July 19, 1995
EPA-SAB-DWC-95-015
Honorable Carol M. Browner
Administrator
U.S. Environmental Protection Agency
401 M Street, SW
Washington, DC 20460
Subject: Science Advisory Board (SAB) Review of Issues Related to the
Regulation of Arsenic in Drinking Water
Dear Ms. Browner:
The Office of Water (OW) is developing the technical foundation for the
regulation of arsenic in drinking water. On August 18-19, 1994, the Drinking Water
Committee (DWC) of the Science Advisory Board (SAB) reviewed several issues of
concern relative to the development of this regulation. Specifically, the Committee
reviewed Agency activities regarding: a) the methods being considered by the Agency
to estimate national occurrence levels of arsenic in drinking water; b) the Agency's
approach to determining Best Available Technology (BAT) for arsenic removal from
drinking water; c) the basis for selection of quantitation detection limits for arsenic; and
d) the Agency's approach to the development of a "decision tree" for a regulatory
impact assessment for arsenic in drinking water.
Although the estimation of the health effects from exposure to arsenic were not
the subject of this review, the Committee wishes to reiterate its concern that the
available data do not allow the estimation of the magnitude of risks associated with
environmental exposures to arsenic, and that a research strategy to resolve these
issues is necessary. We would like to call to your attention earlier reports that
addressed this issue (SAB, 1989, 1992, 1994).
With regard to the analytical issues, the Committee recommended that the
Practical Quantitation Limit (PQL) for arsenic be set using acceptance limits that are
similar to those used for other inorganic substances with currently approved analytical
methods. It was also the Committee's opinion that risk management issues should not
-------
enter into the procedure for determining the allowable error for the PQL. Lastly, if the
PQL set using the acceptance limits described above does not meet EPA's needs, the
Agency should consider and develop alternative analytical methods.
With regard to the use of censored survey data to estimate the occurrence of
arsenic in drinking water, the Committee questioned the suitability of older data
sources, such as those from the National Inorganics and Radionuclides Survey, and
recommended that they be abandoned if sufficient new data become available. To use
existing data, the Committee recommended the use of certain statistical methods that
are described in some detail in the attached report.
This report also addresses a number of issues concerning Best Available
Technology for arsenic removal from drinking water, including the use of ozone as a
pre-oxidant for arsenic, counter-current flow ion exchange processes, reverse osmosis,
and nanofiltration. However, the Committee felt that although the problems with
nanofiltration technology should be solvable, the technology needs to be developed
further if this is to be considered BAT for arsenic.
Finally, the Committee supported the concept of a decision tree to guide the
process of determining the best approach to provide drinking water with acceptable
levels of arsenic. They recommended, however, that the decision tree be expanded to
incorporate a more holistic "optimization" approach, that is; one that seeks to jointly
minimize the levels of all toxic chemicals and microbial agents.
We look forward to your response to the recommendations in the attached report
and to further examination of the difficult technical issues underlying the regulation of
arsenic in drinking water.
Sincerely,
Dr. Genevieve M. Matanoski, Chair
Executive Committee
Science Advisory Board
Dr. Verne A. Ray, Chair
Drinking Water Committee
Science Advisory Board
-------
NOTICE
This report has been written as a part of the activities of the Science Advisory
Board, a public advisory group providing extramural scientific information and advice to
the Administrator and other officials of the Environmental Protection Agency. The
Board is structured to provide balanced, expert assessment of scientific matters related
to problems facing the Agency. This report has not been reviewed for approval by the
Agency and, hence, the contents of this report do not necessarily represent the views
and policies of the Environmental Protection Agency, nor of other agencies in the
Executive Branch of the Federal government, nor does mention of trade names or
commercial products constitute a recommendation for use.
-------
ABSTRACT
The Drinking Water Committee of the Science Advisory Board (SAB) met on
August 18-19, 1994 to review several issues relevant to the Agency's regulation of
arsenic in drinking water; namely analytical methods, best available technology,
national occurrence, and the development of a "decision tree" for a regulatory impact
assessment for arsenic in drinking water.
In its report, the Committee highlights earlier SAB reports regarding the health
effects of arsenic, and makes recommendations regarding the adoption of practical
quantitation limits (PQL) for arsenic, the exclusion of risk management issues from the
procedures for determining the allowable error for the PQL, and for dealing with PQL
error ranges. In addition, the Committee questioned the suitability of older data
sources for estimating the occurrence of arsenic, and recommended the use of certain
statistical methods that are described to derive such estimates from existing data.
The Committee also addressed a number of issues concerning Best Available
Technology (BAT) for arsenic, including the use of ozone as a pre-oxidant for arsenic,
counter-current flow ion exchange processes, reverse osmosis, and nanofiltration.
Finally, the Committee supported the concept of a decision tree to guide the
process of determining the best approach to provide drinking water with safe levels of
arsenic, with some reservations.
Key Words: Arsenic; As; Best Available Technology; BAT; PQL; water quality
-------
ENVIRONMENTAL PROTECTION AGENCY
SCIENCE ADVISORY BOARD
DRINKING WATER COMMITTEE
Arsenic Review Panel
CHAIR
Dr. Verne A. Ray, Medical Research Laboratory, Pfizer Inc., Groton, CT
MEMBERS
Dr. Richard J. Bull, College of Pharmacy, Washington State University, Pullman, WA
Dr. Judy A. Bean, University of Miami School of Medicine, Miami, FL
Dr. Keith E. Cams, Electric Power Research Institute,
Dr. Lenore S. Clesceri, Rensselaer Polytechnic Institute, Materials Research Center,
Troy, NY
Dr. Anna Fan, State of California, OEHHA/PETS, Berkeley, CA
Dr. Charles Gerba, University of Arizona, Tucson, AZ
Dr. Charles C. Johnson, Jr., Bethseda, MD
Dr. Curtis Klaassen, University of Kansas Medical Center, Kansas City, KS
Dr. Edo D. Pellizzari, Research Triangle Institute, Research Triangle Park, NC
Dr. Richard H. Reitz, McClaren Hart, Flint, Ml
Dr. Vernon L. Snoeyink, Department of Civil Engineering, University of Illinois,
Urbana, IL
Dr. James M. Symons, Department of Civil and Environmental Engineering, University
of Houston, Houston, TX
Dr. Marylynn Yates, University of California, Riverside, CA
CONSULTANTS
Dr. Kenny S. Crump, ICF Kaiser, Ruston, LA
Dr. Robert J. Taylor, Houston, TX
Dr. Rhodes Trussell, Montgomery Watson Consulting Engineers, Pasadena, CA
SCIENCE ADVISORY BOARD STAFF
Mr. Manuel R. Gomez, Designated Federal Official, Science Advisory Board (HOOF),
U.S. EPA, 401 M Street, SW, Washington, DC 20460
Mr. Robert Flaak, Designated Federal Official, Science Advisory Board (HOOF),
U.S. EPA, 401 M Street, SW, Washington, DC 20460
Mrs. Mary Winston, Staff Secretary, Drinking Water Committee, Science Advisory
Board (HOOF), U.S. EPA, 401 M Street, SW, Washington, DC 20460
-------
TABLE OF CONTENTS
1. EXECUTIVE SUMMARY 1
2. INTRODUCTION 3
2.1 Background 3
2.2 Charge to the Committee 3
3. HEALTH EFFECTS 4
4. PRACTICAL QUANTITATION LEVELS (PQL) ISSUES 5
4.1 Recommendation #1 5
4.2 Recommendation #2 7
5. BEST AVAILABLE TECHNOLOGY 9
5.1 Pre-oxidation 9
5.2 Ion Exchange 10
5.3 Membrane Treatment 11
5.4 Coagulation, Lime Softening, and Direct Additives 12
6. OCCURRENCE OF ARSENIC 14
6.1 Background 14
6.2 Use of censored data 15
7. REGULATORY IMPACT ASSESSMENT DECISION TREE 21
7.1 Does the SAB agree with the lower removal efficiency assumption for
coagulation/filtration based on full-scale data? 21
7.2 Does the SAB agree that EPA should re-evaluate this assumption if the
American Water Works Association (AWWA) data on existing plants
examines arsenic removal at optimum conditions? 21
7.3 Does the SAB agree with this revision to the design assumptions for
membrane systems? (using 75% recovery) 21
7.4 Does the SAB agree with the elements to be costed for nanofiltration
(including the cost of up-front cartridge filters) when high iron and/or
manganese levels are present with arsenic? 22
7.5 Does the SAB agree that sludges should be classified as non-hazardous
with the current or a revised arsenic standard? 22
7.6 Should alternate source development be used as an alternative even for
low-level options where all potential sources may contain arsenic? 22
7.7 Should regionalization continue to be included at 5% even though it is
generally more expensive than the BATs? 23
References
IV
-------
1. EXECUTIVE SUMMARY
The Drinking Water Committee of the Science Advisory Board (SAB) met on
August 18-19, 1994 to review several issues of concern relative to the regulation for
arsenic in drinking water, namely analytical methods, best available technology, and
occurrence. The Committee report also highlights earlier SAB reports regarding the
health effects of arsenic.
With regard to the analytical issues, the Committee recommended that the PQL
be set using acceptance limits for approved analytical methods that are similar to those
used for other inorganic substances, rather than the ±40 percent that is now being
proposed. It was also the Committee's opinion that risk management issues should not
enter into the procedure for determining the allowable error for the PQL. Lastly, if the
PQL set using the acceptance limits described above does not meet EPA's needs, the
Agency should develop alternative analytical methods.
With the regard to questions of Best Available Technology (BAT), the Committee
felt that pre-oxidation of arsenic with ozone is expected to perform well, although some
issues remain to be investigated. They did not consider that oxidation would be
necessary for most surface waters, but recommended that periodic analysis of
speciation be required to demonstrate that oxidation is not required. The Committee
urged that careful consideration must be given to the formation of disinfection by-
products (DBPs) in any specification that pre-oxidation be part of BAT. They found that
if ion exchange is to be used, the most appropriate choice of systems will probably be
the single, deep bed, counter-current flow ion exchange process, but they questioned
whether ion exchange would be an appropriate process selection for small systems.
Also, they felt that corrosivity of ion exchange product water could be a problem, and
that developmental research would be needed to determine how best to control the
problem.
The Committee concluded that EPA should ensure that appropriate studies are
conducted to determine which processes are BAT. In the Committees' view, it should
be possible to conduct generic studies, as opposed to site specific studies, on reverse
osmosis (RO) and nanofiltration (NF) as BAT. They found that water rejection and
brine disposal would be important issues in most regions, water-scarce or not. It was
not apparent to the Committee that greensand filters will be effective for arsenic
oxidation, and they recommended that performance data be obtained to show the
effectiveness of this approach before it is included as part of BAT. Also, the Committee
-------
felt that although the problems with nanofiltration technology should be solvable, the
technology needs to be further developed if this is to be considered the BAT for
arsenic. In addition, the Committee questioned the suitability of older data sources and
recommended the use of certain statistical methods that are described to derive
estimates from existing data with multiple detection limits.
-------
2. INTRODUCTION
2.1 Background
The Office of Water is developing the technical foundation for the regulation of
arsenic in drinking water. On April 18-19, 1994, the Drinking Water Committee (DWC)
of the Science Advisory Board (SAB) held a public meeting to review several issues of
concern relative to the development of this regulation. The Committee had previously
offered advice concerning the Agency's activities regarding arsenic in drinking water.
Although the health effects of arsenic were not the subject of this review, section three
of this report summarizes the highlights of the Committee's earlier views regarding this
issue (SAB, 1992). The subsequent sections contain the Committee's responses to the
specific questions in the charge in each of the four areas listed below.
2.2 Charge to the Committee
The Committee reviewed Agency activities regarding:
a) the methods being considered by the Agency to estimate national
occurrence levels of arsenic in drinking water;
b) the Agency's approach to determining Best Available Technology (BAT)
for arsenic removal from drinking water;
c) the basis for selection of quantitation detection limits for arsenic; and
d) the Agency's approach to the development of a "decision tree" for a
regulatory impact assessment for arsenic in drinking water.
-------
3. HEALTH EFFECTS
The health hazards associated with environmental and occupational exposure to
arsenic have been the subject of several reviews by the Science Advisory Board (SAB,
1989, 1992, 1994). In these reviews, little question has been raised over the issue of
whether arsenic is a human carcinogen. However, the DWC has repeatedly raised
concerns on the adequacy of available data for purposes of estimating the magnitude
of risks associated with environmental exposures to arsenic. It is not our intent to
reiterate our technical concerns in detail here, but would like to call your attention to
one review (SAB, 1992) in which considerable effort was exerted by the DWC to
identify a research strategy that would help resolve these issues. The Committee
understands that some of this work has been initiated by the Health Effects Research
Laboratory, but most of the more substantive issues identified in the Committee's
earlier report remain unaddressed.
The DWC believes that a research program directed at resolving the quantitative
risk assessment issues with arsenic can be accomplished. The Committee is
concerned that the Agency, in responding to short-term deadlines, will never develop
the information required to make sound decisions based on science.
The relative source contribution should be clearly described in the risk
characterization developed for arsenic in drinking water. The Maximum Contaminant
Level (MCL) for arsenic should be set at a level that reflects understanding of the levels
of arsenic contributed by other sources.
-------
4. PRACTICAL QUANTITATION LEVELS (PQL) ISSUES
4.1 Recommendation #1
The Agency is proposing a PQL of 2 |jg/L for arsenic, with acceptance
limits of ±40 percent. The Committee recommends that the PQL be set
using acceptance limits for approved analytical methods that are similar to
those used for other inorganic substances, rather than the ±40 percent that
is now being proposed.
The PQL is defined by the Agency as the "lowest level of an analyte that can be
reliably measured within specified limits of precision and accuracy during routine
laboratory operations" [50 FR 46906 (November 13, 1985)]. EPA further stated that the
PQL is "analogous to the Limit of Quantitation (LOQ) as defined by the American
Chemical Society," pointing out that the PQL is an inter-laboratory concept while the
LOQ is specific to an individual laboratory [52 FR 25699 (July 8, 1987)]. The
Committee agrees that a Maximum Contaminant Limit (MCL) must be set at a level that
is practical for quantitation with the specified analytical method. If the MCL is not well
based, the public and regulatory agencies will receive erroneous information about the
safety of their drinking water.
The arsenic analytical methods that EPA proposes are not new and have been
in wide use for some time. Except for the hydride method, the methods are also
relatively straightforward using direct injection without derivatization or extraction steps.
These conditions would suggest that a relatively low error from the true value would be
chosen. For example, the analytical methods for selenium are identical to those
proposed for the analysis of arsenic. The acceptance limits EPA used in establishing
the selenium PQL is ±20%. In fact, the analytical procedures for all the other regulated
metals are similar, and the PQL the EPA has accepted for these elements range from
±20% to ±30%. There does not appear to be any reason arsenic should be treated
differently.
The Committee also notes that a plot of the percentages of laboratories passing
versus the concentration of arsenic reveals that the arsenic method begins to
deteriorate rapidly below a level of 4 to 6 ug/L. However, the Agency presented
information that an independent PQL study supported the choice of a 4 ug/L PQL.
-------
The Committee has observed that the Agency on occasion has introduced risk
management policies into the determination of the PQL. For example, in setting the
PQL for vinyl chloride, EPA stated,
"Because of latter factor (vinyl chloride's higher potency than the other
VOCs), EPA believes it is appropriate to accept slightly less precise data
in order to seek to obtain more stringent levels of control [52 FR 25699
(July 12, 1987)]".
This principle has been employed several times since then. In the discussion on
dioxin, EPA stated,
"For 2,3,7,8-TCDD, EPA proposes to set the PQL at five times the Method
Detection Limit (MDL). Lower contaminant levels are associated with
greater difficulty in measurement and consequently less precision and
accuracy. However, EPA believes it is appropriate to accept less
precision if the risk posed by a carcinogenic contaminant at a level of ten
times the MDL is greater than the 1xlO"4 maximum individual lifetime risk
that is generally considered by the Agency to be acceptable [52 FR
30416 (July 25, 1990)]."
Because the concentration of 2 ug/L corresponds to the Agency's current
estimate of the 10"4 risk level, it appears that the Agency is again using risk
considerations to establish the PQL. It is the Committee's opinion that the PQL would
be a stronger scientific concept if risk management issues were not allowed to enter
into the procedure for determining the allowable error for the PQL. The PQL for a
particular analytical method should be determined based on the procedure stated in
EPA's definition, namely the "lowest level that can be reliably achieved within specified
limits of precision and accuracy during routine laboratory operations."
It is the Committee's opinion that the EPA's statement that it will, "accept slightly
less precise data to seek to obtain more stringent levels of control", does not accurately
portray the choice being made. Lowering the hurdle for acceptable precision for an
analytical method does not make it more sensitive nor does it enable the Agency to
provide more stringent control. The overall accuracy and precision would not be
expected to improve as a result of a lowered PQL level. Method performance will be
the same as currently exists. The primary result of a lower PQL will be information of
poorer quality.
-------
The Committee agrees that different errors should be acceptable under some
circumstances, but the Committee recommends that the Agency clarify its policy as to
when an error of a given magnitude is acceptable. Moreover, the Committee
recommends that EPA not consider the proximity to the target risk range in determining
the error allowable in the PQL determination. Rather, EPA should either find or
develop more sensitive methods (see discussion on analytical methods below) when
the PQL which is determined does not provide satisfactory proximity.
The following principles are offered as guidance for EPA in deliberating on its
analytical policy: a) greater error is acceptable when a method is new, when a method
is complex, or when a method has been rarely used; and b) similar errors should be
allowed when the same procedure is used to determine a different analyte unless
special interferences occur. In the case of a new method, the error can be expected to
diminish as it comes into more common use, so greater errors are acceptable in early
Performance Evaluation (PE) data. Greater errors are also acceptable when a method
involves a number of steps, like extraction or derivatization, as these additional steps
inherently increase the variability of the result. Like new methods, methods that are
rarely used are also likely to produce greater variability in a PE survey but will most
likely produce diminished errors after they come into more common use.
4.2 Recommendation #2
If an error of ±20 or ±30 percent results in a PQL that EPA finds
unacceptable, alternative analytical methods should be considered and
developed as approved methods, and the data developed to establish
PQLs for those methods.
Methods are presently available that can provide reliable data at levels down to
and below 0.1 ppb. Many of these methods have been in use in the field of
environmental analytical chemistry for many years, but for a number of reasons, they
have not been used in most regulatory work. Many analysts have generated reliable
data for ambient levels of arsenic in natural waters over the past 20 years. Because
the range of interest in these studies has been in the range of 0.1 to a few ppb, most
have utilized one of the several variations of hydride generation atomic absorption
spectrometry (HGAAS), which provides capability to measure at these low levels.
The analytical community will have to modify existing methods or adopt other,
proven methods in order to provide accurate and precise data at the lower
concentration levels that are now of greater interest. HGAAS already has the potential
-------
to achieve even lower levels, but may require more recent method developments than
are specified in several of the HGAAS methods referenced by EPA (e.g., introduction of
the hydride into a heated quartz cell rather than into an air-hydrogen flame). Graphite
furnace atomic absorption spectrometry (GFAAS) offers the potential for improved data
quality through repeated injection and drying, allowing introduction of larger sample
volumes into the graphite tube prior to measurement. In practice, however, this may
not work since matrix interferences will also increase. Samples with high dissolved
solids may produce unacceptably high background signals when multiple injections are
used. Unfortunately, matrix interferences are most severe at low arsenic
concentrations. Minor modifications to inductively coupled plasma-mass spectroscopy
(ICP-MS) methods are not detailed, but since this analytical tool is not in widespread
use, its ability to support a 2 ppb PQL is limited. A general summary of these methods
is presented in Table 1. While the Committee has no direct experience with detection
levels resulting from a combination of hydride generation and ICP-MS, it seems
reasonable that this method will also provide detection limits of at least 0.05 ppb.
The methods in Table 1 satisfy the requirements for feasibility, in that they are
proven, sensitive, and within the ability of any competent analytical laboratory that
perform analyses at trace levels. Further, several can be used with existing
instrumentation with only the addition of a sample concentration step or a reliable
instrumental module to introduce the sample to the detector.
Table 1. Approximate Detection Limits of
Available Analytical Methods for Measurement of
As in Drinking Water Matrices
Method Approximate MDL
(ppb As)
Column preconcentration/GFAAS 0.05
(graphite furnace atomic absorption
spectrometry)
Manual HGAAS (hydride generation 0.10
atomic absorption spectrometry)
Continuous or flow-injection HGAAS 0.05
(hydride generation atomic absorption
spectrometry)
HGAFS (hydride generation atomic 0.05
fluorescence spectroscopy)
8
-------
5. BEST AVAILABLE TECHNOLOGY
This section contains responses to the questions posed by the Agency regarding
Best Available Technology for arsenic removal. Also included are issues related to the
topic that the Committee believes to be relevant.
5.1 Pre-oxidation
a) If EPA receives performance data on ozone (03), are there other
limitations on the use of ozone as a pre-oxidant for arsenic?
Limitations are probably fewer on the use of ozone as a pre-oxidant for arsenic
than when 03 is used after filtration as a primary disinfectant. Assimilable Organic
Carbon (AOC) formation is ameliorated by coagulation/filtration treatment downstream,
and oxidation of bromide, though still possible, is much less likely during pre-oxidation
because the development of a significant ozone residual is not necessarily required.
Ozone pre-oxidation before nanofiltration could present a problem if the AOC that is
formed has a low molecular weight and passes through the membrane. This issue
should be studied as part of the membrane research effort recommended in Section
5.3.
b) Are there any generic conditions, other than speciation, that could be
used to determine when pre-oxidation would not need to be a component
of BAT?
Oxidation is probably not necessary for most surface waters; however, periodic
analysis of speciation should be required to demonstrate that oxidation is not required.
We do not know, for example, whether As(lll) is found in lakes and reservoirs that
periodically go anoxic.
c) Is there an incompatibility of the requirement for pre-oxidation to convert
As(lll) to As(V) with the requirement to reduce disinfectant byproducts
(DBPs)?
Chlorine is a good oxidant for As(lll), but application must come early in the
treatment train when disinfectant byproduct precursor concentration is high and there is
a danger of producing large concentrations of DBPs. Permanganate may work better,
but we do not have sufficient information on the permanganate demand for arsenic
-------
oxidation relative to the demand exerted by other substances. Ozone may be the most
satisfactory for such waters, but performance data are limited. This is an issue for
surface waters that go anoxic, because As(V) may be converted to As(lll) under these
conditions, and for ground waters with high concentrations of organic matter. Careful
consideration must be given to the formation of DBFs in any specification that pre-
oxidation be part of BAT.
5.2 Ion Exchange
a) Does SAB agree with the "merry-go-round" approach (i.e. series
operation of columns) for ion exchange in order to meet low-level MCL
options?
Series operation of columns can achieve the low levels required, but so can
several other designs/operating methods. For example, operating columns using the
"counter-current flow" principle of operation will increase the usable capacity of the ion
exchange bed and also reduce the "leakage" (effluent level) of arsenic. In most
common ion exchange designs, both the water to be treated and the regenerant go
downflow through the resin bed. This "cocurrent" process can result in high "leakage"
and high regenerant requirements because the regeneration process concentrates the
contaminants on the effluent end of the ion exchange column. Compared to such
operation the "merry-go-round" approach will result in more efficient use of the resin
and achieve lower effluent concentrations. However, over the past twenty years or so,
single-bed, counter-current processes have shown themselves to be more cost
effective options. Instead of the "merry-go-round" approach, EPA will probably find a
single, deep bed, counter-current flow ion exchange process the most appropriate
choice when particularly high removals of arsenic are required.
b) Does the SAB agree with the recommendation of ion exchange alone as a
BAT primarily for small ground water systems?
The ion exchange process has the advantage of being easy to automate.
However, we understand that the anion resins being considered for this process have a
preference for sulfate over arsenate. As a result, if the process is allowed to run
unattended, there is a potential that all the arsenic will be displaced from the column in
a very short time period if the process is run to exhaustion. This problem could,
perhaps, be overcome with proper operation. Further, the ion exchange resin would
probably be used in the chloride form, and thus the process would result in large
increases in chloride concentration in many situations. An increase in chlorides can
greatly increase the corrosivity of the product water and red water problems, thereby
10
-------
introducing undesired secondary effects of this process. Membrane processes may be
a more appropriate choice.
c) Is there a problem with using ion exchange, and coagulation/filtration
followed by ion exchange, for large systems?
A primary issue here is the "corrosivity" of the product water, just as it is for small
systems. Chlorides increase the corrosivity of iron and will lead to more problems with
red water, especially when the high chloride levels are intermittently present.
Modifications of the ion exchange process could be developed that would eliminate or
minimize this problem, but developmental research is needed.
5.3 Membrane Treatment
a) Should pilot-scale studies be included as a component of the treatment
train when EPA prepares costs for options where nanofiltration (NF) is
selected to remove arsenic?
EPA should see that appropriate studies are conducted to determine which
processes are BAT. It should be possible to do generic studies, as opposed to site
specific studies, on reverse osmosis (RO) and NF that show whether they are BAT. NF
is just a variation of RO wherein a looser, alternative membrane is used along with
lower operating pressures. If sufficient information is not available to demonstrate that
NF is an appropriate BAT, EPA should not consider NF and should nominate RO as
BAT. Utilities who are willing to invest in pilot studies can still install NF once they are
convinced it will give satisfactory performance.
b) Is water rejection an issue in water-scarce regions?
Yes. Disposal of reject water is an important issue in most regions, water-scarce
or not. Brine or reject water disposal is one of the principle obstacles to broader
application of ion exchange, reverse osmosis, nanofiltration and electrodialysis and the
constraints are getting tighter, not looser. The EPA should give serious consideration
to this problem.
c) Pre-oxidation before membrane processes.
This is an issue that requires careful research. The EPA assumes that As(lll)
can be oxidized to the more easily removed As(V) using a greensand filter. The
greensand filter is used effectively with permanganate to remove Fe(ll) and Mn(ll), and
11
-------
to accomplish this, permanganate is either applied continuously to the influent to the
greensand or intermittently, during backwash. When applied intermittently, the
mechanism is one of forming a layer of Mn02 on the filter media that adsorbs
multivalent cations. The adsorbed ions are then held until permanganate is applied
during backwash. The adsorbed material is eliminated during backwash, or oxidized to
insoluble forms that remain in layers on the greensand. When permanganate is
applied continuously, the same mechanism applies, except that oxides of Mn and Fe
continuously build up on the greensand media until they are backwashed off. Given
this mechanism of operation, it is not apparent that greensand filters will be effective for
arsenic oxidation. Is co-occurrence of Fe(ll) essential for performance? Does Mn02
oxidize As(lll)? The Committee recommends that performance data be obtained to
show the effectiveness of this approach before it is included as part of BAT.
d) Nanofiltration technology applied to surface waters.
The water industry has great difficulty applying membrane processes to surface
water because of membrane fouling and flux decline. The problems should be
solvable, but the technology needs to be further developed if this is to be considered
BAT for arsenic.
5.4 Coagulation, Lime Softening, and Direct Additives
a) Is the analysis of removal data in the coagulation and lime softening
processes appropriate?
A heavy emphasis has been placed on "percent removal." The mechanism of
removal is probably precipitate formation or formation of a chemical bond that is
concentration dependent. It makes more sense from a chemical mechanism point-of-
view to analyze the data in terms of remaining concentration. For example, the removal
in lime softening may be partially dependent on adsorption onto Mg(OH)2 and may not
be removed if this precipitate is absent. Adsorption is also dependent on solution
concentration, and process effluent concentration would likely be a function of solution
chemistry; percent removal in the coagulation process would then be a function of
surface chemistry.
b) Does the SAB agree that the impact of arsenic in direct additives on water
systems should be minimal unless a very low standard (<1 ug/L) is selected, and
does the SAB agree that the impact of arsenic in indirect additives on water
systems should be minimal?
12
-------
The Committee does not have access to recent data on arsenic levels in direct
additives that would be necessary to make this evaluation. The analysis conducted
several years ago by the National Academy of Sciences (MAS) Committee on direct
and indirect additives indicated that arsenic was generally not a problem in water
treatment chemicals with an MCL of 50 ug/L, and third party certification processes are
established using the guidelines of that committee's effort. A significant effort would be
required to review the matter in light of a revised MCL.
13
-------
6. OCCURRENCE OF ARSENIC
6.1 Background
EPA is considering a change in the MCL for arsenic and would like to use
existing occurrence data on arsenic in drinking water in order to make judgments about
the impact of alternative regulatory targets. The Agency is considering a range for the
new MCL of 2 to 20 ug/L, but the most geographically comprehensive occurrence data
are found in older surveys, particularly the National Inorganics and Radionuclides
Survey (NIRS), and these surveys have a Reporting Limit (RL) of about 5 ug/L. As a
result, the data are censored in the sense that quantitative measurements are reported
for some samples (not censored), but the remainder are only reported as less than the
RL(censored).
Analytical techniques have been improving over time and additional data are
available from the states and from other special purpose surveys. Many of these
surveys have RLs of 0.5 to 1 ug/L, but none is sufficiently comprehensive to serve as a
basis for regulation by themselves. Additional surveys of a more comprehensive nature
may soon become available. EPA would like to use all these data to produce a best
estimate of the national occurrence of arsenic. The problem is not just using censored
data, but using censored data from a variety of sources and a variety of geography's
with multiple reporting limits. EPA suggests that an error in estimating the number of
systems affected by a particular MCL within a factor of two would be acceptable. The
Agency's current estimates suggest that the Nation's capital cost would be $140
millions and $6,200 millions to comply with MCLs of 20 and 5 ug/L, respectively.
Using the older survey data is attractive because of the number of systems that
were tested (some 47,000 in the NIRS, for example). Arsenic is a naturally occurring
element that is probably present in all the samples analyzed. On the other hand,
because arsenic is generally present at low levels and these surveys have relatively
high RLs, the survey results consist almost entirely of censored data (e.g., NIRS 93%,
CWSS 97%, and RWS 90%). The Committee recommends EPA abandon attempts to
use these older data when sufficient new data become available.
The problem EPA faces in using censored data is not a new one and there exists
a substantial literature on the subject. The following discussion will address only the
most important elements of the subject.
14
-------
6.2 Use of censored data
6.2.1 Probability density function (PDF)
If the data were not censored, they could be used to determine the parameters
of a Probability Density Function (PDF) and then that PDF could be used to predict the
number of water systems that would be impacted at alternative MCL values. Data on
environmental chemicals are often modeled with the Log-normal PDF because it
eliminates the risk of estimating negative concentrations and because it predicts mostly
low concentrations except for a long tail to the right, simulating contaminated situations.
The Weibull PDF has these same properties with the additional advantage that it can
be integrated analytically to produce a cumulative distribution function that enables
convenient estimates of the occurrence of specific concentrations (Mackay & Paterson,
1984).
6.2.2 Non-parametric methods
Alternatively, if there were enough data, they could be put in ranked order and
the percentage of the systems exceeding a specified regulatory target could be
estimated without recourse to more complex statistical techniques. The best scientific
approach is to conduct a new survey with reporting limits below the regulatory range
targeted by the risk manager. Using the results of this survey the Agency could
develop a simple ranking of the data to characterize occurrence at levels being
considered for regulation. This approach would avoid the complication of choosing a
PDF altogether.
On the other hand, if the EPA insists on using existing data, the Committee
recommends that the Agency also consider the use of nonparametric methods
(methods that do not require the assumption of a specific PDF model) to estimate the
distribution for concentrations that are in the range of the data. The Kaplan-Meier or
product-limit estimator (Cox and Oakes, 1984), suitably modified for left-censored data,
is the nonparametric maximum likelihood estimator of the distribution function in the
case of censored data. Cox and Oakes (1984) provide methods for computing limits
based on product-limit estimates.
6.2.3 Goodness-of-fit
It appears that the Agency has used comparisons of the estimated averages and
standard deviations as well as probability plots of the data and the Kolmogorov-
15
-------
Smirnov (KS) tests to evaluate the suitability of both the PDF's and the methods of
estimation. The application of the KS test described to the Committee at the August
18th meeting does not appear to be appropriate.
Qualitatively, the "fit" of censored data to any particular distribution (Normal,
Lognormal, Weibull, etc.) is best accomplished by observing a plot of the data on axes
of an appropriate scale, on a probability plot. The Committee is not aware of a general
goodness-of-fit test that is valid for censored data. The Chi-Square goodness of fit test
could be applied if the data were grouped so that all of the non-detects fall in the lowest
interval. However, this restriction may not permit a reasonable grouping if there are
multiple detection limits.
The Kolmogorov-Smirnov test can only be used to select objectively the choice
between PDFs if all the parameters of the PDFs have been determined before-hand
(i.e., not from the data). For example, the KS statistic (and the corresponding plot)
could be used to determine if the crushing strengths of 35 concrete test cylinders meet
a specification: and that they have an average crushing strength of 4,000 psi with a
standard deviation of no more than 400 psi to a 95% level of significance. The
corresponding KS statistic would be 0.21. Using this statistic, a plot could be
constructed on normal probability paper showing the boundaries within which the data
must lie to meet the specification.
6.2.4 Simulation studies
Perhaps a better way to evaluate the suitability of alternative PDFs (and
alternative methods of estimating PDF parameters from a censored distribution) is to
evaluate their effectiveness in producing accurate estimates of low level concentrations
of arsenic using a simulation.
If the EPA wishes to conduct a simulation evaluation of various approaches to
estimating the parameters for the model selected, we recommend that EPA simulate
from a completely determined distribution so that estimated quantities can be compared
to known values. The following specific approach is suggested with the lognormal
distribution in mind.
Using data from a suitable study, estimate the parameters u and s of the log-
normal distribution by fitting a log-normal distribution to the complete data set using the
maximum likelihood estimation (MLE) or regression on ordered statistics (ROS)
approach (but not the Modified Delta-Lognormal, or MDLOG). Using these parameters
and a random number generator, generate by computer simulation the same number of
16
-------
samples as are in the underlying data set. For example, suppose in the database, a)
there are two distinct values for non-detects, 5 ug/L and 10 ug/L, b) there are 1000
samples with concentrations less than or equal to 10 (including detects and non-
detects), 50 of which are non-detects with detection limits of 10 ug/L and 900 of which
are non-detects with a detection limit of 5 ug/L, and c) no detects have concentrations
less than 5 ug/L. First, in turn, take each simulated concentration that is <10 ug/L and,
with probability 50/1000, convert it to a non-detect with detection limit of 10 ug/L. Then
convert each of the remaining simulated samples with concentrations less than 5 ug/L
to non-detects as well. This results in a simulated data set with a pattern of non-
detects that is similar to the original data set.
Generate a large number (e.g., 1000) of such simulated data sets and apply
each of the estimation methods to be compared to each. With each such simulation
and each estimation method evaluated, record the probability that a sample
concentration exceeds 2, 5, 10, 15, 20, 30, 40, and 50 ug/L. For each of the estimation
methods, prepare graphs showing the resulting distributions of probabilities as well as
true probabilities. Also estimate the mean square error and the bias of each estimator.
This simulation should provide a good indication of the relative performances of
the estimators whenever the true distribution is log-normal. If EPA wishes to evaluate
the properties of these estimators when the log-normal assumptions not valid, the
simulation could be repeated, with the only change being that the simulated data are
generated using the new PDF model, presumably one that appears to better describe
the data.
6.2.5 Fitting Censored Data to PDF
During the past decade a substantial literature had developed on methods for
estimating the parameters of a PDF from censored environmental data. The
approaches used break down into three groups that are much like those being
considered by the Agency for arsenic: a) simple substitution; b) regression on order
statistics (ROS); and c) maximum likelihood estimation (MLE).
Suitability of MDLOG. The Modified Delta Log normal method (MDLOG) is but
one example of the simple substitution methods. It produces a distorted distribution
that is not representative of the original censored population and should not be used,
but because the MDLOG estimation method is widely used in the treatment of
environmental data, it deserves further discussion. This method is fundamentally
different from the ROS and MLE methods. Both of the latter two methods assume that
the true concentrations in all samples come from a common population that is log-
17
-------
normally distributed. In particular, non-detects are assumed to come from the same
distribution, albeit the low end of that distribution. The MDLOG method, however
assumes that a) the detects and the non-detects come from different sample
populations; and b) that the detects themselves constitute an uncensored sample from
a log-normal distribution. As discussed later, this latter assumption only applies if it
can reasonably be presumed that the contaminant is present in some samples and
absent in others - an assumption that certainly cannot be applied in the case of a
naturally-occurring element like arsenic.
Whether a sample is a detect or a non-detect is determined by the RL of the
analytical method. In the case of Arsenic, there would be no reason to assume that the
concentrations above and below the RL come from distinct populations. (An illogical
conclusion of such an assumption is that analyzing samples with methods having
different RLs would lead to different underlying concentrations.) Moreover, the detects
cannot represent an uncensored sample from a log-normal distribution because,
whereas a log-normal distribution includes all positive values, a non-censored sample
cannot assume any value less than the smallest RL of any of the samples.
A priori, it is expected that maximum likelihood estimation approach will perform
the best of the these procedures. A number of studies have been published comparing
these methods and showing alternative algorithms for solving the MLE problem
(Hashimoto & Trussell, 1983; Gleit, 1985; Guilliom & Helsel, 1986; El-Shaarawi 1989;
Haas and Scheff, 1990; Helsel, 1990; and El-Shaarawi & Esterby, 1992). These studies
have consistently found the ROS and MLE procedures superior to simple substitution
methods. The practical differences between the MDLOG and the ROS and MLE
estimation methods can be seen from Table 2 displaying the results of applying all
three methods to the NIRS data on very small systems (cf. Exhibits 3-5, 4-1, and 4-5 in
the "occurrence" document given to the Committee). Estimates from the ROS and MLE
methods agree closely and predict a smooth log-normal distribution throughout the
range of 2 to 50 ug/L. Results from the MDLOG methods agree reasonably well with
results from the other two methods in the range where most of the non-censored
samples were located (5 to 20 ug/L). However, it gives very different results in the tails.
The MDLOG procedure predicts that virtually 100% of the systems have arsenic
concentrations greater than 2 ug/L, because of the ad hoc assumption that all non-
detects have actual concentrations of one half the RL (which was 5 ug/L in the NIRS
database). Thus, since the predictions of the MDLOG procedure at concentrations
below the RL are determined solely by this assumption, the MDLOG procedure should
not be used to predict the concentrations below the RL. The MDLOG procedure also
predicts only one-tenth as many systems having concentrations in excess of 50 ug/L as
the other two methods. Although one cannot say with certainty that the MDLOG
18
-------
prediction is in error (since 50 ug/L is beyond the range of the data), the assumption
that led to the smaller tail of the MDLOG distribution is erroneous (that assumption
being that the parameters determined from non-censored values represent entire
population and that all samples below the RL are the same which leads to a log-normal
distribution with a smaller variance than that estimated under a more appropriate
assumption that the non-censored samples represent a censored sample of the
distribution).
Table 2. Comparison of Estimates of Cumulative National Occurrence of Arsenic
in Community Ground Water Supplies (Exhibit 4-5) Using ROS, MLE and MDLOG
Models.
System Size
(Population
Served)
No. of
Systems in
U.S.
Number of Systems with Concentrations (ug/L) in Excess of:
0.5
1
9
3
4
5
10
15
20
30
40
50
ROS Estimates (Exhibit 4-5)
Very Small
25-100
101-500
16634
15422
8467
7850
5634
5223
3266
3028
2216
2054
1629
1510
1259
1167
508
471
276
256
173
160
85
79
49
46
32
30
MLE Estimates
Very Small
25-100
101-500
16634
15422
8223
7624
5449
5052
3155
2926
2143
1987
1578
1463
1222
1133
497
460
272
252
171
159
85
79
50
46
32
30
MDLOG Estimates
Very Small
25-100
101-500
16634
15422
16634
15422
16634
15422
16632
15420
1198
1110
1159
1075
1097
1017
644
597
318
295
153
142
38
36
11
10
4
3
ROS and MLE methods have also been shown to produce almost the same values for
PDF parameters. The ROS procedure is more stable when non-censored data are
limited, but MLE methods generally produce a more precise estimate of the mean. The
Committee recommends that the Agency limit its methods of estimation of PDF
parameters to variations of the ROS and MLE procedures.
19
-------
6.2.6 Confidence Intervals
Standard confidence interval methods are available for the product-limit
estimator (Cox and Oakes, 1984). We recommend that the profile likelihood method
(Cox and Oakes (1984), page 35, procedure (a)) be used to compute confidence limits
for the parameters estimated by the MLE method. The ROS method does not easily
lend itself to confidence limit estimates.
20
-------
7. REGULATORY IMPACT ASSESSMENT DECISION TREE
The Committee believes that the concept of a decision tree to guide the process
of determining the best approach to provide drinking water with safe levels of arsenic is
reasonably good. In addition to guiding the selection of the best available treatment
technology, the decision tree approach will have the added benefit of revealing missing
information about processes needed to make such decisions. As such, it will be helpful
in identifying priorities for further research in order to fill gaps in our knowledge
7.1 Does the SAB agree with the lower removal efficiency assumption for
coagulation/ filtration based on full-scale data?
Yes, given the complexity of full-scale plant operation compared to carefully
designed and run bench and pilot studies, the lower removal efficiency is not
surprising.
7.2 Does the SAB agree that EPA should re-evaluate this assumption if the
American Water Works Association (AWWA) data on existing plants
examines arsenic removal at optimum conditions?
Yes.
7.3 Does the SAB agree with this revision to the design assumptions for
membrane systems? (using 75% recovery)
This issue should be investigated further. Although most systems may be
designed at 75% recovery regardless of the capacity, required maintenance and
downtime are reduced if higher recoveries are allowed. With some designs, recoveries
can be adjusted in the field over a broad range. Considering operating requirements,
lower recoveries would seem appropriate for smaller systems, particularly those
operating on harder water. The question is, are these field adjustments made, or are
the systems just installed and operated as the designer envisioned?
21
-------
7.4 Does the SAB agree with the elements to be costed for nanofiltration
(including the cost of up-front cartridge filters) when high iron and/or
manganese levels are present with arsenic?
Levels of iron (Fe) and manganese (Mn) even below the secondary standards
will interfere with both reverse osmosis and nanofiltration, but recognizing the nature of
the data available, perhaps it is acceptable to presume special treatment on systems
not meeting those standards alone. Such pretreatment should include oxidation,
granular media filtration, and cartridge filtration. Oxidation alternatives deserve further
examination. Where possible, chlorine is most straightforward. For low levels of Fe
and Mn, and where greensand is used, regeneration of the greensand with KMn04 may
be used (see above comments on the greensand process). Chlorine dioxide may be
appropriate in some cases as may ozonation. Continuous permanganate feed is
difficult to control and should only be considered for fairly large systems. For small
systems, greensand may be appropriate, but larger systems will find it more cost-
effective to employ conventional sand or anthracite media and acclimate it with a Mn02
coating.
7.5 Does the SAB agree that sludges should be classified as non-hazardous
with the current or a revised arsenic standard?
Data presented are encouraging, but an adequate explanation for the results is
not available. Assuming a water with 3 mg/L suspended solids, and alum dose of 50
mg/L, and an As level of 1 mg/L, the resulting sludge would be 6% (60,000 mg/kg of dry
solids).
7.6 Should alternate source development be used as an alternative even for
low-level options where all potential sources may contain arsenic?
Utilities accustomed to using groundwater are generally not equipped or trained
to operate treatment facilities, except for the few that already treat water high in
hardness or iron and manganese. As a result they will tend to look at alternate sources
with greater interest. Rarely is an entire region so uniformly high in contaminants such
as arsenic that alternative well fields cannot be found.
22
-------
7.7 Should regionalization continue to be included at 5% even though it is
generally more expensive than the BATs? The focus on system viability in
the SDWA re authorization would suggest that costs continue to be
developed for this option in the RIA.
It is not clear to the SAB that regionalization is consistently higher in cost. The
cost feasibility of regionalization would appear to be related to the proximity of a large
regional system. Perhaps systems in urban areas should have a higher percentage
than those in rural environments.
The Committee also notes that there are additional aspects that should be
incorporated into this decision tree process.
First, the decision tree should be expanded to incorporate a more holistic
approach. Essentially, the goal should be to select appropriate treatment processes
which not only minimize exposure to arsenic but to other toxic agents, both organic
(e.g., trihalomethanes) and metal species as well; other compliance requirements
should be achieved (e.g., disinfection by-product rule, groundwater rule, etc.). The
selection of a treatment process should use an "optimization approach" that will yield
the desired safe levels for all toxic chemicals and microbial agents. From a practical
point of view, a municipality attempting to achieve compliance to several rules may only
have resources to address the issue once; therefore, it needs to "take its best shot."
Furthermore, an estimated economic impact analysis for implementing an
optimal consolidated treatment train may reflect only a modest increment in cost if
distributed across several pollutants, whereas, the economic impact of implementing a
treatment process may be, in fact, highly inflated when determined on a chemical-by-
chemical basis.
Second, a factor that is often considered in selecting a treatment process is the
potential waste products produced. The decision tree should include their production
and disposal issues.
23
-------
References
Cox, D.R. and D. Oakes. 1984. Analysis of Survival Data. Chapman & Hall: New York,
NY.
El-Shaarawi, A. 1989. Inferences about the mean from censored water quality data,
Water Resour. Res., Vol. 25, pp. 685-690.
El-Shaarawi, A. and S. Esterby. 1992. Replacement of censored observations by a
constant: An Evaluation, Water Resour. Res., Vol. 26, pp. 835-844.
Gilliom, R. and D. Helsel. 1986. Estimation of distributional parameters for censored
trace level water quality data, Water Resour. Res., Vol. 22, pp. 135-146.
Gleit, A. 1985. Estimation for small normal data sets with detection limits, ES&T, Vol. 9,
pp. 1201-1206.
Haas, C. and P. Scheff. 1990. Estimation of averages in truncated samples, ES&T,
Vol.24, pp. 912-919.
Hashimoto, L. and R. Trussell. 1983. Evaluating water quality data near the detection
limit, Proceedings of the 1983 Annual Conference of the American Water Works
Association, in Las Vegas, Nevada, p. 1021, AWWA, Denver.
Helsel, D.R. 1990. Less than obvious statistical treatment of data below the detection
limit, ES&T, Vol. 24, No. 12, pp. 1767-1774.
Mackay, D. and S. Paterson. 1984. Spatial concentration distributions, ES&T, Vol. 18,
No. 7, pp. 207a-214b.
SAB. 1989. SAB Review of arsenic issues relating to the Phase II proposed regulations
from the ODW, Science Advisory Board, U.S. Environmental Protection Agency,
Washington, DC. EPA-SAB-DWC-89-038.
SAB. 1992. Review of the ORD's arsenic research recommendations, Science Advisory
Board, U.S. Environmental Protection Agency, Washington, DC. EPA-SAB-
DWC-92-018.
SAB. 1994. Review of the draft drinking water criteria document on inorganic arsenic,
Science Advisory Board, U.S. Environmental Protection Agency, Washington,
DC. EPA-SAB-DWC-94-004.
-------
DISTRIBUTION LIST
Administrator
Deputy Administrator
Assistant Administrators
EPA Regional Administrators
EPA Laboratory Directors
Deputy Assistant Administrator for Water
Director, Office of Ground Water and Drinking Water (OW)
Director, Office of Science and Technology (OW)
EPA Headquarters Library
EPA Regional Libraries
EPA Laboratory Libraries
Library of Congress
National Technical Information Service
Office of Technology Assessment
Congressional Research Service
-------
United States Science Advisory EPA-SAB-DWC-95-015
Environmental Board (1400) July 1995
Protection Agency Washington, DC
&EPA AN SAB REPORT: REVIEW
OF ISSUES RELATED TO THE
REGULATION OF ARSENIC IN
DRINKING WATER
REVIEW BY THE DRINKING WATER
COMMITTEE (DWC) OF THE
SCIENCE ADVISORY BOARD
------- |