United States
Environmental Protection
Agency
Office of Water
4607
EPA 815-D-99-003
November 1999
METHODS, OCCURRENCE AND
MONITORING DOCUMENT FOR
RADON FROM DRINKING WATER
PUBLIC COMMENT DRAFT
-------
-------
Acknowledgments
This document was prepared for the Office of Ground Water and Drinking Water
(OGWDW) of the United States Environmental Protection Agency with substantial input from other
EPA offices. The authors of the document express their thanks to Peter Lassovszky, Ben Smith and
Andrew Schulmann of OGWDW for the advice and comments, Timothy Barry of the Office of
Policy, Planning and Evaluation, and to Professor H. Christopher Frey of North Carolina State
University, Department of Civil Engineering, for peer review comments. William Labiosa of
OGWDW was the principal author of Chapter 2 of this document.
The authors also express their thanks to the Association of State Drinking Water
Administrators (ASDWA) for their Assistence in identifying and gathering data related to radon
occurrence, the Radon Technical Work Group of the American Water Works Association (AWWA)
for constructive comments on early drafts, and to all the individuals and organizations noted in this
document who provided data for the analysis.
Methods, Occurrence and Monitoring Document for Radon
-------
TABLE OF CONTENTS
1. INTRODUCTION M
1.1 Purpose ofThis Document 1~1
1.2 Statutory Requirements !-l
2. ANALYTICAL METHODS 2-1
2.1 Introduction 2-1;
2.1 Inventory of Methods 2-2:
2.1.1 Liquid Scintillation Counting 2-2
2.1.2 The Lucas Cell Technique 2-4
2.2 Major Analytical Methods 2-4
2.2.1 Liquid Scintillation Counting and Lucas Cell Methods 2-4
2.2.2 Standard Method 7500: Radon Liquid Scintillation Counting 2-6
2.3 Other Radon Measurement Techniques 2-8
2.3.1 Delay-Coincidence Liquid Scintillation Counting System 2-8
2.3.2 Activated Charcoal Passive Radon Collector 2-9
2.3.3 Degassing Lucas Cell 2-11
2.3.4 Electret lonization Chamber System 2-12
2.4 Performance Capabilities of the Methods 2-14
2.5 Skill Requirements 2-15
2.6 Practical Availability of Methods 2-16,
2.7 Anticipated Unit Costs 2-19
2.8 Practical Performance and Analytical Uncertainties 2-20
2.9 Degree To Which Each Method Meets EPA's Regulatory Needs 2-22
2.10 References • • 2-22
3. SOURCES OF RADON IN GROUNDWATER 3-1
3.1 Natural Sources of Radon Groundwater Contamination 3-1
3.1.1 Release and Transport Properties of Radon and Radium 3-2
3.1.2 Factors Affecting Distribution of Radon in Groundwater 3-2
3.1.3 Large-Scale Geographic Patterns of Radon Occurrence in Groundwater . 3-3
3.2 Anthropogenic Sources of Radon Contamination in Groundwater 3-4
3.3 Distribution System Sources 3-4
3.3.1 Radon Sources in Distribution Systems 3-4
3.3.2 Radon Sources in Households '. 3-4
3.4 Non-Water Supply Sources of Radon Exposures 3-5
3.5 References 3-5
Methods, Occurrence, and Monitoring Document for Radon
11
-------
4. FATE AND TRANSPORT 4-1
4.1 Physical and Chemical Properties of Radon and Progeny 4-1
4.2 Relationship of Fate and Transport Properties to Human Exposures and Intake
4-1
4.3. Exposures to Radon in Indoor Air After Release During Domestic Water Use
4-2
4.4 Relationship of Fate and Transport Properties to Radon Behavior in Treatment
and Distribution Systems 4-2
4.4.1 Aeration Technologies 4-3
4.4.2 Granular Activated Carbon Treatment 4-3
4.4.3 Radon Release from Pipe Scale 4-3
4.5 References 4-5
5. DISTRIBUTION OF RADON IN GROUND WATER SOURCES 5-1
5.1 Data Availability and Quality 5-1
5.1.1 Previous EPA Data Gathering Efforts Related to Radon Occurrence in
Groundwater Supplies , 5-1
5.1.2 Data Gathering Efforts in Support of the Revised Occurrence
Analysis 5-7
5.1.3 Results of the Data Gathering Effort 5-7
5.2 Methods Used in the Data Analysis 5-13
5.2.1 Statistical Analysis of Radon Distributions 5-13
5.2.2 Distribution Fitting and Goodness-of-Fit Testing 5-18
5.2.3 Hypothesis Testing for Differences in Radon Activity Levels
and Distributions 5-20
5.2.5 Computing Methods 5-22
5.3 Analysis of Radon Occurrence Data: Approach to Stratification 5-22
5.3.1 Stratification by System Size 5-23
5.3.2 Alternative Stratification Variables 5-23
5.4 Distribution of Radon Level in the NIRS Database 5-24
5.4.1 Distribution of Radon in Nationally Aggregated NIRS Data 5-25
5.4.2 Distributions of Radon in Regionally Stratified NIRS Data 5-28
5.4.3 Goodness of Fit Testing of Lognormal and Alternative Distributions of
MRS Data 5-30
5.5 The Distribution of Radon in the Supplementary Data Sets 5-32
5.5.1 Distributions of Radon in Supplemental Data Sets 5-35
5.5.2 Radon Summary Statistics from Supplementary Data Sets 5-37
5.6 Comparison of NIRS and Supplemental Data Sets 5-42
5.6.1 Comparison of Log Mean Radon Levels Between NIRS and Supplemental
Data Sets 5-43
5.6.2 Comparison of Log Standard Deviations 5-45
5.7 Sources and Magnitude of Variability in Groundwater Radon Levels 5-46
5.7.1 Identification of Sources of Variability in Radon Levels 5-47
Methods, Occurrence, and Monitoring Document for Radon
111
-------
5.7.2 Estimating Contributions to Variability 5-47
5.7.3 Magnitude of Contributions to Radon Variability 5-48
5.8 Estimates of Numbers of Groundwater Systems Exceeding Potential Regulatory
Limits 5-49
5.8.1 Characterizing Radon Distributions for States and Regions 5-49
5.8.3 Numbers of Community Water Systems in the U.S 5-59
5.8.4 Numbers of Community Water Systems Exceeding Potential Regulatory
Levels 5-60
5.8.5 Comparison of Predicted Exceedences For States and Regions 5-62
5.8.6 Estimates of Non-Transient Non-Community Systems Exceeding Potential
Regulatory Levels 5-63
5.8.7 Sensitivity Analysis of Estimates of Systems Exceeding Radon Levels .. 5-65
5.9 Comparison of Current Estimates of Radon Exceedences to Previous EPA
Occurrence Analyses 5-70
5.10. References 5-72
6. POTENTIAL EXPOSED POPULATIONS 6-1
6.1 Data Sources 6-1
6.2 Populations Above Regulatory Levels 6-1
6.3 Special Populations 6-2
7. CO-OCCURRENCE ASSESSMENT 7-1
7.1 Data Sources 7-1
7.2 Co-Occurrence of Radon With Other Contaminants 7-1
7.2 Implications of Co-Occurrence 7-2
7.3 References 7-3
8. MONITORING APPROACHES 8-1
8.1 Background 8-1
8.2 Objectives of Monitoring Program 8-1
8.3 Description of Proposed Monitoring Requirements 8-2
8.4 Costs and Effectiveness of the Proposed Monitoring Requirements 8-8
8.4.1 Incremental Skills/Equipment Requirements and Cost of
Radon Monitoring 8-8
8.5 References 8-8
i
APPENDIX A. DATA MANAGEMENT METHODS AND SUPPLEMENTAL DATA SETS A-l
A.1 Data Management and Manipulation A-l
A.2 Supplemental Data Sets A-6
Methods, Occurrence, and Monitoring Document for Radon
IV
-------
APPENDIX B. STATISTICAL METHODS B-l
B.I MLE Approach to Estimating Summary Statistics for Radon Data Sets B-l
B.2 Calculation of Proportions of Systems Above Radon Levels and Confidence Limits on
Proportions (Distributional Approach) B-3
B.3 Estimation of Confidence Intervals (Distribution-Free Method) B-6
APPENDIX C. SOURCES OF VARIABILITY IN RADON MONITORING STUDIES OF
GROUNDWATER SYSTEMS C-l
C.I A Variance Apportionment (ANOVA) Model For Evaluating Radon
Data Sets C-l
C.I.I Identification of Sources of Radon Variance C-2
C.1.2 Estimating Contributions to Variance C-3
C.2. Analytical Variance C-4
C.3 Combined Sampling and Analytical Variance C-6
C.4 Combined Sampling, Analytical and Temporal Variance C-8
C.5 Combined Sampling, Analytical, and Between-Well Variance C-l 1
C.6 Combined Sampling, Analytical, Temporal, and Intra-System Variance C-l2
C.7 Variance from All Sources C-13
C.8 Estimates of Variance Contributions C-15
C.9 References C-17
APPENDIX D. ESTIMATED NUMBERS OF COMMUNITY GROUNDWATER
SYSTEMS EXCEEDING REGULATORY LEVELS, BY REGION, STATE,
AND SIZE D-l
D.I. Proportions of Systems Exceeding Potential Radon Regulatory Levels in
the Eight NIRS regions D-2
D.2. Proportions of Systems Exceeding Potential Radon Regulatory Levels in
Seven States With Supplemental Data D-12
APPENDIX E. ESTIMATED NUMBERS OF NON-TRANSIENT NON-COMMUNITY
GROUNDWATER SYSTEMS EXCEEDING POTENTIAL REGULATORY LEVELS. E-l
E.I Number of Non-Transient Non-Community Systems in the U.S E-l
E.2. Distribution of Radon Levels in Non-Community Non-Transient Systems E-l
E.3 Estimation of Radpn Distributions in Non-Transient Non-Community Systems of
Different Sizes E-4
E.4. Estimated Numbers and Proportions of Non-Transient Non-Community Systems
Exceeding Potential Regulatory Levels E-6
Methods, Occurrence, and Monitoring Document for Radon
-------
-------
1.
INTRODUCTION
1.1 Purpose of This Document
This Methods, Occurrence, and Monitoring (MOM) Document has been developed by
EPA in support of the rulemaking process for radon in drinking water. The Agency is proposing
a Maximum Contaminant Level Goal (MCLG) and National Primary Drinking Water
Regulations (NPDWR) for radon-222 in public water supplies (EPA, 1999a). The purposes of
this document are:
• Identification of available analytical methods for monitoring radon in groundwater
sources and in drinking water,
• Discussion of the patterns of occurrence of radon in groundwater and drinking water, and
• Explanation of alternative monitoring schemes for assuring compliance with the proposed
rule.
1.2 Statutory Requirements
The 1996 Amendments to the Safe Drinking Water Act (PL 104-182) establish a new
charter for public water systems, states, tribes, and EPA to protect the safety of drinking water
supplies. Among other mandates, Congress amended Section 1412 to direct EPA to take the
following actions regarding radon in drinking water.
Withdraw the 1991 Proposed Regulation for Radon
Congress specified that EPA should withdraw the drinking water standards proposed for
radon in 1991.
Arrange for a National Academy of Sciences Risk Assessment.
The amendments in § 1412(b)(13)(B) require EPA to arrange for the National Academy
of Sciences (NAS) to conduct an independent risk assessment for radon in drinking water and an
assessment of the health risk reduction benefits from various mitigation measures to reduce
radon in indoor air. :
Set an MCLG, MCL, and BA Tfor Radon-222
Congress specified in § 1412 (b)(3)(C) that EPA should propose a new MCLG and
NPDWR (an MCL, BAT, and monitoring, reporting, and public notification requirements) for
radon-222 by August, 1999. EPA is also required to finalize the regulation by August, 2000. As
a preliminary step, EPA was required to publish a radon health risk reduction and cost analysis
Methods, Occurrence and Monitoring Document for Radon
1-1
-------
(HRRCA) for possible radon MCLs for public comment by February, 1999. This analysis must
consider seven topics: (1) health risk reduction benefits that come directly from controlling
radon; (2) health risk reduction benefits likely to come from reductions in contaminants that
occur with radon; (3) costs; (4) incremental costs and benefits associated with each MCL
considered; (5) effects on the general population and on groups within the general population
likely to be at greater risk; (6) any increased health risk that may occur as the result of
compliance; and (7) other relevant factors, including the quality and extent of the information,
the uncertainties in the analysis, and factors with respect to the degree and nature of the risk.
Set an Alternative MCL (AMCL) and Develop Multimedia Mitigation (MMM)
Program Guidelines
The amendments in § 1412(b)(13)(F) introduce two new elements into the radon in
drinking water rule: (1) an Alternative Maximum Contaminant Level (AMCL) and (2) radon
multimedia mitigation (MMM) programs. If the MCL established for radon in drinking water is
more stringent than necessary to reduce the contribution to radon in indoor air from drinking
water to a concentration that is equivalent to the national average concentration of radon in
outdoor air, EPA is required to simultaneously establish an AMCL. The AMCL would be the
standard that would result in a contribution of radon from drinking water to radon levels in
indoor air equivalent to the national average concentration of radon in outdoor air. If an AMCL
is established, EPA is to publish guidelines for state multimedia mitigation (MMM) programs to
reduce radon levels in indoor air. Section V describes what a state or public water-system must
have in their multimedia mitigation program.
Evaluate Multimedia Mitigation Programs Every Five Years
Once the MMM programs are established, EPA must re-evaluate them no less than every
five years. [§1412(b)(13)] EPA may withdraw approval of programs that are not expected to
meet the requirement of achieving equal or greater risk reduction.
DevelopMonitoring Requirements and Characterize Contaminant Occurrence
Under every SDWA rule, EPA is required to develop monitoring requirements to assure
compliance with the rule. Water systems are responsible for conducting monitoring of drinking
water to ensure that it meets all drinking water standards. To do this, water systems and states
use analytical methods developed by government agencies, universities, and other organizations.
EPA is responsible for evaluating analytical methods developed for drinking water and
approves those methods that it determines meet Agency requirements Laboratories analyzing
drinking water compliance samples must be certified by the EPA or the state. Chapter 2 of this
document reviews the available analytical methods for radon in drinking water and their
performance and costs.
Methods, Occurrence and Monitoring Document for Radon 1 -2
-------
EPA must also characterize the sources of drinking water contaminants, their fate and
transport properties, and how they relate to potential exposures. Available data related to the
occurrence of contaminants must be evaluated, and the patterns of occurrence across different
regions of the country, different types of water systems (community and non-community) and in
water systems of different sizes, must also be evaluated in ordeir to develop a national picture of
the distribution of contaminants. The degree to which the occurrence of the contaminant is
correlated with that of other contaminants must also be evaluated. Chapters 3 through 7 of this
document address these issues.
Whether addressing a regulated or unregulated contaminants, EPA establishes
requirements as to how often water systems must monitor for the presence of the subject
contaminant. Water systems serving larger populations generally must conduct more monitoring
(temporally and spatially) because there is a greater potential human health impact of any
violation, and because of the physical extent of larger water systems (e.g., miles of pipeline
carrying water). Small water systems can receive variances or exemptions from monitoring in
limited circumstances. In addition, under certain conditions, a state may have the option to
modify monitoring requirements on an interim or a permanent basis for regulated contaminants,
with a few exceptions. Chapter 8 of this document discusses monitoring strategies for
determining compliance with the proposed rule.
Methods, Occurrence and Monitoring Document for Radon
1-3
-------
-------
2.
ANALYTICAL METHODS
2.1 Introduction
This chapter addresses the analytical methods that may be applicable to the measurement
of radon in drinking water samples. It does not recommend a specific method for radon analyses,
but rather, identifies possible candidate techniques and evaluates the extent to which the
performance of those techniques has been demonstrated.
As part of its overall responsibility for regulating the nation's drinking water supplies, in
1991 EPA proposed regulations on various radionuclides iinder 40 CFR Parts 141 and 142 (July
18, 1991, FR 56 [138]: 33050-33127). Although eventually withdrawn, part of that proposal
addressed the regulation of radon (222Rn or radon-222). Among other topics, the proposal
discussed methods for the analysis of radon in drinking water.
As EPA prepares to propose new regulations for radon in drinking water, the Agency has
reviewed and updated the information on the analytical techniques that appeared in the 1991
proposal (EPA 1991). Specifically, in 1998, at EPA's direction, SAIC reviewed the information
in the 1991 proposal and also conducted an electronic literature search to identify additional
analytical techniques that might be used to measure the concentration of radon in drinking water.
The focus of the 1998 effort was to determine if new monitoring techniques had become
available since the 1991 proposal. The techniques identified by that search were further
evaluated to determine their performance capabilities and possible costs. The remainder of this
chapter addresses the following aspects of the techniques:
• Inventory of methods
• Performance capabilities of the methods :
• Skill requirements
• Practical availability of methods
• Anticipated unit costs
• Practical performance and analytical uncertainties
• Degree to which each method meets EPA's regulatory needs
This last section summarizes the results of the review of the analytical techniques relative to
EPA's need for a method for a nationwide compliance monitoring program. The focus of this
section is on techniques for the analysis of radon in drinking water, and as such, does not
attempt to review information relevant to the analysis of other environmental matrices.
Methods, Occurrence, and Monitoring Document for Radon
2-1
-------
2.1 Inventory of Methods
The 1991 EPA proposal focused on two techniques for the analysis of radon in drinking
water: liquid scintillation counting and the Lucas cell. The 1991 discussion of these techniques
is summarized hi Sections 2.1.1 and 2.1.2, below.
Five newer techniques, or combinations of techniques, were identified in an electronic
search of the open literature. Because EPA had reviewed older analytical techniques prior to
proposing the radionuclides rule in 1991, the search was constrained to identify publications that
have appeared since 1990, in an effort to identify newer techniques that may not have been
considered hi conjunction with the 1991 proposed rule on radionuclides. The discussion of the
five newer techniques is presented in Sections 2.3.1 to 2.3.5.
2.1.1 Liquid Scintillation Counting
Radon is an alpha-emitting radionuclide and is just one of 14 radionuclides in what is
known as the "uranium series," the term used to describe the chain of 15 elements that begins
with 238U and ends with 206Pb, a stable (non-radioactive) element. 222Rn is the seventh element in
the series, created as a decay product of 225Ra. Radon undergoes radioactive decay itself,
forming 218Po through the loss of an alpha particle. Polonium decays through the emission of a
beta particle to form 214Pb. The portion of the decay series from radon onward is illustrated in
the Exhibit 2-1, and includes the manner of the decay (alpha or beta particle) and the half-life of
each element.
Exhibit 2-1. Radon Decay Series
Element
222Rn
218p0
214pb
2.4Bj
2I4Po
2.0pb
210Bi
2,0p0
206pb
Decay Emission
alpha
alpha
beta
beta
alpha
beta
beta
alpha
beta
Half-life
3.8 days
3 minutes
27 minutes
20 minutes
1.6 xlO"4 seconds
22.3 years
5 days
138 days
stable
Methods, Occurrence, and Monitoring Document for Radon
2-2
-------
Radon's alpha particle emissions can be used as the basis for measuring radon in a variety
of environmental media. The principal technique for radon analysis considered by EPA in the
1991 proposal was liquid scintillation counting.
Scintillation counting refers to the measurement of the light emitted when an alpha
particle from the sample strikes some form of scintillating material. The two most common
forms of scintillators are the scintillation disk, with is a planchet or metal disk coated with zinc
sulfide, and a liquid scintillation fluid or an organic phosphor. The light emitted from the
scintillator strikes the surface of a photomultiplier tube that is placed next to the sample in a
light-proof container, releasing electrons from the photocathode in the tube at levels proportional
to the intensity of the emitted light. The electrical pulses that result are counted to determine the
number of disintegrations per minute (dpm) that occur, which can be related to the concentration
of a given radionuclide.
In liquid scintillation counting, a volume of sample is mixed with the organic phosphor
contained in a mineral oil solution or "cocktail" in a glass container which is then placed in the
instrument, where it is held against the photomultiplier for counting.
As noted in the 1991 proposal, radon can be measured through a direct, low-volume
liquid scintillation technique in which approximately 10 ml of water is added to a vial with the
scintillation cocktail, mixed, and placed in a liquid scintillation counter. The sample can be left
in the counter for periods ranging from several minutes to several hours, depending on the level
of radon in the sample.
The energy of the alpha particles released by radioactive decay is characteristic of the
radionuclide. In the case of liquid scintillation counting techniques, the counting apparatus can
be configured to measure the scintillations is narrow energy ranges across the emission spectrum.
In the case of radon analyses the counter can set to look in the portion of the energy spectrum
that represents the alpha particles emitted by 222Rn and as well as 218Po and 214Po, the next two
alpha-emitting daughters in the series. Given the short half-lives of these two daughters, their
alpha particle emissions can be measured along with that of the radon itself in less than an hour
of counting time. From a practical standpoint, the emissions of three alpha particles can be
measured and related back to one radon atom, thereby amplifying the signal from that single
radon atom's decay.
It is important to distinguish between an analytical technique and a specific analytical
method. Liquid scintillation counting is a technique. EPA's 1991 proposal stated that the
Agency planned to establish a specific analytical method, EPA Method 913, based on the liquid
scintillation technique. '.
Methods, Occurrence, and Monitoring Document for Radon
2-3
-------
2.1.2 The Lucas Cell Technique
The second technique that EPA considered in the 1991 proposal involved the Lucas cell,
a specially constructed 100- to 125-ml metal cup coated on the inside with zinc sulfide (a
scintillator) and fitted with a transparent window. The Lucas cell replaces the scintillation disk
or planchet in the counting instrument. The analysis of radon in a water sample is accomplished
by purging a volume of the sample with radon-free helium or "aged" air (air in which the radon
has already decayed). The purge gas removes the dissolved radon from the sample and carries it
into a Lucas cell that has been evacuated of any air. After an equilibration period of three to four
hours, the Lucas cell is placed in the counter and the scintillations resulting from the alpha
particles striking the zinc sulfide are counted through the transparent window.
The Lucas cell technique is a modification of other scintillation counting techniques and
was considered by EPA because it can permit the measurement of lower levels of radon than in
the liquid scintillation technique. However, the Agency noted that the method is more difficult
to use than the liquid scintillation method, in particular, requiring specialized glassware and
greater skill on the part of the analyst. It was the Agency's intent to include procedures for the
Lucas cell technique in Method 913, as an adjunct to the liquid scintillation procedures.
2.2 Major Analytical Methods
2.2.1 Liquid Scintillation Counting and Lucas Cell Methods
Subsequent to the 1991 proposal, EPA published a report on its method validation efforts
in fiscal year 1992 (Pia and Hahn 1992). That report described the results of collaborative
studies for the analysis of radon in drinking water and provided performance data on both the
direct low-volume liquid scintillation technique and the Lucas cell technique that the Agency
planned to incorporate into Method 913.
The 1992 study evaluated both the liquid scintillation technique and the Lucas cell
technique for the analysis of performance evaluation samples spiked with radon at levels of 111
and 153 picoCuries per liter (pCi/L).1 The 1992 study also investigated two means of spiking the
samples. The first sample was spiked with radium (226Ra), which produces radon as a decay
product. The second sample was produced using a "radon generator" in which 226Ra was bound
to a strong cation exchange resin. The decay of the radium released radon into the water, while
the remaining radium was still bound to the resin and therefore not present in dissolved form in
1 The Curie is a measure of a quantity of radioactive material. Specifically, a Curie is
defined as the quantity of a radioactive nuclide which produces 3.7 x 10'° atomic disintegrations
per second. The prefix "pico" stands for one trillionth (10~12), thus, a picoCurie would be 3.7x10'
2 atomic disintegrations per second.
Methods, Occurrence, and Monitoring Document for Radon
2-4
-------
the sample. The data from the 1992 study are summarized Exhibit 2-2, for both techniques, both
radon concentrations, and both sources of radon. :
Exhibit 2-2. Summary of EPA 1992 Collaborative Study Data
Technique !
LSCRa
LCSRn
LCRa
LCRn
LCSRa
LSCRn
LCRa
LCRn
Spike
Cone.
(pCi/L)
111
111
111
111
153
153
153
153
Mean Cone.
Found
(pCi/L)
112
112
114
127
156
154
158
174
Mean
Recovery (%)
101
101
102
114
102
101
103
114
Precision
within Lab
(PCi/L)
i 9
14
9
16
1 10
17
: 10
: 17
Reproducibility
(pCi/L)
12
24
12
23
18
28
16
28
% Bias
0.7
1.1
2.3
14.5
2.3
0.9
3.4
13.7
1 LSC Ra = Liquid scintillation counting of samples spiked with 226Ra
LSC Rn = Liquid scintillation counting of samples spiked with radon generator
LC Ra = Lucas cell counting of samples spiked with 226Ra
LC Rn = Lucas cell counting of samples spiked with radon generator
Another important aspect of the EPA 1992 collaborative study were the findings with
regard to sampling, sample containers, and sample handling. EPA conducted single-laboratory
studies that were designed to evaluate factors related to sampling methods for proficiency testing
of radon laboratories. Such performance evaluation (PE) samples have been used as an
important aspect of EPA's certification program for laboratories performing analyses under the
Safe Drinking Water Act monitoring programs. The 1992 report describes studies of four sample
collection techniques (displacement, immersion, catch, and grab sampling). EPA also evaluated
the effectiveness of two types of scintillation vial cap materials (polypropylene and PTFE-lined
caps) at maintaining the integrity of the samples. The effects of headspace or bubbles in the
sample containers were also evaluated.
The analysis of sampling techniques found that the four techniques were statistically
equivalent, in that no systematic error was introduced into the results by any of the four
techniques. The report stated that displacement sampling and immersion sampling were the most
conservative sampling approaches, requiring only that the flow of water from which the sample
is collected not be aerated or turbulent.
Methods, Occurrence, and Monitoring Document for Radon
2-5
-------
With regard to the vial cap materials, EPA found that as much as 10-15% of the radon in
the sample may be lost by its sorption into the polypropylene cap itself. The loss appeared to
occur within the first four hours after the sample was collected. Caps equipped with PTFE liners
did not show this loss of radon over time.
As with volatile organic constituents, radon in water samples may be lost into the
headspace of the sample container. Although careful sampling techniques should result in the
sample container being filled to the top and sealed with no headspace, changes in sample
temperature will affect the solubility of all gases dissolved in the sample, including air and
radon. As the temperature of the sample in the sealed container increases, the solubility of all
gases will decrease and they may come out of solution, forming bubbles at the top of the
container. It is not uncommon to observe air bubbles in a container that form as a result of such
a temperature increase. Given the typical levels of radon in water, it is highly unlikely that a
visible bubble of pure radon would form. However, the concentration of air is much higher and
if radon is present in the sample, then the radon can partition into the headspace created by a
bubble of air and the radon in the headspace would be lost from the sample when the container is
opened.
EPA compared the radon concentrations measured in samples containing six air bubble
volumes ranging from 0 - 5 ml in 63-ml sample bottles. The results of this study indicate that for
bubbles up to 0.25 ml in volume, there was no significant loss of radon from solution. At a
bubble volume of 0.5 ml, the loss of radon was 12%, with even larger losses for larger bubbles.
Based on the solubility of air at 20 °C and 24 °C, EPA concluded that the headspace resulting
from the formation of air bubbles as the sample warmed did not present a problem with respect
to the loss of radon from the sample.
In the 1992 report, Pia and Hahn noted that there was a relatively large positive bias for
the Lucas cell technique when using the radon generator approach (13.7 and 14.5% for the 111
pCi/L and 153 pCi/L sample, respectively). They attributed this bias to a problems with
transferring the radon standard supplied by EPA and calibration of the instrument in the Lucas
cell procedure. They indicated that the systematic error could be addressed by standardizing the
technique used to transfer the sample and the radon standard, and that this issue would be
addressed in EPA Method 913.
2.2.2 Standard Method 7500: Radon Liquid Scintillation Counting
This method is published in Standard Methods for the Examination of Water and
Wastewater, (APHA 1996). The method is specific for 222Rn in drinking water supplies from
groundwater and surface water sources. This method grew out of EPA efforts in connection
with the 1991 radionuclides proposal. In that proposed rule, EPA discussed the development of
EPA Method 913, a liquid scintillation technique for radon analysis. Subsequent to the 1991
radionuclides proposal, EPA submitted the draft procedure to APHA and it was published in
Methods, Occurrence, and Monitoring Document for Radon
2-6
-------
Standard Methods as SM 7500-Rn. Having been published by a consensus organization
(APHA), there was no need for EPA to pursue the promulgation of a separate EPA method.
In Standard Method 7500-Rn, the radon is partitioned selectively into a mineral-oil
scintillation cocktail that is immiscible with the water sample. The sample is held in the dark for
three hours. This "dark adaptation" serves two purposes. Firsts exposure to light can cause the
cocktail to scintillate and this period in the dark allows this light-induced scintillation to dissipate
before sample analysis, thereby reducing the background count. Secondly, the decay of the
radon creates a number of short-lived daughter products. Compared to the half-lives of its
daughter products, the half-life of radon is relatively long, 3.8 days (see the table in Section 2.1).
Thus, during this equilibration period, the alpha emissions due to the daughter products 2I8Po and
214Po become equal to that of the radon itself and the signal from the radon is essentially
amplified by a factor of three. After the equilibration period, the alpha particle emissions from
the sample are counted in a liquid scintillation counter using a region or window of the energy
spectrum optimal for the alpha particles from the three radionuclides. The results are reported in
units of pCi/L. The diffusion of radon is affected by temperature and pressure. Therefore, it is
important to allow the samples to equilibrate to room temperature before processing.
The precision of the method is affected by the background signal in the counting window
used for analysis. A procedure is provided for selection of the analytical window to minimize
the background contribution to the measurement. An important aspect of SM 7500-Rn is that it
does not include any mention of the Lucas cell technique that EPA had planned to include in
EPA Method 913. '
The performance data in SM 7500-Rn shown in Exhibit 2-3 were incorporated from the
1992 EPA collaborative study cited earlier, which included 36 participants. However, the EPA
1992 study data were incorporated without differentiation between the liquid scintillation
counting and Lucas cell techniques, even though, as previously noted, SM 7500-Rn does not ever
mention the use of the Lucas cell.
Exhibit 2-3. Standard Method 7500-Rn Performance Data
Sample Cone.
pCi/L
111
153
Accuracy
%
101 - 102
102-103
Repeatability
pCi/L
9
10
Reproducibility
pCi/L
12
16-18
Bias
%
0.7 - 2.3
2.3 - 3.4
The significance of the inclusion of the Lucas cell data is probably not great. As can be
seen by comparing the data above with that in Exhibit 2-2, the accuracy data reported by EPA
differ only by one percent between the two techniques. At each sample concentration, the
reported precision within a laboratory (repeatability in the table above) is the same for both
Methods, Occurrence, and Monitoring Document for Radon
2-7
-------
techniques and differs by only 1 pCi/L between the two radon activity levels. The most notable
differences are in the reproducibility figures, where the lower value in the 16 - 18 pCi/L range,
and the higher value in the 2.3 - 3.4% bias range both come from the Lucas cell technique.
SM 7500-Rn incorporates other important information from the EPA studies as well. For
example, the method specifies the use of glass sample containers or glass scintillation vials with
PTFE- or foil-lined caps, avoiding the problems associated with the loss of radon into the
polymer caps. The method describes the sample collection and employs the immersion
procedure, although the method does not use that term by name.
23 Other Radon Measurement Techniques
As noted in Section 2.1, EPA's literature search identified several other recently
developed radon measurement techniques, which are discussed in rum in the following sections
2.3.1 Delay-Coincidence Liquid Scintillation Counting System
The literature search performed in 1998 identified a report of an automated liquid
scintillation counting system for determination of 222Rn in ground water (Theordorsson 1996).
The focus of the report was on the use of radon activity levels for earthquake prediction in
Iceland. The report describes an automated radon detection system intended for mostly
unattended operation.
The technique involves a two-part system which includes a prototype assembly for
transferring radon (222Rn) from water to toluene and a single phototube liquid scintillation
counter. The radon in the toluene is detected by liquid scintillation counting, using a method
known as delayed coincidence counting. Delayed coincidence counting takes advantage of the
fact that the next four daughter products of radon all have short half-lives. As shown in Exhibit
2-1, the half-lives of 218Po, 214Po, and 214Bi are all under 30 minutes, and the half-life of 214Po is
only 0.16 milliseconds. The delayed coincidence counter is programmed to respond to the beta
particle decay of an atom of 2I4Bi. Upon detecting that beta particle from 2I4Bi, the system waits
about 5 microseconds and then opens an electronic "gate" to the detector channel that
corresponds to the energy of the alpha particle decay of 214Po and holds that gate open for about 1
millisecond. The result is that the background count measured by the detector is greatly reduced
because the detector is only looking for 214Po scintillations in the very narrow time interval
immediately after the beta particle decay of 214Bi. The detection efficiency for the delayed
coincidence counting of 2I4Po is about 95%.
Most of the other aspects of the technique are modifications of those used in liquid
scintillation counting and the Lucas cell techniques. For example, the transfer of the radon from
the water sample by purging is employed in the Lucas cell, though in this case, the final reservoir
is an organic liquid not unlike that used in liquid scintillation counting.
Methods, Occurrence, and Monitoring Document for Radon
2-8
-------
This technique is designed to permit the use of a much larger water sample than any of
the previously described techniques. The use of a larger sample compensates for the fact that the
percentage of radon transferred from the water to the toluene is only about 40%. In addition,
delayed coincidence counting essentially ignores the alpha decay of the parent radon and 218Po,
thereby reducing the technique's sensitivity by a factor of three. Theodorsson anticipates this
concern, arguing that although the ability to count multiple pulses for each disintegration of a
radon atom is generally considered to increase the sensitivity and accuracy of measurements, that
assumption is in error because the pulses are "not statistically independent." He states that the
delayed coincidence counting
"hardly effects [sic] the resulting accuracy and sensitivity, compared to counting
in a broad alpha-beta window, although the latter may give a pulse rate almost
five times higher."
Unfortunately, Theodorsson does not present any performance data to substantiate this statement.
At the time of the report, the author had only constructed a prototype system that was
designed for primarily unattended operation in the field. This technique may be attractive for
various types of low-level environmental radon measurements since it is relatively simple, very
sensitive, and well protected from disturbances. However, no multiple laboratory data describing
such performance characteristics as sensitivity are provided in the article. Thus, it is not possible
to evaluate this technique more fully.
2.3.2 Activated Charcoal Passive Radon Collector
A technique that measures 222Rn in river water using an activated charcoal passive radon
collector has been described by Yoneda, et al. (1994). Unlike other radon methods that require
the collection of a discrete water sample, the passive radon collector is immersed in the river by
means of a string.
The radon collector used in this study consists of a sealed polyethylene bag containing a
thin layer of activated charcoal. As water passes through the collector, the radon is adsorbed
onto the charcoal and retained there. After a suitable period of immersion in the water of
interest, the bag is removed and sealed in an air-tight plastic container and allowed to stand
overnight until secular equilibrium among the decay products was achieved. The radon on the
charcoal is determined by gamma-ray spectral analysis of its 214Pb and 214Bi daughter products.
The author describes experiments that evaluated the performance of the passive collector,
including an evaluation of bag thickness, amount of charcoal used in the collector, immersion
time and, most importantly, the use of dry and wet charcoal. This method claims to have the
advantage of simplicity, low cost, and the ability to measure the average radon activity in
flowing water over a specified period of time.
Methods, Occurrence, and Monitoring Document for Radon
2-9
-------
The author reported that:
• The mean amount of 222Rn adsorbed by the collector was about reached a maximum
when the quantity of charcoal reached 20 grams, and that the quantity of radon did not
appear to be proportional to the amount of charcoal in the collector. Also, the
charcoal should be fully spread out in a single layer within the bag.
• The thickness of the polyethylene bag did have an impact on the final results (a thin
film collects more 222Rn), but it was noted that when wet charcoal was used, the
effectiveness of the polyethylene film decreases. The general recommendation was
that a relatively thick polyethylene film, 0.005 cm, be used because it is stronger and
less likely to tear.
• Collectors containing dry charcoal collected more 222Rn than those containing wet
charcoal. However, given the difficulty in keeping the charcoal dry during the
immersion phase, it was concluded that, in order for efficient quantitative
measurement of radon, wet charcoal should be used in the collectors. A revised radon
absorption equation was developed to indicate the amount of 222Rn collected in the
wet charcoal collector.
The principal advantage of this method is that a discrete sample is not required, as the
passive collector is immersed directly in the body of water. This method does not measure radon
directly, rather it measures the decay of the daughter ions. An equation is given that allows the
user to quantify the total 222Rn absorbed by fully wet-activated charcoal sealed in a polyethylene
bag in water.
The study report includes data for a variety of tests of the collection device. While some
tests were conducted at lower radon levels, the majority of the performance data were generated
from waters containing greater than 100 Bq/L of 222Rn (>2700 pCi/L). Thus, it is not clear how
well the method would perform at the levels of interest to EPA. The available performance data
described in the article are limited to a single laboratory.
Because of the way that the monitoring is conducted, e.g., immersing the collector in the
water body and monitoring the average radon concentration over a long time period (6-10 days),
it may not be a particularly useful technique for monitoring compliance with a Maximum
Contaminant Level (MCL). Low radon levels over a portion of the monitoring period could
mask higher levels that would violate the MCL. However, if performance data were available for
radon levels near the likely MCL (300 pCi/L), this technique might be useful as a screening
method. If used as a screening method, long-term sample results that averaged over the MCL
could be expected to violate the MCL if a grab sample were analyzed using a method such as
Standard Method 7500-Rn, so no additional testing1 would be needed. In contrast, long-term
sample results below the MCL would still require confirmation using another technique on a
grab sample. However, such screening might not be cost-effective.
Methods, Occurrence, and Monitoring Document for Radon
2-10
-------
In addition, the need to leave the collector in a container of running water for 6-10 days
imposes some practical limitations in comparison to other methods that employ some a sample
collected over a short period (e.g, a few minutes). The adsorption coefficient of radon from
water onto the charcoal can be defined as:
Bq radon per gram of charcoal
adsorption coefficient (k) = •-
Bq radon per mL of water
where Bq, the Becquerel, is the SI unit of radioactivity corresponding to 1 disintegration per
second (approximately 27 picoCuries). It may be possible that the adsorption coefficient reaches
a constant during the exposure period of the collector. However, the study does not provide
sufficient data to determine if that is the case. If the adsorption coefficient is not found to
constant, it would be necessary to determine the total volume of water passing over the collector
during that 6-10 day period. In some monitoring situations, such measurements would likely be
more difficult than the measurement of the radon itself.
No collaborative data were available for this method.
2.3.3 Degassing Lucas Cell ,
A paper by Mullin and Wanty (1991) compares the use of a "degassing Lucas cell"
(DLC) technique with liquid scintillation counting This paper describes the degassing Lucas cell
technique in general terms, noting that a paper by Reimer in the same volume of the USGS
Bulletin provides greater detail. The paper by Reimer was not reviewed directly, as the
comparisons conducted by Mullin and Wanty provided more useful information.
As noted hTSection 2.2., the Lucas cell technique is a well-established method for the
analysis of radionuclides in water, including radon. In the degassing Lucas cell technique, a
water sample is agitated in a closed vessel to extract the radon. The air in the headspace of the
vessel is sampled with a gas-tight syringe and injected into a Lucas cell for counting. The
principal advantage of this technique is that the results can be obtained in the field, at each site,
which was the apparent reason for developing the technique.
The primary disadvantage of this method is that unless the sample is analyzed
immediately, the radon level can be biased low by radon diffusing out of the syringe containing
the air sample. Increased lag time from sampling to analysis via the DLC leads to greater
uncertainty and usually lower radon measurements, both of which were attributed to loss of
radon from the syringes in which the samples were stored. The loss of radon through radioactive
decay during the lag time between sample collection and measurement was accounted for by
using an exponential formula that corrects for the decay of the radon in the sample. However,
that correction factor does account for the diffusion losses of radon from the syringe.
Methods, Occurrence, and Monitoring Document for Radon
2-11
-------
In addition, as written, this method does not expressly include the three-hour equilibrium
period, in an effort to speed the use of the technique for field measurements. The lack of the
equilibration period presents concerns as well. In particular, 2I8Po, one of the short-lived progeny
of 222Rn, closely approaches secular equilibrium with 222Rn in about 10 minutes, and may not be
accounted for adequately in the calibration scheme. Finally, because the DLC analyses is
performed at a tune when the net alpha activity of 218Po is rapidly building, large errors in
apparent radon levels may result.
The authors of the study concluded that the liquid scintillation technique was more
accurate than the degassing Lucas cell technique, but that the degassing Lucas cell may have
utility for reconnaissance sampling, where the results can be used to design sampling schemes
for use of the more accurate liquid scintillation technique. However, field measurements may
not be an important factor relative to SDWA compliance monitoring for radon.
2.3.4 Electret lonization Chamber System
Several articles were found that discuss the use of an electret system for the measurement
of radon (Tai-Pow 1992, Sabol et al. 1995). Additional information was provided to EPA by the
US manufacturer of the electret device, Rad Elec Inc., of Frederick, MD.
An electret is a device which has been treated to hold a stable electrostatic-field potential
(initially 700 to 750 volts). In the case of these two studies, the electret is made of a wafer of
Teflon that is housed in a chamber made of electrically-conducting plastic. The device is called
an electret passive environment radon monitor (E-PERM) by the manufacturer of the device.
The decay products from the radon gas enter the chamber through the filtered inlet at the
top and the alpha particles striking the electret discharge the static charge on the electret. The
surface charge of the electret is measured before and after exposure by using a specially designed
voltage reader. This electric field sensor can detect small changes on the electret. The electret is
designed to handle exposures of two to seven days at levels of 0.04 to 1.85 Bq/L (1 to 50 pCi/L)
of radon in air.
Electret ionization chambers are simple, portable, and easy to use. They are also well-
suited for field measurements, since more than one measurement can be made from the same
electret. Drawbacks to this simple and relatively inexpensive method include poor
reproducibility at lower radon levels, uncertainty in the use of manufacturer-suggested gamma
correction factors, and limited reusability. The electret device lacks specificity for radon. The
surface charge of the electret will change with exposure to gamma radiation from within the
sample chamber or from an external gamma source. It will also change in response to the alpha
decay of other volatile radionuclides that enter the chamber headspace from the water.
When measuring radon concentrations in air, the gamma radiation can be subtracted
through the use of voltage-dependent correction factors, resulting in improved accuracy. In the
Methods, Occurrence, and Monitoring Document for Radon
2-12
-------
studies cited above, the end results showed that a higher concentration of radon in water may
result in elevated airborne radon concentration in the surrounding areas, including increased
radon activity in buildings served by a hot spring water. For routine waterborae radon
monitoring, including use in field conditions, the technique based on electret ion chamber
technology may sometimes be a suitable choice. :
In a 1990, a survey of laboratories conducting radon analyses in drinking water was
performed by Wade Miller Associates, under contract to EPA. The goals of that study were to
identify the types of certification programs that exist for radon analyses in drinking water, to
identify laboratories capable of performing the analyses, and to determine the daily analysis
capacity of each identified laboratory. Of 45 commercial and state laboratories contacted in
1990, only one listed the electret method. ;
Recent information provided by the US manufacturer included cited three additional
studies that were not directly reviewed by SAIC. These include the following papers and
presentations:
• Kotrappa, P. and Jester, W.A.,"Electret Ion Chamber Radon Monitors Measure
Dissolved 222Rn in Water," Health Physics, 64: 397-405 (1993)
• Colle, R. Kotrappa, P., and Hutchinson, J.M.R., "Calibration of Electret-Based
Integral Radon Monitors Using NIST Polyethylene-Encapsulated 226Ra/222Rn
Emanation (PERE) Standards," Journal of Research of National Institute of
Standards and Technology, 100: 629-639(1995). '•
• Budd, G, and Bentley, C., "Operational Evaluation of the EIC Method for
Determining Radon In Water Concentrations," 1993 International Radon Conference,
Hosted by AARST.
Those studies provide precision and bias data on the electret technique over a wide range of
concentrations. According to the manufacturer, the electret technique has recently been certified
by the States of Maine and New Hampshire for monitoring radon in water.
As summarized by the manufacturer, the precision of the electret technique ranged from 4
to 10% across all three of the studies. The bias of the technique was estimated by the
manufacturer to be from -17% to +1% in these three studies, following the application of a
correction factor of 1.15 to the initial sample results. Prior to the use of this correction factor, the
bias ranged from -27% to -9% across these three studies. SAIC contacted the manufacturer and
obtained information on the ranges of radon concentrations that were used in these studies.
According to the manufacturer, the Kotrappa and Jester study examined five radon
activity levels, ranging from a low of about 220 pCi/L to a high of 73,200 pCi/L, and found no
significant change in precision and bias across the range. The Colle et al. study examined only
Methods, Occurrence, and Monitoring Document for Radon 2-13
-------
one radon level of 10 Bq/g, which equates to approximately 270,000 pCi/L. The Budd and
Bentley study examined a variety of activity levels, ranging from about 350 pCi/L to 46,000
pCi/L. The first and third studies included at least some activities near the levels of interest to
EPA.
The correction factor recommended by the manufacturer is intended to relate the electret
results to those expected by the liquid scintillation counting method, although the manufacturer
points out that the liquid scintillation method may not be "accurate with traceability to NIST."
As is the case for the activated charcoal collector method described in Section 2.3.2, the
electret method requires a long exposure of the detector to the sample. The range of exposure
times in the papers reviewed by SAIC is 2-7 days. However, unlike the charcoal collector
technique, the electret is exposed to a discrete sample container in a sealed vessel. Thus,
although the measurement may take up to 7 days to complete, the results represent the
concentration of radon in the discrete water sample.
Although the manufacturer's literature indicates that electret technique performed well in
a 1994 US Department of Energy (DOE) "intercomparison" study, those data appear to be for the
measurement of radon in air. No collaborative data for water samples have been identified.
2.4 Performance Capabilities of the Methods
The performance capabilities of these methods for the analysis of radon were difficult to
evaluate in a consistent manner, in part, because many of the methods were developed in
university settings for purposes other than those envisioned by EPA, i.e, not for compliance
monitoring. Wherever possible, SAIC has reviewed the information on the sensitivity (detection
limit) and precision of these methods. The Selectivity of the procedures for 222Rn is generally
excellent and consistent across most of the methods. This is because most of the methods
measure the alpha particle decay of 222Rn and/or its daughter products, and these particles are
released at discrete alpha energies. In the case of 222Rn, the energy of the alpha particle is 5.49
MeV. The exception is the electret method described by Tai-Pow et al, which measures the
change in the electrical potential of the circuit containing the electret. This technique is less
selective for radon than the other techniques, in that it will respond to both gamma radiation and
other volatile radionuclides in the water sample.
As noted earlier, most of the methods lack data from collaborative studies. The two
exceptions are the liquid scintillation method (SM 7500-Rn) and the Lucas Cell method. Both of
these method were evaluated as part of the 1992 EPA collaborative study. The accuracy,
reproducibility, repeatability, and bias data for Standard Method 7500-Rn and for the Lucas Cell
method are shown in Section 2.3, above.
As noted above, the performance capabilities of some of the other techniques have not
been demonstrated for relatively low activities of 222Rn. Several of the techniques were
Methods, Occurrence, and Monitoring Document for Radon 2-14
-------
described as having poorer performance as low radon activities. A number of the papers did not
present data on the sensitivity of the techniques, and in those cases no attempts were made to
estimate the sensitivities. Rather, the radon levels at which performance was demonstrated were
noted in the discussion. '•
Based on the information provided by the manufacturer, the performance of the electret
method has also been relatively well-characteristics. While the summary information suggests
that the bias is greater than that reported for Standard Method 7500-Rn, even the -17% figure is
not so severe as to rule out this procedure, since methods for some organic analytes can be shown
to have similar bias. However, as noted earlier, no collaborative study data on water samples
were identified. '.
2.5 Skill Requirements ;
The two major techniques employed in most of these methods are liquid scintillation and
Lucas cell counting. Neither of these techniques is technically difficult. Liquid scintillation
counting has been used in medical laboratories and environmental research laboratories for over
30 years. The skills required are primarily the ability to remove an aliquot of the sample from
the original vial and adding an aliquot of the scintillation cocktail, sealing the vial, and placing it
into the counter. The counting process is highly automated and the equipment runs unattended
for days, if needed.
The Lucas cell methods described in the papers considered for this report requires
somewhat more manual skill. As noted in the 1991 proposed rule, EPA expects that this
technique would require greater efforts to train technicians than the liquid scintillation technique.
The Lucas cell technique requires that the counting cell be evacuated to about 10 mTorr pressure.
Then, a series of stopcocks or valves must be manipulated to transfer the radon that is purged
from the sample into the counting cell. Potential problems with the analysis, such as a high
background level of radon that can develop over the course of the day, or aspirating water into
the counting cell, can be minimized by a well-trained analyst. However, as EPA concluded in
1991, the Lucas cell technique is not expected to form the sole basis of a compliance monitoring
program for radon in drinking water.
The electret method is relatively simple to perform. The water sample (<150mL) is
transferred to a larger, leak-tight container housing the electret device. The radon escapes from
the water into the air in the container. The electrical potential (voltage) of the electret must be
measured before and after the analysis, using a specially designed sensor.
Overall, although the requirements vary across the techniques, the skills required to
measure radon using the techniques described here are generally comparable to those required
used to perform gas chromatographic or atomic absorption analyses, methods that EPA has
identified for use in quantifying common organic and inorganic contaminants in water samples.
Methods, Occurrence, and Monitoring Document for Radon
2-15
-------
In discussions between EPA and the water utility industry, concerns have been expressed
about the difficulties in collecting samples and the skills that may be required to do so in a
reproducible fashion. As noted in Section 7.0, the ability to generate useful analytical results for
radon is dependent in an important way on the sample collection process. The 1992 EPA
collaborative study evaluated four sample collection techniques and found them all equally good
at providing equivalent results. The State of California has developed a sampling protocol for
radon in water that employs one of the four techniques evaluated by EPA, namely the immersion
technique. SAIC has reviewed a copy of that protocol that was provided to EPA (Jensen, 1997).
As described in the California protocol, the well is purged for 15 minutes to ensure that a
representative sample is collected. Purging simply means that the water is withdrawn from the
well for this period of time. After purging, a length of flexible plastic tubing is attached to the
spigot, tap, or other connection, and the free end of the tubing is placed at the bottom of a small
bucket. The water is allowed to fill the bucket, slowly, until the bucket overflows. The bucket is
emptied and refilled at least once.
Once the bucket has refilled, a glass sample container of an appropriate size is opened
and slowly immersed into the bucket in an upright position. Once the bottle has been placed on
the bottom of the bucket, the tubing is placed into the bottle to ensure that the bottle is flushed
with fresh water. After the bottle has been flushed, the tubing is removed while the bottle is still
on the bottom of the bucket. The cap is placed back on the bottle while the bottle is still in the
bucket, and the bottle is tightly sealed. As noted in the California protocol, the choice of the
sample container is dependent on the laboratory that will perform the analysis, and will be a
function of the liquid scintillation counter that is employed. If bottles are supplied by the
laboratory, there is no question of what container to employ.
Once the sealed sample bottle is removed from the bucket, it is inverted and checked for
bubbles that would indicate headspace. If there are no bubbles, the outside of the sealed bottle is
wiped dry and cap is sealed in place with electrical tape, wrapped clockwise. After the sample
bottle is sealed, a second (duplicate) sample is collected in the same fashion from the same
bucket. The date and time of the sample collection is recorded for each sample.
As described above, the sample collection procedures are not particularly labor-intensive.
Most of the time is spent allowing the water to overflow the bucket. Likewise, there are no
significant manual skills required. Personnel who can manage to slowly fill a 1-liter glass bottle
to collect a sample for analysis of semivolatile organics, or fill a 40-mL VOA vial without
headspace, can certainly collect samples for radon, using the method described above.
2.6 Practical Availability of the Methods
In order to determine the practical availability of the methods, SAIC considered two
major factors. First, the availability of the major instrumentation was reviewed. Secondly,
Methods, Occurrence, and Monitoring Document for Radon
2-16
-------
several laboratories performing drinking water analyses were contacted to determine their
capabilities to perform radon analyses.
The major instrumentation required for Standard Method 7500-Rn is a liquid scintillation
counter. Automated counters capable of what that method terms "automatic spectral analysis"
are available from at least a dozen suppliers. The Lucas cell apparatus is the same as has been
used for radium analyses for many years. The electret system is used for the measurement of
radon in air as well as in water. Information provided by the manufacturer of the electret system
suggests that there are more than 600 users in the US, of whom, the manufacturer estimates, 10%
measure radon in water.
In order to evaluate the availability of laboratory capacity to perform radon analyses, in
early 1998 SAIC contacted the drinking water certification authorities in the states of California,
Maryland, and Pennsylvania. These states were chosen based on SAIC's knowledge of radon
problems associated with the "Reading Prong" that stretches through parts of Pennsylvania and
Maryland, and the overall status of California's laboratory certification program. A total of eight
commercial laboratories were contacted during this initial, survey. Each laboratory was advised
that SAIC was simply collecting information on the availability and relative costs of radon
analyses for drinking water. SAIC was limited in its ability to perform a broader survey, since an
upper limit of nine was placed on the survey, in order to abide by the Federal information
collection regulations.
Six of the eight laboratories that were contacted in the initial survey perform radon
analyses. All the laboratories were certified in one or more states to perform radiochemical
analyses, though it was unclear if the certifications were specifically for radon or the more
general radiochemical analysis category. •
When asked what specific methods were used, the laboratories responded with either the
technique (liquid scintillation counting) or a specific method citation. EPA Method 913 was
cited by two of the six laboratories. As noted earlier, this method is the precursor to the current
Standard Method 7500-Rn. EPA Method "EERF Appendix B" was cited by another laboratory.
The remaining three laboratories indicated that they performed liquid scintillation analyses and
could accommodate requests for methods employing that technique.
When asked about capacity, the laboratories indicated that they perform between 100 and
12,000 analyses per year. The latter figure came from a laboratory that is currently involved in a
large ground water monitoring project in the western US. The next largest estimate was 300 .
samples per year. However, SAIC expects that like any other type of environmental analysis,
given a regulatory driver to perform the analysis, the laboratory capacity would develop quickly.
The 1992 EPA collaborative study on radon analysis (Pia and Hahn, 1992) included 51
laboratories with the capability to perform liquid scintillation analyses. This suggests that there
already exists a substantial capacity for these analyses. Further, the liquid scintillation apparatus
is used for other radiochemical analyses, including tritium. Information from EPA regarding the
Methods, Occurrence, and Monitoring Document for Radon 2-17
-------
performance evaluation program for tritium analyses suggests that there are approximately 100-
200 laboratories with the necessary equipment. Much of the capacity for tritium analyses could
also be used for radon (EPA 1997). As of September 1997,136 of 171 participating laboratories
achieved acceptable results for tritium. Both the total number of participants and the number
achieving acceptable results vary from study to study, but these data indicate that there is already
a substantial capability for liquid scintillation analysis nationwide.
Recent information provided by the manufacturer of the electret indicate that the States of
Maine and New Hampshire are certifying laboratories for drinking water analysis using the
electret method. Several months after the initial laboratory survey, based on information from
the manufacturer, SAIC contacted a laboratory in New Hampshire that uses the electret method
and obtained information on the analysis price for water samples. The laboratory charges $30
per sample for drinking water analyses. They have been certified for drinking water analyses
using the electret method in New Hampshire for at least three years and in Maine for one year.
They have a current capacity of at least 40 samples per week (2000 per year), and indicated that
they could easily increase that capacity to meet demand.
The availability of laboratories is also dependent on laboratory certification efforts in the
individual states with regulatory authority for their drinking water programs. A major
component of many of these certification programs is continued participation by the laboratory in
the current EPA Water Supply (WS) performance evaluation (PE) program. Efforts are
underway at EPA that will lead to the privatization of all of EPA's PE programs, including the
WS studies. Those efforts will affect laboratory certifications for all analytes regulated under the
SDWA, including radiochemicals such as radon. Any delays in implementing a private PE
program will affect not only radon, but the certification status of laboratories for all regulated
analytes.
Because of the issue involved with safe handling of radiochemical standards, there will
likely be fewer laboratories seeking certification for radon than for other non-radiochemical
parameters. However, there is no fundamental regulatory reason that a radon laboratory in one
state cannot receive certification in another state. Even for more commonly performed analyses,
there are numerous commercial laboratories that are certified in multiple states. Given the
regulatory requirement for radon analyses, one can expect that those laboratories with the
capability for radon analysis will pursue certifications in as many states as practical.
The National Environmental Laboratory Accreditation Conference (NEL AC) is also
evaluating the issues surrounding privatization of the SDWA PE program through its proficiency
testing committee. NELAC serves as a national standard-setting body for environmental
laboratory accreditation, and includes members from both state and Federal regulatory and non-
regulatory programs.
The short holding time for radon, 4 days in Method 7500-Rn, presents a concern relative
to the practical availability as well. The 4-day holding time was also the focus of a number of
Methods, Occurrence, and Monitoring Document for Radon
2-18
-------
comments that EPA received in response to the 1991 proposed rule. Many commenters stated
that if a local laboratory is not available, the only alternative would be to send the samples by
overnight delivery to a laboratory elsewhere. Again, this situation is not unique to the analysis of
radon. Several large commercial laboratories already account for a sizable share of the market
for SDWA analyses for non-radon parameters, including organics, for which the holding times
are often 7 days. Given that a day would be required for shipping the samples, only three days
would remain for the laboratory to perform the radon analysis (the day on which the sample is
collected being "day zero"). Some commenters argued that for a large commercial laboratory
serving the water utilities, this short holding time will make it difficult if not impossible to
perform the necessary analyses within the holding time. However, through common-sense
scheduling efforts between the utility and the laboratory, such as not collecting samples on
Thursdays and Fridays, the holding time issue should be able to be accommodated with relative
ease. At worst, some laboratories may choose to offer analytical services over the weekend,
perhaps at an increased cost.
For the vast majority of other analytes for which EPA has established formal holding
times in its various regulatory programs, the holding times are specified in "days." This is
typically understood to mean "calendar days" with the day of sample collection being "day zero."
Because of the relatively short half-life of radon, the holding time is expected to be proposed as 4
days, beginning at the time of collection. SAIC strongly urges EPA to publish this holding time
as "96 hours" instead of just "4 days," in an effort to reinforce how the holding time is to be
calculated.
2.7 Anticipated Unit Costs
As part of its 1991 proposal, EPA conducted a limited survey of laboratories providing
radon analyses. Four laboratories provided price information to EPA regarding the analysis of a
single SDWA compliance monitoring sample, employing liquid scintillation counting as the
analytical technique. The data from the 1991 survey are in Exhibit 2-4.
As part of the 1998 review of analytical methods for radon, SAIC contacted nine laboratories that
perform radiochemical analyses. Of those nine, seven perform radon analyses. The prices from
the those seven laboratories are shown in Exhibit 2-5. None of the laboratories contacted were
among those contacted by EPA in 1991, but to avoid any confusion, the arbitrary numbers
assigned to each laboratory begin where the 1991 numbers left off.
There was no clear correlation between the estimated price and the method cited by the
laboratory. One of the laboratories that provided an estimate of $40 per sample is certified by the
States of Maine and New Hampshire to perform radon analyses of drinking water using the E-
PERM electret device. The other laboratory that quoted a price of $40 employs liquid
scintillation counting. The 1998 range of prices brackets those collected by EPA in 1991.
Methods, Occurrence, and Monitoring Document for Radon
2-19
-------
Exhibit 2-4. 1991 Radon Cost Survey Data
Arbitrary Lab Number
1
2
3
4
Cost Estimate
$30
$44
$50
$75
Descriptive Statistics
Mean
Median
Std. Dev.
Range
Minimum
Maximum
$49.80
$47.00
$18.80
$45
$30
$75
Exhibit 2-5. 1998 Radon Cost Survey Data
Arbitrary Lab Number
5
6
7
8
9
10
11
Cost Estimate
$75
$50
$40
$75
$45
$55
$40
Descriptive Statistics
Mean
Median
Std. Dev.
Range
Minimum
Maximum
NA
$54.29
$50.00
$15.12
$35.00
$40.00
$75.00
NA
As noted above, one possible response to concerns about the effect of the short holding
time on laboratory capacity would be for some laboratories to offer analyses over the weekend.
The increased cost of such services would likely be due to increased labor costs, particularly if
overtime were paid to the analysts. Assuming a 1.5 multiplier for overtime (e.g., "time and a
half), the unit cost might rise to the range of $60 to $112 per sample, but only for those utilities
that could not arrange to sample at more convenient times.
2.8 Practical Performance and Analytical Uncertainties
The available information on the performance of the various methods is greatest for the
liquid scintillation procedure, SM 7500-Rn, and the Lucas Cell technique. The data from the
1992 EPA collaborative study cited earlier indicate excellent precision and accuracy for liquid
scintillation. The Lucas Cell technique yielded slightly less accurate and less precise results, but
Methods, Occurrence, and Monitoring Document for Radon
2-20
-------
still within the realm of performance that EPA has accepted for the measurement of other
contaminants. Performance data for the electret method are incomplete, with no clear evidence
of a collaborative study in drinking water. :
I
As with many environmental measurements, an overall 'evaluation of the effectiveness of
a monitoring method must also consider the practical aspects of collecting a representative
sample. The analysis of radon presents two specific challenges. First, like many organic
contaminants, radon is volatile, and some radon will come out of solution in a sample if exposed
to the atmosphere for long periods. Secondly, being a radioisotope, 222Rn undergoes radioactive
decay.
The volatility of radon can be addressed in a fashion similar to that for the organic
chemicals, namely careful sample collection techniques that minimize the disturbance of the
sample, and the use of containers that can be sealed tightly.
The conclusions of the 1992 collaborative study indicate that while all four sample
collection techniques examined in that study (displacement, immersion, catch, and grab
sampling) can provide equivalent results, displacement and immersion sampling are the preferred
approaches. Both can be accomplished with little or no specific expertise. Displacement
sampling involves attaching a filling tube attached to the water source, inserting the other end
into the sample container, and allowing the water to fill the container with no aeration until the
container overflows. The filling tube is withdrawn while still running, so that water constantly
overflows the container. The container is then quickly sealed with an appropriate cap (e g
PTFE-lined). ,
Immersion sampling is somewhat similar, in that a sample container is placed in the
bottom of a large container. The filling tube is then inserted into the sample container which is
then filled to overflowing with the water to be sampled. The sample container is removed from
the larger container with forceps and sealed. The use of immersion sampling further reduces the
chances of leaving headspace in the sample container, by allowing the filling tube to be
withdrawn while the sample container is still submerged in the larger container. However, as
noted in the 1992 study report, there was little difference between the results from both sampling
techniques. The sampling procedure developed by California that was described earlier in this
document is an immersion technique. The losses of radon due to sorption on cap liners and in air
bubbles that occur during transportation and storage appear to be minimal for this technique.
The radioactive decay of 222Rn presents some concerns because the half-life of this
isotope is approximately 3.82 days. However, even with this relatively short half life, it is both
possible and practical to calculate the concentration of 222Rn at the time of sampling with a high
degree of accuracy. Depending on the regulatory action level (MCL or other level) that is
specified, the sensitivity of the liquid scintillation method should be sufficient to be used for
compliance monitoring even if the sample is held for several days. Method 7500-Rn currently
specifies a 4-day holding time. For this analyte, sampling documentation must include the time
Methods, Occurrence, and Monitoring Document for Radon
2-21
-------
of sample collection, as well as the date. However, this documentation requirement does not
present any practical difficulty for this technique.
2.9 Degree To Which Each Method Meets EPA's Regulatory Needs
Of the six techniques for the measurement of radon that were evaluated in this report,
only two appear to meet all of EPA's needs relative to compliance monitoring. SM 7500-Rn and
the Lucas Cell technique can achieve reasonable standards for precision and accuracy, are readily
available, and have been subjected to collaborative testing.
The four other techniques lack collaborative testing data, which is a significant problem
in establishing methods for a nationwide compliance monitoring program such as the SDWA.
Of those four other techniques, the electret technique shows greatest promise, and should
collaborative data indicating acceptable performance in water matrices become available in the
future, EPA may wish to consider this technique at a later date.
The other three techniques, the delayed coincidence liquid scintillation counting system,.
the activated charcoal passive collector technique, and the degassing Lucas cell technique may
have some utility in screening samples or in field measurements. The activated charcoal
procedure requires a lengthy exposure to running water and provides an average radon
concentration over the entire sampling period. The extent to which such time-averaged
measurements might be employed in SDWA compliance monitoring is a policy decision that
goes beyond the scope of this evaluation.
In summary, the results of this most recent review of possible analytical techniques for
radon in drinking water has reached the same conclusions as that of the 1991 EPA proposal. The
liquid scintillation counting technique (SM 7500-Rn) is most able to support a SDWA
compliance monitoring program, supported by the possible use of the Lucas cell technique.
2.10 References
Banks, T, WMA, Feb. 21, 1990 Memorandum to G. Helms USEPA-OW on Laboratories
Conducting Analysis of Radon in Drinking Water.
Banks, T, WMA, Nov. 14, 1989, Memorandum to G. Helms, USEPA-OW regarding The Lucas
Cell Method of Testing for Radon in Water.
Budd, G, and Bentley, C., "Operational Evaluation of the EIC Method for Determining Radon In
Water Concentrations," 1993 International Radon Conference, Hosted by AARST.
Che Yang, I. "Sampling and Analysis of Dissolved 222Rn in Water by the De-emanation
Method," US Geological Survey Bulletin, 1991, pp. 227-230.
Methods, Occurrence, and Monitoring Document for Radon
2-22
-------
Colle, R. Kotrappa, P., and Hutchinson, J.M.R., "Calibration of Electret-Based Integral Radon
Monitors Using NIST Polyethylene-Encapsulated ^Ra/^Rn Emanation (PERE) Standards,"
Journal of Research of National Institute of Standards and Technology, 100: 629-639 (1995).
Deloatch, I., Mar. 27, 1991 Memorandum to G. Helms, regarding estimated cost of analyses for
radionuclides.
Jensen, Jane T, 1997 California Department of Health Services Environmental Laboratory
Accreditation Program (ELAP) September 3 letter to William Labiosa, EPA-OGWDW.
Jensen, Jane T, California Department of Health Services Environmental Laboratory
Accreditation Program (ELAP) Attachment to the September 3, 1997 letter to William Labiosa
of EPA-OGWDW. \
Kitto, M.E. et al. "Direct Comparison of Three Methods for the Detection of Radon in Well
Water". Health Physics. Vol. 70, No. 3, pp. 358 - 362. 1996.
Kotrappa, P. and Jester, W.A.,"Electret Ion Chamber Radon Monitors Measure Dissolved 222Rn
in Water," Health Physics, 64: 397-405 (1993). !
Mullan, A. and Wanty, RB ("A Comparison of Two Techniques for 222Rn Measurement in Water
Samples," US Geological Survey Bulletin, 1991, pp. 231-235).;
Pia, S.H. and P. B. Hahn, 1992, "Radiation Research and Methods Validation Annual Report,"
EMSL-LV.
Sabol J, et al., "Monitoring of 222Rn in Taiwanese Hot Springs Spa Waters Using a Modified
Electret Ion Chamber Method," Health Physics, Vol. 68, No. 1,1995, pp. 100-104).
Standard Method 7500-Rn, Standard Methods for the Examination of Water and Wastewater.
19th Edition Supplement. Clesceri, L., A. Eaton, A. Greenberg, and M. Franson, eds. American
Public Health Association, American Water Works Association, and Water Environment
Federation. Washington, DC. 1996.
Tai-Pow J, et al., "The Determination of Dissolved Radon in Water Supplies by the E-PERM
System (Electret lonization Chamber)," InternationalJournal of Radiation Applications and
Instrumentation, Part A, Vol. 43, No. 1-2, 1992, p 95-101. !
Theodorsson, "A New Method for Automatic Measurement of Low-Level Radon in Water".
Journal of Applied Radiation Isotopes". Vol. 47, No. 9/10, pp. i885-895. 1996.
USEPA /'Tritium in Water Performance Evaluation Study, A Statistical Evaluation of the
August 8, 1997 Data," EPA/600/R-97/097, September 1997.
Methods, Occurrence, and Monitoring Document for Radon 2-23
-------
USEPA. "Radiation Research and Methods Validation: Annual Report 1992". EPA 600/X-
93/030. April 1993. (incorrectly cited as "Pia and Hahn 1992" - should be "USEPA 1993")
Yoneda, M., et al. ("Quantitative Measurement of 222Rn in Water by the Activated Charcoal
Passive Collector Method: 1. The Effect of Water in a Collector," Journal of Hydrology, Vol.
155, No. 1-2,1994, pp. 199-223).
Methods, Occurrence, and Monitoring Document for Radon
2-24
-------
3.
SOURCES OF RADON IN GROUNDWATER
3.1 Natural Sources of Radon Groundwater Contamination
Radon is produced in rock, soil and water by the decay of naturally occurring radioactive
elements in minerals. This process transfers radon into air- or water-filled soil pore spaces by
alpha recoil or diffusion. Radon is then transported by air or water until it decays to its progeny
or reaches the atmosphere.
Radon is a member of the "uranium series"of radionuclides all the members of which are
derived from the decay of uranium-238. Each radioactive isotope spontaneously decays to emit a
radioactive particle, radiant energy, and forms "progeny" isotopes. This process continues until a
stable isotope of lead is formed. Radon has three naturally-occurring isotopes, radon-222 (Rn-
222), radon-220 and radon-219. Of the three radon isotopes, Rn-222 is the only one of
environmental concern, because the other isotopes have much shorter half lives which limit their
potential for causing human radiation exposure. Radon-222 decays into Polonium-318 with a
half-life of approximately 3.82 days by alpha emission. The uranium decay series is shown in
Exhibit 3-1.
EXHIBIT 3-1 Uranium Decay Series (Including Rn-222)
SOURCE
Uranium 238
Thallium 234
Palladium 234
Uranium 234
Thorium 230
Radium 226
Radon 222
Polonium 21 8
Lead 2 14
Bismuth 2 14
Polonium 2 14
Lead 210
Bismuth 210
Polonium 210
PRODUCTS
* Thallium 234 + a '
» Palladium 234 + P :
* Uranium 234 + p :
» Thorium 230 + a
» Radium 226 + a
» Radon 222 + a
> Polonium 218 + a
» Lead 214 + a :
» Bismuth 2 14 + P
> Polonium 2 14 + a
> Lead 210 + P
> Bismuth 210 + P
> Polonium 2 10 + P
> Lead 206 + a !
Lead 206
HALF-LIFE
4.46 X 109 years
24.1 days
1.17 minutes
2.45x 10s years
7.5 x 10" years
1622 years
3.825 days
3.11 minutes
26.8 minutes
19.9 minutes
1.6 x 10"4 minutes
22.3 years
5.01 days
138.4 days
Stable
Methods, Occurrence, and Monitoring Document for Radon
3-1
-------
3.1.1 Release and Transport Properties of Radon and Radium
On a microscopic scale, the release of radon into groundwater water is directly related
both to the concentration of radium in the host soil or rock, which determines the amount of
radon generated, and to the emissivity of the mineral (which determines the fraction or the
generated radon that is released from the particle in which it is generated). The physical
condition of the rock (particle size, pore structure) plays a large role in determining emissivity.
Because of the importance of these physical factors in determining radon release, there is often
no strong correlation between radium levels in rocks or soils and radon levels in adjacent
groundwater. The dominant radon route of release into interstitial water is diffusion
along microcrystalline fractures in the rock. However, in most cases (i.e., cases in which the
percolation velocity is greater than 10'5 cm/sec), the mass transport of radon in groundwater
water is governed more by advection than this diffusion (Hess,et al. 1985).
Radium-226 is the immediate radiologic precursor of radon-222. Radium can be released
to groundwater by three routes: the dissolution of aquifer solids; by direct recoil across the
liquid-solid boundary during its formation by radioactive decay of its parent; and by desorbtion.
In contrast to radon, radium has very low solubility in water and very low mobility in
groundwater. Also, radium does not exist as a gas, and vapor phase transport is therefore not
important. Thus, as discussed below the transport patterns of radium generally do not greatly
affect the transport of radon and radium concentrations in groundwater can be a poor predictor of
radon levels.
3.1.2 Factors Affecting Distribution of Radon in Groundwater
The levels of radon in groundwater in specific areas or types of systems are affected by a
number of factors. Geologic regime and geok>gical parameters are strongly associated with radon
levels in groundwater. A number of studies have examined the correlations among radon levels
in groundwater and the occurrence of other elements, aquifer lithology, and the depth to the
groundwater. Analysis has suggested, that for a defined geographic area, relative radon levels
can be inferred from the dominant aquifer lithology and implied activity levels of the parent
isotopes. Loomis (1985) has identified six geologic and hydrologic variables that together can
be used to predict radon activity in groundwater at a regional level. Each variable, except
meteorology, tends to be strongly correlated with lithology type.
• Uranium-radium geochemistry. As noted above aquifer minerals with high uranium or
radium content may exhibit a relatively high rate of radon release.
• Physical properties of source rocks. The escape of radon from rocks into water varies
according to the rock's grain size, degree of weathering, micro fractures, and the
distribution of radon's parent nuclides within the rock's mineral grains. Generally, the
smaller the grain size and more pervasive the fracturing and weathering, the greater the
amount of radon that escapes.
Methods, Occurrence, and Monitoring Document for Radon 3-2
-------
• Dissolved radium. The relationship between dissolved radium in the water and radon in
water is inconclusive; several studies indicate there is little to no correlation in the co-
occurrence of these two nuclides.
• Aquifer properties. The transfer of radon from rocks to the aquifer is largely determined
by the flow characteristics of water through the aquifer. The transfer of radon from rocks
to water is enhanced when the rocks are relatively permeable, weathered, and fractured
and flow rates are relatively high. Given radon's relatively short half-life, groundwater
flow must be relatively rapid for radon to reach water supply wells before it decays
• Meteor ologic factors. Some studies have indicated that radon levels co-vary positively
with precipitation. Moreover, there is some evidence that radon emanation from the
rocks and soils is related to barometric pressure. Several studies that have looked for a
relationship between radon in water and meteorologic factors have found none.
• Well and water system design and use. Several studies have reported that radon levels in
water are inversely proportional to a groundwater system's number of customers and
yield. Reasons for this consistently-seen relationship ate not clear, although it may be
that wells serving smaller numbers of customers may draw from less productive granitic
aquifers with higher levels of radon precursor elements.
The aquifers with the highest radon concentration have: a lithology profile that is
dominated by granite and granite alluvia. These rocks tend to have higher levels of uranium and
a physical structure that facilitates the release of radon into adjacent water. Radon levels are also
often elevated near volcanic ash layers. Lower radon levels are found in basalts and sand
aquifers. This relationship between lithology and radon concentration is illustrated by the
regional differences in radon levels in groundwater between the southern Mississippi valley (a
predominance of basalts and sand results in low radon levels) and Appalachian uplands (a
predominance of granite results in high radon levels).
3.1.3 Large-Scale Geographic Patterns of Radon Occurrence in Groundwater
As noted above, groundwater radon levels in the United States have been found to be the
highest in New England and the Appalachian uplands of the middle Atlantic and southeastern
states. There are also isolated areas in the Rocky Mountains, California, Texas, and the upper
Midwest where radon levels in groundwater tend to be higher than the U.S. average. The lowest
groundwater radon levels tend to be found in the Mississippi valley, lower Midwest, and plains
states. However, even in areas with generally very high or low levels of radon in groundwater,
local differences in geology strongly affect observed radon levels (e.g., not all groundwater radon
levels in New England are high; not all radon levels in the Gulf Coast region are low). For
example, the presence of faults and shear zones in a geographic area characterized by low radon
levels can produce localized areas of high radon levels (Gunderson, et al. 1992). It was found
Methods, Occurrence, and Monitoring Document for Radon
3-3
-------
that radon levels in groundwater were correlated with measured radioactivity of rocks and soils
in the area, the prevalence of rock types know to produce radon in the area, and the area's soil
permeability. The general pattern of groundwater radon occurrence across the US is shown in
Exhibit 3-2. Data related to geographical patterns or radon occurrence are discussed in more
detail in Chapter 5. The potential for radon to co-occur with other pollutants is discussed in
Chapter 7.
3.2 Anthropogenic Sources of Radon Contamination in Groundwater
Radon in the environment is derived primarily from natural sources. Because of its short
half life, there are relatively few anthropogenic sources of groundwater radon contamination.
The most common manmade sources of radon groundwater contamination are wastes from
phosphate or uranium mining or milling operations and from thorium or radium processing.
These sources can results in high groundwater levels in very limited areas if, for instance, homes
are located on soil contaminated with such wastes or tailings, or if a contaminated aquifer is used
as a source of potable water (EPA 1999a). Otherwise, significant groundwater transport of radon,
is limited by its short half-life.
3.3 Distribution System Sources
3.3.1 Radon Sources in Distribution Systems
Radon levels-in distribution systems are usually lower in distribution systems than in
source water because radioactive decay and water treatments involving storage, aeration, or
carbon filtration act to reduce radon levels. As will be discussed in more detail in Section 5.2,
this is not always the case, however. In a number of systems in Iowa, for example, radon levels
in finished water samples were found to be substantially higher than those from the wells
supplying the systems. Detailed studies have shown elevated levels of radium in pipe scale in
these systems. The decay of the radium increases radon levels over and above those already
present in the influent water. The greater the length of old, scaled pipe through which the water
passes, the greater the radon levels. The extent to which this is a general phenomenon is not
known, but it suggests that care should be taken in estimating radon exposures on the basis of
wellhead or point-entry-samples where iron-manganese scaling is likely to be a problem.
3.3.2 Radon Sources in Households
Except to the extent that pipe scale in residences sequesters radium, there are no radon
sources that increase the levels of radon after water enters the household. Radon is released to
indoor air during domestic water use, however, as discussed in Section 4.3.
Methods, Occurrence, and Monitoring Document for Radon
3-4
-------
3.4 Non-Water Supply Sources of Radon Exposures
It has been estimated that only between 1 to 3% of the total residential radon exposures
results from radon in public water supplies (NRC 1998). The most important source of radon
exposure (accounting for approximately 95 percent of exposures) is indoor air contaminated by
radon released from rocks and soils and infiltrating into basements and living spaces. Other
sources of radon exposures include ambient (outdoor) air, fuel gas, and construction material
(primarily gypsum board).
3.5 References Cited
EPA (1999a), Criteria Document for Radon in Drinking Water, Office of Research and
Development, Office of Science and Technology.
Gundersen, L.C.S., et al. (1992) "Geology of Radon in the United States. Geological Society of
America" Special Paper 271.
Hess, et. al. (1985), " The Occurrence of Radioactivity in Public Water Supplies in the United
States. Health Physics." Volume 48, Number 5. (pp. 553-586), May.
Loomis, Dana, (1987) "Radon-222 Concentration and Aquifer Lithology in North Carolina.
Ground Water Monitoring Review." Volume 7, Number 2, (pp. 33-39).
National Academy of Sciences, (1998) Risk Assessment of Radon in Drinking Water, National
Research Council, Committee on Risk Assessment for Radon in Drinking Water, September 14.
Methods, Occurrence, and Monitoring Document for Radon
3-5
-------
-------
4.
FATE AND TRANSPORT
4.1 Physical and Chemical Properties of Radon arid Progeny
Radon, atomic number 86, is a "noble" and chemically inert gas. It does not react with
other elements in the environment. Radon is soluble in water, but also very volatile. It has a
high Henry's Law Constant (>X10~3 m3/!) , indicating a high potential to volatilize from water
solution. Its melting point is -71 °C and its boiling point is -61.8°C. It's solubility in water is
230 cm3/literat20°C. Radon is adsorbed by activated carbon, and therefore presumably to some
extent to other organic matter, although radon partitioning to organic matter in the environment
has not been extensively studied.
As noted in Chapter 3, radon-222 has a half-life of 3.82 days. Radon's progeny
radionuclides (primarily isotopes of lead, polonium, and bismuth) unlike radon, are not gases,
and are less soluble in water than radon. When radon undergoes radioactive decay in water, the
resultant nuclides tend to precipitate out onto suspended particulates or other surfaces. Similarly,
radon progeny in air "plate out" onto airborne particles, and the bulk of radon-related radiation
exposures through the inhalation pathway are often due to the deposition of progeny-bearing
particulates in the respiratory tract.
4.2 Relationship of Fate and Transport Properties to Human Exposures and Intake
-Radon's chemical and physical properties, particularly its radioactive half-life and
volatility, greatly effect its behavior in the environment, and human exposures from domestic
water use.
Because of its short radioactive half-life, the distance over which radon can move in
groundwater is severely limited. In just under four days, the activity of radon will be reduced
about 50 percent, and it will be reduced another 50 percent in the following four days, etc. In an
aquifer where typical horizontal flow velocities are on the order of 10-100 cm/day, this limits the
distance over which radon can be transported and still cause significant exposure to a few meters
or less. In bedrock aquifers, where water flow may be primarily through fractures, this distance
might be larger. As noted in Section 4.1, when radon decays in water, the resulting progeny are
much less soluble and mobile, and do not result in appreciable exposures.
Another consequence of radon's short half-life is that radon levels are reduced when
water is stored for any appreciable time prior to use. Thus, water systems which use storage
devices such as water towers, tanks or reservoirs, are already reducing radon levels in water. The
amount of reduction achieved depends on the average residence time in the storage device, and
whether the storage vessel is open to the atmosphere (see below)
When radon is released to surface water, its high volatility results in rapid release to the
atmosphere. Radon levels in surface water bodies are almost always below measurable levels
Methods, Occurrence, and Monitoring Document
4-1
-------
(NAS, 1998). Systems that store water in contact with the atmosphere therefore achieve radon
reduction both through radioactive decay and volatilization.
4.3. Exposures to Radon in Indoor Air After Release During Domestic Water Use
When water is heated or agitated during domestic use, radon is rapidly released to the air.
NAS (1998) estimates that between 80 and 100 percent of the'radon in tap water remains in
solution to be ingested if the water is consumed immediately and is not heated. Between 60 and
80 percent of dissolved radon is released from water from showers, sinks, and washing machines.
If water is heated to boiling (e.g., during cooking), essentially all of the radon is driven off.
The radon level in indoor air resulting from domestic water use is often estimated using a
transfer factor (TF) approach. This transfer factor is defined as the average increase in long-
term radon in air (pCi/LJ due to a long-term increase of one pCi/Lw radon in water. The value of
the transfer factor depends on three factors:
• Patterns of household water use (amount, timing, duration, agitation, and
temperature);
• Volume and air exchange rate of the room in which the water is being used; and
• Volume and air exchange rate of the entire house.
Measured Transfer factors in typical American houses generally fall between 1:1,000 and
1:100,000, with the mean being between 1:10,000 and 1: 15,000. That is, the domestic water
supply entering a house on average needs to have a radon level of approximately 10,000 pCi/1 to
increase the average indoor air level by 1.0 pCi/1. This value is estimated based on modeling
studies, validated by some of the measurements described above.
More refined models are available for predicting radon levels as a function of water usage
and building design parameters (e.g., "the three-compartment model"). Generally, it has been
found that, while these models provide additional insights into short-term peak exposures in
specific areas of the home (for, example, in the shower), they provide little improvement in the
quality of long-term estimates of inhalation exposures compared to the simpler transfer factor
approach.
4.4 Relationship of Fate and Transport Properties to Radon Behavior in Treatment and
Distribution Systems
As noted above, radon undergoes spontaneous radioactive decay during storage and
residence in distribution systems. Thus, radon levels in distribution systems and at the point of
use are usually lower than in the source water (but see below). In addition, radon's chemical and
physical properties mean that some technologies that are used to remove other contaminants also
Methods, Occurrence, and Monitoring Document
4-2
-------
result in reduced levels of radon. These properties have also been used to design treatment
technologies specifically for removing radon from domestic water. Because radon is an inert
gas, processes which involve chemical treatment of water (e.g., chlorination, iron/manganese
sequestration, chemical coagulation) do not effect radon levels unless they cause it to volatilize
or be removed bound to solids.
4.4.1 Aeration Technologies
Aeration technologies make use of radon's volatility to reduce radon levels in treated
water. In the Proposed Rule, (EPA 1999), high-performance aeration has been selected as the
Best Available Technology (BAT) for radon removal. The specific technologies which have
been identified include packed-tower aeration, multi-stage bubble aeration, and shallow tray
aeration. In addition, there are other aeration technologies that can also cost-effectively achieve
radon reduction in commercial-scale use. All the technologies identified above are capable,
under defined operating conditions, of achieving at least 99.9 percent radon removal from
influent water. Capital and operating costs can be lower if lower removal efficiencies are
required (EPA 19?9b). .
A significant proportion of community groundwater systems already employ aeration
technologies to remove odors or organic chemicals, or as an adjunct iron/manganese removal.
EPA estimates (1999b) that between approximately 16 and 24 percent of groundwater systems
serving 1,000 or more customers currently employ some form of aeration treatment. A smaller
proportion of smaller systems also employ aeration technologies. EPA estimates that these
existing technologies are likely to achieve a 90 percent reduction in radon levels in the majority
of cases.
4.4.2 Granular Activated Carbofl Treatment ;
As noted above, radon also can be adsorbed onto granular activated carbon (GAG). EPA
has indicated (1999b) that GAC technologies, while not BAT for most systems, may be
appropriate for some very small systems where the capital costs of aeration technologies are
prohibitive. Both point-of-entry (POE) and point-of-use (POU) GAC technologies can achieve
up to 99 percent radon removal under certain conditions. However, the amount of carbon and
contact time required to achieve high radon removal efficiencies are considerably greater than
those required to achieve efficient removal of organic chemicals. Thus, at a minimum, changes
in operating conditions would be required to adapt existing GAC systems (which EPA estimates
to be present at about two percent of all small and very small systems) to address radon
contamination.
4.4.3 Radon Release from Pipe Scale
As discussed in Section 3.3, there is evidence that radon can be released from pipe scale
pipes in distribution systems. The best information regarding this phenomenon comes from
Methods, Occurrence, and Monitoring Document 4-3
-------
studies of radon distributions in ground-water systems from Iowa. Information was provided
concerning raw and finished water radon analyses from 150 water systems across the state, from
systems of different sizes (Kelley and Mehrhoff, 1993). The geometric mean radon level in the
raw data samples was 284 pCi/1. As expected, the geometric mean value of water radon levels
from the finished water was lower,' at 176 pCi/1. However, the ratio of the radon levels in the
finished water to the raw water varied considerably. In a substantial proportion of the cases
(Exhibit 4-1), the radon level in the finished water exceeded that from the raw water, by up to
six-fold.
Exhibit 4-1. Ratios of Finished/Raw Radon Levels in 150
Iowa Water Systems
Ratio Finished/Raw Radon
Levels
Less than 1 .0
1.0-1.5
1.5-2.0
2-5
>5
Number of Systems
107
29
6
7
1
Radon levels that were higher in finished water than in raw water occurred with varying
frequency across the types of geological formations. When water was drawn from alluvial
aquifers, finished water levels increased over the wellhead levels only five percent of the time
(3/60 systems). In contrast, this phenomenon was seen in 41 percent (9/22) of the wells finished
in Cambrian/ Ordovician and 40 percent (2/5) of wells finished in Cambrian/ Precambrian units.
Although no specific geochemical data were provided for the systems where the increases
in radon occurred after entry into the systems, the basis for this phenomenon has been previously
described in studies of several of the systems included in the Iowa data (Field et al. 1994, Fisher
et al. 1998). The increases in radon in the distribution system appear to occur as a result of the
accumulation of iron pipe scale in the distribution systems. The scale sequesters radium, and the
resultant buildup of radium results in the releases of radon into the water as it passes through the
system. The ultimate outcome may be in-system radon levels that substantially exceed the levels
seen in the aquifers from which the water is drawn.
There is little evidence concerning the frequency or severity of this phenomenon outside
of Iowa, although there is no reason to think it would not occur wherever the geochemical
conditions are similar. There would a lower likelihood of scaling and radon buildup in systems
drawing from alluvial aquifers, and more potential for problems whenever iron levels are high
Methods, Occurrence, and Monitoring Document
4-4
-------
and eH levels low in the producing aquifer. Systems treating water to reduce iron and
manganese might expect that radon levels would also be reduced in distribution systems.
4.5 References
EPA (1999a) National Primary Drinking Water Regulations; Radon-222, Proposed Rule,
Federal , (Date?)
EPA (1999b), Regulatory Impact Analysis and Revised Health Risk Reduction and Cost Analysis
for Radon in Drinking Water, Office of Groundwater and Drinking Water, August.
Field, R.W., E.L. Fisher, R.L. Valentine, B.C. Kross, (1994), "Radium-Bearing Pipe Scale
Deposits: Implications for National Waterborne Radon Sampling Methods", American Journal of
Public Health, April, pp. 567-570;
Fisher, E.L., L. J. Fuortes, J. Ledolter, D.J. Steck, R.W. Field, (1998), "Temporal and Spatial
Variation of Waterborne Point-of-Use radon in Three Water Distribution Systems", Health
Physics, Vol. 74, No. 2, February , pp. 242-248.
Kelley, R., and M. Mehrhoff, (1993), Radon-222 in the Source and Finished Water of Selected
Public Water Supplies in Iowa, Research Report Number 93-1, University of Iowa Hygienic
Laboratory, January 13.
National Academy of Sciences (1998), Risk Assessment for Radon in Drinking Water,
Committee on Risk Assessment for Exposure to Radon in Drinking Water, National Research
Council, September.
Methods, Occurrence, and Monitoring Document
4.-5
-------
-------
5. DISTRIBUTION OF RADON IN GROUND WATER SOURCES
This chapter of discusses the available information related to the distribution of radon in
groundwater sources in the United States and the numbers and proportions of groundwater
systems with sources that could exceed potential regulatory levels. The following sections discuss
the nature and quality of the available data related to radon distribution, the methods used to
analyze and characterize the distributions of radon from the various data sources, the assumptions
used to develop nationwide estimates of the occurrence of radon in groundwater sources. In
addition, estimates of the proportions of systems that could exceed regulatory levels are
presented, and the potential uncertainty associated with these estimates are discussed. The
material in this chapter comes primarily from the Re-Evaluation of Radon Occurrence in
Groundwater Supplies in the United States: External Review Draft (ICF, 1998). External peer
reviewer comments on that document have been incorporated into the chapter.
5.1 Data Availability and Quality
This section begins with a chronological review of EPA's past efforts to develop data
related to radon occurrence in groundwater. Relevant literature sources, including the Agency's
previous radon occurrence analyses, are cited in the text. We then discuss the data sources that
have been identified since EPA's previous rulemaking effort ended in 1993.
5.1.1 Previous EPA Data Gathering Efforts Related to Radon Occurrence in
Groundwater Supplies
In 1978, EPA's Eastern Environmental Radiation Facility began a pilot study to determine
the need for a nationwide study of radon in drinking water, to demonstrate the feasibility of such a
study; and to develop a limited national database of radon levels in drinking water. In this pilot
study, approximately 6,298 samples of raw and finished water were collected from private and
public surface and groundwater supplies (Horton, 1983). Samples were collected by state or local
personnel from water supplies serving more than 1,000 people in 40 states. Sampling or Quality
Assurance/Quality Control (QA/QC) procedures were not defined in the study protocol, but were
left up to the states. Data collected during the pilot study were analyzed to estimate
representative radon levels in water in each state. Geometric mean radon levels were calculated
for each state represented and for the entire U.S.1 Analysis of these data lead to the following
conclusions:
• radon levels in surface water are very low (geometric mean <2 pCi/1);
• radon levels in groundwater are highest in areas where water is drawn from
granitic aquifers;
1 As will be discussed in Section 5.2, the distributions of radon levels in the various states
and at the national levels were positively skewed, and the investigators thought that the geometric
mean levels provided a better measure for comparison purposes than the arithmetic mean.
Methods, Occurrence, and Monitoring Document for Radon 5-1
-------
higher levels occur in small systems than in larger ones; and
• higher levels of radon are observed in private wells than in public wells.
The pilot study findings were used to help plan a larger-scale survey in support of the
Nationwide Occurrence of Radon and Other Natural Radioactivity in Public Water Supplies
project (USEPA 1985a). This project, which began in November of 1980, was the first attempt
to systematically sample water supplies in the US to characterize the distribution of radioactive
contaminants. The objective of the study was to collect samples that were representative of actual
exposures and thus only finished water was sampled.
More than 2,500 samples were collected. Although the study design called for samples
from all 50 states, only 35 provided data.2 In addition, only public water supplies serving at least
1,000 people were sampled. Thus groundwater supplies representative of the large majority of
the groundwater systems in the US, many of which were likely to have potentially elevated radon
levels (Hess 1985), were not sampled. The population-weighted arithmetic mean radon levels
calculated for each state and for consumers of groundwater in the US as a whole are presented in
the left-hand column of Exhibit 5-1.
Exhibit 5-1. Radon Levels in Public Groundwater Systems From Two Surveys
State
Alabama
Alaska
Arizona
Arkansas
California
Colorado
Connecticut
Delaware
Florida
Georgia
Hawaii
Idaho
Illinois
EPA Eastern Environmental
Radiation Facility Study
(USEPA, 1985)1
Population- Weighted Arithmetic
Mean, pCi/1
160.7
N/A
329.3
N/A
N/A
380.7
N/A
126.4
148.5
147.4
N/A
256.6
167.6
EPA National Inorganics and
Radionuclides Survey (NIRS)
(Longtin, 1987)2
Population- Weighted Arithmetic
Mean pCi/1
420.1
128.5
1,435.1
100.0
228.4
329.9
1,208.9
123.3
127.3
563.4
N/A
437.4
193.2
2 The states not submitting data were Alaska, Arkansas, California, Connecticut, Hawaii,
Iowa, Louisiana, Maryland, Michigan, Missouri, Nebraska, New Jersey, Texas, Washington, and
West Virginia.
Methods, Occurrence, and Monitoring Document for Radon
5-2
-------
State
Indiana
Iowa
Kansas
Kentucky
Louisiana
Maine
Maryland
Massachusetts
Michigan
Minnesota
Mississippi
Missouri
Montana
Nebraska
Nevada
New Hampshire
New Jersey
New Mexico
New York
North Carolina
North Dakota
Ohio
Oklahoma
Oregon
Pennsylvania
Rhode Island
South Carolina
South Dakota
Tennessee
Texas
Utah
Vermont
Virginia
Washington
West Virginia
Wisconsin
Wyoming
U.S.
EPA Eastern Environmental
Radiation Facility Study
(USEPA, 1985)1
Population- Weighted Arithmetic
Mean, pCi/1
105.4
N/A
106.0
107.8
N/A
N/A
N/A
769.4
N/A
210.3
82.0
N/A
328.6
N/A '
550.8
1,183.6
N/A
178.1
132.1
278.6
148.8
169.8
160.0
264.0
719.9
1,511.1
276.9
289.2
23.8
N/A
360.9
656.8
447.8
N/A
N/A
234.4
415.3
232.1
EPA National Inorganics and
Radionuciides Survey (NIRS)
(Longtin, 1987)2
Population- Weighted Arithmetic
Mean pCi/1
! 187.4
136.4
: 369.1
205.5
: 108.2
1,228.4
266.1
587.8
185.2
388.7
104.3
143.7
344.6
351.6
743.2
2,673.5
137.1
: 309.1
: 223.7
; 2,277.7
114.0
175.2
158.0
118.2
: 507.8
1,170.0
557.7 .
281.6
113.7
150.5
226.8
997.1
485.4
432.5
263.6
367.2
558.0
249.0
Sources of Data:
1. Nationwide Occurrence of Radon and Other Natural radioactivity in Public Water Supplies, USEPA-
520-5-85-008
2. Longtin, J.P., "Occurrence of Radon, Radium, and Uranium in Groundwater", Journal of the American
Water Works Association, July, 1987, pp. 84-93.
Methods, Occurrence, and Monitoring Document for Radon
5-3
-------
In the mid-1980s, the need for additional representative and systematically collected
occurrence information on radon and other radionuclides, especially in small systems, motivated
EPA to undertake the National Inorganics and Radionuclides Survey (NIRS). The objective of
NIRS was to characterize the occurrence of radon and other constituents present in community
groundwater supplies in the U.S., its territories, and its possessions. The sampling program was
carefully stratified to reflect the national distribution of community groundwater supplies by size,
as measured in population served. The inventory of the community groundwater supplies
contained the 1983 Federal Reporting Data Systems (FRDS) database was used to select the sites
to be sampled. Non-community systems were not sampled . At the time of the survey, there were
approximately 47,000 community groundwater supplies in the database, from which 1,000 sites,
stratified into the four size categories on the basis of population served, were selected.3 Within
the size categories, an effort was made to develop samples that were geographically
representative of the U.S. The relatively small numbers of medium and large systems in the
sample, however, limited the extent to which this could be accomplished for these size categories.
The four population categories and the number of sites sampled in each category are presented in
Exhibit 5-2. The NIRS radon sampling effort was conducted between July 1, 1984 and October
31, 1986.
Exhibit 5-2 Groundwater System Size Categories and
Selected Sites in the Design of NIRS
Community Water
Supply System
Size
very small
small
medium
large/very large
System
Size (Population
Served)
25-500
501-3,300
3,301-10,000
10,001->100,000
Number of
FRDS Sites*
34,040
10,155
2,278
1,227
Number of Sites
Selected for
NIRS
716
211
47
26
* Based on the FRDS inventory for Fiscal Year 1985.
Of the 1,000 sites selected in the study design, 990 were actually sampled. Although
sampling occurred over a two-year period, only one sample was taken from each water system,
and the MRS data therefore provide a cross-sectional "snap-shot" of radon levels in drinking
water supplies. The locations of these sampling sites are shown in Exhibit 5-3. Samples were
collected from points in the distribution systems selected to as to represent typical radon
exposures. Unlike previous studies, a rigorous QA/QC program was built into NIRS and all
radon analyses were conducted in EPA laboratories. The QA/QC program included field
duplicates, field blanks, split laboratory samples, blind laboratory standards, and laboratory
spikes. Summary results of the NIRS were published in 1987 (Longtin, 1987). The results are
presented in the right-hand column of Exhibit 5-1 and the results are mapped in Exhibit 5-3.
3 The size of the sample was limited to 1,000 because of constraints imposed by available
resources and impending deadlines in the regulatory process.
Methods, Occurrence, and Monitoring Document for Radon
5-4
-------
0>
•*-
I
I
1
O
e
I
I
1
i
o
O
«'
2
a
S
OS
S
-------
The extensive QA/QC program, the large number of samples collected, and the extensive
geographical coverage across the nation (at least of the smaller systems) make the MRS database
the most representative and reliable source of national occurrence data for radon in public water
supplies. EPA has relied exclusively on the NIRS data in its previous analyses of the occurrence
of radon in groundwater supplies and to predict the proportions of groundwater systems that
would exceed possible drinking water standards (Wade Miller 1990, 1993).
In response to the 1991 Proposed Rule, which included EPA's findings on radon
occurrence, the Agency received a number of comments from stakeholders concerning the use of
the NIRS as sole source of radon data for the occurrence analysis. The major concerns included:
• NIRS data represent water quality "at the tap," whereas the proposed rule would
require systems to mitigate radon based on wellhead or point-of-entry levels.
Since levels at the tap are generally lower than levels in source water, according to
the commenters, this could underestimate the numbers of systems affected by
radon regulations;
• NIRS samples were taken after blending of water from different sources in some
distribution systems, further reducing radon levels and obscuring the variability
among different sources ; and
• The numbers of systems sampled in the NIRS are very limited for some states,
leading to a potential underestimation of non-compliance in those states. Also,
radon occurrence in medium and large systems are poorly characterized.
In addition, commenters have raised a number of issues concerning how the cross-
sectional nature of the NIRS data further limit its utility in estimating potential exceedences of
regulatory levels. The limitations include (according to commenters):
• Using the NIRS data directly to estimate regulatory exceedences ignores the potential
uncertainty in the NIRS radon levels introduced by sampling and analytical error. This
uncertainty could result in underestimation of the numbers of systems exceeding
regulatory levels;
• Additional uncertainty is introduced by failing to take into account variations in radon
levels over time in individual wells and systems, which could also increase the proportion
of systems exceeding regulatory levels above that predicted using the cross-sectional
NIRS data.
Finally, as noted above, the NIRS did not attempt to characterize the occurrence of radon
in non-transient non-community water systems (NTNCWS) that could be affected by radon
drinking water standards. These systems, which serve schools., hospitals businesses, commercial
and industrial buildings, and other institutions, could also be significantly affected by the
regulation of radon in drinking water.
Methods, Occurrence, and Monitoring Document for Radon
5-6
-------
All of these considerations influenced EPA to seek additional sources of data related to
radon occurrence in groundwater supplies. Therefore, in 1988, EPA began an effort to update its
database on radon occurrence and to revise its estimates of radon occurrence in groundwater.
5.1.2 Data Gathering Efforts in Support of the Revised Occurrence Analysis
The data gathering effort related to radon occurrence was closely coordinated with EPA's
outreach efforts in support of the development of the 1999 Proposed Rule for radon in drinking
water. These efforts included three Stakeholders Meeting held in Boston, San Francisco, and
Washington, DC in 1998 and 1999. Representatives from many water utilities and state
regulatory agencies took part in the Stakeholders Meetings.
The first major source of data was EPA's files of comments and data submitted in
response to the 1991 Proposal and subsequent efforts to revise the rule. When EPA ceased
rulemaking efforts in 1993, many of these comments had not been fully evaluated or responded
to. These files provided data and information related to individual and groups were involved in
data gathering. Where significant data were provided, the authors of the studies were contacted
to ascertain whether they could provide the raw data in electronic form. In addition, attendees at
the Agency's Stakeholders' Meetings who stated that they had radon occurrence data that they
were willing to share were contacted by telephone to facilitate the transfer of the data.
EPA also conducted a computer literature search of recent publications related to radon
occurrence, and some useful information was identified through the searches. The United States
Geological Survey representative to the Stakeholders Meetings also provided useful background
information and data on radon occurrence.
At first, few new data sets were received from stakeholders, and EPA therefore contacted
the American Society of Drinking Water Administrators (ASDWA) to ascertain whether any
members of this group might have radon occurrence data that they wished to submit. In response
to EPA's request, ASDWA conducted a survey of its members (water utilities and state water
regulators, primarily) asking whether they had any radon occurrence data and whether they would
be willing to provide it to the Agency in support of the revised occurrence analysis. Thirty-five
responses to the survey were received, the majority of which indicated that radon occurrence data
were available. ASDWA then followed up with the positive responders, and data were forwarded
to ICF (EPA's contractor) under the auspices of the Association. In addition to the data received
from ASDWA members, useful data related to radon occurrence was obtained from several
academic researchers.
5.1.3 Results of the Data Gathering Effort
As noted above, supplementary data were sought primarily to:
• Increase the coverage and representativeness of the data, across geographic regions and
system size strata;
Methods, Occurrence, and Monitoring Document for Radon
5-7
-------
• To support the investigation of systematic differences in radon levels between in-system
and well-head or point-of-entry levels; and
• To support an assessment of the magnitude and impacts of sampling, analytical, and
temporal variability on estimated radon distributions.
High priority was therefore given to obtaining data sets that provided information on
radon distributions in a state or region sparsely covered by the NIRS, that provided both in-
system and wellhead samples, and which provided the results of duplicate analyses or multiple
samples over time. In addition, data sets were sought that reported radon levels in large
groundwater systems and in non-transient non-community systems, which were poorly
represented (the former) or absent (the latter) from the NIRS.
To be usable, the data had to be of acceptable quality and traceable back to a specific
water system in a specific state and city. Because system size was a major stratification variable
(see section 5.2), the ability to identify systems (and estimate the populations served) was a key
element in the exploratory data analysis. Data had to come from community groundwater water
systems (systems providing residential water on a year-round basis to 25 or more individuals) or
from non-transient non-community systems (systems providing non-residential drinking water to
25 or more individuals exposed on a year-round or nearly year-round basis.) Data from
residential wells (wells serving individual homes), transient systems (systems serving seasonal
campgrounds, etc.), or monitoring wells were not included in the analysis. Characterizing the
types and sizes of systems reporting data from a given state often presented a major challenge in
the preliminary data analysis.
In addition, data sets were used only if it could be ascertained that the sampling and
analytical procedures used were similar or identical to those recommended by EPA, and if there
was evidence of an acceptable QA/QC program supporting the data gathering effort. In a few
instances, QA/QC plans and/or results from QA/QC replicate analyses were provided, but for
most data sets, only verbal descriptions of QA/AC procedures were provided by the individuals
supplying the data. Finally, for a data set to be included in the analyses described in the analysis,
there had to be a sufficient amount of data to allow meaningful statistical analysis. If only a few
sampling results were provided, then a data set was given low priority or excluded from the data
analysis.
Methods, Occurrence, and Monitoring Document for Radon
5-8
-------
The data gathering efforts resulted in the identification of 20 significant supplementary
data sources related to radon occurrence4. The selected sources are identified in Exhibit 5-4,
along with the general types of data they provide. The sources fall into two general categories.
The first category includes sixteen data sets that have been assembled by state regulatory
agencies for the purpose of characterizing the state-wide distribution of radon in groundwater
supplies. Most of these data sets greatly supplement the NIRS data in terms of the total numbers
of systems sampled, the numbers of systems sampled in each size category, and the geographic
representativeness of radon data. In addition, some of the data sets provided duplicate samples,
which could be used to evaluate sampling and analytical variability, sampling results from more
than one well in the same systems, and samples taken over time from single wells. These data
could be used to characterize intra-system and temporal variability, as discussed in Section 5.7
Some of these sources (e.g, the data on radon distribution in Maine groundwater systems
provided by Hess, et al, and the data from the New York Statewide Surveillance Study) were
available when the previous occurrence analyses were performed; but were not formally evaluated
by the Agency. Data from most of the states, however, has been collected more recently.
The second general type of source that we found report data specifically related to
sampling, analytical, and/or temporal variability in radon levels, but that do not characterize
geographical variability in radon levels. Data sources of this type include the duplicate samples
(samples taken sequentially and analyzed separately) from Alabama rural water systems, the QA
duplicate analytical results (duplicate analyses of the same sample) from the Southern California
Water Survey, the study of radon levels in Missoula, MT municipal wells, and the study of
temporal, sampling, and analytical variability in small-system wells in central North Carolina cited
in Exhibit 5-4.
The only source that provided both raw and finished water data came from the Iowa
Department of Natural Resources, which provided information on raw and finished water radon
levels from 150 community systems. Finally, we identified only six. data sources with significant
data related to radon levels in noncommunity non-transient systems.
In the following sections, data from the NIRS and supplemental sources are used to
evaluate radon occurrence in groundwater systems in the US, and to develop predictions of the
proportions of systems that may exceed radon regulatory levels. Data management and QA/QC
procedures are described in Appendix A. 1. More detailed information concerning the individual
supplemental data sets are provided in Appendix A.2
4 Approximately 10 additional source of radon occurrence data were identified but not
included in the analysis because they failed to meet one or another of the requirements in this
section. The data set for Missouri water systems was received top late to be used in the model for
estimating national radon occurrence, but summary statistics for the Missouri data are presented
in Section 5.2.
Methods, Occurrence, and Monitoring Document for Radon
5-9
-------
0
e.
>*
S
V.
3
Q
§
"3
a
CO
es
Exhibit 5-4. Summary of M;
"3 ;—
CS V 03
Q H >
M 2-5
*"* •" TS K
•3 S T5 "g "i
J? oi c3 5 »^
Q CA os •< >
^>
•s •*••
C c
a a § 2
0 s "a 1
2 a S a S -2
ea o o o 2 >_
O 2 U ^ H <»
e «>
o^B
03 WJ .Si
•^ «• »
5 ^ >>
Q i-4 CA
a
o
a S i -22
5 a 'o «
« 1 -i b S-
£ •§ .5 « E
•a S, S a o
Q CA pu W U
** "o a> a
a es -^ -^^ »s o
i « s-gjf
J ^ "2 2 S •=
n, 55 *: en « 'C
o .a
CA 2 tf O CA Q
.5
di ^
S^^
**»
CA Q
o
CU &£
2 2
O >
Q) o
O U
g
1
03
•*-*
a
0
en
O
O
O
O
ON
ON
i
tN
ON
ON
cd
ca
eg
/_v
Duplicate Samples, Alabama Rural
Water Systems (David Grey,
USEPA, NAERL, Montgomery, AL
en
>
en
CO
>
en
*—
^sg
s ~
t» U
Southern California Radon Survey
QA Duplicates (Metropolitan Water
District of Southern California)
o
o
Z
Z
£
o
S3
c^.
•»-•
g
1
c
c
o
U
Connecticut Community and
Noncommunity Systems (Nancy
McHone, CT Dept. Of
Environmental Protection)
o
Z
0
en
J
O
1
ON
ON
'
ON
ON
O
•s
2
L*
Idaho Community and
Noncommunity Systems (Janet
Crockett, Idaho Department of Wate
Resources)
o
o
0
en
-------
e
o
U
'
o
cc
A
Q
8
I
I
O
a
o
-a
£>
a
~£
a,
sc
o
o
i?
i
02
Tf
ITJ
•a
o
Z
• 0)
•a
<
Ov
OS
o>
c
'3
Maine (Charles T. Hess, University
of Maine, Orono)
"8
£
->
o
Z
o
Z
0
Z
0
Z
CO
CU
^
Os
ON
ON
OS
t3
i-
Maryland (Charles Poreda,
Maryland Department of the
Environment)
•o
£
J
o
o
Z
0
Z
0
Z
cn
>
CN
OS
OS
OS
i
00
I
Michigan (E. Brown, Michigan
Department of Public Health)
e«
>~
Qi
>•
"O
o
E
•J
CO
>
o
Z
o
Z
^.
OS
OS
en
OS
s
«
i=
c/a
Missoula, Montana (Robert Ward,
Mountain Water System, Missoula)
<8
>"
o
Z
w
>
en
^
r-
Os
oo
00
*
o
Z
8
0
CO
cu
o i
OS
OS
oo
OS
o
>-
%
z
New York Community Systems
(New York Statewide Surveillance
for Radon in Selected Community
Water Systems, 1990)
en
CD
>- ,
o
Z
o
z ,
0
Z
o
Z
VI
OS
OS
m
OS
OS
«<-l —
° 5P
~ "cu
:§ J"o
> 0 Z
North Carolina Temporal and
Sampling Analytical Variability
Study (William Drane, Department
of Environmental Sciences and
Engineering, University of North
Carolina)
o
Z
o
Z
o
Z
>
0
Z
V3
>
oo
00
Os
vo
00
OS
_o
6
Ohio Community Water Supplies
(Kathleen Pinto, Ohio EPA Division
of Drinking Water and
Groundwater)
u-i
e
1
1
i
o
a
I
1"
-------
0*23
° cL «
S S '=
03 O 03
Q H >
BD "3 .4?
c a .a .-=
§ a "-P .a
_ O. >» el
ffl *TS •!•*
**"J £3 ^3 2 M
O co « •< >
^•S"
"s B
§ I "1 1
2 B 1 i 1 IS
aO O O bt >.,
Jz; U Z H co
c g
0 « 1
03 WJ 4^
a s&
&
o
« >* -a
* IS b£
> '3 B O
O to CM Ed U
*^ "S fli C
g 5 o 2 .S .2
JSSrSl
Q* £§ "* J") -2 i
3 H "a> a> >L •££
oa 2 tf O co Q
be
**^
CL4 n
a -2
CQ ^
O
3 ^
0
en
o
>•
m
ON
i
ON
ON
ca
"g
J
S
c
(S
Pennsylvania Community Water
Systems (Jerry Rupert, PA
Department of Environmental
Protection)
o
o
o
en
O
£
s
>•
ON
ON
cs
•S ^
oo a
s
a>
h S
South Carolina Community Wat<
Systems (South Carolina Depart!
of Health and Environmental
Control)
0
o
-o
'!
en
>^
O
en
£
ON
ON
«
X
O
en
U
Texas Community and
Noncommunity Systems (Ron
Beardon, Texas Natural Resourc
Conservation Commission)
T3
1>
J
O
1
_'i
1
g
en
£
to
ON
ON
"•"*
ON
ON
C
1
Ic
CO
<~
•o 5
Washington State Community an
Noncommunity Systems (Jim
Hudson, Washington Departmen
Health)
13
1
g
1
en
£
o
en
£
NO
ON
ON
•"•*
ON
ON
~"
e
1
8
fw
i
c
C M*
Wisconsin (Mark Wilson, Wisco
Department of Natural Resource:
CO
CD
O
00
£
DO
O
DO
(U
>H
CO
ON
"C
o
CO
en
S
B
"en
Missouri Department of Natural
Resources (Kenneth Duzan, Divi
of Environmental Quality, Public
Drinking Water Program)
i
1
•*->
C
0>
en
1
*]j*
c
o
1
at
C
E
o
o
c °1
•s'l
§ w
'•is c
13 ""ts
1 £
en 5
•o S
— O
S >C
o -o1
~ 1)
CO -o
^ '>
O Q
? 1
•a «
o tS
E -3
"3 '5
-S §
•*s -2
S S
£ t!
'o
-------
5.2 Methods Used in the Data Analysis
This analysis builds on previous studies of the NIRS and other data sets to estimate the
distribution of radon activity in groundwater sources that are used to supply domestic and
commercial water in the U.S. A second objective of the analysis was to estimate the proportions
of systems that would exceed various radon levels, should they be proposed as regulatory limits
by the EPA. This effort naturally employed a wide range of analytical and statistical tools. This
section describes the approaches that were employed to evaluate the data and to answer specific
questions regarding radon distributions.
5.2.1 Statistical Analysis of Radon Distributions
The specific QA/QC and data management methods used in this analysis are described in
Appendix A. The following sections describe the statistical methods that were used in the
evaluation of radon data from the final data sets that had been subject to QA review.
a. Treatment of Censored Data
In most of the radon data sets, some proportion of the analytical results are censored, that
is, reported as being "less than" some specific value, reflecting an inability of the analytical
method used to measure the analyte in question at low levels. Depending upon the QA
procedures used during the data gathering, the censoring value may constitute the limit of
detection of the analytical method employed, or the lower limit at which analytes can be
quantified. In the NIRS, a Minimum Reporting Limit of 100 pCi/1 was employed. Values below
the censoring limit are referred to as "non-detects". Where a non-detect is reported, it is likely
that the analyte is present at some level greater than zero, but the problem remains as to how to
incorporate these analyses into the calculation of summary statistics (such as mean or standard
deviation).
If the proportion of censored observations is low and the censoring level is low compared
to the levels that are of concern, the presence of censored data will have little effect on the
calculation of summary statistics for the sampled population. However, if the censoring level is
near the level of concern, or if a large proportion of the data are icensored, the potential impact on
such calculations may become significant. In the NIRS data set, approximately 27 percent of the
data are reported as less than the MRL of 100 pCi/1. For larger system sizes, the proportion is
larger. While this has the potential to affect calculations of the representative activity levels (e.g.
the mean or geometric mean values), it has less impact on the utility of the NIRS data for
generating predictions of the proportions of systems above the censoring level, or regulatory
levels near the censoring level. In evaluating radon occurrence from the NIRS data, the censored
data were retained, and subject to the graphical and statistical analyses described below.
As noted in Appendix A. 1, the censoring limits could not be estimated accurately many of
the supplemental databases. Censoring levels were not provided with most of the supplementary
data sets, and it is likely that the quantitative limits varied over time and within the data sets.
Methods, Occurrence, and Monitoring Document for Radon
5-13
-------
Often, reported radon values were much lower than the N1RS MRL. In these case, we used
professional judgement to estimate censoring levels. Values that were reported as "<" were
provisionally included in their databases at one-half the reported "<" level. In addition, reported
values that seemed unreasonably low (generally less than 10-20 pCi/1, depending on the database)
were adjusted to one-half the estimated censoring level. Finally, if a database reported a relatively
large proportion of values at a single lower bound (such as 50 or 100 pCi/1), it was assumed that
all these values were "<", and the data were included in the database provisionally at one-half the
reported levels.
The proportions of censored data in the various supplementary databases are shown in
Exhibit 5-5. (These totals include the data that are reported as being censored and the additional
data points which we identified as reflecting "<" values.) It can be seen that between 7 and 8
percent of the data from three states (Ohio, Pennsylvania, and Washington) are censored. Six
additional state databases (California, Kansas, Maryland, Michigan, New Hampshire, and
Wisconsin) had fewer than about 2.5 percent censored results, while the remainder of the states
reported no censored data. On the whole, 237 of a total of 9,005 analytical results in the state
databases (2.6 percent) were censored. At least one censored result was present among the radon
analyses for about 4.7 percent of the 3,534 systems that were evaluated, and 1.6 percent of all
systems reported only censored results. In almost all cases, these latter systems reported only one
analytical result each. As will be discussed below, the censoring procedures that were employed
appeared to have relatively little impact on the derivation of summary statistics from the various
data sets.
b. Calculation of Summary Statistics for Uncensored Data
As discussed below, the lognormal distribution was the primary model used to
characterize the distribution of radon in groundwater systems. The natural log of the geometric
mean and the natural log of the geometric standard deviation are exactly comparable in their
meaning, for lognormal distributions, to the mean and standard deviation of a normal (Gaussian)
distribution. Thus, the log mean (natural log of the geometric mean) and log standard
deviation (natural log of the geometric standard deviation) are the primary statistics that are used
in the comparison of radon distributions.
For uncensored data, the log mean is simply equal to the arithmetic mean of the natural
logarithms of the radon data, and the log standard deviation is the standard deviation of the
natural logarithms. The relation ship between the geometric mean and standard deviation and the
log mean and standard deviation is thus defined as:
Log Mean = k In (GM) (5-1)
Log Standard Deviation = In (GSD) (5-2)
where GM and GSD are simply the geometric mean and standard deviation of the data,
respectively.
Methods, Occurrence, and Monitoring Document for Radon
5-14
-------
RADON DATA
Ed
H
2
93
pet
g
S
Ed
1
S.
o
r-rl
0
Q
Z
;
DC
.^
s
5
*!
E
«
£
S
•H
CO
COMMENT
CO
85 s
« 2E O
^ 2 id
s > K
u J S
1 0 H
* O
*h
III
*5 H (^
H ^ ^
H z
cn
CtJ
CO
iS
J
o
O ^ CA
as os ^j
« s s
S ^ M
» g 05
2^
C9 ^
w ^
§ J
CO Ed
Z >
td Ed
U J
0 CO
o < S
as > &
Ed CO «
CO Ed ^1
g J CO
— ' s >
^ < o
CO
w
ea
i
Ed
I
«o o"
200,000 pci/1 1
0
r-
»n
2
o
T
|
g
i
u
All results > 1,000 pCi/1
0
o
P!
o
<
C-l
g
1
1
All reported results > 100
0
o
3
o
<
s
OHVOII
S
All reported results > 40 p
o
o
o
o
<
o
I
a
J
•C
a
I
Abilene has three of six re
censorine limit
vo
OS
ON
vo
o
•n
tN
(
S
'KANSAS
,
s
?
X '
Lowest result = 22pCi/l, n
200pCi/l
0
o
s
o
<
s
i
_*
CN
0
n
O
n
r-
o
CN
t-
O
>n
1
&
2
1
O
™~
Is
=1 n
Two results < 0 pCi/1 excli
< 10 pCi/1 adjusted to 5 pC
0
cs
jg
Oj
o
~
MISSOURI
V
a
1
hi
TS
•S
Three medium systems v/i
100 excluded
cs
Jo
p
o
r-
8
1
1
y
s
i
Lowest result =;-14 pCi/1
0
o
t
^
o
<
m
5!
INEW YORK
p;
r-
00
S
OO
cs
s
cs
1
t~
—
00
OO
s
8
'
s
<
i
IPENNSYLVA
s g
3 O
_O OT*
"S ??
One censored sampled del
another sample from thes
was >1 0.000 pCi/1
,_
—
o\
00
—
o
(N
(
S
<
z
3
SOUTH CAR(
i
ts
r^
O
CN
en
8
§
1
ON
2
OO
-------
When data sets contained no censored values, or the frequency of censored values was
low, (< about five percent), the log mean and log standard deviation were calculated as described
above. This procedure was used with the large majority of data sets (see Exhibit 5-5). When the
data contained a higher proportion of censored data, other methods were used to calculate the
summary statistics.
c. Calculation of Summary Statistics for Censored Data (MLE Approach)
The NIRS data in particular, as well as some of the supplemental data sets, contained
substantial proportions of censored observations. When analyzing these data sets, the log mean
and log variance parameters were estimated by a maximum likelihood estimation (MLE) method,
a procedure proven to provide estimators with good statistical properties for reasonably large
samples. The specific approach that was used was the "EM algorithm" (Dempster,et al. 1979).
This algorithm provides a convenient and robust approach to developing MLE estimates of
distributional parameters. It involves the iterative calculation of log likelihood ratios while
updating the estimates of the censored values of the data until the likelihood ratio is optimized
within specified limits. A detailed discussion of the approach is given in Appendix B. 1.
The EM algorithm was implemented on Excel® spreadsheets for each data set and system
size stratum. The spreadsheets were designed to perform 20 iterations of the algorithm starting
from user-specified initial estimates of the sample mean and variance. For all of the data sets and
size strata, the EM estimates of the log mean and log variance converged very rapidly. Usually,
estimates of the log mean and log standard deviation were stable to the fourth decimal place
between the fifth and tenth iteration. The final estimates of log mean and log variance were quite
insensitive to the initial estimates used as inputs to the first iteration of the algorithm. As
discussed further below, the results of the EM algorithm were generally very consistent with the
estimates of distributional parameters derived using other methods.
d. Calculation of Proportions of Systems Above Radon Levels and Confidence
Limits on Proportions (Distributional Approach)
A major focus of this analysis was to predict the proportions of water systems or sources
that would be above potential regulatory limits. Thus, methods had to be found for estimating the
proportions of sources or systems that would exceed regulatory limits, given specified
distributions of radon levels.
In each data set, it is clear possible to estimate the proportions of sources or systems
above a given regulatory level simply by counting. In some data sets, the numbers of sources in a
given range of activity levels may be quite small. Thus, another method may need to be found to
provide an estimate of the number of sources exceeding some regulatory levels. Also, a method is
needed to predict the proportions of sources above regulatory levels for derived distributions (e.g.
for a lognormal distribution derived from NIRS data, adjusted for point-of-entry versus point-of-
use sampling).
Methods, Occurrence, and Monitoring Document for Radon
5-16
-------
In these cases, the proportions of sources/systems above potential regulatory levels were
estimated as:
p(>X) =
$[(LM-ln(X))/LSD]
(5-3)
where:
p(>X)
LM
LSD
The proportion of systems with radon levels exceeding (X) pCi/1
The standard normal cumulative distribution function
Log mean radon (natural logarithm of geometric mean) for sources
or systems being evaluated
Log standard deviation radon (natural logarithm of geometric
standard deviation) for sources of systems being evaluated
This approach simply applies the known properties of the cumulative normal distribution
to the estimated log mean and log standard deviation of radon occurrence in a given population of
sources/systems to estimate the proportion that would be expected to be above the potential
regulatory level, X pCi/L The proportions can be calculated easily using the "NORMDIST"
function in Excel®.
As noted above, the data used to estimate the log mean and log standard deviations of
radon occurrence were quite limited for some data sets. Therefore, it was necessary to calculate
the potential uncertainty in estimates of the proportions, and derive confidence limits for them.
The classical approach to estimating confidence intervals in this case is to estimate the standard
deviation of the estimated proportion, p(>X) , using the formula
SD(p(>X)) =
n
(5-4)
where:
SD = the Standard Deviation of the p(>X)
n = the number of observations upon which p(>X) has been estimated
The 95 percent confidence interval is then given by the large sample approximation
p(>X) ± 1.96 x SD(p(>Xj). ; <5'5)
Methods, Occurrence, and Monitoring Document for Radon
5-17
-------
This formula for SD(p(>XJ) is exactly correct only when the observed fraction of sources
exceeding the level (X) is used to estimate the population fraction (of all sources of the given
size). Since the proportions of sources exceeding potential regulatory levels were usually
estimated using a fitted log-normal distribution, the standard deviation of the estimated proportion
will not be given by the above formula. Equation 5-4 will tend to overestimate SD(p(>X))
because the maximum likelihood estimates have smaller variances than the empirical estimates if
sample sizes are relatively large. A more exact calculation uses the asymptotic distribution of the
maximum likelihood estimators, since the EM algorithm was used to derive maximum likelihood
estimates of the mean and variance of the log radon levels, which in turn were used to estimate
the proportions exceeding potential regulatory levels. We therefore used a more exact calculation
method for determining SD(p(>X)) and its associated confidence interval. This method, which is
described in detail in Appendk B.2, does, in fact yield slightly narrower confidence limits for the
proportions of systems and sources exceeding potential regulatory levels than does the classical
approach.
e. Estimation of Confidence Intervals on Proportions (Distribution-Free Method)
Because the fits of the radon distributions to lognormal distributions were not always very
good (see Section 5.4) estimates of confidence intervals on the proportions of systems exceeding
specified radon levels were also estimated using a distribution-free (also termed "non-parametric")
approach (Johnson and Katz, 1969). This method calculates the upper and lower confidence
limits, Pu and P,, respectively on p*, the estimated proportion of systems above a certain level.
(These parameters are often referred to as Clopper-Pearson confidence limits). This method,
which is based on counts of actual systems/sources above specified levels, makes no assumptions
about the underlying shape of the distribution of radon levels. Confidence intervals on the
proportions developed with this method, which is described in detail in Appendix B.3, are
compared to those derived using the distributional method in section 5.4.
5.2.2 Distribution Fitting and Goodness-of-Fit Testing
A number of approaches were employed to determine the extent to which the radon
occurrence data were consistent with common probability distributions. As discussed will be
seen in Sections 5.4 and 5.5, log-transformation of data from the NIRS and state databases (e.g.,
substituting the natural logarithm of the analytical result for the analytical result itself) results in
distributions that closely approximate "normality." Thus, the primary candidate for the
distribution of radon levels in these cases is the lognormal distribution, and the bulk of the effort
at distribution fitting and goodness-of-fit testing is designed to determine whether, and how well,
the radon data from the NIRS and supplemental data sets fit a lognormal model. Given the mixed
results of our attempts to fit lognormal distributions to the data, we also explored other plausible
distributions to see if the goodness-of-fit to the various data sets could be improved.
As is usual in such analyses, the first approach that is used is a qualitative graphical
method, namely the development of "probability plots" for the data sets in question. This
approach is also called the "regression on order statistics(ROS) approach" and has previously
Methods, Occurrence, and Monitoring Document for Radon
5-18
-------
been applied to the NIRS data by EPA (Barry and Brattin 1998). To develop a probability plot,
the data are ordered in terms of increasing values of the independent variable(x) and plotted as
(x(I), y(I)) pairs, and a regression line is fitted using the following equation:
•jvTT (5-6)
where:
k = the number of the ordered observation (including censored values)
N = total number of observations (including censored values)
u = the intercept estimate from the regression model
o = the slope estimate from the regression model
<& = the cumulative normal distribution
If all of the data points fall on or near the estimated regression line, this is an indication
that the data are consistent with a lognormal model. Curvature in the data in relation to the
regression line can indicate that the lognormal model is not appropriate. Plotting the data against
the regression line also can identify specific data points or groups of points that do not fit the
model well, and additional analyses can be performed on these subsets of the data.
Probability plots provide mainly a qualitative test of the lognormal model. A common
pattern in environmental data sets is that data points in the lower and middle ranges of
concentration or activity fit the lognormal probability plot quite well, but that some data points at
the upper end of the distribution deviate substantially by having log concentrations that are too
high to fit the plotted relationship for the rest of the data (Ott 1998). This pattern holds true to
some extent for the NIRS data, and this has lead one investigator (Burmaster 1998) to explore the
use of mixed lognormal distribution models to better fit these data. While this approach does
improve the modeled fit to the selected data sets, it adds a substantial level of complexity to the
estimation of proportions of systems exceeding specific radon levels and estimating confidence
limits around these proportions. Thus, the mixed models were not used in this analysis.
In addition to the qualitative measure of goodness of fit provided by the lognormal
probability plot, we also employed two quantitative tests to determine the goodness of fit of the
data to lognormal models. The first of these tests is the Shapiro-Wilk's W-test (Shapiro and
Wilk 1965). This procedure calculates a statistic essentially equivalent to the correlation between
the points in the probability plot described above. If the statistic is "significant," the hypothesis
that the underlying data are lognormally distributed should be discarded. This test has some
advantages over some other normality tests in that it is relatively powerful at small sample sizes,
and it is not dependent on selection of appropriate test strata (unlike the conventional x2 test).
When using the W-test, surrogate values equal to one-half the detection limits were substituted
for non-detects when running this test. Thus, the results of this test must be interpreted
cautiously.
Methods, Occurrence, and Monitoring Document for Radon
5-19
-------
The other goodness-of-fit test that has been used to test the hypothesis of lognormality is
the Anderson Darling D-statistic. Like the Shapiro-Wilk W-test, a "significant" result on this test
indicates that the hypothesis that the underlying data are lognormal should be discarded. This test
has the advantages that it is comparatively sensitive to deviation from lognormality in the "tails" of
the distribution. Also, the version of the test that we used (D'Agostino and Stephens, 1989) has
been adapted so that it can accommodate left-censored data. It has the disadvantage, however,
that critical values of the D-statistic have been calculated only for a situation in which the actual
geometric mean and log standard deviation of the distribution from which the data are drawn are
known. In the analyses which follow, these values are not known, but are estimated from the
data. This again introduces some uncertainty into the interpretation of this test. In this case,
using the critical values for the known distribution will probably be less likely to falsely identify a
lognormal data set as not being lognormal, but more likely to falsely identify a non-lognormal data
set as being lognormal. For these reasons, the Anderson-Darling statistic is used primarily to
compare goodness-of-fit among data sets and distributions, rather than as a rigorous test of
hypotheses regarding specific distributions.
5.2.3 Hypothesis Testing for Differences in Radon Activity Levels and Distributions
In Sections 5.4 and 5.5, we make a large number of comparisons of radon levels across
states, size strata, and data sets. In some cases, formal statistical tests are used to evaluate the
significance of the differences between radon levels. Because the various data sets that we
evaluate (national MRS data, state data, data from different system size strata), while generally
consistent with lognormality, vary in their distributional characteristics, we use both classical
"parametric" tests, as well as non-parametric tests of hypotheses.
a. Parametric Tests (Student's t)
For comparing two distributions using the classical Student's t-test, we applied the test to
the natural logarithms of the data. The results of this test indicate whether the log geometric
means of two lognormal distributions are significantly different (i.e., whether the hypothesis of
their equality can be discarded with a given degree of certainty). The t-test is used primarily to
compare data from the same system size strata between the NIRS and the radon data gathered by
the states. The degrees of freedom used in the test is (n(NIRS)+ n(State) -2). In implementing the t-
test, we substituted one-half the quantitation limits for censored observations in both the state and
NIRS data. The t-statistics for independent samples were calculated and the Levene test of
homogeneity of variances was employed to confirm the independence of samples.
For some data sets, we also employed a stratified t-test procedure to test the difference
between the weighted means of the NIRS and state data sets. Using this approach, we first
calculated the weighted mean difference in log radon level:
Weighted Mean Dif. = £(w)s [(log mean(NIRS) - log mean(state)]
(5-7)
Methods, Occurrence, and Monitoring Document for Radon
5-20
-------
where (w)s is the weight for each stratum of the data and log mean denotes the logarithm of the
geometric mean radon. Since the data are stratified by system size, these weights correspond to
the proportions of the various size groundwater systems in various states. The variance of the
weighted mean difference is:
Var(Weighted Mean Dif.) = £(w)s2 [Var(log mean(NIRS) + Var(log mean(state)] (5-8)
where the symbol "Var" indicates "the variance of.
The significance of the difference in means between the two data sets can then be
evaluated using a conventional Z-test (all the comparisons involve more than 30 degrees of
freedom), where Z is:
Z (two-sided) = Weighted Mean Dif.//{Var(Weighted Mean Dif.)}
(5-9)
Critical values for Z corresponding to p = 0.1 (one-sided test) or p = 0.05 (two-sided test) come
from standard statistical tables.
b. Non-Parametric Test for Difference of Means (Mann-Whitney U Test)
In order to confirm the results of the t-test, we also employed a non-parametric test to
evaluate the differences between means in the radon data sets. The Mann-Whitney U-Test was
used to evaluate differences in means in all comparisons where the total number of degrees of
freedom was less than 100, or where the number of samples from either stratum being compared
was less than 10. (For larger data sets, it is generally recognized that the t-test is almost always
more sensitive.) The U-test does not assume any specific distributional form of the data, and the
impact of including the censored data can be expected to be lower than on the t-test results. As
will be seen in Sections 5.4 and 5.5, the U-test and t-test results for the differences between the
NIRS and state data were highly consistent.
c. Test for Differences Between Distributions (Kolmorgorov-Smirnov Test)
The final test that we used to evaluate the differences between radon distributions was the
Kolmorgorov Smirnov test. This procedure compares the entire distributions being evaluated,
rather than estimating the significance of the difference of their mean values. This test was
originally intended to compare known distributions, rather than distributions fitted to
observational data. When used, as in this case, to compare fitted distributions, it is likely that the
test is somewhat conservative, that is, less likely to underestimate the significance of differences in
distributions.
d. Bootstrap Confidence Limits for Ratios of Log Means and Log Standard
Deviations and Numbers of Systems Exceeding Potential Regulatory Levels
Methods, Occurrence, and Monitoring Document for Radon
5-21
-------
The ratios of the log mean radon levels in the state and NIRS data sets, discussed in
Section 5.6, were not expected to be normally distributed. Therefore, we used a numerical
simulation method ("bootstrap") to estimate confidence limits for these ratios. The approach
taken is that described by Davison and Hinkley (1997).
The individual ratios of state to NIRS radon levels are used as inputs to a Monte Carlo
simulation model. For each statistic being analyzed, we took R = 200-500 samples of N values
from the input data, where N is the number of data points. "Studentized" bootstrap confidence
limits are then calculated as:
where:
u* - o* Z*(1_a) and u*-0*Z*(a)
u * = The mean of the R simulation estimates
(5-12)
o*
,Z*«> =
(l-a)'^ (o)
The standard deviation of the R simulation estimates
The a or 1-a percentile of the R ordered values of (u{ - u*) / o*
calculated from the simulation, where u{ is the "ith" ratio estimate
from the simulation and a is the desired confidence level
5.2.5 Computing Methods
Spreadsheet data files were developed and managed using Microsoft Excel® Version 5.0.
Larger data files were manipulated and analyzed using Microsoft Access® Version 2.0.
Probability plots, box plots, multiple regression analyses, single-stratum t-tests, Mann-Whitney U-
Tests, and Kolmorgorov-Smirnov tests were performed using the Statsoft, Inc. Statistica®
program Version 5.0. Pearson rank correlation coefficients were also calculated using this
package.
The Anderson-Darling goodness-of-fit tests, the E/M maximum likelihood algorithms, and
the estimation of confidence limits on the proportions of facilities exceeding specific radon levels
were implemented on Excel spreadsheets, as were the t-tests of the differences between weighted
average radon levels. The Monte Carlo simulation modeling used in the evaluation of temporal
and sampling and analytical variance and in the bootstrap analyses were performed using the
Crystal Ball Pro ® package as an Excel add-in.
5.3 Analysis of Radon Occurrence Data: Approach to Stratification
This section provides a discussion of the methods used in the exploratory data analysis of
the radon data obtained from the sources discussed in Section 5.1. The following sections
discuss, in turn, the approach used to stratify radon data from groundwater systems, comparison
of radon level data from the NIRS and supplementary sources, studies which directly evaluate
radon in raw and finished water, and the use of the lognormal model to predict the proportions of
systems exceeding potential radon regulatory levels.
Methods, Occurrence, and Monitoring Document for Radon
5-22
-------
5.3.1 Stratification by System Size
As discussed in Section 3.1, radon levels in groundwater systems are known to vary as a
function of system size. As early as 1979 (Hess, et al., 1979), it was found that private wells in
Maine generally had higher radon levels than larger public systems. An analysis of data from the
National Uranium Resource Evaluation (NURE) project (Hess, et. al. 1985) confirmed the
national pattern of generally higher radon levels in private wells compared to public wells, and
also provided suggestive evidence that radon levels in smaller public groundwater supply systems
were higher than those in larger systems. The NIRS sampling strategy, including as it did a high
proportion of small and very small systems, was designed in part to more fully investigate the
differences in radon levels across system sizes at the national level. The NIRS sampling strategy
divided groundwater systems into four size categories (Very Small, Small, Medium, and Large)
based on population served.
Data from the NIRS survey confirm that radon levels generally increase with decreasing
system size, as measured by the number of individuals served (Longtin, 1987). EPA's Occurrence
Analysis (Wade Miller 1990) also evaluated radon occurrence in the NIRS systems for these four
size strata of systems. Subsequently, EPA received comments (RCG/Hagler Bailly, 1992) that the
smallest size stratum evaluated in the occurrence analysis was heterogeneous enough (and that the
number of systems in this category was large enough) to merit splitting again into two separate
strata. In 1993, the revised occurrence analysis (Wade Miller, 1993) included evaluation of five
strata rather than four, and in 1995, with its Uncertainty Evaluation of Risks Associated With
Exposure to Radon (EPA 1995a), the agency also evaluated radon levels in the same five size
strata of systems:
Very Very Small (VVS) = serving 25 to 100 people;
Very Small (VS) = serving 101-500 people;
Small (S) = serving 501 to 3,300 people;
Medium (M)= serving 3 3 00-10,000 people; and
• Large (L) = serving more than 10,000 people.
The first two strata represent a subdivision of the original "Very Small" category. In this
analysis, we likewise evaluate radon levels in these five system sizes.
5.3.2 Alternative Stratification Variables
Other stratification variables in addition to system size have been considered. These
include stratification by region, by geological regime, and by other measures of system size, such
as total production (the amount of water pumped per day) or the number of points of entry to the
groundwater system (approximately equivalent to the number of wells). Based on a review of the
NIRS and supplemental data, system size measured in population served was retained as the
primary stratification variable. This decision was based on a number of considerations. Use of the
geologic regime as a predictor of radon activity was ruled out by a lack of geological regime
information for most of wells sampled in the NIRS and supplemental data sets. Alternative
Methods, Occurrence, and Monitoring Document for Radon
5-23
-------
measures of system size, such as number of points of entry, or total production, were found to be
less closely correlated to radon levels than population served in the NIRS data and not available
for the bulk of the supplemental data. Geographical stratification was used, however, to
characterize radon levels. In the first place, most of the supplemental data sets provided data on
specific geographic areas (states), so this was a natural unit of comparison to the NIRS data. In
addition, radon levels were found to vary substantially across different regions of the U.S. As will
be discussed in more detail below, the NIRS data were stratified into eight regions when
substantial variations in radon levels were seen across the country.
Stratification by sampling point was likewise impractical with the available data. The
NIRS samples were all taken at or near the point of use in the distribution system. Almost all the
data from the supplementary sources was taken at either the well head or at the point of entry into
the distribution system.5 Only one data set provided data from both the point of entry and point
of use. As will be discussed in more detail in Section 5.3, data from the point-of-entry samples in
the supplemental data sets were used to estimate the likely point of entry radon levels in
groundwater sources for the systems sampled during the NIRS. To estimate exposure levels from
the supplemental data, the point of entry radon levels were likewise adjusted downward to reflect
the expected losses of radon in the distribution systems.
Because the great majority of the supplemental data were gathered from samples taken at
the wellhead, it was also not possible to stratify these data according to the presence or type of in-
place treatment. This may have resulted in an overestimation of the average radon levels to which
consumers are exposed at systems having in-place treatments that might reduce radon levels
(aeration, granular activated carbon filtration, storage). The NIRS database contained
information on in-place treatments, but no significant relationships were found between the
presence or absence of treatment and radon levels in the distribution system.
5.4 Distribution of Radon Level in the NIRS Database
Like the different radon levels in different size systems, the tendency for radon activity
levels to fit lognormal distributions was noticed in the earliest systematic studies of its occurrence.
Hess, et al. (1979) characterized groundwater levels from in the NURE data from individual states
as being drawn from one or more lognormal distributions, and Longtin (1986) provided
probability plots (see below) for radon levels in groundwater indicating approximate lognormality
for the combined NIRS data from all states and size strata. In addition, there is good theoretical
reason to expect radon levels to approximate such a distribution. Ott (1995), for example, has
illustrated how random multiplicative dilutions of pollutants in the environment tend to produce
concentration distributions that are asymptotically lognormal. In addition, many distributions of
naturally occurring elements and man-made pollutants have been found to be approximate
5 Throughout this report, "point of entry", consistent, with EPA policy is defined as the
location just before water enters the distribution system. For many small and very small systems,
the point of entry is the wellhead. "Point of use" refers to any samples taken in the distribution
system or from the tap.
Methods, Occurrence, and Monitoring Document for Radon 5-24
-------
lognormality. However, since we intend to use the radon occurrence distributions to predict
proportions of systems above specific radon levels, we conducted additional evaluations of the
appropriateness of the lognormal model, and its utility in projecting radon levels.
5.4.1 Distribution of Radon in Nationally Aggregated NIRS Data
When the cumulative occurrence of radon in groundwater systems from the NIRS data is
plotted, without stratification for system size or region, the result is a very broad distribution with
a long "tail" extending to high radon activity levels. While the majority of radon values are less
than 300 pCi/1, there are a substantial number of observations above 5,000 pCi/1. When the data
are "log-transformed" (the natural logarithm of the radon levels are substituted for the raw
values), the distribution becomes somewhat more regular (Exhibit 5-6.) The large group of
systems at the left-hand end of the histogram are the 269 samples reported as being "less than 100
pCi/1". These are shown in the graph as having radon levels of 50 pCi/1 (whose natural logarithm
is 3.912.) The remaining values are distributed in a more or less smooth curve that is somewhat
bell-shaped (it looks a little like a "normal" or Gaussian distribution), but the high-activity tail is
still evident at ln(radon) values greater than 7.0 or so (corresponding to a radon activity of
approximately 1,100 pCi/1.) If the data were "perfectly" lognormal, this histogram would look
like a symmetrical bell-shaped curve. Thus, the evidence suggests that the NIRS data, viewed as
a whole, are nearly, but not perfectly lognormal.
Exhibit 5-6. Distribution of bi(Radon) in NIRS Data Set
(Ml System Sizes, All Regions)
285
<4.0 4.5-5.0 5.5-6.0 8.5-7.0 7:53.0 8.5-9.0 9.5-10.0
4.0-4.5 5.0-5:5 6.0-6.5 7.0-7.5 8.0-8.5 9.0-9.5 > 10.O
bi(Radon)
This impression is confirmed by other tests. Exhibit 5-7 shows a "probability plot" (as
described in Section 5.2.2) of the data from the NIRS, transformed by graphing it on a log scale.
A perfect straight-line fit would indicate a lognormal distribution of radon levels. There is some
Methods, Occurrence, and Monitoring Document for Radon
5-25
-------
Exhibit 5-7. Probability Plot of NIRS Radon Data
(All System Size Categories, All regions)
CSalid line iodkabe* (inarm.)
deviation from linearity, however, particularity near the high end of the data range, consistent with
the long "tail" seen in the conventional histogram. The graph thus suggests that the data are not
perfectly lognormal. In addition, statistical tests designed to test normality indicate that the data
are not consistent with a simple lognormal distribution. Two of the goodness-of-fit tests
discussed in Section 5.2, the Shapiro-Wilk W statistic and Anderson-Darling test, both indicate
that the hypothesis of lognormality for the NIRS data taken as a whole can be discarded with
greater than 95 percent certainty.
When the NIRS data are broken down by size strata, the linearity of the probability plots
improves but is still imperfect (Exhibit 5-8). The W-test again indicates that the data from all five
size categories depart from lognormality, although the larger systems come close to lognormality.
The better goodness-of-fit statistics for the larger systems are more a function of the smaller
number of systems in these categories (there are nine large and 11 medium systems with valid
results altogether) than an inherently better fifto the data.
Exhibit 5-8. ProbabHy Plots of ln(Radon) by System Size
(MRS, iU Regions)
J001
01
as
so
SO
so
as
D01
J01
JOS
to
SD
£0
ss
a 9
SIZE:
Very u«ry Small
12 a
e 9
SIZE:
Very Small
12 a
8 9
SIZE:
Small
12
6 9
SIZE:
Medium
12 3
t2
SIZE:
Large
Methods, Occurrence, and Monitoring Document for Radon
5-26
-------
The very small systems have a particularly high proportion of observations falling below
the linear plot of ln(radon) versus cumulative probability. Closer evaluation of these data indicate
that the highest value (In (radon)= 11.17, radon = 71,400 pCi/1) was more than five times the next
highest radon level in the entire NIRS database, and that the 15 very small systems with the
highest radon levels were concentrated in two states with high average radon levels (Connecticut
and North Carolina). Further, three pairs of data points from these states were taken from water
systems located within 10 miles of one another. Thus, it appears that the NIRS data are not
perfectly representative of the geographic distribution of radon in very small systems, and that this
has affected the goodness-of-fit to the lognormal model. In the estimation of geometric mean and
standard deviations that are discussed below, the highest value in this stratum is omitted as
unrepresentative.
Despite the fact that the probability plots do not show perfect linearity for the different
size systems, the slopes and intercepts of the plots provide approximate estimates of the
geometric mean and geometric standard deviations of the radon analytical results. These results
are summarized in Exhibit 5-9, along with estimates of these statistics developed in previous
analyses. Also included are estimates of the geometric means and standard deviations of radon
levels in the various size strata of the NIRS data developed using the iterative maximum
likelihood approach (EM algorithm) described in Appendix B-2. The different techniques arrive
at generally similar estimates for the radon distributional parameters. The geometric mean values
for the size strata decrease from about 270-285 pCi/1 for the smallest systems (log mean ~ 5.65)
to 125-135 pCi/1 for the largest systems (log mean ~ 4.85). The corresponding arithmetic means
range from approximately 795 pCi/1 for the very very small systems to about 185 pCi/1 for the
large systems.
Compared to the other estimates, the EM algorithm consistently yields slightly higher log
means and slightly lower log variances for all the size strata. The reasons for the differences in
the results between the current study and the 1990 EPA occurrence analyses (Wade Miller
Associates 1990) due to the use of the EM algorithm, but are more likely due to a different
treatment of a few samples with irregular results ("non-detect" values greater than 100), or to the
use of slightly different algorithms to calculate the order statistics for the probability plots. In
addition, EPA's 1990 occurrence analysis did not disaggregate the very small and very very small
strata. Similar factors probably explain the somewhat smaller differences between the results of
EPA's 1995 Uncertainty Analysis (EPA 1995) and our results.
It can be seen from the data in Exhibit 5-9 that there is a clear trend (as expected) of
increasing log radon levels with decreasing system size. The results of a one-way analysis of
variance (ANOVA) indicates that differences in systems size explain a significant proportion of
the overall variance in the data. (The f-statistic for inter-group differences is =15.18, p = 0.000,
indicating a high degree of statistical certainty that system size is associated with log mean radon
level). In addition, the ANOVA post-hoc comparison of means indicates that the log mean radon
levels for the two smallest size strata differ significantly from those of the three larger system
strata. However, the differences among the means of the three large-system strata (large,
medium, and small) are not significant, nor were the differences between the means of the two
Methods, Occurrence, and Monitoring Document for Radon
5-27
-------
EXHIBIT 5-9 ESTIMATED DISTRIBUTIONAL PARAMETERS FOR RADON IN GROUNDWATER SYSTEMS, NIRS
DATA, NATIONALLY AGGREGATED
1. Maximum Likelihood Estimates (This Study)
Log Mean Radon
Geometric Mean
Arithmetic Mean 1
Log Standard Deviation Radon
Number of Systems
Number of Systems With Cenored Data
2, Regression Estimates (This Study)
Log Mean Radon
Geometric Mean
Arithmetic Mean
All Systems
5.325
205
561
1.417
981
269
5.228
•186
572
1.497
Large
4.871
130
191
0.872
335
71
Medium
4.924
138
197
0.847
333
74
4.832
125
181
0.855
4.875
131
199
0.913
Small
4.856
129
275
1.234
232
95
4.811
123
278
1.277
Very Small
5.469
237
629
1.397
53
18
5.372
215
699
1.535
Small
5.645
283
792
1.435
28
11
5.667
262
796
1.492
3. EPA Occurrence Analysis (Wade Miller, 1990)
Loq Moan Radon
Geometric Mean
Arithmetic Mean
4. EPA Uncertainty Analysis (1995)
Log Mean Radon
Geometric Mean
Arithmetic Mean
Log Standard Deviation Radon
—
—
_
4.897
134
205
0.924
_
_
-
4.910
136
199
0.878
4.895
134
206
0.933
4.822
124
284
1.287
5.490
242
758
1.510
4.889
133
200
0.903
4.835
126
281
1.267
5.373
216
677
1.513
-
-
-
-
5.586
267
831
1.508 ;
Notes:
1. Calculated from fitted lognormal distribution
2. Statistics for Very Small category Include Very Very Small Systems
smallest size strata. Tests for differences between the distributions using the Kolmorgorov-
Smirnov test (a procedure for testing the differences between distributions) showed the same
pattern; the distributions of the three largest and two smallest size categories were significantly
different from one another.
5.4.2 Distributions of Radon in Regionally Stratified NIRS Data
As noted above, the NIRS data were also stratified by region. This approach was
originally suggested by commenters on EPA's 1990 and 1993 occurrence analyses (RCG/Hagler
Bailly 1993), and adopted based on the finding that radon distributions varied significantly among
regions of the U.S. The regional stratification proposed by the commenters was:
New England (CT, ME, MA, NH, RI, VT)
Appalachians (DE, GA, MD, NJ, NY, NC, PA, SC, VA, WV)
Plains (AR, KS, KY, MI, MO, NE, ND, OK, SD, TN)
Great Lakes (IL, IN, IA, MN, OH, WI)
Methods, Occurrence, and Monitoring Document for Radon
5-28
-------
Gulf Coast (AL,FL, LA, MS, TX) :
Northwest (OR, WA)
California (CA)
Rocky Mountains (AZ, CO, ID, MT, NV, NM, UT, WY)
EPA had some reservations about adopting these regional divisions (placing Michigan in
the "Plains" region, and Iowa in the "Great Lakes"). Subsequent analysis, however, suggested
that these divisions, despite the counterintuitive assignment of some states to regions, capture
important differences in radon levels, (see below) and, in combination with the data from
individual states, provide a useful basis for predicting the proportions of systems exceeding
potential regulatory levels. In addition, the regional stratification provides another opportunity to
test the issue of lognormality.
Probability plots for the NIRS data stratified by region are shown in Exhibit 5-10.
Subjectively, they appear to be more linear than the plots of radon data stratified by system size.
However, the Shapiro-Wilk W statistic indicates that of all eight regions, the data from the Rocky
Mountain regional is the only combined data set consistent with the lognormal distribution.
Stratifying the data by both region and system size, as is illustrated for the data from the
Appalachian Region in Exhibit 5-11, again seems to improve the linearity of the probability plots
for the smaller size strata. However, this stratification does not; significantly improve the
goodness-of-fit test results, which still indicate that the hypothesis of lognormality may be
excluded with p < 0.05 for the three strata (the smallest systems) with more than a few systems.
Similar results are seen for the other regions.
Exhibit 5-10. Probability Plots of InflFtadon) by Region
(All Size Strata)
•§
f
a
i
s
OH
JOS
3D
SS
otn
as
so
an
x>s
so
ss
6 9 12 3
Appalachian
8 9
California
12 3
8 9
Gulf Coast
12
8 9
Great Lakes
12 a
6 9
New England
12 a
8 9
Northwest
12
6 9 12 3 6 9 12
Plains Rocky Mountains
Methods, Occurrence, and Monitoring Document for Radon
5-29
-------
Exhfeit 5-11. Probabiity Plots of bi(Radon) from MRS Appalachian Region
Data by Size Category
6 9
Very Very Small
3 6 93
Very Small
Small
Medium
Large
noted above, the regional stratification does appear to capture important differences in radon
activity levels. Exhibit 5-12 shows the log mean and log standard deviation values estimated for
the combined systems in the eight regions with statistically significant differences indicated. Most
of the regional log means are significantly different from one another, using the t-statistic as a
criterion. Exceptions include the plains region, whose mean is similar to that for the Gulf Coast,
Great Lakes, and Northwest, and the Rock Mountain region, whose log mean is near that of the
Appalachian region and California. The Kolmorgorov-Smirnov test gives the same
radon distributions as being significantly different from one another (the Gulf Coast compared to
the Great Lakes and Northwest regions) whose means are not significantly different based on the
t-test.
5.4.3 Goodness of Fit Testing of Lognormal and Alternative Distributions of NIRS
Data
Because the previous analyses indicated the possibility that the aggregated and
dissagregated NIRS data might not fit lognormal distributions, we also employed goodness-of-fit
testing to determine whether the lognormal provided the best overall fit to the data. The entire
NIRS data set, and the data broken down by Region and size were fit to a number of analytical
distributions and the Anderson-Darling statistic (see section 5.2) was used to estimate the relative
Methods, Occurrence, and Monitoring Document for Radon
5-30
-------
co
o o
on
S
a
.2
•**
«
Q
•E
-a
a
a
«
w
I
Q
S-
A
a
o
Ml
a
o
S
€
Exhibit 5
-SP
c«
2
"bO
fl
"
o
S,
^
GO O
o
«
Q
.2
bO
"i
02;
(N
o
t
CO
S
oo
oo
Gulf Co
4_> en
S
o "X
fcj «
O »J
S
vo ^^
~ 0
00 °°.
0 0
t
g
vo
CS /-v
^^ CO
CS
g
oo
o
+ i
u-f -5
CN M
O 4)
O /t3
*
03
3
"55 x
ts "
Q
<»
8 ii
O co
00 g
Q +-
S
GO
"53 CM
a S
,£3 O
C8
00
++
,
ov
tg a
e °?
"co o
s
I
B
|
I
I
J
1
s
-------
fits of lognormal and other distributions that were tested.6 The results of the goodness-of-fit
testing are shown in Exhibit 5-13.
Using the relative magnitude of the Anderson-Darling (A-D) statistic as the indicator, the
lognormal distribution provides the best fit to the data as a whole, and to most of the subsets of
the NIRS data when they are stratified either by size or region. The lognormal provides the best
fit to data from three smallest system categories, the third-best fit to the data from medium
systems, and the second-best fit to data from the large system. For data as a whole, and for the
three smallest size categories, the values of the A-D statistics for the lognormal distribution are
such that, using the statistic alone, the lognormal could be discarded as an acceptable fit to the
data (p < 0.05). However, the A-D statistics for the other types of distribution are even larger.
Thus, it appears that the lognormal does as well or better than any of the distributions at
describing the radon data for these data. In the case of the two largest size categories, the
lognormal is not the best-fitting distribution, but the differences in A-D statistic values among the
distributions are so small that there really is no basis to conclude that the lognormal is any worse
then the other contending candidates.
When the NIRS data are stratified by region, the lognormal provides the best fit (as
measured by the A-D statistic) for six of the eight regions. For three regions (California, New
England, and the Rocky Mountains), the A-D value is low enough that the hypothesis of
lognormality cannot be ruled out with any degree of statistical certainty. For another three
regions (Appalachian, Gulf Coast, and Great Lakes), the A-D value could rule out the lognormal
distribution, but the fit for the other distributions appears to be even worse.
In two regions, (the Northwest and Plains), the lognormal distribution apparently does not
provide the best fit to the data. Data from the Plains regions do not fit any of the distributions
very well. This is apparently because of the large proportion (41 percent) of censored
observations in this stratum. The data from tbe Northwest region are characterized by low radon
levels and low levels of variability, compared to other regions. This results in relatively low A-D
statistics for a number of distributions, including the lognormal, even though the lognormal is not
the best fit.
Taken together, we interpret these results to suggest that the lognormal distribution is, on
the whole, the most practical model for describing and predicting radon levels in groundwater
systems. The summary statistics for the radon data for all of the regions and size strata in the
NIRS data are shown in Exhibit 5-14.
5.5 The Distribution of Radon in the Supplementary Data Sets
In this section, we describe the results of our analysis of radon occurrence data from some
of the supplementary data sets obtained during the occurrence analysis. The focus is primarily on
6 The distribution types that were tested included the uniform, triangular, normal
(Gaussian), lognormal, two- and three-parameter Gamma, Beta, Logistic, Type 1 Extreme Value,
Pareto, exponential, and Weibull.
Methods, Occurrence, and Monitoring Document for Radon 5-32
-------
-------
Exhibit 5-14. Summary Statistics for NIRS Data, Broken Down By Region1
R ecj io n
Appalachian
G eometrlc Standard Deviation
Arithmetic Mean. oCi/l
Loo M ean
Loo Standard Deviation
Numbar of Svstem s
Systems With Censored Data
California
Geometric Mean. oCi/l
Geometric Standard Deviation
Arithmetic Mean. pCi/t
Loo Mean
Loo Standard Deviation
Number of Svstems
Systems With Censored Data
Gulf Coast
Geometric Mean, oCi/l
Geometric Standard Deviation
Arithmetic Mean. oCi/l
Loo M ean
^09 Standard Deviation
Number of Svstems
Svstems With Censored Data
G re at Lakes
Geometric Mean. oCi/l
Geometric Standard Deviation
Arithmetic Mean. oCi/l
Loo M ean
Loo Standard Deviation
Number of Svstem s
Svstems With Censored Data
New Enaland
Geometric Mean. oCi/l
Geometric Standard Deviation
Arithmetic Mean. oCi/l
Loo Mean
Loo Standard Deviation
Number of Svstems
Systems With Censored Data
Northwest
Geometric Mean. oCi/l
Geometric Standard Deviation
Arithmetic Mean. oCi/l
Loo Mean
Loo Standard Deviation
Numberof Svstems
Systems With Censored Data
Plains
Geometric Mean. oCi/l
Geometric Standard Deviation
Arithmetic Mean. oCi/l
Loo Mean
Loo Standard Deviation
Number of Svstems
Systems With Censored Data
Rockv Mountains
Geometric Mean. oCi/l
Geometric Standard Deviation
Arithmetic Mean. oCi/l
Loo Mean
Loo Standard Deviation
Number of Svstems
Systems With Censored Data
Svstem Size
ALL
333
4.76
1.127
5.81
1 .56
179
31
ALL
333
3.09
629
5.81
1.13
60
10
ALL
125
3.38
263
4.83
1 .22
189
75
ALL
151
3.01
278
5.02
1.10
191
61
ALL
1.214
3.77
2.933
7.10
1.33
59
4
ALL
161
2.23
222
5.08
0.80
66
20
ALL
132
2.65
213
4.88
0.98
136
52
ALL
361
2.77
607
5.89
1.02
68
6
VVS
378
4.36
1.118
5.94
1 .47
84
16
VVS
359
3.38
754
5.88
1.22
29
5
VVS
183
3.21
362
5.21
1.17
48
1 1
VVS
168
2.43
250
5.13
0.89
40
1 1
VVS
1.657
3.43
3.543
7.41
1.23
26
1
VVS
160
2.14
214
5.07
0.76
36
11
VVS
172
2.77
289
5.15
1.02
34
9
VVS
559
2.54
863
6.33
0.93
24
0
VS
419
5.71
1 .912
6.04
1 .74
56
7
VS
525
2.53
808
6.26
0.93
16
1
VS
181
2.88
318
5.20
1.06
68
18
VS
182
3.54
404
5.20
1.26
65
15
VS
1.169
4.61
3.760
7.06
1.53
24
3
VS
167
2.12
222
5.12
0.75
22
6
VS
127
1.75
148
4.84
0.56
34
13
VS
226
3.32
464
5.42
1.20
22
6
S
199
3.90
502
5.29
1 .36
31
7
S
267
1.84
322
5.59
0.61
7
1
S
64
3.85
1 59
4.16
1 .35
53
32
S
121
2.88
212
4.80
1 .06
65
28
S
479
1.69
550
6.17
0.52
6
0
S
111
3.73
265
4.71
1 .32
6
3
S
83
3.32
170
4.42
1 .20
45
26
S
345
2.34
495
5.84
0.85
1 8
0
M
158
1.67
180
5.06
0.51
6
1
M
145
2.05
187
4.98
0.72
6
2
M
35
3.57
78
3.55
1 .27
10
8
M
140
2.43
208
4.94
0.89
14
5
M
645
1 .23
659
6.47
0.21
2
0
M
112
1.00
1 12
4.72
0.00
1
0
M
142
1.74
166
4.96
0.56
13
3
M
248
1.84
299
5.51
0.61
3
0
1. Regions defined as described in text. Includes states with supplemental data.
Methods, Occurrence, and Monitoring Document for Radon
5-34
-------
those data sets which are intended to be representative of the radon distributions in individual
states, and include the results of statewide radon surveys or ongoing radon monitoring programs.
In Section 5.6, we compare the data from the supplementary data sets to the NIRS results from
the same states and regions in order to develop national estimates of radon occurrence in
groundwater supplies.
5.5.1 Distributions of Radon in Supplemental Dat^ Sets
To more fully investigate the distribution of radon levels in groundwater, we examined the
distributions of radon levels in data from 17 states who had submitted substantial amounts of data.
The results of the analysis were quite variable. Aggregated data from community water systems
in some states (Maine, for example) gave very nearly linear probability plots (Exhibit 5-15). On
the other hand, data from some other states clearly depart from lognormality (data for Texas is
given in Exhibit 5-16.) Goodness-of-fit testing was performed on the aggregated data from the
individual state data sets, and on data from individual size categories when sufficient numbers of
systems had been sampled. (Below about 10 systems, the ability of the Anderson-Darling statistic
to distinguish among the goodness-of-fit for different distributions was found to be quite limited.)
Exhibit 5-15. Probability Plot of InfRadlon) Data From Maine
Community Water Systems (Hess, 1979)
,-. .001
o
* 0.01
£ 0.05
Q>
0>
0.2
Ul 0.5
£• 0.8
£ 0.95
CL 0.99
(403 pCM) (1096pCl/l) (2981 pCi/I) (8103 pCW)
6
8
9
10
ln(Radon}
It can be seen from the data in Exhibit 5-17 that the lognormal distribution provided the best fit
(as measured by the A-D statistic) for 14 out of 17 of the data sets evaluated. In seven of these
cases, the value of the A-D statistic is judged to be consistent with lognormality. In the seven
other statewide data sets where the lognormal is judged to be the best fit, the A-D statistic
indicates that the data are not consistent with the lognormal distribution. In these cases, however,
the other distributional forms that were tested fit the data even more poorly. For one data set
Methods, Occurrence, and Monitoring Document for Radon
5-35
-------
Exhftit 5-16. Probabfity Ptot of Ration Concentrations from Texas
Community Water Systems (MRS Data)
4 5
bKRadon)
(Texas) the lognormal was the second-best fit to the data, but the A-D statistic indicated that the
hypothesis of lognormality could not be discarded at p < 0.05. For two state data sets (Ohio and
Exhibit 5-17. Goodness of Fit Ranking of Lognormal Distribution Applied
to State Radon Data Sets
State
California
Connecticut
Idaho
Iowa
Kansas
Maine
Maryland
Michiqan
Missouri
New Hampshire
New York
Ohio
Pennsylvania
South Carolina
Texas
Washington
Wisconsin
System Size
All
1 + <1-2>
1 +
1 +
1 +
1 -
1 +
1 +
1 +
1 -
1 -
1 -
5 -
1 -
1 -
2 +
8 -
1 -
Large
1 +
—
2 +
—
—
7 +
1 +
1 +
6 +
4 +
3 +
2 +
2 +
9 -
1 +
Medium
6 +
—
1 +
3 +
4 +
—
4 +
7 -
3 +
4 +
6 +
2 +
3 -
1 +
8 +
1 +
Small
__(3)
__
3 +
6 -
7 +
1 +
2 +
2 -
2 -
5 -
4 -
1 +
1 +
1 +
8 +
1 -
Very Small
2 +
—
4 -
1 +
2 +
5 -
1 +
1 +
1 +
1 +
__
2 -
1 +
2 +
8 +
1 +
Very Very
Small
—
1 +
__
-..
7 +
1 +
1 +
1 +
1 +
__
1 +
8 +
1 -
„_
1 -
Notes:
1. Numbers in cells are rankinq of lognormal among nine distributions tested, based on Anderson
2. A (+) indicates that the data are consistent a lognormal distribution at p = 0.05; a (••) indicates
3. (--) indicates not enough data to judge goodness of fit.
Methods, Occurrence, and Monitoring Document for Radon
5-36
-------
Washington), the lognormal was far down on the list of distributions, in terms of goodness of fit.
Sufficient data were available to allow goodness of fit testing on 64 individual size
categories of data within the 17 state data sets (Exhibit 5-17). In 27 cases, the lognormal
provides the best fit, and in another 11 cases, the lognormal is ranked second, in terms of
goodness of fit. In an additional 19 cases, (mostly smaller data sets), the lognormal is ranked
lower than second (e.g, it is at least the third best-fitting distribution), but the A-D statistic cannot
rule out the data being consistent with the lognormal model. In ten of the dissagregated data sets,
the lognormal is ranked worse than second, and the goodness of fit statistic indicates the data do
no fit the lognromal model. In a total of 48 of the 64 cases, the data are indicated to be consistent
with the lognormal distribution. In two of the state data sets (Ohio and Washington), the
lognormal ranks low when the data as a whole nor any of the individual size strata are evaluated.
No specific distribution turned up more frequently in the supplemental data sets as
providing a better fit than the lognormal. The other distributions providing the best fit the data
included, in rough order of frequency, the Weibull, three-parameter Gamma, Type 1 Extreme
Value, logistic, and exponential. None of these distributions appeared as frequently among the
best-fitting distributions, or consistently had lower A-D values, than the lognormal. Thus,
consistent with the pattern seen in the NIRS data, it was decided that the lognormal was the most
suitable distribution for describing the supplementary data sets.
5.5.2 Radon Summary Statistics from Supplementary Data Sets
Exhibit 5-18 summarizes the distributions of system radon levels that were derived from
the 17 supplemental data sets.7 As noted in the previous section, some of the data sets provide
significant information on radon levels in groundwater sources serving all size classes of systems.
In some of the data, sets the number of systems in some strata (usually the larger systems) is
relatively small, and in two state data sets (Connecticut and Idaho) system size data were not
available. Departing from the general pattern, the data sets from California, Ohio, and Wisconsin,
are composed of data predominantly from larger systems. The radon survey in South Carolina
from which the state data were derived was intentionally designed to sample roughly equal
numbers of systems in each size category. ;
The state-wide geometric mean radon levels seen in the data sets vary from a maximum of
8,973 (Connecticut) to a low of 139 (Texas). The geometric mean values for "all" systems in
each state data set are shown in Exhibit 5-19 (except for Connecticut, which would be off the
scale at the top.) Consistent with previous studies, radon levels in the New England states are
seen to be very high, compared to the rest of the country. Only Connecticut, New Hampshire,
7 When more than one measurement was available from one or more sources in a system,
the system geometric mean was calculated with each sample weighted equally. System arithmetic
means are calculated from the derived lognormal distributions of radon levels. State-wide
summary statistics are calculated with each system, irrespective of size, weighted equally. The
radon distributions in Exhibit 5-17 are therefore not population-weighted.
Methods, Occurrence, and Monitoring Document for Radon 5-37
-------
Exhibit 5-18. Summary erf Radon Occurrence Data from 17 State-Wide Data Sets
STATE
CALIFORNIA
GEOMETRIC WEAN
ARITHMETIC MEAN
LOG MEAN
LOG STANDARD DEVIATION
NUMBER OF SYSTEMS
CENSORED1
CONNECTICUT
GEOMETRIC MEAN
ARITHMETIC MEAN
LOG MEAN
LOG STANDARD DEVIATION
NUMBER OF SYSTEMS
CENSORED
IDAHO
GEOMETRIC MEAN
ARITHMETIC MEAN
LOG MEAN
LOG STANDARD DEVIATION
NUMBER OF SYSTEMS
CENSORED
IOWA
GEOMETRIC MEAN
ARITHMETIC MEAN
LOG MEAN
LOG STANDARD DEVIATION
NUMBER OF SYSTEMS
CENSORED
KANSAS
GEOMETRIC MEAN
ARITHMETIC MEAN
LOG MEAN
LOG STANDARD DEVIATION
NUMBER OF SYSTEMS
CENSORED
MAINE
GEOMETRIC MEAN
ARITHMETIC MEAN
LOG MEAN
LOG STANDARD DEVIATION
NUMBER OF SYSTEMS
CENSORED
SYSTE
ALL
284
473
5.65
1.01
74
0
ALL
8.973
14935
9.10
1.01
32
0
ALL
537
716
6.29
0.76
64
0
ALL
283
436
5.65
0.93
150
0
ALL
299
456
5.70
0.92
169
7
ALL
1.300
Z080
7.17
0.97
64
0
WS
153
__
5.03
_
1
0
WS
_
_
—
__
_
WS
_
WS
642
3.245
6.46
1
-------
Exhibit 5-18. Summary of Radon Oc
MARYLAND
GEOMETRIC MEAN
ARITHMETIC MEAN
LOG MEAN
LOG STANDARD DEVIATION
NUMBER OF SYSTEMS
CENSORED
MICHIGAN
GEOMETRIC MEAN
ARITHMETIC MEAN
LOG MEAN
LOG STANDARD DEVIATION
NUMBER OF SYSTEMS
CENSORED
MISSOURI
GEOMETRIC MEAN
ARITHMETIC MEAN
LOG MEAN
LOG STANDARD DEVIATION
NUMBER OF SYSTEMS
CENSORED
NEW HAMPSHIRE
GEOMETRIC MEAN
ARITHMETIC MEAN
LOG MEAN
LOG STANDARD DEVIATION
NUMBER OF SYSTEMS
CENSORED
NEW YORK
GEOMETRIC MEAN
ARITHMETIC MEAN
LOG MEAN
LOG STANDARD DEVIATION
NUMBER OF SYSTEMS
CENSORED
OHIO
GEOMETRIC MEAN
ARITHMETIC MEAN
LOG MEAN
LOG STANDARD DEVIATION
NUMBER OF SYSTEMS
CENSORED
ALL
440
1.490
6.09
1.56
107
1
ALL
187
253
523
0.78
120
5
ALL
156
217
5.05
0.85
691
5
ALL
2.017
4.911
7.61
1.33
788
32
ALL
318
657
5.76
1.21
424
0
ALL
140
228
4.94
0.99
228
23
icurrence Data from 17 S
ws
675
2.759
6.51
1.68
28
1
WS
143
202
4.96
0.83
14
1
WS
197
254
5.28
0.67
209
0
WS
2.507
5.032
7.83
1.18
386
3
WS
290
605
5.67
1.21
82
0
WS
—
vs
517
1.902
6.25
1.61
33
0
VS
183
288
5.21
0.96
30
3
VS
162
219
5.09
0.79
197
0
VS
2,472
5.251
7.81
1.23
279
6
VS
348
1,016
5.85
1.46
141
0
VS
489
492
6.19
0.09
2
0
tate-Wide Data Sets (Continued)
s
281
573
5.64
1.19
35
0
S
211
273
5.35
0.72
42
0
S
137
201
4.92
0.92
208
3
S
680
2.452
6.52
1.60
91
17
S
321
532
5.77
1.01
163
0
S
149
235
5.00
0.96
115
10 .
M
667
1,746
6.50
1.39
7
0
M
168
205
5.12
0.63
20
1
M
106
159
4.67
1.07
56
2
M
593
841
6.39
0.84
16
1
M
312
459
5.74
0.88
22
0
M
134
211
4.90
0.95
65
6
L
214
1.926
5.36
2.10
4
0
L
202
244
5.31
0.62
14
0
L
97
151
4.58
1.01
21
0
L
331
1.270
5.80
1.64
16
5
L
210
298
5.35
0.84
16
0
L
119
218
4.78
1.10
46
7
Methods, Occurrence, and Monitoring Document for Radon
5-39
-------
Exhibit 5-18. Summary of Radon Occurrence Data from 17 State-Wide Data Sets (Continued)
PENNSYLVANIA
GEOMETRIC MEAN
ARITHMETIC MEAN
LOG MEAN
LOG STANDARD DEVIATION
NUMBER OF SYSTEMS
CENSORED
SOUTHCAROLJNA
GEOMETRIC MEAN
ARITHMETIC MEAN
LOG MEAN
LOG STANDARD DEVIATION
NUMBER OF SYSTEMS
CENSORED
TEXAS
GEOMETRIC MEAN
ARITHMETIC MEAN
LOG MEAN
LOG STANDARD DEVIATION
NUMBER OF SYSTEMS
CENSORED
WASHINGTON
GEOMETRIC MEAN
ARITHMETIC MEAN
LOG MEAN
LOG STANDARD DEVIATION
NUMBER OF SYSTEMS
CENSORED
WISCONSIN
GEOMETRIC MEAN
AF«THWlETlCMEAN
IQ&MEAN
LOG STANDARD DEVIATION
NUMBER OF SYSTEMS
CENSORED
ALL
500
1.403
6.21
1.44
488
24
ALL
542
1.767
6.30
1.54
89
1
ALL
139
218
4.94
0.95
120
50
ALL
141
279
4.95
1.17
68
10
ALL
297
416
5.69
0.82
610
7
WS
465
1.310
6.14
1.44
151
5
WS
2,024
3.206
7.61
0.96
22
0
WS
194
284
5.27
0.87
14
4
WS
26
86
3.27
1.54
6
3
WS
260
380
5.56
0.87
176
0
VS
489
1.853
6.19
1.63
153
12
VS
1.250
4.240
7.13
1.56
13
0
VS
142
224
4.96
0.95
33
14
VS
155
296
5.04
1.14
15
2
VS
314
456
5.75
0.87
140
0
S
546
1.243
6.30
1.28
95
2
S
279
680
5.63
1.33
20
1
S
152
265
5.02
1.05
36
14
S
160
244
5.07
0.92
11
0
S
333
468
5.81
0.83
197
0
M
493
1.214
6.20
1.34
4
5
M
325
941
5.78
1.46
18
0
M
138
192
4.93
0.82
18
7
M
139
262
4.93
1.13
15
2
M
273
340
5.61
0.66
67
0
L
575
1.032
6.35
1.08
38
0
L
173
299
5.15
1.04
16
0
L
91
114
4.52
0.66
16
10
L
187
325
523
105
21
3
L
279
304
5.63
0.42
30
0
Notes:
1. Indicates all samples from the systems are censored.
Methods, Occurrence, and Monitoring Document for Radon
5-40
-------
Exhibit 5-19. State-Wide Geometric Mean Radon
Levels from Supplemental Data Sets
2,000-
c
(0 —
Q) :=:
o -
II
c «s
8*
o
and Maine, have overall geometric mean radon levels that exceed 1,000 pCi/1 in the supplemental
data, with the next highest average value being 542 pCi/1 for South Carolina.
Consistent with the pattern seen in the NIRS data, the geometric mean radon levels
increase with decreasing system size for the majority of the data sets (Exhibit 5-20.) In 10 of the
15 supplemental data sets reporting radon values for multiple size strata, the very very small or
very small systems had the highest geometric mean radon values. In the other data sets, the
pattern varied, with the larger systems sometimes having the highest values. The data sets in
which the larger systems had radon levels comparable to those small systems were generally from
states where the average radon levels were low, and where the variation in radon levels among the
size strata was also the lowest. Overall, the levels of radon seen in the supplemental data sets
were similar to those seen in the NIRS data for the same regions (but see below).
Log standard deviations of the radon levels were also calculated for the various
supplemental data sets and are tabulated in Exhibit 5-18. The log standard deviations of radon
levels seen in the combined data sets (all systems sizes) vary from 1.56 for the data from
Maryland to 0.76 for the data from Idaho. Much of the variance in these values conies from
differences between radon levels in the different size strata, and, for most states, the log standard
deviation values for the individual strata are substantially lower than those for the aggregated data
sets.
Methods, Occurrence, and Monitoring Document for Radon
5-41
-------
Exhibit 5-20. Geometric Mean Radon Levels as a
Function of System Size in Supplemental Data Sets
D \fery Very Small
Very Small
• Small
D Medium
Unlike the geometric mean values, there is no consistent pattern in the log standard deviation with
changes in system size. Similarly, the log standard deviation values in the supplemental data sets
are only weakly correlated with the geometric mean values. This implies that the relative
variability of radon levels is more or less constant across groups of systems with different
geometric mean radon levels.
5.6 Comparison of NIRS and Supplemental Data Sets
As noted in Section 5.4, the NIRS data are derived from samples taken within water
distribution systems. In contrast, all of the state supplemental data that we analyzed come from
samples taken at the wellhead or at other sampling locations before the water enters the
distribution system. Thus, we expect to see a systematic difference between the NIRS data and
the supplemental data, with the supplemental data showing higher radon levels than the NIRS for
the same state and system size categories. The magnitude of the difference would reflect the
average reduction in radon that occurs as from existing treatment systems and as a results of
storage and retention of water in the distribution systems. If residence time in the systems is
significant, then a large portion of the influent radon would decay before reaching the consumer.
Characterizing the differences (if any) between the NIRS and state supplemental data has
important implications for estimating the national proportions of systems exceeding potential
regulatory levels, as discussed in detail in Section 5.8.
Methods, Occurrence, and Monitoring Document for Radon
5-42
-------
5.6.1 Comparison of Log Mean Radon Levels Between NIRS and Supplemental
Data Sets
We first compared the log mean radon levels estimated from the NIRS data with those
estimated from the supplemental data sources for the same state and system size strata. In this
analysis we used the log mean radon levels (the log of the geometric mean), rather than the
geometric means, because, with the underlying lognormal distributions of the data, the log mean
values provide a consistently-scaled basis for comparison. The ratios of the geometric means derived
from the NIRS data to those derived from the supplemental data were calculated for all states and
strata covered by both data sources.1 The results of this analysis are summarized in Exhibit 5-21.
For each size stratum, the individual NIRS/state log mean ratios are shown, along with the average
ratios for all of the states. For all of the size strata, the average log mean ratio is close to 1.0. For
"All" systems, the average ratio is 0.97 (the NIRS log means are lower than the corresponding state
log means). The NIRS/state log mean ration for the very very small systems is very similar, at 0.98.
The log mean ratios decrease further as system size increases, to 0.97, 0.95, and 0.93 for the very
small, small, and medium systems, respectively. For the large systems (not shown) only one state
had both sufficient NIRS data and state data to conduct a comparison. In this state (Texas), the ratio
of NIRS to state geometric mean values was 1.08, reversing the
8
8
1.7
1.5
1.3
1.1
0.9
0.7
0.5
Exhibit 5-21. Ratios of Log Mean Values far Individual States
NfftS Versus Supplemental State Data
4
A
Average = 0.97
A
*
i
A
A
A
Average = 0.98
A
A
•*•
A
1
A
Average = 0.97
— = Average Ratio
A = Individual Ratios
4
I
Average= 0.95
.*
Average = 0.93
ALL
WS
VS
System Size
s
M
1 In estimating the NIRS/state log mean and log standard deviation ratios for the
individual states, we eliminated all states and size strata from which five or fewer systems
reported radon values in either the NIRS or the supplemental data sets. The ratios for the strata
with fewer data point tend to have a much higher degree of variability than those having more
data and we did not consider the ratios calculated for these strata to be reliable.
Methods, Occurrence, and Monitoring Document
5-43
-------
pattern of the NIRS log means being generally lower than those from the supplemental data.
The average ratios of state to NIRS log mean radon levels are tabulated in the top row of Exhibit 5-
22, along with the 95 percent confidence intervals on the ratios, derived using the bootstrap method
described in Section 5.2.4. The ratios of the state to NIRS state-wide log means are significantly less
than 1.0 at p = 0.05 when all systems are included in the calculation and for the very small, small, and
medium strata. The state/NIRS ratios for very very small systems were considerably more variable,
and the average state/NIRS ratio for this size category is not significantly different from zero.
Exhibit 5-22. Ratios of State Log Mean and Log Standard Deviation Radon Levels to NIRS
Values (Confidence Limits on the Ratios)
State Log Mean
Radon/NIRS Log
Mean Radon
State Log
Standard
Deviation
Radon/NIRS Log
Standard
Deviation Radon
All
0.97 (1)
(0.96 - 1.00)
1.05
(0.95 - 1.14)
Very Very
Small
0.98
(0.87-1.09)
1.04
(0.85- 1.20)
Very Small
0.97 (2>
(0.92- 1.01)
0.82 (1)
(0.65 - 0.96)
Small
0.95 (1)
(0.93 - 0.98)
0.94
(0.78-1.10)
Medium
0.93(1)
(0.92 - 0.93)
0.84
(0.51 - 1.17)
Notes:
1. Different from 1.00 at p<= 0.05
2. Different from 1.00 at p<= 0.10
The results in Exhibits 5-21 and 5-22 suggest that, on the whole, the geometric means
estimated from the state data are similar to, but lower than, those estimated from the supplemental
data. Within the limitations of the data, there seems to be a trend for the NIRS/supplemental log
mean ratio to decrease with increasing system size. For very very small systems, the average ratio is
close to unity (0.98), whereas the average ratio decreases consistently across the next three size
strata to a value of 0.93 for the medium systems. This means that, as systems get smaller, the NIRS
log means get progressively lower compared to the supplemental data. The continuation of this
trend through the large systems cannot be confirmed (based on these data) because of the lack of
data on NIRS/State ratios for this size system.
The pattern of differences between the radon levels seen in the NIRS and supplemental data
is consistent with a simple physical explanation. First, the radon levels in the smallest systems (very
very small and very small) would be the most similar across the two data sets because:
Methods, Occurrence, and Monitoring Document
5-44
-------
• The NIRS very very small systems would draw from the same relatively low-yield, high-
radon aquifers as the very very small systems in the supplemental data sets;
The NIRS very very small systems are less likely to have treatment systems in place to reduce
radon levels than the larger systems; and :
• The. NIRS very very small systems would have short residence times, and relatively little
radon decay would occur. ;
The combined effect of these fectors would be to reduce the differences between the NIRS
point-of-use samples and the wellhead and point-of-entry samples from the state supplemental data
sets. This would explain why the average (and median) NIRS/state log mean should be close to 1.0
for the smallest systems. ;
In contrast, the radon levels in the larger NIRS systems would be lower than that seen in the
state data sets for the same size systems because:
• A larger proportion of the larger NIRS systems would tend to have treatment in place; and
• The residence times in the larger NIRS systems is appreciable compared to the half-life of
radon. ;
These factors would decrease the average ratio of the NIRS (in-system) to the supplemental
(wellhead) radon log mean ratios, consistent with the observed pattern. These observed differences
between the NIRS and state data are used in the development of national distribution of radon level
in groundwater sources, as discussed in Section 5.8.
5.6.2 Comparison of Log Standard Deviations
The question also arises whether the NIRS data adequately capture the variability of the
groundwater radon levels across the U.S. This issue was addressed by comparing the log standard
deviations estimated from the NIRS data and those estimated from the state data. These
comparisons are shown graphically in Exhibit 5-23, and their confidence limits are shown in the
bottom row of Exhibit 5-22. For all systems combined, the average NIRS/state log standard
deviation ratio was 1.05, indicating a similar average degree of variability in the two data sets. The
average log standard deviation ratio for the very very small systems was also close to unity (1.04.)
For the larger system sizes, the average ratios of log standard deviations in the state and
NIRS data sets are smaller and more variable. For very small, small, and medium systems, the ratios
are 0.82, 0.94, and 0.84, respectively, indicating that the NIRS data are, on average, less variable
than the state supplemental data for these size strata. For Texas, the only state for which a NIR/state
log standard deviation ratio can calculated for large systems, the value is 0.37.
Methods, Occurrence, and Monitoring Document
5-45
-------
Exha* 5-23. Ratios of MRS to State Log Standard Deviations
!••
1.6
1.4
1.2
1.1
I.S
8.4
• 7
*
f
*
— = Mean
A"
A
A
^
A
4
A = State Ratios
i
A
A"""
.4-.
A
A
i
JA.
i
A
t
ALL
ws vs
System Size
The degree of variability in state/NIRS log standard deviation ratios for the different size
strata is greater than that seen for the log mean ratios. Although the average ratios are smaller than
1.0 for all three of the smallest size categories, only the ratio for very small systems achieves
statistical significance at p = 0.05. These data are also used in the development of the national
radon distributions for groundwater sources in Section 5.8.
5.7 Sources and Magnitude of Variability in Groundwater Radon Levels
Until this point, the focus has been on the variability of long-term radon levels in individual
systems and in sources serving these systems. This has been the case for two reasons. First, the
NIRS, which is far and away the most comprehensive data source on radon levels in U.S. water
supplies, comes from a cross-sectional survey, designed to gather data typical of systems, rather than
sources. Second, a primary goal of the occurrence analysis is to determine the proportions of
systems that might be above potential regulatory levels, and the number of customers served by these
systems, so that potential risks and risk reduction can be assessed.
However, as noted in Section 5.1, EPA has received numerous comments regarding the use
of the NIRS data in the characterization of radon levels across the U.S. These comments point out
that the previous occurrence analysis did not address important sources of variability in radon levels,
which might strongly effect the estimates of the numbers of systems affected by radon regulation.
The following sections discuss EPA's analysis of these issues using the NIRS and supplemental state
data sets.
Methods, Occurrence, and Monitoring Document
5-46
-------
5.7.1 Identification of Sources of Variability in Radon Levels ^
When a radon measurements are taken from different sources and systems, the distribution of
the results is affected by a number effectors. For purposes of this analyses, we use the following
terms to identify these sources of variability:
• Var(SYS) — This symbol represents the "true" inter- system variability in radon levels, for
example the differences in radon levels due to differences in location, geologic setting, size,
etc., between a small system in New Hampshire, and a large system in Nebraska.
• Var(W) — This is the variability among different sources (wells) within a given system.
Where a system obtains water from only one well (as do many of the smallest systems), this
source does not contribute to the overall variability in radon observations. This source of
variability is often referred to as "intra-system" variability, although variability in radon levels
within a single system also includes contributions from the following sources as well.
• Var(T) — This fector is the time (temporal) variability of radon levels in a given source or
system. For purposes of this analysis, we use this symbol to refer to differences in radon
levels seen of times scales of greater than one day.
• Var(S) — Refers to all variability associated with the process of taking samples, transferring
them to the laboratory, and performing any other manipulations up to the point at which the
sample enters the scintillation counter. Often, it is not possible to separate this source of
variability from analytical variability, below. \
• Var(A) — The last fector we address is variability associated with the analysis of radon
samples themselves. This includes statistical counting error, as well as any other factors
affecting the precision of the radon analysis. As will be discussed further below, the
magnitude of this variability is measured by examining the differences in the results of
duplicate analyses of the same sample.
By characterizing all of these sources of variability in a given data set, it is possible to predict
the proportions of sources (wells) that would exceed regulatory levels over time, and the numbers of
systems that would be affected by the variability in well radon levels.
5.7.2 Estimating Contributions to Variability
t*
The sources of variability discussed above contribute to the total variability in different data
sets in different ways. Sampling and analytical variability can only be estimated where duplicate
samples are taken; intra-well variability only affects systems that obtain water from more than one
well, etc. The following analysis makes use of a simple generalized model for the variability of radon
to estimate contributions from each of these sources to the overall variability in radon levels.
Methods, Occurrence, and Monitoring Document
5^47
-------
The model that is employed is a additive variance model (also referred to as an ANOVA
model). This model assumes that the sources of variability in any radon data set all act independently
of one another. Under this assumption, it can be shown that the total variance in radon levels can be
expressed as the sum of the contributions to variance from all the individual sources:
Var(Total) = Var(Sys) + Var(W) + Var(T) + Var(S) + Var(A),
(5-13)
where the symbol Var( ) mean "variance due to" the sources which were defined above. An ideal
data set (which does not exist) would consist of a large number of samples over time from a large
number of wells in a large number of systems, with plentiful sampling and analytical duplicates. Such
a database would allow the estimation of the contributions to variance from each source. Because
none of the available data sets meet this ideal, it is necessary to develop estimates of the
contributions to variance from different sources using different data sources. This analysis is
described in detail in Appendix C. The results of the analysis of variance are discussed below.
5.7.3 Magnitude of Contributions to Radon Variability
i
Using the data from a number of studies, the relative contributions of the different sources to
the overall variance in a "typical" radon data sets were estimated. The "typical" data set was defined
as one which includes radon measurements from multiple systems and sources, sampled over a
significant period of time (months to years) from systems located in a region with geological
diversity equivalent to that of an average state data set.
As described in Appendix C, different studies were used to develop ranges of estimates of
total variance, and of the contributions to variance from the individual sources and combinations of
sources. The results of that analysis are summarized in Exhibit 5-24. The average total log variance
seen in the radon data sets that were examined was on the order of 1.35 (top row of the Exhibit).
Combined sampling and analytical variability, VAR(S+A) was found to be quite small in a number of
data sets, typically contributing the equivalent of 0.6 percent to the total variance. Two estimates of
temporal variability, Var(T) were developed using the equivalent of Equation 5-13 on the results
from two different combinations of studies. One estimate was derived by subtracting estimates of
sampling and analytical variability from estimates of combined sampling, analytical, and temporal
variability, Var(S+A+T). The second estimate was derived by subtracting estimates of combined
sampling, analytical, and between-well variance from estimates of combined sampling, analytical,
between-well and sampling variance. In the first case, the estimated contribution of temporal
variability to overall log variability was 0.19, while the second method resulted in an estimate for
temporal variance of about 0.14.
Typical values for the individual contributions of variations among wells and among systems
were also calculated using the relationships shown on the last two rows of Exhibit 5-24. Between-
well variance was found to account for between 12 and 17 percent of the variance, while variations
between systems, as expected, accounted for the bulk (69 percent) of the total variance.
Methods, Occurrence, and Monitoring Document
5-48
-------
Exhibit 5-24. Estimated Contributions to Variance in a Typical Radon Data Set
Source of Variance
Total (All Sources)
Sampling and Analytical
Temporal
Among Wells
Among Systems
Typical Magnitude of Contribution to Log
Variance (Method of Estimation)
— 1.35 (Measured)
~ 0.009 (Measured)
(S+A+T)-(S+A)~0.19
(S+A+T+W) - (S+A+T) ~ 0. 1 4
(S+A+W+T) - (S+A+T) -0.16
(S+A+W) - (S+A) - 0.23
(SYS+S+A+W+T) - (S+A+W+T) - 0.93
(SYS+S+A+W+T) - (S+A) - (W) - (T) - 0.93
Typical
Proportion of
Total Variance,
percent
100
-0.6
13-18
12-17
-69
These results are used in the following sections to help estimate the proportions of systems
above specified radon levels and the effectiveness of monitoring programs in identifying non-
complying sources.
5.8 Estimates of Numbers of Groundwater Systems Exceeding Potential Regulatory
Limits
In this section, we use the distributional approaches discussed in the previous sections to
estimate the proportions of community water systems exceeding potential regulatory levels. Radon
distributional parameters are then developed for eight regions using NTRS data and for 7 states using
data from the supplemental data sources. These distributions are combined to provide an estimate of
the total numbers of systems exceeding potential regulatory levels in the U.S. Potential uncertainties
associated with these estimates are then discussed, and the current estimates of radon exceedences
are compared to the results of previous EPA occurrence analyses.
5.8.1 Characterizing Radon Distributions for States and Regions
This section describes the approaches used to estimate the proportions of systems with radon
levels above potential regulatory levels. The focus of the section is on the development of estimates
of log mean and log standard deviation values that can be used, at the end of the section, for
estimating these proportions using the lognormal model. A large proportion of the discussion is
centered on adjusting the calculated values to account for difference between samples taken in the
distribution system (NIRS) and expected radon levels in influent water, and for differences in
Methods, Occurrence, and Monitoring Document
5-49
-------
variance that would be associated with different monitoring and averaging approaches to estimating
system-wide radon levels.
The general process used to estimate the proportions of systems above potential regulatory
levels (MCLs or AMCLs) is summarized in Exhibit 5-25. Radon occurrence data from NIRS or the
state supplemental data systems (upper left corner of the Exhibit) form the basis for the occurrence
estimates. The NIRS data are used to calculate regional estimates of log mean and log standard
deviation radon levels for each size category of system.
i
The eight regions are those defined in Section 5.1, except that the radon data for the states
with good supplemental data (see below) are excluded.2 In these states (Michigan, New Hampshire:,
New York, Pennsylvania, South Carolina, Texas, and Wisconsin), "raw" state log mean and standard
deviation values are also calculated from the supplementary data sets.
The first step in the process of estimating the proportions of systems above potential
regulatory levels is to adjust the regional log means derived from NIRS data upward to take into
account the difference between the in-system samples and point-of-entry samples discussed in
Section 5.6 The ratios of the NIRS to state log mean and log standard deviation shown in Exhibit 5-
21 and 5-23 are used to calculate adjusted log mean and log standard deviation values
(corresponding to the estimated radon distributions in groundwater sources) for the region where the
NIRS is the primary source of data. The "adjusted" NIRS regional log means are used as inputs to
the lognormal model to estimate the proportions of systems above regulatory levels in the eight
regions. Because the state data sets include only data from groundwater sources and points of entry,
adjustment for in-system sampling is not necessary.
For all data sets where the estimates of log standard deviation are based on single samples
from different systems (as in the NIRS and several of the state data sets), the log variance of radon
levels needs to be reduced to provide a representative estimate of log standard deviation. This is
because, when multiple samples are taken over time (as would be required under EPA's proposed
monitoring scheme), the temporal, sampling, and analytical variance in radon levels would be
canceled out to a large extent. The degree of adjustment in the log variance that is required to
account for multiple sampling is derived from the analyses of variance discussed in Section 5.7
The output of these various adjustment processes are sets of log mean and log standard
deviation estimates for the seven states and eight NIRS regions. These estimates form the basis for
the estimates of the proportions of systems exceeding potential regulatory levels. The proportions
are estimated using the assumption of lognormality of radon levels ("the lognormal model").
2 In addition, the 491 community groundwater systems in Alaska have been added to the
Northwest region totals. Community water systems from Hawaii are excluded because no data
related to radon levels in groundwater in Hawaii were identified.
Methods, Occurrence, and Monitoring Document
5-50
-------
to
3
to
>»'•£; 9? J52
giSS9>
*|E<3
S^ g *
*-»li
2x^-5
l-gs?
c
(0
**- Xv
o Q^
-g o O
< UL D. Q
o: g) co o
— O -*1* '.S
co IS
-------
The final step in the analysis is to multiply the proportions of systems above the various radon
levels by the total numbers of systems in the various states, regions, and size strata. This process
provides stratified estimates of the occurrence of radon in community water systems in the U.S.
a. Regional Estimates of Log Mean and Log Standard Deviations Using Adjusted
NIKS Data
As noted in Section 5.6, the NIRS log mean radon levels in most states are slightly lower
than those from the state data sets for the same size systems. The reason for the differences are most
likely that the NIRS samples were taken from within the distribution systems, while the state data
which we evaluated all came from wellhead or point-of-entry samples. Thus, for purposes of
identifying water systems where influent radon limits are above potential regulatory levels, the radon
levels seen in the NIRS data need to be adjusted to account for differences between in-system and
influent water analyses.
To adjust the NIRS log mean values for the various regions and size strata, the average
national NIRS/state log mean ratio for each size system was multiplied by the NIRS regional log
mean. The NIRS/state ratios of the log geometric mean values are summarized in Exhibit 5-21.
Log mean point-of-entry radon levels for given size system in a specific region was estimated as:
= NIRS
/AF(LM)
(5-12)
where:
AF(LM) = Adjustment factor for log means = national average ratio of NIRS to state
log means for the system size category.
The same adjustment factor was thus used for all systems in a given size category across all regions.
While the NIRS/state log mean ratios varied across the states and across regions of the country,
there were insufficient data to allow the derivation of adjustment factors for each region.
The calculations to adjust the log means derived from the NIRS data are summarized in
Exhibit 5-26. The adjustments increase the log means only slightly, since the adjustment factors are
mostly close to 1.0. However, because these adjustments are carried out in "log space," they result
in appreciable changes in the estimated geometric means. For example, as shown in the Exhibit, the
unadjusted NIRS log mean for medium systems in the Appalachian region is 5.10, corresponding to a
geometric radon level of 163 pCi/1. When the adjustment factor of 0.93 is applied, this value
increases to 5.49, corresponding to a geometric mean radon level of 242 pCi/1, a 48 percent increase.
Owing to the small number of large systems sampled in some regions, the overall level of
uncertainty associated with the log means in these regions is greater than that for the other systems.
Because there are only 29 large systems in the entire NIRS database, there are no data for NIRS
large systems in several regions. For this reason, the national NIRS log mean value for large systems
was used as the estimate of the NIRS large system log mean in all the regions. Similarly, there are
Methods, Occurrence, and Monitoring Document
5-52
-------
only a few medium systems in the New England and Northwest regions. Thus the log mean for these
regions is estimated as the national mean of the log means for this size category of system.
Exhibit 5-26. Adjustment Of NIRS Regional Log Means to Adjust for Point-of-Entry
Versus In-Systems Sampling Using NIRS/Suppiemental Radon Ratios
Unadiustec
Reaion
Appalachian
California
Gulf Coast2
Great Lakes
New Enqland
Northwest
Plains
Rocky Mountains
ALL
5.73
5.81
4.69
4.91
7.00
5.08
4.84
5.89
Reaional Radon Loa Mear
WS
5.81
5.88
5.49
5.04
7.28
5.07
5.07
6.33
VS
6.15
6.26
5.10
5.14
7.03
5.12
4.99
5.42
Values. NIRS Data
S
5.06
5.59
3.78
4.66
6.17
4.71
4.31
5.84
M
5.10
4.98
3.65
4.94
6.47 7
4.72 7
5.04
5.51
L
487
4.87
4.87
4.87
4.87
4.87
4.87
4.87
Adjustment Factor
for Log Means6 0.97 1 .00 0.97 0.95 0.93 0.94
Rea onal Radon Loa Mean Values. Adjusted for NIRS/State Differences
Reaion
Appalachian
California
Guff Coast
Great Lakes
New Enaland
Northwest
Plains
Rocky Mountains
ALL
5.93
6.01
4.85
5.08
7.25
5.26
5.01
6.10
WS
5.80
5.87
5.47
5.03
726
5.06
5.06
6.31
VS
6.35
6.46
5.26
5.30
7.26
5.28
5.15
5.59
's
5.33
5.89
3.99
4:.91
6.51
4:97
4.54
6,16
M
5.49
5.36
3.93
5.32
6.97
5.08
5.43
594
L
5.18
5.18
5.18
5.18
5.18
5.18
5.18
5 18
Notes:
1. Omitting Pennsylvania, South Carolina
2. Omitting Texas
3. Omitting Wisconsin
4. Omitting New Hampshire
5. Omitting Michigan
6. National average NIRS/state log mean rations from Exhibit 5-21
7. Log standard deviation for all large systems is used because of small number of data points.
Methods, Occurrence, and Monitoring Document
5--S3
-------
In addition to the log mean, the log standard deviation also affects the proportions of systems
above given MCL/AMCL levels. Consistent with the approach used for log means, we adjusted the
NIRS log standard deviation values to be more consistent with those seen in the state databases. The
latter presumably better reflect variability in groundwater sources than do the in-system NIRS data.
i
I
To adjust the NIRS regional log standard deviation values, we used the arithmetic average
log standard deviation ratio as follows:
Log Std. Dev.
where:
size) = NIRS Log Std.
/ AF(LSD)
(5-13)
AF(LSD) = Adjustment factor for log standard deviation of in-system radon
level = national average ratio of NIRS to state log standard deviations for
the appropriate system size category.
As shown in Exhibit 5-23, the ratios of the NIRS to state log standard deviations are more '•
variable than the rations of the log mean values ranging from as low as approximately 0.45 to
greater than 1.6 across individual states. The adjustment factors used in this analysis are ranged
between 0.82 and 1.05, however.
The application of the adjustment factors to the log standard deviations of the regional radon
levels is summarized in Exhibit 5-27. The top panel provides the regional log standard deviations
calculated directly from the NIRS data and the bottom panel of the table shows the adjusted log
standard deviation values. The adjustments for many of the regions and system strata are
proportionally larger than those for the log means, and result in both increases and decreases in the
estimated variability in point-of-entry radon levels compared to in-system
values.
Similar to the case for the log mean values, the national NIRS log standard deviation for
large systems was used to estimate adjusted log standard deviations for all of the regions. In
addition, the adjusted log standard deviations for medium systems in New England and the
Northwest were estimated to be equal to the national average log standard deviations due to a lack
of data.
The adjusted log standard deviations tabulated in the bottom panel of Exhibit 5-27 are those
that would be expected to occur assuming single sample were taken from a single well in each
system. As noted above, these values need to be further adjusted if compliance status is to be
evaluated under a monitoring scheme involving multiple samples, because such a scheme reduces the
variance in the estimated radon levels for a given system.
The general approach we have taken to adjusting the log variance estimates (and therefore
Methods, Occurrence, and Monitoring Document
5-54
-------
Exhibit 5-27. Adjustment Of NIRS Regional Geometric Standard Deviations for
Differences Between Point-of-Entry and In-System values Using
NIRS/Supplemental LSD Ratios
Unadjusted Regional Radon Log Standard Deviation Values, NIRS Data
Reaion
Appalachian1
California
Gulf Coast2
Great Lakes3
New Enqland4
Northwest
Plains5
Rocky Mountains
Adjustment Factor for
Log Means8
ALL
1.67
1.13
1.46
1.115
1.34
0.80
0.93
1.02
1.05
WS
1.51
1.22
1.10
0.860
1.18
0.76
1.03
0.93
1.04
VS
1.85
0.93
1.31
1.318
1.66
0.75
0.73
1.20
0.82
S
1.58
0.61
1.78
0.925
0.52
1.32
1.21
0.85
0.94
M
0.56
0.72
1.26
0.923
0.776
0.776
0.57
0.61
0.84
L
0.877
0.87
0.87
0.87
0.87
0.87
0.87
0.87
0.84
Regional Radon Log Standard Deviation Values, Adjusted for NIRS/State Differences
Region
Appalachian
California
Gulf Coast
Great Lakes
New England
Northwest
5lains
Rocky Mountains
ALL
1.598
1.079
1.392
1.066
1.279
0.767
0.885
0.974
WS
1.458
1.174
1.060
0.829
1.137
0.734
0.990
0.899
VS
2.259
1.133
1.601
1.608
2.022
0.917
0.886
1.463
S
1.686
0.652
1.896
0.987
0.558
1.404
1.296
0.907
M
0.669
0.852
1.498
1.098
0.922
0.922
0.680
0.724
L
1.044
1.038
1.038
1.038
1.038
1.038
1.038
1.038
Notes:
1. Omitting Pennsylvania, South Carolina
2. Omitting Texas
3. Omitting Wisconsin
4. Omitting New Hampshire
5. Omitting Michigan
6. Log standard deviation is average of that for other regions (no data).
7. Log standard deviation for all large systems is used because of small number of data points.
8. National average NIRS/state log standard deviation ratio, from Exhibit 5-23
log standard deviation estimates) form multiple samples is illustrated in Exhibit 5-28. The three
Methods, Occurrence, and Monitoring Document
5-55
-------
Exhibit 5-28. Reductions in Variance Achieved by
Combining Multiple Sample Results (Data from California,
New York)
CA (S,A,W,T)
CA (S,A,T)
NY (S,A,T)
0.00
ONE
TWO Tl-REE FOUR
Number of Samples
FIVE
curves in the graph represent the estimated log variance contributions from some of the different
sources discussed in Section 5.7, and how they would be reduced by taking multiple samples. These
curves were derived by Monte Carlo simulation of sampling from the distributions of Var(S),
Var(A), Var(W), and Var(T), or combinations thereof. In each iteration of the simulation, the results
of one, two, three, four, or five samples were averaged (to simulate various potential monitoring
schemes), and the log variance of the averages were calculated.
I
The upper curve represents the simulated results of taking one or more samples from a
lognormal distribution with a variance equal to the average (S+A+W+T) variance for the California
data discussed in Appendix C. Taking many single samples from this population of analyses will
have a combined total (S+A+W+T) log variance of approximately 0.33. As more than one sample is
taken (equivalent to taking multiple samples from randomly selected wells at random times), the
variance is reduced, with a greater degree of reduction as more samples are taken and averaged
together.3
3 If the system average log variance or log standard deviation were calculated by taking
the geometric mean of multiple sample results instead of the arithmetic mean, then the residual log
variance would be exactly equal to (1/n) times the population log variance, where n is the number
of samples. In fact, the results come quite close to this ratio.
Methods, Occurrence, and Monitoring Document
5-56
-------
The middle curve on the graph is the similar procedure repeated again for supplemental data
from California, but in this case starting with only the combined analytical, sampling, and temporal
variance (S+A+T) Finally, the bottom curve is a simulation of multiple sampling of a population of
radon analytical results with a combined log analytical and sampling (S+A)variance equal to that seen
in the supplemental data from New York( Appendix C).
The first step in adjusting log variances for multiple sample averaging was to develop,
implicitly, curves like those in Exhibit 5-28 for all of the data sets that were evaluated. After
separating out the true system variability from the population of reported levels (Var(SYS)), the
calculated log standard deviations (and variances) fall somewhere on one of the three curves shown
in this Exhibit. The calculated log variance from the NIRS data, composed of single samples from
multiple systems, fell at the upper left-hand end of the top curve. The summary statistics from some
of the states, calculated from multiple analyses of multiple samples from multiple wells, correspond
to points near the lower right-hand end of the upper curve.
The following specific rules of thumb were used to estimate state and regional log variances
from the NIRS and supplemental data sets:
• NIRS regional average log standard deviations were derived assuming that averaging four
samples would reduce the total (S+A+T+W) variance by 75 percent. Using the estimate
from Appendix C that (S+A+T+W) variance accounts for an average of 30 percent of the
total variance in a typical population of radon measurements, this means that the observed
NIRS variances were reduced by 22.5 percent to give the adjusted estimates of variance
assuming four samples. For the NIRS data, this works out to a reduction in overall variance
of approximately 13.5 percent.
• For state data sets composed of single samples from multiple systems, the same approach
used for the NIRS data was employed to estimate log standard deviation. For several of the
states, minor corrections first had to be made to account ;for the variance reduction achieved
by taking duplicate samples or conducting duplicate analyses.
• For state data composed of results from multiple wells and/or multiple samples over time
from individual wells, the state log standard deviation estimates as calculated in Section 5.6
were used directly as estimates of log standard deviations. It is recognized that this approach
captures different degrees of variance reduction for different states, depending on the designs
of the sampling programs and numbers of samples taken from individual systems in the
various states.
The resulting estimates of log standard deviations for the NIRS regions are shown in Exhibit
5-29. As noted above, the final log standard deviation values are approximately 13.5 percent lower
than the total variances shown in Exhibit 5-27. The estimates of adjusted log mean values shown in
Exhibit 5-26 and the doubly adjusted log standard deviation values shown in Exhibit 5-29 were used
to estimate the proportions of systems above potential radon regulatory limits later in this section.
Methods, Occurrence, and Monitoring Document
5-57
-------
EXHIBIT 5-29. NIRS Regional Log Standard Deviation Estimates, Adjusted for Averaging of Multiple
Samples
Region
Appalachian
California
Gulf Coast
Great Lakes
New England
Northwest
Plains
Rocky Mountains
System Size
ALL
1.407
0.950
1.226
0.939
1.126
0.675
0.780
0.858
WS
1.284
1.033
0.933
0.730
1.001
0.646
0.871
0.792
VS
1.989
0.997
1.410
1.415
1.780
0.807
0.780
1.288
S
1.484
0.574
1.670
0.869
0.492
1.236
1.141
0.798
M
0.589
0.750
1.318
0.966
0.812
0.812
0.598
0.638
L
0.914
0.914
0.914
0.914
0.914
0.914
0.914
0.914
b. Estimates of Log Mean and Log Standard Deviation Values for States With
Sufficient Supplementary Data
As noted in Section 5.8.1, seven of the state data sets provided sufficient data such that the
state log mean and log standard deviation radon levels could be developed independent of the NIRS
results. For these states, the proportions of systems exceeding potential regulatory levels were
estimated from the state data sets.
All of the states that provided detailed and representative data reported sampling results from
the wellhead or points of entry to the water distribution system. Thus, it was not necessary to adjust
the state log mean radon levels as was done for the NIRS data. Thus, the log mean values that were
used were the same as those shown in Exhibit 5-18, and summarized again in Exhibit 5-30.
Exhibit 5-30. Log Mean Radon Values from State Data Sets
State
MICHIGAN
NEW HAMPSHIRE
NEW YORK
PENNSYLVANIA
SOUTH CAROLINA
TEXAS
WISCONSIN
System Size
ALL
5.23
7.61
5.76
6.21
6.30
4.94
5.69
WS
4.96
7.83
5.67
6.14
7.61
5.27
5.56
VS
5.21
7.81
5.85
6.19
7.13
4.96
5.75
S
5.35
6.52
5.77
6.30
5.63
5.02
5.81
M
5.12
6.39
5.74
6.20
5.78
4.93
5.61
L
5.31
5.80
5.35
6.35
5.15
4.52
5.63
Methods, Occurrence, and Monitoring Document
5-58
-------
The state log standard deviations likewise do not need to be adjusted for the differences between in-
system and point-of-entry sampling. However, they do need to be adjusted for the reduction in
variance associated with the averaging of multiple samples. As vyas the case for the NIRS data, it
was assumed that the arithmetic average of four samples would be used to determine compliance
with potential regulatory levels. The adjustments that were made to the various state log standard
deviations are shown in Exhibit 5-31. These, along with the log mean values shown in Exhibit 5-30,
are used in Section 5.8.4 to develop estimates of the numbers of systems above potential regulatory
levels in seven states.
Exhibit 5-31. Estimates of State Average Log Standard Deviations Adjusted for Multiple
Sampling
Log Standard Deviation (Unadjusted)
State
MICHIGAN
NEW HAMPSHIRE
NEW YORK
PENNSYLVANIA
SOUTH CAROLINA
TEXAS
WISCONSIN
System Size
ALL
0.777
1.334
1.205
1.437
1.537
0.947
0.821
WS
0.834
1.180
1.214
1.438
0.959
0.872
0.870
VS
0.956
1.228
1.463
1.632
1.563
0.953
0.865
S
0.722
1.602
1.007
1.282
1.335
1.053
0.825
M
0.631
0.836
0.880
1.343
1.458
0.815
0.660
L
0.619
1.640
0.837
1.082
1.044
0.657
0.417
Loa Standard Deviation (Adjusted for Multiole Sarrmlina)
State
MICHIGAN
NEW HAMPSHIRE
NEW YORK
PENNSYLVANIA
SOUTH CAROLINA
TEXAS
WISCONSIN
System Size
ALL
0.778
1.343
1.205
1.437
1.537
0.947
0.826
WS
0.839
1.209
1.214
1.438
0.959
0.873
0.445
VS
0.960
1.255
1.463
1.632
1.563
0.954
0.879
S
0.724
1.610
1.007
1.282
1.335
1.053
0.830
M
0.633
0.850
0.880
1.343
1.458
0.816
0.666
L
0.622
1.647
0.837
1.082
1.044
0.658
0.427
5.8.3 Numbers of Community Water Systems in the U.S.
All of the data sets evaluated thus far have been samples from larger population of
groundwater systems. The NIRS is a representative sample of in4system water from approximately
1,000 of the approximately 40,000 community groundwater systems in the U.S, and the state data
Methods, Occurrence, and Monitoring Document
5-59
-------
*.. I
sets, with one exception4, likewise represent relatively small samples of the total populations of
groundwater systems in the states. Therefore, it is necessary to extrapolate the proportions of
systems across the states and the regions that are above regulatory levels to the entire populations of
those states. Data regarding the total numbers of active community groundwater systems in the U.S.
were taken from EPA's Drinking Water Baseline Handbook (EPA 1999), and are summarized in
Exhibit 5-32. The numbers in this table represent the total groundwater systems in each size
category in each state, excluding systems that purchase their groundwater.
i
The data in Exhibit 5-32 show that there are a total of 40,812 active groundwater systems in
the U.S. (excluding Hawaii, for which radon levels were not estimated because no data on radon
levels in groundwater were identified). Consistent with previous estimates, the great majority of the
systems fall in the smallest size categories. The very very small and very small systems each account
for approximately 34 percent of the national total (13,687 and 13,860, respectively). There are
39,389 systems serving less than 10,000 people, amounting to approximately 91 percent of the total
systems. (For purposes of regulatory analysis, EPA defines these systems as "small entities".)
5.8.4 Numbers of Community Water Systems Exceeding Potential Regulatory
Levels
Exhibit 5-33 provides estimates of the numbers of community groundwater systems that
would be above potential regulatory limits, assuming the current distribution of radon levels and
assuming compliance is judged based on the arithmetic average of four duplicate samples. The totals
in this exhibit represent the summed results of calculations for the eight regions where NIRS data
were used to estimate systems above regulatory levels (30,354 systems) and for the seven states
(10,458 systems) where supplemental data sets were used to estimate the numbers of systems above
potential regulatory levels. Calculations for the individual states and regions are shown in Appendix
D.landD.2. \
I
The national proportions of systems predicted to exceed regulatory levels ranges from 76.7
percent (100 pCi/1) to 3.2 percent (4,000 pCi/1). The corresponding numbers of systems exceeding
these levels are 31,307 and 1,312, respectively. Approximately 42.5 percent (17,349) of the systems
are predicted to exceed EPA's preferred regulatory limit of 300 pCi/1, while, as noted above, 1,312
systems (3.2 percent) are predicted to exceed the AMCL values estimated by NAS. Because of their
generally higher radon levels, the two smallest system categories account for a disproportionate
share of the systems exceeding the higher radon levels. Very small and very very small systems
combined account for approximately 85 percent of the systems above 1,000 pCi/1, 91 percent of the
total systems above 2,000 pCi/1, and 94 percent of the systems above 4,000 pCi/1.
4 The data set from New Hampshire appears to present an almost complete census of
groundwater systems in that state. In fact, there are more systems reporting data from New
Hampshire than are identified in EPA's Baseline Handbook, probably because some of the
systems in the New Hampshire data are listed under more than one name or are no longer active.
Methods, Occurrence, and Monitoring Document 5-60
-------
Exhibit 5-32. Number of Community Groundwater Systems By Size and State
System Size
(Population
Served)
Total1
Grand Total
Alabama
Alaska
Arizona
Arkansas
California
Colorado
Connecticut
Delaware
Florida
Georaia
Idaho
Illinois
Indiana
Iowa
Kansas
Kentucky
Louisiana
Maine
Maryland
Massachusetts
Michiaan
Minnesota
Mississiooi
Missouri
Montana
Nebraska
Nevada
NewHamoshire
New Jersey
New Mexico
New York
North Carolina
North Dakota
Ohio
Oklahoma
Oreoon
Pennsylvania
Rhode Island
South Carolina
South Dakota
Tennessee
Texas
Utah
Vermont
Virginia
Washington
WestVirainia
Wisconsin
Vyominq
25-100
13687
40812
3
231
235
40
1098
233
238
89
509
604
340
184
200
214
82
32
235
171
148
115
345
159
72
380
256
112
103
325
113
189
815
895
38
260
82
317
683
22 '
230
103
30
889
81
137
528
942
100
402
78
101-500
13860
500-1,000
4247
1,001-3,300
5260
3,301-10,000
2335
10,001-
50,000
1222
50,001-
100,000
135
100,001-1
Million
64
> 1 Million
2
27
187
255
116
851
191
239
82
633
549
196
378
185
334
212
29
392
79
179
70
428
319
294
333
203
265
80
219
124
238
646
669
83
326
176
221
616
22
160
112
34
1185
124
144
482
735
84
275
79
33
44
95
71
193
42
20
24
181
98
41
203
120
161
83
6
135
20
47
19
127
147
266
128
31
88
22
29
44
54
164
111
32
117
66
38
162
1
46
39
22
429
35
32
81
114
38
141
7
107
20
77
104
213
46
25
20
263
132
38
217
168
165
111
22
240
29
45
37
168
168
365
154
33
81
28
31
71
48
159
95
44
170
102
54
177
4
55
34
44
607
51
20
62
142
39
162
13
78
(>
45
48
164
12
7
7
130
51
23
106
87
57
20
12
118
8
17
62
59
57
109
72
11
24
14
9
68
25
54
4!5
5
s:>
21
17
70
2
31
13
24
262
23
8
13
57
11
87
4
24
: 3
: 28
11
: 178
8
: 8
1 2
132
! 22
'' 10
48
33
15
i 14
5
32
2
1 20
46
15
' 50
35
22
2
9
1
4
; 65
16
44
24
3
57
13
12
25
4
9
3
14
: 60
: 10
; 1
7
40
6
29
1
1
0
2
3
47
0
0
0
32
2
0
1
2
3
0
1
3
0
0
0
3
1
0
2
0
0
0
0
4
1
3
1
0
6
0
1
0
0
2
0
1
5
1
0
0
3
0
4
0
0
0
0
0
21
0
0
0
19
1
0
2
1
1
0
0
2
0
0
0
1
0
1
1
0
1
0
0
0
1
5
0
0
3
0
0
0
0
0
o
1
0
0
0
0
2
0
1
0
0
0
0
0
0
0
0
0
1
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
1
0
0
0
0
0
0
0
Notes:
1. For purposes of this analysis,
Hawaii has been omitted, because no data related to radon levels in Hawaii were identified.
Methods, Occurrence, and Monitoring Document
5-61
-------
Exhibit 5-33. Estimated Numbers and Proportions of Community Groundwater Systems With Long-Term Radon
Above Potential Regulatory Levels
System Size
(Population Served)
VOTY Vorv Smalt
(2?-1QO)
V,n, Crr|a|t f 101-500)
Small;
5Q1-1.QQQ
1.QQ1-3.300
ttodium (3,301-
10,QOQ)
Large:
10.001-100.000
5-100.000
National Total
Proportion of Total
Systems
Total
Systems
13,687
13,860
4,247
5,260
2,335
1,357
66
40,812
100.0%
Systems
Above 100
oCi/l
11,464
10,808
2,849
3,409
1,728
999
49
31,307
76.7%
Systems
Above 300
oCi/1
6,549
6,478
1,452
1,694
769
388
19
17,349
42.5%
Systems
Above 500
oCi/l
4,381
4,608
858
991
384
178
8
11,408
28.0%
Systems
Above 700
pCi/1
3,248
3,602
554
636
227
95
4
8,366
20.5%
Systems
Above 1,000
uCi/1
2,289
2,734
325
370
128
45
2
5,892
14.4%
Systems
Above 2,000
pCi/1
1,047
1,536
103
116
40
9
0
2,852
7.0%
Systems
Above 4,000
DCi/l
415
817
32
36
11
2
0
1,312
3.2%
5.8.5 Comparison of Predicted Exceedences For States and Regions
I
The proportions of systems exceeding potential regulatory levels were estimated differently
for those states having good supplemental data from those that did not. Radon distributions in the
former states were evaluated using the supplemental state data, while the distributions in the other
states were estimated by adjusting the NIRS regional average values using the average NIRS/State
ratios. Thus, a comparison between the two groups of predictions can help to confirm the
consistency of the adjustment approaches for the two types of data.
The proportions of community groundwater systems predicted to be above regulatory levels
using the two data sources are compared in Exhibit 5-34. The predictions of proportions exceeding
regulatory levels derived using the state data are consistently somewhat higher than those derived
from the regional NIRS data for all system sizes. For the lower regulatory levels the two estimates
differ by about four to 4.5 percent, while the proportions predicted to be above the highest
regulatory levels differ by about 2.0-3.5 percent.
The systematic differences between the two sets of predictions appear to indicate a bias in the
approach used to adjust the NIRS data for the differences between raw and in-system data, or in
some other part of the estimation methodology. However, it is likely that a substantial part, if not
all, of the difference in exceedance proportions is driven by real differences in radon levels between
the seven states and the rest of the regions in which they reside. This is because the seven states
having supplemental data appear to be generally among those with above-average radon levels
compared to the rest of the country.
Methods, Occurrence, and Monitoring Document
5-62
-------
Exhibit 5-34. Comparison of Proportions of Systems
Above Potential Regulatory Levels Predicted Using
NIRS and State Data Sets
m
*" or>o/
C OU /o
'•&
OJ
0)
O
X Gr\o/
UJ 60% -
0)
5
U) A no/
>, 40% -
(0
«*-
O
4-1
0) ZU% •
S2
0)
Q.
no/
80.1%
75.6%\
V45-1"
•-*— State Supplemental
A PJIrXo r\Cylons
^v£\_^^ 22.9%
27.2%^^-\^^_ 17.0%
19.7%
^~1>-^^ 9.2%
— >C>~-^*. 4 5%
1 3. 5% ^^^J^rr----~^f
: 6.2
-------
Exhibit 5-35. Comparison of NIRS State Log Mean Levels to Corresponding Regional Log
Means for Seven States
State
Michigan
New Hampshire
New York
Pennsylvania
South Carolina
Texas
Wisconsin
NIRS Data State Log
Mean (all systems)
5.13
7.57
5.41
5.98
5.82
4.93
5.37
NIRS Regional Log
Mean l
4.84
7.00
5.73
5.73
5.73
4.69
4.91
Notes:
1. Regional averages omit states with supplemental data. Thus, they are different from regional
means in Exhibit 5-9.
Data concerning the numbers of NTNCWS active in the U.S. was obtained from EPA's
Drinking Water Baseline Handbook (EPA 1999). Data from this source provide a national total of
NTNCWS of 19,062 in 1998. A breakdown of the numbers of systems by state and size is given in
Exhibit E-l. The distribution of radon levels in NTNCWS systems was estimated using
supplementary data from five states (Maryland, Maine, New Hampshire, Texas, and New
Hampshire). No other states provided information on radon levels in NTNCWS.
National geometric mean radon levels for each NTNCWS size category were estimated using
the average ratios of the geometric mean levels in NTNCWS to the corresponding log mean levels in
community water systems in the five states. Log standard deviations were estimated for each size
category using the average of the log standard deviations seen in the state data sets for each size
category.
The numbers and proportions of NTNCWS exceeding potential regulatory levels were
estimated using the lognormal model, in the same manner as for the community systems. Only
national totals were estimated, as the data were too limited to support detailed regional estimates.
The results of the analysis are shown in Exhibit 5-36. The general pattern of results is similar to that
seen for the community water systems, except that the proportions of NTNCWS exceeding the
regulatory limits are greater, owing to the generally higher radon levels in the latter systems. The
great majority of NTNCWS (over 91 percent) are predicted to have long-term radon levels above
100 pCi/1, and 64.9 percent of the systems would exceed EPA's proposed regulatory level of 300
pCi/1. The proportions of NTNCWS exceeding potential regulatory levels then decline rapidly, until
Methods, Occurrence, and Monitoring Document
5-64
-------
only about 3.1 percent are predicted to exceed the NAS AMCL value of 4,000 pCi/1.
Exhibit 5-36. Estimated Proportions of Non-Community Non-Transient Groundwater Systems With Long-
Term Average Radon Above Potential Flegulatory Levels
State/Region: National
System Size
(Population
Served)
Very Very Small
(25-100)
Very Small (101-
500)
Small:
501-1,000
1,001-3,300
Medium (3.301-
10,000)
Large/Very
Large:
10,001-100,000
>1 00,000
Total
Proportion of
Total Systems
Total
Systems
9,606
6,840
1,891
665
53
7
0
19,062
100.0%
Systems
Above 100
pCi/l
9,137
6,095
1,550
545
45
6
0
17,377
91.2%
Systems
Above 300
pCi/l
6,687
4,511
842
296
26
Systems
Above 500
pCi/l
4,734
3,500
503
177
16
Systems
Above 700
pCi/l j
3,423 !
2,819 j
325 .
114 ;
11 !
Systems
Above 1,000
pCi/l
2,208
2,138
187
66
6
4
0
12,367
64.9%
2
0
8,932
46.9%
2 :
0 i
6,694 !
35.1%:
1
0
4,606
24.2%
Systems
Above 2,000
pCi/l
693
1,074
48
17
2
0
0
1,834
9.6%
Systems
Above 4,000
pCi/l
140
436
8
3
0
0
0
587
3.1%
5.8.7 Sensitivity Analysis of Estimates of Systems Exceeding Radon Levels
In this section, we describe a screening-level analysis of the uncertainty associated with the
estimates of the numbers of systems exceeding potential regulatory levels. A one-dimensional
Monte Carlo simulation method is employed to investigate the combined impacts of the various
sources of uncertainty and variability that affect these estimates, and to develop quantitative
estimates of the overall uncertainty of the estimated numbers of systems exceeding different
regulatory levels.
To implement the Monte Carlo model, the log mean and log standard deviation of the
various state/region/system size categories were modeled as independent random variables:
where:
LM
LM =
LSD =
(NIRS, State)
LM
(NIRS, state)
N(0,SEM)
LSD(NIRS,State)+ N(0,SESD)
(5-14)
(5-15)
N(0,SEM)
= The log mean radon level for the category of systems being
evaluated
= , A random normal variable with mean zero and standard
Methods, Occurrence, and Monitoring Document
5-65
-------
LSD,
(NIRS. State)
N(0,SESD)
deviation equal to the standard error of the log mean radon
The log standard deviation in radon levels for the
category of system being evaluated
A random normal variable with mean zero and standard
deviation equal to the standard error of the log standard
deviation for the category
The approximate normality of the log mean and log standard deviation estimates follows
from the central limit theorem. The approximate independence of these variables follows from
standard statistical theory, which shows that the sample mean and sample variance for non-
censored normal data are independent.
The log mean and log standard deviation values were estimated from the NIRS and state
supplementary' data shown in Exhibit 5-9 and 5-18. The standard errors of the log mean and log
standard deviations of radon levels were estimated as:
SE.\f
SESD =
vr
6
•N/2/7
(5-16)
(5-17)
where:
d
n
the log standard deviation for the state/region/system size category
the number of observations from which the log standard deviation was
estimated.
This approximation (for the standard error of the log mean and the log standard deviation) follows
from the standard large-sample approximation.
In performing the simulation, 5,000 estimates of log mean and log standard deviation were
derived for each set of systems evaluated by random sampling. For each the log mean and log
standard deviation estimates, the proportions of systems above regulatory levels were then
estimated, as described previously, using Equation 5-3. The numbers of systems in each
region/state/system size category exceeding regulatory levels were then estimated as the product of
the proportion above the regulatory limit times the numbers of systems in the various regions and
size categories, as described in Section 5.8.3.
Two sets of simulations were conducted. First, probability distributions were developed for
the total numbers of CWS above the various regulatory levels . The results of that analysis are
summarized in Exhibit 5-37. This Exhibit tabulates the percentiles of the cumulative probability
Methods, Occurrence, and Monitoring Document
5-66
-------
distribution of the estimated numbers of systems exceeding the regulatory levels from 100 pCi/1 to
4,000 pCi. Looking at the second column of the table, for example, it can be seen that the average
estimated number of systems exceeding 300 pCi/1 is 17,349 systems (consistent with Exhibit 5-33),
and that the 5th and 95th percentile estimates of total systems exceeding 300 are 16,713 and 17,999,
respectively.
Exhibit 5-37. Distributions of Monte Carlo Estimates of Total Community Systems
Above Radon Levels
Percentiles
0%
5%
10%
15%
20%
25%
30%
35%
40%
45%
50%
55%
60%
65%
70%
75%
80%
85%
90%
95%
100%
Mean
Potential Radon Regulatory Level, pCi/1
100
30,123
30,678
30,802
30,891
30,954
31,010
31,064
31,113
31,158
31,198
31,239
31,282
31,323
31,368
31,419
31 ,467
31,524
31,591
31,676
31,801
32,362
31,307
300
15,982
16,713
16,860
16,956
17,030
17,097
17,158
17,210
17,263
17,314
17,362
17,408
17,455
17,502
17,561
17,624
17,689
17,766
17,857
17,999
18,766
17,349
500
10,119
10,940
1 1 ,059
11,125
11,186
1 1 ,230
1 1 ,285
11,333
1 1 ,374
11,418
1 1 ,460
11,506
11,543
11,591
1 1 ,656
11,711
1 1 ,763
1 1 ,840
1 1 ,923
12,033
12,726
11,408
700
7,272
7,978
8,083
8,153
8,210
8,262
8,312
8,350
8,389
8,426
8,463
8,501
8,541
8,586
8,626
8,682
8,738
8,797
8,855
8,970
9,569
8,366
: 1,000
5,173
1 5,548
; 5,651
1 5,725
: 5,769
: 5,814
i 5,854
' 5,898
5,932
5,960
: 5,988
6,023
6,060
6,097
6,135
6,172
6,222
: 6,273
, 6,341
: 6,429
6,968
5,892
2,000
2,365
2,633
2,701
2,741
2,778
2,810
2,832
2,857
2,879
2,903
2,925
2,950
2,977
3,001
3,027
3,054
3,086
3,120
3,168
3,239
3,594
2,852
4,000
939
1,155
1,199
1 ,232
1,255
1,279
1,301
1,316
1,332
1,348
1,364
1,382
1,398
1,416
1,429
1,448
1,470
1,492
1,524
1,570
1,853
1,312
On the whole, the spread in the distributions of the estimates are quite narrow. At 100 pCi/1,
the difference between the 5th and 95th percentiles estimates is only about 3.5 percent. As the
regulatory levels increase, the spread in the distributions increasbs, so that at 4,000 pCi/1, the 95th
percentile estimate is about 36 percent greater than the 5th percentile estimate.
Methods, Occurrence, and Monitoring Document
5-67
-------
The second set of simulations looked at the distributions of the numbers of systems in
different size categories above a regulatory limit of 300 pCi/1. The results of that analysis are
summarized in Exhibit 5-38. In this case, the spread in the distributions are also relatively narrow,
and inversely proportional to the average numbers of systems predicted to be above regulatory
levels. For the very very small systems, the 5th and 95th percentile differ by approximately 12
percent around a mean of 6,552 systems above 300 pCi/1. The average number of large systems
above 300 pCi/1, in contrast, is 409, and the 95th percentile is 31 percent higher than the 5th
percentile.
I
Finally, confidence limits were derived for the mean numbers of systems exceeding 300 pCi/1, using
the "bootstrap" method described in Section 5.2. The results of that analysis are summarized in
Exhibit 5-39.
Exhibit 5-38. Distributions of Monte Carlo Estimates of Community Systems
Above 300 pCi/1, by System Size Category
Percentiles
0%
5%
10%
15%
20%
25%
30%
35%
40%
45%
50%
55%
60%
65%
70%
75%
80%
85%
90%
95%
100%
Mean
System Size
VVS
5,669
6,179
6,261
6,315
6,360
6,400
6,430
6,463
6,493
6,521
6,550
6,576
6,605
6,634
6,668
6,706
6,744
6,791
6,849
6,931
7,429
6,552
VS
5,506
6,051
6,147
6,206
6,264
6,304
6,346
6,382
6,416
6,448
6,482
6,512
6,544
6,575
6,613
6,649
6,690
6,735
6,796
6,887
7,331
6,476
S
2,498
2,866
2,925
2,967
3,002
3,030
3,054
3,076
3,106
3,129
3,151
3,171
3,192
3,215
3,239
3,266
3,297
3,326
3,365
3,421
3,824
3,148
M
483
645
672
691
706
719
730
742
753
763
774
784
794
805
816
830
844
861
881
913
1,073
775
L
283
355
367
375
381
387
392
396
400
404
409
413
417
422
426
431
437
444
452
465
533
409
Total
15,982
16,713
16,860
16,956
17,030
17,097
17,158
17,210
17,263
17,314
17,362
17,408
17,455
17,502
17,561
17,624
17,689
17,766
17,857
17,999
18,766
17,349
Methods, Occurrence, and Monitoring Document
5-68
-------
Exhibit 5-39. Confidence Limits on Mean Estimates of the Numbers of Systems
Above 300 pCi/I
Estimate
5th Percentile
Mean
95th Percentile
System Size Category
All
17,316
15,359
17,396
Very Very
Small
6,513
5,549
6,528
Very
Small
6,441
6,482
6,528
Small
3,126
3,149
3,175
Medium
760
774
790
Large
403
409
415
The widest interval between the upper and lower confidence is approximately four percent
for the medium systems. The interval for the large systems is approximately three percent, and the
range between the upper and lower confidence limits for the other size categories are between
approximately 1.0 and 1.6 percent. For all systems combined, the range between the upper and
lower confidence limits is approximately 0.5 percent, or a difference of 80 systems, compared to an
average of 15,359.
The estimates just discussed provide a rough estimate of the degree of uncertainty associated
with the estimates of the numbers of systems above the different potential regulatory levels, the
numbers of systems in different size categories above a regulatory limit of 300 pCi/1, and estimates of
the uncertainty in the mean estimates of these values. On the whole, the distributions of the
estimates are quite narrow, considering the magnitude of variability in radon levels in community
water systems. A major reason for this is that, when averaged over large numbers of systems, much
of this variability is "smoothed out" and the estimates of the overall proportions of systems above the
regulatory levels are quite stable. The validity of these estimates depends on three major
assumptions:
• That long-term radon levels in community water systems are well-represented by the
lognormal models that have been derived for the various states and regions,
• That the potential errors in the summary statistics used to describe these distributions are
normally distributed, and
• That the estimates of the numbers of active community systems in the various states are
regions have been accurately estimated.
To the extent that these assumptions are not true, the uncertainty in the numbers of systems
above potential regulatory levels may have been underestimated.
Methods, Occurrence, and Monitoring Document
5-69
-------
5.9 Comparison of Current Estimates of Radon Exceedences to Previous EPA
Occurrence Analyses
The preceding analyses have employed the same general approaches as those used in EPA's
previous occurrence analyses except that the data base of radon occurrence has been expanded to
include additional data for seven states. In addition, the approaches used to estimate summary
statistics for the radon distributions are slightly different than those employed n the previous
occurrence analyses, and the NIRS data have been adjusted to compensate for differences between
in-system and point-of-entry sampling. Finally, as will be discussed below, the estimated of the
numbers of active community and non-transient non-community systems have changed since the
previous effort. This section provides a brief comparison of the results of the current analysis with
those of EPA's previous efforts.
The data for the 1993 occurrence analysis (Wade Miller Associates 1993) came from EPA's
FRDS data base, current as of 1992. while the estimates used in this analysis come from EPA's
Drinking Water Baseline Handbook. The current estimates are based on SDWIS data from mid-
1998. There are considerable differences between the estimated numbers of community and non-
transient non-community water systems derived from the two sources, as shown in Exhibit 5-40. In
1993, EPA estimated that there were a 45,626 community groundwater systems and 23,865 non-
transient non-community groundwater systems that could be affected by a radon rule. Using the
more recent data, EPA estimates that there are only 40,812 community groundwater systems active
in the U.S., and 19,062 non-transient non-community systems.
Exhibit 5-40. Comparison of Estimated Numbers of Groundwater Systems from the 1993
Occurrence Analysis With Estimates from The Current Analysis
System Size
System Type
Very Very
Small
25-100
Very
Small
101-500
Small
501-1,000
1,001-
3,300
Medium
3,301-
10,000
Large
10,001-
100,000
>1 00,000
Total
1993 Occurrence Analysis
Community
Non-Transient
Non-Communitv
Total
16,634
13,842
30,813
15,422
7,512
23,673
4,691
1,817
6,854
5,261
627
6,232
2,302
63
2,471
1,257
4
1,322
This Analysis
Communitv
Non-Transient
Non-Communitv
Total
13,687
9,606
23,293
13,860
6,840
20,700
4,247
1,891
6,138
5,260
665
5,925
2,335
53
2,388
1,357
7
1,364
59
0
62
45,626
23,865
71,427
66
0
66
40,812
19,062
59,874
Methods, Occurrence, and Monitoring Document
5-70
-------
The reasons for the differences between the earlier estimates and those derived from more
recent data are not clear. Based on the data in Exhibit 5-40, it would appear that the bulk of the
changes have been reductions in the numbers of the smallest systems, both community and non-
transient non-community. It is possible that many of these systems have ceased operations, or
merged with larger systems. Whatever the reasons, the number of community groundwater systems
is estimated to have decreased by approximately 11 percent and the number of non-transient non-
community systems decreased by approximately 20 percent between 1993 and 1998.
In addition to the number of systems affected, the current analysis differs substantially from
the previous one in the proportions of systems predicted to be above the various regulatory levels
(Exhibit 5-41.) The proportions of systems predicted to exceed the three lowest regulatory levels
(100, 300, and 500 pCi/1) in this analysis are all higher than both the lower-bound and upper-bound
estimates from the 1993 analysis. At regulatory limits of 1,000 and 2,000 pCi/1, the predictions of
the current analysis fall between the lower- and upper-bound analyses from 1993, but are closer to
the upper-bound estimates. The major reason for this difference is undoubtedly the upward
adjustment made in the regional log mean values derived from the NIRS data to account for
differences between in-system versus point-of-entry sampling. '
Exhibit 5-41. Comparisons of Estimated Proportions and Numbers of Systems Exceeding
Potential Regulatory Levels from 1993 Occurrence Analysis and the Current Study
1. Estimated Proportions of Systems Exceeding Potential Regulatory Levels
1993 Occurrence Analysis
Reaulatory Level. pCi/l
Lower Bound
Uooer Bound
100
63.5%
75.7%
300
33.4%
46.7%
500
21.4%
33.8%
700
_
1.000
9.7%
19.8%
2.000
3.1%
10.4%
4.000
—
This Analysis
Mean Estimate
81.3%
49.6%
34.0%
25.2%
17.5%
7.8%
3.2%
2. Estimated Numbers of Systems Exceeding Potential Regulatory Levels
1993 Occurrence Analysis
Reaulatorv Level. pCi/l
Lower Bound
Ucoer Bound
100
44.115
52.571
300
23.201
32.472
500
14.898
23.503
700
— !
— '
1.000
6.763
13.749
2.000
2.185
7.229
4.000
—
—
This Analysis
Mean Estimate
48.684
29,716
20.340
15.060
10.498
4.686
1.900
Notes:
1. Includes community and non-transient non-community systems
2. Source: Wade Miller Associates 1993
Methods, Occurrence, and Monitoring Document
5-71
-------
As shown in the bottom panel of Exhibit 5-40, estimates of the numbers of systems above
potential regulatory levels from the current analysis are generally comparable to the 1993 estimates.
For all the regulatory levels where predictions have been made in both studies, the current analysis
predicts that the numbers of systems exceeding regulatory levels will fell almost exactly half-way
between the lower- and upper-bound estimates from the 1993 analysis. The higher proportions
exceeding regulatory levels predicted in the current study are offset by the lower total number of
systems that could be affected by a radon rule. Thus, despite the substantial differences in input data
and estimation methods, this analysis arrives at roughly the same estimates of the numbers of systems
that could be affected by the radon rule as were predicted in EPA's previous study.
5.10. References
Barry, T.M., and Brattin, WJ. (1998), "Distribution of Radon-222 in Community Groundwater
Systems: Analysis of Type I Left-Censored Data With a Single Censoring Point", Human and
Ecological Risk Assessment, 4(2X 579-603.
Burmaster, D.E., and Wilson, A.M.(1998), "Fitting Second-Order Mixture Models to Data With
Many Non-Detects Using Maximum Likelihood Estimation", Submitted to Human and Ecological
Risk Assessment, June
D'Agostino, R.B. and M.A. Stephens, (1989), Goodness of Fit Testing, Marcel Dekker, New York,
pp. 114-115.
Davison, A.C., and D.V. Hinkley, (1997), Bootstrap Methods and Their Application, Cambridge
University Press, Cambridge, UK, pp 193-198.
Dempster, A.P., Laird, N.M., and Rubin, D.B (1979), "Maximum Likelihood Estimation from
Incomplete Data Via the EM Algorithm", Journal of the Royal Statistical Society, Series B, 39, 1-
38.
U.S. Environmental Protection Agency (1995), Uncertainty Analysis of Risk Associated With
Exposure to Radon 'in Drinking Water, Office of Science and Technology, Office of Policy, Planning,
and Evaluation, March, EPA-822-R-96-005
USEPA (1985), Nationwide Occurrence of Radon and Other Natural Radioactivity in Public Water
Supplies, National Air and Environmental Radiation Laboratory, October, EPA-550-5-85-008.
Hess, C.T., J. Michael, T.R. Horton, H.M. Pritchard, and W.A. Coniglio, (1985). "The Occurrence
of Radioactivity in Public Water Supplies in the United States," Health Physics, Vol 48 No. 5 (May)
pp. 553-586.
C.T. Hess, et al.(1979), Radon-222 Variability in Potable Water Supplies in Maine: the Geology,
Hydrology, and Physics and Health Effects, Land and Water Resources Center, University of Maine
at Orono, Project No. A-045-ME,
Methods, Occurrence, and Monitoring Document 5-72
-------
Horton, T.R. (1983), Methods and Results of EPA's Study of Radon in Drinking Water in Drinking
Water, USEPA-520/5-83-027
ICF Kaiser, (1998), Re-Evaluation of Radon Occurrence in grouridwater Supplies in the United
States: External Review Draft, Submitted to the U.S. Environmental Protection Agency Office of
Ground-water and Drinking Water, September 30.
Johnson, N.L., and Katz, M., (1969) Discrete Distributions,.
Longtin, J.P.(1987), "Occurrence of Radon, Radium, and Uranium in Groundwater", Journal of the
American Water Works Association, July,, pp. 84-93.
National Academy of Sciences (1998), Risk Assessment for Radon in Drinking Water, Committee on
Risk Assessment for Exposure to Radon in Drinking Water, National Research Council, September.
Ott, W.R., (1995) Environmental Statistics and Data Analysis, Lewis Publishers, Ann Arbor, pp.
251-255. ;
RCG/Hagler, Bailly, Inc.(1992), Final Report: The Cost of Compliance With the Proposed Federal
Drinking Water Standards for Radionuclides, Prepared for the American Water Works Association,
Section 2.
Shapiro, S.S, and Wilk, M.B.(1965), "An Analysis of Variance Test for Normality (Complete
Samples)", Biometrika, 52, 591 -611.
Wade Miller Associates, Inc.(1990), Occurrence and Exposure Assessment for Radon in Public
Water Supplies, prepared for EPA's Office of Drinking Water, EPA, September 25.
Wade Miller Associates, Inc.(1993), Addendum to: The Occurrence and Exposure Assessments for
Radon, Radium-226, Radium-228, Uranium, and Gross Alpha Particle Activity in Public Drinking
Water Supplies (Revised Occurrence Estimates Based on Comments to the Proposed Radionuclides
Regulation), Final Draft, September 1993.
Methods, Occurrence, and Monitoring Document
5-73
-------
-------
6.
POTENTIAL EXPOSED POPULATIONS
This chapter presents estimates of the populations exposed to radon at various levels, and
discusses exposures to specially sensitive populations. Estimates of the populations exposed
were derived by multiplying the numbers of systems above various radon levels by the average
numbers of individuals served by each size system.
6.1 Data Sources
The data sources used to estimate the numbers of systems of different sizes in the U.S.
and the average populations served per systems was EPA's Baseline Handbook (EPA 1999a).
These data were derived from analyses of the SDWIS database based on information gathered in
mid-1998. Regional and state estimates of the proportions of systems of different sizes above
potential regulatory levels were derived as explained in Section 5.8
6.2 Populations Above Regulatory Levels
Exhibit 6-1 presents estimates of the numbers of populations served by systems with
influent radon above various potential regulatory limits. .
Exhibit 6-1. Populations Potentially Exposed Above Potential Regulatory Levels
By System Size (Thousands)
Radon
level
(pCi/1)
4,000
2,000
1,000
700
500
300
100
Very
Very
Small
25-100
9.4
41
128
202
290
445
733
Very
Very
Small
101-500
46
183
541
848
1,210
1,880
3 290
Very Small
501-3,300
20
119
513
962
1,620
3,140
8.080
Small
3,301-1 OK
0.2
5.7
85.5
267
672
2,080
8.760
Medium
10K-100K
0.9
21.7
: 289
: 859
I 2,070
6,060
23.400
Large
> 100K
0.4
11.0
147
436
1,050
3,070
11.900
Total
77
381
1,695
3,558
6,893
16,641
56.054
Based on SDWIS data, a total approximately 89.7 million people are served by
community water systems. Of these, just over 56 million (62.6 percent) are served by community
groundwater systems with radon levels above 100 pCi/1. Approximately 16.6 million (18.6
percent) are served by systems with long-term radon levels above 300 pCi/1. The proportions
Methods, Occurrence, and Monitoring'Document
6-1
-------
potentially exposed decrease as the regulatory levels increase; only about 77 thousand people
(0.09 percent) are estimated to be exposed above the NAS estimated AMCL level of 4,000 pCi/1.
The proportions of the population exposed above the potential regulatory levels do not
match the proportions of systems above these levels shown in Section 5.8. This is because the
larger systems, while comprising only a small proportion of the total systems, account for large
fractions of the total populations served.
6.3 Special Populations
The numbers of individuals potentially exposed to radon in Exhibit 6-1 includes all
members of the general residential population. EPA has identified only one population that may
be especially sensitive to radon exposures (actually exposure to radon progeny), namely,
smokers. In their risk assessment for radon exposures from drinking water (see EPA 1999b), the
Agency has assumed a historical "ever-smoking" prevalence of 58 percent for males, and 42
percent for females. The agency recognizes that smoking prevalence is currently decreasing, and
thus the numbers of sensitive individuals exposed to radon could decrease in the future.
A further 5.2 million individuals are estimated to be exposed to radon from non-
community non-transient systems (EPA 1999a). In addition, an undetermined number of
individuals are exposed to radon from community water systems in non-residential settings.
There is no data that allows the extent of overlap among these populations to be evaluated. The
levels of exposure associated with a given radon level in water are expected to be lower for non-
residential exposures and exposures to non-community systems than for residential exposures
from community systems.
6.4 References
EPA (1999a), Drinking Water Baseline Handbook, First Edition, Prepared by International
Consultants, Inc, for the Office of Ground Water and Drinking Water, March 2, Draft.
EPA (1999b), Regulatory Impact Analysis and Revised Health Risk Reduction and Cost Analysis
for Radon in Drinking Water, Office of Groundwater and Drinking Water, August.
Methods, Occurrence, and Monitoring Document
6-2
-------
7.
CO-OCCURRENCE ASSESSMENT
7.1 Data Sources
In its analysis of the patterns of co-occurrence of radon with other contaminants, EPA has
relied on two primary data sources. The first is the US Geological Survey (USGS) National
Water Information System (NWIS). NWIS contains data on water quality parameters in surface
and groundwater sources in all fifty states (SAIC 1999). The data are available on-line through
the USGS Water Resources Division, and have been the subject of extensive analysis by the
agency to determine co-occurrence patterns for a wide range of contaminants
In addition to NWIS, the NIRS measured the levels of arialytes other than radon, and thus
provides information concerning radon co-occurrence with other analytes in distribution systems.
While the number of systems sampled is limited, EPA has also analyzed the NIRS data to
investigate the co-occurrence patterns of radon, other radionuclides, and inorganic analytes.
7.2 Co-Occurrence of Radon With Other Contaminants
Screening analysis of the NWIS database (SAIC 1999) have identified statistically
significant correlations between the occurrence of radon and several other contaminants in
certain regions of the U.S. The correlations are summarized in Exhibit 7-1.
Exhibit 7-1. Correlations of Radon With Other Analytes in INWIS DATA
EPA Region
4
7
9
10
Significant Correlation of
Radon Levels With:
Iron :
Manganese
Sulfate
Nitrate ;
Sulfate ;
Beryllium
Selenium '
Sulfate
Barium
Chromium
Manganese
Levels of radon in the INWIS data were found to be significantly correlated with the
levels of one or more inorganics in each of four EPA regions. At the national level, radon is not
Methods, Occurrence, and Monitoring Document for Radon
7-1
-------
significantly correlated with any other analytes in groundwater systems. It is not clear that the
number of correlations seen exceeds the number that would be expected to occur by chance,
given the number of comparisons that were made. No analyses have been conducted of the
correlations between radon and other analytes in systems of different sizes.
Based on only this preliminary analysis, it is not possible to determine whether their
might be any geological factors contributing to the observed patterns of co-occurrence. Given
the greatly differing geochemical properties of the analytes with which radon co-occurs, it is hard
to develop any hypothesis that would explain the co-occurrence. In the NIRS data, radon levels
are significantly correlated with only one analyte (chloride) at the national level, and with none at
the regional level. Taken together, the available data therefore do not suggest that, at the national
level, radon occurrence is correlated with the occurrence of any inorganic contaminants.
Additional analyses are needed to determine if there are any patterns of co-occurrence on smaller
distance scales.
Few data are available regarding the co-occurrence of radon and organic compounds.
Because radon is primarily naturally occurring, and not often associated with human activities,
there is little reason to believe there is a correlation between man-made groundwater pollution
with organic contaminants and radon levels. Generally, groundwater sources tend to have lower
levels of organic compounds than surface water sources. As reported in the proposed
Disinfectants and Disinfection Byproducts Rule (EPA 1994), a survey of surface waters showed
TOC levels with 25th, 50th, and 75th percentiles of 2.6, 4.0, and 6.0 mg/1, respectively. Ground-
waters showed TOC levels at the same percentiles of "non-detect", 0.8, and 1.9 mg/1,
respectively. Nationally, typical ground waters have low TOC levels. However, some areas of
the U.S., e.g., the Southeastern U.S. (EPA Region 4) have some aquifers with high TOC levels.
7.2 Implications of Co-Occurrence
Despite the fact that radon occurrence is not significantly correlated with any other
contaminants at the national level none the less, the existing patterns of its co-occurrence with
some common analytes have important implications for the selection of radon mitigation
technologies and for radon mitigation costs. The most important of these is iron and manganese.
EPA has estimated (EPA 1999) that a substantial fraction of groundwater systems with radon
levels exceeding potential regulatory limits would also have levels of iron (Exhibit 7-2) and
manganese (Exhibit 7-3) exceeding levels that would require sequestration or some other to
protect aeration systems from fowling. This pattern has the potential for increasing the radon
mitigation costs for those systems not already treating for iron and manganese.
One other important instance where co-occurrence is potentially important are those
situations where radon and arsenic co-occur. In these cases, there is the potential that aeration
treatment to reduce radon levels could also aid in the reduction of arsenic exposures because
aeration would oxidize arsenic from soluble trivalent for to the comparatively insoluble
pentavalent form. This would make it easier to remove arsenic from the water (EPA 1999).
Methods, Occurrence, and Monitoring Document for Radon
7-2
-------
Table 7-2. Co-Occurrence of Radon with Dissolved Iron in Raw Ground Water1 (4188
samples)
Radon
(pCi/L)
ND
<100
100-300
300-1,000
1,000-3,000
>3,000
Totals
Dissolved Fe (mg/L)
ND
0.67%
2.17%
7.55%
18.89%
6.42%
2.10%
37.80%
<0.3
0.36%
1 .72%
10.20%
22.61%
9.05%
3.82%
47.76%
0.3-1.5
0.21%
0.53%
2.67%
3.08%3
0.74%
0.31%
7.54%
1.5-2.5
0.02%
0.12%
1 .34%
0.57%
o.jo%
0.02%
2.17%
>2.5
0.31%
0.48%
1 .74%
1.31%
0.62%
0.26%
4.72%
Totals
1 .57%
5.02%
23.50%
46.46%
16.93%
6.51%
100.00%
Notes: :
1. Source: EPA analyses of NWIS data (EPA 1999)
i
Exhibit 7-3. Co-Occurrence of Radon with Dissolved Manganese in Raw Ground Water
(4189 samples) ' ;
Radon
(pCi/L)
ND
<100
100-300
300-1,000
1,000-3,000
> 3,000
Totals
Dissolved Mn (mg/L)
ND
0.69%
2.67%
8.00%
21.99%
6.45%
1 .43%
41.23%
<0.02
0.26%
0.84%
5.97%
11.84%
5.90%
3.39%
28.20%
0.02-0.05
0.05%
0.36%
2.20%
3.17%
1.24%
0.53%
7.55%
> .050
0.57%
: 1.15%
7.33%
; 9.48%3
3.34%
1.17%
i 23.04%
Totals
1.57%
5.02%
23.50%
46.48%
16.93%
6.52%
100.00%
Notes: '.
1. Source: EPA analyses of NWIS data (EPA 1999) ;
7.3 References ;
EPA (1999), Regulatory Impact Analysis and Revised Health Risk Reduction and Cost Analysis
for Radon in Drinking Water, Office of Groundwater and Drinking Water, August.
EPA (1994), Proposed Disinfection and Disinfection Byproducts Rule, 59 Federal Register
38668, July 29.
Methods, Occurrence, and Monitoring Document for Radon
7-3
-------
Science Applications International Corporation (1999), Co-Occurrence of Drinking Water
Contaminants Primary and Secondary Constituents, Submitted to the U.S. Environmental
Protection Agency Office of Ground Water and Drinking Water, May 21.
Methods, Occurrence, and Monitoring Document for Radon
7-4
-------
8.
MONITORING APPROACHES
8.1 Background
The monitoring regulation for radon proposed in 1991 by EPA required that groundwater
systems monitor for radon at each entry point to the distribution system quarterly for one year
initially. Monitoring could be reduced to one sample annually per entry point to the distribution
system if the average of all first quarterly samples was below the MCL. States could allow
systems to reduce monitoring to once every three years if the system demonstrated that results of
all previous samples collected were below the MCL. The proposal also allowed States to grant
waivers to groundwater systems to reduce the frequency of monitoring, up to once every 9 years,
if States determined that radon levels in drinking water were consistently and reliably below the
MCL. Comments made in response to the proposed monitoring requirements for radon were
mainly concerned that the proposed monitoring requirements did not adequately take into
account the effect of seasonal variations in radon levels on determining compliance. Other
commenters felt that sampling at the entry point of the distribution system was not representative
of exposure to radon, and they suggested that sampling for radon should be done at the point of
use. !
Since the 1991 proposal EPA has obtained additional information from states, the water
utilities, and academia on the occurrence of radon, including data on its temporal variability (See
Section 5.7). Utilizing this additional data, the Agency performed extensive statistical analyses
to predict how temporal, analytical variations and variations between individual wells may affect
exposure to radon. The results of these analyses are described iin detail in the previous sections of
this document. As a result of the new information, EPA was able to refine the requirements for
monitoring and address the concerns expressed by the commenters on the 1991 proposal.
8.2 Objectives of Monitoring Program
The objectives of the monitoring program are to provide for rapid, cost-effective
identification of groundwater sources that are above and below proposed regulatory limits. The
proposed monitoring requirements for radon are consistent with the monitoring requirements for
regulated drinking water contaminants, as described in the Standardized Monitoring Framework
(SMF) promulgated by EPA under the Phase II Rule of the National Primary Drinking Water
Regulations (NPDWR) and revised under Phases IIB and V. The goal of the SMF is to
streamline the drinking water monitoring requirements by standardizing them within contaminant
groups and by synchronizing monitoring schedules across contaminant groups.
In developing the proposed compliance monitoring requirements for radon, EPA
considered:
• The likely source of contamination in drinking water;
Methods, Occurrence and Monitoring Document for Radon
8-1
-------
• The differences between ground water and surface water systems;
• The need to collect samples which are representative of consumer exposure;
• Sample collection and analytical methods;
• The use of appropriate historical data to identify vulnerable systems and to specify
monitoring requirements for individual systems;
• The analytical, temporal and intra-system variance of radon levels;
• The use of appropriate historical data and statistical analysis to establish reduced
monitoring requirements for individual systems; and
• The need to provide flexibility to the States to tailor monitoring requirements to site-
specific conditions by allowing them to:
—grant waivers to systems to reduce monitoring frequency, provided certain
conditions are met,
—require confirmation samples for any sample exceeding the MCL/AMCL
—allow the use of previous sampling data to satisfy initial sampling requirements.
—increase or decrease monitoring frequency.
8.3 Description of Proposed Monitoring Requirements
Exhibit 8.1 provides a description of the monitoring requirements proposed for radon,
and compares them to the requirements in the 1991 Proposal and in the Standardized Monitoring
Framework. The major provisions include
Monitoring for Surface Water Systems
Systems relying exclusively on surface water as their water source will not be required to
sample for radon. Systems that rely in part on ground water would be considered groundwater
systems for purposes of radon monitoring. Systems that use ground water to supplement surface
water during low-flow periods will be required to monitor for radon. Ground water under the
influence of surface water would be considered ground water for this regulation.
Sampling, Monitoring Schedule and Initial Compliance for Groundwater Systems
EPA is retaining the quarterly monitoring requirement for radon as proposed initially in
the 1991 proposal to account for variations such as sampling, analytical and temporal variability
Methods, Occurrence and Monitoring Document far Radon
8-2
-------
Exhibit 8.1 Comparison of Monitoring Requirements
- % MONITOR! N<5 REQUIREMENTS FOR RABON - - — l
1991 Proposal
1999 Proposal - MCL/AMCL
SMF for lOCs in Groundwater
INITIAL MONITORING REQUIREMENTS
Four consecutive quarters of
monitoring at each entry point for
one year. Initial monitoring must
be completed by January 1 , 1 999.
Four consecutive quarters of
monitoring at each entry point for
one year. Initial monitoring must
begin by 3 years from after
publication of the final rule in
Federal Register or 4.5 years from •
date of publication of the final rule
in Federal Register (depending on
effective date applicable to the
State).
Four consecutive quarters of
monitoring at each entry point for
sampling points initially exceeding
MCL.
ROUTINE MONlTORINO REQUIREMENTS
One sample annually if average
from four consecutive quarterly
samples taken initially is less than
MCL.
1991 Proposal
One sample annually if average
from four consecutive quarterly
samples is less than MCL/AMCL, .
and at the discretion of State.
1999 Proposal - MCL
One sample at each sample point
during the initial 3 year compliance
period for groundwater systems for
sampling points below MCL.
SMF for lOCs in Groundwater
REDUCED MONITORING REQUIREMENTS j
State may allow groundwater
systems to reduce the frequency of
monitoring to once every three
years provided that they have
monitored quarterly in the initial
year and completed annual testing
in the second and third year of the
first compliance period.
Groundwater systems must
demonstrate that all previous
analytical samples were less than
the MCL.
State may allow groundwater
systems to reduce monitoring
frequency to:
Once every 3 years if average from
four consecutive quarterly samples
is less than !/2 the MCL/AMCL,
provided no samples exceed the
MCL/AMCL or if the system is
determined by State to be "reliably
and consistently below
MCL/AMCL".
State may allow groundwater
systems to reduce monitoring
frequency to:
Once every 3 years if sample
subsequently detects less than MCL
and determined by State to br
"reliably and consistently below
MCL".
Methods, Occurrence and Monitoring Document for Radon
8-3
-------
Exhibit 8-1. Comparison of Monitoring Requirements (Continued)
INCREASED MOMTORIHG REQUIREMENTS
Systems monitoring annually
or once per three year
compliance period that exceed
the radon MCL in a single
sample would be required to
revert to quarterly monitoring
until the average of 4
consecutive samples is less
than the MCL. Ground water
systems with unconnected
wells would be required to
conduct increased monitoring
only at those wells exceeding
the MCL.
The State may require more
frequent monitoring than
specified.
Systems may apply to the State
to conduct more frequent
monitoring than the minimum
monitoring frequencies
specified.
Systems monitoring annually
would be required to increase
monitoring if the MCL/AMCL
for radon is exceeded in a single
sample, the system would be
required to revert to quarterly
monitoring until the average of
4 consecutive samples is less
than the MCL/AMCL.
Systems monitoring once every
three years would be required to
monitor annually if the radon
level is less than MCL/AMCL
but above !/2 MCL/AMCL in a
single sample. Systems may
revert to monitoring once per
three years if the average of the
initial and three consecutive
annual samples is less than !/2
MCL/AMCL
Ground water systems with un-
connected wells would be
required to conduct increased
monitoring only at those wells
which are affected.
If the MCL is exceeded in a
single sample, the system
required to begin sampling
quarterly until State determines
that it is "reliably and
consistently "below MCL. .
" 's" '" MONITOKINt? R|;QWftEM^T$ F0R RAIX>& 'j
1991 Proposal
1999 Proposal - MCL
SMF for lOCs in Groundwater
Methods, Occurrence and Monitoring Document for Radon
8-4
-------
Exhibit 8.1 Comparison of Monitoring Requirements (Continued)
Where the results of sampling
indicate an exceedence of the
maximum contaminant level, the
State may require that one
additional sample be collected as
soon as possible after the initial
sample was taken [but not to
exceed two weeks] at the same
sampling point. The results of the
initial sample and the confirmation
sample shall be averaged and the
resulting average shall be used to
determine compliance.
Systems may collect confirmation
samples as specified by the State.
The average of the initial sample
and any confirmation samples will
be used to determine compliance.
Where the results of sampling
indicate an exceedence of the
maximum contaminant level, the
State may require that one
additional sample be collected as
soon as possible after the initial
sample was taken [but not to exceed
two weeks] at the same sampling
point. The results of the initial
sample and the confirmation sample
shall be averaged and the resulting
average shall be used to determine
compliance.
OFJ&ATA
If monitoring data collected after
January 1, 1985 are generally
consistent with the requirements
specified in the regulation, than
the State may allow the systems to
use those data to satisfy the
monitoring requirements for the
initial compliance period.
If monitoring data collected after
proposal of the rule are consistent
with the requirements specified in
the regulation, then the State may
allow the systems to use those data
to satisfy the monitoring
requirements for the initial
compliance period.
States may allow previous sampling
data to satisfy the initial sampling
requirements provided the data
were collected after January 1,
1990
WAIVERS
State may grant waiver to
groundwater systems to reduce the
frequency of monitoring, up to 9
years, if State determines that
radon levels in drinking water are
"reliably and consistently" below
the MCL.
The State may grant a monitoring
waiver to systems to reduce the
frequency of monitoring to up to
one sample every 9 years based on
previous analytical results,
geological characteristics of source
water aquifer and if a State
determines that radon levels in
drinking water are reliably and
consistently below the
MCL/AMCL.
Analytical results of all previous
samples taken must be below !/2 the
MCL/AMCL.
State may grant waiver to
groundwater systems after
conducting vulnerability assessment
to reduce the frequency of
monitoring, up to 9 years, if State
determines that radon levels in
drinking water are "reliably and
consistently" below the MCL.
System must have 3 previous
samples. Analytical results of all
previous samples taken must be
below MCL.
Methods, Occurrence and Monitoring Document for Radon
J-5
-------
in radon levels. Results of analysis ofdata obtained since 1991, estimating contributions of
individual sources of variability to overall variance in the radon data sets evaluated, indicated
that sampling and analytical variance contributes less than 1% to the overall variance. Temporal
variability within single wells accounts for between 13 and 18% of the variance in the data sets
evaluated, and a similar proportion (12-17%) accounts for variation in radon levels among wells
within systems (Section 5.7).
The Agency performed additional analyses to determine whether the requirement of
initial quarterly monitoring for radon was adequate to account for seasonal variations in radon
levels and to identify non-compliance with the MCL/AMCL. Results of analysis based on radon
levels modeled for radon distribution for groundwater sources and systems in the US (ICF
Consulting 1999) show that the average of the first four quarterly samples provides a good
indication of the probability that the long-term average radon level in a given source would
exceed the preferred regulatory levels.
i
Reduced Sampling Frequency
I
j
Initial compliance with regulatory levels will be determined based on an average of four
quarterly samples taken at individual sampling points in the initial year of monitoring. Systems
with averages exceeding the regulatory levels at any well or sampling point will be deemed to be
out of compliance. Systems exceeding the regulatory levels will be required to monitor quarterly
until the average of four consecutive samples are less than the regulatory levels. Systems will
then be allowed to collect one sample annually if the average from four consecutive quarterly
samples is less than the regulatory levels and if the State determines that the system is reliably
and consistently below regulatory levels.
States will be allowed to reduce monitoring frequency to once every three years (one
sample per compliance period) per well or sampling point, if the average from four consecutive
quarterly samples is less than one-half the regulatory levels and the State determines that the
system is reliably and consistently below one-half the regulatory levels. As shown in Exhibits 8-
2 and 8-3, EPA believes that there is sufficient margin of safety to allow for this since there is a
small probability that long term average radon levels will exceed the regulatory levels.
Systems monitoring annually that exceed the radon regulatory levels in a single sample
will be required to revert to quarterly monitoring until the average of four consecutive samples is
less than the regulatory levels. Ground water systems with unconnected wells will be required to
conduct increased monitoring only at those wells exceeding the regulatory levels. Compliance
will be based on the average of the initial sample and 3 consecutive quarterly samples.
Systems monitoring once per compliance period or less frequently which exceed one-half
the regulatory level (but do not exceed the regulatory level) in a single sample would be required
to revert to monitoring annually. Systems may revert to monitoring once every three years if the
Methods, Occurrence and Monitoring Document for Radon
8-6
-------
Exhibit 8-2. The Relationship Between the First-Year Average Radon Level and the Probability
of the Long-Term Radon Average Radon Levels Exceeding a Regulatory Level of 300 pCi/1
If the average of the first four quarterly
samples from a source is:
Less than 50 pCi/l
Between 50 and 100 pCi/1
Between 100 and 150 pCi/1
Between 1 50 and 200 pCi/1
Between 200 and 300 pCi/1
Then the probability that the long-term
average radon level in that source
exceeds 300 pCi/1 is:
<0.1 percent
<1 percent
<1 percent
7.2 percent
26.8 percent
Exhibit 8-3. The Relationship Between the First-Year Average Radon Level and the Probability
of the Long-Term Radon Average Radon Levels Exceeding the AMCL
If the average of the first four quarterly
samples from a source is:
Less than 2,000 pCi/1
Between 2,000 and 2,500 pCi/1
Between 2,500 and 3,000 pCi/1
Between 3,000 and 4,000 pCi/1
Then the probability that the long-term
average radon level in that source
exceeds 4000 pCi/1 is:
<0.1 percent
9.9 percent
15.1 percent
32.9 percent
average of the initial and three consecutive annual samples is less than one-half the regulatory
level. Ground water systems with unconnected wells will be required to conduct increased
monitoring only at those wells exceeding the regulatory level.
States may grant a monitoring waiver reducing monitoring frequency to once every nine
years (once per compliance cycle) provided the system demonstrates that it is unlikely that radon
levels in drinking water will occur above the regulatory levels. In granting the waiver, the State
must take into, consideration factors such as the geological area where the water source is located,
and previous analytical results which demonstrate that radon levels do not occur above the
regulatory levels. The waiver may be granted for up to a nine year period. (Given that all
previous samples are less than one-half the regulatory levels, then it is highly unlikely that the
long-term average radon levels would exceed these levels.)
Methods, Occurrence and Monitoring Document for Radon
8-7
-------
Confirmatory Samples
If the analytical results from any sampling point are found to exceed the regulatory level
(in the case of routine monitoring) or one-half the regulatory level (in the case of reduced
monitoring), the State may allow the system to collect a confirmation sample(s). The results of
the initial sample and the confirmation sample(s) can then be averaged and the resulting average
used to determine compliance.
Grandfathering of Data
At a State's discretion, sampling data collected since the proposal could be used to satisfy
the initial sampling'requirements for radon, provided that the system has conducted a monitoring
program not less stringent than that specified in the regulation and used analytical methods
specified in the regulation.
8.4 Costs and Effectiveness of the Proposed Monitoring Requirements
8.4.1 Incremental Skills/Equipment Requirements and Cost of Radon Monitoring
The skill requirements required to implement this monitoring scheme are discussed in
detail in Section 2-5. Generally, the skills required to collect water samples for radon analysis
are comparable to those required to sample for gas chromatographic or atomic absorption
analysis.
EPA has conducted two surveys of the potential costs of radon analysis, as discussed in
Section 2.6. The agencies best estimate of radon costs ranges from $60 to $120 per sample. In
the Regulatory Impact Analysis (EPA 1999), the Agency has estimated that the annual costs of
monitoring for radon by community water systems would be 14.1 million. However, these costs
were estimated assuming all sources would need to be monitored quarterly for the indefinite
future. Subsequent analysis of the impacts of reduced monitoring requirements (ICF Consulting
1999) suggests that, after the first-year quarterly monitoring, the numbers of samples required,
and monitoring costs, would be reduced by approximately 89 percent compared to a requirement
for continuous quarterly monitoring.
8.5 References
EPA (1999), Regulatory Impact Analysis and Revised Health Risk Assessment for Radon in
Drinking Water, Office of Ground Water and Drinking Water, August.
ICF Consulting (1999) Methods, Occurrence and Monitoring Document for Radon: Addendum:
Statistical Analysis of Radon Monitoring Requirements, Submitted to the U.S. Environmental
Protection Agency Office of Ground Water and Drinking Water, July 20.
Methods, Occurrence and Monitoring Document for Radon
8-8
-------
APPENDIX A. DATA MANAGEMENT METHODS AND
SUPPLEMENTAL DATA SETS
-------
-------
This appendix provides a description of the methods that were used to manage data
related to radon occurrence and in the estimation of the numbers of groundwater systems that
would exceed regulatory levels. Appendix A.I addresses data management and the development
of data sets for statistical analysis, and Appendix A.2 provides summary descriptions of the
supplemental state data sets. ;
A.1 Data Management and Manipulation i
Data concerning the levels of radon in groundwater and in drinking water supplies were
received from EPA and from over 20 states. Data from the NIRS survey and from 17 states1
were evaluated in detail as part of this occurrence analysis. This section describes how these data
were managed and prepared for analysis. ;
A.1.1 NIRS Data
The NIRS data were received from EPA-OPPE in spreadsheet format. The data set was
the same one that EPA had used in its 1993 and 1995 Uncertainty Analyses (EPA 1993, 1995)
for radon exposures and risks. The numbers of samples and systems of various sizes represented
in the NIRS data are summarized in the first two columns of Exhibit A-l. The NIRS file
Exhibit A-l. NIRS Data File Characteristics
System
Size1
WS
vs
s
M
L
Total
NIRS Data File
Number of
Systems/ Analyses
338
335
232
53
28
986
Number of
Censored Analyses
71
74
95
18
11
269
Numbers of
Systems
Included in
Occurrence
Analysis
335
333
232
53
28
981
Number of
Systems
Included in
EPA 1993
Uncertainty
Analyses
335
334
232
53
28
982
1. WS= very very small (25-100 served), VS = very small (101-500 served), S = Small (501-
3,300), M = Medium (3,301-10,000), and L = Large (>10,000 served)
1 Summary statistics were developed for the radon data from Missouri (see Appendix
A.2), but were received too late to include in the national model for radon occurrence.
Methods, Occurrence and Management Document for Radon: Appendix A
A-l
-------
contains a total of 986 radon analyses, one from each system. Of these, 269 observations were
censored, reporting values less that the Minimum Reporting Level (MRL) of 100 pCi/1. System
size data (population served) came from the Federal Reporting Date System for 1993.
During the initial data evaluation, three systems that reported radon values as "0" were
excluded from the analysis, along with one observation from a system in Puerto Rico. In
addition, a single large system with an anomalously high radon value (greater than 20 times the
next highest value) was removed from the analysis. These procedures, which were essentially
the same as those employed in EPA's previous analyses, resulted in a database of 981 valid
results, compared to the 982 that were used in the previous analysis. The current analysis
includes one less very small system than the previous evaluations, and the impact of the
difference on the summary statistics for this size stratum is minimal.
A. 1.2 Supplementary Radon Data Sets
Additional data related to radon occurrence in groundwater systems were received from
states, water utilities, and academic researchers. Because there was no uniformity in the types,
amount, or form of the data submitted, a great deal of effort was spent in evaluating the data and
putting it into a consistent format for detailed evaluation.
Data Entry, Editing, Formatting and Quality Assurance
The data sets included the results of one-time state-wide surveys intended to evaluate the
extent of the radon problem, results of ongoing monitoring programs implemented to meet state
of federal drinking water standards, or smaller data sets intended to address specific issues
related to radon distribution. Of the 17 state-wide data sets (Exhibit A-2), 12 consisted of single
samples from single sources in each system. The remaining five data sets (California, Michigan,
New Hampshire, Washington, and Wisconsin) provided results from multiple samples from the
same source and/or samples from multiple sources within systems. Three other supplementary
data sets from wells in Missoula, MT, central North Carolina, and residential wells in
Connecticut also included multiple samples from the same and different sources within systems.
Approximately two-thirds of the data sets were provided in electronic format. Five states
provided only hard copy data. In these cases, the data were manually entered into spreadsheets.
When data were entered manually, senior staff first reviewed the data and identified the specific
fields and data items to be entered. Data were then entered and checked by junior staff, and spot-
checked by senior staff. The data were also sorted to identify unusual and outlier values that
could significantly affect the results of statistical analyses, and these entries were verified against
the raw data. Given the expected broad distribution of radon activity levels, however, unusually
high values ("outliers") were not removed from the data unless there was clear evidence that the
reported analytical result was defective (See the "Comments" column of Exhibit A-2).
Methods, Occurrence and Management Document for Radon: Appendix A A-2
-------
EXHIBIT A-I. SUMMARY OF CENSORED OBSERVATIONS AND CENSORING LEVELS FOR STATE RADON DATA
STATE DATA BASE
CALIFORNIA
CONNECTICUT
IDAHO
IOWA
KANSAS
MAINE
MARYLAND
MICHIGAN
MISSOURI
NEW HAMPSHIRE
NEW YORK
OHIO
PENNSYLVANIA
SOUTH CAROLINA
TEXAS
WASHINGTON
WISCONSIN
Total
Voportion
NUMBER OF
SAMPLES
VALID CWS
SAMPLES
1,59?
32
74
150
245
64
259
186
1,184
2,896
425
228
981
121
169
331
1,252
10,189
100.0%
CENSORING
LEVEL, pCi/I
-10
NA
NA
NA
-25
NA
-20
-50
-10
-100
NA
-25
-100
-20
100
-20
-50
-
-
NUMBER OF
CENSORED
RESULTS
13
0
0
0
10
0
3
7
12
70
0
18
64
I
3
25
23
249
2.4%
TOTAL
SYSTEMS
75
32
64
150
169
64
107
120
691
725
425
228
488
89
120
68
610
4,225
100.0%
SYSTEMS WITH
NON-DETECT
RESULTS
7
0
0
0
9
0
2
7
12 '
53
0
17
31
I
3
14
21
177
4.2%
SYSTEMS WITH
ONLY NON-
DETECT
RESULTS
0
0
0
0
6
0
1
5
10
2
0
17
7
1
2
9
7
67
1.6%
COMMENTS
Eight results < 10 pCi/1 adjusted to 5
pCi/1. five values 200,000 pci/1 excluded '
All results > I.OOOpCi/1
All reported results > 100 pCi/1
All reported results > 40 pCi/1
Abilene has three of six results below
censoring limit
Lowest result = 22pCi/l, next lowest -
200 pCi/1
—
-
Two results < 0 pCi/1 excluded, 10
results < 10 pCi/1 adjusted to 5 pCi/1
Three medium systems with all results
< 100 excluded
Lowest result = 14 pCi/1
—
_
One censored sampled deleted
ecause another sample from the
ame system was > 10,000 pCi/1
_
_
-
Development of Files for Statistical Analyses '.
The amounts and types of information provided with the radon analytical results varied
greatly from state to state, ranging from just sample numbers and radon measurements to detailed
information regarding sampling dates and locations, system and source names, PWSIDs, system
and source type, well depths, geological units, numbers of entry points, populations served, total
production, counting errors, GIS coordinates, and QA/QC information. Thus, some data sets
included all the information needed for all the statistical analyses. However, ancillary data had
to be added to many of the files to allow their use in analyses of radon levels across system size,
intra-system and temporal variability, etc.
In formatting the data files for the occurrence analysis, we preserved the following data
elements if they were present:
PWSID or related state ID number
• Sampling date(s)
City and State
Methods, Occurrence and Management Document for Radon: Appendix A
A-3
-------
• System name
• Source name or designation
• System type (community, noncommunity, etc.)
Source type (groundwater, surface water, mixed)
• Population served
• GIS coordinates
• Radon analytical results (separate entries for multiple analyses)
• Detection/Quantification limits for non-detect samples
In most cases, it was necessary to supplement the data with information from other
sources to complete the data files. Only two states provided information related to the
populations served by the systems sampled (an important stratification variable), so it was almost
always necessary to estimate population served for the sampled systems. This was done using
data from searches of EPA's Safe Drinking Water Information System (SDWIS) database for the
states concerned2. Data from SDWIS were matched to systems in each state using either PWSID
number (the preferred method) or by matching utility names. In a small proportion of cases,
system size was assigned solely on the basis of municipality names. In almost all cases, this
procedure resulted hi systems being assigned to the very small or very very small categories.
For some states, SDWIS data also provided information on system and source type.
Where PWSIDs were not available and the system names did not precisely match between
SDWIS and the supplied radon data, judgement was used to match systems to population served
data3. Where matches appeared doubtful, systems were not assigned a population served value
and were not included in calculations related to system size, two state files (from Connecticut
and Idaho) did not provide sufficient information to assign any systems to size categories. Data
from these states were thus used only to calculate state-wide distributions and representative
radon values.
2 The SDWIS data search that was used to identify systems through their PWSID or
system names was conducted in March, 1998.
3 In many cases, the choice was between two or more community groundwater systems
falling into the same size category (e.g., systems serving 150 and 350 people are both "very
small".) This meant that mis-identification would have no affect on the size-related statistical
analyses.
Methods, Occurrence and Management Document for Radon: Appendix A A-4
-------
Water Sources and Types of Systems Evaluated \
Some states provided data only from groundwater supplies, and most provided data only
related to community water systems. Where states did not identify the water source or system
types, SDWIS data were again used to provide this information. As in the case of population
served, judgement was occasionally needed to assign system or source types. Radon data from
significant numbers of non-community non-transient systems were identified in six states (Idaho,
Maine, Maryland, New Hampshire, Texas, and Wisconsin). The radon distributions in the non-
community systems were evaluated separately from the data from community systems, as
discussed in Section 5.8 of the MOM document.
When no informed choice was possible as to source type or system type, no assignment
was made. Any system with a reported radon level of 100 pCi/1 or greater was assumed to be a
groundwater system. This assumption is justified because, as noted in Section 4 of the MOM
document, radon rapidly escapes from surface water, and radon levels in surface water supplies
are very low, significantly lower than 100 pCi/1. Detailed descriptions of the data contained in
the individual data sets, the methods used to prepare the data for analysis, and the resultant data
files themselves, are provided in Appendix A-2.
It can be seen from Exhibit A-2 that the state-wide databases include approximately
10,200 valid analytical results, from 4,225 community water systems, approximately 10 times
the number of results and 4.2 times the number of systems addressed by the NIRS. The presence
of duplicate analyses, multiple samples from single wells and from multiple wells within systems
also provides important information that is used to estimate the proportions of variability
attributed to various sources, as discussed in Section 5.7 and Appendix B-l.
References Cited in Appendix A.1
U.S. Environmental Protection Agency (EPA 1993, 1995), Uncertainty Analysis of Risks
Associated With Radon in Drinking Water, Office of Science and Technology, April 1993 and
March 1995.
Methods, Occurrence and Management Document for Radon: Appendix A A-5
-------
A.2 Supplemental Data Sets
j
This appendix contains summary descriptions of the data contained in the various
supplemental data sets. A one-page data summary is supplied for each major data set. The data
elements provided in each summary include:
.Data Element
Area Covered
Provider
Citation
Purpose/Objective of Data
Explanation
state or other region
name and affiliation
publication reference, if any
compliance, develop cross-sectional profile, etc.
Types of Data Provided:
Approx. Dates of Sampling
Number of Samples
Number of Systems
Sampling Locations
Sampling Procedures
Analytical Methods
PWSID
Lab ID
Sample Date
System Name
Source Type
Source Name
City/County
GIS Coordinates
Radon Analytical Result
Counting Error
Population Served
Points of Entry
Well Depth
Geological Formation
Types of Analysis
Data Entry, Formatting
Requirements
Comments
time span of sample collection
total
total
W = well, POE = point of entry
IF = Inverted Funnel, O = other
LSC = liquid scintillation
water system identification number
identity of laboratory performing analysis
date individual samples were collected, analyzed
utility or company name
groundwater or surface water
well number, etc.
—
—
corrected for background, pCi/1
counts per minute
(by systems)
number in system
(ft)
name
Single or multiple analysis per source, system; samples over time
from same/different sources, analytical duplicates
Manual data entry, SDWIS matching of PWSID, supply names,
addition of population served data, system/source type
Other concerns, features of data
In the categories "Types of Data Provided", responses in the forms may included "Y" (yes), "S"
(sometimes), indicating that the indicated data elements were always of sometimes present.
Where cells are left blank, this indicates that the data are not provided.
Methods, Occurrence and Management Document for Radon: Appendix A
A-6
-------
SUPPLEMENTARY RADON DATA FORM
AREA COVERED:
Alabama
'rovider
David Grey
USEPA
National Air and Environmental Radiation Laboratory
Montgomery. AL
Purpose/Objective
of Dau Collection
Characterize radon activity levels in Alabama rural water supplies, investigate precision of analytical results
Approx. Dates of Sampling
Number of Samples
Number of Systems
Types of Systems
Sampling Location(s)
Sampling Procedures)
Analytical Method(s)
1991-92
1600
800
W.POE
IF
•LS
Types of Analysis
Single Sample. Analysis per System
Multiple Analyses of Single Samples
Multiple Analyses of Consecutive Samples
Multiple Samples Over Time
Multiple Analyses of Samples From Different Days
Raw Versus Finished Water Comparisons
Types of Data Provided
PWS1D
Lab. ID
Sample Date
System Name
System Type
Source Type
Source Name
City/County
CIS Coordinates
Radon Analytical Result
Counting Error
Population Served
Total Production
Points of Entry
Well Depth
Geologic Formation
Data Entry. Formatting Requirements
Manual Data Entry
SDWIS Matching of PWSID Numbers
SDWIS Matching of Supply Names
Addition of Pop. Served Data
Addition of System/Source Type Data
Other (Specify) Data were provided from 800 systems. Due to
resource limitations, only 300 pairs of samples were entered. Because
system names and PWSIDs were not provided, system size could not
be evaluated and these data were used only to evaluate S*A variance
Comments-
Occurrence Analysis Revised Draft
A-l
-------
SUPPLEMENTARY RADON DATA FORM
AREA COVERED:
Provider
Ciuiion
Purpose/Objective
of Dan Collection
California
Dr. David Storm
California Department of Health Services
Collect, compile, evaluate, and report drinking water quality data.
Approx. Dales of Sampling
Number of Samples
lumber of Systems
Types of Systems
Sampling Location(s)
Sampling Procedures)
Analytical Mcthod(s)
Types of Data Provided
PWSID
Lab. ID
Sample Date
System Name
System Type
Source Type
Source Name
City/County
CIS Coordinates
Radon Analytical Result
Counting Error
Population Served
Total Production
Points of Entry
Well Depth
Geologic Formation
1987-1997
1101
64
C.NTNC
NS
Y
Y
Y
Y
Y
Y
Y
S
Types of Analysts
Single Sample. Analysis per System
Multiple Analyses of Single Samples
Multiple Analyses of Consecutive Samples
Multiple Samples Over Time
Multiple Analyses of Samples From Different Days
Raw Versus Finished Water Comparisons
Data Entry. Formatting Requirements
Manual Data Entrv
SDWIS Matching of PWSID Numbers
SDWIS Matching of Supply Names
Addition of Pop. Served Data
Addition of System/Source Type Data
Y
S
S
S
S
Y
Y
Other (Specify)
Comments: More samples were included in the stale data but not used:
this summary does not reflect the state (ftata not used. Point of entry
data only indicates whether sampled before or after treatment; some data
are samples within the distribution system.
Radon Occurrence Analysis Revised Draft
A-2
-------
SUPPLEMENTARY RADON DATA FORM
AREA COVERED:
Southern California (QA Data)
"rovider
USEPA Files
Southern California Ration Survey
Black&Veatch. Prepared for the Metropolitan Water District of Southern California. January 1990
>urpose/0bjective
of Data Collection
Evaluate radon levels in southern California water systems
Appro*. Dates of Sampling
lumber of Samples
\umber of Systems
Types of Systems
Sampling Location(s)
Sampling Procedures)
Analytical Method(s)
1989
200(100 QA duplicates)
many
Types of Analysis
Single Sample. Analysis per System
Multiple Analyses of Single Samples
Multiple Analyses of Consecutive Samples
Multiple Samples Over Time
Multiple Analyses of Samples From Different Days
Raw Versus Finished Water Comparisons
Types of Data Provided
PWS1D
Lab. ID
Sample Date
System Name
System Type
Source Type-
Source Name
City/County
CIS Coordinates
Radon Analytical Result
Counting Error
Population Served
Total Production
Points of Entry
Well Depth
Geologic Formation
Data Entry. Formatting Requirements
Manual Data Entry
SDW1S Matching of PWSID Numbers
SDWIS Matching of Supply Names
Addition of Pop. Served Data
Addition of System/Source Type Data
Other (Specify):
Comments:
These data were used solely in the evaluation of analytical variability.
Radon Occurrence Analysis Revised Draft
A-3
-------
SUPPLEMENTARY RADON DATA FORM
AREA COVERED:
Provider
Citation
Purpose/Objective
of Data Collection
Connecticut
Nancy McHone
Connecticut Department of Environmental Protection
Natural Resources Center
Hartford. CT
-
Characterize bedrock aquifer radon levels statewide, correlate groundwater and indoor air radon levels
Approx, Dates of Sampling
lumber of Samples
Number of Systems
Types of Systems
Sampling Location(s)
Sampling Procedure!*)
Analytical Method(s)
Types of Data Provided
PWSID
Lab, ID
Sample Date
System Name
System Type
Source Type
Source Name
City.'County
CIS Coordinates
Radon Analytical Result
Counting Error
Population Served
Total Production
Points of Entry
Well Depth
Geologic Formation
7
1000
32C. 968 UNK
C.UNK
W
IF
LS
Y(CWS only)
Y(all W)
Y
Y
Y
Types of Analysis
Single Sample. Analysis per System
Multiple Analyses of Single Samples
Multiple Analyses of Consecutive Samples
Multiple Samples Over Time
Multiple Analyses of Samples From Different Days
Raw Versus Finished Water Comparisons
Data Entry. Formatting Requirements
Manual Data Entry
SDWIS Matching of PWSID Numbers
SDWIS Matching of Supply Names
Addition of Pop. Served Data
Addition of System/Source Type Data
Y
Other (Specify):
Comments: Only 32 samples from 15 CWS. data were not matched to
population served categories. Entire data base was used to evaluate radon
occurrence in CT
Radon Occurrence Analysis Revised Draft
A-4
-------
SUPPLEMENTARY RADON DATA FORM
AREA COVERED:
Provider
Citation.
hirpose/Objective
of Data Collection
Idaho
Janet K. Crockett. Sr. Groundwater Quality Analyst
Groundwater Quality Monitoring Division
Idaho Department of Water Resources
Boise. ID
Idaho Department of Water Resources. Idaho Statewide Ground Water Qualm Monitoring Program - Summary of
Results. 1991 Through 1993. Water Information Bulletin No. 50. Part 2. April 1995
Characterization of statewide groundwater quality
Approx. Dates of Sampling
Number of Samples
Number of Svstems
Types of Systems
Sampling Location(s)
Sampling Procedures)
Analytical Method(s)
Types of Data Provided
PWSID
Lab. ID
Sample Date
System Name
System Type
Source Type
Source Name
Cilv/County
CIS Coordinates
Radon Analytical Result
Counting Error
Population Served
Total Production
Points of Entry
Well Depth
Geologic Formation
1991-1993
1040
64 CWS. 27 NCNTWS
C.N
W
IV?
LS
Y
Y
Y
Y
V
Y
V
Types of Analysis
Single Sample. Analysis per Svstem
Multiple Analyses of Single Samples
Multiple Analyses of Consecutive Samples
Multiple Samples O'er Time
Multiple Analyses of Samples From Different Days
Raw Versus Finished Water Comparisons
Data Entry. Formatting Requirements
Manual Data Entry
SDWIS Matching of PWSID Numbers
SDWIS Matching of Supplv Names
Addition of Pop. Served Data
Addition of System/Source Type Data
Y
Y
Y
Y
Other (Specify). Translated use type ID ("Industriai". "cooling", etc.) into
system types
Commentsi
Radon Occurrence Analysis Revised Draft
A-5
-------
SUPPLEMENTARY RADON DATA FORM
AREA COVERED:
Provider
Citation
Purpose. Objective
of Data Collection
Iowa
Mr. Roy Ney
Iowa Department of Natural Resources
Radon- "f*" in the Source and Finished Water of Selected Public Water Supplies in Iowa A Research Report.
January' 13. 1993, No. 93-1 R. Kellcy & M. Mehrhoff. University of Iowa Hygienic Laboratory
To assess the probable extent of elevated radon levels in both the source and finished water of public water
supplies in Iowa.
Approx. Dates of Sampling
Number of Samples
Number of Systems
Types of Systems
Sampling Location(s)
Sampling Procedure(s)
Analytical Method(s)
Types of Data Provided
PWSID
Lab ID
Sample Date
System Name
System Type
Source Type
Source Name
Cityi'Counry
CIS Coordinates
Radon Analytical Result
Counting Error
Population Served
Total Production
Points of Entry
Well Depth
Geologic Formation
1992
150
t
s
s
Y
Y
Y
Y
Y
Y
Types of Analysis
Single Sample. Analysis per System
Multiple Analyses of Single Samples
Multiple Analyses of Consecutive Samples
Multiple Samples Over Time
Multiple Analyses of Samples From Different Days
Raw Versus Finished Water Comparisons
Data Entry. Formatting Requirements
Manual Data Entry
SDWIS Matching of PWSID Numbers
SDWIS Matching of Supply Names
Addition of Pop. Served Data
Addition of System/Source Type Data
Y
Y
Y
Y
Other (Specify) Best judgement was used when adding population
served data.
Comments: The 9 systems are geologic units (or formations). The name
of the lab and an Iowa map with sampling locations were included in the
report.
Radon Occurrence Analysis Revised Draft
A-6
-------
SUPPLEMENTARY RADON DATA FORM
AREA COVERED:
Provider
Citation
Purpose/Objective
of Data Collection
Approx. Dates of Sampl
Number of Samples
Sumber of Systems
Types of Systems
Sampling Location! s)
Sampling Procedure(s)
Analytical Method(s)
Types of Data Provided
PWSID
Lab. ID
Sample Date
System Name
System Type
Source Type
Source Name
City/County
CIS Coordinates
Kansas
Darrel R. Plummer
Public Water Supply Section. Bureau of Water
fciansas Department of Health and Environment
Topeka. KS
-
Ongoing characterization of radon levels in water supply wells throughout the state
ng
Radon Analytical Result
Counting Error
Population Servec
Total Production
Points of Entry
Well Depth
Geologic Formation
3/91-11/97
250
163
C
W.POE
IF
LS
Y
S
Y
Y
Y
Y
Types of Analysis
Single Sample. Analysis per System
Multiple Analyses of Single Samples
Multiple Analyses of Consecutive Samples
Multiple Samples Over Time
Multiple Analyses of Samples From Different Days
Raw Versus Finished Water Comparisons
Data Entry. Formatting Requirements
Manual Data Entry
SDWIS Matching of PWSID Numbers
SDWIS Matching of Supply Names
Addition of Pop. Served Data
Addition of System/Source Type Data
Y
S
Y
Y
Y
Y
Other (Specify): Matching supply names to SDWIS information was
sometimes difficult. Approximately 35 systems with questionable IDs
were omitted data base used from the occurrence analysis.
Comments:
Radon Occurrence Analysis Revised Draft
A-7
-------
SUPPLEMENTARY RADON DATA FORM
AREA COVERED:
Provider
Citation
hirpoie-'Objective
of Data Collection
Maine
Charles T. Hess
Department of Engineering
University of Maine at Orono
Ration 222 in Potable Water Supplies in Maine: The (jeology. Hydrology. Physics and Health Eflecl* . Land
and Water Resources Center. University of Maine at Orono. September 1979
To evaluate the relationships among radon levels in various geological formations in Maine, estimate potential nsks
associated with groundwater radon exposures
Appro*. Dales of Sampling
dumber of Simples
dumber of Systems
Types of Systems
Sampling Locationfs)
Sampling Procedures)
Analytical Methodfs)
Types of Data Provided
PWSID
Lab, ID
Sample Date
System Name
System Type
Source Type
Source Name
City/County
CIS Coordinates
Radon Analytical Result
Counting Error
Population Served
Total Production
Points of Entry
Well Depth
Geologic Formation
1979
550
450
C.NC.P
Wellhead. POE
IF
LS
s
Y
Y
Y
Y
Y
Types of Analysis
Single Sample, Analysis per Svstem
Multiple Analyses of Single Samples
Multiple Analyses of Consecutive Samples
Multiple Samples Over Time
Multiple Analyses of Samples From Different Davs
Raw Versus Finished Water Comparisons
Data Entry. Formatting Requirements
Manual Data Entry
SDWIS Matching of PWSID Numbers
SDWIS Matching of Supply Names
Addition of Pop. Served Data
Addition of System/Source Type Data
i
s
Y
Y
Y
Other (Specify)
Comments: Only a small minority of the noncommunity systems
could be matched through SDWIS with population served
estimates.
Radon Occurrence Analysis Revised Draft
A-8
-------
SUPPLEMENTARY RADON DATA FORM
AREA COVERED:
Maryland
Provider
Mr. Stephen Poreda
Maryland Department of Environment
*urpose/Objective
f Data Collection
Sampling program of survey to collect radon data in anticipation of proposed regulations.
Approx. Dates of Sampling
dumber of Samples
dumber of Systems
Types of Systems
ampling Localion(s)
Sampling Procedurefs)
Analytical Method(s)
1993-1997
413
244
C. NC
NS
Types of Analysis
Single Sample. Analysis per System
Multiple Analyses of Single Samples
Multiple Analyses of Consecutive Samples
Multiple Simples Over Time
Multiple Analyses of Samples From Different Days
Raw Versus Finished Water Comparisons
Types of Data Provided
PWSID
Lab. ID
Sample Date
System Name
System Type
Source Type-
Source Name
City/County
CIS Coordinates
Radon Analytical Result
Counting Error
Population Served
Total Production
Points of Entry
Well Depth
Geologic Formation
Data Entry. Formatting Requirements
Manual Data Entry
SDWIS Matching of PWSID Numbers
SDWIS Matching of Supply Names
Addition of Pop. Served Data
Addition of System/Source Type Data
Other (Specify)
Comments:
Radon Occurrence Analysis Revised Draft
A-9
-------
SUPPLEMENTARY RADON DATA FORM
AREA COVERED:
Provider
Citation
*urpoje/0bjective
of Dan Collection
Michigan
Mr. Elgar Brown
Michigan Department of Public Health
-
To gather information on radon levels throughout the state
Approx. Dates of Sampling
Number of Samples
dumber of Systems
Types of Systems
Sampling Locationfs)
Sampling Procedures)
Analytical Methodfs)
Types of Data Provided
PWSID
Lab. ID
Sample Date
System Name
System Type
Source Type
Source Name
City/County
CIS Coordinates
Radon Analytical Result
Counting Error
Population Served
Total Production
Points of Entry
Well Depth
Geologic Formation
1991-1992
185
120
C
Y
Y
Y
Y
Y
V
Y
Y
Y
Y
Y
S
Types of Analysis
Single Sample. Analysis per Svstem
Multiple Analyses of Single Samples
Multiple Analyses of Consecutive Samples
Multiple Samples Over Time
Multiple Analyses of Samples From Different Days
Raw Versus Finished Water Comparisons
Data Entry. Formatting Requirements
Manual Data Entry
SDWIS Matching of PWSID Numbers
SDWIS Matching of Supply Names
Addition of Pop. Served Data
Addition of System/Source Type Data
Y
c
s
Other (Specify)
Comments:
Radon Occurrence Analysis Revised Draft
A-IO
-------
SUPPLEMENTARY RADON DATA FORM
AREA COVERED:
Provider
Citation
Purpose/Objective
of Data Collection
Missoula. Montana
Robert Ward
Mountain Water Company
Missoula. MT
Ward. Robert B.. Distrihtaion and Occurrence of Radon in the Missoula Valley Aquifer . University of Montana. Master's
Thesis. 1997
Characterize radon levels, temporal and spatial variability in public water supply wells drawing from the Missoula Valley
aquifer, characterize rate of aquifer recharge from adjacent surface water bodies
Appro*. Dates of Sampling
Number of Samples
Number of Systems
Types of Systems
Sampling Location(s)
Sampling Procedures)
Analytical Method! s)
Types of Data Provided
PWSID
Lab. ID
Sample Date
System Name
System Type
Source Type
Source Name
City/County
CIS Coordinates
Radon Analytical Result
Counting Error
Population Served
Total Production
Points of Entry
Well Depth
Geologic Formation
1993-1994
> 1.500
1
C
W
IF
LS
Y
Y
Y
Y
Y
Y
Y
Y
S
Y
Types of Analysis
Single Sample. Analysis per System
Multiple Analyses of Single Samples
Multiple Analyses of Consecutive Samples
Multiple Samples Over Time
Multiple Analyses of Samples From Different Days
Raw Versus Finished1 Water Comparisons
Data Entry. Formatting Requirements
Manual Data Entry
SDWIS Matching of PWSID Numbers
SDWIS Matching of Supply Names
Addition of Pop. Served Data
Addition of Svstem/Source Type Data
Y
Y
Y
Other (Specify) Data were provided in spreadsheet form requiring minimal
formatting and editing
Comments: Almost all samples were from a single large community system
(Mountain Water System). QA/QC of the data were unusually rigorous, all
analyses were conducted in a single university laboratory using standard
methods. Data were used to evaluate temporal, sampling, and analytical
variability.
Radon Occurrence Analysis Revised Draft
A-ll
-------
SUPPLEMENTARY RADON DATA FORM
AREA COVERED:
Provider
Citation
Purpose/Objective
of Dan Collection
New Hampshire
Bernie Lucy
S'ew Hampshire Department of Environmental Services
Concord. NH
-
Results of ongoing state groundwater monitoring program: objective is to periodically simple every groundwaier system
in the state to determine radon levels
Approx, Dates of Sampling
dumber of Samples
dumber of Systems
Types of Systems
Sampling Locations)
Sampling Procedures)
Analytical Method(s)
Types of Data Provided
PWSID
Lab. ID
Sample Date
System Name
System Type
Source Type
Source Name
City/County
G1S Coordinates
Radon Analytical Result
Counting Error
Population Served
Total Production
Points of Entry
Well Depth
Geologic Formation
1988-1977
>5.000
699C. 602NCNT
C.NCNT
W.POE
IF
LS
Y
Y
Y
Y
Y
Y
Y
Y
Y
Y
N
N
N
N
N
N
Types of Analysis
Single Sample. Analysis per System
Multiple Analyses of Single Samples
Multiple Analyses of Consecutive Samples
Multiple Samples Over Time
Multiple Analyses of Samples From Different Davs
Raw Versus Finished Water Comparisons
Data Entry. Formatting Requirements
Manual Data Entry
SDWIS Matching of PWSID Numbers
SDWIS Matching of Supply Names
Addition of Pop. Served Data
Addition of System/Source Type Data
Y
Y
Y
Y
Other (Specify): Data were supplied as very large Hies containing both CWS.
NCNTWS data. Files were sorted by system type, using state type codes.
inactive sources, sytems were removed from data base: consultation with
provide identified system reporting results for treated water.
Comments: There was some ambiguity about source identities within
some systems. This introduces uncertainty into the evaluation of temporal
variability.
Radon Occurrence Analysis Revised. Draft
A-12
-------
SUPPLEMENTARY RADON DATA FORM
AREA COVERED:
Provider
Citation .
Purpose/Objective
of Data Collection
New York
New York State Department of Health. Bureau of Public Waier Supply Protection
Axelrod, David and Kenneth E. Slade. Report of Statewide Surveillance for Radon in Selected Community Water
Systems: New York State, 1989-1990. New York State Department of Health.
Bureau of Public Water Supply Protection. September 1990.
To determine the distribution of radon concentrations in drinJcing water at community water systems in New York
in anticipation of EPA setting an MCL for radon.
Approx. Dales of Sampling
Number of Samples
Number of Systems
Types of Systems
Sampling Location(s)
Sampling Procedures)
Analytical Method(s)
Types of Data Provided
PWSID
Lab. ID
Sample Date
System Name
System Type
Source Type
Source Name
City/County
CIS Coordinates
Radon Analytical Result
Counting Error
Population Served
Total Production
Points of Entry
Well Depth
Geologic Formation
4/3/89 - 4/26/90
429
429
CWS
Tap
Y
Y
Y
Y
Y
Y
Y
Types of Analysis
Single Sample. Analysis per System
Multiple Analyses of Single Samples
Multiple Analyses of Consecutive Samples
Multiple Samples Over Time
Multiple Analyses of Samples From Different Days
Raw Versus Finished Water Comparisons
Data Entry, Formatting Requirements
Manual Data Entrv
SDWIS Matching of PWSID Numbers
SDWIS Matching of Supply Names
Addition of Pop. Served Data
Addition of System/Source Type Data
Y
Y
Other (Specify)
Comments:
Radon Occurrence Analysis Revised Draft
A-13
-------
SUPPLEMENTARY RADON DATA FORM
AREA COVERED:
Provider
Citation
Purpose/Objective
of Data Collection
Chapel Hill. North Carolina Vicinity
William K_ Diane
Department of Environmental Sciences and Engineering
University of North Carolina
Drane. W.K.. J.H. Hightower. J.E. Watson. Jr. "Variations of 222Rn in Public Drinking Water Supplies. Health fhvsic .
73(6). pp 906-9 II, (1997)
Approx. Dates of Sampling
dumber of Samples
Number of Systems
Types of Systems
Sampling Location(s)
Sampling Procedures)
Analytical Method(s)
Types of Data Provided
PWSID
Lab. ID
Sample Date
System Name
System Type
Source Type
Source Name
City/County
CIS Coordinates
Radon Analytical Result
Counting Error
Population Served
Total Production
Points of Entry
Well Depth
Geologic Formation
1993-1995
>2.000
13
C.N
W
O
LS
Y
Y
Y
Y
Y
Types of Analysis
Single Sample. .Analysis per System
Multiple Analyses of Single Samples
Multiple Analyses of Consecutive Samples
Multiple Samples Over Time
Multiple Analyses of Samples From Different Days
Raw Versus Finished Water Comparisons
Data Entry, Formatting Requirements
Manual Data Entry
SDWIS Matching of PWSID Numbers
SDWIS Matching of Supply Names
Addition of Pop. Served Data
Addition of Svstem/Source Tvpe Data
V
Y
Y
Y
Other (Specify):
Comments:
The investigator used a sampling techniques different from the inverted
funnel technique recommended by EPA. The technique appears to be at
least comparable to EPA's in terms of repeatability.
Radon Occurrence Analysis Revised Draft
'A-14
-------
SUPPLEMENTARY RADON DATA FORM
AREA COVERED
Provider
Citation
Purpose/Objective
of Data Collection
Ohio
Kathleen Pinto. Environmental Specialist at Ohio EPA. Division of Drinking and Ground Vtater
(614)644-2752
Radon Sampling Program -- Radon - 222 Groundwater Study Report.
To analyze the impact of an MCL at 200 pCi/L on Ohio's public water supplies.
Approx. Dates of Sampling
Number of Samples
Number of Systems
Types of Systems
Sampling Location(s)
Sampling Procedurc(s)
Analvtical Method(s)
Types of Data Provided
PWSID
Lab. ID
Sample Date
System Name
System Type
Source Type-
Source Name
Cirv/Countv
CIS Coordinates
Radon Analvtical Result
Counting Error
Population Served
Total Production
Points of Entry
Well Depth
Geologic Formation
12/86- 12/88
229
219
RESIDENTIAL
Y
Y
Y
Y
Types of Analysis
Single Sample. Analysis per System
Multiple Analyses of Single Samples
Multiple Analyses of Consecutive Samples
Multiple Samples Over Time
Multiple Analyses of Samples From Different Days
Raw Versus Finished Water Comparisons
Data Entry. Formatting Requirements
Manual Data Entry
SDWIS Manching of PWSID Numbers
SDWIS Matching of Supply Names
Addition of Pop. Served Data
Addition of Svstem/Source Type Data
V
Y
Y
Y
Other (Specify)
Comments:
Radon Occurrence Analysis Revised Draft
A-15
-------
SUPPLEMENTARY RADON DATA FORM
AREA COVERED
Pennsylvania
Provider
Jeny Rupert. Pennsylvania Department of Environmental Protection.
(717)772-2847
Citation
Internal Report: The Occurrence of Radon in Pennsylvania (. 'ommunm Groumlvaler Systems. Division of Drinking
Water Management Pennsylvania Department of Environmental Resources. 5/27'93
Purpose/Objective
of Dili Collection
Impact of EPA's proposed MCL on Pennsylvania Systems
Appro* Dates of Sampling
Number of Samples
Number of Systems
Types of Systems
Sampling Location(s)
Sampling Procedure^)
Analytical Methodfs)
9/92 - 4/93
986
493
CWS
Types of Analysis
Single Sample. Analysis per System
Multiple Analyses of Single Samples
Multiple Analyses of Consecutive Samples
Multiple Samples Over Time
Multiple Analyses of Samples From Different Days
Raw Versus Finished Water Comparisons
Types of Data Provided
PWSID
Lab, ID
Sample Dale
System Name
System Type
Source Type
Source Name
City/County
CIS Coordinates
Radon Analytical Result
Counting Error
Population Served
Total Production
Points of Entry
Well Depth
Geologic Formation
Data Entry. Formatting Requirements
Manual Data Entry
SDW1S Matching of PWSID Numbere
SDWIS Matching of Supply Names
Addition of Pop. Served Data
Addition of System/Source Type Data
Other (Specify)
Comments:
Radon Occurrence Analysis Revised Draft
A-16
-------
SUPPLEMENTARY RADON DATA FORM
AREA COVERED:
Provider
Citation .
Purpose/Objective
of Data Collection
South Carolina
South Carolina Department of Health and Environmental Control
Columbia. South Carolina
(through Max Kukoy). American Water Works Association. Washington. DC)
Results of Special Radon Study of Public Water Supply Wells in South Carolina
David Price. P.E.. Bureau of Drinking Water Protection. SCDHEC. September 30. 1991
Characterize distribution of radon in groundwater systems of various sizes throughout the state.
Approx. Dates of Sampling
Number of Samples
dumber of Systems
Types of Systems
Sampling Location(s)
Sampling Procedures)
Analytical Method(s)
Types of Data Provided
PWSID
Lab. ID
Sample Date
System Name
System Type
Source Type
Source Name
City/County
CIS Coordinates
Radon Analytical Result
Counting Error
Population Served
Total Production
Points of Entry
Well Depth
Geologic Formation
2/91-8/91
. 242
I22(C)
C.NCNT
W.POE
IF
LS
Y
Y
Y
Y
Y
Types of Analysis
Single Sample. Analysis per System
Multiple Analyses of Single Samples
Multiple Analyses of Consecutive Samples
Multiple Samples Over Time
Multiple Analyses of Samples From Different Days
Raw Versus Finished Water Comparisons
Data Entry. Formatting Requirements
Manual Data Enrrv
SDWIS Matching of.PWSID Numbers
SDWIS Matching of Supply Names
Addition of Pop. Served Data
Addition of System/Source Type Data
Y
Y
Y
Y
Y
Other (Specify): Removed sample results from tanks, water towers
Comments: Data were well-stratified across system size categories.
Additional data from Clemson University could not be included in the
occurrence analysis because systems could not be identified.
Radon Occurrence Analysis Revised Draft
A-17
-------
SUPPLEMENTARY RADON DATA FORM
AREA COVERED:
Provider
Ciution
'urpose/Objective
Approx, Dales of SimpI
vi umber of Simples
dumber of Systems
Types of Systems
Sampling Location(s)
Sampling Procedures)
Texas
Mr. Ron Beardon
Texas Natural Resource Conservation Commission
To gather information on radon distribution throughout the state.
ng
Types of Dua Provided
PWSID
Lab. ID
Sample Date
System Name
System Type
Source Type
Source Name
City/County
CIS Coordinates
Radon Analytical Result
Counting Error
Population Served
Total Production
Points of Entry
Well Depth
1995
195
138
CP
NS
Y
Y
Y
Y
Y
S
Y
NS
Types of Analysis
Single Sample. Analysis per System
Multiple Analyses of Single Samples
Multiple Analyses of Consecutive Samples
Multiple Samples Over Time
Multiple Analyses of Samples From Different Days
Raw Versus Finished Water Comparisons
Data Entry. Formatting Requirements
Manual Data Entry
SDWIS Matching of PWSID Numbers
SDWIS Matching of Supply Names
Addition of Pop. Served Data
Addition of System/Source Type Diita
Y
S
Y
Y
Other (Specify)
Comments!
Radon Occurrence Analysis Revised Draft
A-18
-------
SUPPLEMENTARY RADON DATA FORM
AREA COVERED:
Provider
Citation
Purpose/Objective
of Data Collection
Washington (State)
Jim Hudson
Division of Drinking Water
Washington Department of Health
Olympia. WA
'.
Statewide characterization of radon on drinking water supplies
Approx. Dates of Sampling
dumber of Samples
lumber of Systems
Types of Systems
Sampling Location(s)
Sampling Procedures)
Analytical Method(s)
Types of Data Provided
PWSID
Lab. ID
Sample Date
System Name
System Type
Source Type
Source Name
City/County
CIS Coordinates
Radon Analytical Result
Counting Error
Population Served
Total Production
Points of Entry
Well Depth
Geologic Formation
1992-1995
365 fGW
63
C.N
W.POE
IF
LS
Y
Y
Y
Y
S
Y
Y
Types of Analysis :
Single Sample. .Analysis per Svstem
Multiple Analyses of Single Samples
Multiple Analyses of Consecutive Samples
Multiple Samples Over Time
Multiple Analyses of Samples From Different Days
Raw Versus Finished Water Comparisons
Data Entry. Formatting Requirements
Manual Data Entry .
SDWIS Matching of PWSID Numbers
SDWIS Matching of Supply Names
Addition of Pop. Served Data
Addition of System/Source Type Data
Y
5
Y
Y
Y
Y
Other (Specify):
Comments:
Radon Occurrence Analysis Revised Draft
A-19
-------
SUPPLEMENTARY RADON DATA FORM
AREA COVERED:
Provider
ritation
•urpose'Objcctive
Appro*. Dates of SampI
dumber of Simples
sumber of Systems
Types of Systems
Sampling Location!!)
Sampling Procedures)
Mark Wilson
Wisconsin Department of Natural Resources
-
Evaluation of radon levels in water supplies throughout the state
ng
Types of Data Provided
PWSID
Lab. ID
Sample Date
System Name
System Type
Source Type
Source Name
City/County
CIS Coordinates
Radon Analytical Result
Counting Error
Population Served
Total Production
Points of Entry
Well Depth
1990-1996
1256(CWS)
535
C
W.POE
IF
LS
Y
Y
Y
Y
Y
Y
Y
Y
Types of Analysis
Single Sample. Analysis per Svstem
Multiple Analyses of Single Samples
Multiple Analyses of Consecutive Samples
Multiple Samples Over Time
Multiple Analyses of Samples From Different Days
Raw Versus Finished Water Comparisons
Data Entrv. Formatting Requirements
Manual Data Entrv
SDW1S Matching of PWSID Numbers
SDWIS Matching of Supply Names
Addition of Pop. Served Data
Addition of System/Source Type Data
Y
S
Y
Y
Other (Specify): Data had to be extensively reformatted prior to analysis
Comments:
Approximately 350 samples were not included in the final data base
because sytems could not be identified.
Radon Occurrence Analysis Revised Draft
A-20
-------
APPENDIX B
STATISTICAL METHODS
-------
-------
Appendix B. Statistical Methods
This appendix describes three specific statistical approaches that were used to evaluate
radon occurrence data. Appendix B. 1 addresses the "EM" maximum likelihood method for
estimating summary statistics for radon data sets that have "censored" values. Appendix B.2 and
B.3, respectively address distributional (parametric) and non-distributional (non-parametric)
methods used to derive confidence intervals for the estimated proportions of water systems above
potential regulatory levels.
B.I MLE Approach to Estimating Summary Statistics for Radon Data Sets
The NIRS data in particular, and a few of supplemental data sets, contained substantial
proportions of censored observations. In order to make the data generalizable to the U.S.
population of groundwater systems, it was assumed that the distribution of the data was log-
normal, and a separate mean and variance of the logarithm of radon levels was calculated for each
facility size group in each data set. The mean and variance parameters of the log radon levels
were estimated by a maximum likelihood estimation (MLE) method, a procedure proven to
provide estimators with good statistical properties for reasonably large samples. The specific
approach that was used was the "EM algorithm"(Dempster et al. 19 ). This algorithm
provides a convenient and robust approach to developing MLE estimates of distributional
parameters. The following discussion describes the calcukition of distributional parameters for the
radon data, estimation of the proportions of systems that would; exceed regulatory limits, and
development of confidence intervals for those proportions. •
Let x be the natural logarithm of the radon activity and let D be the natural logarithm of
the minimum detection or reporting limit (MDL of MRL). The exact value of x will be unknown
for a "non-detect", i.e. if x is less than equal to D. Let I(x) = 1 if x is a detect and I(x) = 0 if x is
a non-detect. Based on the available data, the log-likelihood for the single observation x is:
/(*;
= /(*) log [c|>((*-u)/o)/o] +[l-/(x)] log
(B-l)
where (z) is the standard normal
cumulative probability distribution function:
V/2lc
<£(z) = f («) du.
(B-2)
The log-likelihood for a single observation is the logarithm of the probability density for
detected values, and is the logarithm of the probability of being at most D for non-detects (since
for a non-detect, the only available information is that the log concentration is at most D). The
log-likelihood for the entire data set is the sum of the individual Jog-likelihoods:
Methods, Occurrence, and Monitoring Document for Radon: Appendix B
B-l
-------
Log-likelihood = S(u,o2) = £ /(*,; u,o2).
(B-3)
The maximum likelihood method estimates the true parameter values u and o2 by finding
the values of n and o2 that maximize the total log-likelihood, S (keeping the observed data fixed).
Instead of a direct maximization, we used the efficient "EM algorithm" which is an
iterative two-step procedure (Dempster et al.). Suppose that our initial estimates of u and o2 are
uokj and (o2)old. If we knew all the concentration values, including values for the non-detects, then
all the I(x) values would be 1 and the complete data log-likelihood (ignoring constant terms)
would be:
Complete data log-likelihood = Y) -log(o) -
M
(x. - \i)2
— - — -
2o2
(B-4)
The E (expectation) step of the algorithm calculates the expected value of the complete data log-
likelihood given the available data, where the expectation is over the log-normal distribution with
the initial estimates a0" and (a2)oW. Let E; be the expected value of X; aid let S; be the expected
value of X; squared (given below). Then the expected complete data log-likelihood equals:
N
N
Expected log-likelihood = -N log o -
1=1
i=\
(B-5)
2o2
The M (maximization) step maximizes this expression for all fi and o2, to give the updated
estimates
(o2)
S, -
E
(B-6)
.new _
where:
.new _
/=!
(B-7)
I i The
iterative scheme starts with any reasonable initial estimated parameter values, such as the mean
and variance of the logged detected concentrations. The E and M steps are used to iteratively
Methods, Occurrence, and Monitoring Document for Radon: Appendix B
B-2
-------
update these estimates until the changes in the parameter values are sufficiently small. The
calculation of Ej and Sj is as follows. If the i'th (logged) observation, Xj, is a detect, then the
expected value is also X;. If the i'th observation is a non-detect, then the expected value is the
conditional mean given that X is at most D:
E(X | non-detect) = E(X \ XzD) = f
00/fl'v/27C
dx.
(B-8)
Integrating by parts gives the expression:
Ej - Xp if xt is a detect,
Ei = u°w - a°ldxF, ifxf is not a detect, where
p _
(B-9)
(B-10)
(B-ll)
A similar calculation for the expected squares gives
Sj ~ (x,)2, if xf is a detect,
S, = (rt2 + (o2)
- o°
), ifxt is not a detect.
(B-12)
(B-13)
The EM algorithm was implemented on Excel© spreadsheets for each data set and system
size stratum. The estimated censoring limits were used to estimate D for each data set. The
spreadsheets were designed to perform 20 iterations of the calculation algorithm starting from
user-specified initial estimates of the sample mean and variance; For all of the data sets and size
strata, the EM estimates of the log mean and log variance converged very rapidly. Usually,
estimates of the log mean and log standard deviation were stable to the fourth decimal place (one
part per 10,000) between the fifth and tenth iteration. The final estimates of log mean and log
variance were quite insensitive to the initial estimates used as inputs to the first iteration of the
algorithm.
B.2 Calculation of Proportions of Systems Above Radon Levels and Confidence Limits
on Proportions (Distributional Approach)
The NIRS and states data were used to estimate the proportions of facilities, p, that
would exceed specified levels assuming that a single measurement was taken from a single
Methods, Occurrence, and Monitoring Document for Radon: Appendix B
B-3
-------
sources from each system.' The classical approach to estimating confidence intervals in this case
is to estimate the standard deviation of the estimated proportion, p, using the formula
SD(p) =
n
(B-14)
The 95 percent confidence interval is then given by the large sample approximation
p ± 1.96 x SD(p).
(B-15)
This formula for SD($) is strictly correct only when the observed fraction of facilities
exceeding the level (u) is used to estimate the population fraction (of all facilities of the given
size). Since the proportions of systems exceeding potential regulatory levels were estimated
using a fitted log-normal distribution, the standard deviation of the estimated proportion will not
be given by the above formula. The above formula for SD(£) will tend to overestimate the
standard deviation because the maximum likelihood estimates have smaller variances than the
empirical estimates if sample sizes are relatively large. A more exact calculation uses the
asymptotic distribution of the maximum likelihood estimators, since the EM algorithm was used
to derive maximum likelihood estimates of the mean and variance of the logged concentrations,
which in turn were used to estimate the proportions. We therefore used a more exact calculation
method for determining SD(£) and its associated confidence interval.
As in above, let x be the natural logarithm (denoted by log) of the concentration and let D
be the natural logarithm of the minimum detection limit (MDL). The exact value of x will be
unknown for a non-detect, Le. if x is less than equal to D. Let I(x) = 1 if x is a detect and I(x) =
0 if x is a non-detect. Then the log-likelihood for the single observation x is
/(*; u,o2) = I(x) log [(j>((x-u)/a)/o] +[1-/(*)] log
(B-16)
is where <|>'(z) is the standard normal probability density function, and (z) is the standard normal
cumulative probability distribution function:
*G?) = f ({>(«) du.
(B-17)
1 The proportions of systems exceeding regulatory levels are estimated using the
cumulative normal distribution as described in Section 5.2 and shown in Equation B-26.
j
Methods, Occurrence, and Monitoring Document for Radon: Appendix B
B-4
-------
Standard maximum likelihood theory shows that for a large number, N, of facilities, the
asymptotic variance-co variance matrix of A and d2 is the inverse of the expected information
matrix:
Cov(A,d2)
Cov(A,62)
N
Tl 1 ^12 I
T2\ T22)
(B-18)
The matrix T is the expected products of the partial derivatives of the log-likelihood with respect
to the parameters:
(B-19)
where X is the log concentration (observed if X > D, otherwise a non-detect). X is assumed to
have a normal distribution. Standard analytical calculations, using integration by parts, show that
the components of T are given by
Tu =
T22 =
- P + d
2o3
(B-20)
(B-21)
(B-22)
where d = (D-u,)/o and P = O(d). P is the probability that an observation is not detected, i.e.
below the MDL. Inverting the 2x2 matrix T gives the variances and covariances:
[22
Var(d2) =
T22 ~
11
n T22 - T122]
Cov(A, 62) =
12
(B-23)
(B-24)
(B-25)
The exceedance probability, p(u), that a facility will exceed level u is estimated by
(B-26)
Methods, Occurrence, and Monitoring Document for Radon: Appendix B
B-5
-------
For large samples, this estimated exceedance probability is given by the first order terms of
the Taylor expansion around the true parameter values u. and o2:
- x /- V^
a) + (u - u.)(—)
o
(B-27)
H are given by:
where y and
7 =
(B-28)
Thus, the variance of the exceedance probability is given (asymptotically) by
Var(p(uj) =
4a4
5 62).
(B-29)
The estimated variance of p(u) is calculated by substituting the estimated values of \i and o2for
the unknown true values.
Finally, we obtain the 95 percent confidence interval for p(u) as:
p(u) ± l.<
(B-30)
The above formulae were derived for the exceedance probability estimates based on a
single value for each facility of a given size. It assumes that each system has only one source of
water (well) and that the decision concerning compliance is made on the basis of a single sample.
In Appendix C and Section 5.8, we discuss the impacts of different sources of variance on these
estimates and how the presence of multiple sources and the use of multiple samples would effect
our estimates of systems exceeding potential radon regulatory levels.
B.3 Estimation of Confidence Intervals (Distribution-Free Method)
For comparative purposes, estimates of confidence intervals on the proportions of systems
exceeding specified radon levels were also estimated using a distribution-free approach (Johnson
and Katz 1969). This method calculates the upper and lower confidence limits, Pu and P,,
respectively on p*, the estimated proportion of systems above a certain level. (These estimates are
often referred to as Clopper-Pearson confidence limits).
Methods, Occurrence, and Monitoring Document for Radon: Appendix B B-6
-------
Again, assume a sample of N systems, and define p* = Probfx, <, x*] = n/N, where n is the
number of systems exceeding x*. Then Pu, the upper confidence limit, is given by
Pu =
n
n + (N - «
* | vl,v2)
(B-31)
where:
vl
v2
P(w*|vl,v2) =
2(N-n)
The F-distribution with vl, v2 degrees of freedom = p = the desired
confidence level
Pi, the lower confidence limit, is:
P • =
(n
(w* | vl, v2)
N - n + (n + l)P -' (w* | vl, v2)
(B-32)
where:
vl
v2
2(N-n+l)
2n
These confidence limits were calculated for the various system size strata in the NIRS data
to help confirm the reasonableness of the parametric calculations described above. As discussed
in Section 5.8, there is generally good agreement between the parametric and empirical
confidence limits on the proportions of systems exceeding specified radon levels.
B.4 Comparison of distributional and Non-Distributional Confidence Limits
Exhibit B-l presents a tabular comparison of calculations of distributional and non-
distributional upper and lower confidence limits on the proportions of systems exceeding potential
regulatory limits for two data sets. The first data set is the comprised of all of the NIRS data for
very very small systems. As noted in Section 5.4, the lognormal distribution provides the best fit
to these data, but the A-D statistic suggests a significant deviation for lognormality. The other
data set is the combined data from the New York radon survey. These data also are best fit by a
lognormal distribution, but the goodness-of-fit test results again suggest that lognormality may be
discarded with a high degree of confidence.
Methods, Occurrence, and Monitoring Document for Radon: Appendix B
B-7
-------
Exhibit B-l. Radon Distributional Data Compared to Parameteric and Non-
Parametric Confidence Limits on Proportions of Systems Exceeding Potential
Regulatory Levels
Radon Level,
pCi/1
300
500
700
1000
2000
4000
Radon Level,
pCi/1
300
500
700
1,000
2,000
4,000
1. NIRS Data (Very Very Small Systems)
Maximum
Likelihood
Lower
Confidence
Limit
47.2%
61.3%
69.8%
77.6%
88.9%
95.4%
Non-
Distributional
Lower
Confidence
Limit
50.0%
63.9%
71.3%
78.0%
87.4%
93.9%
NIRS Data
54.3%
68.1%
75.2%
81.5%
90.1%
95.8%
Non-
Distributional
Upper
Confidence
Limit
58.6%
72.0%
78.8%
84.7%
92.5%
97.3%
Maximum
Likelihood
Upper
Confidence
Limit
56.0%
69.5%
77.4%
84.5%
93.8%
98.1%
2. New York State Data (All Systems)
Maximum
Likelihood
Lower
Confidence
Limit
43.7%
60.5%
70.6%
79.6%
91.6%
97.3%
Non-
Distributional
Lower
Confidence
Limit
38.7%
58.0%
73.0%
85.1% «
93.8%
96.9%
New York Data
42.5%
61.8%
76.4%
87.7%
95.5%
98.1%
Non-
Distributional
Upper
Confidence
Limit
46.3%
65.5%
79.6%
90.1%
96.9%
99.0%
Maximum
Likelihood
Upper
Confidence
Limit
52.5%
68.8%
78.2%
86.3%
95.7%
99.1%
It can be seen from the Exhibit that both sets of confidence limits 'include the observed
numbers of systems measured as exceeding most of the potential radon regulatory levels. The
single exception is maximum likelihood lower confidence limit (5 percent) on the proportion of
systems exceeding 300 pCi/1 in the New York data set. The confidence limit is 43.7 percent,
while the observed value is 42.5 percent. Since the table includes a total of 28 comparisons of
confidence limits to measured values, on the order of one "miss" is not unexpected.
In all cases, the non-parametric confidence limits include the measured values. This is
because these confidence limits take the measured values as their starting points. There do not
appear to be any systematic differences between the distributional and non-distributional
confidence limits.
Methods, Occurrence, and Monitoring Document for Radon: Appendix B
B-8
-------
References Cited in Appendix B
Dempster, A.P., Laird, N.M., and Rubin, D.B, (1977) "Maximum Likelihood Estimation from
Incomplete Data Via the EM Algorithm", Journal of the Royal Statistical Society, Series B, 39,
1-38.
Johnson, N.L., and Katz, M., Discrete Distributions, 1969.
Methods, Occurrence, and Monitoring Document for Radon: Appendix B B-9
-------
-------
APPENDIX C. SOURCES OF VARIABILITY IN RADON MONITORING
STUDIES OF GROUNDWATER SYSTEMS
-------
-------
C. SOURCES OF VARIABILITY IN RADON MONITORING STUDIES OF
GROUND WATER SYSTEMS
Radon activity measurements are known to vary within regions, among sources in
individuals water systems, among groups of systems, and over time. In order to accurately
predict the proportions of systems and sources that might be out of compliance with radon
regulatory levels, the sources and magnitude of this variability needs to be understood. In this
Appendix, we provide a summary and analysis of the available data related to the major sources
of variability in measured radon levels from groundwater supplies. We begin by identifying
sources of variability in radon data sets and establishing a basis; for estimating contributions of
the individual sources to the total variability of the data. This model of variability (actually
variance) is used in Section 5.8 of the MOM document to develop estimates of the proportions of
systems that would exceed potential regulatory levels. We also analyze data from the NIRS and
the supplemental data sets that provide information regarding the relative contribution of the
individual sources to total radon variance.
C.I A Variance Apportionment (ANOVA) Model For Evaluating Radon Data Sets
Assume a data set consisting of individual measurements of radon levels (Xj) from (N)
different groundwater systems (the NIRS is a good example.) Assume that some of the systems
are small, and obtain their water from only one source (well), while others obtain water from
multiple sources. Samples are taken at different times by different personnel (although using
standardized methods) and subject to analysis in different laboratories (again using the standard
liquid scintillation procedure).
When the data from this set are evaluated, they are found to be (roughly) lognormally
distributed, as are most of the data sets we
i /-.,->. identified. The calculated log mean value of
Log Mean = T] ——-—— all the observations is:
N
Log Variance =
(C-l)
y (In X, - In X)2
N
(C-2)
A commonly used measure of the
variation seen within the data set is the variance
of the data, or in this case the log variance:
Methods, Occurrence, and Monitoring Document for Radon: Appendix C
C-l
-------
The log variance simply is simply the sum of the squared differences between the logs of
the individual observations and the mean log radon value. If more than one factor contributes to
the differences between the individual observations and average value, then all of these factors
show contribute to the magnitude of the variance.
That is:
A useful property of variance is that variances of more than one variable are additive.
Var(A) + Var(B) = Var (A + B)
(C-3)
This equation indicates that if A and B are two independent variables, then the variances
of the sum of the variables is equal to the sum of their variances. This relationship holds for the
sum of any number of variables whose distributions are statistically independent (do not depend
on one another). In the following sections, we use this relationship, and the assumption of
independence, to help separate and quantify the contributions of individual sources to the overall
variance in sets of radon observations.
C.1.1 Identification of Sources of Radon Variance
Five types of variance contribute to the overall variance in radon data sets. For the
purpose of the following analysis, we refer to them as:
• Var(SYS) — This represents the "true" inter-system variability in radon levels, that is, the
variability due to differences in location, geologic setting, size, etc;.
• Var(W) — This is the variability among different sources (wells) within a given system.
! I
• Var(T) — This factor is the time-variability of radon levels in a given source or system.
For purposes of this analysis, we use this symbol to refer to long-term differences in
radon levels rather than short-term variability that may result from daily water use
patterns, etc..
• Var(S) — Refers to all sampling variability, or the variations in radon levels associated
with the process of taking water samples, transferring them to the laboratory, and any
other manipulations up to the point at which the sample enters the scintillation counter.
• Var(A) — The last factor we address is variability associated with the analysis of radon
samples themselves. This includes statistical counting error, as well as any other factors
affecting the precision of the radon analysis. As will be discussed further below, the
magnitude of this variability is measured by examining the differences in the results of
duplicate analyses of the same sample.
Methods, Occurrence, and Monitoring Document for Radon: Appendix C
C-2
-------
In the language used above, we define each of these contributors factors as "variables",
the combined variance of which is equal to the sum of their individual variances. For purposes
of this analysis, we assume that the expected value (average) of each of these variables is zero,
that is, they do not introduce any systematic bias into the estimates of radon levels. Using the
additivity property discussed above, we can define the overall variance of the observations in our
radon data set as:
Var(Total) = Var(SYS) + Var(W) + Var(T) + Var(S) + Var(A)
(C-4)
This equation defines the relationships among the variances of the data set composed of
individual radon measurements to the total variance in the data set. Based on this relationship, we
can estimate the total variance of a data set if we have other data sets that allow us to estimate the
contributions to variance of the individual sources. More importantly, if we have data sets that
allow us to estimate the magnitude of the some of the contributions, we can use them estimate
the magnitudes of the others.
Developing estimates of these contributions (for example, temporal variability) is
important in and of itself. We can use such information, for example, example, to determine
whether each specific source is an important contributor to overall variance in radon levels. In
addition, separating the contributions to variance will enable us to develop estimates of the
numbers of systems above potential regulatory levels, given specific assumptions about
monitoring requirements. ;
C.1.2 Estimating Contributions to Variance
As noted above, estimating the magnitudes of individual contributions to variance is
possible if data sets are available that allow the contributions of one or more sources to be
separated from those of the others. This is possible, to some extent, with the various data sources
that are available related to radon occurrence in drinking water systems. The types of data that
are available and the sources that can be characterized by their analysis are summarized in
Exhibit C-l. More complete descriptions of the data sets can be found in the following
sections or in Appendix A.
Each type of data set provides information on one or more contribution to variance. More
importantly, each type of data (with the exception of that listed in the bottom row of the table) is
obtained under conditions that cancels out or excludes the impact of one or more sources of
variance. For example, when duplicate analyses are performed of a single sample, this provides
direct information on the magnitude of Var(A), the analytical variance, and excludes
contributions from any other source (e.g., the scintillation counter does not know where the
sample came from). Interpreting most of the other types of data is not so simple, however.
Analyses of multiple samples taken from the same well at the same time provide information
concerning the magnitude of combined sampling and analytical variability, Var(S) +Var(A). The
independent impacts of S and A on the total variance, however, cannot be determined unless
Methods, Occurrence, and Monitoring Document for Radon: Appendix C
C-3
-------
duplicate analyses are also performed. In this case, the effects of these two variables can be
separated by subtraction. This type of data set (same-time samples from the same source) does
not provide any information about contributions from inter-system, inter-well, and temporal
variability (Var(SYS), Var(W), and Var(T)), however.
Exhibit C-l. Variance Information Provided by Different Types of Radon Studies
Type of Data
Duplicate analyses of single
samples
Multiple samples from the
same wells taken at the
same time
Multiple samples from the
same wells taken over time
Samples taken from
different wells in the same
system at the same time
Samples taken from
different wells in the same
systems over time
Single samples from
different systems
Sources of
Variance Held
Constant !
SYS, W, T, S
SYS, W, T
SYS, W
SYS,*T
SYS
None
Type of Variance
Information
Included
A
S,A
S,A,T
S, W,A
S,W,A,T
SYS, S, W, A, T
Available Studies
Hess, Glick,
Black& Veatch,
Pennsylvania,
Texas
Grey, Ward,
Drane, New York
McHone, Ward,
Drane, California,
Maryland, New
Hampshire
Ward, California,
New Hampshire,
Maryland,
Michigan,
Wisconsin
California, Ward,
New Hampshire
NIRS, States
Notes:
1. A = analytical, S = sampling, SYS = system, T = temporal, W = variance among wells
2. Sources identified in text, reference list.
The rest of the rows in Exhibit C-l show the types of variance information included in,
and excluded from, specific types of data sets, and examples of these types of data sets identified
Methods, Occurrence, and Monitoring Document for Radon: Appendix C
C-4
-------
during the occurrence analysis. In the following sections, thesis data will be described in detail
and used to characterize the approximate contributions to variance from each of the individual
sources at a national level.
C.2. Analytical Variance
Exhibit C-2 identifies six data sets that provide data from duplicate analyses of the same
samples, and thus directly measure analytical variance, or Var(A). The studies are presented in
roughly chronological order, with the earliest example being the data from Maine gathered by
Hess (1979). In this study, a total of 139 duplicate results were reported, the average radon
activity in individual sources ranged from less than 100 pCi/1 to greater than 20,000 pCi/1. The
log variance for individual duplicate samples1 was found to increase with decreasing radon
levels, and was more ten times higher for samples with average radon less than 1,000 pCi/1 than
for samples with average radon levels above 5,000 pCi/1. This would be consistent with reduced
analytical reproducibility and accuracy at low radon levels. For the data set as a whole, the
average relative standard deviation (RSD)2 of the paired samples was 8.7 percent, but the average
RSDs for the analyses below 1,000 pCi/1 were all above 10 percent, which is the suggested
reproducibility criterion in EPA's Standard Method (MOM document, Section 2). Altogether,
approximately 29 percent of the paired analyses had RSDs above 10 percent.
The log variance for duplicate analyses was calculated as (ln(X,)-ln(X2))2/2.
2 The Relative Standard Deviation, which is equivalent to the coefficient of variation, is
the standard deviation of the measurements divided by the average of the measurements.
Methods, Occurrence, and Monitoring Document for Radon: Appendix C C-5
-------
The distribution of log variance from the paired analyses in the Hess (1979) data is
summarized in the top row of Exhibit C-2. The average log variance between paired analyses
was 0.023, and the median was 0.0033. The mean log variance in this data set exceeds the 75th
percentile value, reflecting the fact that a large proportion of the total varismce was contributed
by a relative small number of high-variance sample.
The next row of Exhibit C-2 summarizes the estimates of analytical variance from the
Southern California Radon Survey (Black & Veatch 1990). The QA report for this study
includes the results of 100 duplicate analyses. The average RSD for the peiired samples is 5.7
percent, corresponding to a log variance of 0.017. Because the study covered a much narrower
range of radon levels than the data from Maine, the trend of increasing variance with decreasing
radon levels was not as apparent in this study. In general, the log variance was much lower than
in the Hess study, and the distribution was much less skewed. The mean and median log
variance were almost the same, 0.0017 and 0.0016, respectively.
Exhibit C-2. Distribution of Analytical Variance from Six Data Sets
Data Set1
Hess, et ai. (Maine)
Black&Veatch
(Southern California)
Click (NIRS)
Pennsylvania
Texas
Ward (Missoula, MT)
Average
Percentile Log Variance
5th
1.57E-05
3.00E-06
-
0.00002
0.00000
0.00057
0.00012
25th
0.00044
0.00020
-
0.00048
0.014
0.0022
0.0035
50th
0.0033
0.0016
-
0.0023
0.052
0.0045
0.013
Mean
0.023
0.0017
0.0043
0.032
0.087
0.0057
0.026
75th
0.011
0.0045
-
0.010
0.063
0.0081
0.019
95th
0.058
0.016
-
0.079
0.248
0.0137
0.083
Notes;
1. Sources defined in text, reference list
The next source of data related to analytical variance is the Mid-Term Quality Assurance
Report for the NIRS (Glick, 1985). This report is of interest primarily because it gives direct
information into the analytical precision achieved by the laboratories conducting analyses for the
survey that has become the major national source of data on radon in groundwater systems. The
analysis reports an RSD of 5.3 percent, corresponding to a log variance of 0.0043, for 25
duplicate samples analyzed by liquid scintillation. The average radon activity in the samples
ranged from less than 100 pCi/1 to 4,270 pCi/1, but no information was presented as to the
Methods, Occurrence, and Monitoring Document for Radon: Appendix C
C-6
-------
variability as a function of radon level. No data on the distribution of analytical precision (e.g.,
the proportion of analyses with RSDs above 10 percent) was provided.
Two of the other supplemental data sets also provide information on analytical
variability, for Pennsylvania (PADER 1993), and Texas (TNRCC 1998), respectively. The
former study provided results from duplicate analyses from a cross-section of 488 water systems
throughout the state, while the latter data set provided multiple analyses of samples from 120
systems. The distribution of log variance for duplicate results in the Pennsylvania data set is
very similar to that seen in the Hess data from Maine. The variance seen in the Texas data has a
broader distribution, and both the mean and median are substantially higher than those seen in
either the Maine or Pennsylvania data.
The final source of analytical variance data is a study by Ward (1997) of radon levels in
supply wells for a single water utility in Missoula, Montana. The data the used to evaluate
analytical variability consist of the results of quadruplicate analyses of 32 samples taken from the
same well on the same day. As shown in Exhibit C-2, the average and median log variance of
the quadruplicate analyses are low relative to those for the two cross-sectional state studies, but
comparable to that seen in the NIRS and Southern California QA reports.
The bottom row of Exhibit C-2 averages the results of the various studies. The average
median log variance from these studies 0.013 and the average log mean variance is 0.026. Based
solely on the mean and median results, it appears that the Hess and Pennsylvania data are the
most "typical" of the data sets, with the California and Ward data having the lowest variances,
and the Texas data set having the highest variance.
C.3 Combined Sampling and Analytical Variance
After reviewing the studies of analytical variance, it would be highly desirable to have
data sources that provided information solely on sampling variance. This is, however, not
possible because once separate samples are taken, separate analyses must be performed. The best
that can be done, therefore, is to review studies that evaluate the results of duplicate samples
(analyses of samples taken from the same source at the same time using the same methods.) This
provides estimates of the combined magnitude of sampling and analytical variability. Sampling
variability can then be estimated indirectly using Equation C-4 or its variants.
Exhibit C-3 summarizes the results of several studies of duplicate samples. The first of
these involved the analyses of paired samples from rural water systems in Alabama collected in
1991 and 1992 under the supervision of personnel from EPA's Eastern Environmental Radiation
Facility (Gray 1998). Approximately 800 pairs of samples were collected, but due to resource
limitations, only 300 (100 consecutive samples each from the beginning, middle, and end of the
sampling period) were analyzed for variability. Samples were collected using EPA's standard
methods, and mailed to EERF for analysis by liquid scintillation counting. The results.of this
study summarized in the top row of Exhibit C-3. the median and mean log variance (which
Methods, Occurrence, and Monitoring Document for Radon: Appendix C C-7
-------
include contributions from both sampling and analytical variability) were 0.011 and 0.058. This
again reflects a skewed distribution of log variances, with a relatively few samples contributing a
large proportion of the total variability.
The average log variance was found to increases by about 200-fold from the highest to
lowest radon levels, with a corresponding increase in RSD of about 15-fold. The variance of this
data set is reduced by about 50 percent if the ten worst pairs of samples are removed, and the
proportions of samples with RSD above 10 percent is 25.8, approximately the same as seen in
the data set from Maine.
The next source of data on sampling and analytical variability is another set of data
gathered by Ward (1997). In this case, the data that were used were 320 pairs of duplicate (same-
day) samples taken from 32 wells in the same Missoula, MT system previously mentioned. As
can be seen from the exhibit, the log variance seen in this study was considerably lower than that
seen in the Alabama data set. The median and mean log variance were 0.0054 and 0.0067,
respectively.
Another recent study also provides information on combined sampling and analytical
variance, this time from a group of repeatedly-sampled wells in central North Carolina (Drane, et
al. 1997) . Five samples were taken from each well per sampling event by allowing water to run
directly from the source down the side of a scintillation vile, and analyzed using standard liquid
scintillation methods. This sampling method is different from that recommended by EPA and is
Exhibit C-3. Sampling Plus Analytical Variability Estimates from Four Data Sets
Data Set1
Grey (Alabama)
Ward (Missoula, MT)
Drane (North Carolina)
New York
Average
Percentile Log Variance
5th
0.0001
2.8E-05
0.0008
0<2>
0.0003
25th
0.0017
0.0013
0.0014 .
0.00032
0.0012
50th
0.011
0.0054
0.0022
0.0024
0.0052
Mean
0.0058
0.0067
0.0039
0.019
0.0088
75th
0.038
0.015
0.0042
0.011
0.017
95th
0.17
0.049
0.011
0.074
0.075
Notes:
1. Sources defined in text, reference list
2. Not detectible within analytical precision
claimed by the author to be superior in terms of consistency and precision of analytical results.
Wells were sampled as many as 39 times, giving a total of 304 sampling events. Most of the
sampled wells serve small subdivisions or mobile home parks, and all were within convenient
driving distance of Chapel Hill. Although the wells were all located in the Piedmont province,
average radon levels varied from 136 pCi/1 to over 36,000 pCi/1.
Methods, Occurrence, and Monitoring Document for Radon: Appendix C C-8
-------
The sampling results from the individual wells are summarized in the third row of Exhibit
C-3. The average variability seen in this study (log variance = 0.0039) is smaller than that seen
in any of the others. The median and various percentile values are also low compared to most of
the other studies. The data showed a general, but not entirely consistent, increase in sampling
and analytical variability with decreasing radon levels across wells. The pooled analyses from
the various wells all have RSDs less than 10 percent and only 11.8 percent of the individual
sampling events have RSDs at this level or above. Part of the explanation for the low variance
may be the improved sampling method employed, but, as was the case in the Montana study, this
effort also involved sampling from a relatively small geographic area under well-controlled
conditions, and analyses in a single university laboratory.
The final data set used to evaluate sampling and analytical variance was the results of a
survey of radon levels in community water systems undertaken by the New York State
Department of Health in 1989-1990 (NYDOH 1990). Duplicate samples were analyzed from
411 of 424 of the community water systems that were surveyed. The distribution of log variance
of the duplicate sample results, as shown in the exhibit, is comparable to that from the other
studies at the low end, but the mean variance (0.019) is somewhat higher than that of the other
studies.
The bottom row of Exhibit C-3 averages the results of the various studies.
C.4 Combined Sampling, Analytical and Temporal Variance
The next set of studies to be discussed are those that provide information about the
variability in radon levels from multiple samples from the same sources, taken over time. Thus,
the studies in Exhibit C-4 provide information about the combined magnitude of analytical,
sampling, and temporal variability.
Methods, Occurrence, and Monitoring Document for Radon: Appendix C
C-9
-------
The first three data sets are derived from state-wide
California, and Maryland. The data from New Hampshire (NHDES
surveys in New Hampshire,
1998) provide information
Exhibit C-4. Estimates of Sampling Plus Analytical Plus Temporal Variance
From Six Data Sets (Samples From The Same Sources Over Time)
Data Set 1
New Hampshire
California
Maryland
McHone (Connecticut)
Ward (Missoula, MT)
Drane
Mean of All Studies
Mean of Statewide Studies
Percentile Log Variance
5th
0.0006
0.0003
0.0003
0.009
0.0021
0.0010
0.0022
0.0004
25th
0.018
0.0034
0.0026
0.010
0.0068
0.0032
0.0073
0.0079
50th
0.074
0.012
0.037
0.019
0.0098
0.0080
0.027
0.041
Mean
0.35
0.19
0.24
0.20
0.024
0.012
0.17
0.26
75th
0.26
0.045
0.15
0.16
0.015
0.013
0.11
0.15
95th
1.60
0.56
0.52
0.76
0.078
0.032
0.59
0.89
Notes:
1. Sources defined in text and reference list
on over 5,000 radon measurements from over 1,300 community and non-community
groundwater systems in the state, gathered between 1988 and 1997. A total of 849 wells serving
community water systems were sampled more than once on different dates. The data from
California (CADHS, 1998) include approximately 1,500 sampling results from 524 sources in 75
community systems. Of these, 316 sources are sampled more than once on different dates. The
data from Maryland (MDOE 1998) come from 107 community systems, with multiple samples
(usually two) taken from 43 sources.
The mean and median log variance from individual sources in New Hampshire were 0.35
and 0.074, respectively. These comparatively large values reflect the large combined sampling,
analytical, and temporal variability associated with the high proportion of small systems and
bedrock wells in this data set. The distributions of log variance from the two other states are
lower than that seen in the New Hampshire data. The average log variance in same-source,
different-day samples from the California data was 0.19, the mean value from the Maryland data
was 0.24.
i ;
The other three data sets that provide data on multiple samples from the same sources, in
contrast, concentrate on much smaller geographic areas. The first of these is a study of five
private residential wells in Connecticut (McHone 1993). This study is included in our analysis,
even though it reports data from residential wells, because it is one of the best sources of long-
Methods, Occurrence, and Monitoring Document for Radon: Appendix C C-10
-------
term variability data for individual wells in the New England region. The study addresses both
very short-term variations (hours to days) as well as longer-term (weekly) fluctuations in radon
level. The five residential wells that were sampled were located in two different geologic
provinces in Connecticut. Radon activities in the various wells fell into three ranges. Radon
levels in the first well ranged from approximately 3,300 pCi/1 to 6,000 pCi/1. Three wells had
intermediate radon levels, from about 12,000 pCi/1 to 31,000 pCi/1. Finally, analyses from one
well, in a different geologic unit, indicated radon levels from approximately 33,000 pCi/1 to
600,000 pCi/1. All of these values are greater than those typically seen in water supply wells
outside New England or other high-radon areas. Long-term variability in radon levels in the five
wells was evaluated by taking weekly samples from August 1991 through August 1992.
Ward (1997) also studied the long-term variability of radon levels in water supply wells
in Missoula, MT. His study (in addition to the other data previously discussed) monitored radon
levels in 32 wells over the period of approximately two years. An average of 39 samples
(maximum 57) was taken from each well, which with one exception, drew from the same alluvial
aquifer. The radon levels in all the wells averaged between 250 and 1,000 pCi/1. The final data
set with multiple samples from the same sources is the study by Drane (also previously
discussed). This study evaluated the changes in radon activity over approximately two years in
14 water supply wells serving very small systems, with an average of 20 sampling events per
well.
The distributions of log variance for these studies are shown in the various rows of
Exhibit C-4. While the study by McHone shows an average log variance similar to that seen in
the three state-wide data sets, the variances seen in the more recent studies of limited areas by
Ward and Drane are much lower. These lower variances may results in part from more carefully
controlled sampling and analyses in the latter two studies, but the smaller spatial areas covered
may also have something to do with it as well. The relatively high variances seen in the McHone
study may be associated with the fact that all of the wells studied were low-capacity wells
drawing from bedrock aquifers, with higher inherent variability in radon levels. While
substantial short-term (hours to days), as well as long-term variability was seen in this study,
there were no consistent indications of seasonal variations in radon levels.
One useful feature of the data set developed by Drane is that the availability of multiple
samples per sampling event (generally five each), as well as the results of multiple analyses per
sample, allows the estimation of the variance contributions of sampling and analytical error and
true temporal variability. This process is illustrated in Exhibit C-5. The second column of the
table shows the log variance among the daily sampling events for the 14 wells. The values for
individual wells range from 0.0018 for Well 13 to 0.047 for Well 2, with a combined average log
variance of 0.0142. The second column of the table shows the average sampling and analytical
variance for each well. The log temporal variance (the log variance in daily radon levels minus
the weighted sampling and analytical variance) is shown in the next-to-last column of the table.
The estimated temporal variances for the individual wells range from 0.0002 to 0.0462, with a
weighted average-log variance of 0.0135. :
Methods, Occurrence, and Monitoring Document for Radon: Appendix C C-11
-------
This pattern indicates that, in this study, the contribution of sampling and analytical
variability to overall variance in radon levels is relatively small compared to that from temporal
Exhibit C-5. Calculation of Temporal Variance from Drane, et al. (1997) North Carolina Community
Water Systems Data
Well
Number
1
2
3
5
6
7
8a
8b
9
10
11a
11b
12
13
Weighted
Average
Log Variance in
Daily Geometric
Means, Var(S+A) +
Var(T)
0.017
0.047
0.014
0.0078
0.0069
0.0026
0.012
0.024
0.012
0.0021
0.0047
0.0023
0.0072
0.0018
0.0142
Average Log
Variance of Daily
Analyses, Var(S+A
0.0019
0.0032
0.00090
0.013
0.010
0.0025
0.0011
0.0018
0.0013
0.00069
0.0050
0.010
0.0015
0.0022
0.0035
Analyses Per
Sampling
Events
5
5
5
5
5
5
5
5
5
5
5
5
5
5
-
Number of
Sampling
Events
36
36
39
22
20
23
23
5
16
23
14
4
14
10
285
Log Tempora
Variance
Var(T)
0.017
0.046
0.014
0.0053
0.0049
0.0021
0.012
0.023
0.012
0.0020
0.0037
0.0002
0.0069
0.0014
0.0135
Geometric
Mean Radon
Level, pCi/l
36,089
24,382
13,194
1,845
182
514
3,252
1,072
2,232
2,662
352
136
1,061
406
10,348
I I
variability. In addition, the estimated temporal variance is quite low compared to those implied
by the results of the other studies. (This issue will be discussed in more detail below.)
The last two rows of Exhibit C-4 show the average and median log variances across all
the studies. Averages are calculated for all of the data sets, as well as just for the state-wide data
sets, to show differences between the two types of studies. If the McHone study of residential
wells is excluded, the average log variance for the state-wide studies would be approximately 10
times higher than that for the more localized (and more recent) studies.
C.5 Combined Sampling, Analytical, and Between-Well Variance
The next type of data set reports radon levels from different sources in the same systems,
taken at the same time. These data sets contain information on the combined variability
associated with analytical and sampling error and provide "snap shots" of the variability in radon
levels in multiple wells in the same system ("intra-system variability") at the same point in time.
We identified only two data sets than contain such information. Data received from
Wisconsin (WIDNR, 1998) reported radon analytical results from over 530 community
groundwater systems. Of these, in 121 cases multiple sources (wells) from the same systems
Methods, Occurrence, and Monitoring Document for Radon: Appendix C
C-12
-------
were sampled on the same date. Between two and 20 sources (average = 3.1) were sampled per
system. Data from Michigan (MIDPH, 1998) provided information on same-day samples from
multiple sources at 43 groundwater systems. Usually, no more:than two sources were sampled in
each system.
The variances distributions for these data sets are reported in Exhibit C-6. The variance
distributions are not dissimilar to those seen in the previous section. The median and mean log
variance for systems in Wisconsin were 0.13 and 0.30, respectively, and the median and mean
log variances for Michigan were 0.051, and 0.18. These data sets did not provide information
from duplicate samples, so the contribution of sampling and analytical variability to the total
variance could not be estimated.
C.6 Combined Sampling, Analytical, Temporal, and Intra-System Variance
Exhibit C-6. Estimates of Combined Sampling Analytical, and Between-Well Variance
(Samples Taken from Different Sources in the Same Systems on the Same Date)
Data Set1
Wisconsin
Michigan
Average
Percentile Log Variance
5th
0.0046
0.0010
0.0028
25th
0.044
0.0057
0.025
Median
0.13
0.051
0.091
Mean
0.30
0.18
0.24
75th
0.30
0.25
0.27
95th
1.14
0.78
0.96
Notes: ;
1. Sources defined in text and in reference list '
Exhibit C-7 provides summaries of the variance distributions for those data sets that carry
information related to combined analytical, sampling, temporal, and intra-system (between-well)
variability. These studies report the results of radon analyses from different sources in the same
systems taken at different times. All of the data sets in this group except one are subsets of data
sources that have been previously described. First, the supplemental data sets for California,
New Hampshire, and Maryland to identify and evaluate the variability among the sources in the
same systems over time. In addition to the state data sets, the other study that provided
information on combined analytical, sampling, temporal, and intra-system variability that was
Ward's extensive study of a single system in Missoula, MT.
All three of the state-wide data sets have very similar log variance distributions, the mean
log variance among sources being 0.30, 0.50, and 0.53, for California, New Hampshire, and
Maryland, respectively. In contrast, the log variance estimate from the single Missoula system is
only 0.066. Again, average percentile and mean log variance values are presented both for all
data sets, and for just the statewide data.
Methods, Occurrence, and Monitoring Document for Radon: Appendix C
C-13
-------
C.7 Variance from AH Sources
Exhibit C-7. Estimates of Sampling Plus Analytical Plus Temporal Plus Intra-Systen
Variability from Four Data Sets (Samples from Different Sources at Different Times
in the Same Systems)
Data Set1
California
New Hampshire
Ward (Missoula, MT)
Maryland
Average of All Data Sets
Average of State-Wide Studies
Percentile Log Variance
5th
0.0050
0.0016
~
0.0019
0.0029
0.0029
25th
0.041
0.030
—
0.016
0.029
0.029
50th
0.13
0.11
—
0.075
0.11
0.11
Mean
0.30
0.50
0.066
0.53
0.35
0.42
75th
0.34
0.38
-
0.36
0.36
0.36
95th
1.32
2.24
-
2.29
1.95
1.95
Notes:
1. Sources defined in text and in the reference list
The last piece of data needed to estimate the contribution of the individual sources of
variance (using Equation C-4) is an estimate of the total combined variance from all sources. As
discussed in Section C.I, estimates of total variance may be developed by analyzing the variance
of individual samples taken from different systems over time. Exhibit C-8 summarizes the log
variance estimates for all of the 16 state-wide supplemental data sets and for the NIRS.
: ! I
This table includes log variance estimates from the seven studies discussed in detail in
Section 5.5 that have substantial data related to systems if all sizes, as well as from studies that
do not provide information on all system size strata. The studies are ranked in order of the total
log variance for all systems combined, and the distributions of the variances are summarized in
the six rows at the bottom of the exhibit. Variance values from six of the studies are entered in
bold face. This designation is applied to all data sets where the system-wide radon levels
reported in Section 5.5 of the MOM used to calculated variance were derived by averaging the
results from multiple sources, or samples, or by averaging radon measurements over time. These
values have been adjusted to offset the reduction in total variance that results from the averaging
process. As discussed in more detail in Section 5.8, talcing multiple samples cancels out a large
portions of variance arising from the specific sources. For example, taking the average of
duplicate analyses greatly reduces the analytical component of variance; averaging multiple
samples from different sources greatly reduces the variance contribution from inter-well
variability. (This procedure is
equivalent to going back to the original data set and calculating variance across all radon
measurements, without averaging the measurements at individual sources, etc.)
Methods, Occurrence, and Monitoring Document for Radon: Appendix C
C-14
-------
In some cases (such as New Hampshire), adding this variance back into the data set
greatly increases the overall variance compared to that reported in Exhibit 5-14, because much
variance is lost through calculating averaged system radon values. In other cases, such as New
Exhibit C-8. Estimates of Variance From Ail Sources (Single Samples From
Different Systems) ;
Data Set1
IDAHO
MICHIGAN
KANSAS
IOWA
MAINE
WISCONSIN
OHIO
TEXAS
CONNECTICUT
CALIFORNIA
WASHINGTON
NEW YORK
NIRS
PENNSYLVANIA
NEW HAMPSHIRE
SOUTH CAROLINA
MARYLAND
5th Percentile
25th Percentile
Median
Mean
75th Percentile
95th Percentile
System Size
ALL
0.57
0.78
0.84
0.86
0.94
0.98
0.98
0.98
1.02
1.02
1.36
1.47
2.01
2.10
2.28
2.36
2.44
0.74
0.94
1.02
1.35
2.01
2.38
WS
—
0.88
1.07
3.24
0.32
1.06
—
0.85
—
—
2.38
1.49
0.76
2.10
1.90
0.92
2.82
0.59
0.88
1.07
152
2.10
2.99
VS
—
1.10
1.17
0.61
1.01
1.05
—
1.00
—
3.16
1.30
2.16
0.72
2.70
2.01
2.44
2.60
0.68
1.02
1.24
1.64
2.37
2.86
S
. —
0.70
•0.75
0.72
,0.95
0.98
0.91
1.20
—
0.32
0.85
1.03
±52
1.68
3.07
1.78
1.42
0.59
0.80
0.98
179
1.47
2.17
M
—
0.58
0.19
0.80
0.72
0.74
0.91
0.75
—
1.75
1.27
0.79
1.95
1.84
1.20
2.13
1.92
0.46
0.75
0.91
177
1.79
2.00
L
—
0.57
0.49
0.00
1.23
0.48
1.21
0.52
—
0.62
1.10
0.72
2JJ6.
1.20
3.19
1.09
4.40
0.33
0.54
1.09
,7.26
1.22
3.56
Notes:
1. Sources defined in text and in reference list
York and Pennsylvania, only a small change (log variance increases of about 0.09) occurred.
Methods, Occurrence, and Monitoring Document for Radon: Appendix C
C-15
-------
The corrected total variances summarized in Exhibit C-8 provide the basis for estimating the
contributions of individual variances sources, as discussed in the following section.
C.8 Estimates of Variance Contributions
: ! ! !
As discussed in Section C.I, Equation C.4 and variations thereon can be used to estimate
the contributions of individual sources to the overall variance seen in radon data sets. This
section describes the procedure was used to estimate the relative contributions from the various
sources to the total variance in a "typical" data set that captures the contributions from all of the
sources. This approach makes use of the estimates of the contributions to log variance discussed
in Sections C.2. through C.I.
Exhibit C-9 tabulates information related to variance contributions discussed in previous
sections. The left-hand column of the exhibit indicates the type(s) of variance that were
estimated through analysis of the various types of data sets, and indicates the table where the data
are summarized. The four right-hand columns of the exhibit summarize the central portions
(25th-75th percentile) of the distributions of log variance for the various types of data sets.
Each row in Exhibit C-9 identifies data sets that provide information about specific
sources of variance in the radon data. Note that, despite the fact that the reported variances are
averages of results from multiple studies, estimates do not all come from the same studies. Thus,
in using these date to estimate variance contributions, we are implicitly assuming that the
Exhibit C- 9. Estimates of Log Variance Contributions From Different Studies
Type of Variance
A (Exhibit C-2)
A (Ward)
S (Ward)
S,A (Exhibit C-3)
S,A,T (Exhibit C-4,
State-Wide Studies)
S,A,W (Exhibit C-6)
S,A,W,T (Exhibit C-7,
Statewide Studies)
SYS, S,A,W,T
(Exhibit C-8)
Percentile Estimate
25th
0.003
—
—
0.0012
0.008
0.025
0.029
0.94
MEDIAN
0.013
—
—
0.0052
0.041
0.091
0.11
1.02
MEAN
0.023
0.0057
0.0029
0.0088
0.26
0.24
0.42
1.35
75th
0.011
—
—
0.017
0.15
0.27
0.36
2.01
Methods, Occurrence, and Monitoring Document for Radon: Appendix C C-16
-------
behavior of radon variance in the different studies are consistent.
Our approach to estimating typical variance contributions from each source is
summarized in Exhibit C-10. At the top, we put our best (average) estimate of total log variance
(1.35) in a data set that reflects contributions from all sources of variance. Below it, we provide
the equations used to develop estimates of the individual variance contributions. A good
estimate of sampling and analytical variance Var(S+A) is the value 0.009, derived both from
Var(S) Var(A) in the Ward study, as well as from the other six studies of sampling and analytical
variance discussed in Section C.3. The studies of analytical variance themselves (Section C.2)
give a higher value (about 0.02), but some of these studies are rather old, and it the more recent
well-controlled studies seem to do better than this.
Exhibit C-10.Estimation of Contributions to Variance from Different Sources
Source of Variance
Total (All Sources)
Sampling and
Analytical
Temporal
Among Wells
Among Systems
Typical Magnitude of Contribution to
Log Variance (Method of Estimation)
~ 1.35 (Measured)
~ 0.009 (Measured)
(S+A+T)-(S+SA)~0.19
(S+A+T+W) - (S+A+T) -0.14
(S+A+W+T) - (S+A+T) -0.16
(S+A+W) - (S+A) ~ 0.23
(SYS+S+A+W+T) - (S+A+W+T) ~ 0.93
(SYS+S+A+W+T) - (S+A) - (W) - (T) ~
0.93
Typical Proportion
of Total Variance,
percent
100
-0.6
13-18
12-17
-69
A typical contribution of temporal variance, Var(T), is calculated in two ways. First we
subtracting log Var(S+A) from the log variances of the data sets expressing sampling, analytical,
and temporal variability. Second, we subtract log Var(S+A+W) from Var(S+A+W+T). Since
this exercise was conducted using average values of quantities that can vary over orders of
magnitude, the relatively close agreement of the estimates of temporal variance derived by these
two approaches (0.19 versus 0.14) using different data sets is quite encouraging, and suggests
that the additive model for variance calculations is appropriate for this data. Similarly, Var(W),
the variance contribution from variations between wells, and Var(SYS), the variance across
Methods, Occurrence, and Monitoring Document for Radon: Appendix C C-17
-------
systems, are also calculated using two sets of relationships. In each case, the estimates of the
total variance contributions are quite consistent.
The last column of Exhibit C-10 summarizes the approximate proportional contribution
of each variance source to the overall variance seen in radon data sets we evaluated. As
expected, sampling and analytical variance contribute only a small proportion (less than one
percent) of the total variance. We estimate that temporal variability within single wells typically
accounts for between 13 and 18 percent of the variance in the data sets evaluated, and a similar
proportion (12-17 percent) is accounted for by variation in radon levels among wells within
systems. Variations among systems related to geographic and hydrogeologic setting account for
the bulk (about 69 percent) of the variance seen in the data sets we evaluated.
Within their limitations, these'results are generally consistent with the previous
knowledge about patterns of radon occurrence in the U.S. We see a large-scale geographic
affect, with levels in New England, the Appalachian Region, and isolated areas of the west
having average radon levels far greater than those in the other areas. In addition, short-range
spatial variability can also be significant. Between-well variability accounts for an average of
about 15 percent of total variability in a typical data set. Similarly, temporal variability is also
important, accounting for about the same average proportions of variance as variability between
wells. For any given data set, the relative contributions of these sources of variance could differ.
The values derived here are intended to be typical of data sets that are representative of large,
geologically diverse regions, such as states, which have a significant proportion of multi-well
systems.
These results have important implications for the development of monitoring plans for
establishing radon compliance. As discussed Section 5.8, the relatively large contributions of
well-to-well and temporal variability argue strongly for taking multiple samples from more than
one well per system spread over time, if representative levels of radon in groundwater systems
are to be established accurately. On the other hand, given the small variance contribution of
analytical and sampling variability, it would appear that beyond normal QA/QC considerations,
performing duplicate analyses or taking duplicate samples from the same source at the same time
may not improve the accuracy of estimates of long-term radon levels very much.
C.9 References
Black & Veatch, (1990) Southern California Radon Survey, Prepared for the Metropolitan Water
District of California, January.
California Department of Health Services, (CADHS 1998), Radon monitoring data provided by
Dr. David Storm.
Drane, W.K., E.L. York, J.H. Hightower, and J.E. Watson, Jr., (1997), "Variation of 222Rn in
Public Drinking Water Supplies", Health Physics, 73(6). 906-911, December; See also Drane,
Methods, Occurrence, and Monitoring Document for Radon: Appendix C
C-18
-------
William K., Variation of222Rn in Drinking Water in Drinking Water Supplies, master's thesis,
University of North Carolina, Chapel Hill, NC, 1996.
Glick, E.M, (1985) Mid-Term Quality Assurance Report for the National Inorganic and
Radiologic Survey, September. !
Grey, David, (1998) Radon data packages for Alabama rural water systems, USEPA National Air
and Environmental Radiation Laboratory, Montgomery, AL. ;
Hess, C.T., (1979) Radon-222 in Potable Water Supplies in Maine: The Geology, Hydrology,
Physics, and Health Effects, Land and Water Resources Center, University of Maine at Orono,
September. :
Maryland Department of the Environment, (MDOE 1998), Data provided by Mr. Stephen
Poreda, Drinking Water Program.
McHone, N.W., and Thomas, M.A., (1993) Temporal Variations in Private Well Radon and
Radium Levels in Specific Connecticut Geological Formations and Effect on Indoor Radon,
Connecticut Department of Environmental Protection Natural Resources Center.
Michigan Department of Public Health (MIDPH 1998), Radon monitoring data received from
Mr. Elgar Brown. :
New Hampshire Department of Environmental Services, (NHDES 1998) Radon monitoring data
from New Hampshire supplied by Mr. Bernie Lucy, January.
New York State Department of Health (NYDOH 1990), Report of Statewide Surveillance of
Radon in Selected Community Water Systems 1989-1990, Bureau of Public Health Protection,
September.
Pennsylvania Department of Environmental Resources, (FADER 1993), The Occurrence of
Radon in Pennsylvania Community Water Systems, Division of Drinking Water Management
Texas Natural Resources Conservation Commission (TNRCC 1998), Radon data from Texas
community and non-community systems were provided by Mr. Ron Bearden
Ward, Robert B., (1997) The Distribution and Occurrence of Radon in The Missoula Valley
Aquifer, Master's thesis, University of Montana. •,
Wisconsin Department of Natural Resources. (WIDNR, 1998), Radon monitoring data supplied
by Mr. Mark Wilson.
Methods, Occurrence, and Monitoring Document for Radon: Appendix C C-19
-------
-------
Appendix D: Estimated Numbers of Community Groundwater
Systems Exceeding Potential Regulatory Levels
by Region, State, and Size
Methods, Occurrence, and Monitoring Document for Radon: Appendix D
-------
Appendix D.I. Proportions of Systems Exceeding
Levels in the Eight NIRS regions
Potential Radon Regulatory
Methods, Occurrence, and Monitoring Document for Radon: Appendix D
-------
.e
^
(/) CO
c "55
o >
4= 0)
™0
* i
CO ^
3 **
2 «
o 1
•^ H
3 Q.
1 >
fS
0 <
•R »
w —
(A >
1 3 ;
j^
^^B [
"O o)
iS E
&_
^ o
^_ i
Q i1
*- O
S -i
X
UJ
1
-2
CO
c
V
0)
(0
c
i
(/r
c
o
'5>
0)
Q£
(0
O)
LU
O
State/Region:
Scenario:
0 0
l§-t
^ '*'
£ 0 —
.i°£
< CM a
2> o —
|°-t
(0 ~
£ > o
3 0 D.
« ^ O
CO N.
in —
£ §1 o
^ o a.
V) J2 o
CO 10
1 >§
® o a-
W JD O
>> < 0
CO co
w —
£ S> o
0) | Q.
W -Q o
^^ ^t ^^
CO T""
(0
n
»- >»
CO
II
^« {Q ^"^
.2 i s
^£ i
CO t CO
<
1
^
<
I
1
r
c
c
c
c
c
T
c
c
h
^
T1
u
p
o
a
a
c
o
R
CO
1
CO
CO
DO
T—
3)
X)
f)
r-
O
n
^
•s*.
o
M
M"
M
o"
D
•^.
t"
•>
0
3"
3
•>
^
•>
O
If)
o
IO
co
m
CO
CO
CM
m
^
CD
CD
CO
CD
CO
CM"
o
CD
CO
0
0
co_
co"
0
in
m
CM
co
CO
co
CD
co
oo
CM
CD
in
in
oo
CM_
CO
D
o"
5
o
CO
3
n
o
in
co
^
r"
O5
OO
CM
CO
a>
oo
in
"*"
o
0
o
o
T-
o
o
o~
o
0
CM
h.
CD
CO
OO
m
o
§
1
A
CO
CO
CO
in
CO
co
*~
1
CM
xj"
s-
-------
J2
0)
0)
c
o
CO
CD
D)
2
o
E
0)
G)
C
O
r*
to J2
E °>
"to _i
>* .
? o
undwatel
MCL/AM
O "Z3
IM ***
^1
'£ °
§ o-
E §
E o
O .Q
0 <
UH
0
c
o
'•E
o
Q.
O
Q.
•a
IS
'"to
LLJ
CM"
I
Q
la
IE
X
LU
w)
0)
Q.
E
(S
CO
*J
Q.
Q
1
re
t§
Q
+•»
.Q
to
o
u
Q>
C "0
1 £
J2 «J
BS £i
Q. C
Q. Q>
< 0
State/Region:
Scenario:
o
c °
I •* s
-SoO
£.§ °-
to .2
o
« 0
-2 *« o
W > D.
>» o
CO .Q
0
§
2, > Q.
CO .Q
o
M o
£ 2T!
S S 0
>, o °-
rf
W o
E w S
*•• § O
J.S tt
o
(0 o
£ co ^
*^ ^ c^
to* -^
<
o
flj fl) *~
"S o Q.
W-" J3
-f
<
(0
co
0)
System Size
(Population Se
CO
CO
en
T —
0
OO
^~
o
CD
CD
CO
CM
O>
?5_
•*"
0
o
CM"
f**.
f**»
CM"
IO
"re
CO
0>
£ 0
> ?
in
in
CO
CM
in
CO
GO
O
T—
T—
o>
in
CO
T~
m
in
O)
CD
CM"
S"
o
in
o
T-
15
E
to
0>
1
o
OO
CM
CD
S
CM
CM
OO
in
o
CO
CO
o
0
1 •
o
m
-
CM
CD
in
o>
OO
CM
CO
oo
CO
CD
••3-
0
o
to"
o
T"
o
o
CM
OO
CO
CM
CO
OO
O>
CM
CM
t
O
CO
CO
1 ' Medium 1
o
o
0
o"
0)
o
Large/Verv Lai
o
-
in
o
o
CM
CO
CO
CO
in
0
o
o
o
1
T-
o
o
o
o
o
o
o
o
-
•r—
O
O
0_
p
A
O
0)
CM
OO
O
CD
T—
O
0
TO-
CO
in
CO
CM"
CO
in
co"
o
CO
O)
5
in"
Is
0
^
-?
o
^~
N-
CM
vO
O>
CO
CO
O>
CO
CO
CO
in
CT>
cri
^P
o
o
o
T
5
o
Proportion of "
Systems
-------
JO
O
^
0)
o
•0
CO
a:
0)
en
re
^v
0)
^>
^^
^
d)
•?
D)
C
o
1
_J
£
CO (0
i >
+* Q
JW MOJ
C/) — J
^^
m IS
13
"O O
1 «
O *5
fl
E *"
E o
O J3
0 <
*o
CO
c
o
r
o
a
o
CL
"«
E
.~
to
UJ
CO
1
ci
*->
1
X
UJ
in
a>
a.
to
CO
IB
"EL
Q
Quarterly
j£
O
"S
.Q
'd
4«l
V)
Q
|
0)
•o
§
IE
*- •<-•
•• c
(B 0)
O O
I&iate/Kegion:
Scenario:
0
o
o
it
s > —
lEo
CO < Q.
O
o
o
2 CM"
S S>
"(0 O ••==
>, J3 O
CO < Q.
O
O
o_
c ^~
m 0>
^J >
(O O -"^=
>. J3 O
CO < Q.
O
IS
& , ^3 O
CO < Q.
If
£ > —
>.5 5
CO < Q.
O
w g
w o S
>.A O
CO < Q.
O
£ 0)
*"* rt ^^
>,^i O
CO < Q.
(0
i
*•»
5L
CO
15
*^
H
¥
o>
System Size
(Population S
o
UD
CM
T—
O)
o
oo
CO
oo
en
0
n
N
v^
5
E
n
£•
s>
>
£•0
si
00
CM
OO
O
T~
o>
CM
m
o>
co
05
o
m
S
CO
CM
oo
in
00
o
IO
'r
T-
To
CO
0)
>
T
c
u
u
T-
o
T-
t-
a
T-
cc
a:
T~
0
||
^ S
O 0
0 0
-- oo
M CM
•> CD
J CO
- o
) 1-
- CM
CO
T—
CM
O
*"?.
^~
O
0
o
CO
03
CM
i
',
1
f
CO
CO
v—
CD
T —
O O
j£ o"
1
a>
E
n
"
1 Large/Very
o
r*** *
i^j
CM °
5 «
| *
m _
8! ^
c
c
c
o) v
o
o
1
T™
o
o
o
0 $
T_
O CO
T —
fv.
•^-
o
CN
0 §
CD
3 CO
) CM
- m
CM"
in
. CD
i*-_
CM
S
••2
0
o
o
o
0
-
T~
T-
ra
5
5
c
o ..
E §
&l
O V)
- >l
-------
to
CO
0)
E
O
T!
CO
tr
0)
D)
S
0)
^
E
flj
,*•*
O)
c
O
MJ
r*
w J2
E »
(D ^
Wi
«J
*•« i
;. o
o> S
iundwat
1 MCL/A
2 «
•I *!^
£Z
^* jjj
'n o
1 a>
E o
O .Q
0 <
0
to
c
o
r
o
Q.
o
D_
•a
1c
c
._
"to
UJ
Q
1
CO
"5.
E
re
CO
&
"re
"Q.
3
Q
>i
1
C§
.0
o
"to
a
o
o
•o
*• ?
(0 *
re H-
o —
O 2
— Tz
3 0)
O 0
State/Region:
Scenario:
o
0
o
iS
(D >
>> X5 O
CO < Q.
O
o
o
E£
Q} ^>
"tn O S
>% .0 O
CO < Q.
O
o
o_
1 ®
W 0 S
>- ja O
CO < Q.
o
i «
(0 C5 ~
>. -Q O
CO < Q.
O
.- O
w to
£Z o>
Us
co < a.
o
w S
Q> -,
I.!!
CO < Q.
O
E T]
•2 > =;
>> -Q O
CO < 0.
CO
i
to
CO
"ta
0
H-
'o'
0)
System Size
(Population Se
-
O)
0
o
S
oo
CM
CO
M-
CO
O)
CO
o
re
CO
0)
> r-
CM
in
CD
CO
CO
CO
CM
CO
O
in
i
CD
o"
o
2
0
Ti
CO
1
o
i
CO
0>
in
CM
oo
CO
CD
in
CO
0)
o>
CM
in
Oj
o
o
1
o
in
in
in
CO
CO
O)
oo
oo
r--
co
in
r-
o
o
o
o
o
-
in
o
oo
O)
CO
CM
CO
in
CO
T- O
o o
« o
co" o"
•~* T".
E
3
T3
a>
o
0)
S
ra
| Large/Very !
0
-
oo
f-
co
CO
S
o>
0)
in
CM
Ol
o
0
o
o
T"
1
o
o
o
T"
o
o
-
CM
CO
h-
£:
CO
CM
O
o
p
o'
o
/v
o
CO
o
T —
O5
CM
CM
OT
O
CO
O>
^j.
CN
CM
S
|2
0
O
o
0
o
o
r-
-
15
5
1 Proportion of 1
Systems
-------
_co
0)
0)
o
•u
co
01
0)
O)
5
0
^^
^-*
^
E
La
fli
H
O>
C
o
_J
(—
1
co j«
1 5
*j? ^"*
v7 _J
4/5 d
o> SE
13
•o o
C 5
. 3 •=
2 «
0 ~
•£? *J
'E °
E ®
1 S
O .Q
O <
**-
O
CO
C
0
r
O
Q.
O
£
•o
0
CO
_E
"co
HI
id
I
Q
X
UJ
u>
jo)
Q.
re
CO
"re
"5.
s
Q
.>»
1
re
a
o
"3
.a
"w
b
o
Q>
w 12
S o
5*
re *±
o c
i_ 0)
O 0
1 State/Reg ion:
Scenario:
o
o
o
E ^
m —
CO O "=:
>, JO O
CO < Q.
O
O
O
2 CM"
^j •>
CO O ^^
>, J2 O
CO < Q.
O
O
• o
CO __"
is-
to o ^=
>. -Q O
CO < Q.
0
Ew N.
0)
w o S
•S..Q 0
CO < Q.
0
1 «
111
CO < Q.
O
(0 M)
E"
o
,2 >
£3c5
CO < Q.
O
1 ^
0) > __
w o ^
>. -Q O
CO < Q.
CO
E
j2
CO
S
i2
f
£
System Size
(Population Se
o
o
in
o>
CO
in
co
CO
o
IO
CN_
5
5
X
5
>
II
r-
CM
CO
CO
o
T—
CM
0
O)
if)
CO
CO
0_
•»-
CM
o"
O
10
0
"5
CO
1
o
re
CO
o
T~
CO
CM
CM
O
IO
^5
J
CO
o
T
m
o
"*""
0
CD
CM
O)
IO
S
CO
CD
IO
CO
CO
CO
o
o
CO
co"
T-
o
CD
"3-
CJ)
en
CO
00
(O
CO
CO
CO
05
CO
T- O
O 0
CO O
co" o"
1 Medium (
1
\
(.
,
i
•
o
b
S
ro
1
S
re
o
^~
CD
?
CO
CM
CO
cn
IO
CD
CM
Ol
o
o
CJ
o
1
T"
o
o
o
T-
0
o
o
o
-
CM
IO
r-
9
3
N-
CM
CO
CM
CO
o
CD
CD
CO
h-
CM_
O
CO
co"
_
3
S
o
*~
0
o
o
o
o
o
T-
T —
5
5
1 Proportion oil
(Systems
-------
£
CD
i
c
o
73
CO
D-
O
0>
CO
Q)
>
^%
E
k»
0
D)
C
O
^J
«•
JM
1
E 1
CD >
•J3 CD
(*/ ajl
^* 1
O
flj 5"
•*^ S
i is
•o O
c 5
3 «=
2 "«
of
•^ ff
1 ^
E §>
1 1
0 <
n-
o
CO
o
kl
O
a
0
a.
•o
IS
E
7n
LJJ
CD
Q
*-•
la
IE
X
UJ
Q>
Q.
E
(0
CO
73
o
*o
3
Q
>»
1
(0
a
_o
'=
•s
5
>^
O
C
Q)
11
03 Q)
LU S
:> ^
0) 0)
Z 0
[State/Region:
(Scenario:
o
o
o
03 •*"
C ,£! O
CO < Q.
0
O
O
g CM"
S m
Elo
CO < Q.
O
O
o
to ,_-
fll
"to o n=
>^ J3 O
CO < Q.
O
0) > __
Tfl C3 ~
>, .0 O
CO < Q.
O
IK °
W 10
0 °
w 0 5
>,.Q O
CO < D.
o
1s
CM ^^J
111
CO < Q.
O
1 «
® > —
in O ^35
>. -Q O
CO < Q.
(0
I
4-1
(0
CO
Is
o
System Size
(Population Served)
CO
o
0
CN
IO
CO
O)
CN
CO
in
CN
••a-
CD
o
CO
CD
CO
00
CD
1
in
«
CO
e-
0>
^0
0,0
If)
to
t —
in
CO
CN
0
CNJ
CO
CO
CD
CO
o
o
CO
CD
If)
IO
ID
o"
o
in
0
7s
E
CO
1
o
!
0
-
O)
CO
r-
CD
CO
CN
O)
CN
0)
o
o
o
^7
o
m
o
-
CN
CO
in
CO
CO
a>
o
IO
IO
o
0
co_
o
^.
03
CD
S
CN
CN
CO
£5
N-
CO
T- 0
O O
CO O
co) o"
E "~
3
'•B
Q)
o
LargeA^ery Large;
o
o
CN
^
CO
Jv».
to
CD
O
O
p
o
o
o
p
Q
o
o
o
o
o
o
o
o
o
o
p
o
o
T—
A
CO
CD
CN
f-
O
LO
CD
cS
CN
O_
O
CN_
in
CO
co_
IO
CO
to
"*
CN
o>
10_
s
o
0
o
t —
*-
-1-
-
^
I^-
Proportion of Total
Systems
-------
(0
Q)
^
0)
^J
c
o
TJ
(2
0)
c>
2
o
^
^^
^
E
^H
,™
|^™
O)
c
o
^J
^
i
to J2
E «
0 0)
E^
o
>undwater
MCL/AM
S «
01
*•• *••
'c O
1 a.
E «
E o
O -Q
0 <
>4_
O
(0
o
o
Q.
2
a.
•0
0
CO
E
"w
LU
Exhibit D.1-
to
a>
a.
E
(0
(O
S
m
.0
"5.
Q
<5
t
(0
a
^
o
y Distribul
o
fli
P™ LM
r c
o a>
Z O
1 State/Region:
Scenario:
o
0
o
w ^
^ S-
tfi O «— 5
>.^ 0
CO < CL
o
0
o
§5
"oo O -"^=
>, £1 O
CO < Q.
0
O
O
en T-"
S ^
P^ > —
co o rs
>. J3 O
CO < Q.
o
s > —
I|S
CO < Q.
O
^5
i-2
«1§
>> -Q O
co < n.
o
i!
** o ~
CO < Q.
0
ET~
o
"S o S
>. -Q O
CO < Q.
(0
to
CO
(0
2
f
System Size
(Population Se
o
o
CO
CD
in
10
CO
CNJ
o
CO
T—
o
o>
to
CM
•^1^
5
5
0
>
D ^3
0
CM
m
CM
CO
CO
CN
•^f
CO
C>
CO
o"
o
in
o
"55
CO
I
,
•<-
CO
^1
o
CM
CO
in
N
CO
Oj
=>
ol
T-
^r
oo
CM
CM
CO
»
O
CO
CO
co
CD
CM
O
s
o
o
T-
co
-
CO
T —
00
in
o
T- O
o o
o o
Medium {
1
OJ
E
(C
I LargeA/ery I
o
o
CM
^
CO
_..t
CO
en
in
3
3
i
3
3
O
O
O
o
o
T-
-
CM
S
o
T-
A
CM
O
in
in
o
CO
T~
CO
f--
CM
O
D
CM"
CO
CO
co"
1
o
o
o
o
o
o
-
T-
§
5
ll
2 S.
CL CO
-------
J2
0)
^
0)
c
o
re
QL
(0
O)
2
0)
^
^^
^
E
fli
U/
f^
O)
c
3
JC
to J2
E ®
"to Jj
m* — J
O
0 5
13
•O o
5 cE
o "zr
IM *»*
O*iZ!
C
£-.2
ommuni
bove Po
o <
"5
CO
c
o
"C
o
f\
U.
2
Q.
T3
CO
E
to
UJ
CO
1
Q
*••
X
HI
in
o;
c
E
re
to
to
"5
3
Q
j>%
n, Quarter
.2
Distribul
>»
o
c
fl>
•o
0)
w 2
11
D. O
1 State/Region:
Scenario:
o
o
0
{2 -a-"
~tn o ~
co"< c
o
o
o
2 CM"
ji) §
to o ~
co"< c
o
o
o
to ,_-
to o iS
>. .n O
CO < Q.
o
>» -O o
CO < Q.
| S
S > _
w o S
>.-a O
CO < Q.
o
ECO
o
CO < Q.
O
>i J3 O
CO < Q.
(0
E
%
V)
3
,2
S"
1
0)
•> =
*i
E -2
0) =
w ^
J^* ^
CO ^^-
o
CM
to
O>
CO
CO
CO
CD
o
CM
CO
CM
CD
CD
CO
1
I
i
1
•* cT
o
T—
0
T-
CD
T—
O)
g
CM
CO
O
CO
O_
O
CD
CO_
O
to
0
15
E
to
1
o
1
0
CM
°
CM
CC
CM
CO
D
M
in
n
s
i
9
0
CO
CO
:M
n
°
CO
CO
CO
CO
O)
CO
3
n
5
0
0
CM
-
CO
CM
h-
c\i
o>
CO
CM
T- O
0 0
CO O
eo~ o
Medium (
1
c
r
c
T
C
r
r>
•r"
0 C
T-
Q>
s
n
Large/Verv I
io.ooi-mn nnn
o o
Z> 0
o o
- 0
Q
D _
g ^
* CM
> CO
3
3_
3
A
*-
CO
o
CD
O
in
CO
CM
CO
M
CO
CM
10
CM"
CO
CO
CO
,
Q. CO
-------
to
CD
CD
c
o
re
o*
CD
D)
2
o
^
J^
^»
E
LM
fli
W
O)
C
0
^^
_
i
w 5?
re $
c °
3 =
2 15
o -g
£•-8
1£
P CD
E o
O .Q
0 <
14-
0
to
c
o
kZ
o
Q.
O
£
•c
"re
E
to
LJJ
0)
Exhibit D.1-
(A
CO
Q.
(0
CO
*•*
"5.
Q
>,
1
ra
d
_o
Distribut
to >>
c *2
ro §
? T
3 C
1 1
U C
O CO
C£ U
State/Region:
Scenario:
o
o
o
i£
•2 > —
>-v -O C3
CO < Q-
O
O
O
<§ CM"
0}
1/5 o —
>, J2 O
CO < Q.
O
O
0
if
"w o ~
>sJ2 O
CO < Q.
O
"55 o S
>, -Q O
CO < Q.
o
(0 g
>. J3 O
CO < Q.
O
<2 co
>.^2 O
CO < Q.
O
C fl>
CO >
w o S
>,-n o
CO < Q.
(0
0
co
to
p
0)
System Size
(Population Se
a
CO
o
S
oo
CM
OO
T—
CD
IO
T—
in
n
"a
CO
a>
> °
IO
CO
o
CM
CM
CO
O
CO
CD
CO
CO
CO
CD
O
CD
CO
CO
t —
o"
o
in
o
"re
CO
£
0)
o
CO
-
CM
IO
CM
O
to
t —
co
CM
a>
CO
CM
CO
Ol
o
q
o
m
-
CM
OO
to
O
00
IO
^c —
a>
co
CM
IO
CM
CO
S'
CO
o
o
o
T-
o
-
0
CO
CM
CM
If)
o
10
IO
^ —
T- O
o o
crT o
1 Medium I
1
o
CO
E
n
1
0)
E
ra
O
o
CM
IO
o
CO
CM
o>
to
o
oo
CH
o
o
o
o
1
o
q
o
o
0
o
o
0
o
-
-
o
o
o~
o
/r
CO
CO
CO
oo
r-
co
CM
T"~
CO
CO
co_
o
CM"
CM
co"
o
so
co"
"o
o
0
o
o
o
"-
-
-
s
1 Proportion of 1
Systems
-------
-------
Appendix D.2. Proportions of Systems Exceeding Potential Radon Regulatory
Levels in Seven States With Supplemental Data
Methods, Occurrence, and Monitoring Document for Radon: Appendix D
-------
-------
o>
0
JD
<
_w
Q)
» s
to I]
o ^
CO _
§ co
1 c
3 «
o o
ty
o
CO
. c
o
'•c
0
Q.
2
Q.
TJ
CD
1o
E
to
UJ
CN
d
1
I
'3T
(D
s
CO
c
i
w
Q.
E
re
CO
42
"w
"5.
0
2>*
1
(0
3
o
c"
.2
3
<*^
OT
Q
^t
o
c
0>
TJ
0) i 1—
r E
_• -4^
co ^
2;. ** Q-
CO J
"*
o
w S
c
§ ^ IS
•K * O
s. § =-
CO a
<
o
£ §
£ T-"-
t>'t
>< o
co J5
o
CO 0
E >- -
5 2 o
&r
o
li1
o
W 0
E M «:
| § 0
^** .fi
W <
O
O) O
tjl
^f
(/}
1 1
1- >.
CO
^^
^
0)
1
System Size
(Population S
CM
CM
00
in
"3-
CD
00
T—
0_
•t —
CO
J--
CM
T —
CO
00_
T""
co"
CD
00
CD
•co"
1
to
CVJ
•l-^-
15
E
n
V
>
si
CD
O
CM
T—
CD
CO
CD
CD
CO
00
CO
CO
t —
CD
CD
t —
CO
CM
CM"
CD
CM
in
co"
^^
o
o
2
•r-
"5
CO
0)
i
i
CD
S
CO
CM
CD
x—
CD
00
CN
r--
CD
CO
CO
CD
0_
O
o
^3
I.
T—
O
10
CO
CM
tn
CD
CO
CM
CM
O
CO
CO
V™
in
0
CD
o_
T—
CD
in
CO
O
o
CO
crP
i
o
CD
CD
1
O
•sf
CD
i
'•
O
0
CD
CO
CM
m
i
o"
0
o
eT
m .
0)
o
Lame/Very La
T_
^
T—
CO
CD
CM
CD
O
in
CO
0
CM
o
o
Q
o
o
1
T-
o
o
0
o
o
o
-
CM
CD
CO
O
o
o
o
in
h-
'
-------
CD
o
o
in
CD
CD
c
o
TJ
CO
OL
CD
U)
(0
I_
CD
>
<
C
&•
0)
IT
c
_J
±i w
(A 5>
E -J
3 _i
to O
>s «»
?1
*•* "*•».
1 5
§ S
TJ —
C 5
2 S
O o
& **~
3
E
h»
o
O
*o
o
r
o
Q.
O
Q.
"O
to
E
1/>
LU
CN
d
CO
Q)
C
E
(Q
OT
"re
"a.
3
Q
>>
1
n
3
a
E£
.0
13
JQ
g
ts
o
o
c
0>
•a
c
3 _
P S
; 'c
J= «
S O
1 State/Region:
Scenario:
o
« g
o ^ S
*i 0) C
(/> > r
>» O
OT ^
<
o
c °
C js|* aj
** o> C
OT 1 ft
% S
OT ^
^
W o
E h~ _
km K^
a g o
|,| °-
|-1 s
*^ i» O
>, ° D-
05 5
o
« o
E « a:
•S 2 0
>,| o.
W <
(0 g
3^0
>. o a
*^
to
ll
£ >,
OT
System Size
(Population Served)
o
0
•*
o
CO
CM
in
CD
O)
CM
CM
in
I
3
3
>
I g
X T-
O
CO
CD
in
CO
CO
CD
o
CO
CO
CO
CM
I-B^S
o
0
IO
0
"5
(0
1
>
o
o
CM
CO
in
o
•vf
CO
o
r-
CM
0
1
o
m
j.
o
o
CO
CO
o
CM
CO
in
CM
CO
CD
o
o
CO
2
T-
0
o
o
T~
CO
-
*
o>
in
)
t
1
U^fT
o~
— O
0
o"
T-
1
01
E
ra
o
0
o
o
*-
in
CD
CO
0
o
C3_
cT
o
T-
I
O
o
o"
o
o
o
o
o
0
•<-
ir—
O
O
cT
o
/v
o
CO
in
CM
o
CD
CM
CM
0
CO
in
CO
CO
S
£
cf^
O
O
0^
CO
o
CM
CM
~s
CM
Ifj
SS
OJ
CD
^».
NT
CO
CM
00
O
0
o
Proportion of Totai
[Systems
-------
0)
0
^t
(0
0)
QJ ^1
1o O
>» S
IM "^
•2 Q
Groundw;
otential M
*°-
c
3
E
E
o
0
(|M
0
(0
C
_0
o
Q.
O
^
Q.
•o
0)
•*->
n
E
'•ja
V)
Hi
CO
CM
Q
o
Q.
(0
CO
're
_o
o
k^
3
Q
_>»
1
«
C§
O
'&
Distribu
o> "
~ o>
•£ "o
£ c
15
5 1
0) 0>
z o
1 State/Region:
Scenario:
o
.. o
1 -
•2 o> o
!x rt ^
(O .a
•*
o
1§--
2 CD O
>% O ^^
co .a
^
0
£ o
E •«-" s
*^ Q.
"|
^^
M 0
E.I- s
In > ^
>» »
W <
. 0
V> o
s » =
S 2 5
« o o.
a^» ^
C/3 "S
<
lS-
s > «
§1 ft
<
o
s i o
|. o tt
<
m
ll
V)
^
System Size
(Population Serve
?
oo
CM
in
CM
CO
CM
in
O)
CM
CM
CO
s
CO
in
CM
CO
,
IO
V)
1
> ?
r-
§
CD
in
00
o>
0
o
CM
CO
CM
G)
CM
_
^3
i
0
CO
1
o
i
CO
^r
t-
CM
T—
?
-
O
CM
CO
CM
O)
CM
O
^>
o
7
o
IO
«*
CO
CO
in
CO
T—
CM
CM
CM
CO
O
CO
o
0
o
CM
* ,
10 i
-
0>
o
o
CO
CO~
If
•o o
® 0
2 ^
O
(Large/Very Large;
o
*-
-
CM
CM
CO
-
O
o
0
o
0
o
o.
cT
T-
o
0
o
o
0
o
o
o
o
o
o.
o
o
T-
A
en
0)
CM
CO
00
o
CO
in
CM
in
0
CD
CD
5
ff
^5
CM
CM
CO
0
CD
O
"«
Proportion of Tot
Systems
-------
10
CD
CD
C
0
•u
C5
eg
CD
O)
2
CD
l^f
QJ
i
O)
C
i
(0 CO
£ 'o>
.2 ?
to _j
w _j
i_ O
OB g
ii
o —
fc: re
j* 1
1 5
o °
0 <
M™
o
o
•c
0
Q.
LM
Q.
•D
CO
'43
CO
HI
CM
d
4->
la
UJ
M
O
Q.
E
R
£.
to
o
Q.
3
Q
Quarterly
o
i -
3
£
4-1
M
Q
U
c
0>
•D
C
> E
> o
z o
5
1 state/Reg
Scenario:
o
1 §-
** o>~ C
OT 5
"*
o
i §
^ •* &•
CO ^
a
CO .Q
(0 §
E •- -
S S o
i* o Q.
« 1
.. o
1 s _
a a o
!•!
^
0
| CO _
155
>, ° D-
o
(0 0
E •<- s=
155
>• ° a-
w <
(0
« 1
^^ ^\
CO
I
a>
•S §
to -2
1-1
CN
in
CM
T —
c>
CO
CD
CM
CO
cn
CO
o
CD
CD
t —
CO
>
•
II
CO
in
CM
in
in
o
CM
o
CD
CM
en
CO
in
CO
CO
o
o
up
CO
CO
0)
o
1
r-
co
OJ
CO
CO
•<*•
in
CO
CO
|
CO
o
0
0
1
o
10
,-
m
CM
m
CO
CM
m
S
en
CO
T—
m
o
o
co
co"
§
O
,_
m
o
CO
CO
CM
en
In c
a
^ &•
o ™
i§ I
in
0
o
T—
*
-
CO
CO
CO
rj
o
o
o
o"
0
^.
0
o
o
o
0
o
o
-
CM
*
in
o
o
o
o
A
in
CO
CO
T~
CO
CM
CO
en
CD
m
CD
CO
CO
en
CO
in
in
T
o
en
CO
^
3s
°
?
CM
O
h-
?g
in
CM
h~
CO
en
CD
m
CN
CM
CO
0
0
CD
O
O
«•
O
1 =
0 §
a. «-•
S £
X. CO
-------
tfi
. ,
00 — 1
i_ O
O ^5
il
5*
o — *•
*•• CO
0 5|
= P
3 Q-
E o
1 o
o ^
^ **
to
o
'•E
o
f*l
u.
o
^
a.
tS
_E
to
UJ
in
i
CM
a
15
IE
X
LU
.
1
,
U
.<" ^
C C
o>
a. u
1 State/Region:
Scenario:
o
05 o
111
^ > Q.
w E
0
E °
S «"|
5** o ^^
w ^
o
1 ^"g
I.S °-
«o ^
o
W 0
E ^ =
155
>> ° °-
(n •£
^
. o
(0 o
E •*> =
% 5.5
CA *^
O
<2 o
'I 5§
!••§ °"
<
o
W ^3
c; ^_
1 5o
>, ° °-
w <
(A
ii
° 5,
CO
s~
o
System Size
(Population Se
CO
CO
o
CO
o
CM
in
CO
CN
00
CM
CO
CO
CM
CO
CO
in
CO
CO
CD
in
' — '
"S
V)
£
0)
2 o
I?
CO
0
CN
t —
§
CN
CN
in
o
CO
CO
CO
tfi
CD
s
^_ ^
o
0
•9
^•n
o
"5
tn
I
0
1
0
in
CN
CM
in
O)
CO
in
00
0
r-
CM
CO
o
o
o^
T"
1
V
O
in
T —
CO
CN
CD
in
in
r-
co
C3>
o
CN
CD
I"1*"*"
Is**.
o
o
co_
CO"
T™
o
^.
o
CM
CO
CN !
in
CO
in
CM
CD
o
Is*"-
1
o
it
D O_
o
o3
Q
|Large/Very Lar
T-
co
CO
£
-
00
s
in
CM
o
o
o_
eT
o
T
T-
O
O
T-
o
0
o
0
0
0
0
o
o
0
o_
cT
o
CM
CO
"*""
CM
O3
CN
CO
in
CN
o
0
CD
00
CO
O)
o_
CO
CO
CO
r-
5
i2
CD
N-'
^o
O5
CD
CO
tf>
CD
CO
oi
^
«r>
CO
CD
>^
^ —
CD
CO
o
o
0
"re
<->
o
1 Proportion of 1
Systems
-------
CD
^
o
J2
to
CD
CD
•"•
C
O
"O
re
DC
CD
2?
&H*
CD
<
£
^
CD
F-
1
O)
O
_J
•C
±f to
CO CD
|J
|i
* o
•5 ^^
•2 MM
1 s
JJ
CD o
>, CL
'E
3
E
o
O
I+M
O
to
c
,o
o
D.
0
turn
Q.
"7"J
CD
-^«
re
"to
111
CN
Q
w
CO
a.
re
CO
0)
<*-!
(0
o
Q.
3
Q
1
re
3
a
o
^f
.a
™
*••
(0
Q
o
We
£»
0 g
re 1—
j _
"3 "c
O CO
CO O
1 State/Region:
Scenario:
o
1 °-
IP
tO .Q
<
O
W §
CO" £t
**
o
1—^
c o
C, *.
*~ T~ -^-
^* O
to ^j
ll_
c • 5H
d) O "~
O o o
"
£.£ °-
^3
w §
E PO t^M
« " i^
tS > O
w 0 Q.
to -°
o
|5 o Q.
to^-0
U)
° ^.
CO
??
^*
System Size
(Population Serve
to
If)
CD
^-
^
o>
CO
CO
T —
CM
in
CM
CM
o
CO
CM
o
CO
CM
m
M
0
5
r. o"
U O
>l T"
h-
co
s
O3
CO
CO
0
in
X —
CO
CM
in
o
CD
^_^
o
o
IO
0
"re
CO
CO
o
,
•«-
CO
CO
T-
"*
in
CM
CM
CD
CO
CD
O
o
o
TT~
o
to
•r-
^
CO
CO
"*
CO
CD
CM
CO
in
in
0
o
CO
n
T-
o
cs_
T~
co
1-
CO
CM
CD
in
CM
CO
1
5
n,
I f
•o o
I?
o
ILargeA/ery Largej
0
0
-
^_
CM
CO
CO
^
o
o
o
o"
o
1
o
0
eT
T-
CD
CD
O
0
o
o
o
o
o
0
o
o"
A
in
CO
CO
00
0
CO
CM
CO
CO
CD
CO
5
CM
CO
co •
in
1
—
o^
CO
r-.'
xO
CM
in
co
&)
in
^
CO
CO
CD
&>
CD
-S
in
o>
r-
^
CO
c\i
o>
^o
0
O
O
z
Proportion of Tot
{Systems
-------
O
o
&
^
in
0
0)
c
o
•o
CO
D£
0)
••»
Ol
0)
^
<
E
o
O)
C
O
_J
Wti)
^tf
c
CU ^J
^^
O — 1
-3 9
Groundw
otential IV
.£*
'E
i
E
o
O
**•
O
(0
c
o
o
D.
O
a!
^M
fli
"«
"w
HI
CN
Q
W
0>
Q.
re
CO
^1
re
_o
"5.
O
1
re
a
o
.n
*z?
4-*
(0
b
o
0)
•o
c
w 2
* 1
a> a>
State/Region:
Scenario:
o
_ o
e =»-
*^ fl) O
^ > D.
CO ^
**
0
«2 §
c ^» ...
4^ Q) CO
>> O ^"
CO jo
•*
0
t—i
E --
•S a O
S) ° °~
<
1§ =
o a> ~
w <
o
W 0
E «> =
fl) O JTT
0) S Q
>» 5
^C
w§
E « =:
S §> 0
JM °"
<
o
S 5 |
ix ° °-
<
in
il
i_ (/>
i^ >^
CO
*^^
System Size
(Population Served
o
CO
f-
CM
CO
CD
CM
CM
O
CO
CD
CO
OO
CO
in
"re
£
0)
£ o
go
o
CO
CM
CD
in
T —
in
CM
CM
CD
in
CO
T —
T—
o"
S
T"
T-
co
1
o
"re
CO
o
CO
CO
CM
CO
in
m
J:
CO
CM
CO
CM
O
o
o
in
T-
-3-
CM
CM
in
-'
oo
in
CO
CO
o
CD
o
o
co"
1
T"
o
o ,
o ;
CM I
CO
in
in
E
CM
CO
CM
i
3
n
n
If
•a o.
o
JLargeA/ery Large:
0
o
o
o
o
CM
CO
CM
in
CO
o
o
o
o
0
T™
1
o
cf
o
0
o
o
o
o
o
-
o
o
o
o
0
T™
A
CM
5:
T —
CO
o
CM
CO
CO
CO
CO
•sr
co
^
CO
CO
CM"
oo
CO
co"
To
c^
O
O
xp
^r
o
-vP
O
CO
c\i
-e
o**
CO
in
CM
T_
SD
^5
CO
CO
^p
0
CD
o
T~
1 Proportion of Total
Systems
-------
0)
0
•Q
^
CO
0)
c
o
•o
C8
o:
I
0)
E
o
D)
C
O
_J
£ M
*J CO
CO Q)
-2 H
10 (-'
>,g
"S
0) — 1
ii
5 c
^^
•H Ctf
t '^
li
O o
>, CL
*E
3
E
o
O
*o
to
c
o
o
Q.
O
CL.
•a
3
CO
E
"•P
to
LU
•
CO
CN«
Q
en
u
Q.
CO
"5.
3
Q
>.
1
n
3
a
c"
o
Distribu
o
03
•o
„ c
C CO
» t~
o 5
L> •*-»
2 g
5 O
IState/Regior
Scenario:
o
<2 o
E 5" =
S o> o
>« o °"
CO JJ
<
(A §
S o o
2? S ^^
CO 5
o
(0 §
Sog
j.s"
CO jo
<
0) o
E ^^ ™""
s §> g
co" ^
cog
E «» as
S S o
l-l
o
W O
E <*> —
Q) Q) w
11 ft
<
o
15-
1 5o
8, o a-
"*
u
rj
It
CO
'o*
0)
0>
CO
System Size
(Population
o
o
0
m
o>
CM
T —
in
CD
O)
CO
CM
o
in
^
5
5
5
if
T-
m
CD
CM
0
m
CM
CO
CO
CO
CM
l«O
f^fc-
CM
_ _^
O
o
in
0
15
CO
£
0)
o
J
o
CM
CO
T—
CO
CM
5
CO
N-
v—
CO
T—
T—
O
O
1
O
o
CM
in
T—
O
CO
S
O)
CO
o
in
CM
CD
o
o
eo
o
o
CM
1-
CD
O
CO
CO
h-
00
>
£
I 0
o o
1?
0
ail
S
(0
|LarqeA/ery L
o
o
o
-
CO
?
CO
CO
CO
CO
o
o
o
o"
o
T-
1
o
o
0
o
0
o
o
o
o
__
-
-
o
0 i
o
0
0
A
T-
o
r-
in
CO
CM
CM
S
o
o_
^— •
o
15
2
v^
T™~
C)
CT
o
in
_
CO
o
^
CO
CD
CM
•^
CD
I
o^
o
o'
o
T~
5
3
(Proportion o
(Systems
-------
Appendix E: Estimated Numbers of Non-Transient Non-Community
Groundwater Systems Exceeding Potential Regulatory Levels
Methods, Occurrence, and Monitoring Document for Radon, Appendix E
-------
ik lUnl I I I il I
-------
Appendix £: Estimated Numbers of Non-Transient Non-Community Groundwater
Systems Exceeding Potential Regulatory Levels
In this appendix, the numbers of non-transient non-community groundwater systems
(NCNTWS) exceeding potential regulatory levels are estimated. Estimates of the national total
of NTNCWS are developed, the radon distributions in NTNCWS of different sizes are
characterized, and the numbers and proportions of NTNCWS exceeding the various radon levels
are estimated. The data supporting the estimation of radon occurrence in NTNCWS are quite
limited, and thus the estimates of NTNCWS exceeding potential regulatory levels are less
detailed than those developed for community systems, and are subject to much more uncertainty.
E.I Number of Non-Transient Non-Community Systems in the U.S.
EPA's Drinking Water Baseline Handbook (EPA 1999) provides estimates of the
numbers of NTNCWS by state and system size, as summarized in Exhibit E-l. Using data from
the SDWIS database, EPA estimates that there were a total of J 9,062 active NTNCWS in the
U.S. in 1998. Consistent with previous analysis, the great majority of NTNCWS are very small;
just over 50 percent serve fewer than 100 customers, 86 percent serve fewer than 500 customers,
and 99.6 percent serve fewer than 3,300 customers. Based on the SDWIS data, there are no
NTNCWS systems serving more than 50,000 customers in the;U.S.
E.2. Distribution of Radon Levels in Non-Community Non-Transient Systems
The NIRS data come solely from CWS, and do not provide any information regrading
radon levels occurring in NTNCWS. Among the data sets received from the states, only six
(from Idaho, Maine, Maryland, New Hampshire, Texas, and Wisconsin) contained information
on radon levels in NTNCWS. The data from Idaho do not identify NTNCWS by system size and
therefore were not used in the analysis.
These geometric mean and log mean radon levels in NTNCWS in the five states are
summarized in Exhibit E-2, along with summary statistics relating to radon levels from the same
size CWS in the same states1. Generally, the radon levels in the NTNCWS systems are
considerably higher than those in the same size CWS in the same states, although the differences
are not always statistically significant2 due to the small numbers of systems involved. In the
supplemental data from New Hampshire, the CWS and NTNCWS radon levels seem to be
comparable, although the relatively small differences in log means for some strata are significant
1 The numbers in the "All" column for NTNCWS may be greater than the sums of the entries
for some sizes because they may include systems for which size data were not available.
2 Student's t-test for independent samples applied to the log means, two-tailed. "NO" denotes
differences that were not significant at the p = 0.10 level.
Methods, Occurrence, and Monitoring Document for Radon, Appendix E E-l
-------
Exhibit E-1 . Estimated Numbers of Active Non-Transient Non-Community Groundwater Systems by State
Total in Size
Category
National Total
By State:
Alabama
Alaska
Arizona
Arkansas
California
Colorado
Connecticut
Delaware
Florida
Georgia
Idaho
Illinois
Indiana
Iowa
Kansas
Kentucky
Louisiana
Maine
Maryland
Massachusetts
Michigan
Minnesota
Mississippi
Missouri
Montana
Nebraska
Nevada
NewHampshire
New Jersey
New Mexico
New York
North Carolina
North Dakota
Ohio
Oklahoma
Oregon
Pennsylvania
Rhode Island
South Carolina
South Dakota
Tennessee
Texas
Utah
Vermont
Virginia
Washington
West Virginia
Wisconsin
Wyoming
Size (Customers Served)
25-100
9,606
19,062
101-500
6,840
501-1000
1,891
1001-3300
665
3301-10000
53
10001-50000
7
1 00000
0
11
0
96
20
491
68
426
46
725
143
140
167
349
55
36
27
104
138
235
104
961
601
28
95
96
143
56
256
543
78
299
259
15
503
75
162
406
31
126
15
29
329
22
1
278
134
82
556
47
19
0
66
19
381
52
150
25
269
99
92
210
250
57
16
44
76
184
163
75
607
67
40
92
101
33
24
124
305
44
237
281
7
460
40
149
572
25
94
9
25
281
22
0
328
110
83
406
27
13
0
26
3
92
7
50
9
80
32
23
44
83
14
9
8
28
19
76
37
125
11
37
25
12
8
5
35
105
21
111
88
0
123
6
14
191
11
19
1
3
69
3
0
115
25
11
61
3
3
0
21
3
30
2
15
7
31
16
8
21
11
7
6
1
24
3
20
7
23
2
16
13
2
3
5
6
41
4
' 41
23
0
24
0
7
73
3
7
0
1
45
2
0
45
12
5
25
1
0
0
2
0
8
0
0
0
2
1
2
2
0
0
0
0
0
0
1
1
2
1
4
1
0
1
1
0
5
0
2
1
0
2
1
0
4
1
0
0
0
5
1
0
2
0
0
0
0
0
0
1
0
3
1
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
1
0
0
0
0
0
0
0
0
1
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
Notes:
|1. Source: USEPA,
Drinking Water Baseline Handbook First Edition, 1999
E-2
-------
(D
E
(0
to
c
M
E
O
(0
/^%
L.
Q>
t
c
E
o
O
.£
(0
o>
.**
§
M
(0
(O
"c
(J) (0
w £
C (0
2 35
T
c
o
J^
c
3
E
E
o
C
o
.S
"o
s
o
•o
•IS
(£
Comparison ol
CM
UJ
4->
!E
X
LU
CO
?
UJ
I-
co
co
in
tr
5
CO
UJ
h-
5
5
LU
co
co
co
g
i
<(
co
>
CO
5
01
1
(n
ALLSY
in
^
o
O
CO
^
z
CJ
z
o
co
z
o
z
1
ro
O
1
z
n
z
o
NCNTWS
I
STATISTIC
UJ
Z
0
-
h-
3
8
CD
CO
s
CD
in
CO
CM
8
fe
NUMBER OF SYSTEM!
to
OJ
en
CO
in
in
a>
CO
55
1
CM
O
en
CO
CO
en
m
CM
CO
o
at
ARITHMETIC MEAN
!
CM
'
§
CO
CD
o_
CO
CM
CO
CO
tn
§
in
CO
i
o
GEOMETRIC MEAN
1
CO
o
m
cb
en
CO
en
CO
en
to
to
in
CM
Ol
cn
CO
CO
CM
to
en
8
CD
in
cb
1
CD
1
to
LOG MEAN
i
CM
CO
CO
§
i
s
CO
TT
to
o
en
in
CO
CO
CO
s
s
in
LOG STD. DEV
!
1
!
o
a.
O
z
O
z
b
o
V
a.
DIFF. OF LOG MEANS
SIGNIFICANT?
a
z
>-
a:
E
0
m
o
-
in
s
r-.
CO
-
*
CM
CM
CD
NUMBER OF SYSTEM
I
CO
CO
cn
CM
CO
3
CO
in
tn
CM
en
to
o
CO
CM
r-
CM
i
GEOMETRIC MEAN
!
CO
|
en
m
cq
m
m
CO
cb
CO
te
CO
o
CO
CO
cn
CO
CO
CO
CO
h-
cb
Ol
1
LOG MEAN
i
|
r*.
CO
b
CO
in
b
in
en
b
cq
CO
o
q
CO
to
to
b
CO
to
to
b
m
CO
CO
o
cn
b
LOG STD. DEV
t
o
z
Q
^
O
z
in
o
V
a.
DIFF. OF LOG MEANS
SIGNIFICANT?
LU
£
i—
to
t
CM
CO
o
CO
en
o
CM
r-
m
CM
0
CO
§
CN
O
CO
Ol
Ol
CO
rn
NUMBER OF SYSTEM
o
*
CN
Ol
at
CM
0
CM
0
O
CO
CO
CN
CM
r--
co
a
CM
CM
N-
O
CN
cn
CO
GEOMETRIC MEAN
CO
CO
tr
g
CM
in
m
CO
to
in
CO
i!
CM
m
m
CD
in
CO
CO
CM
in
to
CO
CO
1
1
1
•
tn
r*.
o
CM
CM
S
CO
tn
01
CM
w
m
CO
in
«
to
§
CO
CO
CO
LOG STD. DEV
!
in
0
0
V
0.
O
z
O
Z
in
o
»
Q.
DIFF. OF LOG MEANS
SIGNIFICANT?
UJ
x
CO
§ <
Z X
•
11
m
CM
h-
CM
'I
O
in
in
CO
CO
o
CM
CO
CM
CO
cn
•NUMBER OF SYSTEM
co
5
;
to
o
CO
CO
to
f-
CO
CO
CO
in
CO
CO
(GEOMETRIC MEAN
CD
CO
CO
o
CN
O
in
!
0
m*
en
in
CO
CO
CO
o
r-
I
in
Ol
CM
O
in
cn
CO
LOG MEAN
1
CO
9
0
cn
CO
0
•
§
tn
CM
CO
b
CO
at
CO
b
0
en
b
i
b
CD
m
CO
CO
cn
b
LOGSTD DEV
,
O
in
q
O_
O
DIFF. OF LOG MEANS
SIGNIFICANT?
«
X
t-
*
8
to
-
r-.
CO
CM
CO
CM
r-
co
CM
8
CO
in
CO
NUMBER OF SYSTEM
,
1
CO
CM
OJ
8
CM
CO
r--
CO
cn
CD
CD
cn
CM
CO
o
CM
CM
o
CM
GEOMETRIC MEAN
*
o
in
CM
CN
rr
m'
s
in
TT
in
CM
m
cb
cn
to
in
to
o
o
cb
CO
CO
tn
tn
o
to
CM
CO
in
ILOG MEAN
*
CM
CO
b
o
m
tn
CO
b
CM
0>
b
q
CO
en
b
ff
b
CO
in
q
TT
O
q
tn
o>
b
LOGSTD. DEV
.
,
(O
o
b
n
a.
tn
b
V
a.
o
m
o
b
v
a.
DIFF. OF LOG MEANS
SIGNIFICANT?
z
CO
z
O
o
CO
£
•a
«
£
o
3
r independent samples, t
•S
"c
cu
Qj "=
r> to
z .
E-3
-------
because of the large numbers of systems. Breaking the general pattern, in the supplemental data
from Texas, the log mean radon level for the very small NTNCWS is significantly lower than the
log mean for very small CWS.
i
£.3 Estimation of Radon Distributions in Non-Transient Non-Community Systems of
Different Sizes
As was the case with the community systems, it is necessary to extrapolate these data to
develop nationwide estimates of the proportions of NTNCWS above potential regulatory levels.
Because the data are so sparse, a much simpler approach has been used for the NTNCWS than
for the community systems. Single values of the log mean and log standard deviation for long-
term radon levels have been developed for each system size category across the U.S., based on
the data from the five states identified above. The assumption is made that radon levels in the
NTNCWS hi the rest of the states will bear the same relationship the radon levels in community
systems as they do in the five states for which radon data are available for NTNCWS.
To estimate the log mean radon levels for each size category of NTNCWS, the average
ratios of the geometric mean radon levels3 in the NTNCWS to the geometric mean radon in CWS
from the same states were calculated, as shown in Exhibit E-3.
Exhibit E-3. Ratios of Geometric Mean Radon Levels in Non-Community Non-Transient Systems
to Radon Levels in Community Systems in the Same States
NCNTWS/CWS GEOMETRIC MEAN RATIOS
State
MARYLAND
MAINE
NEW HAMPSHIRE
TEXAS
WISCONSIN
Average
ALL
2.271
2.108
1.158
1.144
1.496
1.635
WS
1.575
1.862
0.968
0.531
1.679
1.323
VS
2.104
2.112
1.022
1.359
2.294
1.778
S
3.780
2.234
1.728
..
0.557
2.075
M
_
_
_
„
_
-
L
„
_
_
_
-
It can be seen that the ratios of the geometric means are highly variable across system
sizes and among the states. In some cases, this is due to the small numbers of states where
comparisons are possible, and the small numbers of NTNCWS systems in some size categories.
In fact, average ratios were not calculated for medium or large systems because data are available
for only one state each (and one system each) in these size categories.
3 This model assumes a constant multiplicative relationship between the radon levels in CWS
and in NTNCWS. It is equivalent to assuming that the log mean radon levels in NTNCWS can
be described by the log means radon levels from CWS plus a constant.
Methods, Occurrence, and Monitoring Document for Radon, Appendix E
E-4
-------
Because of the small number of data points, it was decided to use the overall average
NTNCWS:CWS ratio of log means (1.635) to characterize national radon distributions for the
various size NTNCWS systems. For each NTNCWS system size category, the estimated
geometric mean radon level was estimated as the national average geometric mean radon level in
the same size community system, multiplied by 1.635, as shown in Exhibit E-4. The resulting
geometric mean (log mean) radon levels range from 520 pCi/1 (6.25) for the very small systems
to 260 pCi/1 (5.56) for the small system category. [
Exhibit E-4. Estimates of Log Mean Radon Levels for Non-Transient Non-
Community Groundwater Systems Using CWS/NTNCWS Ratio
CWS/NTNCWS Geometric Mean Ratio:
System Size
WS
VS
S
M
L
Community Systems
Log Mean
5.71
5.76
5.07
5.19
5.25
Geom. Mean
300
318
159
180
190
1.635
Estimates for Non-Community Non-
Transient Systems
Log Mean
6.20
6.25
5.56
5.68
5.74
Geom. Mean
491
520
260
294
311
Single, national estimates of log standard deviation values were also estimated for each
size category of NTNCWS. As was the case with the log mean estimates, the data were limited,
owing to the small number of states where comparisons could be made between NTNCWS
systems and CWS of the same size. The average log standard deviation values seen in the five
states and the average across all the states are shown in Ejdiibit E-5
Exhibit E-5. Log Standard Deviation Radon Levels in Non-Transient Non-Community
Systems
State
MARYLAND
MAINE
NEW HAMPSHIRE
TEXAS
WISCONSIN
Average
System Size
ALL
1.568
1.335
1.441
1.156
1.004
1.301
WS
1.590
0.638
1.365
0.490
0.722
0.961
VS
1.659
1.647
1.545
0.825
1.014
1.338
S
1.430
0.758
1.350
—
0.645
1.046
M
-
—
-
—
-
L
—
-
--
-
-
Methods, Occurrence, and Monitoring Document for Radon, Appendix E
E-5
-------
In the estimation of the proportions of NTNCWS exceeding potential regulatory levels,
the average log standard deviation for each of the respective size classes was used. The average
value for small systems (1.046) was also used as the estimate of log standard deviations for the
medium and large NTNCWS systems as well. This decision has relatively little impact on the
total numbers and proportions of systems exceeding regulatory levels because the medium and
large systems make up such a small proportion of the total NTNCWS. The log standard
deviations used in this analysis were not adjusted for multiple sampling compliance schemes, as
was done for the community systems. Thus, the system log standard deviations and the
proportions of systems exceeding the various radon levels may be slightly overestimated.
i , i
E.4. Estimated Numbers and Proportions of Non-Transient Non-Community Systems
Exceeding Potential Regulatory Levels
The numbers and proportions of NTNCWS exceeding potential regulatory limits were
calculated in the same way as described in Section 5.8 for the community systems. For each size
category, the lognormal model was used to estimate the proportions of systems exceeding the
various regulatory levels. These proportions were then multiplied by the total numbers of
NTNCWS in each size category to give estimates of the numbers of systems exceeding
regulatory levels. The results of this process are summarized in Exhibit E-6.
Exhibit E-6. Estimated Proportions of Non-Community Non-Transient Groundwater Systems With Long-
Term Average Radon Above Potential Regulatory Levels
State/Region: National
System Size
(Population
Served)
Very Very Small
(25-100)
Very Small (101-
500)
Small;
501-1,000
1,001-3,300
Medium (3,301-
10,000)
Larsa:
10,001-100,000
>1 00,000
Total
Proportion of
Total Systems
Total
Systems
9,606
6,840
1,891
665
53
7
0
19,062
100.0%
Systems
Above 100
pCi/l
9,137
6,095
1,550
545
45
Systems
Above 300
pCi/l
6,687
4,511
842
296
26
Systems
Above 500
pCi/l '
4,734
3,500
Systems
Above 700
pCi/l
3,423
2,819
503
177
16
6
0
17,377
91 .2%
4
0
12,367
64.9%
2
0
8,932
46.9%
325
114
11
2
0
6,694
35.1%
Systems
Above 1,000
pCi/l
2,208
2,138
187
66
6
1
0
4,606
24.2%
Systems
Above 2,000
pCi/l
693
1,074
48
17
2
Systems
Above 4,000
pCi/l
140
436
8
3
0
0
0
1,834
9.6%
0
0
587
3.1%
The general pattern of results is similar to that seen for the community water systems,
except that the proportions of NTNCWS exceeding the regulatory limits are greater, owing to the
generally higher radon levels in the latter systems. The great majority of NTNCWS (over 91
percent) exceed 100 pCi/l, and 64.9 percent exceed EPA's proposed regulatory level of 300
Methods, Occurrence, and Monitoring Document for Radon, Appendix E
E-6
•ill
-------
pCi/1. The proportions of systems exceeding the potential regulatory levels then decline rapidly,
until only approximately 3.1 percent of the NTNCWS are predicted to exceed the NAS AMCL
value of 4,000 pCi/1.
As noted above, the estimates of the numbers of NTNCWS systems exceeding regulatory
limits are very uncertain, and depend heavily on the consistency !of the relationship between
radon levels in NTNCWS and community systems being the same throughout the country as they
are in the five states for which data are available. The extent to which this is true is not known.
Methods, Occurrence, and Monitoring Document for Radon, Appendix E
E-7
-------
i; iiilti!
------- |