You are on page 1of 15

Natural Resources Research, Vol. 16, No.

2, June 2007 ( 2007)


DOI: 10.1007/s11053-007-9038-5

Development and Implementation of a Bayesian-based


Aquifer Vulnerability Assessment in Florida

Jonathan D. Arthur,1,4 H. Alex R. Wood,2 Alan E. Baker,2 James R. Cichon,2


and Gary L. Raines3

Received 15 February 2007; accepted 15 February 2007


Published online: 2 June 2007

The Florida Aquifer Vulnerability Assessment (FAVA) was designed to provide a tool for
environmental, regulatory, resource management, and planning professionals to facilitate
protection of groundwater resources from surface sources of contamination. The FAVA
project implements weights-of-evidence (WofE), a data-driven, Bayesian-probabilistic
model to generate a series of maps reflecting relative aquifer vulnerability of Floridas
principal aquifer systems. The vulnerability assessment process, from project design to map
implementation is described herein in reference to the Floridan aquifer system (FAS). The
WofE model calculates weighted relationships between hydrogeologic data layers that
influence aquifer vulnerability and ambient groundwater parameters in wells that reflect
relative degrees of vulnerability. Statewide model input data layers (evidential themes)
include soil hydraulic conductivity, density of karst features, thickness of aquifer confine-
ment, and hydraulic head difference between the FAS and the watertable. Wells with
median dissolved nitrogen concentrations exceeding statistically established thresholds serve
as training points in the WofE model. The resulting vulnerability map (response theme)
reflects classified posterior probabilities based on spatial relationships between the evidential
themes and training points. The response theme is subjected to extensive sensitivity and
validation testing. Among the model validation techniques is calculation of a response theme
based on a different water-quality indicator of relative recharge or vulnerability: dissolved
oxygen. Successful implementation of the FAVA maps was facilitated by the overall project
design, which included a needs assessment and iterative technical advisory committee input
and review. Ongoing programs to protect Floridas springsheds have led to development of
larger-scale WofE-based vulnerability assessments. Additional applications of the maps in-
clude land-use planning amendments and prioritization of land purchases to protect
groundwater resources.
KEY WORDS: Floridan, aquifer, vulnerability, Florida, Bayesian, weights-of-evidence (WofE).

INTRODUCTION
1
Florida Department of Environmental ProtectionFlorida Geo-
logical Survey, 903 W. Tennessee St., Tallahassee, FL 32312,
In many regions of the world, water-quality
USA. declines are the result of anthropogenic activities
2
Advanced Geospatial, Inc., 2441 Monticello Dr., Suite 600, Tal- that threaten groundwater sustainability. For more
lahassee, FL 32303, USA. than three decades, regulatory, growth-manage-
3
US Geological Survey, Western Mineral Resources Team, MS ment, and land-use planning agencies at all levels
176, c/o Mackay School of Mines University of Nevada at Reno,
Reno, NV 89557, USA.
of government have focused on groundwater pro-
4
To whom correspondence should be addressed; e-mail: jonathan. tection. These efforts have become more effective
arthur@dep.state.fl.us with improved characterization of the natural

93
1520-7439/07/0600-0093/0  2007 International Association for Mathematical Geology
94 Arthur, Wood, Baker, Cichon, and Raines

hydrogeologic system and increased collaboration three principal aquifer systems: the surficial aquifer
between policy makers and geoscientists. Among system, the intermediate aquifer system or the Flo-
the scientific resources required in this process are ridan aquifer system (FAS).
maps of aquifer vulnerability. Application of these Aquifer protection in Florida is critically
maps can minimize risk of groundwater contamina- important, resulting not only from the states vul-
tion through the implementation of developmental nerable hydrogeologic setting, but also because
standards, of best management practices, restrictions more than 90% of Floridas drinking water comes
on high-impact land uses, and natural area conser- from groundwater (Marella, 2004). In Florida, the
vation programs and public education. To facilitate average person uses more than 140 gallons of
lasting and practical groundwater resource protec- groundwater per day (Solley and others, 1998).
tion, geoscientists are encouraged to work closely Floridas population exceeds 17.8 million (U.S.
with planning and regulatory officials while devel- Census, 2005), and increases an average of 1,060
oping vulnerability assessment models, recognizing people per day (Vogel, 2006). The U.S. Census
that the assessment is not an end product, but Bureau (press release CB-05-52) projects that Flor-
rather a dynamic and crucial component of a larger idas population will increase by 79.5% between 2000
effort to enhance water-resource protection. and 2035. Moreover, according to Visit Florida
Using a Geographic Information System (GIS), Research (2006), the state had more than 83 million
a relatively new application of a Bayesian probabi- visitors in 2005.
listic approach, weights-of-evidence (WofE; Raines, While water-resource managers in Florida are
Bonham-Carter, and Kemp, 2000), was imple- exploring alternative drinking-water supplies such as
mented to develop aquifer vulnerability assessments desalination, reservoirs, and managed underground
in Florida. The modeling technique and results are storage, regulatory and planning agencies and elec-
described herein as well as the overall aquifer ted officials have become increasingly aware of the
vulnerability assessment process from project design need to manage growth and land use to minimize
to map implementation. Discussion of the model in potential contamination of aquifers. A heightened
context of the overall project helps underscore awareness with regard to increase in nutrients issu-
factors involved in successful implementation of the ing from Floridas springs (Florida Springs Task
assessment toward more effective water-resource Force, 2000; Scott, 2002) helped strengthen the link
protection. The maps are intended to serve as sci- between land-use planning and groundwater pro-
entifically defensible planning and best manage- tection, especially in spring-recharge basins
ment practice tools to minimize adverse impacts on (springsheds). For example, the following language
groundwater quality. was adopted into Florida law in 2004: The Legis-
lature further finds that springs and groundwater
once damaged by overuse can be restored through
Floridas Water Quality Challenge: Karstic Aquifers good stewardship, including effective planning
and a Growing Population strategies and best management practices to pre-
serve and protect the spring and its springshed.
Relative to most regions of the United States, Prudent land use planning decisions can protect and
Floridas aquifers are among the most vulnerable to improve quality and quantity, as well as upland re-
surface sources of contamination (Berndt and oth- sources of a springshed. Managing land use types
ers, 1998). Florida is comprised of a Paleocene and and their allowable densities and intensities of
younger carbonate platform overlain by variably development, followed by specific site planning to
permeable siliciclastic materials. The terrain is further minimize impacts, rank as an important
characterized by paleoshoreline features, fluvial and goal (369.315 [2] Florida Statutes [F.S.]).
internally drained watersheds, karst topography and
rapid timescales of groundwater/surface-water ex-
change. Evidence of karst is widespread, including PREDICTING AQUIFER VULNERABILITY
more than 700 springs (Scott and others, 2004),
thousands of cave systems (Kincaid and others, In this paper, aquifer vulnerability is defined
2004) and tens of thousands of sinkholes. Most sur- as the tendency or likelihood for contaminants to
face-water bodies in Florida receive groundwater reach the top of a specified aquifer system after
baseflow (Winter and others, 1998), which resides in introduction at land surface based on existing
A Bayesian-based Aquifer Vulnerability Assessment in Florida 95

knowledge of natural hydrogeologic conditions calculating a weighted score. Many variants of this
(modified from National Resource Council [NRC], index method exist (e.g., Edet, 2004; Navulur, Engel,
1993). An appropriate model on which the vulner- and Mamillapalli, 1995; Rupert, 1997). Simulation
ability assessment is based must be able to address models consider the role of hydrologic and hydrog-
data gaps, variably resolved spatial data coverages eologic processes such as transport and dispersion
(i.e. resolutions that differ by orders of magnitude) (e.g., Connell and van den Dale, 2003; Huaming and
and the complex and dynamic nature of hydrogeo- Wang, 2004; Stewart and Loague, 2003). Multivari-
logic systems. Essential to the assessment are the ate, fuzzy logic, neural networks and probability
three laws of groundwater vulnerability: (1) all analyses are among the statistical group of models.
groundwater is vulnerable, (2) uncertainty is inher- Nolan (2001), for example, applied logistic regres-
ent in all vulnerability assessments, and (3) the sion to U.S. Geological Survey National Water-
obvious may be obscured and the subtle indistin- Quality Assessment data to assess aquifer suscepti-
guishable (NRC, 1993). bility to contamination. An R-mode factor analysis
was used by Lawrence and Upchurch (1982) to
associate water-quality analyses in terms of pro-
DRASTIC ORIGINS cesses affecting aquifer recharge, such as carbonate
dissolution, ion exchange in confining sediments,
An early and significant groundwater protection and land use. Dixon and others (2001) are among
resource for environmental managers and planners researchers applying a neural network approach to
in the United States was DRASTIC, a methodology predicting vulnerability with an emphasis on soil
designed to identify areas of relative vulnerability to structure. Witkowski and others (2003) applied a
anthropogenic contamination based on hydrogeo- hybrid approach by using MODFLOW calibration
logic characteristics (Aller and others, 1985). The parameters as input data for a DRASTIC-type index
components of DRASTIC include: Depth-to- model. An emerging statistical approach involves
watertable, net Recharge, Aquifer media, Soil application of Bayesian probability (Alberti and
media, Topography, Impact of the vadose zone, and others, 2001; Cichon and others, 2005; Masetti, Poli,
hydraulic Conductivity of the aquifer. Implementa- and Sterlacchini, 2005, 2007). Empirical models
tion of DRASTIC mapping has been useful and has focus on properties of geologic media comprising
increased awareness with regard to aquifer protec- the aquifer, such as sorption capacity (Bekesi
tion; however, the modeling technique has several and McConchie, 2000) or grain size (Bekesi and
shortcomings (Soller, 1992; Merchant, 1994; Ross McConchie, 2002). The fifth model-type, hybrid, is
and others, 2004). In a coastal plain hydrogeologic simply any combination of these other methods.
setting dominated by karst topography, results of This cross section of studies underscores the diver-
DRASTIC modeling overemphasize topography sity of approaches available for aquifer vulnerability
and do not account for karst features, which are assessments.
significant contributors to aquifer vulnerability (e.g., Although these types of models describe the
Arthur and others, 2003; Beck and Jenkins, 1988; framework on which vulnerability assessments are
COST Action 65, 1995; Doerfliger, Jeannin, and based, the manner in which input information is
Zwahlen, 1999). Sinkholes, for example, may func- considered falls into a different classification:
tion as unobstructed shortcuts that allow surface knowledge driven or data driven. Knowledge-driven
contamination to enter an aquifer system. models (also known as expert models) rely on
expert scientific opinion, insight and perhaps anec-
dotal information. Data-driven models are based on
MODEL TYPES measured observations. DRASTIC, for example, is a
knowledge-driven index model.
Models of aquifer vulnerability generally fall
into five broad categories: index, simulation, statis-
tical (i.e., probabilistic, experimental), empirical and METHODS
hybrid models (Bonham-Carter, 1994; Focazio and
others, 2002; NRC, 1993; Rupert, 1997, 1999). Index The Florida Aquifer Vulnerability Assessment
models, such as DRASTIC combine spatial data (FAVA) project was designed to address the con-
layers (i.e., maps reflecting different parameters) by cerns outlined here regarding the need for a more
96 Arthur, Wood, Baker, Cichon, and Raines

advanced series of maps that reflect relative aquifer  Scientifically defensible results
vulnerability in Florida using a data-driven approach.  Results can be validated by hydrochemistry
A more comprehensive understanding of the project
and implementation of the WofE-derived assessment
is facilitated by description of selected details,
including expert review during the project, assess- Model Selection
ment of end-user needs and model selection criteria.
Throughout the development of the FAVA Several models were evaluated relative to the
project, a policy of adaptive management was above criteria as potential frameworks upon which
implemented. Part of this process involved the FAVA maps would be developed. There was con-
assembly and collective input from a multiagency sensus among the investigators and the TAC that an
Technical Advisory Committee (TAC). Project index model or a purely knowledge-driven model
TAC members participated through semiannual would introduce relatively higher bias compared to a
workshops, provided technical review of interim data-driven (i.e., statistical or empirical) model.
model outputs from pilot studies, and generally Moreover, index models are difficult to evaluate in
served as a sounding board as the project pro- terms of validation and co-dependent input data.
gressed. The TAC members were points of contact A pilot study was initiated to evaluate different
for agency resources (i.e., GIS coverages and raw modeling techniques in three Florida counties with
data). Expertise among TAC members included diverse hydrogeologic settings (Baker and others,
water quality, hydrologic modeling, hydrogeology, 2002). Three models were ultimately considered: (1)
and land-use planning. Three members of the com- a semi-empirical model applying composite vertical
mittee had first-hand experience in development of hydraulic conductivity (Kv) and influence of proxi-
the Florida DRASTIC model. As feedback from the mal karst features to estimate conservative con-
TAC was received, course corrections in the data taminant travel time to top of the aquifer system; (2)
development and project plans were made. a fuzzy-logic model utilizing expert knowledge; and
A Delphi study also was completed to reach (3) WofE. Upon completion of vulnerability maps
consensus among a subgroup of potential end-users for pilot study areas using all three methods, the
(land-use planners) regarding aspects of the overall WofE model was determined to best fit the estab-
assessment (Ferguson, 2002). The study included a lished criteria. The fuzzy logic model is knowledge-
comparison of relative anticipated benefits of the driven, rather than primarily data-driven, and did
project as compared to DRASTIC. Delphi-study not meet the easy to explain criterion. The
results included the development of a list of criteria empirical model was excluded because of requisite
that potential end-users believed would maximize assumptions of vertical hydraulic conductivity for
usefulness of the assessment: large areas. Further details regarding the develop-
ment and assessment of these models is described in
 Appropriate list of parameters Arthur and others (2007, in review). The aquifer
 Sensitivity of scale (e.g., GIS grid-cell size vulnerability assessment model applied in this study
adequate to represent karst) is WofE, implemented in Arc Spatial Data Modeler
 Address and reduce uncertainties (ArcSDM; Kemp and others, 2001), an extension to
 Well-documented methodology ESRIs ArcView (3.x). More recent versions of
 Easy to upgrade given future data ArcSDM (Sawatzky and others, 2004) have been
 Easy to comprehend developed for ESRIs ArcGIS (8.x and 9.x).
 Clarity in presentation of results
 Use of existing data
Weights-of-Evidence (WofE) and Bayes Rule of
In addition, the following model selection Probability
criteria were identified by the TAC:
Use of the WofE modeling technique involves
 Easy to explain the combination of diverse spatial data that are used
 Meet identified end-users needs to describe and analyze interactions and generate
 GIS format (scaleable, updateable and predictive models (Bonham-Carter, 1994; Raines,
flexible) Bonham-Carter, and Kemp, 2000). Weights-of-
A Bayesian-based Aquifer Vulnerability Assessment in Florida 97

evidence is a data-driven model that utilizes known Bonham-Carter (1994) provides further details of
occurrences as model training sites to create maps the WofE technique.
from weighted input data layers and reduce some of
the inherent bias that can be associated with
knowledge-driven models. Masetti, Poli and Ster- RESULTS
lacchini (2005) describe historical GIS application of
WofE for mineral-potential mapping, and note Although the FAVA project included vulnera-
that only recently has the model been applied bility assessments for Floridas three major aquifer
to aquifer vulnerability assessments. Weights-of- systems, this paper focuses on the FAS (Fig. 1) to
evidence is based on Bayes Rule of Probability with illustrate the major components of WofE modeling:
an assumption of conditional independence. The training points, evidential themes and the response
model is implemented in a log-linear form so the theme. Implementation of the assessment is de-
weights from the evidential themes can be added. scribed in the Discussion. In the FAS WofE analysis,
The model combines the weights of the predictor the unit area was 1 km2 and the calculated prior
variables from the input data to express a proba- probability equaled 0.0013. The final response theme
bility that a unit cell will contain a training point grid cell size is 900 m2, which is dictated by the most
(Bonham-Carter, 1994). highly resolved evidential theme.

Components of a WofE Analysis Training Points

Delineation of a study area (i.e., aquifer system In FAVA, training points are based on water-
areal extent) is the critical first step in implementing quality analyses from wells that represent ambient
a WofE-based aquifer vulnerability assessment be- (background) groundwater quality conditions
cause the identified area is used in the calculation of throughout the study area. At the time the project
weights and probabilities throughout the modeling began, the Florida Department of Environmental
process. Water-quality data from wells serve as Protection (FDEP) ambient groundwater monitor-
model training points (known occurrences of aquifer ing network provided the most comprehensive
vulnerability) which are combined with spatial data dataset that had undergone significant quality
layers, or evidential themes, representing hydrogeo- assurance/quality control (QA/QC), including con-
logic factors such as soil hydraulic conductivity, sistent sampling and analytical protocols. Available
karst features, thickness of confinement, etc. A prior water-quality parameters that reflect some measure
probability is calculated by dividing the number of of relative aquifer recharge (i.e., vulnerability) were
training points by the total study area, which is evaluated. Although dissolved oxygen and tritium
characterized by unit cells of a predetermined size. are useful parameters in this regard, dissolved
The prior probability represents the probability that nitrogen is among the highest precision and most
a training point will occupy any given unit area widely available water-quality parameters in the
within that study area, prior to the input of evidence. networks database. Naturally occurring nitrogen is
Weights calculated in WofE establish spatial asso- generally considered ubiquitous at land surface;
ciations between training points and evidential further, relatively low dissolved nitrogen concen-
themes. Following weights calculations, evidential trations occur in well-protected (i.e., effectively
themes are generalized (as two or more classes) to confined or less vulnerable) aquifer systems. As
reveal the areas having relative degrees of spatial such, dissolved nitrogen in ambient groundwater is a
association with training points. Upon completion of good indicator of relative recharge or vulnerability
the WofE analysis, the response theme output rep- (Navulur, Engel, and Mamillapalli, 1995).
resents the spatial distribution of posterior proba- A training point set of dissolved nitrogen con-
bilities (i.e., probability that a unit area contains a centrations was derived from more than 780 wells in
training point based on input evidential themes). the ambient network. Median dissolved nitrogen
Higher posterior probability values indicate that an concentrations were calculated to represent each
area is more likely to contain a training point, training point (monitor-well location) in the model.
whereas lower posterior probability values indicate An upper fence statistical analysis justified re-
that an area is less likely to contain a training point. moval of 152 outlier median values, which likely
98 Arthur, Wood, Baker, Cichon, and Raines

Figure 1. Floridan aquifer system (FAS) WofE study area.

represented non-ambient conditions (i.e., anthropo- used as input in the FAS FAVA model. These
genic sources). The dissolved nitrogen concentration coverages provided a minimal degree of overlapping
at the 75th percentile of the distribution (0.036 mg/ information (i.e., conditional independence; see also
L) represents the threshold above which median Discussion). Details regarding development of these
concentrations are considered to reflect aquifer statewide coverages (i.e., well data analyses, polygon
vulnerability. A final data set of 148 wells (median attribution, surface kriging, watertable linear
concentrations within the upper 25th percentile) regression, etc.) are described in Arthur and others
comprise the training points for the FAS WofE (2007, in review).
model. Development and QA/QC of these statewide
spatial coverages, along with the development of a
statewide 15-m digital elevation model (DEM), ab-
Model Evidence sorbed a majority of financial resources available for
this project. Although not an evidential theme, the
Several evidential themes were considered DEM was of critical importance as it was utilized to
during iterative runs of the WofE model, including calculate statewide depth to water, confinement
surface geology, soil drainage, soil hydraulic con- thickness and identify closed topographic depres-
ductivity, aquifer head difference, vadose zone sions. The latter coverage was refined significantly to
lithology, aquifer confinement thickness, karst fea- serve as a proxy for karst feature density. Data used
ture density, and other relevant spatial data sets. In for development of evidential themes have English
the final analysis, evidential themes representing soil and metric units, depending on the data type. The
hydraulic conductivity, karst features, thickness of DEM, for example, is in meters whereas confine-
aquifer confinement, and hydraulic head difference ment thickness is in feet. Although mixed units
between watertable and potentiometric surface were are included in this report, the GIS platform and
A Bayesian-based Aquifer Vulnerability Assessment in Florida 99

ArcSDM are transparent to these variables, thus no ential pathways for infiltration of contaminants to an
need exists for standardization. aquifer system from land surface. The WofE analysis
Soil hydraulic conductivity is a measure of the indicated that a binary classification was justified.
rate at which water travels through the soil horizon Based on the location of training points relative to
(Fig. 2A). Areas with high soil hydraulic conduc- the closed topographic depressions, areas greater
tivity are normally associated with higher aquifer than 3,420 m from effective karst are associated
vulnerability. The WofE analysis indicated that the with areas of lower vulnerability (Fig. 2D). Areas
highest and most significant contrast in hydraulic located within 3,420 m of a karst feature are
conductivity occurs at 19.7 inches/hour, indicating associated with high vulnerability.
that a binary classification using this value best re- In Florida, aquifer confinement overlying the
flects the spatial relation between this parameter and FAS occurs in the form of the Intermediate Aquifer
training points (Fig. 2B). Areas underlain by soil System/Intermediate Confining Unit (Southeastern
hydraulic conductivity ranging from 0.1 to 19.7 in/hr Geological Society, 1986). This confinement
are associated with areas of lower vulnerability (Fig. 2E) reduces flow of groundwater where it is
based on the location of training points, whereas thick and laterally continuous between the overlying
areas underlain by soil hydraulic conductivity aquifer (or land surface) and the FAS. Areas
exceeding 19.7 in/hr are associated with areas of underlain by relatively thin to absent confining
higher vulnerability. sediments are associated with higher aquifer vul-
Areas in close proximity to a closed topo- nerability. The results of the WofE analysis for this
graphic depression (Fig. 2C), a proxy for karst-fea- evidential theme justified delineation of a multi-class
ture density, (or effective karst, Arthur and others, theme (Fig. 2F): (1) areas underlain by greater than
2006, 2007) are associated with higher aquifer vul- 451 feet of confinement; (2) areas underlain by be-
nerability, as these features tend to provide prefer- tween 160 and 451 feet of confinement; and (3) areas

50 25 0 50
Miles
50 25 0 50
Kilometers

Soil Hydraulic Conductivity


Soil Hydraulic Conductivity
(in/hr)
(in/hr)
20.0 19.7 - 20.0
0.1 - 19.7
0.1
STATSGO data used

A B

Karst Features (meters)


0 - 3,420
> 3,420

Closed Topographic
Depressions

C D

Figure 2. Evidential themes and their generalizations based on WofE weights analysis: A, soil hydraulic conduc-
tivity; B, binary classification of soil hydraulic conductivity; C, closed topographic depressions; D, binary classifi-
cation of karst-feature density.
100 Arthur, Wood, Baker, Cichon, and Raines

50 25 0 50
Miles
50 25 0 50
Kilometers
Confinement Thickness (ft)
1 -100
701 - 800
101 - 200 Confinement Thickness (ft)
801 - 900
201 - 300
901 - 1,000 (feet)
301 - 400
1,001 - 1,100 0 - 160
401 - 500
1,101 - 1,200 160 - 451
501 - 600
601 - 700 1,201 - 1,226 >451
Area outside of IAS extent; subject to
local and variable confining conditions

E F

Water Table - FAS Head Difference


(feet)
-89 - -50 46 - 60 Head Difference (feet)
-49 - -30 61 - 80
-8 to 224
-29 - -20 81 - 100
-19 - 0 101 - 120 -89 to -8
0 - 20 121 - 140
21 - 30 141 - 175
31 - 45 176 - 224

G H

Figure 2 (continued). Evidential themes and their generalizations based on WofE weights analysis: E, aquifer
confinement thickness; F, multi-class confinement thickness; G, hydraulic head difference between watertable
and the FAS potentiometric surface; negative values indicate areas where FAS potentiometric surface exceeds
watertable elevation; and H, binary classification of hydraulic head difference.

underlain by less than 160 feet of confinement, the ity that a unit area contains a training point based on
latter class being associated with more vulnerable the spatial relationships between the evidential
areas. theme classifications and the training points, which
Areas where the calculated watertable eleva- are reflected in the calculated weights (Wl, W2, and
tion exceeds the FAS potentiometric surface W3; Table 1). Aquifer confinement thickness has
(Fig. 2G; hydraulic head difference) indicate the three weights because it was a multiclass theme
potential for increased downward recharge to the (n = 3), whereas the other evidential themes were
FAS, which is associated generally with higher binary classifications. The larger absolute values of
aquifer vulnerability. The WofE analysis supported the negative weights (W2 and W3 for confinement
a binary classification for this evidential theme thickness and W2 for other evidential themes)
(Fig. 2H), where a hydraulic head difference greater indicate that the response theme is a better predictor
than ) 8 ft is associated with areas of higher vul- of where training points are not likely to occur. In
nerability. other words, the FAS response theme more strongly
predicts where the FAS is less vulnerable to con-
tamination rather than where it is more vulnerable
Response Theme to contamination. Contrast (Table 1) is a measure of
correlation of each evidential themes relative
After numerous iterations of the WofE model importance with respect to the model output (i.e.,
were completed to optimize selection and general- strength in predicting locations of training points).
ization of evidential themes and response-theme The highest contrast value corresponds to aquifer
confidence, the final response theme was generated. confinement, meaning this evidence shares the
As noted, the response theme reflects the probabil- greatest association with the training points and was
A Bayesian-based Aquifer Vulnerability Assessment in Florida 101

Table 1. Weights-of-Evidence Output Parameters

Evidential Theme W1 W2 W3 Contrast Condence

Aquifer Confinement 0.4127 ) 1.7500 ) 2.7121 3.1248 3.1136


Karst-Feature Density 0.4794 ) 1.1573 1.6367 7.0812
Hydraulic Head Difference 0.2736 ) 1.5470 1.8206 5.2923
Soil Hydraulic Conductivity 0.7336 ) 0.0529 0.7865 2.7967

therefore the primary determinant in predicting ing sediments, dense closed topographic depression
areas of relative vulnerability. Soil hydraulic con- distribution, positive hydraulic head difference, and,
ductivity had the least influence on the model out- to a lesser degree, high soil hydraulic conductivity.
put. Confidence provides a useful measure of The overall test of conditional independence
significance of the contrast because of the uncer- (CI) of the WofE analysis equals 0.64, which falls
tainties of the weights and areas of possible missing outside of the ideal range (1.0 0.15). Expert
data (Raines, 1999). All confidence values in Table 1 knowledge, including TAC input, supports the con-
exceed approximately a 99.5% level of significance clusion that evidential themes do not provide sig-
based on Student t-test values. nificantly redundant information. In such situations,
Figure 3 reveals step-wise increases in posterior application of logistic regression is appropriate to
probability when compared with increasing cumu- address inflated posterior probabilities associated
lative area. Based on input from the TAC, potential with non-ideal conditional independence (Agterberg
end-users and assessment of Figure 3, posterior and others, 1993). Nearly identical response themes
probabilities reflected in the FAS response theme from WofE analyses with and without the logistic
were generalized into three classes (Figs. 3 and 4A). regression option also support the application of this
Although additional classes could have been se- technique.
lected, too many or too few classes of vulnerability
may be difficult for the end user to interpret or
implement. Higher posterior probabilities indicate Sensitivity and Validation
that an area is more likely to be vulnerable to con-
tamination. The highest posterior probability class Among the many strengths of the WofE tech-
generally corresponds with areas of thinner confin- nique is self-validation, which results from use of the
0.01000

More Vulnerable

0.0013
Posterior Probability

0.00100
Vulnerable

0.00029

0.00010

Less Vulnerable

0.00001
0 10 20 30 40 50 60 70 80 90 100
Cumulative Area (%)

Figure 3. Vulnerability of contamination class breaks, represented by dashed lines, are


based on prior probability (0.00130) and occurrence of significant increase in probability
relative to cumulative area.
102 Arthur, Wood, Baker, Cichon, and Raines

Relative Vulnerability
Confidence
More Vulnerable
< 95%
Vulnerable
95% - 99%
Less Vulnerable > 99%
Surface Water Bodies

A B
N

50 25 0 50 Miles

50 25 0 50 Kilometers

Relative Vulnerability Relative Vulnerability


More Vulnerable More Vulnerable
Vulnerable Vulnerable
Less Vulnerable Less Vulnerable
Surface Water Bodies Surface Water Bodies

C D

Figure 4. A, Three-class relative vulnerability of FAS based on posterior probability values (dissolved nitrogen); B,
confidence map based on results in A; C, response theme sensitivity based on random 75% subset of original training points
(dissolved nitrogen); and D, response theme validation applying dissolved oxygen training points.

training point data set. Of the 148 training points, regard to item (l) a qualitative, non-statistical com-
the location of 121 (82%) were correctly predicted parison of land use relative to posterior probability
by the analysis (i.e., the points occur in areas where demonstrated lack of a notable relationship. Only
the posterior probability exceeds the prior proba- one land-use classification, upland forests, slightly
bility). Additional assessment of model output, exceeded the prior probability by 0.0001. The con-
however, is necessary to improve defensibility of the fidence map (Fig. 4B) corresponding to the FAS
results and facilitate implementation. For the FAS response theme, and generated for item (2), indi-
response theme in the FAVA project, this includes: cates that more than 80% of the study area exceeds
(1) comparison of land use with posterior probability 95% confidence as calculated by the WofE analysis.
to assess potential bias; (2) a calculated confidence In addition to the iterative WofE analyses that
map; (3) use of a random 75% subset of training helped identify the most appropriate evidential
points; (4) comparison of dissolved nitrogen values themes, sensitivity analysis of the final response
with posterior probability; (5) comparison with a theme was accomplished by using a random 75%
response theme based on a different training point subset of the original training points in an otherwise
set (dissolved oxygen); and (6) technical review by identical WofE analysis. The resulting response
the TAC (see Methods) and scrutiny through theme (Fig. 4C) was compared to the original FAS
numerous professional presentation venues. With response theme using spatial statistics. Assessment
A Bayesian-based Aquifer Vulnerability Assessment in Florida 103

Figure 5. Relation between three-class relative vulnerability (as posterior prob-


ability) and dissolved nitrogen in FAS (R2 = 0.9346).

of the two responses yielded a kappa correlation potential for misapplication and misinterpretation
coefficient of 0.84 (1.00 equals perfect correlation), because many end-users do not have scientific, or
indicating almost perfect agreement (Landis and more specifically, geoscience backgrounds. The FAS
Koch, 1977). An area-weighted Spearmans Rank vulnerability assessment, although strengthened by
correlation coefficient for the original FAS response various sensitivity and validation methods, has cer-
theme and the random subset response theme equals tain limitations and associated caveats about which
0.985, which is significant at a 99% confidence level. the end-users must be informed. Fundamental
A useful validation tool involves assessment of among these issues are:
dissolved nitrogen relative to posterior probabilities
in the three-class response theme. As expected,  Results of the assessment reflect a snapshot
increasing posterior probabilities (e.g., increased of relative vulnerability of the aquifer system
relative vulnerability) correspond with increasing using best available data at the time during
concentrations of dissolved nitrogen in ambient which the model was completed in March,
groundwater of the FAS (Fig. 5). Aside from this 2005.
technique and critical review the results by the TAC,  Strictly speaking, because of the nature of the
perhaps the most convincing validation of the WofE methodology, the response theme is a
assessment is comparison of the FAS response map of relative vulnerability of the FAS to
theme with that of an independent water-quality dissolved nitrogen; contaminant-specific
parameter indicative of recharge or vulnerability: maps can be generated, which may yield
dissolved oxygen (Fig. 4D). Almost perfect agree- widely differing results.
ment (Landis and Koch, 1977) exists between the  The conceptual model is based on infiltra-
two response themes based on calculation of the tion of a conservative contaminant from
kappa coefficient (0.811). A 99% level of confidence land surface to the top of the FAS, given
exists based on an area-weighted Spearmans Rank hydrogeological characteristics of the natu-
correlation coefficient (0.997) for the FAS response ral system. As a result, if groundwater
theme and the dissolved oxygen response theme. contamination risk is the ultimate goal,
additional factors such as loading/discharge,
groundwater flow gradients, transport,
DISCUSSION dilution, chemical transformation of con-
taminants (i.e., volatilization, denitrification,
Aquifer vulnerability assessments have a broad etc.) also should be taken into consider-
application potential. With this benefit comes the ation.
104 Arthur, Wood, Baker, Cichon, and Raines

 The assessment does not apply to surface- quantitative probabilities. In a more conservative
water bodies including groundwater dis- sense, the maps reflect favorability with regard to
charge areas, such as seeps or springs. hydrogeologic conditions that are conductive to in-
 The response theme is as accurate as the most creased aquifer vulnerability.
detailed evidential theme and as inaccurate The question of an appropriate scale in which to
as the least detailed evidential theme. As apply the assessment results often arises. Given that
discussed next, an appropriate scale at which the response theme is a GIS coverage with a small
to apply the maps is approximately 1:110,000. cell size, there may be a tendency for the end-user to
 Results of the assessment are not intended to make decisions at this large scale. Contrast values
replace site-specific investigations of aquifer for each evidential theme can be used to estimate an
vulnerability. appropriate applied scale for the vulnerability map.
 Disturbed lands (e.g., mined areas), closed Normalized contrast values provide scaled weights
basins that encompass recharge wells, re- for each theme. These weights can be multiplied by
cently formed sinkholes, and regions of sig- published or applied (i.e., approximate) scales for
nificant potentiometric-surface depression each evidential theme. The weighted (re-scaled)
because of historic groundwater mining are map scales can then be summed (Table 2). This
among the factors that may result in under- qualitative approach suggests that application of the
prediction of relative vulnerability. FAS FAVA map is appropriate at a scale of
 The model results reflect vulnerability of the 1:110,000. Examples of land-use planning activities
natural system (i.e., intrinsic vulnerability) at this scale include determination of potential sites
and do not account for human activities. For for landfills, pipelines and residential areas using
example, the presence of a leaky under- septic systems.
ground storage tank, or contaminant plume Another aspect of scale is the study area, which
does not, by definition, affect the vulnerabil- is important with regard to resolution of the re-
ity of the aquifer system, although their sponse theme. Several advantages exist with regard
occurrence has the potential to impact the to larger scale aquifer vulnerability assessments as
quality of groundwater. The results of the compared to the statewide assessment developed in
analysis can be overlain with known contam- the FAVA project. Evidential themes are likely to
inated sites in a susceptibility analysis to be of higher resolution as the study area is reduced.
visualize which of these sites occurs in more For example, the DEM on which spatial evidence
vulnerable areas and therefore have a greater may be based can take advantage of light detection
potential to impact groundwater. and ranging (LIDAR) data as opposed to a coarser
DEM. Arthur and others (2007) provide a compar-
Based on the premise that all aquifers are vul- ison of closed topographic depressions based on 7.5-
nerable to contamination (NRC, 1993), the three minute quadrangle maps versus LIDAR coverage to
classes reflected in FAS response theme (Fig. 4A) emphasize the opportunity for significant data
are labeled less vulnerable, vulnerable, and more refinement. Moreover, assessment of karst-related
vulnerable, the latter corresponding to higher pos- features in large-scale studies can be field-validated
terior probabilities. No classification exists stating and classified with respect to depression type (i.e.,
that the FAS is not vulnerable. Moreover, the des- water-filled sinkhole, karst windows, cover-collapse
ignations emphasize that the map shows relative sinkholes, etc.). Other data sets that may only be
degrees of vulnerability, which are supported by available at the subregional scale include lineaments

Table 2. Approximation of Appropriate Applied Scale for the Response Theme

Evident Theme Applied Scale* Contrast Normalized contrast Contributing scale factor

Aquifer Confinement 1:150K 3.12 0.42 63,000


Karst-Feature Density 1:24K 1.64 0.22 5,300
Hydraulic Head Difference 1:150K 1.82 0.25 37,500
Soil Hydraulic Conductivity 1:20K 0.79 0.11 2,200
Estimated applied scale 1:110,000

*Subjective.
A Bayesian-based Aquifer Vulnerability Assessment in Florida 105

and cave (dry and saturated) maps. Training-point likely to be fully understood. Successful implemen-
data also may exhibit greater spatial density and tation also is significantly enhanced through inter-
contain more constituents. In such situations, the action with forward-thinking decision makers who
training data set may be expanded further or refined are open to adaptive management.
to include temporal changes in water quality or al- Design and completion of the FAVA project
low use of more ideal parameters such as tritium. facilitated implementation within various local and
Minimal refinements in evidential theme resolution, state programs in part because of the membership
when applied in larger scale WofE study areas, can and role of the TAC. The Florida Department of
yield a more highly resolved response theme (Ci- Community Affairs (FDCA), an agency represented
chon and others, 2005). on the TAC, administers Floridas growth-manage-
The WofE methodology as applied to aquifer ment program and statewide land-use planning
vulnerability provides a highly adaptable, scalable, activities. The FDCA developed a guideline docu-
and useful tool for refinement of ongoing ground- ment in partnership with the FDEP: Protecting
water resource protection programs. Because the Floridas Springs: Land Use Planning Strategies and
assessments are based on snapshots of best available Best Management Practices (FDCA and others,
data, the results are static representations and 2002). The FAVA project is referenced therein as a
should be updated periodically as new or more re- resource in development of springs protection plans.
fined data becomes available. New or refined evi- Water quality concerns regarding Wekiva Springs
dential hydrogeological data may reveal more (near Orlando, Florida) provide a more specific
complex or highly resolved patterns of vulnerability. example of FAVA implementation. At the request
of the FDEP, the FAVAWofE methodology was
applied to produce an aquifer vulnerability assess-
IMPLEMENTATION AND APPLICATION ment of the Wekiva area (Cichon and others, 2005).
In 2006, the FDEP incorporated the Wekiva
Although several geoscientists recognize the assessment results in a rule outlining waste-water
inherent vulnerability of aquifers as well as the management requirements in the area.
concepts of model uncertainty and aquifer hetero- Additional applications include use of the
geneity, many end-users of the assessment may have FAVA maps to rank aquifer recharge as part of a
a tendency to seek black and white solutions or decision matrix to identify priority areas for pur-
apply the vulnerability maps at scales larger than chase under the states land conservation program
scientifically defensible. This inclination may be Florida Forever (Florida Natural Areas Inventory,
driven by program needs or may be related to mis- 2006). In addition, the FDEP applies FAVA maps
conceptions about resolution of the assessment. during source-water assessments (e.g., municipal
Successful implementation of vulnerability assess- well-head protection) and review of land-use
ments requires education, which should be viewed as amendments in the context of groundwater protec-
a process rather than an event. Opportunities for tion. The FAVA maps are being applied similarly by
education of end-users can be sought through the Florida Department of Health to identifying
interaction with planning and regulatory agencies priority areas of concern for potable-well contami-
and professional associations in disciplines of envi- nation by point and nonpoint nitrate contamination
ronmental health, drinking water, groundwater, sources. That department is considering use of the
land-use planning, growth management, agriculture vulnerability assessments to guide requisite levels of
and natural resource conservation, management, sewage treatment and sewage disposal activities.
and protection. Key elements of the education pro- Results of the FAVA project can be applied as
cess include workshops, presentations and written a surrogate for aquifer recharge, a concept that has a
materials that minimize technical jargon while strong presence in Florida law, for example:
explaining fundamental pre-requisite concepts
such as map scale and resolution, principals of  Chapter 373.0395 F.S groundwater recharge
hydrogeology and modeling. Once this foundation areas
of knowledge is conveyed, advanced concepts such  Chapter 163.3177(6)(c) F.S. Comprehensive
as description of the specific modeling technique, planshigh recharge areas
scientific defensibility of the assessment, and  Chapter 369.315 (2) F.S.pertaining to
appropriate application of the assessment are more groundwater and springshed protection.
106 Arthur, Wood, Baker, Cichon, and Raines

Related federal-level references include the REFERENCES


Environmental Protection Agency Source Water
Assessment and Protection program and the Safe Agterberg, F. P., Bonham-Carter, G. F., Cheng, Q., and Wright,
Water Drinking Act, both of which prescribe aquifer D. F., 1993. Weights of evidence modeling and weighted lo-
gistic regression for mineral potential mapping. in Davis, J.
protection. C., and Herzfeld, U. C., eds., Computers in Geology, 25 years
Aquifer vulnerability maps also may be applied of Progress: Oxford Univ. Press, Oxford, p. 1332.
in environmental risk assessments. Murray and Alberti, L., De Amicis, M., Masetti, M., and Sterlacchini, S., 2001,
Bayes rules and GIS for evaluating sensitivity of ground-
Rogers (1999), for example note that groundwater water to contamination: Proc. Intern. IAMG Conference
vulnerability mapping provides a starting point for 2001-Cancun, Mexico, unpaginated.
quantifying anticipated environmental risk at a Aller, L., Bennett, T., Lehr, J. H., and Petty, R. J., 1985,
DRASTIC: A standardized system for evaluation of
particular site and can help identify those sites that ground water pollution potential using hydrogeologic set-
merit additional information. In lieu of specific site tings: U.S. Environmental Protection Agency, EPA/600/2-
information, this can serve as a proxy for anticipated 85/018, 63 p.
Arthur, J. D., Cichon, J. R., Baker, A. E., Wood, H. A. R., and
future clean-up costs. Additionally, this method can Rudin, A., 2003, The Florida Aquifer Vulnerability Assess-
be used by other stakeholders in the recycling of ment (FAVA) model: A tool for aquifer protection in karst
industrial sites (a.k.a. brownfield redevelopment) to settings (abst.): The 16th National Cave and Karst Manage-
ment Symposium; http://www.nckms.com/abstracts.htm, last
estimate the anticipated liability of sites (Stiber, accessed 10 August 2006.
Small, and Fischbeck, 1998). Arthur J. D., Baker A. E., Cichon J. R., Wood H. A. R., and
Several Florida county government agencies are Rudin A., 2007, Florida, Aquifer Vulnerability Assessment
(FAVA): Contamination potential of Floridas principal
preparing to initiate FAVA-WofE based on projects aquifer systems: Florida Geol. Survey Bull. 67, in review.
to support local land-use planning and groundwater Baker, A. E., Cichon, J. R., Arthur, J. D., and Raines, G. L., 2002,
protection programs. Some of these projects are Florida Aquifer Vulnerability Assessment (FAVA) (abst.):
Geol. Soc. America Abstracts with Programs, v. 34, no. 6,
underway. Integration of the FAVA maps may oc- p. 346.
cur in activities such as well-head protection, delin- Beck, B. F., and Jenkins, D. T., 1988, Potential for groundwater
eation of animal carcass disposal sites, the design of pollution of the Floridan aquifer, based upon surficial
drainage, karst development, and, overburden characteristics:
groundwater monitoring plans and public education Florida Sinkhole Research Institute, Univ. Central Florida,
to promote stewardship of groundwater resources. Map Series 87-88-1, scale 1:250,000, 15 sheets.
Moreover, WofE model outputs can produce con- Bekesi, G., and McConchie, J., 2000, Empirical assessment of the
influence of the unsaturated zone of aquifer vulnerability,
taminant-specific vulnerability maps, and may be Manawatu region, New Zealand: Ground Water, v. 38, no. 2,
coupled with groundwater flow models (i.e., p. 193199.
groundwater susceptibility models), which would Bekesi, G., and McConchie, J. A., 2002, The use of aquifer media
characteristics to model vulnerability to contamination,
provide an even more powerful tool for aquifer Manawatu region, New Zealand: Hydrogeology Jour., v. 10,
protection. p. 322331.
Berndt M. P., Oaksford, E. T., Mahon, G. L., and Schmidt, W.,
1998, Groundwater, in Fernald, E. A., and Purdum, E. D.,
eds., Water Resources Atlas of Florida: Florida State Univ.,
ACKNOWLEDGMENTS Inst. Public Affairs, 312 p.
Bonham-Carter, G. F., 1994, Geographic Information Systems for
The authors are grateful for the constructive geoscientists, modeling with GIS: Pergamon Press, Oxford,
398 p.
comments by reviewers of this paper: Richard Cichon J. R., Baker A. E., Wood H. A. R., and Arthur J. D., 2005,
Deadman, Florida Department of Community Af- Wekiva aquifer vulnerability assessment. Florida Geol. Sur-
fairs, Tim Hazlett, Hazlett-Kincaid, Inc., and Sam vey Rept. Invest. No.104, 36 p.
Connell, L. D., and van den Daele, G., 2003, A quantitative
Upchurch, SDII Global, Inc. We are also thankful approach to aquifer vulnerability mapping: Jour. Hydrology,
for invaluable input by the FAVA TAC, members of v. 276, no. 14, p. 7188.
which are listed in Arthur and others (2007). Sug- COST Action 65, 1995, Hydrogeological aaects of groundwater
protection in karstic areas final report: European Com-
gestions by Marco Masetti, Department of Earth mission, Rept. EUR 16574 EN, Directorat-General: Science,
Sciences, University of Milan, Italy also were help- Research and Development, Brussel, Luxemburg, 446 p.
ful. Financial support for this project was provided Dixon, B., Scott, H. D., Brahana, J. V., Mauromoustakos, A., and
Dixon, J. C, 2001, Application of neuro-fuzzy techniques to
by the Division of Water Resource Management, predict ground water vulnerability in northwest Arkansas:
Florida Department of Environmental Protection, Arkansas Water Resource Center, Univ. Arkansas, Publ. No.
which administered funds originating from the U.S. PUB-183, 64 p.
Doerfliger, N., Jeannin, P. -Y., and Zwahlen, F., 1999, Water
Environmental Protection Agency, Source Water vulnerability assessment in karst environments: A new
Assessment and Protection Program. method of defining protections areas using a multi-attribute
A Bayesian-based Aquifer Vulnerability Assessment in Florida 107

approach and GIS tools (EPIK method): Environmental Navulur, K. C. S., Engel, B. A., and Mamillapalli, S., 1995,
Geology, v. 39, no. 2, p. 165176. Groundwater vulnerability evaluation to nitrate pollution on
Edet, A. E., 2004, Vulnerability evaluation of a coastal plain sand a regional scale using GIS, in Applications of GIS to the
aquifer with a case example from Calabar, southeastern Modeling of Non-Point Source Pollutants in the Vadose
Nigeria: Environmental Geology, v. 45, no. 8, p. 10621070. Zone, SSSA Spec. Publ. No. 48, ASA-CSSA-SSSA Bou-
Ferguson, S., 2002, DRASTIC vs FAVAA comparison of two youcos Conference (Mission Inn, Riverside, CA) 20 p.
available methodologies for aquifer protection in Florida: A Nolan, B. T., 2001, Relating nitrogen sources and aquifer sus-
case study in Orange County: unpubl. research paper, Florida ceptibility to nitrate in shallow ground waters of the United
State Univ., 67 p. States: Ground Water, v. 39, no. 2, p. 290299.
Florida Department of Community Affairs (FDCA), Florida Raines, G. L., 1999, Evaluation of weights of evidence to predict
Department of Environmental Protection and 1000 Friends epithermal-gold deposits in the great basin of the western
of Florida, 2002, Protecting Floridas springs land-use United States: Natural Resources Research, v. 8, no. 4,
planning and best management practices, 124 p., http:// p. 257276.
www.dca.state.fl.us/fdcp/DCP/publications/springsmanual.pdf, Raines, G. L., Bonham-Carter, G. F., and Kemp, L., 2000, Pre-
last accessed 24, September 2006. dictive probabilistic modeling using ArcView GIS: ArcUser,
Florida Natural Areas Inventory, 2006, Florida Forever Conser- v. 3, no. 2, p. 4548.
vation Needs Assesment Tech. Rept., Version 2.1, 116 p., Ross, M., Martel, R., Lefebvre, R., Parent, M., and Savard, M.,
http://www.fnai.org/PDF/FF_CAN_technical_report.pdf, last 2004, Assessing rock aquifer vulnerability using downward
accessed 5, October 2006. advective times from a 3D model of surficial geology: A case
Florida Springs Task Force, 2000, Floridas Springs, in Harnett, F. study from the St. Lawrence Lowlands, Canada: Geofsica
M., ed., Strategies for Protection & Restoration, 59 p. http:// Internacional, v. 43, no. 4, p. 591602.
www.dep.state,fl.us/springs/reports/FloridaSpringsReport.pdf Rupert, M. G., 1997, Nitrate (NO2+NO3N) in ground water of
last accessed 10, September 2006. the upper Snake River Basin, Idaho and western Wyoming,
Focazio, M. J., Reilly, T. E., Rupert, M. G., and Helsel, D. R., 1991-95: U.S. Geological Survey Water Resources Invest.
2002, Assessing ground-water vulnerability to contamination: Rept. 97-4174, 47 p.
Providing scientifically defensible Information for decision Rupert, M. G., 1999, Improvements to the DRASTIC ground-
makers: U.S. Geol. Survey Circ., v. 1224, p. 33. water vulnerability Mapping method: U.S. Geol. Survey Fact
Huaming, G., and Wang, Y., 2004, Specific vulnerability assess- Sheet FS-066-99, 6 p.
ment using the MLPI model in Datong city, Shanzi province, Sawatzky, D. L., Raines, G. L., Bonham-Carter, G. F., and
China: Environmental Geology, v. 45, no. 3, p. 401407. Looney, C. G., 2004, ARCSDM3.1: ArcMAP extension for
Kemp, L. D., Bonham-Carter, G. F., Raines, G. L., and Looney, spatial data modeling using weights of evidence, logistic
C. G., 2001, Arc-SDM: Arcview extension for spatial data regression, fuzzy logic and neural network analysis. http://
modeling using weights of evidence, logistic regression, fuzzy www.ige.unicamp.br/sdm/ArcSDM31/default_e.htm, last acc-
logic and neural network analysis, http://www.ige.unicamp.br/ essed 5 October, 2006.
sdm/defaultSDM e.htm, last accessed 4, October 2006. Scott. T. M, 2002, Floridas springs in jeopardy: Geotimes,
Kincaid, T. R., Denizman, C., Arthur, J. D., and Hazlett, T., 2004, May, 2002, http://www.agiweb.org/geotimes/may02/feature_
The Florida cave database: A GIS of underwater caves for springs.html, last accessed 12 September, 2006.
hydrogeological characterizations (abst.): Geol. Soc. America Scott, T. M., Means, G. H., Meegan, R. P., Means, R. C.,
Abstracts with Programs, v. 36, no. 2, p. 85. Upchurch, S. B., Copeland, R. E., Jones, J., Roberts, T., and
Landis, J. R., and Koch, G. G., 1977, The measurement of Willet, A., 2004, Springs of Florida: Florida Geol. Survey
observer agreement for categorical data: Biometrics, v. 33, Bull. No. 66, 377 p.
p. 159174. Soller, D. R., 1992, Applying the DRASTIC Model a review of
Lawrence, F. W., and Upchurch, S. B., 1982, Identification of county-scale maps: U.S. Geol. Survey Open-File Rept. 92
recharge areas using factor analysis: Groundwater, v. 20, no. 297, 36 p.
6, p. 680687. Stiber, N. A., Small, M. J., and Fischbeck, P. S., 1998, The rela-
Marella R. L., 2004, Water withdrawals, use discharge, and trends tionship between historic industrial site use and environ-
in Florida, 2000: U.S. Geol. Survey Scientific Invest. Rept., mental contamination: Jour. Air & Waste Management
v. 20045151, 50 p. Assoc., v. 48, no. 9, p. 809818.
Masetti, M., Poli, S., and Sterlacchini, S., 2005, Aquifer vulnera- Stewart, I. T., and Loague, K., 2003, Development of type transfer
bility assessment using weights of evidence modeling tech- functions for regional-scale nonpoint source groundwater
nique: Application to the province of Milan, northern Italy: vulnerability assessments: Water Resources Research, v. 39,
Proc. IAMG05. GIS and Spatial Analysis, v. 1, p. 499504. no.12, SBH17.
Masetti, M., Poli, S., and Sterlacchini, S., 2007, The use of the Southeastern Geological Society, 1986, Hydrogeological units of
weights-of-evidence modeling technique to estimate the Florida: Florida Geol. Survey Spec. Publ. No. 28, 8 p.
vulnerability of groundwater to nitrate contamination: Nat- U.S. Census Bureau: State and County Quick Facts, 01-Feb-2005,
ural Resources Research, v.16, no.2, this issue, doi: 10.1007/ 15:48:47 EST, http://www.census.gov.
s11053-007-9045-6. Witkowski, A. J., Rubin, K., Kowalczyk, A., Ro_zkowski, A., and
Merchant, J. W., 1994, GIS-based groundwater pollution hazard Wrobel, J., 2003, Groundwater vulnerability map of the
assessment; a critical review of the DRASTIC model: Pho- Chrzanow karst-fissured Triassic aquifer (Poland): Environ-
togrammetric Engineering and Remote Sensing, v. 60, no. 9, mental Geology, v. 44, no. 1, p. 5967.
p. 11171127. Visit Florida Research, 2006, http://www.media.visitflorida.org/
Murray, K. S., and Rogers, D. T., 1999, Groundwater vulnerability, about/research/, last accessed 10, September 2006.
Brownfield Redevelopment and Land Use Planning: Jour. Vogel, M., 2006, Good Migrations: Florida Trend, http://
Environmental Planning and Management, v. 42, p. 801810. www.floridatrend.com/article.asp?id=5795, last accessed 24,
National Research Council (NRC) 1993, Ground water vulnera- September 2006.
bility assessment: Predicting relative contamination potential Winter, T. C., Harvey, J. W., Franke, O. L., and Alley, W. M.,
under conditions of uncertainty: National Acad. Press, 1998, Ground water and surface waterA single resource:
Washington, 204 p. U.S. Geol. Survey Circ., v. 1139, p. 79.

You might also like