STATISTICAL EVALUATION OF VARIABLES AFFECTING OCCURRENCE

OF HYDROCARBONS IN AQUIFERS USED FOR PUBLIC SUPPLY, CALIFORNIA1

Matthew K. Landon, Carmen A. Burton, Tracy A. Davis, Kenneth Belitz, and Tyler D. Johnson2

ABSTRACT: The variables affecting the occurrence of hydrocarbons in aquifers used for public supply in Califor-nia were assessed based on statistical evaluation of three large statewide datasets; gasoline oxygenates also wereanalyzed for comparison with hydrocarbons. Benzene is the most frequently detected (1.7%) compound among17 hydrocarbons analyzed at generally low concentrations (median detected concentration 0.024 lg/l) in ground-water used for public supply in California; methyl tert-butyl ether (MTBE) is the most frequently detected (5.8%)compound among seven oxygenates analyzed (median detected concentration 0.1 lg/l). At aquifer depths used forpublic supply, hydrocarbons and MTBE rarely co-occur and are generally related to different variables; in shal-lower groundwater, co-occurrence is more frequent and there are similar relations to the density or proximity ofpotential sources. Benzene concentrations are most strongly correlated with reducing conditions, regardless ofgroundwater age and depth. Multiple lines of evidence indicate that benzene and other hydrocarbons detected inold, deep, and/or brackish groundwater result from geogenic sources of oil and gas. However, in recentlyrecharged (since ~1950), generally shallower groundwater, higher concentrations and detection frequencies ofbenzene and hydrocarbons were associated with a greater proportion of commercial land use surrounding thewell, likely reflecting effects of anthropogenic sources, particularly in combination with reducing conditions.

(KEY TERMS: organic chemicals; drinking water; environmental sampling; groundwater hydrology; geochemis-try; hydrocarbons; gasoline oxygenates.)

Landon, Matthew K., Carmen A. Burton, Tracy A. Davis, Kenneth Belitz, and Tyler D. Johnson, 2013. Statisti-cal Evaluation of Variables Affecting Occurrence of Hydrocarbons in Aquifers Used for Public Supply, California.Journal of the American Water Resources Association (JAWRA) 1-17. DOI: 10.1111/jawr.12129

INTRODUCTION

Biofuels, including ethanol, are expected to providea greater contribution to the fuel supply in the Uni-ted States (U.S.) in the future. Research in the pastdecade (Molson et al., 2000, 2002; Deeb et al., 2002;Ruiz-Aguilar et al., 2003; Mackay et al., 2006; Gomezand Alvarez, 2010; Corseuil et al., 2011) has shownthat ethanol in gasoline may cause plumes of benzenein groundwater to be larger than they would other-

wise be. Benzene mobility in the presence of ethanolor other fuel alcohols may result from several factors,including a decrease in benzene degradation ratesdue to consumption of dissolved oxygen by preferen-tial microbial oxidation of ethanol (Molson et al.,2000, 2002; Corseuil et al., 2011). A key step in deter-mining the vulnerability of public-supply wells to theincreased use of biofuels is to identify public-supplywells in key settings of the U.S. that have hadhistorical detections of fuel components and to evalu-ate what variables may explain those detections.

1Paper No. JAWRA-12-0177-P of the Journal of the American Water Resources Association (JAWRA). Received August 2, 2012; acceptedJuly 2, 2013. © 2013 American Water Resources Association. This article is a U.S. Government work and is in the public domain in the USA.Discussions are open until six months from print publication.

2Hydrologists, U.S. Geological Survey, 4165 Spruance Road, Suite 200, San Diego, California 92101 (E-Mail/Landon: landon@usgs.gov).

JOURNAL OF THE AMERICAN WATER RESOURCES ASSOCIATION JAWRA1

JOURNAL OF THE AMERICAN WATER RESOURCES ASSOCIATION

AMERICAN WATER RESOURCES ASSOCIATION

California provides an ideal location to evaluate therelation of environmental factors to the occurrence offuel components in public-supply wells for severalreasons. First, there are thousands of public-supplywells, providing the statistical mass necessary forcomprehensive analysis. Second, there is an abun-dance of ancillary data available to facilitate statisti-cal analysis; the U.S. Geological Survey (USGS) hasassembled extensive ancillary data for the CaliforniaState Water Resources Control Board’s (CA SWRCB)Groundwater Ambient Monitoring and Assessment(GAMA) program Priority Basin Project (PBP) (CASWRCB, GAMA. Accessed on July 29, 2011, at http://www.swrcb.ca.gov/gama/; USGS, GAMA PBP.Accessed on December 8, 2011 at: http://ca.water.usgs.gov/gama/). Third, California is an importantsetting in the national landscape, superimposinglarge urban populations dependent on groundwatersupplies, with a high density of fuel sources, andlarge groundwater withdrawals.

The objective of this study was to identify relationsof source, transport, and receptor variables to theoccurrence of hydrocarbons in groundwater used forpublic supply in California. Relations of variables tohydrocarbon distributions are compared to relationsfor gasoline oxygenates.

Previous studies have discussed the occurrence ofvolatile organic compounds (VOCs) and relations ofoccurrence to selected explanatory variables ingroundwater in parts of California (Shelton et al.,2001; Dawson et al., 2003; Belitz et al., 2004; Dens-more et al., 2004; Wright et al., 2004; Hamlin et al.,2005; Johnson and Belitz, 2009) or the U.S. (Grady,2003; Moran et al., 2005; Zogorski et al., 2006; Squil-lace and Moran, 2007). In comparison, this studyfocuses specifically on organic constituents derivedfrom spilled fuel and explores relations of the occur-rence of these constituents in aquifers primarily usedfor public supply in California to potential explana-tory variables. This study also includes analysis ofthree disparate datasets, having varying analyticalreporting limits and spatial distributions, to compareresults between different depths, land uses, and con-centration ranges of fuel constituents. In a previousstudy, Landon and Belitz (2012) described that ben-zene and other hydrocarbons detections in relativelydeep groundwater used for public supply in Californiaare more commonly attributable to geologic sources ofpetroleum (geogenic) than anthropogenic sourcesbased on analysis of two datasets. In contrast, thefocus of this study is to describe the results of statis-tical analysis of the relations of hydrocarbons andoxygenates to a broader range of explanatory vari-ables, including relations in relatively shallowergroundwater in more urban areas in California, andto compare results for hydrocarbons and oxygenates.

METHODS

Data Sources

This study, a partnership of the USGS PBP andU.S. Environmental Protection Agency (USEPA),includes data from three sources. First, the studyincludes data collected by the USGS as part of the Cal-ifornia State Water Board’s GAMA program PBP,hereafter referred to as GAMA data (USGS, GAMAPBP Publications, Data Series Reports. Accessed onDecember 8, 2011, http://ca.water.usgs.gov/gama/publications.html). Second, data assembled for drinkingwater regulatory monitoring purposes by the Califor-nia Department of Public Health (CDPH) wereincluded and are hereafter referred to as CDPH data(CDPH, Drinking Water Program. Accessed on Sep-tember 7, 2011, http://www.cdph.ca.gov/programs/Pages/DWP.aspx). Third, the study analyzed data fromthe USGS National Water Information System (NWIS)hereafter referred to as NWIS data (USGS, NWIS.Accessed on December 8, 2010, http://waterdata.usgs.gov/ca/nwis/). The GAMA and NWIS data include agetracers and ancillary water chemistry data, and use oflow laboratory reporting levels (LRLs) at a smallernumber of wells (~1,000-2,000) compared to the CDPHdata, which has less extensive ancillary data and LRLsthat are typically one to two orders of magnitudehigher, at a larger number of wells (~12,000).

GAMA and NWIS data were collected using proto-cols to ensure that representative samples are col-lected while minimizing the potential for samplecontamination (Koterba et al., 1995; USGS, variouslydated). Detailed descriptions of sample collection, anal-ysis, and quality-assurance (QA) methods and resultsfor GAMA data are available in previous publications(USGS, GAMA PBP Publications, Data Series Reports.Accessed on December 8, 2011, http://ca.water.usgs.gov/gama/pub lications.html). About 80% of the wellsin the GAMA data were public-supply wells or othertypes of large withdrawal wells, 10% were domesticwells, and 10% were observation wells. The CDPHdatabase lists wells used for public drinking water sup-plies, and includes wells from systems classified ascommunity (such as cities, towns, and mobile-homeparks), nontransient, noncommunity (such as schoolsand workplaces), and transient, noncommunity (suchas campgrounds and parks). About 45% of the wells inthe NWIS data were public-supply or other types ofproduction wells, 22% were domestic wells, and 33%were monitoring wells.

Because GAMA data were collected from wells thatwere randomly selected from spatially distributed gridcells covering priority groundwater basins (stratifiedrandom design), and had low LRLs and associated QA

JAWRA JOURNAL OF THE AMERICAN WATER RESOURCES ASSOCIATION2

LANDON, BURTON, DAVIS, BELITZ, AND JOHNSON

data (Belitz et al., 2003, 2010), GAMA data were usedfor evaluating occurrence and distribution of hydro-carbons and gasoline oxygenates in California ground-water. The more extensive CDPH data, with highermethod detection levels (MDLs), were used for assess-ing the distribution at higher concentrations and toevaluate explanatory relations identified using theGAMA data. The detection frequencies based on CDPHdata need to be regarded with caution because theMDLs were not always recorded in the database; thedetection frequencies in this study are reported rela-tive to the most common MDLs. The NWIS data, withlow LRLs similar to GAMA for most of the data, have amore clustered spatial distribution than the GAMAand CDPH data; the NWIS data are primarily for wellslocated in Southern California and the Central Valley.As a consequence, the NWIS data were not used forthe purpose of evaluating distribution, but were usedfor assessing relations to explanatory variables.

Analytes

Because of expected low detection frequencies ofhydrocarbons in public-supply wells (Hadley andArmstrong, 1991; Williams et al., 2002; Zogorski et al.,2006; Kulongoski et al., 2010; Landon et al., 2010;Bennett et al., 2011; Burton et al., 2011), this studycompiled data for 17 hydrocarbons and seven gasolineoxygenates or degradates included in standard USGSVOC analyses (Bender et al., 1999) (Table S1) in anattempt to have sufficient statistical mass to evaluatevariables explaining detections. Detection frequenciesof the gasoline oxygenate methyl tert-butyl ether(MTBE) have been slightly higher than for individualhydrocarbons in Southern California (e.g., Sheltonet al., 2001; Hamlin et al., 2005) and nationally (Zogor-ski et al., 2006). Benzene and MTBE were selected asthe primary dependent variables for hydrocarbons andoxygenates, respectively, because they were: (1) themost commonly occurring hydrocarbon and oxygenatein the GAMA data, (2) the most reliably measuredconstituents based on QA analysis. The sum-of-hydro-carbons concentration was selected as a secondarydependent variable (Table 1) because of the commonco-occurrence of multiple hydrocarbons.

Data Processing

A single analysis at each well was selected to avoidbiasing the data to those wells having many analysesover time. About 90% of the wells sampled by GAMAwere visited once; for the 10% of wells sampled morethan once, the sample date with the most completedata was selected. The GAMA data included 1,973

wells sampled during May 2004 through August 2010(median of November 2006). For CDPH data, themost recent concentration of each VOC and other tar-get constituents was selected and evaluated using theQA steps described below. The CDPH data included12,417 wells sampled between July 1984 and August2010 (median of March 2005); >75% of the data was

TABLE 1. Detection Frequencies for Benzene andSum of Hydrocarbons in GAMA and CDPH Data.

BenzeneSum of

Hydrocarbons MTBE

ConstituentCAS number 71-43-2 na 1634-04-4Threshold type MCL-CA na MCL-USThreshold value (lg/l) 1 na 13MaximumLT-MDL (lg/l)

0.013 na 0.08

GAMANumber of wells 1,972 1,973 1,972Detections at or aboveLT-MDL

33 53 114

Detection frequency(LT-MDL), %

1.67 2.69 5.78

Detections at or aboveCDPH MDL

4 na 1

Detection frequency(CDPH MDL), %

0.20 na 0.05

Detectedconcentrations,maximum, lg/l

78.9 79.8 28.3

Detectedconcentrations,median, lg/l

0.024 0.09 0.10

Detections abovethreshold

3 na 1

Detection frequencyabove threshold, %

0.15 na 0.05

CDPHNumber of wells 12,417 12,441 10,792Most frequent MDL,lg/l

0.5 0.5 3

Detections at or aboveCDPH MDL

16 71 32

Detection frequency(CDPH MDL), %

0.13 0.57 0.30

Detectedconcentrations,maximum, lg/l

140 200 500

Detectedconcentrations,median, lg/l

3.45 0.60 8.60

Detections abovethreshold

11 na 15

Detection frequencyabove threshold, %

0.09 na 0.14

Note: GAMA, Groundwater Ambient Monitoring and Assessment;CDPH, California Department of Public Health; MTBE, methyltert-butyl ether; CAS, chemical abstract service; LT-MDL, USGSlong-term method detection level; MCL-CA, California Depart-ment of Public Health maximum contaminant level; MCL-US,U.S. Environmental Protection Agency maximum contaminantlevel; na, not available; lg/l, micrograms per liter.

JOURNAL OF THE AMERICAN WATER RESOURCES ASSOCIATION JAWRA3

STATISTICAL EVALUATION OF VARIABLES AFFECTING OCCURRENCE OF HYDROCARBONS IN AQUIFERS USED FOR PUBLIC SUPPLY, CALIFORNIA

from the 2000s. For NWIS data, the sample date hav-ing the most complete data was selected for eachwell. The NWIS data included 1,105 wells sampledbetween August 1984 and June 2010 (median ofMarch 2001); >70% of the data was from the 2000s.

Study-specific reporting levels (SRLs) for individ-ual VOCs were developed by the GAMA program onthe basis of evaluation of blank samples (Fram et al.,2012). These SRLs for VOCs were used in this studyto exclude GAMA results with a reasonable likelihoodof being false positives. GAMA SRLs were applied toNWIS data to maintain consistency in the analysis ofGAMA and NWIS data. Application of GAMA SRLsresulted in analytical detections at low concentrationsof toluene, ethylbenzene, o-xylene, m- and p-xylene,and 1,2,4-trimethylbenzene being treated as non-detections in 290 GAMA and 155 NWIS samples (seeSupporting Information); these constituents weredetected in blanks at concentrations and frequenciesthat overlapped with environmental samples (Framet al., 2012). Hydrocarbons other than the five listedabove were not detected in GAMA blank samples andthe interpretation of detections of these constituentswas not affected by application of SRLs. In 25 GAMAand 51 NWIS samples, data for the five constituentswere not screened to SRLs because of independent evi-dence that detections of these compounds were likelyto be environmental rather than sampling artifacts.The evidence included: detection of one or more co-occurring hydrocarbons that were not censored basedon detections in blanks, historical detections of hydro-carbons in that well in the CDPH database, or concen-trations of the five constituents above SRLs.

Because the CDPH data are collected for regula-tory purposes at thousands of wells by different indi-viduals and laboratories, there is not an opportunityto review the quality of these data as rigorously asthe GAMA data. Following CDPH recommendations,a single detection of a constituent is not considered torepresent an actual occurrence until the detection isverified by subsequent sample results. For all CDPHwells with hydrocarbon detections in the most recentsample, the historical data were evaluated for previ-ous detections. If there were previous detections, themost recent result was used to represent that well.Single detections of a single constituent with no otherhistorical detections were treated as nondetections inthe most recent CDPH data to avoid false positives;results from wells with only a single analysis wereretained in the CDPH data analyzed. The screeningprocedures resulted in apparent detections beingtreated as nondetections in 55 samples; 29 of thesesamples had single detections of toluene only in themost recent sample. This screening was done toremove detections from the data that were not likelyto represent aquifer conditions.

Explanatory Factors

Potential explanatory variables were organized intothree conceptual groups: source, transport, and recep-tor variables. Source variables include direct or proxymeasures of the density of sources of fuel constituentsat or near the land surface within areas surroundingwells or the proximity of these sources to wells. Trans-port variables include factors that affect the hydrogeo-logic conditions between the recharge area and asampled well, including geochemical conditions,groundwater age, groundwater pumping and recharge,geology, density of public-supply wells, soil properties,and depth to water. Receptor variables are characteris-tics of the sampled well (receptor) itself, including wellconstruction and well type.

Table 2 lists 19 continuous potential explanatoryvariables discussed in this study. Data for many ofthe explanatory variables have been used in previousstudies in California (Johnson and Belitz, 2009; Jur-gens et al., 2010; Wright and Belitz, 2010; Fram andBelitz, 2011; Landon et al., 2011). Leaking Under-ground Fuel Tank (LUFT) density and proximity andpopulation density were selected because they havebeen found to be important in previous studies ofVOCs (Moran et al., 2005). Relations to land use wereassessed using data in 30-m cells interpreted fromearly 1990s satellite imagery and aerial photography(Nakagaki et al., 2007) and subsequent calculationsof the percent of the total land area within 500 m ofeach well in selected categories including urban, com-mercial, and high-intensity residential land use(Johnson and Belitz, 2009). The commercial land-usecategory represents commercial/industrial/transporta-tion areas consisting of roads, railroads, and highlydeveloped areas (constructed materials account for>80% of area) not classified as high-intensity residen-tial (apartment complexes and row houses). Land-usedata from the 1990s were used rather than morerecent data because water entering the perforationsof wells is more likely to be influenced by land usetwo or more decades ago than current land use; the1990s land-use data represented the earliest compre-hensive land-use data available across the entirestate. A Geographic Information System was used toattribute GAMA, CDPH, and NWIS wells with ancil-lary spatial data. Additional continuous and categori-cal variable data compiled for this study that did notindicate consistent relations with hydrocarbon distri-bution are described in the Supporting Information.

GAMA data were classified on the basis of ground-water age, well depth, reduction-oxidation (redox) con-ditions, and proximity to oil and gas fields; detectionfrequencies were compared between the resulting databins. For tritium (3H), a tracer of groundwaterrecharged by precipitation during the period of above

JAWRA JOURNAL OF THE AMERICAN WATER RESOURCES ASSOCIATION4

LANDON, BURTON, DAVIS, BELITZ, AND JOHNSON

TABLE 2. Spearman’s q Correlation Test Results Between Primary Continuous Explanatory Variablesand Concentrations of MTBE, Benzene, and Sum of Hydrocarbons.

Variable, Units (reference)

GAMAGAMAOld gw

GAMARecent gw CDPH CDPH NWIS

MTBE Benz HC Benz Benz MTBE Benz MTBE Benz HC

SourceAnthropogenic

6LUFT density, no./km2 +1 ns ns ns ns ns ns +1 +1 +16Distance to nearest LUFT, m �1 ns ns ns ns �1 ns �1 �1 �1

7Population density, people/km2 +1 ns ns ns ns ns ns +1 +1 +18Commercial land use, % +1 +1 ns ns +1 ns ns +1 +1 +18High-intensity residential land use, % +1 ns +1 ns ns ns ns +1 +1 +18Urban land use, % within 500 m of well +1 ns ns ns ns ns ns +1 +1 +1

Geogenic9Distance to nearest oil or gas field, m ns �2 �2 ns ns ns ns � ns ns

Receptor10,11,12Depth to top of perforations, m bls �4 ns +2 ns ns ID ID ns ns ns10,11,12Well depth, m bls ns +2 ns ns ns ID ID �4 ns ns

TransportAge/depth indicators10,11,12pH, standard pH units �4 ns +2 ns ns �4 +2 �4 ns ns10,12Tritium, pCi/l +4 �2 �2 ns ns ID ID ns ns ns10Carbon-14, pmc +4 �2 �2 ns ns ID ID ID ID IDAquifer stress/aquifer characteristics11CDPH well density, wells/km2 +5 ns ns ns ns ns ns +5 ns ns13Soil permeability, cm/h +5 ns ns ns ns ns ns ns ns ns14Groundwater recharge minus pumping �5 ns ns ns ns ns ns �5 ns nsOxidation-reduction conditions10,12Dissolved oxygen, mg/l ns �3 �3 �3 ns ID ID �3 ns ns10,11,12Nitrate, mg/l as N ns �3 �3 �3 ns +4 �3 ns �3 �3

10,11,12Manganese, lg/l ns +3 +3 +3 ns +3 ns +3 +3 ns10,11,12Iron, lg/l ns +3 +3 +3 +3 ns ns ns ns ns

Notes: +, significant positive correlation; �, significant inverse correlation; ns, correlation not significant; ID, insufficient data to evaluatecorrelation; significant correlations were determined on the basis of p values (significant level of the Spearman’s test) less than the thresholdvalue (a) of 0.05; GAMA, Groundwater Ambient Monitoring and Assessment; CDPH, California Department of Public Health; NWIS,National Water Information System; MTBE, methyl tert-butyl ether; Benz, benzene concentrations; HC, sum-of-hydrocarbon concentrations;gw, groundwater; km2, square kilometers; no., number; m, meters; %, percent; LUFT, Leaky Underground Fuel Tank; pCi/l, picocuries perliter; pmc, percent modern carbon; cm, centimeters; mg/l, milligrams per liter; lg/l, micrograms per liter; bls, below land surface; N,nitrogen.

Group of relations:1Correlations with anthropogenic source variables consistent with increasing concentrations with increasing source density and proximity.2Correlations of benzene with variables indicating increasing concentrations with proximity to oil and gas fields and greater depth and age.3Correlations with redox conditions indicating increasing concentrations with reducing groundwater.4Correlations of MTBE with variables indicating increasing concentrations with shallower and younger groundwater.5Correlations of MTBE with variables indicating increasing concentrations with greater aquifer stress and permeability.Sources:6California State Water Resources Control Board, 2003, Geographic Environmental Information Management System GeoTracker (GEIMS)Leaking Underground Fuel/Storage Tank database (LUFT) [digital data]: data received via email, currently now available via web, accessedon November 18, 2011 at URL: http://geotracker.waterboards.ca.gov/data_download.asp. Sacramento, California, California State WaterResources Control Board, Division of Water Quality.

7California Department of Finance, 2000, Demographic Research Unit, 1970-1980-1990-2000 Comparability File, digital dataset of USCENSUS data normalized to the 1990 tract boundaries for the State of California, available for download, accessed on April 19, 2004 atURL: http://www.dof.ca.gov/html/demograp/scdc%5Fproducts.htm.

8Nakagaki et al. (2007); (Johnson and Belitz, 2009).9California Division of Oil, Gas, and Geothermal Resources, 2005, Oil and Gas Fields. California Department of Conservation, Sacramento, CA.10U.S. Geological Survey, California Groundwater Ambient Monitoring and Assessment Program, Priority Basin Project, Publications, DataSeries Reports. Accessed on December 8, 2011 at: http://ca.water.usgs.gov/gama/publications.html.

11California Department of Public Health, Drinking Water Program. Accessed on September 7, 2011, http://www.cdph.ca.gov/programs/Pages/DWP.aspx.

12U.S. Geological Survey, National Water Information System. Accessed on December 8, 2010: http://waterdata.usgs.gov/ca/nwis/.13USDA (1994).14California Department of Water Resources (2005).

JOURNAL OF THE AMERICAN WATER RESOURCES ASSOCIATION JAWRA5

STATISTICAL EVALUATION OF VARIABLES AFFECTING OCCURRENCE OF HYDROCARBONS IN AQUIFERS USED FOR PUBLIC SUPPLY, CALIFORNIA

ground nuclear testing that began after 1950, a thresh-old activity of 1 pCi/l was selected for distinguishingwater that was likely to have recharged aquifers priorto 1950 (old groundwater, <1 pCi/l) from post-1950water (recent-age groundwater, ≥1 pCi/l) (Michel,1989; Plummer et al., 1993; Michel and Schroeder,1994; Clark and Fritz, 1997; Manning et al., 2005).Well depths were classified into three bins, <30, 30-180, and >180 m. These depth values were selected toseparate upper and lower parts of the depth distribu-tions from most of the interquartile range (Figure S1)while using consistent threshold values for all threedatasets. Three depth bins were used because of theobservation of a bimodal relation of hydrocarbon occur-rence to well depth; these three bins were further sim-plified to a binary classification of ≥180 m (deep) and<180 m (shallow and intermediate) for use with othervariables. Redox conditions were classified based onmeasured concentrations of dissolved oxygen (O2),nitrate (NO3

�), manganese (Mn+2), iron (Fe+2), andsulfate (McMahon and Chapelle, 2008; Jurgens et al.,2009). The redox categories of oxic, suboxic, mixed, andanoxic (McMahon and Chapelle, 2008) were furtherlumped into a binary redox classification of oxic, andreducing, the latter of which included suboxic, mixed,and anoxic categories, for testing the variation in fre-quency of detection of contaminants (Table S2). Basedon analysis of the relation between hydrocarbon occur-rence and minimum distance to the nearest oil and gasfield, a threshold value of 5 km was identified for bin-ary classification. Density of CDPH wells (wells/km2)is considered an indicator of aquifer utilization for sup-ply (aquifer stress), although groundwater withdraw-als for irrigation dwarf those for public supply inCalifornia. However, the locations and withdrawalsfrom irrigation wells throughout California are notavailable. Additional proxy variables for assessingrelations to aquifer stress were areal mean ground-water recharge, pumping, and recharge minus pump-ing rates (cm/yr), which were computed fromestimated annual volumes and areas reported by Cali-fornia Department of Water Resources (2005) for 56planning areas covering the state.

Statistical Methods

Nonparametric rank-based methods were used forstatistical analysis because these techniques are gen-erally not affected by outliers and do not require thatthe data follow a normal distribution (Helsel andHirsch, 2002). The significance level (p) used forhypothesis testing was compared to a threshold value(a) of 5% (a = 0.05) to evaluate whether the relationwas statistically significant (p < a). Correlations wereinvestigated using Spearman’s method to calculate

the rank-order correlation coefficient (q) betweenconcentrations and continuous explanatory variables.The Wilcoxon rank-sum test was used to evaluate thedifferences between two groups (Helsel and Hirsch,2002). The Kruskal-Wallis test was used to testdifferences among more than two groups (Conover,1980). A Pearson’s chi-square (v2) contingency tabletest was used to evaluate whether two categoricalvariables are related. All statistical analysis was doneusing S-PLUS for Windows, version 8.1, ProfessionalEdition (TIBCO Software Inc., Somerville, Massachu-setts).

Principal components analysis (PCA) was also usedto distinguish the characteristics of wells in whichbenzene and other hydrocarbons were detected. PCAis a multivariate statistical technique that reducesthe number of variables into a smaller number ofcomponents (Kshirsagar, 1972; Gnanadesikan, 1997;Manly, 2005). The transformed variables, known asprincipal components, are uncorrelated linear combi-nations of the original variables. PCA provides anobjective way to combine variables to account for asmuch of the variability in the data as concisely aspossible. The magnitude and direction (positive ornegative) of the contribution of each variable to theprincipal component score is described by an eigen-vector (Manly, 2005). PCA was applied iterativelybased on conceptual models of important variablesdeveloped from initial nonparametric statistical anal-ysis. Although both correlation and covariance matri-ces were calculated, PCA results based on correlationmatrices are presented in this study. PCA was con-ducted using GAMA data that excluded samples withmissing data for one or more variables; the data ana-lyzed included 1,473 samples for benzene and sum ofhydrocarbons and 1,461 samples for MTBE. PCA wasconducted using both ranked and log-transformeddata; results for these two datasets were similar andonly results for log-transformed data are reportedin this study. PCA was performed using the Multi-Variate Statistical Package (MVSP) v. 3.13p byKovach Computing Services (Kovach, 2007).

RESULTS

Both the GAMA and CDPH data primarily repre-sent depth zones of aquifers used for public supplyacross California, whereas the NWIS data generallyrepresent shallower groundwater in more urban set-tings. The median well depth was 120 m for the GAMAdata, 140 m for CDPH data, and 92 m for NWIS data(Figure S1). Public-supply wells are primarily perfo-rated at depths sufficient to protect the wells from

JAWRA JOURNAL OF THE AMERICAN WATER RESOURCES ASSOCIATION6

LANDON, BURTON, DAVIS, BELITZ, AND JOHNSON

surface contamination and are located in or near thecommunities that they serve. The GAMA data are pri-marily collected from randomly selected wells withincells spatially distributed across entire groundwaterbasins (Belitz et al., 2010) and are not always locatedin or near communities, as CDPH wells generally are.Studies in which VOC data were collected and enteredinto the NWIS database by USGS have primarily beenconducted in urban areas using samples from a varietyof monitoring, domestic, and supply wells to meet theobjectives of local studies. These differences in the spa-tial distribution of wells result in significant differ-ences in well depth and urban land-use percentage(within 500 m of well) between the datasets. Welldepths were significantly greater for CDPH data thanfor GAMA data (Wilcoxon test statistic Z = 36.3, p <0.0001) and well depths were significantly greater forGAMA data than NWIS data (Wilcoxon Z = 6.3, p <0.0001). Similarly, urban land-use percentage was sig-nificantly greater for CDPH data than for GAMA data(Wilcoxon test statistic Z = 16.7, p < 0.0001) and urbanland use was significantly greater for NWIS than CDPHdata (Wilcoxon Z = 12.8, p < 0.0001) (Figure S2).

Occurrence of Oxygenates and Hydrocarbons

Based on the GAMA and CDPH data, detectionfrequencies and concentrations of gasoline oxygenatesand hydrocarbons were generally low in public-supplyaquifers of California. Spatial weighting of the datacould result in slight changes in the detection fre-quencies reported below, but raw frequency estimatesrarely fall outside the 90% confidence intervals ofspatially weighted estimates (Belitz et al., 2010).

MTBE was detected in about 5.8% of the GAMAdata (Table 1) and was the most frequently detectedoxygenate. Four other oxygenates were detected in theGAMA data at frequencies ≤0.5% (Table S1). One ormore oxygenates were detected in 6.2% of GAMA sam-ples. The median detected concentration for MTBEwas 0.10 lg/l (Table 1). Oxygenates were detectedabove a regulatory threshold in one GAMA sample, afrequency of 0.05%; this well had MTBE above theCalifornia Maximum Contaminant Level (MCL) of13 lg/l. MTBE detections are distributed across sev-eral agricultural and natural areas of California aswell as densely urbanized areas of Southern California

(a) (b)

FIGURE 1. Maps Showing Groundwater Ambient Monitoring and Assessment Data Overlainon Land Use for (a) Methyl Tert-Butyl Ether (MTBE) and (b) Benzene.

JOURNAL OF THE AMERICAN WATER RESOURCES ASSOCIATION JAWRA7

STATISTICAL EVALUATION OF VARIABLES AFFECTING OCCURRENCE OF HYDROCARBONS IN AQUIFERS USED FOR PUBLIC SUPPLY, CALIFORNIA

(Figure 1a, Table S3). MTBE rarely co-occurred withother oxygenates; of the GAMA samples in whichMTBE was detected, only 2.6% (3 of 114) had detec-tions of other oxygenates.

Benzene was detected in about 1.7% of the GAMAdata (Table 1) and was the most frequently detectedhydrocarbon. Fifteen other hydrocarbons weredetected in the GAMA data at frequencies ≤1.0%(Table S1). One or more hydrocarbons were detected inabout 2.7% of GAMA samples. The median detectedconcentration was 0.024 lg/l for benzene and 0.09 lg/lfor sum of hydrocarbons (Table 1). Hydrocarbons weredetected above a regulatory threshold in three GAMAsamples, a frequency of detection of 0.15%; thesewells had benzene concentrations above the CaliforniaMCL of 1.0 lg/l and were located in the southwesternportion of the Central Valley (Figure 1b, Table S1).Benzene detections were widely distributed acrossCalifornia, but were especially prevalent in parts ofthe Central Valley and Southern California (Table S3).

CDPH MDLs were 2-100 (median 20) times higherthan the GAMA long-term method detection levels(LT-MDLs); consequently, detection frequencies forthe CDPH data were lower than for the GAMA data.However, when GAMA data were censored to the mostcommon MDLs of the CDPH data, detection frequen-cies of hydrocarbons for the GAMA data were gener-ally similar to those in the CDPH data (Tables 1 andS1). Detected concentrations in the CDPH data werehigher than in the GAMA data, reflecting higherMDLs and sample numbers about 10 times larger forCDPH, but the frequency of detections above regula-tory thresholds were similar for both datasets. In theCDPH data, four hydrocarbons or gasoline oxygenateswere detected above regulatory thresholds (detectionfrequency above threshold): benzene (0.09%), toluene(0.01%), MTBE (0.14%), and tert-butyl alcohol (0.05%)(Table S1). CDPH wells with MTBE concentrationsabove the MCL were widely distributed across Califor-nia (Figure S3a). CDPH wells with benzene concentra-tions above the MCL were primarily located in inlandsouthern California, the southern Central Valley, andCentral California coastal areas (Figure S3b), a similardistribution to GAMA data.

Hydrocarbons and oxygenates rarely co-occur(Figure 2). In the CDPH data, hydrocarbons andoxygenates were not detected together (0% frequencyof co-occurrence in 10,792 samples). At or above theCDPH MDLs of 0.5 lg/l, hydrocarbons and gasolineoxygenates evidently do not co-occur. At LT-MDLs forthe USGS analyses, which are one to two orders ofmagnitude lower than CDPH MDLs, hydrocarbonsand oxygenates sometimes co-occur. For GAMA data,there was co-occurrence in 9 of 166 (5.4%) wells withdetection of either class of compounds; for NWISdata, there was co-occurrence in 16 of 141 (11.3%)

wells with detection of either class of compounds. Thefrequency of co-occurrence may increase at lower con-centrations as a result of mixing of water of differentages and source areas in supply wells, producingrelatively low concentrations. Co-occurrence alsoincreases at shallower aquifer depths, as the NWISdata represent shallower groundwater than theGAMA and CDPH data. The rare co-occurrence ofhydrocarbons and oxygenates implies that these con-stituents can be treated as being independent of eachother for the purposes of statistical analysis.

Relation to Explanatory Variables

In GAMA and CDPH data, MTBE and benzene wererelated to different variables or had opposing relationsto variables (Table 2), indicating fundamental differ-ences in the sources and processes controlling occur-rence of these compounds at aquifer depths used forpublic supply. Consequently, relations of MTBE andbenzene to explanatory factors are discussed sequen-tially below. Results of different nonparametric tests,correlations, and tests for differences in values ofexplanatory variables between wells with and withoutdetections, generally indicated the same relations ofbenzene or MTBE prevalence to explanatory variables(Tables S4 and S5). Although statistical correlationbetween variables does not indicate causality, analysisof correlations with many variables provides insightconcerning controlling variables in a regional context.

MTBE

Higher MTBE concentrations were correlated withincreasing density and proximity of sources, youngergroundwater age (higher values of 3H and 14C),increasing density of public-supply wells, higher soil

FIGURE 2. Detection Frequency of Hydrocarbons, Oxygenates,and Co-occurrence of Hydrocarbons and Oxygenates for theGroundwater Ambient Monitoring and Assessment (GAMA), Cali-fornia Department of Public Health (CDPH), and National WaterInformation System (NWIS) Data.

JAWRA JOURNAL OF THE AMERICAN WATER RESOURCES ASSOCIATION8

LANDON, BURTON, DAVIS, BELITZ, AND JOHNSON

permeability, more negative water balance (recharge-pumping), and shallower perforations in receptorwells. Thus, MTBE is related to source, transport,and receptor variables, although relations with sourceand transport variables are strongest. These relationsare consistent with expectations that MTBE isrelated to anthropogenic sources. Relations were mostevident for GAMA and NWIS data; fewer relationswere evident from analysis of CDPH data.

For the GAMA data, MTBE concentrations were sig-nificantly correlated with many source variables,including increasing concentrations with increasingurban, high-intensity residential, and commercial landuse, density of LUFTs, and population density, anddecreasing distance to the nearest LUFT (Table 2).Nearly all source terms are correlated with each other,including positive correlations of LUFT density,population density, and percentages of urban, high-intensity residential, and commercial land use (TableS6); distance to the nearest LUFT was negativelycorrelated with the source density variables. MTBEwas also significantly related to several transport vari-ables, including positive correlations with 3H and 14C(decreasing age), density of CDPH wells, and soil per-meability, and a negative correlation with rechargeminus groundwater pumping (Table 2). Among recep-tor variables, MTBE concentrations were negativelycorrelated with depth to the top of perforations(Table 2). MTBE concentrations were also negativelycorrelated with pH, which may serve as a surrogate fordepth; pH usually increases from recharge at the watertable to greater depths in a groundwater flow systemwith increasing residence time (Koh et al., 2006; Jur-gens et al., 2008). In the GAMA data, pH is negativelycorrelated with 3H and 14C, and positively correlatedwith well depth (Table S6).

For the CDPH data, higher MTBE concentrationswere significantly correlated with increasing sourceproximity (but not density as for GAMA) and indica-tors of younger groundwater age (decreasing pH,increasing NO3

�). There were fewer relations evidentfrom analysis of CDPH than GAMA data. In both theCDPH and GAMA data, there was a negative correla-tion with pH and with distance from the well to thenearest LUFT (Table 2). As with the GAMA data,increasing pH is a surrogate for increasing groundwa-ter age and depth in the CDPH data; pH is available atmany more CDPH wells than depth or age. Two othervariables, NO3

� and Mn+2, were positively correlatedwith MTBE in the CDPH data but not in the GAMAdata. The correlation of increasing MTBE with increas-ing NO3

� in the CDPH data may reflect that relativelyyounger, shallower groundwater with elevated NO3

�

concentrations from surficial sources is also vulnerableto MTBE. The positive correlation of MTBE and Mn+2

in the CDPH data is consistent with the fact that

MTBE is more readily biodegraded under oxic thananoxic conditions (Landmeyer et al., 2001; McMahonand Chapelle, 2008). It is not possible to evaluate therelations of MTBE to receptor variables given limitedCDPH well construction data. Other source and trans-port variables with significant relations for GAMAdata were not significant for CDPH data (Table 2). Therelative scarcity of the relations for the CDPH datacompared to the GAMA data probably reflects that theCDPH data have the following: (1) higher MDLs, (2)greater spatial clustering, and (3) more limited ancil-lary variable information.

For NWIS data, relations of MTBE to explanatoryvariables followed expected patterns, with concentra-tions increasing with increasing source proximity anddensity, decreasing depth, increasing well density,more negative water balance, and more reducinggroundwater conditions (Table 2). For NWIS data,MTBE was correlated with the same source densityand proximity variables as for the GAMA data. MTBEwas negatively correlated with well depth for theNWIS data, similar to the negative correlation withdepth to the top of the perforations for the GAMA data(Table 2). MTBE and pH were negatively correlatedfor both the NWIS and GAMA data, reflecting that pHis positively correlated with depth (Tables 2 and S6).MTBE was positively correlated with density of CDPHwells and negatively correlated with recharge minusgroundwater pumping. These relations of MTBE withindicators of aquifer stress were significant for boththe GAMA and NWIS data. The correlation of increas-ing MTBE with these variables is reasonable asgreater pumping and aquifer development would beexpected to increase groundwater velocities and MTBEmovement to wells. However, density of CDPH wells(positive) and recharge minus groundwater pumping(negative) also were correlated with LUFT density andother source density variables (Table S6). Therefore, itis possible that the relations of MTBE concentrationsto well density and negative water balance may resultfrom cross-correlations between source and aquiferstress variables. MTBE was negatively correlated withO2, and positively correlated with Mn+2, indicatinghigher MTBE concentrations under anoxic conditions.McMahon and Chapelle (2008) also noted higher detec-tion frequencies of MTBE under anoxic than oxic con-ditions. These patterns are consistent with theobservation that MTBE is more readily biodegradedunder oxic than anoxic conditions (Landmeyer et al.,2001). It is possible that the association of MTBE andanoxic conditions could be influenced by the source ofMTBE, gasoline spills, which often result in anoxicwater near the source as a result of abundant organiccarbon from gasoline as an electron donor, and trans-port of both MTBE and anoxic water to downgradientwells. However, plumes near gasoline spills are

JOURNAL OF THE AMERICAN WATER RESOURCES ASSOCIATION JAWRA9

STATISTICAL EVALUATION OF VARIABLES AFFECTING OCCURRENCE OF HYDROCARBONS IN AQUIFERS USED FOR PUBLIC SUPPLY, CALIFORNIA

typically localized near the spill site (Rice et al., 1995)and are unlikely to commonly affect ambient ground-water redox conditions in downgradient productionwells represented by the GAMA, CDPH, and mostNWIS wells.

Benzene and Other Hydrocarbons

The explanatory variables for benzene generally dif-fered from those for MTBE, which include variablesrelated to anthropogenic sources. Benzene was corre-lated with fewer explanatory variables (Table 2) andexhibited more complex relations with controlling vari-ables than MTBE. Concentrations of benzene and sumof hydrocarbons were correlated with each other(q = 0.78, p < 0.0001) and were correlated with thesame explanatory variables in most cases (Table 2).Detections of benzene and other hydrocarbons in Cali-fornia groundwater appear to be related to both geo-genic sources and anthropogenic surficial fuel sources.

The presence of benzene and other hydrocarbons inold (3H < 1 pCi/l), deep (>180 m), and brackish (spe-cific conductance >1,600 lS/cm) groundwater in aqui-fers used for public supply in California was attributedto geogenic sources (Landon and Belitz, 2012). Evi-dence that many of the detections of benzene and otherhydrocarbons in groundwater are related to geogenicrather than anthropogenic sources includes: higherconcentrations and detection frequencies of benzeneand sum of hydrocarbons with increasing well depth,groundwater age, and proximity to oil and gas fields;and geochemical evidence, including salinity and chlo-ride/iodide ratios, that old groundwater samples withdetections of benzene and other hydrocarbons haveinteracted with oilfield brines. Detection frequencies ofbenzene were highest (20%) in samples with brackish,old, and reducing groundwater in proximity (<5 km) tooil or gas fields, and in nonbrackish, old, deep, andreducing groundwater <5 km from oil or gas fields(15.4%); benzene detected in these settings is likely tobe geogenic and accounts for about 45% of detections ofbenzene in the same GAMA data analyzed in thisstudy (Landon and Belitz, 2012). Of the remainingbenzene detections in the GAMA data, those in recent-age groundwater (3H ≥ 1 pCi/l) at shallow to interme-diate depths (<180 m) account for about 27% ofdetections; anthropogenic sources may explain manyof these detections (Landon and Belitz, 2012). Theremaining detections (28%) occur in groundwater withmixed age and depth characteristics and may resultfrom geogenic, anthropogenic, or a mixture of sources.The results below emphasize analysis of relations ofbenzene to explanatory variables in recent-age ground-water, which includes samples representing a mixtureof groundwater ages.

GAMA Data. Benzene concentrations and detec-tion frequencies were strongly correlated with lowerconcentrations of O2 and NO3

� and higher concentra-tions of Mn+2 and Fe+2 (Tables 2 and S7). Samplesclassified as reducing (including redox categories ofsuboxic, mixed, and anoxic, Table S2) had a signifi-cantly higher (v2 = 19.1, p < 0.0001) benzene detec-tion frequency (4.1%) than samples classified as oxic(0.9%) (Figure 3). Benzene and sum-of-hydrocarbonconcentrations and detection frequencies were higherin reducing than oxic groundwater in both old(3H < 1 pCi/l) and recent-age (3H ≥ 1 pCi/l) ground-water (Tables 2, S4, and S7). These results indicatethat redox conditions influence hydrocarbon persis-tence across a range of groundwater residence times.The relation to redox conditions is expected becausehydrocarbons readily biodegrade under oxic or aero-bic conditions (Howard et al., 1991; Suarez and Rifai,1999); biodegradation rates under reducing or anaero-bic conditions are substantially lower than thoseunder oxic conditions (Kauffman and Chapelle, 2010).

Benzene detections exhibited a bimodal relationwith respect to well depth, with significantly higherdetection frequencies for wells <30 m (3.13%) and>180 m (3.92%) than for wells with intermediatedepths (30-180 m) (1.1%) (v2 = 13.7, p = 0.0011) (Lan-don and Belitz, 2012). These bimodal relations areconsistent with two sources of hydrocarbons, deepand shallow, to groundwater.

In recent-age groundwater samples, benzene andhydrocarbon concentrations and (or) detection frequen-cies are significantly correlated with commercial landuse. Among GAMA samples of recent-age groundwa-ter, benzene concentrations were positively correlatedwith commercial land-use percentage (Table 2); inaddition, the detection frequency for benzene (3.2%) inwells with commercial land use >10% was higher(v2 = 9.3, p = 0.0023) than the detection frequency(0.7%) in wells with commercial land use <10%

0%

1%

2%

3%

4%

5%

Reducing Oxic

Benz

ene

dete

con

freq

uenc

y n=1,285n= 467

FIGURE 3. Detection Frequency of Benzene inGroundwater Ambient Monitoring and Assessment Data

Categorized by Redox (Table S2).

JAWRA JOURNAL OF THE AMERICAN WATER RESOURCES ASSOCIATION10

LANDON, BURTON, DAVIS, BELITZ, AND JOHNSON

(Figure 4a). Among GAMA recent-age samples thatwere also reducing, the benzene detection frequencyfor wells with commercial land use >10% was 7.1%,higher (v2 = 7.8, p = 0.0053) than for wells with com-mercial land use <10% (1.1%) (Figure 4b). Similarly,detection frequencies of sum of hydrocarbons werehigher for wells with commercial land use >10% andreducing conditions than other recent-age groundwa-ter samples (Table S7). Concentrations of sum ofhydrocarbons in GAMA recent-age samples were alsopositively correlated with another land-use surrogate,high-intensity residential (Table S4), which was corre-lated with other indicators of urban land use, includingcommercial land use (Table S6).

The results of PCA were consistent with the non-parametric statistical analysis of correlations anddetection frequencies described above and indicatedthat benzene detections occurred in two differentgroups of wells with contrasting characteristics, (1)wells that sample old, deep, and (or) brackish ground-water and (2) wells that sample recent, shallow tointermediate depth, and nonbrackish groundwater(Figure 5). For benzene, the first two PCA axesaccounted for 46% of the variance in the data. The firstprincipal axis separated samples by location and depthvariables (well depth, distance to oil and gas fields, andcommercial land-use percentage). The second principalcomponent is mostly influenced by groundwater ageand water quality variables (tritium, dissolved oxygen,and specific conductance).

a

b

0

5

10

> 10% < 10%

Benz

ene

dete

con

freq

uenc

y, %

Commercial land use

0

5

10

> 10% < 10%

Benz

ene

dete

con

freq

uenc

y, %

Commercial land use

n= 42 n=191

n=251 n=1,085

FIGURE 4. Detection Frequency of Benzene in GroundwaterAmbient Monitoring and Assessment Samples Categorized by

Commercial Land-Use Percentage for (a) Recent-Age Groundwaterand (b) Recent-Age and Reducing Groundwater.

Axi

s 2

Axis 1

-0.03

-0.06

-0.09

-0.12

0.03

0.06

0.09

0.12

0.15

-0.03-0.06-0.09 81.051.021.090.060.030.021.0-WellDepth

Lcommpct

DO

Trit

SPC

OnG_DIST_m

Vector scaling: 0.26

brackish

recent, deep

recent, shallow

no benzene

old, deep

old, shallow

FIGURE 5. Principal Components Analysis Plot of Groundwater Ambient Monitoring and Assessment Data Showing Characteristics ofWater Samples from Wells with Benzene Detections, Organized into Two Major Groups: Old, Deep, and Brackish Samples (purple polygon)and Recent, Shallow to Intermediate Depth Samples (orange polygon). Variables contributing to principal component axes 1 and 2 are shown:tritium concentration (Trit), well depth, salinity (SPC), commercial land use (commpct), dissolved oxygen (DO), and distance to nearest oiland gas field (OnG_DIST_m).

JOURNAL OF THE AMERICAN WATER RESOURCES ASSOCIATION JAWRA11

STATISTICAL EVALUATION OF VARIABLES AFFECTING OCCURRENCE OF HYDROCARBONS IN AQUIFERS USED FOR PUBLIC SUPPLY, CALIFORNIA

CDPH Data. The higher MDLs of the CDPH dataand generally low frequency of occurrence of hydro-carbons limit the use of these data to discern rela-tions to explanatory variables. In the CDPH data,there were fewer statistically significant relationsbetween benzene concentrations and explanatoryvariables than for the GAMA or NWIS data, andsome important variables such as age-tracer concen-trations and perforation depth were absent or infre-quent in the CDPH data. However, Landon andBelitz (2012) noted that the positive correlation ofbenzene to pH, a surrogate for depth, and higherdetection frequency of benzene in wells <5 km fromoil and gas fields than in wells ≥5 km away was con-sistent with evidence from the GAMA data that manydetections in public-supply wells are related to geo-genic sources of petroleum. The negative correlationof benzene and sum-of-hydrocarbon concentrationswith NO3

� and positive correlation of sum of hydro-carbons with Mn+2 and Fe+2 (Tables 2 and S4) areconsistent with relations indicating persistence ofhydrocarbons in reducing groundwater noted in theGAMA data.

NWIS Data. The NWIS data generally exhibiteddifferent relations between hydrocarbons and explan-atory variables than those indicated by the GAMAand CDPH data, likely as a consequence of represent-ing shallower and more urban settings. A commonal-ity with the GAMA and CDPH data is that analysisof the NWIS data indicated that benzene concentra-tions were greater in more reducing groundwater,with a negative correlation with NO3

� and a positivecorrelation with Mn+2 (Table 2).

In contrast to the GAMA data, benzene and sum-of-hydrocarbon concentrations for the NWIS datawere not significantly correlated with depth, ground-water age (3H and 14C), or proximity to oil and gasfields (Table 2). Although the NWIS data had a bimo-dal relation for benzene and well depth similar to theGAMA data, the higher detection frequencies for welldepth <30 m (4.0%) and >180 m (3.8%) than for wells30-180 m deep (2.6%) were not significantly different.Thus, in contrast to results of analysis of GAMA andCDPH data, analysis of the NWIS data did not yieldstatistical evidence for benzene occurring in somedeep groundwater as a result of geogenic sources.This difference is likely to reflect the more limitedspatial distribution of the NWIS data (Figure S4b).

For NWIS data, benzene was related to more sourcevariables than in the GAMA and CDPH data. Benzeneconcentrations and detection frequencies were signifi-cantly correlated with source density and proximity(Table 2). The relations to increasing density and prox-imity of fuel sources suggest that anthropogenicsources explain most detections of benzene in the

NWIS data. Commercial land use, the single measureof source density correlated with benzene concentra-tions in GAMA recent-age groundwater data, also wassignificantly correlated with benzene and sum ofhydrocarbons for NWIS data (Table 2). In the NWISdata, benzene detection frequency was significantlyhigher (contingency table v2 = 8.4, p = 0.0038) amongwells with commercial land use >10% (5.1%) thanamong wells with commercial land use <10% (1.8%).The larger number of relations for the NWIS data com-pared to the GAMA data probably reflects that NWISdata have similar LRLs to GAMA but is more clusteredin shallow and urban settings.

Analysis of the NWIS data indicated that MTBEand benzene and other hydrocarbons were related toa number of common variables, particularly sourceterm and redox variables (Tables 2 and S4). Althoughrelations of benzene and sum of hydrocarbons tosource variables were statistically significant for theNWIS data, greater absolute values of Spearman’s q,smaller p values, and visual plots of data indicatedthat MTBE concentrations had stronger relations tosource terms than benzene and sum of hydrocarbonsdid (Table S4). MTBE and benzene in the NWIS datawere generally related to different receptor andtransport variables, with the exception of a positivecorrelation of both constituents with Mn+2, consistentwith expected greater persistence of both compoundsunder more reducing conditions. In the NWIS data,higher concentrations of MTBE were negatively cor-related with well depth and groundwater rechargeminus pumping and positively correlated with CDPHwell density (Table 2), the latter two variables serv-ing as proxy indicators of increased aquifer stress. Incontrast, benzene and sum of hydrocarbons were notsignificantly related to receptor and transport vari-ables other than redox conditions.

The relations of MTBE and benzene and otherhydrocarbons to common source term variables observedfor the NWIS data contrast with the results of analysesof the GAMA and CDPH data, which indicated com-mon relations to only one source variable, commercialland use. This contrast likely reflects differences in theareal and vertical distribution of wells between the dif-ferent datasets (Figures 1, S3, and S4). The NWISdata are primarily restricted to Southern Californiabasins, Bay area, Central Valley, and a few samples inthe Southern California Desert. Thus, the NWIS dataare more focused on urban and agricultural areas morelikely to be affected by anthropogenic activity than themore widely distributed GAMA and CDPH data.

Although co-occurrence of hydrocarbons and MTBEwas more common in the NWIS data than in theGAMA and CDPH data, co-occurrence was still rela-tively infrequent, occurring in about 1.5% of NWISsamples (Figure 2) and about 11.4% (16 of 141) of

JAWRA JOURNAL OF THE AMERICAN WATER RESOURCES ASSOCIATION12

LANDON, BURTON, DAVIS, BELITZ, AND JOHNSON

samples that had detections of either hydrocarbonsand (or) MTBE. Thus, correlations with the samesource variables in the NWIS data do not indicate co-occurrence of hydrocarbons and MTBE. The wells inwhich co-occurrence did occur did not have signifi-cantly different characteristics from wells with detec-tions of either hydrocarbons or MTBE without co-occurrence.

DISCUSSION AND CONCLUSIONS

The infrequent co-occurrence of oxygenates withbenzene and other hydrocarbons in samples fromaquifer depths used for public supply (GAMA andCDPH), and the divergent relations of these constitu-ents to explanatory variables, provide insight regard-ing the differing variables that control theiroccurrence. MTBE followed a pattern expected for aconstituent derived from anthropogenic surficialsources, with larger concentrations associated with:proximity and density of sources; aquifers with youn-ger-age groundwater, increasing public-supply welldensity, more negative water balance (recharge-pumping), and larger soil permeability; and receptorwells with shallower perforations. Benzene and otherhydrocarbons followed a pattern expected for constit-uents derived from both anthropogenic and geogenicsources, with concentrations positively correlatedwith commercial land use, but also with depth,groundwater age, and proximity to oil and gas fields.Larger concentrations and detection frequencies ofbenzene were also associated with reducing condi-tions regardless of groundwater age.

Differences in the spatial distribution of MTBEand hydrocarbons indicate that these compoundshave different transport characteristics and sources.MTBE has lower biodegradation rates under oxic andanoxic conditions than hydrocarbons (Landmeyeret al., 2001), and is expected to move faster to supplywells when spilled fuel products containing both setsof constituents reach groundwater (Kauffman andChapelle, 2010). Detection frequencies of MTBE inCalifornia — about five times greater than benzeneand two times greater than all hydrocarbons com-bined (Table 1) — are consistent with results fromacross the U.S. (Zogorski et al., 2006) and reflect thediffering physiochemical properties of these constitu-ents. MTBE also has a different use history andsource distribution than gasoline hydrocarbons (Mo-ran et al., 2005). MTBE was a fuel additive in useduring the early 1990s through the mid 2000s (Shihet al., 2004; Zogorski et al., 2006), a much shorterperiod of use than gasoline, which has been used for

more than 100 years. MTBE in groundwater mayalso result from nonpoint sources such as partitioningfrom the atmosphere to recharge water (Squillaceet al., 1998; Baehr et al., 1999; Belitz et al., 2004; Zo-gorski et al., 2006). For example, in confined aquifersof the coastal plain of Southern California, whichreceive little vertical recharge from the overlyinglandscape, the distribution of MTBE is predominantlycontrolled by lateral movement of groundwater alongflow paths away from artificial recharge areas towardpumping centers (Shelton et al., 2001; Dawson et al.,2003); the artificial recharge primarily comprised sur-face water that contained MTBE that may have beenderived from atmospheric or other surficial sources.

In groundwater shallower than that typically usedfor public supply, represented by the NWIS data, com-mon relations to source variables and more frequentco-occurrence (about 11%) imply that the conditionspermitting co-occurrence of MTBE and hydrocarbonsbecome more prevalent closer to sources of these com-pounds near the land surface. In the NWIS data,MTBE and benzene were both related to sourcestrength and proximity variables and were both moreprevalent under reducing than oxic conditions.

There are multiple lines of evidence indicating thatdetections of hydrocarbons in some groundwater arerelated to geogenic (geologic reservoirs of petroleum)rather than anthropogenic sources (Landon andBelitz, 2012). Concentrations and detection frequen-cies of benzene and sum of hydrocarbons are posi-tively correlated with well depth, groundwater age,and other depth/age proxies such as pH; also, concen-trations and detection frequencies in old groundwaterare greater in proximity to oil or gas fields (Landonand Belitz, 2012). Geochemical evidence, includingsalinity and Cl/I ratios, indicate that old groundwaterwith detections of benzene and other hydrocarbonshave interacted with oilfield brines (Landon andBelitz, 2012). In addition, PCA indicates that wellswith benzene and other hydrocarbon detectionsoccurred in two different groups of wells with con-trasting characteristics: one group of wells that sam-ple old, deep, and (or) brackish groundwater andanother group of wells that sample recent, shallow tointermediate, and nonbrackish groundwater. Theseresults support the nonparametric statistical analysisindicating both geogenic and anthropogenic sourcesof hydrocarbons to groundwater used for public sup-ply in California.

The higher detection frequencies of benzene andother hydrocarbons in reducing groundwater, signifi-cant negative correlations with O2 and NO3

�, and sig-nificant positive correlations with Mn+2 and Fe+2 in allevaluated data (GAMA, CDPH, NWIS) are consistentwith laboratory and field studies indicating that bio-degradation rates of hydrocarbons are typically lower

JOURNAL OF THE AMERICAN WATER RESOURCES ASSOCIATION JAWRA13

STATISTICAL EVALUATION OF VARIABLES AFFECTING OCCURRENCE OF HYDROCARBONS IN AQUIFERS USED FOR PUBLIC SUPPLY, CALIFORNIA

under anoxic than oxic conditions (Howard et al.,1991; Kazumi et al., 1997; Weiner and Lovley, 1998;Suarez and Rifai, 1999; USEPA, 1999; Kauffman andChapelle, 2010). Within a plume of dissolved hydrocar-bons from spilled petroleum, more reducing conditionsdevelop in the interior of the plume as a result oforganic carbon serving as electron donors for redoxreactions (Baedecker et al., 1993; Chapelle et al.,2003). The resulting redox zonation can vary oversmall spatial scales (McNab, 1999; Bekins et al., 2001).Previous studies noting that plumes from fuel spillsare generally of limited extent (Rice et al., 1995) andinfrequently reach public-supply wells in California(Hadley and Armstrong, 1991) have attributed theseresults to the effects of microbially facilitated biodegra-dation. In this study, the measured redox parametersare representative of the ambient conditions of thegroundwater system, often at the depths used for pub-lic supply, rather than the zone of altered redox thatoften occurs around spill sites.

In recent-age groundwater (3H ≥ 1 pCi/l), benzeneand hydrocarbon concentrations and (or) detectionfrequencies are positively correlated with commercialland use. Wells having a combination of >10% com-mercial land use within 500 m of the well and reduc-ing groundwater conditions had the highest detectionfrequency of benzene (7.1%) and hydrocarbons(11.9%) among any combination of variables inGAMA recent-age samples. In the NWIS data, repre-senting shallower and more urban groundwater thanthe GAMA data, benzene and hydrocarbon concentra-tions were more strongly correlated with commercialland use and were related to five additional sourcestrength variables, including LUFT density, distanceto the nearest LUFT, population density, percenturban land use, and percent high-intensity residen-tial land use. These variables are proxy measures forfuel source strength and proximity. In shallowgroundwater (NWIS data), relations of benzene andother hydrocarbons to source terms were as strong asrelations to redox indicators; in deeper groundwater(GAMA and CDPH), hydrocarbons were morestrongly correlated with redox indicators than withsource terms.

Differences and similarities of results of analysesof the three datasets provided greater insight regard-ing explanatory variables than would have been pos-sible with analysis of any single dataset. Analysis ofGAMA data, with lower reporting limits and moreextensive ancillary, age, and chemistry data thanavailable for the CDPH data, provided greater insightregarding variables explaining occurrence of hydro-carbons and MTBE in public-supply wells (Table 2).Relations to explanatory variables identified fromanalysis of the much larger CDPH data were consis-tent with analysis of the GAMA data, but fewer rela-

tions to source, transport, and receptor variableswere evident (Table 2). Comparison of analyses ofGAMA and NWIS data indicated that effects ofanthropogenic sources of hydrocarbons are more evi-dent in shallow groundwater than in deeper ground-water, and that deeper groundwater is affected moreby geogenic sources of hydrocarbons than by anthro-pogenic sources. For MTBE, the similarity of rela-tions identified using the GAMA and NWIS datareflects that MTBE followed patterns expected for amobile constituent with surficial anthropogenicsources; in both datasets, larger concentrations wereassociated with greater source proximity and density,shallower depths, younger groundwater age, andincreasing aquifer pumping. For benzene, with amore complex distribution of geogenic and anthropo-genic sources, differences in explanatory relationsidentified between the GAMA and NWIS data reflectthat the GAMA data are from wells that are deeperand less commonly in urban areas; in the GAMAdata, relations to both geogenic and anthropogenicsources were detected, whereas the shallower, moreurban NWIS data were primarily suited for detectingrelations with anthropogenic sources.

Occurrence of benzene in public-supply wells inCalifornia is infrequent (1.7%) and these occurrencesare related more to natural variables — ambientreducing groundwater conditions and proximity to ge-ogenic sources — rather than anthropogenic vari-ables. It is possible that anthropogenic source control/remediation and groundwater protection activitiescontribute to these relations. The limited impact offuel spills on aquifers used for public-water supply inCalifornia is notable given the widespread distribu-tion of petroleum storage and distribution systemsacross the landscape; the number of leaking gaso-line tanks in California has been estimated toexceed 11,000 or 6.5% of underground tanks in thestate (Hadley and Armstrong, 1991). Hadley andArmstrong (1991) concluded that the most likelyexplanation of infrequent occurrence of benzene inpublic-supply wells was biodegradation of spilledpetroleum near its source. In a simulation study ofVOC transport in selected principal aquifers in theU.S., Kauffman and Chapelle (2010) determinedthat simulated aquifers in the western U.S. hadlow vulnerability to VOCs, particularly petroleumhydrocarbons such as benzene and toluene because oflow-recharge rates, long travel times, and thepredominantly oxic conditions, which facilitated bio-degradation of the hydrocarbons before reaching pub-lic-supply wells. The results of this study areconsistent with these interpretations, but clarify thegeogenic and anthropogenic variables that explainthe occurrence of hydrocarbons in wells in those caseswhere they occur.

JAWRA JOURNAL OF THE AMERICAN WATER RESOURCES ASSOCIATION14

LANDON, BURTON, DAVIS, BELITZ, AND JOHNSON

SUPPORTING INFORMATION

Additional Supporting Information may be foundin the online version of this article, including: supple-mental methods description, description of occurrenceof selected constituents, relations to secondaryexplanatory variables, and supporting figures andtables.Figure S1. Well-depth distribution for CDPH,

GAMA, and NWIS data.Figure S2. Urban land use distribution around

CDPH, GAMA, and NWIS wells.Figure S3. (a) Map showing results of analyses for

MTBE in CDPH data overlain on land use. (b) Mapshowing results of analyses for benzene in CDPHdata overlain on land use.Figure S4. (a) Map showing results of analyses for

MTBE in NWIS data overlain on land use. (b) Mapshowing results of analyses for benzene in NWIS dataoverlain on land use.Figure S5. (a) Map showing estimated areal average

groundwater pumping for 2,000 (cm) in 56 planningareas (California Department of Water Resources,2005) and CDPH wells with detections of any hydro-carbon (blue) or no detections (black). (b) Map showingestimated areal average groundwater recharge for2,000 (cm) in 56 planning areas (California Depart-ment of Water Resources, 2005) and CDPH wells withdetections of any hydrocarbon (blue) or no detections(black). (c) Map showing estimated areal averagegroundwater recharge minus pumping for 2,000 (cm)in 56 planning areas (California Department of WaterResources, 2005) and CDPH wells with detections ofany hydrocarbon (blue) or no detections (black).Table S1. Hydrocarbons and gasoline oxygenates

included in study, comparative thresholds, reportinginformation, and detection frequencies relative toselected thresholds.Table S2. Criteria and threshold concentrations for

identifying redox processes in groundwater (Jurgenset al., 2009; McMahon and Chapelle, 2008) and distri-bution of wells by redox category and redox process.Table S3. Detections of MTBE and benzene by

lumped hydrogeologic province in GAMA and CDPHdata.Table S4. Spearman’s q and significance levels for

correlation tests between continuous explanatory fac-tors and benzene, sum of hydrocarbons and MTBE.Table S5. Results of Wilcoxon nonparametric tests

for differences in explanatory variables between wellswith and without detections of MTBE and benzene,GAMA data.Table S6. Spearman’s correlation coefficients (q)

between explanatory variables for GAMA data.Table S7. Detection frequencies of benzene and

summed hydrocarbons in GAMA data by explanatoryvariable categories.

Please note: Neither AWRA nor Wiley Blackwell isresponsible for the content or functionality of anysupporting information supplied by the authors. Anyqueries (other than missing content) should be direc-ted to the corresponding author for the article.

ACKNOWLEDGMENTS

This study was funded by the U.S. Environmental ProtectionAgency, Office of Research and Development under interagencyagreement DW-14-95779601-0. This study was coordinated withthe California GAMA program, funded by State bonds administeredby the California State Water Resources Control Board. We thankthe well owners who allowed the USGS to collect samples and per-sonnel who collected and managed the data. The use of brandnames in the manuscript is for identification purposes only anddoes not imply endorsement by the U.S. Government.

LITERATURE CITED

Baedecker, M.J., I.M. Cozzarelli, R.P. Eganhouse, D.I. Siegel, andP.C. Bennett, 1993. Crude Oil in a Shallow Sand and GravelAquifer. III. Biogeochemical Reactions and Mass BalanceModeling in Anoxic Groundwater. Applied Geochemistry 8:569-586.

Baehr, A.L., P.E. Stackelberg, and R.J. Baker, 1999. Evaluation ofthe Atmosphere as a Source of Volatile Organic Compounds inShallow Groundwater. Water Resources Research 35(1):127-136.

Bekins, B.A., I.M. Cozzarelli, E.M. Godsy, E. Warren, H.I. Essaid,and M.E. Tuccillo, 2001. Progression of Natural AttenuationProcesses at a Crude Oil Spill Site: II. Controls on Spatial Dis-tribution of Microbial Populations. Journal of ContaminantHydrology 53:387-406.

Belitz, K., N.M. Dubrovsky, K.R. Burow, B.C. Jurgens, and T.D.Johnson, 2003. Framework for a Ground-Water Quality Moni-toring and Assessment Program for California. U.S. GeologicalSurvey Water-Resources Investigations Report 03-4166, 78 pp.

Belitz, K., S.N. Hamlin, C.A. Burton, R. Kent, R.G. Fay, and T.D.Johnson, 2004. Water Quality in the Santa Ana Basin Califor-nia, 1999-2001. U.S. Geological Survey Circular 1238.

Belitz, K., B.C. Jurgens, M.K. Landon, M.S. Fram, and T.D. John-son, 2010. Estimation of Aquifer Scale Proportion Using EqualArea Grids: Assessment of Regional Scale Groundwater Quality.Water Resources Research 46:W11550, doi: 10.1029/2010WR009321.

Bender, D.A., J.S. Zogorski, M.J. Halde, and B.L. Rowe, 1999.Selection Procedure and Salient Information for VolatileOrganic Compounds Emphasized in the National Water-QualityAssessment Program. U.S. Geological Survey Open-File Report99-182, 32 pp.

Bennett, G.L., M.S. Fram, and K. Belitz, 2011. Status and Under-standing of Groundwater Quality in the Southern, Middle, andNorthern Sacramento Valley Study Units, 2005–08—CaliforniaGAMA Priority Basin Project. U.S. Geological Survey ScientificInvestigations Report 2011-5002, 86 pp.

Burton, C.A., J. Montrella, M.K. Landon, and K. Belitz, 2011.Status and Understanding of Groundwater Quality in the SantaClara River Valley, 2007—California GAMA Priority BasinProject. U.S. Geological Survey Scientific Investigations Report2011-5052, 86 pp.

JOURNAL OF THE AMERICAN WATER RESOURCES ASSOCIATION JAWRA15

STATISTICAL EVALUATION OF VARIABLES AFFECTING OCCURRENCE OF HYDROCARBONS IN AQUIFERS USED FOR PUBLIC SUPPLY, CALIFORNIA

California Department of Water Resources, 2005. California WaterPlan Update 2005. A Framework for Action, Bulletin 160-05,December 2005. http://www.waterplan.water.ca.gov/previous/cwpu2005/index.cfm, accessed November 2011.

Chapelle, F.H., M.A. Widdowson, J.S. Brauner, E. Mendez, and C.C.Casey, 2003. Methodology for Estimating Times of RemediationAssociated with Monitored Natural Attenuation. U.S. GeologicalSurvey Water Resources Investigations Report 03-4057, 51 pp.

Clark, I.D. and P. Fritz, 1997. Environmental Isotopes in Hydro-geology. Lewis Publishers, New York City, New York, 328 pp.

Conover, W.J., 1980. Practical Nonparametric Statistics (SecondEdition). Wiley, New York City, New York.

Corseuil, H.X., A.L. Monier, M. Fernandes, M.R. Schneider, C.C.Nunes, M. Rosario, and P.J.J. Alvarez, 2011. BTEX PlumeDynamics Following an Ethanol Blend Release: GeochemicalFootprint and Thermodynamic Constraints on Natural Attenua-tion. Environmental Science and Technology 45:3422-3429, doi:10.1021/es104055q.

Dawson, B.J., K. Belitz, M.T. Land, and W.R. Danskin, 2003. Sta-ble Isotopes and Volatile Organic Compounds Along SevenGround-Water Flow Paths in Convergent and Divergent FlowSystems, Southern California, 2000. U.S. Geological SurveyWater-Resources Investigations Report 03-4059.

Deeb, R.A., J.O. Sharp, A. Stocking, S. McDonald, K.A. West, M.Laugier, P.J.J. Alvarez, M.C. Kavanaugh, and L. Alvarez-Cohen, 2002. Impact of Ethanol on Benzene Plume Lengths:Microbial and Modeling Studies. Journal of EnvironmentalEngineering 128(9):868-875.

Densmore, J.N., K. Belitz, M.T. Wright, B.J. Dawson, T.D. John-son, and G.B. Hudson, 2004. Evaluation of Volatile OrganicCompounds in Two Mojave Desert Basins – Mojave River andAntelope Valley – San Bernardino, Los Angeles, and Kern Coun-ties, California, June – October 2002. U.S. Geological SurveyScientific Investigations Report 2004-5248, 40 pp.

Fram, M.S. and K. Belitz, 2011. Probability of Detecting Perchlo-rate Under Natural Conditions in Deep Groundwater in Califor-nia and the Southwestern United States. EnvironmentalScience and Technology 45:1271-1277, doi: 10.1021/es103103p.

Fram, M.S., L.D. Olsen, and K. Belitz, 2012. Review of VolatileOrganic Compound Blank Data Collected for the CaliforniaGroundwater Ambient Monitoring and Assessment (GAMA)Program, May 2004 – October 2009. U.S. Geological Survey Sci-entific Investigations Report 2012-5139.

Gnanadesikan, R., 1997. Methods for Statistical Data Analysis ofMultivariate Observations (Second Edition). Wiley, New YorkCity, New York, 384 pp.

Gomez, D.E. and P.J.J. Alvarez, 2010. Comparing the Effects ofVarious Fuel Alcohols on the Natural Attenuation of BenzenePlumes Using a General Substrate Interaction Model. Journalof Contaminant Hydrology 113:66-76.

Grady, S.J., 2003. National Survey of Methyl Tert-Butyl Ether andOther Volatile Organic Compounds in Drinking-Water Sources:Results of the Random Survey. U.S. Geological Survey Water-Resources Investigations Report 02-4079, 85 pp.

Hadley, P.W. and R. Armstrong, 1991. Where’s the Benzene? –Examining California Ground-Water Quality Surveys. GroundWater 29(1):35-40.

Hamlin, S.N., K. Belitz, and T.D. Johnson, 2005. Occurrence andDistribution of Volatile Organic Compounds and Pesticides inGround Water in Relation to Hydrogeologic Characteristics andLand Use in the Santa Ana Basin, Southern California. U.S. Geo-logical Survey Scientific Investigations Report 2005-5032, 40 pp.

Helsel, D.R. and R.M. Hirsch, 2002. Statistical Methods in WaterResources. U.S. Geological Survey Techniques of Water-Resources Investigations, book 4, chap. A3, 510 pp. http://water.usgs.gov/pubs/twri/twri4a3/.

Howard, P.H., R.S. Boethling, W.F. Jarvis, W.M. Meylan, and E.M.Michalenko, 1991. Handbook of Environmental DegradationRates, 725. Lewis Publishers, Inc., Chelsea, Michigan.

Johnson, T.D. and K. Belitz, 2009. Assigning Land Use to SupplyWells for the Statistical Characterization of Regional Ground-water Quality: Correlating Urban Land Use and VOC Occur-rence. Journal of Hydrology 370:100-108.

Jurgens, B.C., K.R. Burow, B.A. Dalgish, and J.L. Shelton, 2008.Hydrogeology, Water Chemistry, and Factors Affecting theTransport of Contaminants in the Zone of Contribution of aPublic-Supply Well in Modesto, Eastern San Joaquin Valley, Cali-fornia. U.S. Geological Survey Scientific Investigations Report2008-5156, Reston, Virginia. http://pubs.usgs.gov/sir/2008/5156.

Jurgens, B.C., M.S. Fram, K. Belitz, K.R. Burow, and M.K. Lan-don, 2010. Effects of Ground-Water Development on Uranium:Central Valley California, USA. Ground Water 48(6):913-928,doi: 10.1111/j.1745-6584.2009.00635.x.

Jurgens, B.C., P.B. McMahon, F.H. Chapelle, and S.M. Eberts,2009. An Excel� Workbook for Identifying Redox Processes inGround Water. U.S. Geological Survey Open-File Report 2009-1004, 8 pp.

Kauffman, L.J. and F.H. Chapelle, 2010. Relative Vulnerability ofPublic Supply Wells to VOC Contamination in HydrologicallyDistinct Regional Aquifers. Ground Water Monitoring andRemediation 30(4):54-63.

Kazumi, J., M.E. Caldwell, J.M. Suflita, D.R. Lovley, and L.Y.Young, 1997. Anaerobic Degradation of Benzene in DiverseAnoxic Environments. Environmental Science and Technology31(3):813-818.

Koh, D.C., L.N. Plummer, D.K. Solomon, E. Busenberg, Y.J. Kim,and H.W. Chang, 2006. Application of Environmental Tracers toMixing, Evolution, and Nitrate Contamination of Ground Waterin Jeju Island, Korea. Journal of Hydrology 327(1-2):258-275.

Koterba, M.T., F.D. Wilde, and W.W. Lapham, 1995. Ground-WaterData-Collection Protocols and Procedures for the NationalWater-Quality Assessment Program—Collection and Documen-tation of Water-Quality Samples and Related Data. U.S. Geolog-ical Survey Open-File Report 95-399, 113 pp.

Kovach, W.L., 2007. MVSP – A Multivariate Statistical Package forWindows, Ver. 3.1. Kovach Computing Services, Wales, UnitedKingdom.

Kshirsagar, A.M., 1972. Multivariate Analysis. Dekker, New YorkCity, New York, 534 pp.

Kulongoski, J.T., K. Belitz, M.K. Landon, and C. Farrar, 2010. Sta-tus and Understanding of Groundwater Quality in the NorthSan Francisco Bay Groundwater Basins, 2004—CaliforniaGAMA Priority Basin Project. U.S. Geological Survey ScientificInvestigations Report 2010-5089, 88 pp.

Landmeyer, J.E., F.H. Chapelle, H.H. Herlong, and P.M. Bradley,2001. Methy Tert-Butyl Ether Biodegradation by IndigenousAquifer Microorganisms Under Natural and Artificial Oxic Con-ditions. Environmental Science and Technology 35(6):1118-1126.

Landon, M.K. and K. Belitz, 2012. Geogenic Sources of Benzene inAquifers Used for Public Supply, California. EnvironmentalScience and Technology 46:8689-8697.

Landon, M.K., K. Belitz, B.C. Jurgens, J.T. Kulongoski, and T.D.Johnson, 2010. Status and Understanding of GroundwaterQuality in the Central–Eastside San Joaquin Basin, 2006:California GAMA Priority Basin Project. U.S. Geological SurveyScientific Investigations Report 2009-5266, 97 pp.

Landon, M.K., C.T. Green, K. Belitz, M.J. Singleton, and B.K.Esser, 2011. Relations of Hydrogeologic Factors, GroundwaterReduction-Oxidation Conditions, and Temporal and Spatial Dis-tributions of Nitrate, Central-Eastside San Joaquin Valley, Cali-fornia. Hydrogeology Journal 19:1203-1224, doi: 10.1007/s10040-011-0750-1.

JAWRA JOURNAL OF THE AMERICAN WATER RESOURCES ASSOCIATION16

LANDON, BURTON, DAVIS, BELITZ, AND JOHNSON

Mackay, D.M., N.R. De Sieyes, M.D. Einarson, K.P. Feris, A.A.Pappas, I.A. Wood, L. Jacobson, L.G. Justice, M.N. Noske, K.M.Scow, and J.T. Wilson, 2006. Impact of Ethanol on the NaturalAttenuation of Benzene, Toluene, and o-Xylene in a NormallySulfate-Reducing Aquifer. Environmental Science and Technol-ogy 40(19):6123-6130.

Manly, B.F.J., 2005. Multivariate Statistical Methods: A Primer(Third Edition). Chapman and Hall, New York City, New York,214 pp.

Manning, A.H., D.K. Solomon, and S.A. Thiros, 2005. 3H/3He AgeData in Assessing the Susceptibility of Wells to Contamination.Ground Water 43(3):353-367.

McMahon, P.B. and F.H. Chapelle, 2008. Redox Processes andWater Quality of Selected Principal Aquifer Systems. GroundWater 46(2):259-271.

McNab, Jr., W.W., 1999. Comparisons of Geochemical Signatures ofBiotransformation of Fuel Hydrocarbons in Groundwater. Envi-ronmental Monitoring and Assessment 59:257-274.

Michel, R.L., 1989. Tritium Deposition in the Continental UnitedStates, 1953–83. U.S. Geological Survey Water-Resources Inves-tigations Report 89-4072, 46 pp.

Michel, R.L. and R. Schroeder, 1994. Use of Long-Term TritiumRecords from the Colorado River to Determine Timescales forHydrologic Processes Associated with Irrigation in the ImperialValley, California. Applied Geochemistry 9:387-401.

Molson, J.W., J. Barker, E.O. Frind, and M. Schirmer, 2002. Mod-eling the Impact of Ethanol on the Persistence of Benzene inGasoline-Contaminated Groundwater. Water ResourcesResearch 38(1):4-1-4-12.

Molson, J.W., J. Barker, and M. Schirmer, 2000. Modeling theImpact of Ethanol on the Persistence of BTEX Compounds inGasoline-Contaminated Groundwater. A Report prepared for theCalifornia MTBE Research Partnership. Center for Groundwa-ter Restoration and Protection, National Water Research Insti-tute, Fountain Valley, California.

Moran, M.J., J.S. Zogorski, and P.J. Squillace, 2005. MTBE andGasoline Hydrocarbons in Ground Water of the United States.Ground Water 43(4):615-627.

Nakagaki, N., C.V. Price, J.A. Falcone, K.J. Hitt, and B.C. Ruddy,2007. Enhanced National Land Cover Data 1992 (NLCDe 92).U.S. Geological Survey: Raster Digital Data. http://water.usgs.gov/lookup/getspatial?nlcde92, accessed September 2010.

Plummer, L.N., R.L. Michel, E.M. Thurman, and P.D. Glynn, 1993.Environmental Tracers for Age-Dating Young Groundwater. In:Regional Groundwater Quality, W.M. Alley (Editor). Van No-strand Reinhold, New York City, New York, pp. 255-294.

Rice, D.W., R.D. Grose, J.C. Michaelsen, B. Dooher, D.H. Mac-Queen, S.J. Cullen, W.E. Kastenburg, L.G. Everett, and M.A.Marino, 1995. California Leaking Underground Fuel Tank(LUFT) Historical Case Analyses. Lawrence Livermore NationalLaboratory submitted to the California State Water ResourcesControl Board, UCRL-AR-122207.

Ruiz-Aguilar, G.M.L., K. O’Reilly, and P.J.J. Alvarez, 2003. A Com-parison of Benzene and Toluene Plume Lengths for Sites Con-taminated with Regular vs. Ethanol-Amended Gasoline. GroundWater Monitoring & Remediation 23(1):48-53.

Shelton, J.L., K.R. Burow, K. Belitz, N.M. Dubrovsky, M. Land,and J.M. Gronberg, 2001. Low-Level Volatile Organic Com-pounds in Active Public Supply Wells as Ground-Water Tracersin the Los Angeles Physiographic Basin, California, 2000. U.S.Geological Survey Water-Resources Investigations Report01-4188, 29 pp.

Shih, T., Y. Rong, T. Harmon, and M. Suffet, 2004. Evaluation ofthe Impact of Fuel Hydrocarbons and Oxygenates on Groundwa-ter Resources. Environmental Science and Technology 38(1):42-48.

Squillace, P.J. and M.J. Moran, 2007. Factors Associated withSources, Transport, and Fate of Volatile Organic Compoundsand Their Mixtures in Aquifers of the United States. Environ-mental Science and Technology 41(7):2123-2130.

Squillace, P.J., J.F. Pankow, N.E. Korte, and J.S. Zogorski, 1998.Environmental Behavior and Fate of Methyl Tert-Butyl Ether(MTBE). U.S. Geological Survey Fact Sheet FS-203-96.

Suarez, M.P. and H.S. Rifai, 1999. Biodegradation Rates for FuelHydrocarbons and Chlorinated Solvents in Groundwater. Bio-remediation Journal 3(4):337-362.

USDA (U.S. Department of Agriculture), 1994. STATSGO, StateSoil Survey Geographic Data Base ARC/INFO 7.0 Coverage,UNIX. USDA, Fort Worth, Texas [CD-ROM].

USEPA (U.S. Environmental Protection Agency), 1999. AnaerobicBiodegradation Rates of Organic Chemicals in Groundwater,184. U.S. Environmental Protection Agency Office of SolidWaste, Washington, D.C.

USGS (U.S. Geological Survey), variously dated. National FieldManual for the Collection of Water-Quality Data. U.S. Geologi-cal Survey Techniques of Water-Resources Investigations Book9, Ch. A1-A9. http://pubs.water.usgs.gov/twri9A, accessed June2009.

Weiner, J.M. and D.R. Lovley, 1998. Anaerobic Benzene Degrada-tion in Petroleum-Contaminated Aquifer Sediments After Inocu-lation with a Benzene-Oxidizing Enrichment. Applied andEnvironmental Microbiology 64(2):775-778.

Williams, P., L. Benton, J. Warmerdam, and P. Sheehan, 2002.Comparative Risk Analysis of Six Volatile Organic Compoundsin California Drinking Water. Environmental Science and Tech-nology 36(1):74-81.

Wright, M.T. and K. Belitz, 2010. Factors Controlling the RegionalDistribution of Vanadium in Groundwater. Ground Water48(4):515-525, doi: 10.1111/j.1745-6584.2009.00666.x.

Wright, M.T., K. Belitz, and T. Johnson, 2004. Assessing the Sus-ceptibility to Contamination of Two Aquifer Systems Used forPublic Water Supply in the Modesto and Fresno MetropolitanAreas, California, 2001 and 2002. U.S. Geological Survey Scien-tific Investigations Report 2004-5149, 35 pp.

Zogorski, J.S., J.M. Carter, T. Ivahnenko, W.W. Lapham, M.J. Mo-ran, B.L. Rowe, P.J. Squillace, and P.L. Toccalino, 2006. VolatileOrganic Compounds in the Nation’s Ground Water and Drink-ing-Water Supply Wells. U.S. Geological Survey Circular 1292,98 pp.

JOURNAL OF THE AMERICAN WATER RESOURCES ASSOCIATION JAWRA17

STATISTICAL EVALUATION OF VARIABLES AFFECTING OCCURRENCE OF HYDROCARBONS IN AQUIFERS USED FOR PUBLIC SUPPLY, CALIFORNIA