Environ. Sci. Technol.
2007, 41, 2123-2130
Factors Associated with Sources,
Transport, and Fate of Volatile
Organic Compounds and Their
Mixtures in Aquifers of the United
States
PAUL J. SQUILLACE AND
MICHAEL J. MORAN*
U.S. Geological Survey, 1608 Mountain View Road,
Rapid City, South Dakota 57702
Factors associated with sources, transport, and fate of
volatile organic compounds (VOCs) in groundwater from
aquifers throughout the United States were evaluated using
statistical methods. Samples were collected from 1631
wells throughout the conterminous United States between
1996 and 2002 as part of the National Water-Quality
Assessment (NAWQA) Program of the U.S. Geological
Survey. Water samples from wells completed in aquifers
used to supply drinking water were analyzed for more than
50 VOCs. Wells were primarily rural domestic water
supplies (1184), followed by public water supplies (216);
the remaining wells (231) supplied a variety of uses. The
median well depth was 50 meters. Age-date information shows
that about 60% of the samples had a fraction of water
recharged after 1953. Chloroform, toluene, 1,2,4-trimethyl-
benzene, and perchloroethene were some of the
frequently detected VOCs. Concentrations generally were
less than 1 g/L. Source factors include, in order of
importance, general land-use activity, septic/sewer density,
and sites where large concentrations of VOCs are
potentially released, such as leaking underground storage
tanks. About 10% of all samples had VOC mixtures that
were associated with concentrated sources; 20% were
associated with dispersed sources. Important transport factors
included well/screen depth, precipitation/groundwater
recharge, air temperature, and various soil characteristics.
Dissolved oxygen was strongly associated with VOCs
and represents the fate of many VOCs in groundwater.
Well type (domestic or public water supply) was also an
important explanatory factor. Results of multiple analyses
show the importance of (1) accounting for both dispersed
and concentrated sources of VOCs, (2) measuring
dissolved oxygen when sampling wells to help explain the
fate of VOCs, and (3) limiting the type of wells sampled
in monitoring networks to avoid unnecessary variance in
the data, or controlling for this variance during data analysis.
Introduction
Groundwater is used by about one-half of the population of
the United States as a source of potable water, including
nearly all of the 40 million or more people served by domestic
* Corresponding author phone: (605) 394-3244; fax: (605 355-
4523; e-mail: mjmoran@usgs.gov.
10.1021/es061079w Not subject to U.S. Copyright. Publ. 2007 Am. Chem. Soc.
Published on Web 03/06/2007
water supplies (1). Concern about the quality of this heavily
used resource has led to many small-scale investigations that
define the risk and remediation of concentrated sources
where contaminants are released at one location. A comple-
mentary interest in dispersed sourcesscontaminants re-
leased over large areas from activity prevalent in a given
settingshas led to large-scale water-quality investigations
of aquifers.
As part of the National Water-Quality Assessment
(NAWQA) Program of the U.S. Geological Survey (USGS), the
source, transport, and fate of VOCs in shallow groundwater
beneath new residential/commercial areas was previously
investigated and reported (2). These results could be helpful
for well head protection strategies in new development
surrounding major metropolitan areas. Also as part of the
NAWQA program, VOCs in deeper major aquifers were
investigated (3) by sampling existing water-supply wells.
These large-scale studies can provide an indication of aquifer
vulnerability and the quality of the water in the aquifer as
a whole, as well as identify contaminants that present the
greatest risk to aquifers. The purpose of this paper is to
summarize factors associated with the sources, transport,
and fate of 10 frequently detected VOCs in these aquifers
using various statistical techniques. These factors were
derived from ancillary data and were categorized as being
related to sources, transport, and fate processes. Water
samples for this analysis were collected between 1996 and
2002 as part of 55 NAWQA aquifer studies across the United
States.
Experimental Section
NAWQAs Aquifer Studies. Groundwater samples from 55
aquifer studies were collected between 1996 and 2002 as
part of the NAWQA Program of the USGS (Figure 1 and Table
1 in the Supporting Information). About 30 existing wells
were sampled for each aquifer study. These studies have
three unique characteristics. First, samples were collected
before treatment to define source-water quality in the aquifer.
Second, sampled wells were spatially distributed and ran-
domly selected among existing wells within the targeted
aquifer without respect to land use. However, sampled wells
mostly were in rural areas of the nation because urban areas
constitute a small fraction of the total landscape. Aquifers
were selected for investigation because of their importance
as a supply of potable water. Third, most of the samples were
collected from domestic water supplies; 73% of the samples
(1, 184) were collected from domestic water supplies, 13%
(216) were from public water supplies, and the remaining
14% (231) were from a variety of other well types such as
irrigation or monitoring wells (Table 1 in the Supporting
Information). Domestic water supplies were sampled more
frequently because they were most commonly available in
many parts of the investigated aquifers.
NAWQA aquifer studies are intended to be resource
assessments of the selected aquifer; consequently, sampled
wells are spatially distributed and not stratified on the basis
of factors such as water use, land use, or population density
(4). Sampled wells were in areas where population density
is higher on average than for the conterminous United States
as a whole; the median population density near wells sampled
for aquifer studies (about 20 people/km2) is one order of
magnitude larger than the median density for the conter-
minous United States (about two people/km2) but most of
the sampled wells were still in rural areas (Figure 2 in the
Supporting Information). These aquifer studies, conse-
quently, are groundwater resource assessments of rural areas
VOL. 41, NO. 7, 2007 / ENVIRONMENTAL SCIENCE & TECHNOLOGY 9 2123
FIGURE 1. Statistical summary of detection frequencies of chloroform, methylene chloride, and chloromethane in young oxic water
compared with old anoxic groundwater along a hypothetical groundwater flowpath. (Detection frequencies were calculated using no
common assessment level.)
where groundwater is used principally for domestic water
supply. More information about the design of these studies
has been published by Gilliom and others (4).
Sampling and Analytical Methods. VOC analyses were
done for more than 50 VOCs selected by the VOC National
Synthesis Team of NAWQA (5). All analyses were done at the
USGS National Water Quality Laboratory in Denver, Colo.,
using the method that provided the lowest concentration
information currently in use (2006) at the USGS National
Water Quality Laboratory (6). Analysis was done by use of
purge-and-trap, capillary column gas chromatography/mass
spectrometry. All concentrations reported by the laboratory
were used in this analysis including the smallest concentra-
tions with estimated values. Nondetect values were recorded
as less than the long-term method detection level. A subset
of 10 VOCs listed in the Supporting Information Table 2 were
frequently detected VOCs and were selected for analysis in
this paper. All VOC data are available at http://water.usgs.gov/
nawqa/vocs/national_assessment/ and have been discussed
in more detail by Moran and others (7). Sampling protocols
and quality-assurance/quality-control plans are described
by Koterba and others (8) and Moran and others (7).
Ancillary Data. A variety of ancillary data were used to
represent hydrogeologic and anthropogenic factors that can
affect the source, transport, and fate of VOCs in groundwater.
Twenty-four hydrogeologic and 24 anthropogenic factors
were tested (7). A 500 m and 1000 m radius buffer was used
in calculating land-use characteristics around the well. The
selection of a 500 m buffer is based on previous investigations
(8-10). A 1000 m radius buffer was used when counting the
number of potential concentrated sources of VOCs (for
example, leaking underground storage tanks) associated with
a sampled well because of uncertainty in locating these
concentrated sources. Previous publications (2, 7, 11) discuss
in more detail the limitations, assumptions, and sources of
the ancillary data.
2124 9 ENVIRONMENTAL SCIENCE & TECHNOLOGY / VOL. 41, NO. 7, 2007
Well depths ranged from 4 to 825 m, with a median depth
of 50 m. Domestic water supplies had a median well depth
of 45 m, and public water supplies had a median depth of
120 m. There were 922 wells finished in unconfined aquifers
and 381 wells in confined aquifers; for the remaining 328
wells, aquifer types were mixed or not identified. Dissolved-
oxygen concentrations in groundwater were classified as oxic
or anoxic. Water was classified as oxic if dissolved-oxygen
concentrations were greater than 0.5 mg/L and anoxic if
dissolved-oxygen concentrations were less than or equal to
0.5 mg/L.
Ancillary information was classified as relating to source,
transport, or fate of VOCs, with the exception of well type,
which was classified as an indeterminate factor because the
reason why it affects VOC detection is uncertain. For example,
pumping rates, travel times in groundwater, and size of
contributing areas are generally different for domestic and
public water supplies, and all of these factors can affect VOC
detection (12).
Statistical Methods. Statistics used for this analysis are
described in detail by Menard (13) and Helsel and Hirsch
(14), and are summarized by Squillace and others (2). An
alpha level of 0.05 was used for all statistical analyses to
determine significant differences and associations. The
concentration data used in this analysis were not censored
to a common reporting level, for example 0.2 g/L, because
(1) the same laboratory and analytical methods were used
for all samples, (2) censoring would reduce the ability to
show associations between a particular VOC and explanatory
factors, and (3) statistical methods such as logistic regression
are done for each VOC individually.
Logistic regression models were used to look for associa-
tions between groundwater quality at a particular sampled
well and current land-use activity within a circular buffer of
500-m radius around the well. Various hydrogeologic,
geochemical, and land-use conditions could adversely affect
the results (2); nevertheless, several previous studies have
shown a relation between land use and water quality (9, 10,
15-19). For each VOC, various logistic regression models
are possible, but information from a variety of statistical
measures (2) was used to select the final model. Every final
model for every VOC had a significant overall likelihood ratio
statistic p-value (<0.05), and every variable in every final
model had a significant Wald statistic p-value (<0.05).
The results of these models show associations only; they
were not developed to be predictive models. The Hosmer-
Lemeshow (H-L) test was used as a diagnostic for the logistic
regressions; however, for situations where there is colinearity
among covariates, the H-L statistic can perform poorly.
Although the H-L test value is low for some regression
models, this does not mean that there is not a valid association
between the response variable and the explanatory variables,
as indicated by the Wald statistic for each individual variable.
For logistic regression models, unstandardized coefficients
were recalculated in standard deviation units so that the
magnitude of the standardized coefficients could be directly
compared (11, 13).
Quantile plots were used to show the distribution of the
sample data. The quantile of a sample is the data point
corresponding to a given fraction of the data at or less than
a particular concentration (14). Quantile plots are sample
approximations of the cumulative distribution function of a
continuous variable. Direct comparisons between concen-
tration distributions in two data sets also were displayed by
plotting the quantiles of one data set in relation to the second
data set in quantile/quantile plots (14).
A mixture is defined as a unique combination of two or
more particular compounds, regardless of the presence of
other compounds that may occur in the same sample. The
ratio of the actual detection frequency of the mixture to the
theoretical probability of detecting a mixture, called the
detection frequency ratio (DFR), can be used to show
associations with possible sources. If a mixture had a DFR
close to 1, then the frequency of these VOCs occurring
together can be explained by chance and, consequently,
random co-occurrence of VOCs is implied. The DFR can be
somewhat larger than 1 for some random mixtures if there
are variations in aquifer vulnerability given a uniform
contaminant load. In vulnerable parts of the aquifer, many
VOCs can reach the groundwater because transport to the
water table is quick, with little degradation. In less vulnerable
parts of the aquifer, only the more mobile and persistent
VOCs reach the groundwater. If a mixture had a DFR much
greater than 1, then the frequency of these VOCs occurring
together cannot be explained by chance; consequently,
nonrandom co-occurrence of VOCs is implied. These non-
random mixtures could be derived from, for example, leaking
underground storage tanks where multiple VOCs can be
present. DFR could also be large for some mixtures if one
VOC is transformed to a second and they both occur together
in one sample.
Results and Discussion
Logistic regression analyses were used to identify factors
associated with the sources, transport, and fate of 10 VOCs
(Table 1). The factors associated with each VOC are listed in
rank order of importance based on the value of the
standardized coefficient. The relative importance of each
factor (or groups of factors) is revealed by the number of
times they were used in the final 10 logistic regression models
and the individual ranks of the factors (Table 3 in the
Supporting Information).
Sources. Concentrations of VOCs were generally less than
1 g/L in aquifer studies and were associated with multiple
dispersed sources and concentrated sources. These results
are for primarily domestic wells in rural areas of the nation,
and different results would be expected had the data been
from public supply wells in urban areas of the nation.
Dispersed sources were more strongly associated with the
10 VOCs in groundwater than concentrated sources; all 10
VOCs were associated with dispersed sources such as land
use, densities of septic/sewer systems, and areas of methyl
tert-butyl ether (MTBE) use (Table 1 and Supporting Infor-
mation Table 3). Land-use factors (year of construction,
percent urban land, and percent undeveloped land) indicated
the effect of age and intensity of urban development. VOCs
were all associated with more urban development, with the
exception of chloromethane (which was positively associated
with undeveloped land, see ref 11 for more discussion on
this association). Seven VOCs (all except bromodichlo-
romethane, chloromethane, and methylene chloride) were
associated with a concentrated source; however, according
to the standardized coefficients these sources were not as
important as dispersed sources (Table 1).
Although data used to represent various sources could be
erroneous in some situations, the individual factors probably
best represent the intended source. Some factors are
moderately associated with each other but can be used
together in the same model if they represent different sources
(20) and p values are significant (20). The amount of urban
land use is thought to be a surrogate for several dispersed
sources; on the other hand, counts of Resource Conservation
and Recovery Act (RCRA) sites, leaking underground storage
tanks (LUST), and underground storage tanks (UST) sites
are probably the best indicators of potential concentrated
sources.
The detection of VOCs found in gasoline (1,2,4-trimeth-
ylbenzene (TMB), toluene, and MTBE) were associated with
MTBE-use areas, septic-system density, age of construction,
and number of LUST and UST gasoline sites near the sampled
well. The intensity of MTBE use in gasoline was strongly
associated with MTBE detection in groundwater. The number
of LUST or UST gasoline sites within 1000 m of the sampled
well was a significant source factor associated with TMB,
toluene, and MTBE. Subsurface leakage or surface runoff
from these sites may be a source of these three gasoline-
related VOCs.
The detection of solvents and disinfection byproducts
were associated with septic-system density, urban land use,
and number of RCRA sites near the sampled well. Septic
systems had a positive association with the detection of five
VOCs (chloroform, methylene chloride, 1,1,1-trichloroethane
(TCA), trichloroethene (TCE), and perchloroethene (PCE)).
Three VOCs (chloroform, TCE, and PCE) were associated
with septic systems and public water supplies. For these three
VOCs, it is possible that public water supplies sampled for
these studies are more affected by septic systems than even
domestic water supplies. Although chloroform was more
strongly associated with public water supplies, it was still
detected in 26% of the domestic water supplies. Chloroform
can be formed in septic systems if, for example, household
bleach is present. Triclosan, an antimicrobial agent used in
household dishwashing soaps, also reacts with chlorinated
water to produce chloroform (21).
Transport. A VOC detected in groundwater would, in
general, first require a release from a source and then
subsequent transport to groundwater. Factors that potentially
affect the transport of VOCs to groundwater are dependent
on where the release occurred. If the release is at land surface
(for example, a spill), climate and near-surface soil charac-
teristics could affect contaminant transport to groundwater.
If the release is below land surface (for example, LUST sites
or septic systems/sewer lines), climate would be less
important; furthermore, soil characteristics at land surface
also might not be relevant unless they are representative of
material deeper in the unsaturated zone.
VOL. 41, NO. 7, 2007 / ENVIRONMENTAL SCIENCE & TECHNOLOGY 9 2125
TABLE 1. Logistic Regression Analyses for 10 VOCs Frequently Detected in NAWQA Aquifer Studies
Hosmer/Lemeshow
statistic (p-value,
larger value
type of wald statistic standardized indicates
compound explanatory variable variablea p-value coefficient a better model)
bromodichloromethane oxic/anoxic (oxic ) 0, anoxic ) 1) F 0.000 -0.18 0.38
people using sewer systems/km2 (natural log) S 0.000 .17
groundwater recharge (mm/yr) T 0.025 -.10
DWS/PWS (DWS ) 0, PWS ) 1) I 0.004 .09
chloroform percent urban land use (NLCDE) S 0.024 .91 <.05
oxic/anoxic (oxic ) 0, anoxic ) 1) F 0.000 -.31
annual precipitation (cm/yr) T 0.000 .19
DWS/PWS (DWS ) 0, PWS ) 1) I 0.000 .17
percent hydric soils T 0.000 -.16
people using septic systems/km2 (natural log) S 0.000 .13
count USEPA RCRA sites S 0.014 .08
chloromethane oxic/anoxic (oxic ) 0, anoxic ) 1) F 0.000 .12 .57
percent silt in soil T 0.004 .08
percent undeveloped land (NLCDE) S 0.025 .08
methylene chloride DWS/PWS (DWS ) 0, PWS ) 1) I 0.043 -.10 .19
well depth (m) T 0.014 -.10
people using septic systems/km2 (natural log) S 0.000 .08
percent sand in soil T 0.001 -.07
TCA oxic/anoxic(oxic ) 0, anoxic ) 1) F 0.000 -.24 .86
depth to top of well screen (m) T 0.000 -.23
soil organic matter, percent by weight T 0.002 -.16
annual air temperature (C) T 0.000 -.15
people using septic systems/km2 (natural log) S 0.000 .14
percent urban land use (NLCDE) S 0.000 .12
count USEPA RCRA sites S 0.000 .09
median year of construction S 0.029 -.05
TCE percent urban land use (NLCDE) S <0.05 .91 .43
oxic/anoxic (oxic ) 0, anoxic ) 1) F 0.000 -.31
annual precipitation (cm/yr) T <0.05 .19
DWS/PWS (DWS ) 0, PWS ) 1) I <0.05 .17
percent hydric soils T <0.05 -.16
people using septic systems/km2 (natural log) S <0.05 .13
count USEPA RCRA sites S <0.05 .08
PCE depth to top of well screen (m) T 0.000 -.21 0.34
oxic/anoxic (oxic ) 0, anoxic ) 1) F 0.000 -.18
DWS/PWS (DWS ) 0, PWS ) 1) I 0.000 .15
percent urban land use (NLCDE) S 0.000 .15
people using septic systems/km2 (natural log) S 0.005 .09
count USEPA RCRA sites S 0.031 .06
MTBE annual precipitation (cm/yr) T 0.000 .27 .48
MTBE use area S 0.000 .26
depth to top of well screen (m) T 0.005 -.26
DWS/PWS (DWS ) 0, PWS ) 1) I 0.000 .20
annual air temperature (C) T 0.001 -.15
oxic/anoxic (oxic ) 0, anoxic ) 1) F 0.008 -.11
count VISTA gasoline LUST sites S 0.000 .10
TMB people using septic systems/km2 (natural log) S 0.000 -.17 .07
annual air temperature (C) T 0.000 -.11
annual precipitation (cm/yr) T 0.012 -.08
count VISTA gasoline UST sites S 0.007 .07
toluene annual air temperature (C) T 0.000 -.13 .22
median year of construction S 0.000 -.10
count VISTA gasoline LUST sites S 0.000 .10
DWS/PWS (DWS ) 0, PWS ) 1) I 0.008 -.10
oxic/anoxic (oxic ) 0, anoxic ) 1) F 0.006 -.08
percent hydric soils T 0.002 .08
a Type of variable: S, source; T, transport; F, fate; I, indeterminate. km2, square kilometers; mm/yr, millimeters per year; cm/yr, centimeters

per year; m, meters; C, degree Celsius. DWS, domestic water supply; PWS, public water supply; NLCDE, national land cover data enhanced;
USEPA, U.S. Environmental Protection Agency; RCRA, Resource Conservation and Recovery Act; TCA, 1,1,1-trichloroethane; TCE, trichloroethene;
MTBE, methyl tert-butyl ether; TMB, 1,2,4-trimethylbenzene; LUST, leaking underground storage tank; UST, underground storage tank.

Important transport factors include air temperature,
various soil characteristics, groundwater recharge/precipita-
tion, and well/screen depth. Cool air temperature was an
important transport factor for the detection of four VOCs
(TCA, MTBE, TMB, and toluene, Table 1). In cool climates,
volatilization of MTBE, TMB, and toluene from a gasoline
spill at land surface would be reduced, increasing the
potential of these VOCs to infiltrate to groundwater with
recharge water.
Several soil characteristics were important for the detec-
tion of six VOCs (chloroform, TCE, toluene, chloromethane,
methylene chloride, and TCA). A soil is classified as hydric

2126 9 ENVIRONMENTAL SCIENCE & TECHNOLOGY / VOL. 41, NO. 7, 2007

if it is sufficiently wet in the upper part to develop anoxic
conditions during the growing season; consequently, this
soil characteristic could be more related to the fate of VOCs
than to transport. Nevertheless, chloroform and TCE were
associated with non-hydric soils, and toluene was associated
with hydric soils. These associations are consistent with the
stability of these VOCs in oxic and anoxic conditions. Two
VOCs (chloromethane and methylene chloride) were as-
sociated with less permeable soils. Slow movement of water
through the unsaturated zone may promote the degradation
of chloroform and formation of these two degradation
products. TCA was the only VOC associated with low organic
carbon content in soils.
High rates of groundwater recharge/precipitation were
associated with the detection of chloroform, TCE, and MTBE;
in contrast, low rates were associated with the detection of
bromodichloromethane and TMB. VOCs also are more
frequently detected in more populated areas for two rea-
sons: (1) the number of sources increase, and (2) recharge
significantly increases ( ref 11 and Supporting Information
Figure 3), hastening transport of VOCs through the unsatur-
ated zone to groundwater and reducing the opportunity for
degradation. It is not known why population density had a
monotonic association with recharge in rural and low-density
residential areas (Supporting Information Figure 3); in fact,
this association may not be related to anthropogenic activity
but could simply show that population density is larger in
areas where water is more abundant. For example, popu-
lation density and groundwater recharge were larger in the
northeast; in contrast, population density and ground-
water recharge were smaller in the southwest. VOCs associ-
ated with low recharge rates are more difficult to explain,
but they could be related to the source. If a source (for
example, a leaky sewer line) releases a small volume and
small concentrations of particular VOCs (for example,
bromodichloromethane and TMB), then low amounts of
recharge would not dilute concentrations to below the
reporting level.
Shallow well depth and shallow depth to the top of the
well screen (Table 1) were important factors in the detection
of four VOCs (methylene chloride, TCA, PCE, and MTBE).
Travel times to deeper wells are generally longer, allowing
more time for degradation along the flowpath or not allowing
sufficient time for VOCs to reach the well; there was a
significant association between age of the water and well
depth and depth to the top of the screen interval.
Confined aquifers were not associated with fewer VOC
detections compared with unconfined aquifers. Paired
detection frequencies of 10 VOCs listed in Table 1 were
compared between aquifers that were classified as confined
or unconfined. According to Wilcoxon signed rank, there
was no statistically significant difference in the detection
frequencies of these 10 VOCs in the two types of aquifers;
furthermore, the detection frequency of any VOC in uncon-
fined aquifers (67%) was not significantly different than
confined aquifers (62%). One may expect that water in
confined aquifers would be older than unconfined aquifers,
because longer flow paths from the recharge areas should be
associated with confined aquifers; however, the age of
groundwater samples collected from unconfined aquifers (n
) 386) was not significantly different than the age of samples
collected from confined aquifers (n ) 256). It seems that
leakage through the confining layers was sufficient to allow
some recent (post-1953) recharge into the confined aquifers.
Two logistic regression models were developed and compared
to determine whether dissolved oxygen or aquifer type
(confined/unconfined) was the best explanatory variable.
Both models used chloroform (the most frequently detected
VOC) as the dependent variable. The seven explanatory
variables for chloroform listed in Table 1 were used in one
model, and the other model used the same variables except
aquifer type was used instead of oxic/anoxic. The model
using dissolved oxygen was significantly better than the model
using aquifer type as indicated by the Akaikes Information
Criteria. Furthermore, among the seven explanatory variables,
dissolved oxygen was the second strongest in one model and
aquifer type was the seventh strongest factor in the other
model. These results show that although dissolved oxygen
concentration is associated with the type of aquifer, dissolved
oxygen also affects the fate of VOCs in groundwater;
consequently, dissolved oxygen is a better explanatory factor
than type of aquifer.
Fate. Young groundwater in recharge areas of aquifers
generally has larger dissolved-oxygen concentrations com-
pared to old groundwater in discharge areas because
dissolved oxygen becomes depleted along an aquifer flowpath
(22). Samples collected in the NAWQA studies show that
groundwater dated younger than 1953 (post bomb water)
had significantly larger dissolved-oxygen concentrations
(Figure 4 in the Supporting Information) and more VOCs per
sample than groundwater dated 1953 or older. Based on 720
samples with age-date information, about 60% of sampled
groundwater contained some fraction of recharge that
entered the aquifer after 1953.
Two-thirds of the samples from aquifers were oxic;
consequently, compounds that biodegrade more slowly in
oxic groundwater should be more persistent, contributing
to a greater detection frequency. In general, halogenated
aliphatic VOCs, such as TCE, biodegrade more slowly than
petroleum hydrocarbons in oxic conditions; consequently,
it is not surprising that among the 10 frequently detected
VOCs, halogenated aliphatic VOCs were the most commonly
detected, followed by petroleum hydrocarbons, and then
the fuel oxygenate MTBE (Table 4 in the Supporting
Information). The use of VOCs has increased since 1953 and
also could contribute to more frequent detection of some
VOCs in oxic water.
Chloroform, methylene chloride, and chloromethane were
frequently detected VOCs (when using no common assess-
ment level), but the biodegradation of chloroform can
partially explain the detection of methylene chloride and
chloromethane in anoxic conditions. By reductive dechlo-
rination reactions, chloroform can be transformed to meth-
ylene chloride, which can be further transformed to chlo-
romethane (23). The transformation of at least some chlo-
roform to methylene chloride and chloromethane in ground-
water is supported by three observations. First, among
samples with young oxic water (recharged after 1953), the
detection frequency of chloroform was about 5 times higher
compared to old (recharged during or before 1953) anoxic
water (Figure 1). Although chloroform loading to groundwater
may not be identical for this oxic and anoxic water, the large
difference in detection frequencies is compelling evidence
that chloroform transformation has occurred in anoxic water.
Second, among samples with oxic water, chloroform makes
up more than 95% of the total molar concentrations of
chloroform, methylene chloride, and chloromethane; how-
ever, in anoxic water, methylene chloride and chloromethane
make up more than one-half of the total molar concentrations
of these three compounds (Figure 5 in the Supporting
Information). Third, detection frequencies of methylene
chloride or chloromethane in young oxic water were sig-
nificantly lower than detection frequencies in old anoxic
water, a finding that supports the possible formation of
methylene chloride and chloromethane with time and redox
condition (Figure 1).
Dissolved-oxygen concentration was the most frequently
occurring explanatory factor in logistic regression models
for the 10 VOCs investigated. Seven VOCs were associated
with oxic conditions (bromodichloromethane, chloroform,
VOL. 41, NO. 7, 2007 / ENVIRONMENTAL SCIENCE & TECHNOLOGY 9 2127
FIGURE 2. Quantile/quantile plot of the concentrations of 10 volatile organic compounds (VOCs) in oxic and anoxic conditions. For all
VOCs, except toluene and 1,2,4-trimethylbenzene, the concentrations are significantly different between oxic conditions and anoxic
conditions.
TCA, TCE, PCE, MTBE, and toluene), and only one VOC
(chloromethane) was associated with anoxic conditions
(Table 1). The results of this multivariate analysis can be
different than univariate analysis; however, multivariate
analysis is probably more reliable because it controls for
differences in source and transport factors that also can affect
the detection of VOCs in groundwater.
Dissolved-oxygen concentrations in groundwater not only
affected the detection (as shown by logistic regression
analysis) but also the concentrations of most VOCs. All VOCs
shown in the quantile/quantile plot in Figure 2 had signifi-
cantly different concentrations in oxic conditions compared
with anoxic conditions (except toluene and TMB). MTBE,
bromodichloromethane, TCE, TCA, PCE, and chloroform had
significantly larger concentrations in oxic conditions than in
anoxic conditions. These larger concentrations are expected
on the basis of degradation studies that show slower
degradation rates for many of these VOCs in oxic water when
compared with anoxic water (24, 25). Dissolved oxygen seems
to have the greatest effect on the concentrations of TCE;
concentrations of TCE in oxic water generally were about 2
orders of magnitude larger than those in anoxic water at an
equal quantile (Figure 2). The slopes of the PCE, chloroform,
and bromodichloromethane curves increase as concentra-
tions decrease (Figure 2). This change in slope could indicate
greater stability of these VOCs in oxic water (with decreasing
concentrations) compared with anoxic water. Chloroform,
for example, seems to become more stable in oxic conditions
at concentrations less than 0.5 g/L. Other VOCs (MTBE,
bromodichloromethane, TCA, and toluene) show a similar
pattern but the inferred slopes inflect at different concentra-
tions. Chloromethane and methylene chloride had signifi-
cantly larger concentrations in anoxic conditions than in
2128 9 ENVIRONMENTAL SCIENCE & TECHNOLOGY / VOL. 41, NO. 7, 2007
oxic conditions. Gasoline hydrocarbons, such as toluene and
TMB, are known to be more stable in anoxic conditions than
in oxic conditions (25). The concentrations detected in these
studies, however, generally were less than 1 g/L, and at
these concentrations, degradation rates may slow consider-
ably (26) and may become similar to those in oxic conditions.
Quantile plots showed that urban land use and people
using septic systems had a greater effect on concentrations
of chloroform in groundwater when dissolved oxygen
concentrations favor the preservation of chloroform (Figure
6 in the Supporting Information). Chloroform is more stable
in oxic than anoxic conditions, and the effect of urban land
use and the use of septic systems on the concentrations of
chloroform was greater for oxic conditions than anoxic
conditions (Figure 6 the in Supporting Information).
Well Type. Well type (domestic/public water supplies)
was a significant factor for seven VOCs. Five VOCs (bro-
modichloromethane, chloroform, MTBE, TCE, and PCE) were
associated with public water supplies, and two VOCs (meth-
ylene chloride and toluene) were associated with domestic
water supplies (Table 1). Well type was still a significant factor
even after controlling for a number of obvious differences
like well depth and land use around the well. This indicates
that there are a number of other important factors that were
not available as ancillary information (like pumping rates
and land use in the actual capture zone) that are probably
necessary to better explain the differences in VOC detections
between these well types. More VOCs were associated with
public water supplies than domestic water supplies even
though public water supplies generally were much deeper
than domestic water supplies. Larger contributing areas
associated with public water supplies (12) and larger open
intervals could contribute to the stronger associations of five
VOCs with public water supplies. Public water supplies,
however, also had significantly larger amounts of urban land
use within 500 m of the sampled well compared with domestic
water supplies; the median amount of urban land use for
domestic water supplies was 0.2% compared to 55% for public
water supplies. Water from public water supplies generally
contain a blend of water with a continuum of ages, but the
apparent age of water from 84 public water supplies was
significantly younger than water from 528 domestic water
supplies. Faster travel times associated with water from public
water supplies would decrease the opportunity for degrada-
tion of VOCs and also could partially explain why five VOCs
were associated with public-water supplies compared with
two VOCs associated with domestic-water supplies.
For some VOCs, the association with a particular well
type could be evident only when dissolved-oxygen conditions
favor the preservation of the VOC. Figure 7 in the Supporting
Information, for example, shows that the detection frequency
of chloroform was much larger in public water supplies (80%)
than in domestic water supplies (30%) in oxic conditions
and that the detection frequencies in the two types of wells
were nearly identical (12%) in anoxic conditions. These results
show the importance of limiting the type of wells sampled
in monitoring networks to avoid unnecessary variance in
the data, or controlling for this variance during data analysis.
Factors such as well depth and land use are not adequate to
control all the variance between domestic and public water
supplies.
Mixture Analysis. The detection of mixtures of VOCs is
associated with well type. Samples from public water supplies
had the largest detection frequency of mixtures (53%),
followed by monitoring, irrigation, and other types of wells
(36%), and domestic water supplies (34%). The greater
incidence of mixtures in samples from public water supplies FIGURE 3. Quantile plot of number of Resource Conservation and
than from domestic water supplies could be related to differ- Recovery Act (RCRA) sites within 1000 m of sampled wells with
ences in land use and pumping rates. Public water supplies random mixtures, nonrandom mixtures, and all remaining samples.
are more likely than domestic water supplies to capture The detection frequencies of VOCs found in these mixtures
groundwater that has both short and extended flow paths. are shown in Table 4 in Supporting Information. Mixtures
Integrating these flow paths makes it more likely to detect were ranked according to their detection frequency ratio
multiple VOCs in a single public water supply sample (12). (actual divided by theoretical probability of detection, DFR),
Most releases from concentrated sources, for example, which ranged from 1 to about 3000 (Table 5 in Supporting
land disposal and waste storage facilities (27), contain Information). Almost all mixtures had a DFR greater than 1,
multiple VOCs. The detection of mixtures could indicate a indicating that mixtures normally had detection frequencies
concentrated source with the potential for larger concentra- greater than what would be expected by chance. The most
tions of VOCs; however, analysis of data indicated no frequently detected mixtures had some of the lowest DFRs,
difference in concentrations of VOCs if they occurred alone indicating a fairly random co-occurrence pattern; for ex-
or in mixtures. Among the 10 frequently detected VOCs, five ample, the mixture of PCE and chloroform was the most
compounds (TCA, MTBE, TCE, PCE, and bromodichlo- frequently detected mixture (8.5%) and had a DFR of 2
romethane) rarely occurred by themselves (less than about (mixture 23, Table 5 in the Supporting Information). Some
10% when detected). For these five VOCs, it is impossible to of the less frequently detected mixtures had very large DFRs;
compare concentrations of VOCs occurring alone with those for example, the mixture of chloroform, PCE, 1,1-dichloro-
in mixtures because of insufficient data. The concentrations ethene, TCA, and TCE (mixture 102, Table 5 the in Supporting
of the remaining five VOCs (TMB, chloromethane, methylene Information) was detected in about 1% of the samples but
chloride, toluene, and chloroform) were not significantly was detected about 3000 times more frequently than can be
different, or the difference in median concentrations was explained by chance.
only 0.01 g/L, if these VOCs were found alone or in mixtures. Mixtures that had a DFR of less than about 3 were
In the following analysis, mixtures were identified without considered random; in other words, the frequency of these
regard to dissolved-oxygen concentration. This approach VOCs occurring together can be explained by chance (see
gives the best indication of mixtures that are most likely to the previous section on statistical methods). Mixtures that
be detected in the sampled groundwater; however, it will had a DFR greater than about 3 were considered nonrandom;
result in a bias toward mixtures common within the dominant in other words, the frequency of these VOCs occurring
redox condition (oxic conditions for data used in this together cannot be explained by chance. This division was
analysis). Dissolved oxygen, in fact, has a dramatic effect on based on differences in the chemical composition and the
the composition of mixtures. In oxic water, for example, the number of VOCs in samples with random and nonrandom
concentrations of TCE were significantly correlated with mixtures. Chloroform had a dominant effect on the com-
five other VOCs; in anoxic water, TCE was not correlated position of mixtures with DFRs less than about 3. Among the
with any VOC because TCE is not stable in anoxic water 37 mixtures rank ordered with the lowest DFRs (less than
(Table 1). about 3, Table 5 in the Supporting Information), chloroform
There were 102 mixtures that had detection frequencies is paired with another VOC in about one-half of these
of about 1% or more (Table 5 in Supporting Information). mixtures. Among the remaining 65 mixtures with the largest

VOL. 41, NO. 7, 2007 / ENVIRONMENTAL SCIENCE & TECHNOLOGY 9 2129

DFRs, only one mixture had chloroform paired with just one
other VOC. Samples with only random mixtures have fewer
VOCs (median ) 2.0) and generally have only one random
mixture (median ) 1.0); in contrast, samples with nonrandom
mixtures generally have more VOCs (median ) 4.5) and have
more than one nonrandom mixture (median ) 3.0). For more
discussion on random and nonrandom mixtures, see Squil-
lace and others (11).
Identification of random and nonrandom mixtures is
important because they are associated with different potential
sources of VOCs. There were significantly more RCRA sites
(concentrated sources) within 1000 m of sampled wells having
nonrandom mixtures than sampled wells having random
mixtures or any other sampled well (Figure 3). To create this
figure, all samples were placed into one of three groups:
samples with random mixtures, samples with nonrandom
mixtures, or all remaining samples. About 10% of all samples
had nonrandom VOC mixtures and about 20% of all samples
had random mixtures. Samples with nonrandom mixtures
were also associated with (in order of strength of association)
oxic conditions, public water supplies (as opposed to
domestic water supplies), more RCRA sites, and increased
groundwater recharge when compared to samples with only
random mixtures (Table 6 in the Supporting Information).
The factors in this model also show the importance of
understanding the source, transport, and fate of VOCs to
explain the detection of nonrandom mixtures.
Acknowledgments
We thank the many USGS personnel who collected and
compiled the water-quality data used in this paper. We also
thank Kerie Hitt, Barbara Ruddy, Curtis Price, and David
Wolock for compiling and analyzing data using geographic
information systems. Last, we thank Michael Focazio, Bernard
Nolan, Jon Scott, and Brian Katz for providing insightful
reviews that substantially improved this paper.
Note Added after ASAP Publication. The assignment of
copyright for the version published ASAP March 6, 2007 was
in error; the corrected version was published ASAP March
12, 2007.
Supporting Information Available
Tables indicating wells sampled, detection frequencies of
VOCs, logistic regression results, and information on mixtures
of VOCs in groundwater. Figures showing locations of well
networks and figures illustrating relations between occur-
rence of VOCs and various ancillary variables. This material
is available free of charge via the Internet at http://
pubs.acs.org.
Literature Cited
(1) Alley, W. M.; Reilly, T. E.; Franke, O. L. Sustainability of ground-
water resources. U. S. Geol. Surv. Circ. 1186, 1999.
(2) Squillace, P. J.; Moran, M. J.; Price, C. V. VOCs in shallow
groundwater in new residential/commercial areas of the United
States. Environ. Sci. Technol. 2004, 38 (20), 5327-5338.
(3) Zogorski, J. S.; and others. Volatile organic compounds in the
Nations major aquifers and in samples from domestic and
public wells. U. S. Geol. Surv. Circ. 1292, 2006.
(4) Gilliom, R. J.; Alley, W. M.; Gurtz, M. E. Design of the National
Water-Quality Assessment ProgramsOccurrence and distribu-
tion of water-quality conditions. U. S. Geol. Surv. Circ. 1112,
1995.
(5) Bender, D. A.; Zogorski, J. S.; Halde, M. J.; Rowe, B. L. Selection
procedure and salient information for volatile organic com-
pounds emphasized in the National Water-Quality Assessment
Program. U. S. Geol. Surv. Open-File Rep. 99-182, 1999.
(6) Connor, B. F.; Rose, D. L.; Noriega, M. C.; Murtagh, L. K.; Abney,
S. R. Methods of analysis by the U.S. Geological Survey National
Water Quality LaboratorysDetermination of 86 volatile organic
2130 9 ENVIRONMENTAL SCIENCE & TECHNOLOGY / VOL. 41, NO. 7, 2007
compounds in water by gas chromatography/mass spectrom-
etry, including detections less than reporting limits. U. S. Geol.
Surv. Open-File Rep. 97-829, 1998.
(7) Moran, M. J.; Zogorski, J. S.; Rowe, B. L. Approach to an
assessment of volatile organic compounds in the Nations
ground water and drinking-water supply wells. U. S. Geol. Surv.
Open-File Rep. 2005-1452, 2006.
(8) Koterba, M. T.; Wilde, F. D.; Lapham, W. W. Ground-water data-
collection protocols and procedures for the National Water-
Quality Assessment ProgramsCollection and documentation
of water-quality samples and related Data. U. S. Geol. Surv.
Open-File Rep. 95-399, 1995.
(9) Moran, M. J.; Zogorski, J. S.; Squillace, P. J. Chlorinated solvents
in groundwater of the United States. Environ. Sci. Technol. 2007,
41, 74-81.
(10) Nolan, B. T.; Hitt, K. J. Vulnerability of Shallow Ground Water
and Drinking-Water Wells to Nitrate in the United States.
Environ. Sci. Technol. 2006, 40 (24), 7834-7840.
(11) Squillace, P. J.; Moran, M. J. Factors associated with sources,
transport, and fate of volatile organic compounds in aquifers
of the United States and implications for ground-water man-
agement and assessments. U. S. Geol. Surv. Sci. Invest. Rep.
2005-5269, 2006.
(12) Stackelberg, P. E.; Kauffman, L. J.; Ayers, M. A.; Baehr, A. L.
Frequently co-occurring pesticides and volatile organic com-
pounds in public supply and monitoring wells, southern New
Jersey, USA. Environ. Toxicol. Chem. 2001, 20 (4), 853-865.
(13) Menard, S. Applied Logistic Regression Analysis, 2nd ed.; Sage
Publications: Thousand Oaks, CA, 2002.
(14) Helsel, D. R.; Hirsch, R. M. Statistical Methods in Water Resources;
Elsevier: New York, 1992.
(15) Eckhardt, D. A.; Stackelberg, P. E. Relation of ground-water
quality to land use on Long Island, New York. Ground Water
1995, 33 (6), 1019-1033.
(16) Grady, S. J.; Mullaney, J. R. Natural and human factors affecting
shallow water quality in surficial aquifers in the Connecticut,
Housatonic, and Thames River Basins; U. S. Geol. Surv. Water-
Resour. Invest. Rep. 98-4042, 1998.
(17) Nolan, B. T. Nutrients in ground waters of the conterminous
United States, 1992-1995. Environ. Sci. Technol. 2000, 34 (7),
1156-1165.
(18) Barbash, J. E.; Thelin, G. P.; Kolpin, D. W.; Gilliom, R. J. Major
herbicides in ground water results from the national water-
quality assessment. J. Environ. Quality 2001, 30 (3), 831-845.
(19) Robinson, J. L. Ground-water quality beneath an urban resi-
dential and commercial area, Montgomery, Alabama, 1999-
2000. U. S. Geol. Sur. Water-Resour. Invest. Rep. 02-4052, 2002.
(20) Nolan, B. T.; Hitt, K. J.; Ruddy, B. C. Probability of nitrate
contamination of recently recharged ground waters in the
conterminous United States. Environ. Sci. Technol. 2002, 36
(10), 2138-2145.
(21) Rule, K. L.; Ebbett, V. R.; Vikesland, P. J. Formation of chloroform
and chlorinated organics by freechlorine-mediated oxidation
of triclosan. Environ. Sci. Technol. 2005, 39, 3176-3185.
(22) Chapelle, F. H.; Widdowson, M.A.; Brauner, J.S.; Mendez, E., III;
Casey, C. C. Methodology for estimating times of remediation
associated with monitored natural attenuation. U. S. Geol. Surv.
Water-Resour. Invest. Rep. 03-4057, 2003.
(23) Egli, C.; Tschan, T.; Schlotz, R.; Cook, A. M.; Leisinger, T.
Transformation of tetrachloromethane to dichloromethane and
carbon dioxide by Acetobacterium woodi. Appl. Environ. Mi-
crobiol. 1988, 54 (11), 2819-2824.
(24) Kotzias, D.; Sparta, C. VOCs and water pollution. In Chemistry
and Analysis of Volatile Organic Compounds in the Environment;
Bloemen, H. J., Burn, J., Eds.; Blackie Academic and Profes-
sional: London, 1993; pp 175-199.
(25) Suarez, M. P.; Rifai, H. S. Biodegradation rates for fuel
hydrocarbons and chlorinated solvents in groundwater. Biorem.
J. 1999, 3 (4), 337-362.
(26) Roch, F.; Alexander, M. Inability of bacteria to degrade low
concentrations of toluene in water. Environ. Toxicol. Chem.
1997, 16, 1377-1383.
(27) Huling, S. G. Facilitated Transport. U.S. Environmental Protec-
tion Agency EPA/540/4-89/003; U.S. Environmental Protection
Agency: Washington, DC, 1989.
Received for review May 5, 2006. Revised manuscript re-
ceived January 30, 2007. Accepted January 30, 2007.
ES061079W