b
Comparison of Arsenic Exceedance Probabilities, GW Systems
5 10
Arsenic Concentration (ug/L)
20
AOED
NAOS-Lg
MWDSC
USGS
NAOS-Sm
Wade Miller
107
-------
The values associated with the surface water findings shown graphically in Figure 6-la
are presented in Table 6-8a. Table 6-8a also includes 95% confidence intervals for the AOED
estimates that are developed in section 6-4. At arsenic concentrations of 2 |ig/L, exceedance
estimates in AOED are similar to, though somewhat higher than, those for NAOS-small systems,
NAOS-large systems, and MWDSC estimates. At 5 |ig/L, the AOED, NAOS-small, and NAOS-
large exceedance estimates are similar (3.0, 1.3, and 1.8, respectively), and somewhat above the
MWDSC estimate (0.0 percent). At 10 |ig/L, the AOED and NAOS-large exceedance estimates
are similar (0.8 and 0.6 percent, respectively), and above the NAOS-small and MWDSC
estimates (both at 0.0 percent). At 20 |ig/L, there are no reported data for NAOS, and AOED and
MWDSC estimates are low and similar (0.3 and 0.0 percent, respectively).
The values associated with the ground water findings shown graphically in Figure 6-lb
are presented in Table 6-8b. Table 6-8b also includes 95% confidence intervals for the AOED
estimates that are developed in section 6-4. At arsenic concentrations of 2 |ig/L, the exceedance
estimates for AOED are similar to the NAOS-small, NAOS-large, and the USGS estimates (with
the four estimates ranging from 28.8 percent for NAOS-large to 23.5 percent for NAOS-small),
with MWDSC and Wade Miller estimates somewhat lower. At 5 |ig/L, the AOED, NAOS-
small, NAOS-large, MWDSC, and USGS estimates are similar, ranging from 15.4 percent
(NAOS-large) to 12.1 percent (AOED), with the Wade Miller estimate at 6.9 percent. At 10
|ig/L, the AOED, NAOS-small, NAOS-large, MWDSC, and USGS estimates again are similar,
ranging from 7.6 percent (USGS) to 5.1 percent (NAOS-small), with Wade Miller estimates at
2.9 percent. And at 20 |ig/L, there are no reported data for NAOS, and the AOED, MWDSC,
and USGS exceedance estimates again are similar, ranging from 3.1 percent (USGS) to 1.9
percent (MWDSC), with Wade Miller study results at 1.1 percent. At all arsenic concentrations
of interest, the Wade Miller (1992) estimates are lower than those based on AOED, NAOS,
MWDSC, and USGS. The Wade Miller estimates relied upon MRS data for ground water (and
1978 CWSS, NOMS and RWS data for surface water). These data were highly censored, and
had relatively high reporting limits (generally 5 |ig/L). Therefore, these estimates are probably
less accurate than the estimates from the other studies and databases, particularly at
concentrations of 2 and 5 |ig/L.
On the national level, at arsenic concentrations ranging from 2 to 10 |ig/L, exceedance
estimates are similar for the AOED, NAOS (small and large systems), MWDSC, and USGS
studies. In addition, for the studies with data available (AOED, MWDSC, and USGS),
exceedance estimates are also similar at arsenic concentrations of 20 and 50 |ig/L. The fact that
these studies are in general agreement, especially given that each study used different data sets
and different methods for calculating exceedance estimates, indicates that the arsenic occurrence
estimates presented in Section 6.2 are reasonably representative at the national level. This
comparison of exceedance probabilities across studies suggests that arsenic occurrence
projections based on compliance monitoring data are relatively close to other recently developed
projections through the range of this comparison.
108
-------
Table 6-8a
Comparison of AOED, MWDSC, and NAOS
Surface Water Arsenic Occurrence Estimates
Percent of Systems with Mean Arsenic Concentrations
Exceeding Specific Limits (ug/L):
Studv
AOED
AOED 95% Conf. Int.
MWDSC
NAOS-Sm
NAOS-Lg
2
9.8
(9.1,25.4)
8.0
6.2
7.5
5
3.0
(1.8,9.7)
0.0
1.8
1.3
10
0.8
(0.5, 1.6)
0.0
0.0
0.6
20
0.3
(0.1,0.8)
0.0
NR
NR
50
0.1
(0.0, 0.6)
0.0
NR
NR
Note: NAOS-Sm includes systems serving < 10,000 people, NAOS-Lg includes systems
serving > 10,000 people. NR = Not Reported.
Table 6-8b
Comparison of AOED, MWDSC, NAOS, USGS and Wade Miller
Ground Water Arsenic Occurrence Estimates
Percent of Systems with Mean Arsenic Concentrations
Exceeding Specific Limits (ug/L):
Studv
AOED
AOED 95% Conf. Int.
MWDSC
NAOS-Sm
NAOS-Lg
USGS
Wade Miller
2
27.3
(26.6, 29.9)
19.2
23.5
28.8
25.0
17.4
5
12.1
(11.7, 13.0)
13.5
12.7
15.4
13.6
6.9
10
5.3
(5.2, 5.9)
5.8
5.1
6.7
7.6
2.9
20
2.0
(1.9,2.3)
1.9
NR
NR
3.1
1.1
50
0.4
(0.4, 0.5)
0.0
NR
NR
1.0
0.2
Note: NAOS-Sm includes systems serving < 10,000 people, NAOS-Lg includes systems
serving > 10,000 people. NR = Not Reported.
109
-------
Comparisons of estimated exceedances of arsenic concentrations at the regional level (for
ground water systems) are presented in Figure 6-2 (for 5 |ig/L) and Figure 6-3 (for 20 |ig/L).
Figure 6-2 shows that the AOED exceedance probabilities at 5 |ig/L are lower than the USGS
exceedance probabilities in all regions except New England (Region 1). Also, the AOED
estimates at 5 |ig/L are lower than NAOS estimates in all regions except New England (Region
1) and North Central (Region 6). The three studies agree reasonably well regarding exceedance
probabilities in the Midwest (Region 4) and South Central (Region 5) Regions. All three studies
indicated that the lowest exceedance probabilities occurred in the South East Region (Region 3),
and that the highest exceedance percentages at a concentration of 5 |ig/L occur in the West
Region (Region 7).
A similar comparison is presented in Figure 6-3 for regional exceedance probabilities
relative to the arsenic concentration of 20 |ig/L . These probabilities are significantly lower than
at 5 ng/L. The results of the three studies appear to agree less well at 20 jig/L than at 5 |ig/L.
As at 5 |ig/L, the South East Region (Region 3) shows low exceedances probabilities at 20 |ig/L,
and the West Region (Region 7) shows higher exceedance probabilities than most of the other
regions at 20 |ig/L. The highest exceedance probabilities at 20 |ig/L were indicated to be in the
North Central Region (Region 6) according to the USGS study, but the other studies did not
support this finding.
For surface water systems, USGS did not develop exceedance estimates, and NAOS
estimated that surface water systems in only two regions would have arsenic concentrations
above 5 |ig/L in finished surface water. These Regions include the South Central (Region 5),
where 7 percent of systems were predicted to exceed 5 |ig/L, and the North Central (Region 6),
where 12 percent of systems were predicted to exceed concentrations of 5 |ig/L. Surface water
exceedance estimates based on AOED are presented in Table 6-2. Based on these estimates,
some systems would have arsenic concentrations above 5 |ig/L in six Regions (exceedance
probabilities in parentheses): New England (9 percent); Mid Atlantic (0.1 percent); Midwest (1.2
percent); South Central (1 percent); North Central (3.8 percent), and West (7.4 percent). AOED-
based estimates indicate some systems will exceed arsenic concentrations of 20 jig/L in New
England (0.4 percent); Midwest (0.1 percent); North Central (0.1 percent); and the West (1.1
percent).
6.3.2 Comparison of AOED, Kennedy-Jenks, and Saracino-Kirby Occurrence Estimates
for California
In addition to the occurrence results presented above, two additional studies have
developed estimates for arsenic occurrence in California. The additional studies are a Kennedy-
Jenks Consultants report of the cost of compliance prepared for the Association of California
Water agencies (ACWA, 1996), and the Saracino-Kirby, Inc., report on arsenic occurrence and
conjunctive use in California (ACWA, 2000). The two studies used similar, though not identical,
sources of data. For the Kennedy-Jenks report, data sources included the ACWA Low-Level
Arsenic Database, data from the California Department of Water Resources (DWR) and the
Department of Heath Services (DHS), and California-specific data from the USGS. For the
110
-------
0
Figure 6-2
Comparison of Ground Water Systems Exceeding Arsenic Concentrations of 5ug/L
AOED
4
Region
NAOS
USGS
in
-------
Figure 6-3
Comparison of Ground Water Systems Exceeding Arsenic Concentrations of 20ug/L
4
Region
ACD
NAOS
USGS
-------
Saracino-Kirby report, data sources included the DWR, the DHS, California-specific data from
the USGS, data from the USGS stream water quality monitoring network, and data from the
Sacramento River Trace Metals Study.
The Saracino-Kirby study data included both raw and treated water sample results for
ground water occurrence data, but used only raw water sample analytical results for surface water
occurrence data. In the Kennedy-Jenks study, the type of water sample results included in the
data (raw versus treated) was not clearly defined. Though the report's use of USGS data and
other references suggest the data represent raw water, other references (such as results reported
from "surface water plants") imply the use of treated water sample results.
Figure 6-4a presents arsenic occurrence estimates in surface water systems from
California derived in this study (see Appendix B-l) compared to results from the Kennedy-Jenks
(ACWA, 1996) and Saracino-Kirby (ACWA, 2000) at concentrations of 2, 5, 10, 20, and 50
|ig/L. Figure 6-4b presents this same comparison for arsenic occurrence estimates in ground
water. For surface water arsenic occurrence, generally the Saracino-Kirby results are higher than
the AOED results which, in turn, are higher than the Kennedy-Jenks results. For ground water
arsenic occurrence, this same general pattern is apparent, although the results of the AOED and
Kennedy-Jenks are in close agreement (especially for arsenic values greater than or equal to 5
|ig/L). The values associated with the surface water findings shown graphically in Figure 6-4a
are presented in Table 6-9a, and the values associated with the ground water findings in Figure 6-
4b are presented in Table 6-9b.
6.4 UNCERTAINTY ANALYSIS
The uncertainty analysis described in this section was used to develop the confidence
intervals presented in Section 6.2. Due to project constraints, these estimates were based on an
earlier version of the database than the one used to develop the occurrence estimates. However,
the similarity of the estimated arsenic occurrence distributions for the two databases suggests that
the earlier confidence intervals can be used as a reasonable approximation to the expected results
from a revised simulation analysis. Confidence intervals were only computed for CWS systems.
6.4.1 Purpose of Uncertainty Analysis
An uncertainty analysis was conducted to determine the potential amount of error in the
exceedance probability estimates that are presented in Section 6.2. The ROS method that was
used to estimate system means has a potential drawback, in that it does not allow the calculation
of confidence intervals in a straightforward manner. Therefore, to determine 95-percent
confidence intervals, it was necessary to perform a statistical simulation to quantify the potential
sources of uncertainty. Three sources of uncertainty were identified and simulated: 1) sampling
variability, both within and between systems; 2) the fill-in of censored observations in the
estimation of system means; and 3) fitting of lognormal distributions to populations of system
means within each State. The combined effect of these uncertainties was modeled through a
simulation, and the results of this simulation were used to establish confidence intervals for the
arsenic occurrence estimates.
113
-------
Figure 6-4a
Comparison of CA Arsenic Exceedance Probabilities, SW Systems
AOED
Arsenic Concentration (ug/L)
Kennedy-Jenks
Saracino-Kirby
Figure 6-4b
Comparison of CA Arsenic Exceedance Probabilities, GW Systems
Arsenic Concentration (ug/L)
AOED
Kennedy-Jenks
Saracino-Kirby
114
-------
Table 6-9a
Comparison of California Arsenic Occurrence Estimates from AOED,
Kennedy-Jenks, and Saracino-Kirby Studies Surface Water
Percent of Systems Estimated to Exceed Arsenic Concentrations
fog/L):
Study
AOED
Kennedy-Jenks
Saracino-Kirbv
2
18.1
15
N/A
5
8.2
1
22
10
4.0
<1
9
20
1.7
<1
3.5
50
0.5
N/A
2.0
Table 6-9b
Comparison of California Arsenic Occurrence Estimates from AOED,
Kennedy-Jenks, and Saracino-Kirby Studies Ground Water
Percent of Systems Estimated to Exceed Arsenic
fog/L):
Studv
AOED
Kennedy-Jenks
Saracino-Kirbv
2
43.8
56
N/A
5
20.5
19
45
10
9.2
6
22
20
3.3
3
9.5
Concentrations
50
0.6
N/A
1.9
115
-------
The purpose of this uncertainty analysis is to provide a conservative estimate of the total
uncertainty attributable to the effects of sampling variability, the fill-in methods for censored
data, and the lognormality assumption. For this reason, the uncertainty analysis was deliberately
designed to increase the variability of the estimates. For example, the uncertainty analysis
lessened the number of detects required for the ROS method (instead of substitution) from 5 or
more down to 2 or more. Also the censored values were randomly substituted from a probability
distribution instead of using fixed values. However, there are various other sources of
uncertainty not addressed by the uncertainty analysis. Such sources include measurement error,
the use of regional exceedance distributions (which effectively assumes homogeneity within each
region), and uncertainties in the SDWIS database.
6.4.2 Uncertainty Analysis Methodology
Briefly summarized, the uncertainty simulation first simulated a population of systems,
then a mean for each system, and then estimated State exceedance probabilities based on the
simulated population of system means. Sets of exceedance probabilities obtained from many
repetitions of the simulation were then used to estimate non-parametric confidence intervals for
the concentrations of interest. In other words, bootstrap confidence intervals (Davison and
Hinkley, 1997) were computed for the estimation procedure described in Section 6.1. To
evaluate the influence of the lognormal model used to fill in censored observations, the entire
simulation was repeated twice: the first time, a uniform distribution was used to fill in censored
observations, and the second time, a Weibull distribution was used. Also, in order to evaluate
the influence of the right-tailed ROS method used to estimate the State exceedance probabilities,
the probabilities were computed using both ROS and an empirical method.
As mentioned above, the first step in the uncertainty analysis was to simulate a population
of systems. Systems were sampled with replacement from the list of systems in each State. The
second step was to simulate a system mean for each selected system. In this step, the combined
set of detected and non-detected concentrations in each system was re-sampled, such that if the
system had d detected and c non-detected concentrations, a random sample of size d+ c was
selected at random with replacement from the d + c observations. The calculation involves real-
space and log-space means and variances. The real-space mean and variance are the mean and
variance of the untransformed concentration values. The log-space mean and variance are the
mean and variance of the logarithms of the concentration values. When two or more distinct
detected concentrations were drawn, the log-space mean and variance were estimated by the ROS
method described in Appendix A; non-detects were filled in by drawing from a truncated
lognormal distribution with the given log-space mean and variance, such that the filled-in values
fell in the range from zero to the reporting level; and a real-space mean was computed for the
system. When all detects were equal or there was only one detect, each non-detect was filled in
by drawing from a uniform distribution from zero to the reporting level, and the real space
system mean was calculated. When there were no detects, the system was treated as a non-detect
at the modal reporting level.
116
-------
Using the above procedure, a population of system means was generated for each State.
From these populations, national exceedance probability distributions were then computed
exactly as before: State exceedance probability distributions were estimated by the right-tailed
ROS method, at concentrations of 2, 3, 5, 10, 15, 20, 25, 30, 40, and 50 jug/L; State distributions
were combined into regional distributions; and regional distributions were combined into a
national distribution, as described in Sections 6.1.2-6.1.4.
The entire procedure above was repeated 1,000 times, in order to generate 1,000
simulated national exceedance probability distributions. At each concentration (e.g., at 10 jug/L),
the 1,000 exceedance probability estimates were then sorted in increasing order, and the interval
from the mean of the 25th- and 26th-largest estimates to the mean of the 975th- and 976th-largest
estimates was taken as a nonparametric 95% confidence interval for the true exceedance
probability at that concentration.
The confidence intervals described above quantify the uncertainty in the exceedance
probability estimates due to the sampling variability of detected arsenic concentrations. They do
not include uncertainty due to the use of the lognormal distribution to fill in the censored
observations. In order to evaluate this additional uncertainty, the confidence intervals were
recomputed twice, using two alternative distributions to fill in the censored observations: the
two-parameter Weibull (a long-tailed) distribution and the uniform (a flat) distribution. In the
first repeat of the simulation, censored values that were previously filled in by drawing from a
lognormal distribution were instead drawn from a uniform distribution on the interval from zero
to the reporting level. In this case it was not necessary to fit the distribution. In the second
repeat of the simulation, censored values were replaced by draws from a truncated Weibull
distribution, with parameters estimated by a censored maximum likelihood algorithm in SAS. In
systems with only one detect or all detects equal, the maximum likelihood algorithm could not be
meaningfully applied, so non-detects were drawn from a uniform distribution. From each of the
uniform and Weibull simulations, 1,000 replicates were independently generated and confidence
intervals were computed as above.
Another source of uncertainly in the probability estimates is the use of the right-tailed
ROS method, with its lognormal assumption, to estimate State probability distributions.
In order to evaluate this uncertainty, exceedance probabilities and their confidence intervals were
computed using both the right-tailed ROS method, as described above, and an empirical method.
The empirical method estimates exceedance probabilities as the empirical fraction of detected
concentrations above each concentration level. Empirical estimates are computed only at
concentrations of 10 mg/L or higher (since the highest censoring limit in the AOED is 10 mg/L)
so that all non-detects may be unambiguously counted as less than the concentrations of interest.
6.4.3 Uncertainty Analysis Results
This discussion focuses on the national ground water and surface water uncertainty
analysis results. Uncertainty analyses were conducted for lognormal, uniform and Weibull
distributions, using both right-tailed ROS and empirical methods. The results of the right tailed
ROS uncertainty analysis for the national exceedance probabilities using a lognormal distribution
are found in Table 6-3a. The 95 percent confidence intervals are not centered directly on the
117
-------
point estimates for the national ground water and surface water exceedance probabilities. Rather,
the point estimates are located closer to the lower limits of the 95 percent confidence intervals.
In a comparison of the uncertainty analysis results from the three distributions, some variation is
evident among the distributions. In particular, the percentages for the lognormal distributions are
slightly higher than for the uniform or Weibull distribution at most concentrations. While there
are some differences among the confidence intervals for the lognormal, Weibull and uniform
distributions, these differences are minor. One exception is the surface water CIs at 5 mg/L.
Overall, these differences suggest that the analysis results are not particularly sensitive to the
method used to fill in the censored observations. Figures 6-4 and 6-5 depict plots of the 95
percent confidence intervals from the three distributions for ground water and surface water at
selected concentrations. While the widths of the CIs from the different distributions are
relatively similar, some minor variations are seen. For example, for the ground water analysis,
both the lognormal and Weibull CIs are wider than the uniform distribution CI. For surface
water, at low concentrations, the CI widths from the log normal distribution fall between those
for the Weibull and uniform CIs, but at higher concentrations are greater than the CI widths of
the Weibull and uniform distributions. All the CI widths for both surface water and ground
water narrow with increasing concentration. The 95 percent confidence intervals from the right-
tailed ROS lognormal distribution analysis were used to calculate the confidence intervals
presented in Tables 6-4, 6-5, and 6-7.
A comparison of the uncertainty analysis results for ground water and surface water
reveals different widths of the 95 percent confidence intervals. The national surface water
confidence intervals are significantly wider than the intervals for ground water exceedance
probabilities. For example, the CI at 5 mg/L for ground water is [11.74, 13.04], while the CI for
surface water at 5 mg/L is [1.8, 9.66]. The increased width is due to the smaller amount of data
available from surface water systems and the increased level of censoring of the surface water
data. Consequently, the results of the surface water uncertainty analysis are more sensitive to the
analytical methods employed.
In this uncertainty analysis, both a right-tailed ROS and an empirical analysis were
conducted for all distribution types. The results of the two methods differ. The empirical
percentages are lower than those for the right-tailed ROS analyses, for both surface water and
ground water. For ground water, the 95 percent confidence intervals from the empirical analysis
are wider than those from the right-tailed ROS analysis, for all distribution types. However, for
surface water, the confidence intervals for the right-tailed ROS method are wider than those from
the empirical analysis. In the log normal distribution, for both methods, the CIs become
increasingly narrow at higher concentrations. The lognormal results of these two analyses for
ground water and surface water at selected concentrations are presented in Figures 6-6 and 6-7.
118
-------
Figure 6-4 Comparison of Ground Water 95% Confidence Intervals by 3
Distribution Types
25
20
S
,£
«
.a
o
PLH
0>
u
C
«
«
o>
o>
u
*
^
10
[3]
[5]
1 1
[10]
1 1 1
[20]
[50]
1: Log-Normal
2: Uniform
3: Weibull
Arsenic Concentration in [ig/L
(presented as [x] in figure)
-------
Figure 6-5 Comparison of Surface Water 95% Confidence Intervals by 3
Distribution Types
25
20
§ 15
o
10
[3]
[5]
[10]
I I I
[20]
I I I
Arsenic Concentration in pg/L
(presented as [x] in figure)
[50]
1: Log-normal
2: Uniform
3: Weibull
-------
Figure 6-6 Comparison of Ground Water 95% CI by
Right-tailed ROS vs. Empirical Methods
[10]
.0
o
PLH
[20]
[30]
J_L
[50]
Arsenic Concentration in p.g/L
(presented as [x] in figure)
1: Right-tailed ROS
and
2: Empirical
-------
Figure 6-7 Comparison of Surface Water 95% Confidence Intervals by Right-
tailed ROS vs. Empirical Method
1 fi
>-^ 1 A
£
« 12-
5S '-z
*-
2
CQ 1
5 1
2
^08
a^ u-°
u
c
«
'O n R
5 u-°
o>
u
(^
W 04 -
00
n
1
[10]
[20]
[30] [50]
2
Arsenic Concentrations Oig/L) ^Right-tailed ROS
(presented as fx] in the figure)
VF l J & ' 2: Empmcal
-------
7. Intra-system Variability
This Chapter presents the results of analyses of intra-system variability that were
conducted using subsets of the AOED data. Section 7.1 defines the purpose of the intra-system
variability analysis. Section 7.2 provides an overview of the data that were available for these
analyses. Section 7.3 presents the methods by which the intra-system variability analyses were
conducted, and the results of the analysis. Section 7.4 briefly summarizes the intra-system
variability analyses.
7.1 Purpose of Analyses
The purpose of the intra-system analysis is to facilitate prediction of the number of
points-of-entry or POE that will be affected by various MCL alternatives. Compliance with the
arsenic standard is measured at the point-of-entry to the distribution system, and individual
systems can have multiple points-of-entry. Thus, one system may need to install one, two, three,
or more treatment systems or blend its water sources, depending upon its configuration and POE
mean arsenic levels. If arsenic levels in all POE in a system are below regulatory limits, it will
not need to install any treatment technologies for arsenic. Thus, arsenic levels in POE drive
compliance costs and risk reduction benefits more directly than do system mean arsenic levels.
The ideal analysis would be a survey designed to estimate arsenic levels in POE
throughout the United States. However, data are not currently available to support development
of such estimates: SDWIS does not catalogue information at the POE level; the earlier arsenic
occurrence studies focused on arsenic concentrations in systems rather than POE; and only a
third of the data sets in AOED link sample results to POE identification numbers. So the ideal
analysis is currently infeasible.
Since the ideal is infeasible, a reasonable and feasible alternative is to quantify a
relationship between POE means and system means where the data are suitable, and to use this
relationship, to estimate, from a population of system means, the number of POE means that are
likely to exceed specific regulatory alternatives. The analyses discussed in this Chapter were
designed to quantify the relationship between system means and POE means. This relationship
was quantified as an estimated coefficient of variation (CV), or relative standard deviation. This
is the CV of the distribution of the POE means for any given system. The CV values that were
calculated under these analyses are being applied under a different work assignment in a RIA to
estimate the number of POE that may exceed MCL alternatives. Under that work assignment,
the distribution of system mean arsenic concentrations, the probability distributions of POE for
systems of different sizes, and the CV of the relationship between the system mean arsenic
concentration and the POE arsenic concentration means are used to estimate the number of POE
in the United States that may exceed specific arsenic concentrations.
7.2 Available Data
To evaluate the relationship between POE mean arsenic levels and system mean arsenic
levels, it is necessary to have data sets that include distinct POE identifiers that associate each
123
-------
sample with the POE where it was collected. As indicated in Chapter 4, a total often States
provided data sets that included suitable POE identifiers:
Alabama
Arkansas
California
Indiana
Illinois
New Mexico
North Carolina
Oklahoma
Texas
Utah
Since the purpose of the intra-system variability analyses was to estimate variability
between POE, only those systems with two or more POE were used for these analyses. The
variable SRC_ID in the AOED INTRA database identifies the POE within each system, and a
unique POE identifier is the combination of PWSID and SRC_ID. A second restriction was that
all systems where there were any completely censored POE were not used. A completely
censored POE is a POE for which all the measurements are below the reporting limit. A third
restriction was to exclude data from surface water NTNCWS systems; the limited data for that
group of systems precluded the development of independent estimates of occurrence and intra-
system variability for surface water NTNCWS systems. Table 7-1 shows the numbers of
samples, non-detects, systems, and POE for systems with at least two POE that were used for the
intra-system variability analyses. This data subset contains a total of 4390 samples from 638
ground water CWS systems, 542 samples from 32 surface water CWS systems, and 237 samples
from 49 ground water NTNCWS systems. Note that none of the systems in Arkansas and only
one system in Alabama met the requirements for inclusion in these analyses. Estimates of intra-
system variability that were developed using these data are described in the following sections of
this report.
7.3 Analytical Methods and Results
The methods used to estimate intra-system variability was based on a fitted statistical
model that represented the two sources of variability for multiple measurements on the same
system: intra-system variability and within-POE variability. The intra-system variability is the
variability between POE for the same system. The within-POE variability is the temporal and
analytic variability between multiple measurements taken at the same POE. The first step of the
analysis was to estimate the concentrations for non-detects using the ROS method applied to the
data for each POE. This is described in section 7.3.1. The statistical model formulation and the
results of fitting that model are described in sections 7.3.2 and 7.3.2.
124
-------
Table 7-1
Data Used for Intra-system Variability Analyses
Type of
Water System
cws
cws
cws
cws
cws
cws
cws
cws
cws
Source
Type
GW
GW
GW
GW
GW
GW
GW
GW
GW
States
California
Illinois
Indiana
North Carolina
New Mexico
Oklahoma
Texas
Utah
All
Number of
Systems
208
22
1
5
194
51
103
54
638
Numbe
ofPOE
890
54
6
9
655
145
241
153
2153
Number of
Samples
2257
65
7
22
1120
183
402
334
4390
Number of
Non-Detects
479
0
0
12
75
3
34
59
662
cws
cws
cws
cws
cws
cws
cws
SW
SW
SW
SW
SW
SW
SW
Alabama
California
Illinois
New Mexico
Texas
Utah
All
1
5
1
5
16
4
32
4
21
2
11
37
12
87
95
209
4
22
169
43
542
89
84
1
3
69
10
256
NTNCWS
NTNCWS
NTNCWS
NTNCWS
NTNCWS
GW
GW
GW
GW
GW
California
New Mexico
Texas
Utah
All
14
25
6
4
49
28
125
13
10
176
53
151
19
14
237
5
2
0
2
9
125
-------
7.3.1 Estimation of Concentrations for Non-Detects
Arsenic concentrations for non-detects (values below the reporting limit) were estimated
using either substitution or the adapted ROS method described in Chapter 6. The actual values
for concentrations above the reporting limit were used. The method that was applied to fill-in the
non-detect concentrations depended on the number of detected values in the data set for the
particular POE. The method used for estimating non-detects was the same as that used in Chapter
6, except that the adapted ROS method was separately applied to the data for each POE, instead
of being applied to all the data for a system. As a result, arsenic concentrations for values below
the reporting limit were estimated as follows:
1-4 detects for the POE: Half the reporting level was substituted for non-detects.
All detects equal: Half the reporting level was substituted for non-detects.
5 or more detects: Adapted ROS was used to estimate the non-detect arsenic
concentrations as follows: the detects for the POE were plotted as in Helsel and Cohn
(1988); non-detects were uniformly plotted from zero to the estimated probability, from
the fitted log-normal model, at which Y is less than or equal to the censoring level; the
estimated concentration for each non-detect was estimated from the fitted log-normal
distribution.
Note that every POE in the database used for the intra-system variability analyses had at least one
detected value.
7.3.2 Log-Normal Mixed Model
The approach used to estimate the CV for intra-system variability in arsenic
concentrations relies on a log-normal mixed model. The log-normal model assumes that for a
given POE, in a given system, the logarithm of the arsenic concentration is normally distributed,
with a log-mean that depends on the system and the POE, and a constant log-variance, V, that is
the same for every system and POE. The terms "log-mean' and "log-variance" are used to define
the log-space mean and variance, i.e., the mean and variance of the logarithms of the
concentrations. The log-variance V measures the within-POE residual variation, which
represents the temporal and analytic variability of the multiple measurements for the same POE.
Temporal variability is the variation between measurements made at different dates or times.
Analytic variability is the variability between arsenic concentration measurements made by the
same or different measuring instruments. The other part of the model is the intra-system
variation between the POE, which is the main focus of the analysis. The statistical model
assumes that for any given system, the POE log-means are normally distributed around the
system log-mean. The variance of this distribution of the POE log-means summarizes the intra-
system variability.
The statistical model used to describe the intra-system and within-POE variability
therefore follows the equation:
Log(arsenic) = System Log-mean + POE Effect + Residual Error
126
-------
where:
System Log-mean
POE Effect
Residual Error
the log-mean for the given system;
POE log-mean - System log-mean, which is normally distributed
with a mean of zero and a variance Var(POE) = o2; and
Log(arsenic) - POE log-mean, which is normally distributed with a
mean of zero and a variance Var(residual) = V.
The POE Effect is the intra-system variability and the Residual Error is the combined temporal
and analytic variability. This model was fitted by the method of maximum likelihood estimation,
separately for each combination of water system type and source type.
Under this statistical model, the POE means will be log-normally distributed with a log-
variance of o2, and the CV of the intra-system variability is given by the equation:
CV(POEMean)=
-1 XlOO% .
7.3.3 Results
The estimated model coefficients and their standard errors are given in Table 7-2. The
last two columns give the CV of the intra-system variability and its standard error. The estimated
CVs are 37 % (standard error = 2 %) for ground water CWS systems, 53 % (standard error = 9%)
for surface water CWS systems, and 25 % (standard error = 6 %) for ground water NTNCWS
systems. For the surface water NTNCWS systems, the estimated CV for the surface water CWS
systems was used.
Table 7-2
Results of Intra-System Variability Analyses
Type of
Water
System
CWS
CWS
NTNCWS
NTNCWS1
Source
Type
GW
SW
GW
SW
Intra-system
Variability
Var
(POE)
0.1287
0.2446
0.0614
0.2446
Standard
Error
0.01274
0.07693
0.0266
0.07693
Within-POE
Variability
o2
0.9602
0.3545
0.2064
0.3545
Standard
Error
0.02352
0.02287
0.02841
0.02287
Intra-system
Variability CV
CV
37%
53%
25%
53%
Standard
Error
2%
9%
6%
9%
'Intra-system variability estimates for surface water NTNCWS systems were copied
from the estimates for surface water CWS systems.
127
-------
7.4 Summary of Intra-system Analyses
A statistical model was used to estimate the intra-system variability of arsenic
concentrations at different POE for the same system. This model accounts both for the intra-
system variability between POE and for the temporal and analytic variability of the multiple
measurements at the same POE. The results tabulated in Table 7-2 can be used in a RIA to
estimate variability between POE. In conjunction with the arsenic national occurrence estimates
given in Chapter 6, the intra-system variability estimates can be used to estimate distributions of
POE mean arsenic concentrations in public water supply systems.
128
-------
8. Temporal Variability
8.1 Purpose of Analysis
The purpose of the temporal variability analysis is to examine the variability of arsenic
concentrations over time in a source. This information may be used in the Regulatory Impact
Analysis to determine the probability that a single arsenic sample or the average arsenic level in a
given source would exceed regulatory levels and to estimate national annual monitoring costs.
8.2 Available Data and Results
There were insufficient data in the AOED to analyze the temporal variability of arsenic
concentrations. However, USGS had data from 353 wells with 10 or more arsenic analyses
collected over different time periods. USGS examined its raw water arsenic data to assess the
variability of arsenic levels over time and to determine whether there are temporal trends
(Focazio, et a/., 2000). These wells were used for various purposes, such as public supply,
research, agriculture, industry, and domestic supply, and encompassed non-potable and potable
water quality. USGS conducted a regression analysis of arsenic concentration and time for each
well and found that most of the wells had little or no temporal trend (low r-squared values when
arsenic concentrations were regressed with time). Arsenic levels for most of the wells probably
do not consistently increase or decrease over time. In addition, USGS examined the relationship
of well depth and temporal variability by analyzing the relationship between standard deviation
and well depth for wells with mean arsenic concentrations less or equal to 10 |ig/L. They found
no relationship (Figure 8-1).
To determine the extent of the temporal variability, EPA analyzed the CVs for the mean
arsenic level in the wells. 116 wells had a CV and standard deviation of zero. Most of these
wells consistently had arsenic concentrations below the detection limit of 1 |ig/L. EPA examined
the CVs for the other wells in relation to the mean arsenic level and found a relatively constant
CV on the lognormal scale (Figure 8-2) The geometric mean of the CVs, excluding the CVs that
are zero, is 0.388 or 38.8%. The range of the non-zero CVs is 5.5% to 236.3% and the mode is
27.6%. Focazio, et al. (2000) listed several factors that may contribute to this variability,
including natural variability in geochemistry or source of contamination, sampling technique, and
changes in pumping over time.
129
-------
100.000
10.000
)
g
U-*
CO
1.000
Q
g
I 0.100
I
W
0.010
0.001
1.00
Figure 8-1: Standard Deviation in Relation to Well Depth
* * **
*
10.00
100.00
1000.00
10000.00
Well Depth (feet)
Note: Samples restricted to mean arsenic concentrations less than or equal to 10 ug/L
Figure adapted from Focazio, et al., in press.
130
-------
Figure 8-2: Coefficient of Variation in Relation to Mean Arsenic Level
1000.00
100.00
o
?
.2
'c
ra
c
.0)
'o
I
o
O
10.00
1.00
0.10
0.01
1.00
* * S* *?
4 V
10.00 100.00
mean arsenic level (ug/L)
Note: Does not include wells with zero CVs.
1000.00
10000.00
131
-------
This page intentionally left blank
132
-------
9. References
Agency for Toxic Substances and Disease Registry, 1998. Draft Toxicological Profile for
Arsenic. Prepared for the U.S. Department of Health and Human Services, ATSDR, by the
Research Triangle Institute. August, 1998.
Agency for Toxic Substances and Disease Registry, 1997. List of Top 20 Hazardous Substances.
Available on the Internet at http://www.atsdr.cdc.gov/. ATSDR, Research Triangle Park, NC.
November, 1997.
Aschbacher, Peter W. and VernonJ. Feil. 1991. Fate of ] 14C] Arsanilic Acid in Pigs and
Chickens. J. Agric. Food Chem. 1991, 88, 146-149.
Azcue, J.M. and J.O. Nriagu. 1994. Arsenic: Historical Perspectives. In: Arsenic in the
Environment, Part I: Cycling and Characterization, Edited by J. O. Nriagu. John Wiley and
Sons, Inc. New York, NY. pp 1-16.
Berger, BJ. and A.H. Fairlamb. 1994. High-performance liquid chromatographic method for the
separation and quantitative estimation of antiparasitic melaminophenyl arsenical compounds.
Trans. R. Soc. Trop. Med. Hyg. 88:357-359.
Budavari, S., MJ. O'Neil, A. Smith, and P.E. Heckelman. 1989. Merck Index, ed. Merck &
Company.
Clifford, D. and Z. Zhang. 1994. Arsenic Chemistry and Speciation. Paper presented at the
American Water Works Association Annual Conference. New York, NY. June 19-23, 1994.
Clifford, D. 1986. Removing dissolved inorganic contaminants from water. Environ. Sci.
Techol.,20: 1072-1080.
Cohen, A.C. 1991. Truncated and Censored Samples: Theory and Applications, New York:
Marcel Dekker.
Cullen, W.R. and K.J. Reimer. 1989. Arsenic speciation in the environment. Chem. Rev., 89:
713-764.
Davenport, J.R. and F.J. Peryea. 1991. Phosphate fertilizers influence leaching of lead and
arsenic in a soil contaminated with lead and arsenic in a soil contaminated with lead arsenate.
Water, Air and Soil Pollution. 57/58: 101-110.
Davison, A.C. and D.V. Hinkley. 1997. Bootstrap methods and their application. Cambridge
University Press.
Federal Register. Vol. 58, No. 234. (58 FR 64579) Inorganic Arsenicals; Conclusion of Special
Review. (December 8, 1993) 64579-64582.
133
-------
Fergusson, J. 1990. The Heavy Elements: Chemistry. Environmental Impact and Health Effects.
Oxford: Pergamon Press, 1990.
Focazio, M., A. Welch, S. Watkins, D. Helsel, and M. Horn. 2000. A Retrospective Analysis of
the Occurrence of Arsenic in Ground Water Resources of the United States and Limitations in
Drinking Water Supply Characterizations. U.S. Geological Survey. Water Resources
Investigations Report 99-4279.
Frey, M.M. and M.A. Edwards. 1997. Surveying arsenic occurrence. J. AWWA, 89: 105-117.
Fuhrer, G. J., DJ. Cain, S.W. McKenzie, J.F. Rinella, J.K. Crawford, K.A. Skach, and
M.I. Hornberger, and M.W. Gannett. 1996. Surface-Water-Quality Assessment of the Yakima
River Basin in Washington: Spatial and Temporal Distribution of Trace Elements in Water,
Sediment, and Aquatic Biota, 1987. U. S. Geological Survey Open-File Report 95-440, 190 p.
Gilliom and Helsel, "Estimation of Distributional Parameters for Censored Trace Level Water
Quality Data 1." Estimation Techniques, Water Resources Research, 22(2), 135-146, 1986.
Gulledge, J.H. and J.T. O'Connor. 1973. Removal of Arsenic (V) from water by adsorption on
aluminum and ferric hydroxides. J. AWWA., 65: 548-552.
Helsel, D.R. and T.A. Cohn. 1988. "Estimation of Descriptive Statistics for Multiply Censored
Water Quality Data." Wat. Resour. Res., 24, pp. 1997-2004.
Hess, R. E. and R.W. Blanchar. 1977. Dissolution of arsenic from waterlogged and aerated soil.
SoilSci. Soc. Am. J., 41(5): 861-865.
Hinkle, S.R. and DJ. Polette. 1999. Arsenic in Ground Water of the Willamette Basin, Oregon.
U.S. Geological Survey Water-Resources Investigations Report 98-4205, 32 p.
Irgolic, KJ. 1994. Determination of total arsenic and arsenic compounds in drinking water, pp.
51-60 in Arsenic: Exposure and Health, W.R. Chappell, C.O. Abernathy, and C.R. Cothern, eds.
Northwood, U.K.: Science and Technology Letters.
Isaac, R.A., S.R. Wilkenson, and J.A. Stuedemann. 1978. Analysis and fate of arsenic in broiler
litter applied to coastal Bermuda grass and Kentucky-31 tall fescue. In : D.C. Adriano and I.L.
Brisbin, Jr. (eds.), Proc. Symp. On Environmental Chemistry and Cycling Processes, U.S.
Dept. of Energy, pp. 207-220.
Jekel, M.R. 1994. Removal of Arsenic in Drinking Water Treatment. Chapter 6 in Nriagu, J.O.,
Ed., Arsenic in the Environment Part I: Cycling and Characterization. New York: John Wiley &
Sons, Inc. pp 119-132.
Jordan, D., M. McClelland, A. Kendig, and R. Frans. 1997. Monosodium methanearsonate
influence on broadleaf weed control with selected postemergence-directed cotton herbicides. J.
Cotton Sci., 1: 72-75.
134
-------
Kennedy Jenks Consultants. 1996. Final Report. Cost of Compliance with Potential Arsenic
MCLs. Prepared for the Association of California Water Agencies. November 27, 1996.
Kirk-Othmer Encyclopedia of Chemical Technology, 1992. 4th Edition, Volume 3. New York,
New York, John Wiley and Sons, Inc.
Konefes, J.L. and M.K. McGee. 1996. Old cemeteries, arsenic, and health safety. Cult. Resour.
Mgmt. 19(10): 15-18.
Kroll, C.N. and J.R. Stedinger. 1996. Estimation of Moments and quantiles using censored data.
Water Resources Research. 32(4): 1005-1012.
Loebenstein, J.R. 1994. The Materials Flow of Arsenic in the United States. U.S. Department
of the Interior, Bureau of Mines, pp 1-12.
Longtin, J.P. 1988. Occurrence of Radon, Radium, and Uranium in Groundwater. J. AWWA.,
80(7):84.
Maclean, K.S. and W.M. Langille. 1981. Arsenic in orchard and potato soils and plant tissue.
Plant Soil 61(3): 413-418.
Marvinney, R.G., M.C. Loiselle, J.T. Hopeck, D. Braley, and J.A. Krueger. 1994. Arsenic in
Maine Ground Water: An Example From Buxton, Maine. 1994 Focus Conference on Eastern
Regional Ground Water Issues, pp. 701-714.
Mok, W.M. and C.M. Wai. 1994. Mobilization of arsenic in contaminated river waters. In:
Arsenic in the Environment, Parti: Cycling and Characterization, Edited by J. O. Nriagu. John
Wiley and Sons, Inc. New York, NY. pp 99-115.
Mok, W.M., and C.M. Wai. 1989. Distribution and mobilization of arsenic species in the creeks
around the Blackbird mining district, Idaho. Wat. Resour. Res., 23(1): 7-13.
Moody, J.P. and R.T. Williams. 1964. The Fate of Arsanilic Acid and Acetylarsanilic Acid in
Hens. FoodCosmet. Toxicol. 1964, 2:687-693.
Morrison. 1975. Distribution of Arsenic from Poultry Litter in Broiler Chickens, Soils, and
Crops. J. Ag. and Food Chem., 23 (4).
National Academy of Sciences (NAS). 1977. Arsenic. Medical and biological effects of
environmental pollutants. Washington, D.C. pp. 332.
National Research Council (NRC). 1999. Arsenic in Drinking Water. National Academy Press.
Washington, D.C. pp. 310.
135
-------
Nimick, D.A., J.N. Moore, C.E. Dalby, and M.W. Savka. 1998. The fate of geothermal arsenic
in the Madison and Missouri Rivers, Montana and Wyoming. Wat. Res. Research, 34(11): 3051-3067.
Ogden, P.R. 1990. Arsenic behavior in soil in and ground water at a Superfund site. In:
Superfund '90. Hazardous Materials Control Res. Inst, Silver Spring, MD. pp 123-127.
Onishi, Y. 1978. Arsenic. In Wedephol, K. H., ed., Handbook of Geochemistry. Berlin. 11/3:
33-A-l to 33-0-1.
Pacyna, J.M., M.T. Scholtz, and Y.F. Li. 1995. Global budget for trace metal sources. Environ.
Rev., 3: 145-159.
Peryea, FJ. and R. Kammereck. 1997. Phosphate-enhanced movement of arsenic out of lead
arsenate-contaminated topsoil and through uncontaminated subsoil. Water, Air and Soil
Pollution. 93(1-4): 243-254.
Peryea, FJ. 1991. Phosphate-induced release of arsenic from soils contaminated with lead
arsenate. SoilSci. Soc. Am. J. 55: 1301-1306.
Peters, S.C., J.D. Blum, B. Klaue, and M.R. Karagas. 1999. Arsenic Occurrence in New
Hampshire Drinking Water. Environ. Sci. & Technol. 33(9), 1328-1333.
Reese, R.G., Jr. 1999. Arsenic. In: United States Geological Survey Minerals Commodities
Summaries, 1999. Fairfax, VA.
Reese, R.G., Jr. 1998. Arsenic. In: United States Geological Survey Minerals Yearbook, 1998.
Fairfax, VA.
Robertson, F.N. 1989. Arsenic in ground-water under oxidizing conditions, south-west United
States. Environ. Environ. Geochem. and Health. 11(3/4): 171-186.
Rubel, F. and S.W. Hathaway. 1987. Pilot study for the removal of arsenic from drinking water
at Fallen, Nevada, Naval Air Station. J. AWWA., 79: 61-65.
Saracino-Kirby Inc. 2000. Arsenic Occurrence and Conjunctive Management in California.
Prepared for the Association of California Water Agencies. September, 2000.
Science Applications International Corporation (SAIC). 1999. Geometries and Characteristics
of Public Drinking Water Systems. Draft Document. Prepared for the USEPA OGWDW under
USEPA Contract 68-C6-0059. May, 1999.
Shen, Y.S. 1973. Study of arsenic removal from drinking water. J. AWWA., 65: 543-548.
Simo, J.A., P.G. Freiberg, K.S. Freiburg. 1996. Geologic constraints on arsenic in ground water
with applications to ground water modeling: Ground water Research Rept. WRC GRR 96-01,
University of Wisconsin, pp.60.
136
-------
Smith, S.C., J.G. Britton, J.D. Enis, K.C. Barnes, and K.S. Lusby. 1992. Mineral levels of
broiler house litter and forages and soils fertilized with Litter. (In review).
Stauffer, R.E. and J.M. Thompson. 1984. Arsenic and antimony in geothermal waters of
Yellowstone National Park, Wyoming, USA. Geochim. Cosmochim. Ada, 48: 2547-2561.
Kroll, C.N. and J.R. Stedinger. 1996. Estimation of moments and quantiles using censored data,
Water Resources Research, 32(4), 1005-1012.
Steevens, D.R., L.M. Walsh, and D.R. Keeney. 1972. Arsenic residues in soil and potatoes from
Wisconsin potato fields - 1970. Pestic. Monit. J., 6(2): 89-90.
Thompson, W.T. 1973. Agricultural Chemicals. Book 1-Insecticides. Thompson Publications,
Indianapolis. 300 pp.
United States Environmental Protection Agency. 1999a. Toxic Release Inventory Data Report
for 1999. USEPA Office of Prevention, Pesticides, and Toxic Substances. Washington, D.C.
United States Environmental Protection Agency. 1999b. US EPA List of Pesticides Banned and
Severely Restricted in the United States, Original U.S. Nominations to the U.N. PIC Procedure.
USEPA Office of Prevention, Pesticides, and Toxic Substances. Washington, D.C. At
http://www.epa.gov/oppfeadl/international/piclist.htm February 1999.
United States Environmental Protection Agency, 1998a. International Pesticide Notice: USEPA
Cancels the Last Agricultural Use of Arsenic Acid in the United States. USEPA Office of
Prevention, Pesticides, and Toxic Substances. Washington, D.C. At
http://www.epa.gov/oppfeadl/17b/r2.htm. February 1998.
United States Environmental Protection Agency. 1998b. Locating and Estimating Air Emissions
From Sources of Arsenic and Arsenic Compounds. USEPA Office of Air Quality Planning and
Standards, Research Triangle Park, NC. Document No. EPA-454-R-98-013. June 1998.
United States Environmental Protection Agency. 1997a. Chromated Copper Arsenicals (CCA)
and Its Use as a Wood Preservative. USEPA Office of Prevention, Pesticides and Toxic
Substances, Washington, D.C. Available at http://www.epa.gov/opp00001/citizens/lfile.htm.
May 1997.
United States Environmental Protection Agency. 1997b. Pesticide Industry Sales and Usage,
1994 and 1995 Market Estimates. USEPA Office of Prevention, Pesticides and Toxic
Substances, Washington, D.C. Document No. EPA-733-R-97-002. August 1997.
United States Environmental Protection Agency. 1995. Office of Pesticide Programs Annual
Report for 1994. USEPA Office of Prevention, Pesticides and Toxic Substances, Washington,
D.C. Document No. EPA-735-R-95-001. January 1995.
137
-------
United States Environmental Protection Agency. 1993. Draft Drinking Water Criteria
Document on Arsenic. Prepared by Life Systems, Inc. U.S. Environmental Protection Agency,
Office of Drinking Water, Washington, D.C.
United States Environmental Protection Agency. 1984. Wood Preservative Pesticides: Creosote,
Pentachlorophenol, Inorganic Arsenicals; Position Document 4. Office of Pesticides and Toxic
Substances. EPA 540/9-84/003. July, 1984.
United States Environmental Protection Agency. 1975. Interim Primary Drinking Water
Regulations. Federal Register, December 24, 1975, 59,556.
Wade Miller Associates, Inc. 1992. Occurrence Assessment for Arsenic in Public Drinking
Water Supplies. Prepared for USEPA under Contract 68-CO-0069. September, 1992.
Wade Miller Associates, Inc. 1989. Estimated National Occurrence and Exposure to Arsenic in
Public Drinking Water Supplies. Prepared for USEPA under Contract 68-01-7166. June, 1989.
Waslenchuk, D. 1979. The geochemical controls on arsenic concentrations in southeastern
United States rivers. Chem. Geol. 24: 315-325.
Welch, A.H.; M. Lico, and J. Hughes. 1988. Arsenic in ground water of the western United
States. Groundwater. 26(3): 333-347.
Westjohn, D.B., A. Kolker, W.F. Cannon, and D.F. Sibley. 1998. Arsenic in ground water in the
"Thumb Area" of Michigan. The Mississippian Marshall Sandstone Revisited, Michigan: Its
Geology and Geologic Resources, 5th symposium, pp. 24-25.
Woolson, E.A., J.H. Axley, and P.C. Kearney. 1973. The chemistry and phytotoxicity of arsenic
in soils: II. Effects of time and phosphorous. Soil Sci. Soc. Am. Proc. 37(2): 254-259.
Yan-Chu, H. 1994. Arsenic Distribution in Soils. Chapter 2 in Nriagu, J.O., Ed., Arsenic in the
Environment Part I: Cycling and Characterization. New York: John Wiley & Sons, Inc. pp 17-
49.
138
------- |