PDF ( 6 MBs ) ( .pdf )

uf00094054-1_Page_34.txt

uf00094054bw_Page_39.txt

uf00094054-1_Page_01.txt

uf00094054bw_Page_10.txt

uf00094054bw_Page_12.txt

uf00094054-1_Page_29.txt

uf00094054-1_Page_36.txt

uf00094054bw_Page_21.txt

uf00094054bw_Page_07.txt

uf00094054-1_Page_11.txt

uf00094054-1_Page_25.txt

uf00094054-1_Page_18.txt

uf00094054-1_Page_08.txt

uf00094054-1_Page_15.txt

uf00094054bw_Page_41.txt

uf00094054-1_Page_24.txt

uf00094054-1_Page_17.txt

uf00094054-1_Page_32.txt

uf00094054bw_Page_35.txt

uf00094054-1_Page_16.txt

uf00094054-1_Page_28.txt

uf00094054-1_Page_33.txt

uf00094054bw_Page_37.txt

uf00094054-1_pdf.txt

uf00094054bw_Page_13.txt

uf00094054-1_Page_30.txt

uf00094054-1_Page_27.txt

uf00094054bw_Page_26.txt

uf00094054bw_Page_02.txt

uf00094054-1_Page_23.txt

uf00094054-1_Page_31.txt

uf00094054-1_Page_19.txt

uf00094054bw_Page_42.txt

uf00094054bw_Page_14.txt

uf00094054bw_Page_06.txt

uf00094054bw_Page_05.txt

uf00094054bw_Page_09.txt

uf00094054bw_Page_38.txt

uf00094054bw_Page_40.txt

uf00094054-1_Page_04.txt

uf00094054bw_Page_03.txt

uf00094054-1_Page_22.txt

uf00094054-1_Page_20.txt

WEKIVA AQUIFER VULNERABILITY ASSESSMENT


-~



Wittm


FLORIDA GEOLOGICAL SURVEY

REPORT OF INVESTIGATION 104










STATE OF FLORIDA
DEPARTMENT OF ENVIRONMENTAL PROTECTION
Colleen M. Castille, Secretary





DIVISION OF RESOURCE ASSESSMENT AND MANAGEMENT
Edwin J. Conklin, Director





FLORIDA GEOLOGICAL SURVEY
Walter Schmidt, State Geologist and Chief





Report of Investigation 104


WEKIVA AQUIFER VULNERABILITY ASSESSMENT





By

James R. Cichon, Alan E. Baker, Alex R. Wood, and Jonathan D. Arthur


Published for the

FLORIDA GEOLOGICAL SURVEY
Tallahassee, Florida
2005

























































Printed for the
Florida Geological Survey

Tallahassee
2005

ISSN 0160-0931




111









PREFACE


FLORIDA GEOLOGICAL SURVEY

Tallahassee, Florida
2005

The Florida Geological Survey (FGS), Division of Resource Assessment and Management,
Department of Environmental Protection, is publishing as Report of Investigation No. 104,
Wekiva Aquifer Vulnerability Assessment, authored by James R. Cichon, Alan E. Baker, Alex R.
Wood, and Jonathan D. Arthur, all staff of the FGS.

This publication is a localized offshoot of the state-wide Florida Aquifer Vulnerability
Assessment project (a Bulletin of the Florida Geological Survey, in press). Unbiased and
credible scientific assessment of the potential vulnerability of our ground water aquifers to
contamination and other surface modifications is critical to the future sustainability of Florida's
environment. State, regional, county and local governmental agencies, as well as private land
owners and the public will find this information useful in support of informed land-use planning
and natural resource conservation and preservation decisions.




Walter Schmidt, Ph.D., P.G.
State Geologist & Chief
Florida Geological Survey











TABLE OF CONTENTS


IN TR O D U C T IO N ................ ...................... .................. .................. 1
A quifer System s........................... ................ ............... 1
V uln erab ility ...................................................... .......... ...... 2
A CK N O W LED GEM EN TS ........................... .................... .......................... ..................... .............. 3
WEKIVA AQUIFER VULNERABILITY ASSESSMENT MODEL ...................................................... 4
W eights of Evidence .............. ... ........................ ................ ......... .... ... ... ... .. ................. 4
Study Area ................................................................... 4
Training Points....................................... .... .......... 6
Evidential Themes ... ................................... ............. 6
Generalization of Evidential Themes........................ ........................ 7
Soil Permeability ......................... ............. 8
Buffered Effective Karst Features .................................................... 8
Thickness of the Intermediate Aquifer System ................. .............. 10
Hydraulic Head Difference between SAS and FAS ............................ 15
R esp on se T h em e .................. ... ...................... ... ... ... ..... ..................... .. .... ........ 15
C onfi dence M ap ... ............................................ .................. ........... ... ........... ... 25
V alidation and Sensitivity Analysis.............................................. ..... .......... ............ 25
Dissolved Oxygen Data vs. Posterior Probability....................................... 28
Random 75% Subset of Training Points............... ............ ..................... 28
D iscu ssion an d Sum m ary .................................................................................. ....................... 30
D isclaim er ......... .............. .................. .. ............. .... ..... ........... ... .................... 32
W EIG H TS OF EV ID EN CE GLO SSA RY .......................................... ................................................. 33
REFEREN CES ................. ... ........ ........................ .................... .. 35

FIGURES
1. Hydrogeologic conceptual model of the W ekiva study area .......................................... ............ .... 2
2. Extent of the WAVA study area and training point set developed based on dissolved oxygen .............. 5
3. Soil perm ability m ap of the W AVA study area ............................................... .................................... 9
4. Soil permeability values (in/hr) plotted against contrast values calculated using WofE........................ 10
5. Map showing binary generalization of the soil permeability evidential theme. ................................... 11
6. Closed topographic depressions underlain by less than 100 feet (30.5 m) of FAS overburden. ............ 12
7. Map showing binary generalization of the buffered effective karst features evidential theme. ......... 13
8. Thickness grid of IAS (in feet) and the distribution of wells used ....... ........ .............. 14
9. Map showing binary generalization of the IAS thickness evidential theme....................................... 16
10. Hydraulic head difference between the SAS and the FAS potentiometric surface............................... 17
11. Map showing binary generalization of the hydraulic head difference evidential theme....................... 18
12. Relative vulnerability of the FA S FA V A m odel........................................................ .... ................. 19
13. W ofE conceptual m odel of the FA S. .............. ...................... ................... .. ................ 21
14. Posterior probability class breaks in W AVA m odel...........................................................................22
15. Relative vulnerability predicted by the WAVA model for the FAS ..........................................23
16. Relative vulnerability predicted by the FAS WAVA model showing hydrography........................... 24
17. Relative vulnerability of the WAVA model showing missing soil permeability data........................ 26
18. Distribution of confidence values calculated for WAVA response theme. ..................................27
19. Posterior probability vs. dissolved oxygen median concentration values..................... ............ 28
20. Relative vulnerability predicted by the WAVA model using a subset of training point data............. 30











TABLES
1. Test values calculated in WofE and their respective studentized T values .............. .................
2. WofE final output table listing weights calculated for each evidential theme........................................22
3. Cross-tabulation matrix of the WAVA response theme versus the 75% subset response theme .......... 29
4. Kappa coefficient values and associated interpretation (Landis and Koch, 1977). ............................. 29






REPORT OF INVESTIGATION NO. 104


WEKIVA AQUIFER VULNERABILITY ASSESSMENT

By

James R. Cichon, Alan E. Baker, P.G. 2324, Alex R. Wood, and Jonathan D. Arthur, P.G. 1149



INTRODUCTION

The Floridan Aquifer System is among the most vast and vulnerable hydrogeologic resources in the
nation. A remarkable 90% of Florida's drinking water originates from these ground-water resources
(Berndt et al., 1998). More than 700 springs are known to exist in Florida, allowing a unique window
into the aquifer systems (Scott et al., 2004). These springs have long been a resource for recreation and
fresh water. Protection of Florida's springs becomes an increasingly important issue as the population of
Florida continues to grow at a rate of almost 900 people per day (U.S. Census, 2005). As a result,
pressures on these unique natural resources are intensifying.

It is important to note that water quality of Florida's springs is a reflection of the condition of the
surrounding ground-water system and not just an indication of individual spring health. Springsheds, or
the spring's area of ground-water contribution, must also be taken into consideration for any vulnerability
assessment if proper protection of the spring itself is to take place. An assessment of the vulnerability of
the ground water, which includes the springshed, can therefore be an important tool for decision making,
development of rules or policies regarding environmental conservation, protection, growth management,
and land-use planning. To fully understand the relationship between ground water and vulnerability, an
understanding of the aquifer systems present in the study area is needed.

Aquifer Systems

The hydrogeologic framework of the Wekiva study area is comprised of three aquifer systems: the
Surficial Aquifer System (SAS), the Intermediate Aquifer System (IAS) and the FAS. According to
Florida Geological Survey Special Publication No. 28 (Southeastern Geological Society, 1986), the
Surficial Aquifer System is "the permeable hydrogeologic unit contiguous with land surface that is
comprised principally of unconsolidated to poorly indurated plastic deposits. It also includes well-
indurated carbonate rocks..." that are not part of the FAS.

The intermediate aquifer system/intermediate confining unit consists of highly-variable siliciclastic and
carbonate deposits that are relatively low-permeability, fine-grained sediments and collectively retard the
exchange of water between the overlying SAS and the underlying FAS. The term "intermediate confining
unit" applies to those areas where this unit yields little to no water, whereas the term "intermediate aquifer
system" applies to those areas where one or more low to moderate-yielding aquifers occur. Special
Publication No. 28 (Southeastern Geological Society, 1986) is currently under revision by the Florida
Geological Survey (FGS), and the forthcoming update suggests the use of the term "Intermediate Aquifer
System" for this entire unit. Moreover, the revision calls for the elimination of the use of "intermediate
confining unit." Instead, the "intermediate confining unit" is considered to be confining beds within the
IAS. This newer convention currently under review is hereby adopted in this report.

The FAS is comprised of a thick carbonate sequence of rocks that collectively function as a regional
water-yielding hydraulic unit. Where overlain by the IAS, the FAS is confined; however, sinkholes that






FLORIDA GEOLOGICAL SURVEY


breach the IAS can provide direct pathways for surface water to enter the FAS (Figure 1). Ground water
also recharges the FAS by infiltration through overlying sediments of the IAS or SAS. In areas where the
FAS is directly overlain by the SAS, the FAS may or may not contain ground water under confined
conditions, depending on the thickness and extent of low permeability material at the base of the Surficial
Aquifer System (Southeastern Geological Society, 1986).


Figure 1. Hydrogeologic conceptual model of the Wekiva study area depicting the major aquifer
systems and karst features.


Vulnerability

Due to Florida's unique hydrogeologic setting, all of Florida's ground water is vulnerable to
contamination. In fact, this statement, in a more broad sense, is considered the "First Law of Ground
Water Vulnerability" by the National Research Council (NRC, 1993) which states: "All ground water is






REPORT OF INVESTIGATION NO. 104


vulnerable." Furthermore, the NRC defines the phrase "ground-water vulnerability to contamination" as
the tendency or likelihood for contaminants to reach a specified position in the ground-water system after
introduction at some location above the uppermost aquifer. In this report, a similar definition of aquifer
vulnerability is adopted: The tendency or likelihood for contaminants to reach the top of the FAS after
introduction at land surface.

The Wekiva River Coordinating Committee Final Report identified numerous studies by Florida's water
management districts and the United States Geological Survey (USGS) that clearly demonstrate
contamination attributable to changes in land use. In other words, what occurs on the land directly and
indirectly affects the quality of water moving through the subsurface conduits. Many of these conduits
flow to surface-water bodies via springs.

The Florida Geological Survey (FGS) applied a modeling technique used in the statewide Florida Aquifer
Vulnerability Assessment (FAVA) project to the Wekiva study area in order to delineate relative degrees
of vulnerability for the FAS based on available data. This project is known as the Wekiva Aquifer
Vulnerability Assessment model or WAVA. The modeling technique used is known as Weights of
Evidence (WofE), which is based entirely in a geographic information system (GIS) and is summarized
below. For a more thorough explanation of this method refer to FGS Bulletin 67, Florida Aquifer
Vulnerability Assessment: Contamination potential of Florida's principal aquifer systems (Arthur et al.,
2005, in preparation). The relative vulnerability zones identified in the WAVA model may be applied to
facilitate designation of "protection zones" by regulatory agencies. These "protection zones," however,
would likely include other physical and chemical (i.e., water quality) factors affecting the hydrologic
system, including ground-water flow rates and patterns and proximity to impacted areas.

The software used to execute WofE and the Florida Department of Environmental Protection (FDEP)
Albers projection require certain data layers to be listed in meters. However, some data layers are listed
in units standard to that particular data set. For reader convenience we list the primary unit followed by
the alternative unit in parentheses.


ACKNOWLEDGEMENTS

The authors are grateful for the work of the Wekiva River Basin Coordinating Committee (WRBCC),
which was appointed by the Governor in July, 2003. This committee follows a series of legislation and
Governor Executive Orders, all with the ultimate goal of protecting surface- and ground-water resources
of the Wekiva River system. The research presented herein was funded by the Florida Springs Initiative
through the FDEP, and was facilitated by the support of Vivian Garfein, Director, FDEP Central District,
Richard Drew, Bureau Chief, FDEP Water Facilities Regulation and Mike Bascom, FDEP Springs
Coordinator. We are thankful for the insightful review comments provided by individuals listed above,
including members of the WRBCC, as well as Richard Deadman, Florida Department of Community
Affairs. Moreover, we gratefully acknowledge the following individuals for providing technical and/or
editorial review of the report: Chris Ferraro, Lou Ley and Alan Stodghill (FDEP), William Pollock and
Dr. Eberhard Roeder, Florida Department of Health, and several FDEP/Florida Geological Survey staff,
including Carol Armstrong, Dr. Rick Copeland, Tom Greenhalgh, Jackie Lloyd, Frank Rupert, Dr. Tom
Scott and Dr. Walt Schmidt.






FLORIDA GEOLOGICAL SURVEY


WEKIVA AQUIFER VULNERABILITY ASSESSMENT MODEL

Weights of Evidence

Use of WofE requires the combination of diverse spatial data which are used to describe and analyze
interactions and generate predictive models (Raines et al., 2000). When applied in the WAVA project,
WofE was used to generate maps of aquifer vulnerability, or response themes. These response themes
were generated in the Environmental Systems Research Institute ArcView 3.2 environment. WofE was
executed using the Arc Spatial Data Modeler which is available as an internet download (Kemp et al.,
2001).

A primary benefit of applying WofE to the WAVA project is that it is data-driven, rather than expert-
driven. The data that "train" the model consist of known occurrences of parameters (water quality values)
that reflect relative aquifer vulnerability, such as high levels of dissolved oxygen in ground-water wells.
These wells were the training points used to calculate weights for laterally continuous input data layers, or
evidential themes, which are then combined to yield a response theme.

When reviewing the model results, it is important to note that all aquifers, to some degree, are vulnerable
to contamination from land surface. The model results simply identify those areas within the study area
that are more vulnerable or less vulnerable based on the evidential themes and training points used in the
model. So, the vulnerable areas are assigned values relative to each other within the study area. These
"relative" values are not directly transferable or correlated to another study area.

Study Area

The initial step in the development of the WAVA model was the delineation of a study area extent. The
study area, comprised of 30 square meters (m2)(322.9 ft2) grid cells, was used in the calculation of
weights and probabilities throughout the modeling process as described below. The extent of the study
area used for input into the WAVA model was based on a combination of the Wekiva study area (as
identified in the Wekiva River Basin Coordinating Committee, 2004) and the area contributing ground
water to the Wekiva River (Figure 2).

Large water bodies were omitted from the WAVA model because a well would never be drilled in these
areas therefore, they would never contain a training point. Additionally, if lakes were left in the model,
the surface area was increased with no chance of increasing the number of training points. This would
unnecessarily bias the model. Further, large water bodies typically have no soils or other input data
associated with them, thus the model output omits these areas due to lack of data or potential bias in the
calculated probabilities. Removal of large water bodies does not change the relative results (i.e., relative
vulnerability) of the model; however, it would have the effect of increasing weights calculated during the
modeling process. Regardless, the model is more valid and defensible by removing the large water
bodies.

Due to the geostatistical framework and evidential layers (spatial hydrogeological data) of WAVA,
aquifer systems in discharge areas were sometimes predicted by the output model to be low in
vulnerability, even though the discharging surface waters are highly vulnerable to contamination. These
discharging surface waters are not part of the aquifer, although they can originate from it. The WAVA
project was designed to focus on the ability for a contaminant to travel through soils, overburden, karst
features, etc. to enter into the aquifer system. As a result, it is very important that the WAVA model
never be applied to assess contamination of surface waters or discharge areas.







REPORT OF INVESTIGATION NO. 104


U


Leesburg

* K


LAKi.

'CY


0.


U


Training points
Wells sampled for dissolved oxygen
Springs
SWekiva study area
Area contributing groundwater
to the Wekiva River (Source: SJRWMD)
SWAVA study area

Water features


Enlarged Area


N
F~A


10 5 0


10 Miles

10 Kilometers
U


Figure 2. Extent of the WAVA study area and training point set developed based on dissolved
oxygen measured in wells. Large water bodies have been omitted from the analysis to avoid biasing
the model (see text). Wekiva River Basin Coordinating Committee (WRBCC; March 16, 2004).






FLORIDA GEOLOGICAL SURVEY


Training Points

In WofE, training points are locations of known measured occurrences (i.e., locations where quantitative
data have been obtained). In mining applications for example, identified ore reserves are known
occurrences (training points). In an aquifer vulnerability assessment, wells with water quality data
indicative of high recharge (i.e., dissolved oxygen, nitrates above an identified background threshold,
tritium) are potential known occurrences (i.e., areas where a good connection exists between the top of
the aquifer and land surface). Training points are used in WofE to calculate the following parameters (see
Weights of Evidence Glossary): prior probability, weights for each evidential theme, and posterior
probability of the response theme. Training points are converted to represent a unit area (i.e., a 30 m2 grid
cell) of the study area within a GIS application.

The water-quality parameter selected for the WAVA training data set was dissolved oxygen. In general,
native ground-water concentrations of dissolved oxygen in the FAS are typically low where the aquifer
system is not affected by activities at land surface or active recharge. Therefore, where dissolved oxygen
occurs at concentrations above background levels in an aquifer system, it can generally be assumed a
relatively greater hydrologic connection exists between land-surface activities and ground water.
Dissolved oxygen analytical values served as the training point set for the WAVA model.

More than 200 wells completed in the FAS for the WAVA study area were compiled from the FDEP
Background Water Quality Monitoring Network, the FDEP STATUS network, and the St. John's River
Water Management District monitoring network. Of these wells, 28 were measured for dissolved oxygen.
Using statistical methods, described in Arthur et al. (2005, in preparation), two wells were identified as
outliers and removed from the dataset leaving 26 wells for further analysis (Figure 2). The 75th percentile
value was the statistical break used to identify relatively high dissolved oxygen concentrations in ground
water, which is considered herein as a proxy for areas of higher recharge and vulnerability. Median
dissolved oxygen concentrations for wells with multiple analyses were calculated. These median values,
along with data from wells with one dissolved oxygen analysis were then used to calculate a grand
median value. The 75th percentile combined median value for dissolved oxygen concentration was
0.4 mg/L. Six wells existed in the dataset that had a measured median dissolved oxygen value greater
than the 0.4 mg/L threshold. These six wells comprised the training point theme for input into the
WAVA model (Figure 2).

The probability that a training point will occupy a defined unit area within the study area, independent of
any evidential theme data, is known as prior probability. The prior probability is calculated by dividing a
unit area representing the training points by the total study area. In less complex terms, the prior
probability is based on prior knowledge of the problem without the benefit of supporting evidence. The
prior probability for the WAVA model was calculated at 0.0017. The distribution of all wells sampled for
dissolved oxygen, and the wells meeting training point criteria are displayed in Figure 2.

Evidential Themes

Several evidential themes were considered for use in the WAVA model due to their potential influence on
ground-water quality. They are:

Soil permeability
Buffered effective karst features (based on closed topographic depressions)
Thickness of Intermediate Aquifer System (IAS)
Hydraulic head difference between Surficial Aquifer System (SAS) and FAS
Vertical leakage rate to and from the FAS






REPORT OF INVESTIGATION NO. 104


Vertical leakage rate to and from the FAS ultimately was not used because it was based on model
simulations with a grid-cell size of 1,000 meters and there was large variability in the calculated leakage
rate between adjacent grid cells. Weights were calculated for each of the other four evidential themes and
generalized as discussed below.

Generalization of Evidential Themes

Evidential themes were generalized in an effort to assess which areas of the evidence shared a greater
association with locations of training points. During calculation of weights for each evidential theme used
in the WAVA project, a contrast value was calculated for each class of the theme by combining the
positive and negative weights (positive weight negative weight). Contrast is a measure of a theme's
significance in predicting the location of training points and helps to determine the threshold or thresholds
that maximize the spatial association between the evidential theme map pattern and the training point
theme pattern (Bonham-Carter, 1994). Arthur et al., (2005, in preparation) provide a more comp-
rehensive explanation of theme generalization.

Confidence of the evidential theme equals the contrast divided by the standard deviation (a student T-test)
for a given evidential theme and provides a useful measure of significance of the contrast due to the
uncertainties of the weights and areas of possible missing data (Raines, 1999). A confidence value of
1.282 corresponds to a 90% level of significance (see Table 1), which was the value selected as the
minimum acceptable confidence level for the WAVA project evidential themes. Confidence values
approximately correspond to the statistical levels of significance listed in Table 1.

Table 1. Test values calculated in WofE and their respective studentized T values expressed as level
of significance in percentages.




99.5% 2.576
99% 2.326
97.5% 1.960
95% 1.645
90% 1.282
80% 0.842

75% 0.674
70% 0.542

60% 0.253


Contrast values were used to determine where to sub-divide evidential themes into generalized categories.
The most common method of categorizing an ordered evidential theme was to select the maximum
contrast as a threshold value to create a binary generalized evidential theme. In some WofE models,
categorization of more than two classes may be justified (Arthur et al., 2005, in preparation). For
evidential themes used for the WAVA project, a binary break was typically defined by the WofE analysis
thereby creating two spatial categories: one with stronger association with the training point theme and
one with weaker association with the training point theme.






FLORIDA GEOLOGICAL SURVEY


Soil Permeability

Soil permeability values were obtained from the soils statewide grid layer developed for FAVA (Arthur et
al., 2005, in preparation), which was compiled from the United States Department of Agriculture -
Natural Resources Conservation Service Soil Service Geographic (SSURGO) data set. This database
was mapped at a scale of 1:24,000 and contains the most detailed soil permeability data. However,
different permeability values can exist along some county boundaries due to the different mapping
techniques used by different soil scientists. The development of this layer included the calculation of
average soil permeability values for each soil horizon layer. Then, based on soil horizon thicknesses,
weighted-average permeability values were calculated for the entire soil column. This allowed the
generation of a data coverage of soils containing a single permeability value per soil polygon. This data
coverage is displayed in Figure 3. Missing-data areas on the figure represent urban land, pits and water
features.

Soil permeability is a measure of the rate at which water travels through the vadose (unsaturated) zone.
Areas with high soil permeability values are normally associated with higher aquifer vulnerability.
Weights were calculated for soil permeability using the cumulative descending method (refer to Weights
of Evidence Glossary) of the WofE model technique. The highest contrast of any class was calculated at
13.1 in/hr (9.2 x 10-' cm/sec) as displayed in Figure 4. Based on the calculated weights, the most
appropriate break in the soil permeability evidential theme was at 13.1 in/hr (9.2 x 103- cm/sec), creating a
binary generalized theme for input into the WAVA model. This contrast break indicated that values
exceeding 13.1 in/hr (9.2 x 103- cm/sec) were strongly correlated with aquifer vulnerability as defined by
the training point data, whereas values less than 13.1 in/hr (9.2 x 103- cm/sec) were less significant. The
generalized theme is displayed in Figure 5.

Buffered Effective Karst Features

To represent karst features (i.e., sinkholes) affecting vulnerability in the WAVA model study area, an
effective karst GIS grid was generated based on a combination of closed topographic depressions
identified in USGS 7.5-minute quadrangle maps and thickness of overburden on FAS (Figure 6). Closed
topographic depressions (considered an estimation of karst feature distribution) are represented on USGS
7.5-minute maps as enclosed hachured contour lines. The effective karst feature evidential theme was
developed by evaluating which of these closed topographic depressions were more likely to be
hydrologically connected to the underlying FAS. This was accomplished by filtering out depressions
underlain by more than 100 feet (30.5 m) of FAS overburden thickness. The 100-ft (30.5 m) threshold of
overburden thickness has been used to identify karst-prone areas by Cichon et al. (2004) and Wright
(1974). Though the location of training points was not used to select this filter threshold, the lack of their
occurrence in areas underlain by more than 100 feet (30.5 m) of FAS overburden thickness lends support
to the use of this filter.

Areas nearer to an effective karst feature are normally associated with higher aquifer vulnerability due to
the increased chance of overland flow and the infiltration into the depression. Therefore, buffer zones of
3,600-m (11,811 ft) divided into 30-m (98.4 ft) intervals were generated around each karst feature, and
weights were calculated for the effective karst feature evidential theme using the cumulative ascending
method. The highest contrast of any class was calculated at a distance of 60 m (196.9 ft) from a
depression creating a binary generalized theme for input into the WAVA model. In other words, the
analysis indicated that this threshold of proximity to karst features maximized the spatial association
between the map pattern and the training point pattern. The generalized theme is displayed in Figure 7.






REPORT OF INVESTIGATION NO. 104


u011 rermeauiiity ininr) N
S|0-1.5 10.5- 13.0 W $ E
1.5- 4.5 13.0- 20.0 s
S4.5- 7.0 No Data 10 0 10 Miles
7.0 9.5 Water Features 10 0 10 Kilometers
9.5- 10.5
Figure 3. Soil permeability map of the WAVA study area (Florida Geographic Data Library, 2003).
Note differences across county lines due to different mapping techniques. Missing-data areas
represent urban land, pits and water features.







FLORIDA GEOLOGICAL SURVEY


2.00
---.----- --- 1 1



1.50



1.00



0.50



0.00



-0.50
---------------------- -----------------------0 .0 0



--------------------------------------------- *-0.50
25.0 20.0 15.0 10.0 5.0 0.0

Soil Permeability (in/hr)

Figure 4. Soil permeability values (in/hr) plotted against contrast values calculated using WofE.
The highest contrast value was calculated at 13.1 in/hr (9.2 x 10-3 cm/sec).





Thickness of the Intermediate Aquifer System

For the purposes of the WAVA project, "Intermediate Aquifer System," or IAS, refers to the statewide
hydrostratigraphic unit that provides variable confinement to the FAS. The initial step in the
identification of the IAS was to create a database of well core and cuttings data. The well database was
then used to create a hydrostratigraphic surface for the top of the IAS and the top of the FAS (which
coincides with the base of the IAS). Following creation of these surface models, it was necessary to
resolve the IAS with land-surface elevation in localized areas where the IAS surface interpolation
extended above land surface. After the hydrostratigraphic surfaces were developed and refined,
calculation of a thickness map was completed by subtraction of the IAS surface from the FAS surface.
The final output yielded a continuous thickness map of the IAS; however the map could not take into
account areas of zero IAS thickness due to the grid calculation. As a result, based on data from more than
24 wells within and immediately adjacent to the study area, an approximate extent of the IAS was
delineated. These areas, which are located within the northeastern and southwestern part of the WAVA
study area are depicted on Figure 8 as "IAS thin to absent" and are juxtaposed against the calculated grid
results.






REPORT OF INVESTIGATION NO. 104


N
Soil Permeability (in/hr) w
0.1 13.1 s
S 13.1 -20.0 10 0 10 Miles
Water Features 10 0 10 Kilometers


Figure 5. Map showing binary generalization of the soil permeability evidential theme, based on a
contrast value of 13.1 in/hr (9.2 x 10-3 cm/sec).








FLORIDA GEOLOGICAL SURVEY


q;r


. -


4.

-V


- -N


S ,t~



;t
,Yc
%'t._


b~b


Effective Karst Features

S Closed topographic
Depressions

Water Features


~IV.
$.ra >t

I~ 9:
c -. -t


N



S
10 0 10 Miles

10 0 10 Kilometers


Figure 6. Closed topographic depressions underlain by less than 100 feet (30.5 m) of FAS
overburden.









12


V





I;0


r 0.







REPORT OF INVESTIGATION NO. 104


Buffered Effective Karst
Features (meters)

0 -60
> 60
Water Features


s
0 5 0 10 Miles


10 5 0


10 Kilometers


Figure 7. Map showing binary generalization of the buffered effective karst features evidential
theme based on a contrast buffer distance of 60 m (196.9 ft).


I







FLORIDA GEOLOGICAL SURVEY


S


/


Legend
IAS Thickness Grid
m IAS thin to absent
1 -20
F-20- 40
= 40- 60


II 60 80
S80 100
100 120
120 140
Water Features


* Wells completed in top of FAS and ICU
Wells completed in top of ICU only
* Wells completed in top of FAS only


N
WSE
S


0 0 10 Miles
10 0 10 Kilometers


Figure 8. Thickness grid of IAS (in feet) and the distribution of wells used. Data from FGS
lithologic database combined with geophysical well logs (Davis et al., 2001) were utilized to develop
this coverage. See text for explanation of "IAS thin to absent" areas.






REPORT OF INVESTIGATION NO. 104


Areas underlain by thinner IAS sediments are normally associated with higher aquifer vulnerability.
Weights were therefore calculated for the IAS evidential theme using the cumulative ascending method.
The highest contrast of any class was calculated at a thickness of 48 feet (14.6 m) creating a binary
generalized theme for input into the WAVA model (Figure 9). In other words, the analysis indicated that
this threshold of IAS thickness maximized the spatial association between the map pattern and the
training point pattern.

The best available data indicated that the area including Wekiwa Spring is underlain by 50 feet (15.2 m)
or more of IAS thickness. This data included well core and cuttings data and geophysical (gamma) well
logs from FGS lithologic database and FGS Special Publication No. 50 (Davis et al., 2001). As a result,
for this analysis, the spring is located outside the area which is strongly correlated with aquifer
vulnerability as defined by the training point data (i.e., greater than 48 feet or 14.6 m thick). In fact, the
exposed rocks within the spring pool are the Coosawhatchie Formation of the Hawthorn Group which is
part of the IAS (Harley Means, personal communication, 2004).

Hydraulic Head Difference between SAS and FAS

Hydraulic head difference was calculated by subtracting the FAS 1993-1994 potentiometric surface
(Sepulveda, 2002) from the water-table surface (Arthur et al., 2005, in preparation). Although more
recent potentiometric surfaces for the FAS have been published, the data control in Sepulveda's (2002)
map was more dense and the potentiometric surface was more highly resolved. Areas with a positive
hydraulic head difference value indicated that the FAS is receiving recharge, whereas areas with a
negative value indicated the FAS is discharging to the overlying aquifer system (Figure 10).

Areas of greater hydraulic head difference between the SAS (water table) and FAS indicate greater
potential for downward recharge to the FAS, which is normally associated with higher aquifer
vulnerability. Weights were therefore calculated for the hydraulic head difference evidential theme using
the cumulative descending method. The highest contrast of any class was calculated at a hydraulic head
difference value of 17 ft (5.2 m) creating a binary generalized theme for input into the WAVA model. In
other words, the analysis indicated that this threshold of hydraulic head difference maximized the spatial
association between the map pattern and the training point pattern. Figure 11 displays the binary
generalization used for the hydraulic head difference evidential theme.

Response Theme

The WAVA response theme is an output map, calculated using WofE, showing the probability (posterior
probability) that a unit area is vulnerable to contamination from land surface based on the evidence
provided. A response theme is portrayed as relative vulnerability and is classified based on the inflections
in a chart in which cumulative study area is plotted against posterior probability. The chart may allow for
the selection of several classes of posterior probability. The more vulnerable areas correspond with
higher posterior probabilities, while the less vulnerable areas are associated with lower posterior
probabilities. In essence, a higher posterior probability indicates that an area is more likely to contain a
training point, or more likely to be contaminated, and therefore more vulnerable to contamination from
land surface.

As described in the Introduction, the WAVA model was based on the modeling technique used in the
FAVA project. The FAVA project identified relative vulnerability of Florida's principal aquifer systems
and consolidated the results into three classes (more vulnerable, vulnerable and less vulnerable). The
FAVA response theme clipped to the Wekiva study area (Figure 12) shows that the majority of the






FLORIDA GEOLOGICAL SURVEY


N
IAS Thickness (feet) we E
S 0-48 s
> 48 10 0 10 Miles
Water Features 10 0 10 Kilometers

Figure 9. Map showing binary generalization of the IAS thickness evidential theme based on a
contrast value of 48 ft (14.6 m).






REPORT OF INVESTIGATION NO. 104


N
Hydraulic Head Difference (feet) W* E
-32--12 50 70 s
-12- 10 70 90 10 0 10 Miles
10-30 90-112 10 0 10 Kilometers
I 130 50 Water Features

Figure 10. Hydraulic head difference between the SAS (Arthur et al., 2005, in preparation) and the
FAS potentiometric surface (Sepulveda, 2002).







FLORIDA GEOLOGICAL SURVEY


Head Difference (feet)
17-112
-32 17
Water Features


S
0 5 0 10 Miles


10 5 0


10 Kilometers


Figure 11. Map showing binary generalization of the hydraulic head difference evidential theme
based on a contrast value of 17 ft (5.2 m).


I







REPORT OF INVESTIGATION NO. 104


Relative Vulnerability
SMore Vulnerable
Vulnerable
Wekiva Study Area
I | Water Features


N


S
0 50 Miles
0 5 0 10 Miles


10 5 0


10 Kilometers


Figure 12. Relative vulnerability of the FAS FAVA model (Arthur et al., 2005, in preparation)
showing zones of vulnerability based on the extent of the Floridan Aquifer System.


I






FLORIDA GEOLOGICAL SURVEY


Wekiva study area is located in the "more vulnerable" zone. A smaller part of the study area is located in
the "vulnerable" zone and none of the study area is located in the "less vulnerable" zone. By applying the
FAVA modeling technique to the Wekiva study area (i.e., applying the WAVA model) it was possible to
delineate refined zones of relative vulnerability by considering only the range of hydrogeological
conditions in the study area as opposed to statewide hydrogeological conditions. Although the Wekiva
study area should be considered "vulnerable" to "more vulnerable" in the context of statewide FAS
vulnerability, the WAVA model refines these two degrees of vulnerability into three classes.

Using the four evidential themes discussed above, a response theme was generated showing the posterior
probability that a unit area contained a training point based on the evidential themes provided. A
conceptual model showing the association between the training points and the evidential themes is shown
in Figure 13. The posterior probabilities of the response theme ranged from 0.0007261 to 0.1042028
across the model domain. Plotting posterior probability against cumulative area as a percentage (Figure
14) allowed the delineation of class breaks for display of protection zones in the final response theme
(Figure 15). The breaks for these protection zones were selected where a notable stepwise increase in
posterior probability relative to the cumulative study area occurred. The first break, which delineated the
less vulnerable zone from the vulnerable zone, occurred at a posterior probability value of 0.0000990.
The less vulnerable zone represents approximately 25.7% of the study area. The second break delineating
the vulnerable zone from the more vulnerable zone occurred at the next significant stepwise increase in
posterior probability at a value of 0.0017000, which corresponded with the prior probability. This
vulnerable zone represents approximately 55.8% of the study area. The remainder of the study area
includes the more vulnerable zone and represents approximately 18.4% of the study area. This more
vulnerable zone contains the greatest probability of containing a training point, and is therefore
considered the most vulnerable to contamination from land surface.

For the purposes of this project, three vulnerability classes were delineated; however, based on the
methodology used to select these classes, a third probability break could be delineated, representing less
than 3% of the study area. This area would represent the most vulnerable part of model study area and
would be located principally in the "more vulnerable" area defined in the FAVA study.

The response theme indicated that the areas of highest vulnerability (high probabilities) tended to be
associated with IAS sediments that are thin to absent, dense karst-feature distribution, high positive
hydraulic head difference, and high soil permeability. Conversely, areas of lowest vulnerability (low
probabilities) tended to be determined by thick IAS sediments, sparse karst-feature distribution, hydraulic
head difference less than 17 feet (5.2 m) and low soil permeability values.

In addition to the large surface-water bodies present in the study area, a multitude of other surface-water
features occur (Figure 16). Many of these features represent areas of discharge. It is important to note
that these discharging surface waters are not part of the aquifer, although they originate from it. The
WAVA project was designed to predict the relative ability for a contaminant introduced at land surface to
travel through soils, overburden, karst features, etc. to enter into the FAS. As a result, it is essential that
the WAVA model never be applied to assess contamination of surface waters or discharge areas.

As described earlier, an assumption is made when using WofE that there is conditional independence
between the layers used as predictors. Conditional independence is violated when the presence of one
evidential theme influences the probability of another evidential theme. The validity of posterior
probability values depends upon the degree of conditional independence calculated for the model.






REPORT OF INVESTIGATION NO. 104


Soil Permeabiliti






Karst Features -





IAS Thickness




Hydraulic Head
Difference




WAVA
Response Theme
















Figure 13. WofE conceptual model of the FAS. The top four layers are evidential themes and the
bottom layer is the response theme divided into zones of relative vulnerability. Yellow lines
represent training points (wells) projected throughout the layers.







FLORIDA GEOLOGICAL SURVEY


cc 001000
0


I
0 -' 0001-
I-
0001001



0000100 4 0 5 0 6 0 7 0 8 0 9 0
0 0000Uc



000001
000 1000 2000 3000 4000 5000 6000 7000 8000 9000 10000

Cumulative Area (%)
Figure 14. Posterior probability class breaks (green dashed lines) representing the boundaries
between zones of relative vulnerability. A fourth optional break occurs at a posterior probability of
-0.017, which would represent less than 3% of the study area.


Evidential themes are considered independent of each other if the conditional independence value is
around 1.00. For the WAVA project, appropriate conditional independence values fell within the range of
1.00 0.15 (Gary Raines, personal communication, 2003). Conditional independence was calculated at
1.15 indicating minimal dependence between evidential themes.

Weights calculated for the evidential themes used in the WAVA model are included in Table 2. This
table displays the evidential themes used, weights calculated for those evidential themes, as well as the
contrast and confidence of the evidential theme. A positive weight indicates areas where training points
were likely to occur, while a negative weight indicates areas where training points were not likely to
occur. The contrast column is a combination of the highest and lowest weights (positive weight -
negative weight) and is a measure of how well the generalized evidential themes predict training points. A



Table 2. WofE final output table listing weights calculated for each evidential theme and their
associated contrast and confidence values of the evidential themes.



Hydraulic Head Difference 0.7412 -1.2857 2.0269 1.8494

Buffered Effective Karst 1.3546 -0.5550 1.9096 2.3343

Soil Permeability 1.2564 -0.5396 1.7960 2.1955

IAS Thickness 0.8739 -0.7726 1.6465 1.8994







REPORT OF INVESTIGATION NO. 104


Relative Vulnerability
SMore Vulnerable W E
Vulnerable s
Less Vulnerable 10 5 0
Wekiva Study Area 10 5 0
I I Water Features


10 Miles


10


SKilometers


Figure 15. Relative vulnerability predicted by the WAVA model showing more vulnerable,
vulnerable and less vulnerable areas of the FAS.







FLORIDA GEOLOGICAL SURVEY


Relative Vulnerability
SMore Vulnerable Hydrography W E
Vulnerable s
Less Vulnerable 10 5 0


Wekiva Study Area
I I Water Features


10 5 0


Figure 16. Relative vulnerability predicted by the WAVA model showing more vulnerable,
vulnerable and less vulnerable areas of the FAS. Hydrography represents seasonal and permanent
wetlands, ponds and small lakes.


10 Miles


10


SKilometers


II






REPORT OF INVESTIGATION NO. 104


positive contrast that is significant, based on its confidence, suggests that a generalized evidential theme
is a useful predictor. The confidence of the evidential theme is the contrast divided by its standard
deviation and provides a useful measure of significance of the contrast because of the uncertainties of the
weights and areas of missing data (Raines, 1999).

The hydraulic head difference evidential theme had a stronger association with the training points (i.e.,
highest contrast) than the other evidential themes and was therefore the primary determinant in predicting
areas of vulnerability in this model. Three of the evidential themes (buffered effective karst, confining
unit thickness and soil permeability) indicated where training points were more likely to occur because
the positive weights (W1) were stronger (had a greater absolute value) than the negative weights (W2).
Confidence values, for each evidential theme, exceeded the target value of 1.282, which corresponds to a
level of significance of approximately 90% (Table 1). In fact, all confidence values for the evidential
themes exceeded the 95% level of significance (1.645). For the dissolved oxygen response theme all of
the training points occurred within areas of higher vulnerability, thus supporting the results of the model
output.

The soil permeability evidential theme contains large areas of missing data particularly along the Orange
and Seminole county border. To compensate for this in the statewide FAVA model, a nearest neighbor
GIS function was employed, which was used to apply spatial statistics (Euclidean distance functions) to
predict soil permeability data values for these areas. Although this worked well on a statewide scale, use
of this process on a smaller area skewed the vulnerability where there were large areas of missing data. In
the WAVA model for example, use of this nearest neighbor function caused these areas along the Orange
and Seminole county border to be included in the more vulnerable zone of the response theme. However,
if the nearest neighbor function was not used, the calculated response theme included these same areas in
the vulnerable zone (Figure 17). Based on the uncertainty due to missing data, the nearest neighbor
function was not used in the WAVA model. The response theme values in these areas were therefore
based solely on IAS thickness, buffered effective karst features, and hydraulic head difference. As a
result, the relative vulnerability in these areas of missing data has a higher degree of uncertainty when
compared to other areas of the model, and as new data becomes available, the vulnerability of these areas
can be re-assessed.

Confidence Map

There are two types of confidence used on the WofE model. Confidence of the evidential theme, as
reported in Table 2 and discussed above in Generalization of Evidential Themes, equals the contrast
divided by the standard deviation (a student T-test) for a given evidential theme (Raines, 1999). The
second type of confidence can be calculated for each response theme by dividing the theme's posterior
probability by its total uncertainty (standard deviation). A confidence map can be generated based on
these calculations. The confidence map for the WAVA response theme is displayed in Figure 18.

Areas with a high posterior probability tend to have higher confidence values and therefore have a higher
level of certainty with respect to predicting aquifer vulnerability. A small population of training points
along with missing data raises the total uncertainty for the response theme which in turn lowers the
confidence. A confidence (of posterior probability) map containing all evidential themes (Figure 18) can
therefore contain a lower level of significance than those calculated for each separate evidential theme.

Validation and Sensitivity Analysis

For other aquifer vulnerability assessments completed by the FGS (Arthur et al., 2005, in preparation;
Baker et al., 2005) models were validated by creating a training point theme based on another parameter
that reflected vulnerability yet was independent of the original training point set. As mentioned above, the







FLORIDA GEOLOGICAL SURVEY


Relative Vulnerability
SMore Vulnerable No Soil Permeability Data W E
Vulnerable s
Less Vulnerable 10 5 0


Wekiva Study Area
I I Water Features


10 5 0


Figure 17. Relative vulnerability predicted by the WAVA model showing areas where no soil
permeability data exists. The relative vulnerability predicted in these areas could change (i.e.,
higher or lower) as new data becomes available.


10 Miles


10


SKilometers


II







REPORT OF INVESTIGATION NO. 104


Confidence

I 70%- 75%
I 75%- 80%
M 80% 90%
SI Water Features


N
W$E


0


10 5 0


Figure 18. Distribution of confidence values calculated for WAVA response theme.


10 Miles


10 Kilometers


I


I


~ ~`~

i


r"*iflr'f
Y'
\*'I







FLORIDA GEOLOGICAL SURVEY


WAVA response theme, based on calculated weights, was a better predictor of where training points were
more likely to occur because the positive weights (W1) were stronger than the negative weights (W2). In
contrast, results from using a dissolved nitrogen training point set as validation for the WAVA model,
indicated where training points were less likely to occur because the negative weights (W2) were stronger
than the positive weights (W1). This produced a different kind of model, and, as a result, was not
comparable to the original WAVA response theme. The following methods, however, did prove effective
in completing reliable validation and sensitivity analyses: comparison of dissolved oxygen and posterior
probability, and comparison of the WAVA model with a model based on a subset of training-point data.

Dissolved Oxygen Data vs. Posterior Probability

Posterior probability values were compared with the dissolved oxygen dataset from which the training
point theme was extracted. Dissolved oxygen median concentration values were plotted versus posterior
probability values (Figure 19). As expected a positive trend was observed between posterior probability
and dissolved oxygen values and a calculated Pearson's correlation coefficient (r) of 0.689 indicates more
than a 99% degree of statistical significance.


0.025



0.020



" 0.015

g

S0.010
0


0.005


0.000 4-
0.00


0.20 0.40 0.60 0.80 1.00 1.20 1.40
Dissolved Oxygen (mg/L)


Figure 19. Posterior probability vs. dissolved oxygen median concentration values

Random 75% Subset of Training Points

If the WAVA evidential themes and training points were robust (i.e., not sensitive to subtle changes in the
training data set), one would expect the response theme patterns for the full training data set and a subset
to be similar. For this sensitivity test, a training point theme consisting of a random subset of 75% of the
original training points was generated and the model was re-executed. Response themes generated for the
random subset of points were broken into the three relative vulnerability zones using the methodology
described in the Response Theme section. The subset response theme was used to evaluate the degree of
correlation between the WAVA response theme and the response theme developed using the 75% subset







REPORT OF INVESTIGATION NO. 104


of training points. This random subset yielded a prior probability of 0.0014. Weights were then
recalculated for each evidential theme, class breaks were selected, and a response theme was generated
(Figure 20). The pattern of posterior probabilities was nearly identical to the original total dissolved
oxygen response theme.

In order to compare quantitatively the WAVA response theme to the random 75% subset response theme,
the kappa coefficient (Cohen, 1960) was applied. This parameter is used to measure the amount of spatial
agreement between response themes while taking into account agreement that could have occurred by
chance.

Cross-tabulation matrices were used to calculate the kappa coefficient (see also Weights of Evidence
Glossary). Values along the diagonal in the initial matrix (Table 3; see bold numbers) reflect the amount
of agreement between response themes by area (in square meters). The remaining values in the table
reflect where the response themes were mismatched. The matrix allows further matrix calculations of
observed and expected proportions (i.e., agreement). The kappa coefficient between the two response
themes equals 0.979 and less than 2% of the study area is mismatched.

Kappa coefficient results can range between -1 (perfect disagreement) and 1 (perfect agreement). A value
of zero indicates that the agreement was no better than that expected due to chance (Bonham-Carter
1994). Kappa coefficients calculated in the WAVA project were all positive values. Positive kappa
coefficients can be interpreted using Table 4. Based on results of the kappa coefficient, the two response
themes are in "almost perfect agreement." This finding, as well as the correlation between dissolved
oxygen and posterior probability provides strong validation for the WAVA model.

Table 3. Cross-tabulation matrix of the area in square meters per class of the WAVA response
theme and 75% subset response theme. Values along the diagonal reflect the amount of agreement.



608,123,700 43,521,300 0 651,645,000

0 1,974,464,100 0 1,974,464,100

0 0 910,533,600 910,533,600

608,123,700 2,017,985,400 910,533,6001 3,536,642,700

Table 4. Kappa coefficient values and their associated interpretation (Landis and Koch, 1977).


Kappa Interpretation
< 0 No agreement

0.0 0.19 Poor agreement

0.20 0.39 Fair agreement

0.40 0.59 Moderate agreement

0.60 0.79 Substantial agreement

0.80 1.00 Almost perfect agreement







FLORIDA GEOLOGICAL SURVEY


Relative Vulnerability
1 More Vulnerable W*E
SVulnerable s
Less Vulnerable 10 5 0


Wekiva Study Area
Water Features


10 5 0


Figure 20. Relative vulnerability of the FAS predicted by the WAVA model using a random 75%
subset of the original dissolved oxygen training point theme.


10 Miles


10 Kilometers


I


I






REPORT OF INVESTIGATION NO. 104


Discussion and Summary

Results of the statewide FAVA model (Arthur et al., 2005, in preparation) predict that the Wekiva study
area lies within the "vulnerable" to "more vulnerable" zones of relative vulnerability (Figure 12). The
WAVA map, however, provides more detail due to the smaller study area and refined datasets. Using
FAVA as a point of statewide reference, the entire Wekiva study area should be considered at a minimum
"vulnerable." Even the "less vulnerable" zones predicted by the WAVA map are "vulnerable" with
respect to the findings of the FAVA study. In agreement with hydrogeological principles, the "more
vulnerable" zones predicted by the WAVA model (Figure 15) correspond to areas of relatively thin
confinement, high karst-feature density, higher soil permeability and larger differences in hydraulic head
between the water table and the FAS potentiometric surface.

Within the Wekiva study area, the largest and most laterally continuous zone predicted as "more
vulnerable" occurs approximately three miles (-4.8 kilometers [km]) north-northeast of Lake Apopka;
whereas much of the "less vulnerable" areas are predicted to occur along the northeastern part of the
Wekiva study area as well as east and south of Lake Apopka. Several "more vulnerable" areas located
outside of the Wekiva study area yet within the WAVA study area correspond to population centers, such
as Orlando. Future land-use planning in these areas will hopefully consider the results of the WAVA
model as measures are developed to protect local ground-water resources.

In summary, all aquifer systems in Florida are vulnerable to contamination. This is not only attributable
to Florida's hydrogeologic setting but also to human modifications of the natural system. The WAVA
project results provide a probability map that identifies zones of relative vulnerability in the study area
based on the evidential themes and training points used as input for the model. Because the model input
consists only of natural systems characteristics that affect aquifer vulnerability, the WAVA output map is
consequently an approximation of intrinsic aquifer vulnerability (i.e., it does not account for sources of
contamination or behavior of contaminants). The methodology used to produce the WAVA map is a
data-driven process, which minimizes the potential for scientific bias.

The maps are as accurate as the most detailed input layer, and as inaccurate as the least detailed layer. It
is the responsibility of the end-users of these maps to determine specific and appropriate applications (see
also Disclaimer). It is recommended that the WAVA vulnerability map should be applied at scales
smaller than 1:200,000 (i.e., 1:500,000). Implementing this scale recommendation prohibits application
of the vulnerability map to assess individual land parcels. Although much of the input data and the output
vulnerability map are accurate to a raster resolution of 30 m2 (322.9 ft2), use of the map at this scale is not
suggested. Application of the vulnerability map on the order of greater than .75 square miles (1.94 km2)
is more appropriate. In other words, the WAVA vulnerability map is no substitute for site specific
hydrogeological investigation.

The relative vulnerability zones delineated in this report may be applied to facilitate designation of
"protection zones" by regulatory agencies. These "protection zones," however, would likely include other
factors affecting the hydrologic system, including ground-water flow gradients and proximity to impacted
areas. The WAVA vulnerability map can also be used to develop measures designed to protect the
Floridan Aquifer System, which provides flow to the Wekiva River and Wekiwa Spring. The model
results have far reaching implications and can be applied to a variety of environmental management,
protection and conservation activities, including:

Wellhead protection
Source-water protection
Recharge protection






FLORIDA GEOLOGICAL SURVEY


Vulnerability indices
Contaminant-specific maps
Land conservation acquisition
Total maximum daily loads (TMDLs)
Surface-water/ground-water interactions
Water-quality management tool
Resource planning strategies and policies
Prioritization of areas of critical concern
Design of monitoring plans
Best Management Practices

The results of the WAVA vulnerability model are useful for development and implementation of ground-
water protection measures; however, the vulnerability output map included in this report should not be
viewed as a static evaluation of the vulnerability of the Floridan Aquifer System. The model was built
using the best data available at the time this project was completed and as a result, represents a "snapshot"
in time. As new and more refined data become available for the WAVA study area, this data can be
incorporated into future modeling scenarios. The advantages of including new or enhanced data include a
more refined output vulnerability map with higher confidence in the results. Data improvements could
include input of detailed topographic data (i.e., LIDAR), collection of soil permeability data for the "no
data" areas along the Seminole-Orange County border (Figure 17) or strategically collecting water-level
data to improve the hydraulic head difference evidential theme. Enhancement of these data layers would
yield significant improvements in refinement, accuracy and statistical confidence of the final model
output. The more fine-tuned the input data, the more detailed and reliable the model results can be. The
weights of evidence modeling approach to aquifer vulnerability is a highly adaptable and useful tool for
implementing ongoing protection of Florida's vulnerable ground-water resources.

Disclaimer

The WAVA map was developed by the FDEP/FGS to carry out agency responsibilities related to
management, protection, and responsible development of Florida's natural resources. Although efforts
have been made to make the information in these maps accurate and useful, the FDEP/FGS assumes no
responsibility for errors in the information and does not guarantee that the data are free from errors or
inaccuracies. Similarly FDEP/FGS assumes no responsibility for the consequences of inappropriate uses
or interpretations of the data on these maps. As such, these maps are distributed on an "as is" basis and the
user assumes all risk as to their quality, the results obtained from their use, and the performance of the
data. FDEP/FGS further makes no warranties, either expressed or implied as to any other matter
whatsoever, including, without limitation, the condition of the product, or its suitability for any particular
purpose. The burden for determining suitability for use lies entirely with the user. In no event shall the
FDEP/FGS have any liability whatsoever for payment of any consequential, incidental, indirect, special,
or tort damages of any kind, including, but not limited to, any loss of profits arising out of use of or
reliance on the maps or support by FDEP/FGS. FDEP/FGS bears no responsibility to inform users of any
changes made to this data. Anyone using this data is advised that resolution implied by the data may far
exceed actual accuracy and precision.

Comments on this data are invited and FDEP/FGS would appreciate that documented errors be brought to
the attention of our staff. The development of these data sets represents a major investment of staff time,
effort and funds. As a professional responsibility, we expect that the FDEP/FGS will receive proper credit
when these data sets are used. Further, since part of this data was developed and collected with State of
Florida funding, no proprietary rights may be attached to it in whole or in part, nor may it be sold to the
U.S. Government or the Florida State Government as part of any procurement of products or services.






REPORT OF INVESTIGATION NO. 104


WEIGHTS OF EVIDENCE GLOSSARY

Binary Refers to the generalization or simplification of evidential themes or data layers. Binary layers
are reclassified from the original dataset into presence/absence type themes or into two classes.

Conditional Independence Occurs when an evidential theme does not affect the probability of another
evidential theme. Evidential themes are considered independent of each other if the conditional
independence value calculated is within the range 1.00 + 0.15 (Raines, personal communication, 2003).
Values that significantly deviate from this range can inflate the posterior probabilities resulting in
unreliable response themes.

Confidence of evidential theme Contrast divided by its estimated standard deviation; provides a useful
measure of significance of the contrast.

Confidence of Posterior Probability A measure based on the ratio of posterior probability to its
estimated standard deviation.

Contrast W+ minus W- (see weights), which is an overall measure of the spatial association
(correlation) of an evidential theme with the training points.

Cumulative Ascending Calculates the cumulative weights from the first class to the last class while
increasing the area. Areas nearest a training point have a stronger association, and those farthest away
have a weaker association. This method is applicable for themes where the training points are mainly
associated with the lower values of the evidential theme (e.g., higher vulnerability correlates with lower
confinement thickness).

Cumulative Descending Calculates the cumulative weights from the last class to the first class while
increasing the area (opposite of cumulative ascending). This method is applicable for themes where the
training points are mainly associated with the higher values of the evidential theme (e.g., higher
vulnerability correlates with higher soil permeability).

Evidential Theme A set of continuous spatial data that is associated with the location and distribution
of known occurrences (i.e., training points); these map data layers are used as predictors of vulnerability.

Kappa Coefficient Allows statistical comparison of map patterns. It is a multivariate accuracy
assessment technique developed by Cohen (1960) to determine if one error matrix is significantly
different than another.

LIDAR Light detection and ranging; an airborne remote sensing system that uses laser light to measure
land-surface elevation with a high degree of accuracy and resolution. LIDAR data can be used to
generate digital elevation models.

Model The characteristics of a set of training points, and the relationships of the training points to a
collection of evidential themes.

Posterior Probability The probability that a unit cell contains a training point after consideration of the
evidential themes. This measurement changes from location to location depending on the values of the
evidence.






FLORIDA GEOLOGICAL SURVEY


Prior Probability The probability that a unit cell contains a training point before considering the
evidential themes. It is a constant value over the study area equal to the training point density (total
number of training points divided by total study area in unit cells).

Response Theme An output map that displays the probability that a unit area would contain a training
point, estimated by the combined weights of the evidential themes. The output is displayed in classes of
relative aquifer vulnerability or favorability to contamination (i.e., this area is more vulnerable than that
area) or favorability. The response theme is the relative vulnerability map.

Spatial Data Information about the location and shape of, and relationships among, geographic
features, usually stored as coordinates and topology.

Study Area A grid theme that acts as a mask to define the area where the model is developed and
applied. It may be irregular in outline and may contain interior holes (e.g., lakes and no data areas).

Training Points A set of locations (points) reflecting a parameter used to calculate weights for each
evidential theme, one weight per class, using the overlap relationships between points and the various
classes. In an aquifer vulnerability assessment, training points are wells with one or more water quality
parameters indicative of relatively higher recharge which is an estimate of relative vulnerability.

Weights A measure of an evidential-theme class. A weight is calculated for each theme class. For
binary themes, these are often labeled as W+ and W-. For multiclass themes, each class can also be
described by a W+ and W- pair, assuming presence/absence of this class versus all other classes. Positive
weights indicate that more points occur on the class than due to chance, and the inverse for negative
weights. The weight for missing data is zero. Weights are approximately equal to the proportion of
training points on a theme class divided by the proportion of the study area occupied by theme class,
approaching this value for an infinitely small unit cell.






REPORT OF INVESTIGATION NO. 104


REFERENCES

Arthur, J.D., Baker, A.E., Cichon, J.R., Wood, H.A.R., and Rudin, A., 2005, in preparation, Florida
Aquifer Vulnerability Assessment (FAVA): Contamination potential of Florida's principal
aquifer systems: Florida Geological Survey Bulletin 67.

Baker, A.E., Wood, A.R., Cichon, J.R., and Arthur, J.D., 2005, Alachua Aquifer Vulnerability
Assessment; a final report submitted to Alachua County, January, 2005, 36 p. (unpublished).

Berndt., M.P., Oaksford, E.T., Mahon, G.L., and Schmidt., W., 1998, Groundwater, in Femald, E.A., and
Purdum, E. D., eds., Water Resources Atlas of Florida: Florida State University, Institute of
Public Affairs, 312 p.

Bonham-Carter, G. F., 1994, Geographic Information Systems for Geoscientists, Modeling with GIS:
Oxford, Pergamon, 398 p.

Cichon, J.R., Wood, H.A.R., Baker, A.E., and Arthur, J.A., 2004, Application of Geologic Mapping and
Geographic Information Systems to Delineate Sensitive Karst Areas for Land-Use Decisions:
American Geological Institute website, http://www.agiweb.org/environment/publications/
mapping/graphics/florida.pdf, 2004.

Cohen, J., 1960, A coefficient of agreement for nominal scales: Educational and Psychological
Measurement, v. 20, no. 1, p. 37-46.

Davis, J., Johnson, R., Boniol, D., and Rupert, F., 2001, Guidebook to the Correlation of Geophysical
Well Logs within the St. Johns River Water Management District: Florida Geological Survey
Special Publication No. 50, 114 p.

Florida Geographic Data Library, 2003, GeoPlan Center, Gainesville, University of Florida:
www.fgdl.org/.

Kemp, L.D., Bonham-Carter, G.F., Raines, G.L. and Looney, C.G., 2001, Arc-SDM: Arcview extension
for spatial data modeling using weights of evidence, logistic regression, fuzzy logic and neural
network analysis: http://ntserv.gis.nrcan.gc.ca/sdm/, 2002.

Landis, J.R. and Koch, G.G., 1977, The measurement of observer agreement for categorical data:
Biometrics, v. 33, p. 159-174.

National Research Council, 1993, Ground Water Vulnerability Assessment: Predicting Relative
Contamination Potential under Conditions of Uncertainty: Washington, National Academy Press,
204 p.

Raines, Gary L., 1999, Evaluation of Weights of Evidence to Predict Epithermal-Gold Deposits in the
Great Basin of the Western United States: Natural Resources Research, v. 8, no. 4, p. 257-276.

Raines, G. L., Bonham-Carter, G. F., and Kemp, L., 2000, Predictive Probabilistic Modeling Using
ArcView GIS: ArcUser, v. 3, no.2, p. 45-48.






FLORIDA GEOLOGICAL SURVEY


Scott, T.M., Means, G.H., Meegan, R.P., Means, R.C., Upchurch, S.B., Copeland, R.E., Jones, J.,
Roberts, T., and Willet, A., 2004, Springs of Florida: Florida Geological Survey Bulletin No. 66,
377 p.

Sepulveda, N., 2002, Simulation of Ground-Water Flow in the Intermediate and Floridan Aquifer
Systems in Peninsular Florida: U.S. Geological Survey Water-Resource Investigation Report 02-
4009, 130 p.

Southeastern Geological Society Ad Hoc Committee, 1986, Hydrogeological Units of Florida: Florida
Geological Survey Special Publication No. 28, 8 p.

U.S. Census Bureau: State and County Quick Facts, 01-Feb-2005, 15:48:47 EST: www.census.gov.

Wekiva River Basin Coordinating Committee, 2004, Wekiva River Basin Coordinating Committee Final
Report: Recommendations for Enhanced Land Use Planning Strategies and Development
Standards to Protect Water Resources of the Wekiva River Basin, March 16, 2004, 50 p.
(unpublished).

Wright, A.P., 1974, Environmental Geology and Hydrology of the Tampa Area, Florida: Florida
Geological Survey Special Publication No. 19, 93 p.

PAGE 1

WEKIVA AQUIFER VULNERABILITY ASSESSMENT FLORIDA GEOLOGICAL SURVEY REPORT OF INVESTIGATION 104

PAGE 2

iiSTATE OF FLORIDA DEPARTMENT OF ENVIRONMENTAL PROTECTION Colleen M. Castille, Secretary DIVISION OF RESOURCE ASSESSMENT AND MANAGEMENT Edwin J. Conklin, Director FLORIDA GEOLOGICAL SURVEY Walter Schmidt, State Geologist and Chief Report of Investigation 104 WEKIVA AQUIFER VULNERABILITY ASSESSMENT By James R. Cichon, Alan E. Baker, Alex R. Wood, and Jonathan D. Arthur Published for the FLORIDA GEOLOGICAL SURVEY Tallahassee, Florida 2005

PAGE 3

iii Printed for the Florida Geological Survey Tallahassee 2005 ISSN 0160-0931

PAGE 4

iv PREFACE FLORIDA GEOLOGICAL SURVEY Tallahassee, Florida 2005 The Florida Geological Survey (FGS), Division of Resource Assessment and Management, Department of Environmental Protection, is publishing as Report of Investigation No. 104, Wekiva Aquifer Vulnerability Assessment , authored by James R. Cichon, Alan E. Baker, Alex R. Wood, and Jonathan D. Arthur, all staff of the FGS. This publication is a localized offshoot of the state-wide Florida Aquifer Vulnerability Assessment project (a Bulletin of the Florida Geological Survey, in press). Unbiased and credible scientific assessment of the potential vulnerability of our ground water aquifers to contamination and other surface modifications is critical to the future sustainability of Florida’s environment. State, regional, county and local governmental agencies, as well as private land owners and the public will find this information useful in support of informed land-use planning and natural resource conservation and preservation decisions. Walter Schmidt, Ph.D., P.G. State Geologist & Chief Florida Geological Survey

PAGE 5

v TABLE OF CONTENTS INTRODUCTION........................................................................................................................................1 Aquifer Systems...............................................................................................................................1 Vulnerability....................................................................................................................................2 ACKNOWLEDGEMENTS..........................................................................................................................3 WEKIVA AQUIFER VULNERABILITY ASSESSMENT MODEL.........................................................4 Weights of Evidence........................................................................................................................4 Study Area..........................................................................................................................4 Training Points....................................................................................................................6 Evidential Themes..............................................................................................................6 Generalization of Evidential Themes.....................................................................7 Soil Permeability......................................................................................8 Buffered Effective Karst Features............................................................8 Thickness of the Intermediate Aquifer System.......................................10 Hydraulic Head Difference between SAS and FAS...............................15 Response Theme...............................................................................................................15 Confidence Map................................................................................................................25 Validation and Sensitivity Analysis..................................................................................25 Dissolved Oxygen Data vs. Posterior Probability................................................28 Random 75% Subset of Training Points..............................................................28 Discussion and Summary...............................................................................................................30 Disclaimer......................................................................................................................................32 WEIGHTS OF EVIDENCE GLOSSARY.................................................................................................33 REFERENCES...........................................................................................................................................35 FIGURES 1. Hydrogeologic conceptual model of the Wekiva study area ....................................................................2 2. Extent of the WAVA study area and training point set developed based on dissolved oxygen ..............5 3. Soil permeability map of the WAVA study area .....................................................................................9 4. Soil permeability values (in/hr) plotted against contrast values calculated using WofE........................10 5. Map showing binary generalization of the soil permeability evidential theme......................................11 6. Closed topographic depressions underlain by less than 100 feet (30.5 m) of FAS overburden.............12 7. Map showing binary generalization of the buffered effective karst features evidential theme..............13 8. Thickness grid of IAS (in feet) and the distribution of wells used.........................................................14 9. Map showing binary generalization of the IAS thickness evidential theme..........................................16 10. Hydraulic head difference between the SAS and the FAS potentiometric surface...............................17 11. Map showing binary generalization of the hydraulic head difference evidential theme.......................18 12. Relative vulnerability of the FAS FAVA model...................................................................................19 13. WofE conceptual model of the FAS.....................................................................................................21 14. Posterior probability class breaks in WAVA model.............................................................................22 15. Relative vulnerability predicted by the WAVA model for the FAS.....................................................23 16. Relative vulnerability predicted by the FAS WAVA model showing hydrography.............................24 17. Relative vulnerability of the WAVA model showing missing soil permeability data..........................26 18. Distribution of confidence values calculated for WAVA response theme...........................................27 19. Posterior probability vs. dissolved oxygen median concentration values.............................................28 20. Relative vulnerability predicted by the WAVA model using a subset of training point data...............30

PAGE 6

vi TABLES 1. Test values calculated in WofE and their respective studentized T values...............................................7 2. WofE final output table listing weights calculated for each evidential theme........................................22 3. Cross-tabulation matrix of the WAVA response theme versus the 75% subset response theme...........29 4. Kappa coefficient values and associated interpretation (Landis and Koch, 1977).................................29

PAGE 7

REPORT OF INVESTIGATION NO. 104 1WEKIVA AQUIFER VULNERABILITY ASSESSMENT By James R. Cichon, Alan E. Baker, P.G. 2324, Alex R. Wood, and Jonathan D. Arthur, P.G. 1149 INTRODUCTION The Floridan Aquifer System is among the most vast and vulnerable hydrogeologic resources in the nation. A remarkable 90% of Florida’s drinking water originates from these ground-water resources (Berndt et al., 1998). More than 700 springs are known to exist in Florida, allowing a unique window into the aquifer systems (Scott et al., 2004). These springs have long been a resource for recreation and fresh water. Protection of Florida’s springs becomes an increasingly important issue as the population of Florida continues to grow at a rate of almost 900 people per day (U.S. Census, 2005). As a result, pressures on these unique natural resources are intensifying. It is important to note that water quality of Florida’s springs is a reflection of the condition of the surrounding ground-water system and not just an indication of individual spring health. Springsheds, or the spring’s area of ground-water contribution, must also be taken into consideration for any vulnerability assessment if proper protection of the spring itself is to take place. An assessment of the vulnerability of the ground water, which includes the springshed, can therefore be an important tool for decision making, development of rules or policies regarding environmental conservation, protection, growth management, and land-use planning. To fully understand the relationship between ground water and vulnerability, an understanding of the aquifer systems present in the study area is needed. Aquifer Systems The hydrogeologic framework of the Wekiva study area is comprised of three aquifer systems: the Surficial Aquifer System (SAS), the Intermediate Aquifer System (IAS) and the FAS. According to Florida Geological Survey Special Publication No. 28 (Southeastern Geological Society, 1986), the Surficial Aquifer System is “the permeable hydrogeologic unit contiguous with land surface that is comprised principally of unconsolidated to poorly indurated clastic deposits. It also includes wellindurated carbonate rocks” that are not part of the FAS. The intermediate aquifer system/intermediate confining unit consists of highly-variable siliciclastic and carbonate deposits that are relatively low-permeability, fine-grained sediments and collectively retard the exchange of water between the overlying SAS and the underlying FAS. The term “intermediate confining unit” applies to those areas where this unit yields little to no water, whereas the term “intermediate aquifer system” applies to those areas where one or more low to moderate-yielding aquifers occur. Special Publication No. 28 (Southeastern Geological Society, 1986) is currently under revision by the Florida Geological Survey (FGS), and the forthcoming update suggests the use of the term “Intermediate Aquifer System” for this entire unit. Moreover, the revision calls for the elimination of the use of “intermediate confining unit.” Instead, the “intermediate confining unit” is considered to be confining beds within the IAS. This newer convention currently under review is hereby adopted in this report. The FAS is comprised of a thick carbonate sequence of rocks that collectively function as a regional water-yielding hydraulic unit. Where overlain by the IAS, the FAS is confined; however, sinkholes that

PAGE 8

FLORIDA GEOLOGICAL SURVEY 2 breach the IAS can provide direct pathways for surface water to enter the FAS (Figure 1). Ground water also recharges the FAS by infiltration through overlying sediments of the IAS or SAS. In areas where the FAS is directly overlain by the SAS, the FAS may or may not contain ground water under confined conditions, depending on the thickness and extent of low permeability material at the base of the Surficial Aquifer System (Southeastern Geological Society, 1986). Figure 1. Hydrogeologic conceptual model of the Wekiva study area depicting the major aquifer systems and karst features. Vulnerability Due to Florida’s unique hydrogeologic setting, all of Florida’s ground water is vulnerable to contamination. In fact, this statement, in a more broad sense, is considered the “First Law of Ground Water Vulnerability” by the National Research Council (NRC, 1993) which states: “All ground water is S i n k h o l e

PAGE 9

REPORT OF INVESTIGATION NO. 104 3 vulnerable.” Furthermore, the NRC defines the phrase “ground-water vulnerability to contamination” as the tendency or likelihood for contaminants to reach a specified position in the ground-water system after introduction at some location above the uppermost aquifer. In this report, a similar definition of aquifer vulnerability is adopted: The tendency or likelihood for contaminants to reach the top of the FAS after introduction at land surface. The Wekiva River Coordinating Committee Final Report identified numerous studies by Florida’s water management districts and the United States Geological Survey (USGS) that clearly demonstrate contamination attributable to changes in land use. In other words, what occurs on the land directly and indirectly affects the quality of water moving through the subsurface conduits. Many of these conduits flow to surface-water bodies via springs. The Florida Geological Survey (FGS) applied a modeling technique used in the statewide Florida Aquifer Vulnerability Assessment (FAVA) project to the Wekiva study area in order to delineate relative degrees of vulnerability for the FAS based on available data. This project is known as the Wekiva Aquifer Vulnerability Assessment model or WAVA. The modeling technique used is known as Weights of Evidence (WofE), which is based entirely in a geographic information system (GIS) and is summarized below. For a more thorough explanation of this method refer to FGS Bulletin 67, Florida Aquifer Vulnerability Assessment: Contamination potential of Florida’s principal aquifer systems (Arthur et al., 2005, in preparation). The relative vulnerability zones identified in the WAVA model may be applied to facilitate designation of "protection zones" by regulatory agencies. These “protection zones,” however, would likely include other physical and chemical (i.e., water quality) factors affecting the hydrologic system, including ground-water flow rates and patterns and proximity to impacted areas. The software used to execute WofE and the Florida Department of Environmental Protection (FDEP) Albers projection require certain data layers to be listed in meters. However, some data layers are listed in units standard to that particular data set. For reader convenience we list the primary unit followed by the alternative unit in parentheses. ACKNOWLEDGEMENTS The authors are grateful for the work of the Wekiva River Basin Coordinating Committee (WRBCC), which was appointed by the Governor in July, 2003. This committee follows a series of legislation and Governor Executive Orders, all with the ultimate goal of protecting surfaceand ground-water resources of the Wekiva River system. The research presented herein was funded by the Florida Springs Initiative through the FDEP, and was facilitated by the support of Vivian Garfein, Director, FDEP Central District, Richard Drew, Bureau Chief, FDEP Water Facilities Regulation and Mike Bascom, FDEP Springs Coordinator. We are thankful for the insightful review comments provided by individuals listed above, including members of the WRBCC, as well as Richard Deadman, Florida Department of Community Affairs. Moreover, we gratefully acknowledge the following individuals for providing technical and/or editorial review of the report: Chris Ferraro, Lou Ley and Alan Stodghill (FDEP), William Pollock and Dr. Eberhard Roeder, Florida Department of Health, and several FDEP/Florida Geological Survey staff, including Carol Armstrong, Dr. Rick Copeland, Tom Greenhalgh, Jackie Lloyd, Frank Rupert, Dr. Tom Scott and Dr. Walt Schmidt.

PAGE 10

FLORIDA GEOLOGICAL SURVEY 4 WEKIVA AQUIFER VULNERABILITY ASSESSMENT MODEL Weights of Evidence Use of WofE requires the combination of diverse spatial data which are used to describe and analyze interactions and generate predictive models (Raines et al., 2000). When applied in the WAVA project, WofE was used to generate maps of aquifer vulnerability, or response themes. These response themes were generated in the Environmental Systems Research Institute ArcView 3.2 environment. WofE was executed using the Arc Spatial Data Modeler which is available as an internet download (Kemp et al., 2001). A primary benefit of applying WofE to the WAVA project is that it is data-driven, rather than expertdriven. The data that “train” the model consist of known occurrences of parameters (water quality values) that reflect relative aquifer vulnerability, such as high levels of dissolved oxygen in ground-water wells. These wells were the training points used to calculate weights for laterally continuous input data layers, or evidential themes, which are then combined to yield a response theme. When reviewing the model results, it is important to note that all aquifers, to some degree, are vulnerable to contamination from land surface. The model results simply identify those areas within the study area that are more vulnerable or less vulnerable based on the evidential themes and training points used in the model. So, the vulnerable areas are assigned values relative to each other within the study area. These “relative” values are not directly transferable or correlated to another study area. Study Area The initial step in the development of the WAVA model was the delineation of a study area extent. The study area, comprised of 30 square meters (m2)(322.9 ft2) grid cells, was used in the calculation of weights and probabilities throughout the modeling process as described below. The extent of the study area used for input into the WAVA model was based on a combination of the Wekiva study area (as identified in the Wekiva River Basin Coordinating Committee, 2004) and the area contributing ground water to the Wekiva River (Figure 2). Large water bodies were omitted from the WAVA model because a well would never be drilled in these areas – therefore, they would never contain a training point. Additionally, if lakes were left in the model, the surface area was increased with no chance of increasing the number of training points. This would unnecessarily bias the model. Further, large water bodies typically have no soils or other input data associated with them, thus the model output omits these areas due to lack of data or potential bias in the calculated probabilities. Removal of large water bodies does not change the relative results (i.e., relative vulnerability) of the model; however, it would have the effect of increasing weights calculated during the modeling process. Regardless, the model is more valid and defensible by removing the large water bodies. Due to the geostatistical framework and evidential layers (spatial hydrogeological data) of WAVA, aquifer systems in discharge areas were sometimes predicted by the output model to be low in vulnerability, even though the discharging surface waters are highly vulnerable to contamination. These discharging surface waters are not part of the aquifer, although they can originate from it. The WAVA project was designed to focus on the ability for a contaminant to travel through soils, overburden, karst features, etc. to enter into the aquifer system. As a result, it is very important that the WAVA model never be applied to assess contamination of surface waters or discharge areas.

PAGE 11

REPORT OF INVESTIGATION NO. 104 5 Figure 2. Extent of the WAVA study area and training point set developed based on dissolved oxygen measured in wells. Large water bodies have been omitted from the analysis to avoid biasing the model (see text). Wekiva River Basin Coordinating Committee (WRBCC; March 16, 2004).

PAGE 12

FLORIDA GEOLOGICAL SURVEY 6 Training Points In WofE, training points are locations of known measured occurrences (i.e., locations where quantitative data have been obtained). In mining applications for example, identified ore reserves are known occurrences (training points). In an aquifer vulnerability assessment, wells with water quality data indicative of high recharge (i.e., dissolved oxygen, nitrates above an identified background threshold, tritium) are potential known occurrences (i.e., areas where a good connection exists between the top of the aquifer and land surface). Training points are used in WofE to calculate the following parameters (see Weights of Evidence Glossary ): prior probability, weights for each evidential theme, and posterior probability of the response theme. Training points are converted to represent a unit area (i.e., a 30 m2 grid cell) of the study area within a GIS application. The water-quality parameter selected for the WAVA training data set was dissolved oxygen. In general, native ground-water concentrations of dissolved oxygen in the FAS are typically low where the aquifer system is not affected by activities at land surface or active recharge. Therefore, where dissolved oxygen occurs at concentrations above background levels in an aquifer system, it can generally be assumed a relatively greater hydrologic connection exists between land-surface activities and ground water. Dissolved oxygen analytical values served as the training point set for the WAVA model. More than 200 wells completed in the FAS for the WAVA study area were compiled from the FDEP Background Water Quality Monitoring Network, the FDEP STATUS network, and the St. John’s River Water Management District monitoring network. Of these wells, 28 were measured for dissolved oxygen. Using statistical methods, described in Arthur et al. (2005, in preparation), two wells were identified as outliers and removed from the dataset leaving 26 wells for further analysis (Figure 2). The 75th percentile value was the statistical break used to identify relatively high dissolved oxygen concentrations in ground water, which is considered herein as a proxy for areas of higher recharge and vulnerability. Median dissolved oxygen concentrations for wells with multiple analyses were calculated. These median values, along with data from wells with one dissolved oxygen analysis were then used to calculate a grand median value. The 75th percentile combined median value for dissolved oxygen concentration was 0.4 mg/L. Six wells existed in the dataset that had a measured median dissolved oxygen value greater than the 0.4 mg/L threshold. These six wells comprised the training point theme for input into the WAVA model (Figure 2). The probability that a training point will occupy a defined unit area within the study area, independent of any evidential theme data, is known as prior probability. The prior probability is calculated by dividing a unit area representing the training points by the total study area. In less complex terms, the prior probability is based on prior knowledge of the problem without the benefit of supporting evidence. The prior probability for the WAVA model was calculated at 0.0017. The distribution of all wells sampled for dissolved oxygen, and the wells meeting training point criteria are displayed in Figure 2. Evidential Themes Several evidential themes were considered for use in the WAVA model due to their potential influence on ground-water quality. They are: Soil permeability Buffered effective karst features (based on closed topographic depressions) Thickness of Intermediate Aquifer System (IAS) Hydraulic head difference between Surficial Aquifer System (SAS) and FAS Vertical leakage rate to and from the FAS

PAGE 13

REPORT OF INVESTIGATION NO. 104 7 Vertical leakage rate to and from the FAS ultimately was not used because it was based on model simulations with a grid-cell size of 1,000 meters and there was large variability in the calculated leakage rate between adjacent grid cells. Weights were calculated for each of the other four evidential themes and generalized as discussed below. Generalization of Evidential Themes Evidential themes were generalized in an effort to assess which areas of the evidence shared a greater association with locations of training points. During calculation of weights for each evidential theme used in the WAVA project, a contrast value was calculated for each class of the theme by combining the positive and negative weights (positive weight – negative weight). Contrast is a measure of a theme’s significance in predicting the location of training points and helps to determine the threshold or thresholds that maximize the spatial association between the evidential theme map pattern and the training point theme pattern (Bonham-Carter, 1994). Arthur et al., (2005, in preparation) provide a more comprehensive explanation of theme generalization. Confidence of the evidential theme equals the contrast divided by the standard deviation (a student T-test) for a given evidential theme and provides a useful measure of significance of the contrast due to the uncertainties of the weights and areas of possible missing data (Raines, 1999). A confidence value of 1.282 corresponds to a 90% level of significance (see Table 1), which was the value selected as the minimum acceptable confidence level for the WAVA project evidential themes. Confidence values approximately correspond to the statistical levels of significance listed in Table 1. Table 1. Test values calculated in WofE and their respective studentized T values expressed as level of significance in percentages. Studentized T Value (confidence expressed as level of significance) Test Value 99.5% 2.576 99% 2.326 97.5% 1.960 95% 1.645 90% 1.282 80% 0.842 75% 0.674 70% 0.542 60% 0.253 Contrast values were used to determine where to sub-divide evidential themes into generalized categories. The most common method of categorizing an ordered evidential theme was to select the maximum contrast as a threshold value to create a binary generalized evidential theme. In some WofE models, categorization of more than two classes may be justified (Arthur et al., 2005, in preparation). For evidential themes used for the WAVA project, a binary break was typically defined by the WofE analysis thereby creating two spatial categories: one with stronger association with the training point theme and one with weaker association with the training point theme.

PAGE 14

FLORIDA GEOLOGICAL SURVEY 8 Soil Permeability Soil permeability values were obtained from the soils statewide grid layer developed for FAVA (Arthur et al., 2005, in preparation), which was compiled from the United States Department of Agriculture Natural Resources Conservation Service Soil Service Geographic (SSURGO) data set. This database was mapped at a scale of 1:24,000 and contains the most detailed soil permeability data. However, different permeability values can exist along some county boundaries due to the different mapping techniques used by different soil scientists. The development of this layer included the calculation of average soil permeability values for each soil horizon layer. Then, based on soil horizon thicknesses, weighted-average permeability values were calculated for the entire soil column. This allowed the generation of a data coverage of soils containing a single permeability value per soil polygon. This data coverage is displayed in Figure 3. Missing-data areas on the figure represent urban land, pits and water features. Soil permeability is a measure of the rate at which water travels through the vadose (unsaturated) zone. Areas with high soil permeability values are normally associated with higher aquifer vulnerability. Weights were calculated for soil permeability using the cumulative descending method (refer to Weights of Evidence Glossary ) of the WofE model technique. The highest contrast of any class was calculated at 13.1 in/hr (9.2 x 10-3 cm/sec) as displayed in Figure 4. Based on the calculated weights, the most appropriate break in the soil permeability evidential theme was at 13.1 in/hr (9.2 x 10-3 cm/sec), creating a binary generalized theme for input into the WAVA model. This contrast break indicated that values exceeding 13.1 in/hr (9.2 x 10-3 cm/sec) were strongly correlated with aquifer vulnerability as defined by the training point data, whereas values less than 13.1 in/hr (9.2 x 10-3 cm/sec) were less significant. The generalized theme is displayed in Figure 5. Buffered Effective Karst Features To represent karst features (i.e., sinkholes) affecting vulnerability in the WAVA model study area, an effective karst GIS grid was generated based on a combination of closed topographic depressions identified in USGS 7.5-minute quadrangle maps and thickness of overburden on FAS (Figure 6). Closed topographic depressions (considered an estimation of karst feature distribution) are represented on USGS 7.5-minute maps as enclosed hachured contour lines. The effective karst feature evidential theme was developed by evaluating which of these closed topographic depressions were more likely to be hydrologically connected to the underlying FAS. This was accomplished by filtering out depressions underlain by more than 100 feet (30.5 m) of FAS overburden thickness. The 100-ft (30.5 m) threshold of overburden thickness has been used to identify karst-prone areas by Cichon et al. (2004) and Wright (1974). Though the location of training points was not used to select this filter threshold, the lack of their occurrence in areas underlain by more than 100 feet (30.5 m) of FAS overburden thickness lends support to the use of this filter. Areas nearer to an effective karst feature are normally associated with higher aquifer vulnerability due to the increased chance of overland flow and the infiltration into the depression. Therefore, buffer zones of 3,600-m (11,811 ft) divided into 30-m (98.4 ft) intervals were generated around each karst feature, and weights were calculated for the effective karst feature evidential theme using the cumulative ascending method. The highest contrast of any class was calculated at a distance of 60 m (196.9 ft) from a depression creating a binary generalized theme for input into the WAVA model. In other words, the analysis indicated that this threshold of proximity to karst features maximized the spatial association between the map pattern and the training point pattern. The generalized theme is displayed in Figure 7.

PAGE 15

REPORT OF INVESTIGATION NO. 104 9 10 0 10Miles 10 0 10KilometersSoil Permeability (in/hr) 10.5 13.0 13.0 20.0 0 1.5 1.5 4.5 7.0 9.5 9.5 10.5 4.5 7.0 No Data Water Features Figure 3. Soil permeability map of the WAVA study area (Florida Geographic Data Library, 2003). Note differences across county lines due to different mapping techniques. Missing-data areas represent urban land, pits and water features.

PAGE 16

FLORIDA GEOLOGICAL SURVEY 10 -0.50 0.00 0.50 1.00 1.50 2.00 0.0 5.0 10.0 15.0 20.0 25.0Soil Permeabilit y ( in/hr ) 13.1 Figure 4. Soil permeability values (in/hr) plotted against contrast values calculated using WofE. The highest contrast value was calculated at 13.1 in/hr (9.2 x 10-3 cm/sec). Thickness of the Intermediate Aquifer System For the purposes of the WAVA project, “Intermediate Aquifer System,” or IAS, refers to the statewide hydrostratigraphic unit that provides variable confinement to the FAS. The initial step in the identification of the IAS was to create a database of well core and cuttings data. The well database was then used to create a hydrostratigraphic surface for the top of the IAS and the top of the FAS (which coincides with the base of the IAS). Following creation of these surface models, it was necessary to resolve the IAS with land-surface elevation in localized areas where the IAS surface interpolation extended above land surface. After the hydrostratigraphic surfaces were developed and refined, calculation of a thickness map was completed by subtraction of the IAS surface from the FAS surface. The final output yielded a continuous thickness map of the IAS; however the map could not take into account areas of zero IAS thickness due to the grid calculation. As a result, based on data from more than 24 wells within and immediately adjacent to the study area, an approximate extent of the IAS was delineated. These areas, which are located within the northeastern and southwestern part of the WAVA study area are depicted on Figure 8 as “IAS thin to absent” and are juxtaposed against the calculated grid results.

PAGE 17

REPORT OF INVESTIGATION NO. 104 11 10 0 10Miles 10 0 10KilometersSoil Permeability (in/hr) 0.1 13.1 13.1 20.0 Water Features Figure 5. Map showing binary generalization of the soil permeability evidential theme, based on a contrast value of 13.1 in/hr (9.2 x 10-3 cm/sec).

PAGE 18

FLORIDA GEOLOGICAL SURVEY 12 Effective Karst Features Closed topographic depressions Water Features 10 0 10Miles 10 0 10Kilometers Figure 6. Closed topographic depressions underlain by less than 100 feet (30.5 m) of FAS overburden.

PAGE 19

REPORT OF INVESTIGATION NO. 104 13 10 0 10 5 Miles 10 0 10 5 KilometersBuffered Effective Karst Features (meters) > 60 0 60 Water Features Figure 7. Map showing binary generalization of the buffered effective karst features evidential theme based on a contrast buffer distance of 60 m (196.9 ft).

PAGE 20

FLORIDA GEOLOGICAL SURVEY 14 10 0 10Miles 10 0 10KilometersLegendIAS Thickness Grid IAS thin to absent 1 20 20 40 40 60 60 80 80 100 100 120 120 140 Water FeaturesWells completed in top of FAS only Wells completed in top of ICU only Wells completed in top of FAS and ICU Figure 8. Thickness grid of IAS (in feet) and the distribution of wells used. Data from FGS lithologic database combined with geophysical well logs (Davis et al., 2001) were utilized to develop this coverage. See text for explanation of “IAS thin to absent” areas.

PAGE 21

REPORT OF INVESTIGATION NO. 104 15 Areas underlain by thinner IAS sediments are normally associated with higher aquifer vulnerability. Weights were therefore calculated for the IAS evidential theme using the cumulative ascending method. The highest contrast of any class was calculated at a thickness of 48 feet (14.6 m) creating a binary generalized theme for input into the WAVA model (Figure 9). In other words, the analysis indicated that this threshold of IAS thickness maximized the spatial association between the map pattern and the training point pattern. The best available data indicated that the area including Wekiwa Spring is underlain by 50 feet (15.2 m) or more of IAS thickness. This data included well core and cuttings data and geophysical (gamma) well logs from FGS lithologic database and FGS Special Publication No. 50 (Davis et al., 2001). As a result, for this analysis, the spring is located outside the area which is strongly correlated with aquifer vulnerability as defined by the training point data (i.e., greater than 48 feet or 14.6 m thick). In fact, the exposed rocks within the spring pool are the Coosawhatchie Formation of the Hawthorn Group which is part of the IAS (Harley Means, personal communication, 2004). Hydraulic Head Difference between SAS and FAS Hydraulic head difference was calculated by subtracting the FAS 1993 potentiometric surface (Sepulveda, 2002) from the water-table surface (Arthur et al., 2005, in preparation). Although more recent potentiometric surfaces for the FAS have been published, the data control in Sepulveda’s (2002) map was more dense and the potentiometric surface was more highly resolved. Areas with a positive hydraulic head difference value indicated that the FAS is receiving recharge, whereas areas with a negative value indicated the FAS is discharging to the overlying aquifer system (Figure 10). Areas of greater hydraulic head difference between the SAS (water table) and FAS indicate greater potential for downward recharge to the FAS, which is normally associated with higher aquifer vulnerability. Weights were therefore calculated for the hydraulic head difference evidential theme using the cumulative descending method. The highest contrast of any class was calculated at a hydraulic head difference value of 17 ft (5.2 m) creating a binary generalized theme for input into the WAVA model. In other words, the analysis indicated that this threshold of hydraulic head difference maximized the spatial association between the map pattern and the training point pattern. Figure 11 displays the binary generalization used for the hydraulic head difference evidential theme. Response Theme The WAVA response theme is an output map, calculated using WofE, showing the probability (posterior probability) that a unit area is vulnerable to contamination from land surface based on the evidence provided. A response theme is portrayed as relative vulnerability and is classified based on the inflections in a chart in which cumulative study area is plotted against posterior probability. The chart may allow for the selection of several classes of posterior probability. The more vulnerable areas correspond with higher posterior probabilities, while the less vulnerable areas are associated with lower posterior probabilities. In essence, a higher posterior probability indicates that an area is more likely to contain a training point, or more likely to be contaminated, and therefore more vulnerable to contamination from land surface. As described in the Introduction , the WAVA model was based on the modeling technique used in the FAVA project. The FAVA project identified relative vulnerability of Florida’s principal aquifer systems and consolidated the results into three classes (more vulnerable, vulnerable and less vulnerable). The FAVA response theme clipped to the Wekiva study area (Figure 12) shows that the majority of the

PAGE 22

FLORIDA GEOLOGICAL SURVEY 16 10 0 10Miles 10 0 10KilometersIAS Thickness (feet) 0 48 > 48 Water Features Figure 9. Map showing binary generalization of the IAS thickness evidential theme based on a contrast value of 48 ft (14.6 m).

PAGE 23

REPORT OF INVESTIGATION NO. 104 17 10 0 10Miles 10 0 10KilometersHydraulic Head Difference (feet) -32 -12 -12 10 10 30 30 50 50 70 70 90 90 112 Water Features Figure 10. Hydraulic head difference between the SAS (Arthur et al., 2005, in preparation) and the FAS potentiometric surface (Sepulveda, 2002).

PAGE 24

FLORIDA GEOLOGICAL SURVEY 18 Head Difference (feet) 17 112 -32 17 Water Features 10 0 10 5 Miles 10 0 10 5 Kilometers Figure 11. Map showing binary generalization of the hydraulic head difference evidential theme based on a contrast value of 17 ft (5.2 m).

PAGE 25

REPORT OF INVESTIGATION NO. 104 19 10 0 10 5 Miles 10 0 10 5 Kilometers Relative Vulnerability Water Features Wekiva Study Area More Vulnerable Vulnerable Figure 12. Relative vulnerability of the FAS FAVA model (Arthur et al., 2005, in preparation) showing zones of vulnerability based on the extent of the Floridan Aquifer System.

PAGE 26

FLORIDA GEOLOGICAL SURVEY 20 Wekiva study area is located in the “more vulnerable” zone. A smaller part of the study area is located in the “vulnerable” zone and none of the study area is located in the “less vulnerable” zone. By applying the FAVA modeling technique to the Wekiva study area (i.e., applying the WAVA model) it was possible to delineate refined zones of relative vulnerability by considering only the range of hydrogeological conditions in the study area as opposed to statewide hydrogeological conditions. Although the Wekiva study area should be considered “vulnerable” to “more vulnerable” in the context of statewide FAS vulnerability, the WAVA model refines these two degrees of vulnerability into three classes. Using the four evidential themes discussed above, a response theme was generated showing the posterior probability that a unit area contained a training point based on the evidential themes provided. A conceptual model showing the association between the training points and the evidential themes is shown in Figure 13. The posterior probabilities of the response theme ranged from 0.0007261 to 0.1042028 across the model domain. Plotting posterior probability against cumulative area as a percentage (Figure 14) allowed the delineation of class breaks for display of protection zones in the final response theme (Figure 15). The breaks for these protection zones were selected where a notable stepwise increase in posterior probability relative to the cumulative study area occurred. The first break, which delineated the less vulnerable zone from the vulnerable zone, occurred at a posterior probability value of 0.0000990. The less vulnerable zone represents approximately 25.7% of the study area. The second break delineating the vulnerable zone from the more vulnerable zone occurred at the next significant stepwise increase in posterior probability at a value of 0.0017000, which corresponded with the prior probability. This vulnerable zone represents approximately 55.8% of the study area. The remainder of the study area includes the more vulnerable zone and represents approximately 18.4% of the study area. This more vulnerable zone contains the greatest probability of containing a training point, and is therefore considered the most vulnerable to contamination from land surface. For the purposes of this project, three vulnerability classes were delineated; however, based on the methodology used to select these classes, a third probability break could be delineated, representing less than 3% of the study area. This area would represent the most vulnerable part of model study area and would be located principally in the “more vulnerable” area defined in the FAVA study. The response theme indicated that the areas of highest vulnerability (high probabilities) tended to be associated with IAS sediments that are thin to absent, dense karst-feature distribution, high positive hydraulic head difference, and high soil permeability. Conversely, areas of lowest vulnerability (low probabilities) tended to be determined by thick IAS sediments, sparse karst-feature distribution, hydraulic head difference less than 17 feet (5.2 m) and low soil permeability values. In addition to the large surface-water bodies present in the study area, a multitude of other surface-water features occur (Figure 16). Many of these features represent areas of discharge. It is important to note that these discharging surface waters are not part of the aquifer, although they originate from it. The WAVA project was designed to predict the relative ability for a contaminant introduced at land surface to travel through soils, overburden, karst features, etc. to enter into the FAS. As a result, it is essential that the WAVA model never be applied to assess contamination of surface waters or discharge areas. As described earlier, an assumption is made when using WofE that there is conditional independence between the layers used as predictors. Conditional independence is violated when the presence of one evidential theme influences the probability of another evidential theme. The validity of posterior probability values depends upon the degree of conditional independence calculated for the model.

PAGE 27

REPORT OF INVESTIGATION NO. 104 21 Figure 13. WofE conceptual model of the FAS. The top four layers are evidential themes and the bottom layer is the response theme divided into zones of relative vulnerability. Yellow lines represent training points (wells) projected throughout the layers. Soil Permeability Karst Features IAS Thickness Hydraulic Head Difference WAVA Response Theme

PAGE 28

FLORIDA GEOLOGICAL SURVEY 22 0.00001 0.00010 0.00100 0.01000 0.10000 1.00000 0.0010.0020.0030.0040.0050.0060.0070.0080.0090.00100.00Cumulative Area (%) 0.000099 0.0017 More Vulnerable Vulnerable Less VulnerableFigure 14. Posterior probability class breaks (green dashed lines) representing the boundaries between zones of relative vulnerability. A fourth optional break occurs at a posterior probability of ~0.017, which would represent less than 3% of the study area. Evidential themes are considered independent of each other if the conditional independence value is around 1.00. For the WAVA project, appropriate conditional independence values fell within the range of 1.00 0.15 (Gary Raines, personal communication, 2003). Conditional independence was calculated at 1.15 indicating minimal dependence between evidential themes. Weights calculated for the evidential themes used in the WAVA model are included in Table 2. This table displays the evidential themes used, weights calculated for those evidential themes, as well as the contrast and confidence of the evidential theme. A positive weight indicates areas where training points were likely to occur, while a negative weight indicates areas where training points were not likely to occur. The contrast column is a combination of the highest and lowest weights (positive weight – negative weight) and is a measure of how well the generalized evidential themes predict training points. A Table 2. WofE final output table listing weights calculated for each evidential theme and their associated contrast and confidence values of the evidential themes. Evidential Theme W1 W2 Contrast Confidence Hydraulic Head Difference 0.7412 -1.2857 2.0269 1.8494 Buffered Effective Karst 1.3546 -0.5550 1.9096 2.3343 Soil Permeability 1.2564 -0.5396 1.7960 2.1955 IAS Thickness 0.8739 -0.7726 1.6465 1.8994

PAGE 29

REPORT OF INVESTIGATION NO. 104 23 10 0 10 5 Miles 10 0 10 5 Kilometers Relative Vulnerability Wekiva Study Area More Vulnerable Vulnerable Less Vulnerable Water Features Figure 15. Relative vulnerability predicted by the WAVA model showing more vulnerable, vulnerable and less vulnerable areas of the FAS.

PAGE 30

FLORIDA GEOLOGICAL SURVEY 24 10 0 10 5 Miles 10 0 10 5 Kilometers HydrographyRelative Vulnerability Wekiva Study Area More Vulnerable Vulnerable Less Vulnerable Water Features Figure 16. Relative vulnerability predicted by the WAVA model showing more vulnerable, vulnerable and less vulnerable areas of the FAS. Hydrography represents seasonal and permanent wetlands, ponds and small lakes.

PAGE 31

REPORT OF INVESTIGATION NO. 104 25 positive contrast that is significant, based on its confidence, suggests that a generalized evidential theme is a useful predictor. The confidence of the evidential theme is the contrast divided by its standard deviation and provides a useful measure of significance of the contrast because of the uncertainties of the weights and areas of missing data (Raines, 1999). The hydraulic head difference evidential theme had a stronger association with the training points (i.e., highest contrast) than the other evidential themes and was therefore the primary determinant in predicting areas of vulnerability in this model. Three of the evidential themes (buffered effective karst, confining unit thickness and soil permeability) indicated where training points were more likely to occur because the positive weights (W1) were stronger (had a greater absolute value) than the negative weights (W2). Confidence values, for each evidential theme, exceeded the target value of 1.282, which corresponds to a level of significance of approximately 90% (Table 1). In fact, all confidence values for the evidential themes exceeded the 95% level of significance (1.645). For the dissolved oxygen response theme all of the training points occurred within areas of higher vulnerability, thus supporting the results of the model output. The soil permeability evidential theme contains large areas of missing data particularly along the Orange and Seminole county border. To compensate for this in the statewide FAVA model, a nearest neighbor GIS function was employed, which was used to apply spatial statistics (Euclidean distance functions) to predict soil permeability data values for these areas. Although this worked well on a statewide scale, use of this process on a smaller area skewed the vulnerability where there were large areas of missing data. In the WAVA model for example, use of this nearest neighbor function caused these areas along the Orange and Seminole county border to be included in the more vulnerable zone of the response theme. However, if the nearest neighbor function was not used, the calculated response theme included these same areas in the vulnerable zone (Figure 17). Based on the uncertainty due to missing data, the nearest neighbor function was not used in the WAVA model. The response theme values in these areas were therefore based solely on IAS thickness, buffered effective karst features, and hydraulic head difference. As a result, the relative vulnerability in these areas of missing data has a higher degree of uncertainty when compared to other areas of the model, and as new data becomes available, the vulnerability of these areas can be re-assessed. Confidence Map There are two types of confidence used on the WofE model. Confidence of the evidential theme, as reported in Table 2 and discussed above in Generalization of Evidential Themes , equals the contrast divided by the standard deviation (a student T-test) for a given evidential theme (Raines, 1999). The second type of confidence can be calculated for each response theme by dividing the theme’s posterior probability by its total uncertainty (standard deviation). A confidence map can be generated based on these calculations. The confidence map for the WAVA response theme is displayed in Figure 18. Areas with a high posterior probability tend to have higher confidence values and therefore have a higher level of certainty with respect to predicting aquifer vulnerability. A small population of training points along with missing data raises the total uncertainty for the response theme which in turn lowers the confidence. A confidence (of posterior probability) map containing all evidential themes (Figure 18) can therefore contain a lower level of significance than those calculated for each separate evidential theme. Validation and Sensitivity Analysis For other aquifer vulnerability assessments completed by the FGS (Arthur et al., 2005, in preparation; Baker et al., 2005) models were validated by creating a training point theme based on another parameter that reflected vulnerability yet was independent of the original training point set. As mentioned above, the

PAGE 32

FLORIDA GEOLOGICAL SURVEY 26 10 0 10 5 Miles 10 0 10 5 Kilometers No Soil Permeability Data Relative Vulnerability Wekiva Study Area More Vulnerable Vulnerable Less Vulnerable Water Features Figure 17. Relative vulnerability predicted by the WAVA model showing areas where no soil permeability data exists. The relative vulnerability predicted in these areas could change (i.e., higher or lower) as new data becomes available.

PAGE 33

REPORT OF INVESTIGATION NO. 104 27 10 0 10 5 Miles 10 0 10 5 Kilometers Confidence 70% 75% 75% 80% 80% 90% Water Features Figure 18. Distribution of confidence values calculated for WAVA response theme.

PAGE 34

FLORIDA GEOLOGICAL SURVEY 28 WAVA response theme, based on calculated weights, was a better predictor of where training points were more likely to occur because the positive weights (W1) were stronger than the negative weights (W2). In contrast, results from using a dissolved nitrogen training point set as validation for the WAVA model, indicated where training points were less likely to occur because the negative weights (W2) were stronger than the positive weights (W1). This produced a different kind of model, and, as a result, was not comparable to the original WAVA response theme. The following methods, however, did prove effective in completing reliable validation and sensitivity analyses: comparison of dissolved oxygen and posterior probability, and comparison of the WAVA model with a model based on a subset of training-point data. Dissolved Oxygen Data vs. Posterior Probability Posterior probability values were compared with the dissolved oxygen dataset from which the training point theme was extracted. Dissolved oxygen median concentration values were plotted versus posterior probability values (Figure 19). As expected a positive trend was observed between posterior probability and dissolved oxygen values and a calculated Pearson’s correlation coefficient (r) of 0.689 indicates more than a 99% degree of statistical significance. 0.000 0.005 0.010 0.015 0.020 0.025 0.000.200.400.600.801.001.201.401.60 Dissolved Oxygen (mg/L) Figure 19. Posterior probability vs. dissolved oxygen median concentration values Random 75% Subset of Training Points If the WAVA evidential themes and training points were robust (i.e., not sensitive to subtle changes in the training data set), one would expect the response theme patterns for the full training data set and a subset to be similar. For this sensitivity test, a training point theme consisting of a random subset of 75% of the original training points was generated and the model was re-executed. Response themes generated for the random subset of points were broken into the three relative vulnerability zones using the methodology described in the Response Theme section. The subset response theme was used to evaluate the degree of correlation between the WAVA response theme and the response theme developed using the 75% subset

PAGE 35

REPORT OF INVESTIGATION NO. 104 29 of training points. This random subset yielded a prior probability of 0.0014. Weights were then recalculated for each evidential theme, class breaks were selected, and a response theme was generated (Figure 20). The pattern of posterior probabilities was nearly identical to the original total dissolved oxygen response theme. In order to compare quantitatively the WAVA response theme to the random 75% subset response theme, the kappa coefficient (Cohen, 1960) was applied. This parameter is used to measure the amount of spatial agreement between response themes while taking into account agreement that could have occurred by chance. Cross-tabulation matrices were used to calculate the kappa coefficient (see also Weights of Evidence Glossary ). Values along the diagonal in the initial matrix (Table 3; see bold numbers) reflect the amount of agreement between response themes by area (in square meters). The remaining values in the table reflect where the response themes were mismatched. The matrix allows further matrix calculations of observed and expected proportions (i.e., agreement). The kappa coefficient between the two response themes equals 0.979 and less than 2% of the study area is mismatched. Kappa coefficient results can range between -1 (perfect disagreement) and 1 (perfect agreement). A value of zero indicates that the agreement was no better than that expected due to chance (Bonham-Carter 1994). Kappa coefficients calculated in the WAVA project were all positive values. Positive kappa coefficients can be interpreted using Table 4. Based on results of the kappa coefficient, the two response themes are in “almost perfect agreement.” This finding, as well as the correlation between dissolved oxygen and posterior probability provides strong validation for the WAVA model. Table 3. Cross-tabulation matrix of the area in square meters per class of the WAVA response theme and 75% subset response theme. Values along the diagonal reflect the amount of agreement. More VulnerableVulnerableLess VulnerableTotal More Vulnerable 608,123,70043,521,300 0 651,645,000Vulnerable01,974,464,10001,974,464,100Less Vulnerable0 0910,533,600910,533,600Total608,123,7002,017,985,400910,533,6003,536,642,700Class 75% Subset Response Theme Table 4. Kappa coefficient values and their associated interpretation (Landis and Koch, 1977). Interpretation of kappa values for positive agreement Kappa Interpretation < 0 No agreement 0.0 – 0.19 Poor agreement 0.20 – 0.39 Fair agreement 0.40 – 0.59 Moderate agreement 0.60 – 0.79 Substantial agreement 0.80 – 1.00 Almost perfect agreement

PAGE 36

FLORIDA GEOLOGICAL SURVEY 30 10 0 10 5 Miles 10 0 10 5 Kilometers Relative Vulnerability More Vulnerable Vulnerable Less Vulnerable Wekiva Study Area Water Features Figure 20. Relative vulnerability of the FAS predicted by the WAVA model using a random 75% subset of the original dissolved oxygen training point theme.

PAGE 37

REPORT OF INVESTIGATION NO. 104 31 Discussion and Summary Results of the statewide FAVA model (Arthur et al., 2005, in preparation) predict that the Wekiva study area lies within the “vulnerable” to “more vulnerable” zones of relative vulnerability (Figure 12). The WAVA map, however, provides more detail due to the smaller study area and refined datasets. Using FAVA as a point of statewide reference, the entire Wekiva study area should be considered at a minimum “vulnerable.” Even the “less vulnerable” zones predicted by the WAVA map are “vulnerable” with respect to the findings of the FAVA study. In agreement with hydrogeological principles, the “more vulnerable” zones predicted by the WAVA model (Figure 15) correspond to areas of relatively thin confinement, high karst-feature density, higher soil permeability and larger differences in hydraulic head between the water table and the FAS potentiometric surface. Within the Wekiva study area, the largest and most laterally continuous zone predicted as “more vulnerable” occurs approximately three miles (~4.8 kilometers [km]) north-northeast of Lake Apopka; whereas much of the “less vulnerable” areas are predicted to occur along the northeastern part of the Wekiva study area as well as east and south of Lake Apopka. Several “more vulnerable” areas located outside of the Wekiva study area yet within the WAVA study area correspond to population centers, such as Orlando. Future land-use planning in these areas will hopefully consider the results of the WAVA model as measures are developed to protect local ground-water resources. In summary, all aquifer systems in Florida are vulnerable to contamination. This is not only attributable to Florida’s hydrogeologic setting but also to human modifications of the natural system. The WAVA project results provide a probability map that identifies zones of relative vulnerability in the study area based on the evidential themes and training points used as input for the model. Because the model input consists only of natural systems characteristics that affect aquifer vulnerability, the WAVA output map is consequently an approximation of intrinsic aquifer vulnerability (i.e., it does not account for sources of contamination or behavior of contaminants). The methodology used to produce the WAVA map is a data-driven process, which minimizes the potential for scientific bias. The maps are as accurate as the most detailed input layer, and as inaccurate as the least detailed layer. It is the responsibility of the end-users of these maps to determine specific and appropriate applications (see also Disclaimer ). It is recommended that the WAVA vulnerability map should be applied at scales smaller than 1:200,000 (i.e., 1:500,000). Implementing this scale recommendation prohibits application of the vulnerability map to assess individual land parcels. Although much of the input data and the output vulnerability map are accurate to a raster resolution of 30 m2 (322.9 ft2), use of the map at this scale is not suggested. Application of the vulnerability map on the order of greater than .75 square miles (1.94 km2) is more appropriate. In other words, the WAVA vulnerability map is no substitute for site specific hydrogeological investigation. The relative vulnerability zones delineated in this report may be applied to facilitate designation of "protection zones" by regulatory agencies. These “protection zones,” however, would likely include other factors affecting the hydrologic system, including ground-water flow gradients and proximity to impacted areas. The WAVA vulnerability map can also be used to develop measures designed to protect the Floridan Aquifer System, which provides flow to the Wekiva River and Wekiwa Spring. The model results have far reaching implications and can be applied to a variety of environmental management, protection and conservation activities, including: Wellhead protection Source-water protection Recharge protection

PAGE 38

FLORIDA GEOLOGICAL SURVEY 32 Vulnerability indices Contaminant-specific maps Land conservation acquisition Total maximum daily loads (TMDLs) Surface-water/ground-water interactions Water-quality management tool Resource planning strategies and policies Prioritization of areas of critical concern Design of monitoring plans Best Management Practices The results of the WAVA vulnerability model are useful for development and implementation of groundwater protection measures; however, the vulnerability output map included in this report should not be viewed as a static evaluation of the vulnerability of the Floridan Aquifer System. The model was built using the best data available at the time this project was completed and as a result, represents a “snapshot” in time. As new and more refined data become available for the WAVA study area, this data can be incorporated into future modeling scenarios. The advantages of including new or enhanced data include a more refined output vulnerability map with higher confidence in the results. Data improvements could include input of detailed topographic data (i.e., LIDAR), collection of soil permeability data for the “no data” areas along the Seminole-Orange County border (Figure 17) or strategically collecting water-level data to improve the hydraulic head difference evidential theme. Enhancement of these data layers would yield significant improvements in refinement, accuracy and statistical confidence of the final model output. The more fine-tuned the input data, the more detailed and reliable the model results can be. The weights of evidence modeling approach to aquifer vulnerability is a highly adaptable and useful tool for implementing ongoing protection of Florida’s vulnerable ground-water resources. Disclaimer The WAVA map was developed by the FDEP/FGS to carry out agency responsibilities related to management, protection, and responsible development of Florida's natural resources. Although efforts have been made to make the information in these maps accurate and useful, the FDEP/FGS assumes no responsibility for errors in the information and does not guarantee that the data are free from errors or inaccuracies. Similarly FDEP/FGS assumes no responsibility for the consequences of inappropriate uses or interpretations of the data on these maps. As such, these maps are distributed on an "as is" basis and the user assumes all risk as to their quality, the results obtained from their use, and the performance of the data. FDEP/FGS further makes no warranties, either expressed or implied as to any other matter whatsoever, including, without limitation, the condition of the product, or its suitability for any particular purpose. The burden for determining suitability for use lies entirely with the user. In no event shall the FDEP/FGS have any liability whatsoever for payment of any consequential, incidental, indirect, special, or tort damages of any kind, including, but not limited to, any loss of profits arising out of use of or reliance on the maps or support by FDEP/FGS. FDEP/FGS bears no responsibility to inform users of any changes made to this data. Anyone using this data is advised that resolution implied by the data may far exceed actual accuracy and precision. Comments on this data are invited and FDEP/FGS would appreciate that documented errors be brought to the attention of our staff. The development of these data sets represents a major investment of staff time, effort and funds. As a professional responsibility, we expect that the FDEP/FGS will receive proper credit when these data sets are used. Further, since part of this data was developed and collected with State of Florida funding, no proprietary rights may be attached to it in whole or in part, nor may it be sold to the U.S. Government or the Florida State Government as part of any procurement of products or services.

PAGE 39

REPORT OF INVESTIGATION NO. 104 33 WEIGHTS OF EVIDENCE GLOSSARY Binary – Refers to the generalization or simplification of evidential themes or data layers. Binary layers are reclassified from the original dataset into presence/absence type themes or into two classes. Conditional Independence – Occurs when an evidential theme does not affect the probability of another evidential theme. Evidential themes are considered independent of each other if the conditional independence value calculated is within the range 1.00 0.15 (Raines, personal communication, 2003). Values that significantly deviate from this range can inflate the posterior probabilities resulting in unreliable response themes. Confidence of evidential theme – Contrast divided by its estimated standard deviation; provides a useful measure of significance of the contrast. Confidence of Posterior Probability – A measure based on the ratio of posterior probability to its estimated standard deviation. Contrast – W+ minus W(see weights), which is an overall measure of the spatial association (correlation) of an evidential theme with the training points. Cumulative Ascending – Calculates the cumulative weights from the first class to the last class while increasing the area. Areas nearest a training point have a stronger association, and those farthest away have a weaker association. This method is applicable for themes where the training points are mainly associated with the lower values of the evidential theme (e.g., higher vulnerability correlates with lower confinement thickness). Cumulative Descending – Calculates the cumulative weights from the last class to the first class while increasing the area (opposite of cumulative ascending). This method is applicable for themes where the training points are mainly associated with the higher values of the evidential theme (e.g., higher vulnerability correlates with higher soil permeability). Evidential Theme – A set of continuous spatial data that is associated with the location and distribution of known occurrences (i.e., training points); these map data layers are used as predictors of vulnerability. Kappa Coefficient – Allows statistical comparison of map patterns. It is a multivariate accuracy assessment technique developed by Cohen (1960) to determine if one error matrix is significantly different than another. LIDAR – Light detection and ranging; an airborne remote sensing system that uses laser light to measure land-surface elevation with a high degree of accuracy and resolution. LIDAR data can be used to generate digital elevation models. Model – The characteristics of a set of training points, and the relationships of the training points to a collection of evidential themes. Posterior Probability – The probability that a unit cell contains a training point after consideration of the evidential themes. This measurement changes from location to location depending on the values of the evidence.

PAGE 40

FLORIDA GEOLOGICAL SURVEY 34 Prior Probability – The probability that a unit cell contains a training point before considering the evidential themes. It is a constant value over the study area equal to the training point density (total number of training points divided by total study area in unit cells). Response Theme – An output map that displays the probability that a unit area would contain a training point, estimated by the combined weights of the evidential themes. The output is displayed in classes of relative aquifer vulnerability or favorability to contamination (i.e., this area is more vulnerable than that area) or favorability. The response theme is the relative vulnerability map. Spatial Data – Information about the location and shape of, and relationships among, geographic features, usually stored as coordinates and topology. Study Area – A grid theme that acts as a mask to define the area where the model is developed and applied. It may be irregular in outline and may contain interior holes (e.g., lakes and no data areas). Training Points – A set of locations (points) reflecting a parameter used to calculate weights for each evidential theme, one weight per class, using the overlap relationships between points and the various classes. In an aquifer vulnerability assessment, training points are wells with one or more water quality parameters indicative of relatively higher recharge which is an estimate of relative vulnerability. Weights – A measure of an evidential-theme class. A weight is calculated for each theme class. For binary themes, these are often labeled as W+ and W-. For multiclass themes, each class can also be described by a W+ and Wpair, assuming presence/absence of this class versus all other classes. Positive weights indicate that more points occur on the class than due to chance, and the inverse for negative weights. The weight for missing data is zero. Weights are approximately equal to the proportion of training points on a theme class divided by the proportion of the study area occupied by theme class, approaching this value for an infinitely small unit cell.

PAGE 41

REPORT OF INVESTIGATION NO. 104 35 REFERENCES Arthur, J.D., Baker, A.E., Cichon, J.R., Wood, H.A.R., and Rudin, A., 2005, in preparation, Florida Aquifer Vulnerability Assessment (FAVA): Contamination potential of Florida’s principal aquifer systems: Florida Geological Survey Bulletin 67. Baker, A.E., Wood, A.R., Cichon, J.R., and Arthur, J.D., 2005, Alachua Aquifer Vulnerability Assessment; a final report submitted to Alachua County, January, 2005, 36 p. (unpublished). 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 University, Institute of Public Affairs, 312 p. Bonham-Carter, G. F., 1994, Geographic Information Systems for Geoscientists, Modeling with GIS: Oxford, Pergamon, 398 p. Cichon, J.R., Wood, H.A.R., Baker, A.E., and Arthur, J.A., 2004, Application of Geologic Mapping and Geographic Information Systems to Delineate Sensitive Karst Areas for Land-Use Decisions: American Geological Institute website, http://www.agiweb.org/environment/publications/ mapping/graphics/florida.pdf, 2004. Cohen, J., 1960, A coefficient of agreement for nominal scales: Educational and Psychological Measurement, v. 20, no. 1, p. 37-46. Davis, J., Johnson, R., Boniol, D., and Rupert, F., 2001, Guidebook to the Correlation of Geophysical Well Logs within the St. Johns River Water Management District: Florida Geological Survey Special Publication No. 50, 114 p. Florida Geographic Data Library, 2003, GeoPlan Center, Gainesville, University of Florida: www.fgdl.org/. Kemp, L.D., Bonham-Carter, G.F., Raines, G.L. and Looney, C.G., 2001, Arc-SDM: Arcview extension for spatial data modeling using weights of evidence, logistic regression, fuzzy logic and neural network analysis: http://ntserv.gis.nrcan.gc.ca/sdm/, 2002. Landis, J.R. and Koch, G.G., 1977, The measurement of observer agreement for categorical data: Biometrics, v. 33, p. 159-174. National Research Council, 1993, Ground Water Vulnerability Assessment: Predicting Relative Contamination Potential under Conditions of Uncertainty: Washington, National Academy Press, 204 p. Raines, Gary L., 1999, Evaluation of Weights of Evidence to Predict Epithermal-Gold Deposits in the Great Basin of the Western United States: Natural Resources Research, v. 8, no. 4, p. 257-276. Raines, G. L., Bonham-Carter, G. F., and Kemp, L., 2000, Predictive Probabilistic Modeling Using ArcView GIS: ArcUser, v. 3, no.2, p. 45-48.

PAGE 42

FLORIDA GEOLOGICAL SURVEY 36 Scott, T.M., Means, G.H., Meegan, R.P., Means, R.C., Upchurch, S.B., Copeland, R.E., Jones, J., Roberts, T., and Willet, A., 2004, Springs of Florida: Florida Geological Survey Bulletin No. 66, 377 p. Sepulveda, N., 2002, Simulation of Ground-Water Flow in the Intermediate and Floridan Aquifer Systems in Peninsular Florida: U.S. Geological Survey Water-Resource Investigation Report 024009, 130 p. Southeastern Geological Society Ad Hoc Committee, 1986, Hydrogeological Units of Florida: Florida Geological Survey Special Publication No. 28, 8 p. U.S. Census Bureau: State and County Quick Facts, 01-Feb-2005, 15:48:47 EST: www.census.gov. Wekiva River Basin Coordinating Committee, 2004, Wekiva River Basin Coordinating Committee Final Report: Recommendations for Enhanced Land Use Planning Strategies and Development Standards to Protect Water Resources of the Wekiva River Basin, March 16, 2004, 50 p. (unpublished). Wright, A.P., 1974, Environmental Geology and Hydrology of the Tampa Area, Florida: Florida Geological Survey Special Publication No. 19, 93 p.