ARTICLE IN PRESS

Journal of Environmental Management 81 (2006) 405–419 www.elsevier.com/locate/jenvman

Sustainability of ground water quality considering land use changes and public health risks Navin K.C. Twarakavi, Jagath J. Kaluarachchi Civil and Environmental Engineering and Utah Water Research Laboratory, Utah State University, Logan, UT 84322-8200, USA Received 17 March 2005; received in revised form 14 November 2005; accepted 16 November 2005 Available online 4 April 2006

Abstract One of the major environmental issues of concern to policy-makers is the increased vulnerability of ground water quality (GWQ). Another issue of equal interest is the sustainability of natural resources for future generations. To understand the sustainability of the natural resources such as water in general, one needs to understand the impact of future land use changes on the natural resources. This work proposes a methodology to address sustainability of GWQ considering land use changes, aquifer vulnerability to multiple contaminants, and public health risks. The methodology was demonstrated for the Sumas-Blaine aquifer in Washington State. The land transformation model predicted that nearly 60 percent of the land use practices would change in the Sumas-Blaine Aquifer by the year 2015. The accuracy of the LTM model predictions increased to greater levels as the spatial resolution was decreased. Aquifer vulnerability analysis was performed for major contaminants using the binary logistic regression (LR) method. The LR model, along with the predicted future land use, was used to estimate the future GWQ using two indices—carcinogenic and non-carcinogenic ground water qualities. Sustainability of GWQ was then analyzed using the concept of ‘strong’ sustainability. The sustainability map of GWQ showed improvements in many areas where urbanization is expected to occur. The positive impact of urbanization on GWQ is an indication of the extensive damage caused by existing agricultural activities in the study area. r 2006 Elsevier Ltd. All rights reserved.

1. Introduction Ground water is one of the nation’s most important natural resources. Ground water provides drinking water for more than one-half of the nation’s population (Solley et al., 1998) and is the sole source of drinking water for many rural communities and some large cities. In 1990, ground water accounted for 39 percent of the public water supply for cities and towns and 96 percent for self-supplied systems for domestic use (Solley et al., 1998). Ground water is also the source of much of the water used for irrigation. Ground water is a major contributor to flow in many streams and rivers and has a strong influence on river and wetland habitats for plants and animals. Fresh ground water withdrawal in the US in 1995 was estimated to be approximately 77 billion gallons per day (Solley et al., 1998), which is about 8 percent of the estimated 1 trillion Corresponding author. Tel.: +1 435 797 3918; fax: +1 435 797 3663.

E-mail address: [email protected] (J.J. Kaluarachchi). 0301-4797/$ - see front matter r 2006 Elsevier Ltd. All rights reserved. doi:10.1016/j.jenvman.2005.11.008

gallons per day of natural recharge to the ground water system of the US (Nace, 1960). Even though ground water resources appear to be ample, spatial availability of ground water varies at large (Alley et al., 1999). Considering the variability of ground water resources, the question of availability for future generations needs to be carefully analyzed. The concept of analyzing the availability of a natural resource in the future is also referred to as ‘sustainability’ (Alley et al., 1999). The term ‘‘sustainability’’ has become a catch-all phrase that refers to almost everything. Pearce et al. (1989) have collected the various definitions of sustainability. The most commonly used definition for sustainability comes from the World Commission of Environment and Development of 1987 as ‘‘development that meets the needs without compromising the ability of future generations to meet their economic needs.’’ Sustainability, in short, is about being able to achieve intergenerational equity by developing policies that optimize the various factors involved such as land use changes (Pearce et al., 1989). Defining the

ARTICLE IN PRESS 406

N.K.C. Twarakavi, J.J. Kaluarachchi / Journal of Environmental Management 81 (2006) 405–419

sustainability of a natural resource such as ground water is often a difficult and elusive subject (Alley et al., 1999). Alley et al. (1999) discussed that ground water is neither a non-renewable resource, such as mineral or petroleum deposits, nor is it completely renewable in the same manner as solar energy. Also, defining sustainability of ground water needs to take into consideration the various attributes of ground water resources such as quality, quantity, and its impact on the dependent ecosystems. Therefore, the definition of ground water sustainability needs to be multipronged to accommodate the various attributes described earlier; and at the same time, the definition should be simple for the purpose of performing a study on large spatial domains. In this work, ground water sustainability is defined using the concept commonly referred to as ‘strong sustainability’ which recognizes the fact that natural resources cannot be substituted by manmade technologies. Following the definition from Alley et al. (1999), ground water sustainability is defined here as ‘‘use and development of ground water in such a way that results in no unacceptable damage in the future to the quantity, quality, and the dependent ecosystems.’’ Ground water sustainability-related studies in the past have concentrated on the quantity issues (Alley et al., 1999; US EPA, 2002). Quantity-related studies can be traced back to as early as the 1960s. Walton (1964) estimated the ‘sustainable yield’ of a confined aquifer for maintaining a healthy future supply of water. Bacchus (1998) studied the effects of pumping on seasonal fluctuations in ground water levels near wetlands. A few studies in the past concentrated on the development of methods for sustainable ground water by analyzing the influence of ground water pumping on the reduction of the ground water levels, lake levels, wetlands, and availability of drinking water sources (Downing, 1998; Sophocleous, 1998; Gelt et al., 1999). Morgan and Jones (1999) performed a numerical model analysis to understand the effects of ground water withdrawals on the discharge to streams and springs in small basins. Alley and Emery (1986) studied the effect of ground water pumping on streamflow and ground water levels for the Blue River Basin, a heavily irrigated area in southeastern Nebraska, and analyzed the complexities of prediction capabilities. McGuire and Sharpe (1997) analyzed the historical changes in water levels in the High Plains aquifers in parts of Colorado, Kansas, Nebraska, New Mexico, Oklahoma, South Dakota, Texas, and Wyoming, and found that land use changes tend to have significance on the sustainability of ground water quantity. It is clear that previous studies have not given ground water quality (GWQ) enough focus in sustainability analysis. The need to consider GWQ as a major factor influencing sustainability has been highlighted by Alley et al. (1999). Ground water quantity is not the only major factor one has to consider to address ground water sustainability. For example, consider the case of brines. Even though the quantity of brine-affected ground water may be large, these

sources are not included as available water resources as the utility of such poor quality water is negligible. Therefore, one needs to analyze the changes in GWQ, as well as ground water quantity, to understand true ground water sustainability. In other words, sustainability of ground water resources needs to consider at least two major attributes—ground water quantity sustainability and GWQ sustainability. GWQ sustainability has not been given its due attention in previous work. The goal of this study is to address sustainability of GWQ such that one could couple the proposed framework with existing methods for ground water quantity sustainability to better understand and manage ground water as a natural resource. One may consider GWQ to be sustainable if the regional-scale GWQ in the future is at least as good as the present state. Quantifying regional-scale GWQ is a vague concept because ‘GWQ’ at any location is a function of multiple contaminants and the public health risk posed by the contaminants. Some of the major contaminants found typically in ground water at regional-scales include agricultural chemicals, such as pesticides, herbicides, and nitrates, and heavy metals, such as arsenic, cadmium, mercury, lead, and zinc. A key consideration in understanding regional-scale GWQ is its vulnerability to multiple contaminants whose sources may be anthropogenic (located primarily at and near the land surface) as well as natural. A variety of techniques has been used in the past to estimate ground water vulnerability to contaminants. The nature of techniques for vulnerability estimation varies from simple overlay/index approaches to complex simulation models. The selection of the best method of vulnerability estimation is a function of data availability, the nature of the contaminant, types of sources, and the geochemical characteristics. Overlay/index methods involve combining the maps of various physiological attributes such as geology, depth to water table, aquifer properties, and net recharge and assigning weights to each property to obtain a final score. The methods are driven largely by data availability and expert judgment rather than the existing physical processes (NRC, 1993). Processbased simulation models require analytical and/or numerical solutions to the governing mathematical equations that represent coupled processes of contaminant transport (NRC, 1993). These methods are computationally costly and demand substantial data. On the other hand, statistical methods are flexible and better suited to accommodate uncertainty in data than the former methods (NRC, 1993). Based on previous studies (Eckhardt and Stackelberg, 1995; Nolan, 2001; Nolan et al., 2002; Rupert, 1998; Tesoriero and Voss, 1997; Tesoriero et al., 1998; Twarakavi and Kaluarachchi, 2005, 2006), logistic regression (LR) has been successfully used to assess vulnerability of shallow ground water to different regional-scale contaminants. The goal of this work is to develop a methodology to assess the sustainability of GWQ by considering land use

ARTICLE IN PRESS N.K.C. Twarakavi, J.J. Kaluarachchi / Journal of Environmental Management 81 (2006) 405–419

changes and public health risk due to major contaminants found in large watersheds. The future land use changes occurring in a watershed are a result of the policies adopted by land managers to accommodate the future population growth and economic activity (Pijanowski et al., 2000). These future land use changes may affect GWQ. For example, consider a scenario of a new highway proposed in a rural agricultural watershed. One would definitely expect agricultural activity to decrease in the future as a result of highway construction. This activity would result in a change in the spatial distribution of long-term GWQ. Many other factors, such as the one explained above, could influence future land use changes, which in turn indirectly affect GWQ. Therefore, a land manager in a large watershed indirectly controls ground water sustainability from a water quality standpoint. In this work, the results will demonstrate how to model future land use changes and assess the sustainability of GWQ. This paper only discusses the sustainability of GWQ. There are other important factors that influence the decisions of a land manager such as ground water quantity, atmospheric pollution, surface water pollution, soil waste generation, and ecological balance. In reality, the influence of the factors mentioned before can outweigh the implications from a GWQ sustainability analysis. Sustainability is a broad issue and the results and discussion from this work only hold true from the perspective of GWQ. 2. Methodology The objective of this work is to understand the sustainability of GWQ. The framework adopted here has three important phases. Fig. 1 shows the flowchart describing the methodology proposed to describe sustainability of GWQ. The first phase involves predicting the future spatial distribution of land use. The output from the first phase is maps of land use patterns at different time periods. A host of models is available to simulate future land use changes (Agarwal et al., 2002). A suite of complex factors, including population change, policy, economics, and natural and environmental characteristics, drives land use changes. In this work, the Land Transformation Model (LTM) of Pijanowski et al. (1997, 2000, 2002) is used to develop the spatial–temporal distribution characterizing land use changes. The LTM uses driving variables such as the population growth, transportation factors, proximity or density of important landscape features (such as rivers, lakes, and recreational sites) along with historical land use data to model future land use changes. The LTM has been used in previous environmental studies. Wayland et al. (2002) coupled the LTM with ground water models to examine the relationship between water quality and historical land use patterns. The inverse flow and solute modeling produced a reasonable distribution of ground water travel times across a watershed given the hydrology of the system. Boutt et al. (2001) used the LTM along with

407

Start

Phase1

Simulate future land use changes

Identify major contaminants Phase 2 Perform vulnerability analysis

Estimate carcinogenic and noncarcinogenic ground water quality Phase 3 Determine sustainability of ground water quality

End Fig. 1. Flowchart showing the methodology used to determine sustainability of ground water quality.

ground water models to examine potential relationships between the land use derived solutes and baseflow surface water quality. Wayland et al. (2002) used the LTM to study stream water quality with the aid of ground water transport models. Wayland et al. (2003) used the LTM along with ground water models to examine the usefulness of the synoptic sampling approach for identifying the relationship between complex land use configurations and stream water quality. The second phase of the framework involves quantifying the regional-scale GWQ at future time periods. GWQ may be viewed as a function of vulnerability of shallow ground water to multiple contaminants and the health risk posed to the residents by these contaminants. For example, high concentrations of non-toxic contaminants such as chlorides, does not necessarily contribute as much to poor water quality than the presence of toxic contaminants even in lower concentrations. Often, vulnerability of ground water to a contaminant or a class of contaminants has been used to predict the GWQ in an area (Eckhardt and Stackelberg, 1995; Nolan, 2001; Nolan et al., 2002; Rupert, 1998; Tesoriero et al., 1998; Twarakavi and Kaluarachchi, 2005, 2006). However, a thorough understanding of GWQ due to all available major contaminants is needed to address true vulnerability and also to extend the work to the concept of sustainability. The distribution of major contaminants in a watershed depends heavily on the land use distribution. Therefore, a methodology to predict vulnerability of ground water to major contaminants is needed to define ‘true’ GWQ. The ‘quality’ aspect is difficult to quantify as

ARTICLE IN PRESS 408

N.K.C. Twarakavi, J.J. Kaluarachchi / Journal of Environmental Management 81 (2006) 405–419

the risk posed by a contaminant is related to the type of toxic end-point of the chemical, exposure duration and pathways, daily intake, population characteristics, and a host of other factors. Human health risk is dependent on whether the contaminant is a carcinogen which has the potential adverse health effects of causing cancer during the lifetime of an individual. In the proposed methodology, vulnerability maps for major contaminants are grouped using the health risk-based indexing approach to quantify GWQ for each time period. The third phase of the methodology involves comparing GWQ between successive time periods to estimate the sustainability of GWQ of the study area. 2.1. Land transformation model Detailed descriptions of the LTM may be found elsewhere (Pijanowski et al., 2000, 2002). The core of the LTM is based on Geographic Information Systems (GIS) and Artificial Neural Networks (ANN) routines. ANN uses a machine learning approach to solve relationships between inputs and outputs. GIS is a powerful spatial data analysis tool that can be used to perform spatial–temporal modeling. The information derived from historical land use changes and the driving variables are used to forecast future land use changes. The LTM currently employs a multilayer perceptron (MLP) neural net topology with one or more hidden layers; each layer has at least the same number of nodes as the number of inputs. Inputs are derived from raster spatial layers of driving variables that are normalized to values between 0 and 1 by dividing by the maximum value in each data layer. Output is composed of locations that transitioned to urban areas (coded as a 1) and those that did not transition (coded as a 0). Several driving variables are considered in the LTM representing different areas such as policy, environment, and landscape features. The possible driving variables could be classified into the following categories: Transportation: The Euclidean distance from each cell in the region to the nearest highway and other roads was created as separate spatial layers. Another transportationrelated driving variable may be the density of roads within a defined radius and this may be used to represent the ease of acquiring urban services to a cell. Landscape features: The distance from lakes, shoreline, recreational spots/landmarks, and rivers was created as separate spatial layers. Pijanowski et al. (2000) have found that the landscape topography is an influential factor contributing towards residential use. Therefore, a ‘‘rolling hills’’ driving variable grid was created as a spatial layer that was calculated as the standard deviation of all cell elevations within a defined radius. Urban services: The distance of each cell from the nearest urban cell in the base land use layer was calculated. Cells containing larger values probably indicate a higher cost of connecting to urban services.

Exclusionary zones: Some areas may need to be excluded from development. These areas include local parks, national forests, public lands, existing urban land, water bodies and protected wetlands, existing transportation infrastructure, Indian reservations, and state forests and state parks. The driving variables listed above were used in the current analysis. The LTM follows four sequential steps: (a) processing/coding of data to create spatial layers of predictor variables; (b) applying spatial rules that relate predictor variables to land use transitions for each location in an area, the resultant layers contain input variable values in grid format; (c) integrating all input grids using one of three techniques; and (d) temporal scaling of the amount of transition in the study area to produce a time-series of possible future land uses. In Step (a), processing of spatial data and inputs are generated from a series of base layers that are stored and managed within GIS. These base layers represent land uses (such as agricultural parcels and urban areas) or features in the landscape (e.g., roads, rivers, and lakeshores). Grid cells are coded to represent predictors as either binary (presence ¼ 1 or absence ¼ 0) or continuous variables depending on the type of attribute. For Step (b), inputs are developed using a set of spatial transition rules that quantify the spatial effects that predictor cells have on land use transitions (see Pijanowski et al. (2000) for details). This study used four classes of transition rules: neighborhoods or densities, patch size, site-specific characteristics, and distance from the location of a predictor cell. Certain locations are coded so that these do not undergo transitions. This is necessary for areas within which development is prohibited, such as public lands. Cells were coded with ‘‘0’’ if a transition cannot occur; all other locations are assigned ‘‘1.’’ All such layers are then multiplied together to generate one single layer of ‘‘exclusionary zones.’’ In Step (c), ANN is used to develop the simulation model. The output from the ANN model is a map of ‘‘change likelihood values,’’ which specifies the relative likelihood of change for each cell based on ANN results. The results will be computed given the aggregate value of a cell for change derived from the total of predictor variable values. In Step (d) of temporal indexing, the amount of land that is expected to transition to urban over a given time period is determined using a ‘‘principle index driver’’ or PID (Pijanowski et al., 2000). The PID value is estimated based on the population growth over a time interval (i.e., number of years) for the region. The PID is used to determine the number of cells that need to transition to urban. Projections are made by selecting the appropriate number of cells in priority order, which is based on the change likelihood value for that cell (determined in Step (c)). 2.2. Ground water quality The methodology to quantify GWQ may be divided into the following steps: (a) identify the major contaminants in

ARTICLE IN PRESS N.K.C. Twarakavi, J.J. Kaluarachchi / Journal of Environmental Management 81 (2006) 405–419

the study area (carcinogens and non-carcinogens); (b) estimate the vulnerability of ground water to different contaminants; and (c) group the different vulnerability estimates of multiple contaminants based on toxicology endpoint. The methodology results in two ‘GWQ’ maps each indicating the carcinogenicity and non-carcinogenicity of ground water. In Step (a), water quality data of the study area are analyzed for the major contaminants detected. Typically, three classes of regional-scale contaminants are observed; pesticides, heavy metals, and nitrate. While pesticides have no background concentration, heavy metals and nitrate have background concentrations in ground water due to the presence of natural sources. Using water quality data, one may identify the major contaminants in ground water. The major contaminants are identified in the study area by analyzing the occurrence of contaminants for the range of detected concentrations, MCL, and the spatial distribution of occurrence. In Step (b), ground water vulnerability to major contaminants is estimated. One of the common statistical methods to estimate aquifer vulnerability is the technique of binary LR or commonly called logistic regression or LR (Eckhardt and Stackelberg, 1995; Nolan, 2001; Nolan et al., 2002; Rupert, 1998; Tesoriero and Voss, 1997). Nolan (2001) used LR to estimate aquifer vulnerability to nitrate contamination in the US. Binary LR has been used extensively in epidemiological studies and more recently, LR is becoming a common technique in environmental research (Hosmer and Lemeshow, 1989). Other regression techniques, such as classical linear regression, relate the response variables to the influencing variables. LR, however, relates the probability of the response to be less than a threshold value due to a set of influencing variables (Afifi and Clark, 1984; Helsel and Hirsch, 1992; Hosmer and Lemeshow, 1989; Kleinbaum, 1994); for example, the probability of arsenic to be less than the MCL for different land use types and/or soil classes. In a LR model, regression is linear between the natural logarithm of the odds ratio for the probability of response to be less than the threshold value and influencing variables. The odds ratio, O, is defined as O¼

p , 1
p

(1)

where p is the probability of the response to be less than a given threshold value. Binary LR states that the natural logarithm of the odds ratio, or logit, is linearly related to the influencing variables and can be written as Log ðO ¼ÞlogitðpÞ ¼ a þ bx;

(2)

where a is a constant, b is a vector of slope coefficients, and x is the vector of influencing variables. The threshold concentration used in this study is the MCL. Binary LR was used to develop vulnerability maps for each major contaminant. The maps indicate the probability of a

409

contaminant to be detected above the MCL in the study area. In Step (c), the vulnerability maps for the major contaminants developed in Step (b) are used to characterize the GWQ of the study area. As discussed earlier, the GWQ estimate for any location is best described by two separate quantities—carcinogenic GWQ and non-carcinogenic GWQ. While all major contaminants known to be carcinogenic are grouped to estimate the carcinogenic GWQ, the non-carcinogenic contaminants are grouped to estimate the non-carcinogenic GWQ. In case more than one contaminant is present in the carcinogen or non-carcinogen group, it is necessary to use a weighting procedure to estimate the GWQ. The weights would add more meaning if they can indicate the health risk posed by the representative contaminants. Reference dose (RfD) is commonly used in risk estimation where RfD is an estimate of the daily exposure to the human population that is likely to present no appreciable risk during the lifetime. RfD is expressed in units of milligrams per kilogram of body weight per day (mg/kg/day). RfD is useful as a reference point from which to gauge the potential effects of the chemical at other doses. Usually, doses less than the RfD are not likely to be associated with adverse health risks and are therefore, less likely to be of regulatory concern. As the frequency and/or magnitude of the exposures exceed the RfD, the probability of adverse effects to the public increases. One may use RfD along with the MCL for developing weights indicating the risk posed by a contaminant. One option is to use a weight of MCLi/ RfDi. The RfD is an estimate that indicates the daily intake concentration of a contaminant that could potentially cause adverse health effects. The MCL however, is estimated considering the health risks as well as the economics of cleaning contaminated water. Therefore, in the case of costly treatment technologies, one may expect MCL to be higher while the RfD is the same. Also, it is worth noting that the MCL is a federal standard that is enforced for drinking water supplies. For example, US EPA enforces drinking water suppliers to treat water such that arsenic concentrations are below the MCL of 10 mg/L. It is clear that the MCL considers the inherent risk associated with contaminated water. Therefore, a ratio of MCLi =RfDi gives an estimate of the intrinsic risk that may be associated with a contaminant in drinking water supply. At any location, if the pci is the probability for carcinogen i to exceed the MCLi, then the carcinogenic GWQ, GWQc, may be estimated as  ,X  n
n
X MCLi MCLi GWQc ¼ pci ; (3) RfDi RfDi i¼1 i¼1 where i ¼ 1,y, n are the major carcinogenic contaminants, RfDi is the RfD of contaminant i, and MCLi is the MCL for the contaminant i. Similarly, one may estimate the non-carcinogenic GWQ, GWQnc, at the same

ARTICLE IN PRESS N.K.C. Twarakavi, J.J. Kaluarachchi / Journal of Environmental Management 81 (2006) 405–419

410

location as GWQnc ¼

 k
X MCLi i¼1

RfDi

, pnci

 k
X MCLi ; RfDi i¼1

, Sc ¼ GWQtc
GWQtþDt c (4)

where i ¼ 1,y, k are the major non-carcinogenic contaminants, and pnci is the probability for non-carcinogen i to exceed the MCLi. It may be noted that higher the value of GWQc and GWQnc, poorer the GWQ. Therefore, one may estimate the carcinogenic and noncarcinogenic GWQ through using the vulnerability maps of the contaminants and risk data available in the US EPA IRIS network (http://www.epa.gov/iris/). A weight of MCLi =RfDi is assigned to contaminant i to indicate the relative toxicity of the contaminant. However, there are assumptions involved in using such a weight. Some of these assumptions are similarity in the toxic endpoints and exposure pathways. Inclusion of toxic endpoints and exposure pathways into the proposed methodology is difficult as it demands more information about the contaminants and the population characteristics. The vulnerability maps for carcinogens and non-carcinogens are grouped using Eqs. (3) and (4) to indicate GWQc and GWQnc of the study area. 2.3. Sustainability of GWQ The concept of ‘strong’ sustainability is considered where this definition is often used for understanding the intergenerational equity of natural resources. This definition is used to estimate the sustainability of GWQ between two time steps. The concept of ‘strong sustainability’ rejects the idea that existing and new infrastructure adequately compensates future generations for ecological and environmental losses. To express it mathematically, if Kn(t) is the natural capital of a resource at any time t, then for strong sustainability to exist: Knðt þ dtÞXKnðtÞ.

(5)

Estimation of sustainability of GWQ between current and a future generation is performed by comparing GWQ at each location spatially. In the previous section, estimation of carcinogenic and non-carcinogenic GWQ maps was discussed. The core of the procedure was the vulnerability analysis using the binary LR method. To estimate the carcinogenic and non-carcinogenic GWQ for the future generation, one could apply the binary LR model developed with the historical land use to the modeled future land use derived from the LTM. Such an analysis is valid as the binary LR model considers that the influencing variables such as land use and soil type, are independent of each other (Hosmer and Lemeshow, 1989; Afifi and Clark, 1984). Once the carcinogenic and non-carcinogenic GWQ maps are available for the year t and t+Dt, the difference between the GWQ values is estimated at each location as S nc ¼ GWQtnc
GWQtþDt nc ,

(6a)

(6b)

where Snc and Sc are the GWQ sustainability scores between two time steps for non-carcinogens and carcinogens, respectively; GWQtc is the carcinogenic GWQ estimate for year t; and GWQtnc is the non-carcinogenic GWQ estimate for year t. Sc and Snc give an estimate of the change in GWQ with respect to carcinogens and noncarcinogens. A positive value implies an improvement in future GWQ at a location while a negative value indicates that the future generation is not served properly from the viewpoint of GWQ. 3. Study area The study area is the Sumas-Blaine aquifer located in Whatcom County, Washington State (see Fig. 2). The approximate area of the Sumas-Blaine aquifer is about 150 square miles. Ground water from the aquifer is used mainly for residential and agricultural purposes and serves nearly 100,000 residents. 3.1. Land use pattern The study area falls within the Water Resources Inventory Area 1 (WRIA 1). The Sumas-Blaine aquifer is the most exploited aquifer in the watershed with a majority of the agricultural practices concentrated in the aquifer. Major crops grown in WRIA 1 include raspberries, strawberries, seed potatoes, blueberries, beans, corn, carrots, peas, and cauliflower. Due to the intensive agricultural activities in the westernmost part of WRIA 1, GWQ has degraded and concentrations of contaminants, such as nitrate, are increasing (Kaluarachchi et al., 2002). Apart from agriculture, other land uses that are present are forestry, industrial, commercial, residential, and dairy farming-related practices. Land use databases for the years 1975 and 1995 were obtained through personal communication with state and local agencies in WRIA 1. Land use data of 1975 were obtained from the National Land Class Database (NLCD) prepared by the USGS that has 21 different land use classes. For convenience of analysis, the land use distribution was grouped into the following classes: (1) agricultural, (2) forested/barren, (3) urban/built-up, and (4) others. All land use classes except others are self-explanatory. The land use class others includes wetlands, water bodies, rivers, and streams. 3.2. Regional hydrogeology The Sumas-Blaine aquifer consists largely of highly permeable Sumas outwash deposits. In general, these deposits overlay older and less permeable glacio-marine silt and bedrock (Cox and Kahle, 1999). The aquifer consists predominantly of sand and gravel glacial outwash deposits and alluvial gravel, sand, silt, and clay deposits of the Nooksack and Sumas Rivers. The aquifer thickness is

ARTICLE IN PRESS N.K.C. Twarakavi, J.J. Kaluarachchi / Journal of Environmental Management 81 (2006) 405–419

411

Fig. 2. Location of Water Resources Inventory Area 1 and the Sumas-Blaine aquifer of Washington State.

between 12 and 24 m in the western part of the study area, but it can exceed 60 m in the northeastern part (Kaluarachchi et al., 2002). Predicting the occurrence of heavy metals in ground water is complicated by the natural occurrence of mineral deposits in the subsurface. As detailed information on the occurrence of natural sources of heavy metals in the subsurface is not available, it has been assumed that mineral deposits are horizontally stratified within the aquifer. Natural occurrence of heavy metal sources, therefore, has been quantified by elevation and well depth of monitoring locations as the SumasBlaine aquifer is continuous. The average annual precipitation in the watershed is around 100 cm, of which only 30 percent usually falls from April through September. The growing season for most crops falls within this period (NRCS, 1992). The water table is shallow, typically less than 20 ft (Tooley and Erickson, 1996). Variables such as precipitation and temperature are not considered as influencing variables in the vulnerability model, as these variables are nearly uniform throughout the study area. The study area has a gentle topographic slope. Higher elevations are encountered in the eastern part of the aquifer. Elevation data were obtained from a digital elevation model (DEM) of the USGS. 3.3. Soil types Soil type data of the study area were obtained from the soil survey geographic (SSURGO) database, developed by the US Department of Agriculture (USDA). The database contained the spatial distributions of hydraulic conductivity, clay content, and soil hydrologic groups (SHGs). A SHG is a group of soils with similar runoff potential under similar storm and cover conditions. Soil properties that

influence runoff potential are those that influence the minimum rate of infiltration for a bare soil after prolonged wetting and when not frozen. These properties are depth to the seasonally high water table, soil composition, intake rate, and hydraulic conductivity after prolonged wetting, and depth to a low permeable layer. The infiltration rate increases from SHG A through D. 3.4. Water quality Water quality data were collected from a variety of sources using the databases of Whatcom County Department of Health, Washington State Department of Health, Washington State Department of Ecology, and the USGS. The wells sampled for each heavy metal are generally uniformly scattered throughout the study area. The sampling well density for each heavy metal is around 1 per square mile, which indicates a reasonable spatial sampling distribution. The depths of the sampling wells are mostly shallow with depths ranging from near the surface to 150 ft with a median depth of 40 ft. The shallow depths of the wells ensure sampling of the unconfined aquifer in the study area. Ground water sampling and testing were conducted according to the protocols established by the US Geological Survey’s National Water Quality Assessment Program (Koterba et al., 1995).The samples were collected over screened intervals with depth. Most of the well screens were usually less than 2 m in length. The well depth and heavy metal concentrations are screen-averaged estimates. To minimize the effects of altering the water chemistry of ground water samples, wells were purged for sufficient time before sampling. In the case of large diameter wells, samples were collected as close to the well screen as possible. Verification of data quality has been performed

ARTICLE IN PRESS 412

N.K.C. Twarakavi, J.J. Kaluarachchi / Journal of Environmental Management 81 (2006) 405–419

by the respective data collecting agencies as per regulations listed in Koterba et al. (1995). Water quality data included records of concentrations for various contaminants and water quality indicators such as nitrates, pesticides, heavy metals, pH, phosphorus, and other organic and inorganic contaminants. Water quality data between 1990 and 2000 were used in this work (Kaluarachchi et al., 2002). Nearly 4500 wells have been sampled in the Sumas-Blaine Aquifer between 1975 and 2000. Since the data were not collected as a part of a structured monitoring plan, wells do not have records for all the contaminants. The two major contaminants in the Sumas-Blaine Aquifer are nitrate and arsenic. Kaluarachchi et al. (2002) analyzed nitrate occurrences in the study area for the period from 1990 to 2000. They found that the nitrate concentrations in ground water are increasing during the past decade. In addition, the results showed an increase in the number of wells with nitrate concentrations exceeding the MCL. A detailed statistical analysis of nitrate in the Sumas-Blaine Aquifer may be found in Kaluarachchi et al. (2002). Twarakavi and Kaluarachchi (2005) analyzed the heavy metal concentrations in the study area and have shown a high occurrence of arsenic in the study area. Twarakavi and Kaluarachchi (2005) estimated the maximum and median arsenic concentrations of 1700 and 13 mg/L, respectively in the Sumas-Blaine Aquifer. These concentrations are high compared to the MCL of 10 mg/L. Readers are referred to Twarakavi and

Kaluarachchi (2005) for a detailed discussion of the heavy metal concentrations in the study area. 4. Results and discussion The LTM was used to simulate the future land use patterns of WRIA 1, which includes most of Whatcom County. Land use change simulation was performed for WRIA 1 instead of the Sumas-Blaine aquifer because population statistics and projections are more efficiently used in the LTM with political boundaries than hydrogeological boundaries. Sustainability analysis for GWQ was performed for the Sumas-Blaine aquifer. The future land use projections for the years 2015 and 2025 were extracted from the LTM results for the WRIA 1. Sustainability of GWQ can be analyzed between any time step of years 1995, 2015, and 2025. However, GWQ for the years 1995 and 2015 was analyzed for simplicity. 4.1. Land use changes Fig. 3 shows the driving variables considered as inputs to the LTM. The results of the LTM simulation for WRIA 1 are provided in Fig. 4. Land use data used for training the neural network were from data of 1975 and 1995. Population forecast data for WRIA 1 were used to project the future land use changes for the years 2015 and 2025. Future intermediate projections from population data were

Fig. 3. Description of data used in the land transformation model: (a) an example of a driving variable, distance to rivers/streams and (b) exclusionary zones of WRIA 1.

ARTICLE IN PRESS N.K.C. Twarakavi, J.J. Kaluarachchi / Journal of Environmental Management 81 (2006) 405–419

413

Fig. 4. Historical land use data and the predicted future land use changes by the LTM in WRIA 1.

Fig. 5. A map showing a small part of WRIA 1 to demonstrate the prediction capability of the LTM.

obtained from the Washington County census data. Historical populations for the years 1975 and 1995 are 91,700 and 149,114, respectively. Projected intermediate estimates of population in 2015 and 2025 are 213,246 and 246,636, respectively. As a validation, the LTM was used to forecast land use changes in 1995 and a map of the differences between the observed and predicted land use changes is shown in Fig. 5. During the model development phase, these maps were used as visualization tools to discern the type of spatial error pattern that may indicate missing driving variables. A match of 55 percent between the observed and predicted land use change to urban was obtained in the final model simulation as shown in Fig. 5 (note that correctly identified changes are in gray; locations not correctly identified are in black). The errors may be attributed to a number of reasons; first, the grid size of the model is 30 m  30 m which is relatively fine based on the poor resolution of the available data. Second, the quality of the land use databases in 1975 and 1995 was not good

affecting the accuracy of the model predictions. Owing to the significant differences in technologies and data acquisition methods, one may expect the 1995 land use database to be superior in quality compared to the 1975 land use database. Since errors in the spatial–temporal model tend to be auto-correlated, a scalable window goodness-of-fit algorithm (Pijanowski et al., 2000) was applied to assess the predictability across spatial scales. The approach produced windows of 90 m  90 m, 150 m  150 m, etc., across the entire WRIA 1 and searched for the accuracy of model results. The model results improved to 79 percent between the observed and predicted at a window size of 1 km (34 cells). The increase in the match suggests that the ANN is learning well about the conditions that lead to land use changes. The LTM model was then used to predict the future land use changes in WRIA 1 for the years 2015 and 2025. The future predictions produced interesting results. Fig. 6

ARTICLE IN PRESS 414

N.K.C. Twarakavi, J.J. Kaluarachchi / Journal of Environmental Management 81 (2006) 405–419

Fig. 6. Percentages of historical and predicted distributions of different land use classes in WRIA 1 and the Sumas-Blaine aquifer.

shows the historical and predicted percentages of different classes for WRIA 1 and the Sumas-Blaine aquifer. It was observed that highways and road infrastructure followed by distance to urban seem to have the maximum influence on land use changes. The area, where urban development seems to encroach into other land use classes, is of interest. Future land use development in the watershed is anticipated along the highway running across the watershed from the northwest to the southeast. It is also interesting to note an increase in future urban development in areas (such as the eastern part of WRIA 1) that have been forested in the past. The proportion of land under each land use class seems to indicate a significant increase in urban/built-up land in WRIA 1 consistent with an increase in future population. 4.2. Sustainability of GWQ On a regional-scale, three classes of contaminants are commonly detected in ground water—heavy metals, pesticides, and nitrate. The same is observed in ground waters of Sumas-Blaine Aquifer. As a first step, the major contaminants in the study area were identified from the available database. Kaluarachchi et al. (2002) give a detailed statistical analysis of nitrate, pesticides, and other regional-scale contaminants found in the area. Twarakavi and Kaluarachchi (2005) performed a statistical analysis of the heavy metal concentration data. Out of all pesticides, Ethylene Dibromide (EDB) and 1,2-DCP were detected above the respective MCLs. However, the data were found to be inconsistent and statistically not robust enough and, therefore, was not used in the analysis. Data for nitrate were adequately available for analysis. Similarly, arsenic was selected for heavy metal analysis as it was found to be the major heavy metal present in the Sumas-Blaine aquifer.

Other metals, such as chromium, cadmium, mercury, and lead, were found in insignificant quantities and posed no health risk. Statistical analysis was performed to check for the robustness of data. Finally, nitrate and arsenic were used in representing the GWQ of the area due to the abovementioned data limitations. Nitrate represents the noncarcinogens while arsenic represents the carcinogens. Binary LR analysis was performed to estimate the probability of nitrate and arsenic to exceed the respective MCLs. The MCL of nitrate and arsenic is 10 mg/L as N and 10 mg/L, respectively. The concentration data for nitrate and pesticides were checked for temporal consistency with respect to the threshold concentration of the MCL. Also, only shallow wells were selected such that the wells were truly located within the shallow Sumas-Blaine aquifer. Details of the well sampling procedures are explained elsewhere (Twarakavi and Kaluarachchi, 2005). A number of influencing variables was considered in the binary LR model; these variables are: (a) well depth, (b) surface elevation, (c) clay content, (d) SHG, (e) hydraulic conductivity, and (f) land use. A radius of one mile was used to estimate the land use class and soil type surrounding a given monitoring well. Significance of an influencing variable to the response was checked by performing a univariate LR analysis between the influencing variable and the response and estimating the Wald statistic and w2-test. A Wald statistic, W, greater than two and a w2 p-value o0.25 imply that the variable is significant (Hosmer and Lemeshow, 1989). The results of the univariate analysis were used to select the significant influencing variables. Among the land uses, the influence of agriculture in the occurrence of arsenic and nitrates in ground water seems to be the most significant. Similarly, good draining soils seem to promote high vulnerability which is expected. The best influencing variables were selected for subsequent multivariate LR analysis. The multivariate binary LR model was developed following the methodology described in Hosmer and Lemeshow (1989). Measures of association between the predicted and observed probabilities had a percentage of concordant pairs between 70 and 85 percent for arsenic and nitrate, respectively. The final multivariate binary LR model relating the probability of concentrations to exceed the MCLs of nitrate and arsenic is shown as Logit ðconcentration4 ¼ MCLÞ ¼ b0 þ bE E þ bA A þ bU U þ bF F þ bO O þ bGA G A þ bGB G B þ bGC G C þ bGD GD ,

ð7Þ

where b0 is a constant; bE is the slope coefficient for elevation; bA, bU, bF, and bO are slope coefficients for variables representing fraction of land use classes agriculture (A), urban/built-up (U), forested/barren (F), and others (O), respectively; and bGA, bGB, bGC and bGD are slope coefficients for fractions of soil hydrologic groups (SHG) A (GA), SHG B (GB), SHG C (GC), and SHG D (GD), respectively.

ARTICLE IN PRESS N.K.C. Twarakavi, J.J. Kaluarachchi / Journal of Environmental Management 81 (2006) 405–419

Table 1 provides the value of the coefficients for the influencing variables for nitrate and arsenic vulnerability models. Spatial maps showing the probability of arsenic and nitrate to exceed the corresponding MCLs in the study area were developed for the years 1995 and 2015. The vulnerability map for 1995 considered land use data of 1995; the map of 2015 used the predicted land use results of the LTM. Arsenic vulnerability maps for the years 1995 and 2015 are shown in Fig. 7. Vulnerability of ground water to arsenic contamination shows a high dependence on elevation indicating the presence of natural sources (Twarakavi and Kaluarachchi, 2006). However, there is a

Table 1 Expected values of slope coefficients of influencing variables in the logistic regression equations for arsenic and nitrate Coefficient

Arsenic

Nitrate

b0 bE bA bU bF b0 bGA bGB bGC bGD


0.62 0.04 1.00 0.65
0.01 0.25 2.01 1.85 0.49 0.00


5.02 — 10.61 3.89 5.14 7.49 0.00 0.04
1.60
0.383

415

negligible change in GWQ from arsenic contamination in 2015. A detailed explanation of the vulnerability analysis of arsenic is discussed in Twarakavi and Kaluarachchi (2006). Nitrate vulnerability maps for the years 1995 and 2015 are shown in Fig. 8. Land use, especially agriculture, is the driving force for ground water contamination by nitrate. Agricultural activities seem to contribute heavily towards nitrate contamination of ground water. The simulated results showed that there would be a major decrease in vulnerability of ground water to nitrate in 2015 as a result of urban development. The next step towards understanding GWQ sustainability is to estimate the carcinogenic and non-carcinogenic GWQ maps for the study area. As mentioned earlier, in spite of the proposed methodology to consider multiple carcinogens and non-carcinogens, the lack of data for pesticides and the absence of data for other major contaminants have reduced the number of contaminants to two, i.e., nitrate and arsenic. Due to the presence of a single contaminant in each group, the need to use a weighting system described in Eqs. (3) and (4) is not necessary. Therefore, the vulnerability maps for nitrate and arsenic can be considered as carcinogenic GWQ and noncarcinogenic GWQ maps, respectively. Figs. 9a and b show maps illustrating how the carcinogenic and non-carcinogenic GWQ is changing between the years 1995 and 2015, respectively. It can be noticed that in the Sumas-Blaine aquifer, urbanization

Fig. 7. Probability of arsenic exceeding the MCL in shallow ground waters of the Sumas-Blaine aquifer for years 1995 and 2015.

ARTICLE IN PRESS 416

N.K.C. Twarakavi, J.J. Kaluarachchi / Journal of Environmental Management 81 (2006) 405–419

Fig. 8. Probability of nitrate exceeding the MCL in shallow ground waters of the Sumas-Blaine aquifer for 1995 and 2015.

Fig. 9. Changes in carcinogenic (a) and non-carcinogenic (b) ground water quality between 1995 and 2015 in the Sumas-Blaine aquifer.

tends to help GWQ of the area. The results indicate that GWQ is sustainable in the future. This is somewhat obvious as the aquifer is currently heavily used in agricultural activities; and any change in the land use pattern from agriculture to urban can improve GWQ. The

carcinogenic GWQ of the study area between 1995 and 2015 seems to change by less than 0.01. While 24.2 percent of the study area undergoes an improvement in GWQ, the remaining part of the study area experiences degradation in carcinogenic GWQ. The area undergoing improvement in

ARTICLE IN PRESS N.K.C. Twarakavi, J.J. Kaluarachchi / Journal of Environmental Management 81 (2006) 405–419

417

Fig. 10. Map showing sustainability of ground water quality between the years 1995 and 2015.

carcinogenic GWQ is located in the western part of the Sumas-Blaine aquifer, which is predicted to have new urban land development in the year 2015. The non-carcinogenic GWQ in the year 2015 is predicted to have a drastic change. While 81 percent of the study area is predicted to have an improvement in non-carcinogenic GWQ between 0 and 0.5, 10 percent of the study area is estimated to have an improvement of between 0.5 and 1.0. Also, less than 10 percent of the study area is predicted to face degradation in non-carcinogenic GWQ. As is the case with carcinogenic GWQ, the area facing improvements in non-carcinogenic GWQ is located in the western part of the Sumas-Blaine aquifer, which will undergo major urbanization in the future. Maps can now be developed that show the areas with sustainable GWQ. Fig. 10 shows sustainability of GWQ in the Sumas-Blaine aquifer for the year 2015. It may be observed that areas that are sustainable are mostly in the western part of the Sumas-Blaine aquifer. Almost 60 percent of the Sumas-Blaine aquifer has sustainable GWQ in 2015, i.e., the same quality as 1995. Nearly 17.5 percent of the area is predicted to have an improvement in carcinogenic and non-carcinogenic GWQ indicating high sustainable GWQ while 6.5 percent is predicted to experience degradation in carcinogenic and non-carcinogenic GWQ. Also, 12 percent of the study area is predicted to have an improvement in non-carcinogenic GWQ and degradation in non-carcinogenic GWQ, while only 2 percent is predicted to experience the converse. In this work, maps produced by the proposed methodology showing carcinogenic GWQ and non-carcinogenic GWQ and sustainability can be used for a variety of purposes. The maps provide an understanding of the impact of various land use relevant factors on future GWQ. The results derived from these maps could be effectively used to blend GWQ in developing land use and land management policy. For example, in the SumasBlaine aquifer, it is observed that highways and urbanization seem to improve GWQ. One may use this result in the designing process for future construction in the land occupied by the aquifer. Some of the results of the Sumas-Blaine aquifer may sound contradictory to common

philosophy about environmental quality such as urbanization which seems to help improve GWQ. This result, however, is valid for the following reasons: (1) future urbanization encroaches into agricultural lands that are currently loading ground water with agricultural chemicals such as nitrate and arsenic more than any other land use class; (2) microbial contamination and other possible regional-scale contaminants are not considered due to lack of data; and (3) only GWQ is considered and impacts due to air quality degradation, solid waste production, and urban water quality issues have not been addressed. 5. Conclusions Sustainability of GWQ has been a topic of interest in the recent past. In this paper, a methodology to address sustainability of GWQ was proposed and demonstrated. The major strength of this approach is the consideration of multiple contaminants and their respective health risks together with land use changes. The LTM performed relatively well for the Sumas-Blaine aquifer of Washington State, which is located in the northwest corner of the state with an area of about 350 square miles. The model predicted the land use change up to 60 percent accuracy and the accuracy increased to greater levels as the spatial resolution was decreased. The results of this work and the proposed methodology have broad implications in policy and decision-making. The methodology proposed here can be used readily to identify major sources of GWQ degradation due to various land use practices and the impact of these land use practices on the future of GWQ. The information derived from this methodology is useful to land managers and policy-makers in addressing land management decisions that are of interest to the public and the surrounding ecosystem. Some of the major site-specific conclusions derived for the Sumas-Blaine aquifer are as follows: 1. Urbanization is most prevalent in areas with a high proximity to urban centers or highways. 2. Agricultural activities seem to have the worst impact on GWQ of all land use classes while well-draining soil

ARTICLE IN PRESS N.K.C. Twarakavi, J.J. Kaluarachchi / Journal of Environmental Management 81 (2006) 405–419

418

3.

4.

5.

6.

hydrologic groups A and B have the worst impact on GWQ. Compared to the current carcinogenic and non-carcinogenic GWQ, it is predicted that the GWQ will improve significantly in the future, as urbanization encroaches into agricultural lands. Non-carcinogenic GWQ is predicted to improve significantly by the year 2015 compared to 1995. The improvement in carcinogenic GWQ is negligible. Sustainable GWQ for the year 2015 is predicted in the western part of the aquifer, where the density of highways is greater and more urban centers are present. It is predicted that almost 60 percent of the SumasBlaine aquifer will have sustainable GWQ in 2015. Nearly 17.5 of the area is predicted to have an improvement in carcinogenic and non-carcinogenic GWQ indicating high sustainable GWQ, while 6.5 percent is predicted to experience degradation in carcinogenic and non-carcinogenic GWQ.

There are some limitations in the proposed methodology due to the uncertainties of the LTM and the binary LR model. LTM models the changes in land use under the assumption of high transition probability for one land use class. In other words, LTM models land use changes one land use class at a time. Therefore, the historical land uses need to be analyzed for selecting the land use class with a high transition probability. In a case where more than one land use has a high transition probability, the LTM model may not be effective. However, it is typical in landscape ecology that one land use class, typically urban/built-up, has a high transition. The method of binary LR to estimate aquifer vulnerability to heavy metal contamination is, however, subject to the quality of data available. The temporal and spatial variability of data has to be analyzed for each data input. The method, however, is superior over the other vulnerability estimation techniques as discussed earlier. Acknowledgments The funding for this work was provided by the Inland Northwest Research Alliance of Idaho and the US Department of Energy. The authors would like to thank the WRIA 1 technical team and Mark Winkelaar of the Utah Water Research Laboratory for the data. The authors would also like to thank Dr. Bryan Pijanowski and Snehal Pithadia of Purdue University for the Land Transformation Model and their support in this research. References Afifi, A.A., Clark, V., 1984. Logistic Regression in Computer-aided Multivariate Analysis. Lifetime Learning Publications, Belmont, CA. Agarwal, C., Green, G.M., Grove, J.M., Evans, T.P., Schweik, C.M., 2002. A review and assessment of land-use change models: dynamics of space, time, and human choice. General Technical Report NE-297, US

Department of Agriculture, Forest Service, Northeastern Research Station, Newton Square, PA, 61pp. Alley, W.M., Emery, P.A., 1986. Groundwater model of the Blue River Basin, Nebraska—twenty years later. Journal of Hydrology 85, 225–249. Alley, W.M., Reilly, T.E., Frank, O.L., 1999. Sustainability of groundwater resources. US Geological Survey Circular 1186, 1p. Bacchus, S.T., 1998. Determining sustainable yield for karst aquifers of the southeastern coastal plain—a need for new approaches. In: Borchers, J.W. (Ed.), Land subsidence case studies and current research, Proceedings of the Dr. Joseph F. Poland Symposium on Land Subsidence. Association of Engineering Geologists, Special Publication No. 8, pp. 503–519. Boutt, D.F., Hyndman, D.W., Pijanowski, B.C., Long, D.T., 2001. Identifying potential land use-derived solute sources to stream baseflow using ground water models and GIS. Ground Water 39 (1), 24–34. Cox, S., Kahle, S., 1999. Hydrology, water quality, and sources of nitrate in lowland glacial aquifers of Whatcom County, Washington and British Columbia, Canada. US Geological Survey, Water Resources Investigation Report 98-4195. Downing, R.A. (Compiler), 1998. Groundwater Our Hidden Asset. Earthwise Series. British Geological Survey, Keyworth, Nottingham, UK. Eckhardt, D.A.V., Stackelberg, P.E., 1995. Relation of ground-water quality to land use on Long Island, New York. Ground Water 33 (6), 1019–1033. Gelt, J., Henderson, J., Seasholes, K., Tellman, B., Woodard, G., others, 1999. Water in the Tucson area: seeking sustainability. Water Resource Research Center, University of Arizona, Issue Paper No. 20, 155pp. Helsel, D.R., Hirsch, R.M., 1992. Statistical Methods in Water Resources. Elsevier, New York, NY. Hosmer, D.W., Lemeshow, S., 1989. Applied Logistic Regression. Wiley, New York, NY. Kaluarachchi, J.J., Kra, E., Twarakavi, N., Almasri, M.N., 2002. Nitrogen and pesticide contamination of ground water in Water Resources Inventory Area 1. Ground Water Quality Report for WRIA 1, Phase II Report, Utah State University, Logan, UT. Kleinbaum, D.G., 1994. Logistic Regression—A Self-Learning Text. Springer, New York, NY. Koterba, M.T., Wilde, F.D., Lapham, W.W., 1995. Groundwater datacollection protocols and procedures for the National Water-Quality Assessment Program: collection and documentation of water-quality samples and related data. US Geological Survey Open-File Report 95399. McGuire, V.L., Sharpe, J.B., 1997. Water-level changes in the High Plains aquifer—Predevelopment to 1995. US Geological Survey WaterResources Investigations Report 97-4081. Morgan, D.S., Jones, J.L., 1999. Numerical model analysis of the effects of ground-water withdrawals on discharge to streams and springs in small basins typical of the Puget Sound Lowland, Washington. US Geological Survey Water-Supply Paper 2492, 73pp. Nace, R.L., 1960. Water management, agriculture, and ground-water supplies. US Geological Survey Circular 415. Nolan, B.T., 2001. Relating nitrogen sources and aquifer susceptibility to nitrate in shallow ground waters of the United States. Ground Water 39 (2), 290–299. Nolan, B.T., Hitt, K.J., Ruddy, B.C., 2002. Probability of nitrate contamination of recently recharged ground waters in the conterminous United States. Environmental Science and Technology 26 (10), 2138–2145. NRC, 1993. Ground Water Vulnerability Assessment: Contamination Potential Under Conditions of Uncertainty. National Academy Press, Washington, DC. NRCS, 1992. Soil Survey Report of Whatcom County area, Washington, Soil Conservation Service, US Department of Agriculture.

ARTICLE IN PRESS N.K.C. Twarakavi, J.J. Kaluarachchi / Journal of Environmental Management 81 (2006) 405–419 Pearce, D., Markandya, A.I., Barbier, E.B., 1989. Blueprint for a Green Economy. Earthscan Publications Ltd. Pijanowski, B.C., Long, D.T., Gage, S.H., Cooper, W.E., 1997. A land transformation model: conceptual elements, spatial object class hierarchies, GIS command syntax and an application to Michigan’s Saginaw Bay Watershed. Land Use Modeling Workshop, Sioux Falls, SD, June 3–5. Pijanowski, B.C., Gage, S.H., Long, D.T., 2000. A land transformation model: integrating policy, socioeconomics and environmental drivers using a geographic information system. In: Harris, L., Sanderson, J. (Eds.), Landscape Ecology: A Top Down Approach. Pijanowski, B.C., Brown, D.G., Manik, G., Shellito, B., 2002. Using neural nets and GIS to forecast land use changes: a land transformation model. Computers, Environment and Urban Systems 26 (6), 553–575. Rupert, M.G., 1998. Probability of detecting atrazine/desethyl-atrazine and elevated concentrations of nitrate in ground water in the Idaho part of the Upper Snake River Basin. US Geological Survey Water Resources Investigations Report 98-4204. Solley, W.B., Pierce, R.R., Perlman, H.A., 1998. Estimated use of water in the United States in 1995. US Geological Survey Circular 1200, 71pp. Sophocleous, M. (Ed.), 1998. Perspectives on sustainable development of water resources in Kansas. Kansas Geological Survey Bulletin 239. Tesoriero, A.J., Voss, F.D., 1997. Predicting the probability of elevated nitrate concentrations in the Puget Sound Basin: implications for aquifer susceptibility and vulnerability. Ground Water 35 (6), 1029–1039.

419

Tesoriero, A.J., Inkpen, E.L., Voss, F.D., 1998. Assessing ground water vulnerability using logistic regression. In: Proceedings of the Source Water Assessment and Protection ’98 Conference, National Water Research Institute, Fountain Valley, CA. Tooley, J., Erickson, D., 1996. Nooksack watershed surficial aquifer characterization, Washington State Department of Ecology, Ecology Report #96-311. Twarakavi, N.K.C., Kaluarachchi, J.J., 2006. Arsenic in ground waters of conterminous United States: assessment, health risk, and cost. Journal of American Water Resources Association 42 (2), 1–24. Twarakavi, N.K.C., Kaluarachchi, J.J., 2005. Aquifer vulnerability assessment to heavy metals using ordinal logistic regression. Ground Water 43 (2), 200–214. US EPA, 2002. The clean water and drinking water infrastructure gap analysis, EPA-816-R-02-020. Walton, A., 1964. Future water-level declines in deep sandstone wells in Chicago region. Ground Water 2 (1), 13–20. Wayland, K.G., Hyndman, D.W., Boutt, D., Pijanowski, B.C., Long, D.T., 2002. Modeling the impact of historical land uses on surface water quality using ground water flow and solute transport models. Lakes and Reservoirs 7, 189–199. Wayland, K.G., David, T., Hyndman, D.W., Pijanowski, B.C., Woodhams, S.M., Haack, S.K., 2003. Identifying relationships between baseflow geochemistry and land use with synoptic sampling and R-mode factor analysis. Journal of Environmental Quality 32, 180–190.