Citation
Geo-Spatial Assessment of the Impact of Land Cover Dynamics and Distribution of Land Resources on Soil and Water Quality in the Santa Fe River Watershed

Material Information

Title:
Geo-Spatial Assessment of the Impact of Land Cover Dynamics and Distribution of Land Resources on Soil and Water Quality in the Santa Fe River Watershed
Creator:
SABESAN, AARTHY
Copyright Date:
2008

Subjects

Subjects / Keywords:
Agriculture ( jstor )
Land cover ( jstor )
Land use ( jstor )
Montane forests ( jstor )
Pixels ( jstor )
Soils ( jstor )
Surface water ( jstor )
Water quality ( jstor )
Watersheds ( jstor )
Wetlands ( jstor )
Suwannee River, FL ( local )

Record Information

Source Institution:
University of Florida
Holding Location:
University of Florida
Rights Management:
Copyright Aarthy Sabesan. Permission granted to the University of Florida to digitize, archive and distribute this item for non-profit research and educational purposes. Any reuse of this item in excess of fair use or other copyright exemptions requires permission of the copyright holder.
Embargo Date:
6/30/2005
Resource Identifier:
436098606 ( OCLC )

Downloads

This item is only available as the following downloads:


Full Text

PAGE 1

GEO-SPATIAL ASSESSMENT OF THE IMPACT OF LAND COVER DYNAMICS AND DISTRIBUTION OF LA ND RESOURCES ON SOIL AND WATER QUALITY IN THE SANTA FE RIVER WATERSHED By AARTHY SABESAN A THESIS PRESENTED TO THE GRADUATE SCHOOL OF THE UNIVERSITY OF FLOR IDA IN PARTIAL FULFILLMENT OF THE REQUIREMENTS FOR THE DEGREE OF MASTER OF SCIENCE UNIVERSITY OF FLORIDA 2004

PAGE 2

To my mom and dad

PAGE 3

ACKNOWLEDGMENTS No work in science can afford to be a total personal achievement. First, I thank my advisor, Dr. Sabine Grunwald, for all the generous guidance, suggestions and enthusiasm that were crucial for the completion of this thesis. The research and teaching experiences that I have gained in the last two years are invaluable and I am grateful to her for giving me all these wonderful opportunities. I would specially like to thank my committee members, Dr. Michael Binford and Dr. Mark Clark, for their valuable input and guidance. Funding for this thesis from the USDA is greatly acknowledged. I thank all my wonderful colleagues at the GIS research lab for making this journey an enjoyable one. Special thanks go to Sanjay Lamsal for all his efforts in ground truth data collection. I thank Dr. Christine Bliss and Dr. Isabelle Lopez for the laboratory analysis. A very special note of thanks goes to Kathleen McKee and Rosanna Rivero for all the stimulating debates and guidance. I would like to thank my family for all their support and motivation that kept me going. My heartfelt gratitude goes to Mr. Joseph, my high school chemistry teacher, for inspiring me to perform better than I thought I could. Last but not least I thank my friends Latha, Manu, Suresh, Aparajithan, and Kavitha for keeping me sane during the whole time. iii

PAGE 4

TABLE OF CONTENTS page ACKNOWLEDGMENTS.................................................................................................iii LIST OF TABLES............................................................................................................vii LIST OF FIGURES...........................................................................................................ix ABSTRACT.....................................................................................................................xiv CHAPTER 1 INTRODUCTION........................................................................................................1 1.1 Research Significance.............................................................................................6 1.2 Hypotheses..............................................................................................................6 1.3 Objectives...............................................................................................................6 2 WATERSHED CHARACTERISTICS........................................................................8 2.1 Introduction.............................................................................................................8 2.2 Population...............................................................................................................8 2.3 Climate..................................................................................................................12 2.4 Topography...........................................................................................................12 2.5 Physiographic Regions.........................................................................................12 2.6 Hydro-Geologic Framework.................................................................................13 2.7 Soils......................................................................................................................17 2.8 Land Use...............................................................................................................19 3 LANDCOVER DYNAMICS IN THE SFRW...........................................................23 3.1 Definitions of Land Use and Land Cover.............................................................23 3.2 Impact of Land Cover Change in the SFRW........................................................24 3.3 Significance..........................................................................................................24 3.4 Objectives.............................................................................................................25 3.5 Materials...............................................................................................................25 3.5.1 Landsat Imageries.......................................................................................25 3.5.2 Software......................................................................................................26 3.5.3 Auxiliary Data............................................................................................26 3.5.4 Field Sampling Instruments........................................................................27 iv

PAGE 5

3.6 Methods................................................................................................................27 3.6.1 Design of Classification Scheme................................................................27 3.6.2 Ground Truth Data Collection....................................................................30 3.6.3 Image Analysis...........................................................................................33 3.6.4 Accuracy Assessment.................................................................................71 3.6.5 Change Detection.......................................................................................77 3.7 Discussions and Conclusions................................................................................84 3.7.1 Accuracy of Image Classification..............................................................84 3.7.2 Land Cover Change Analysis.....................................................................92 4 ANALYSIS OF SOIL NITRATE-NITROGEN VALUES......................................103 4.1 Introduction.........................................................................................................103 4.2 Sampling Site Selection......................................................................................105 4.2.1 Assumptions.............................................................................................105 4.2.2 Materials...................................................................................................106 4.2.3 Methods....................................................................................................106 4.3 Replacement Sites...............................................................................................118 4.4 Field Sampling....................................................................................................121 4.4.1 Materials...................................................................................................121 4.4.2 Methods....................................................................................................122 4.5 Laboratory Analysis............................................................................................123 4.6 Analysis of Soil Sampling Results.....................................................................124 4.6.1 Exploratory Spatial Data Analysis...........................................................124 4.6.2 Interpolation Based Prediction of Nitrate-Nitrogen Values in SFRW.....135 4.6.3 Pixel Based Prediction of Nitrate-Nitrogen Values..................................153 4.7 Discussions and Conclusions..............................................................................156 5 ANALYSIS OF SURFACE WATER QUALITY...................................................161 5.1 Introduction.........................................................................................................161 5.2 Objectives...........................................................................................................166 5.3 Materials.............................................................................................................166 5.3.1 Software....................................................................................................166 5.3.2 Water Quality Data...................................................................................167 5.3.3 Attributes for the Sub-Watershed Characteristics Database....................167 5.4 Methods..............................................................................................................167 5.4.1 Processing Surface Water Quality Data...................................................167 5.4.2 Delineating Upslope Drainage Area.........................................................169 5.4.3 Creating Sub-Watershed Characteristics Database..................................175 5.4.4 Analysis of Correlations...........................................................................176 5.4.5 Comparison of Watershed Characteristics...............................................176 5.5 Discussion and Conclusions...............................................................................188 6 SYNTHESIS.............................................................................................................197 v

PAGE 6

APPENDIX A HEADER FILE OF LANDSAT IMAGES...............................................................202 B FLUCCS LULC CLASSES......................................................................................207 C LOCATION OF GROUND CONTROL POINTS...................................................211 D U.S.G.S SURFACE WATER QUALITY MONITORING REPORTS...................219 LIST OF REFERENCES.................................................................................................223 BIOGRAPHICAL SKETCH...........................................................................................231 vi

PAGE 7

LIST OF TABLES Table page 3.1 Spectral locations of Landsat bands.........................................................................26 3-2 Image interpretation formats for Anderson LULC classes......................................29 3-3 Aggregated LULC classes to create final land cover classes...................................32 3.4 Accuracy assessment report for 2000 classified image............................................75 3.5 Accuracy assessment report for 2003 classified image............................................76 3-6 Areal extents of land cover classes in 1990.............................................................78 3-7 Areal extents of land cover classes in 2000.............................................................79 3-8 Areal extents of land cover classes in 2003.............................................................80 3.9 Comparison of the areal extend of land cover classes..............................................81 3-10 Land cover shifts between 1990-2000.....................................................................82 3-11 Land cover shifts between 2000-2003.....................................................................82 3-12 Land cover categories of change..............................................................................83 3-13 Aerial extend of land cover change class.................................................................86 4.1 Projection parameters.............................................................................................108 4-2 Reclassification of LULC classes..........................................................................111 4-3 Area under each LULC-soil combination category...............................................116 4-4 Recalculated LULC-soil percentages.....................................................................118 4-5 Allocation of sample sites......................................................................................118 4.6 Sample sites per each LULC-soil combination......................................................119 4-7 Summary statistics of nitrate-nitrogen values........................................................128 vii

PAGE 8

4-8 Summary of resulting prediction errors using different interpolation techniques for layer 1 nitrate-nitrogen concentrations...........................................144 4-9 Summary of resulting prediction errors using different interpolation techniques for layer 2 nitrate-nitrogen concentrations...........................................146 4-10 Summary of resulted prediction errors using different interpolation techniques for layer 3 nitrate-nitrogen concentrations...........................................148 4-11 Summary of resulting prediction errors using different interpolation techniques for layer 4 nitrate-nitrogen concentrations...........................................150 4-12 Summary of resulting prediction errors using different interpolation techniques for average nitrate-nitrogen concentrations..........................................152 4-13 Average nitrate-nitrogen values for each LULC-soil combination........................154 4-14 Average nitrate-nitrogen value for the LULC classes............................................156 5-1 Description of station ID........................................................................................168 5-2 Results of the correlation analysis..........................................................................177 5-3 Stream discharge at monitoring stations................................................................191 viii

PAGE 9

LIST OF FIGURES Figure page 1-1 Location of the SFRW within the SRWMD..............................................................3 1-2 Contributions of SFRW to nitrate-nitrogen loads into the Gulf of Mexico...............3 2-1 Location of the SFRW in the State of Florida............................................................9 2-2 Counties and prominent city limits in the SFRW....................................................10 2-3 Population growth in the SRWMD..........................................................................11 2-4 Population density in SFRW....................................................................................11 2-5 Elevation across SFRW............................................................................................13 2-6 Slope across SFRW..................................................................................................14 2-7 Physiographic regions in SRWMD..........................................................................14 2-8 Environmental geology in SFRW............................................................................15 2-9 Cross-section of a karst section in the Suwannee River Basin................................16 2-10 DRASTIC Index values for SFRW..........................................................................18 2-11 Soil orders in SFRW................................................................................................21 2-12 1995 LULC map of the SFRW................................................................................22 3-1 LULC classes in 1995 SFRW..................................................................................31 3-2 Location of ground control points............................................................................33 3-3 Pictures of land cover classes in SFRW...................................................................34 3-4 Illustration of the accuracy of the geometric correction process..............................39 3-5 Geometrically corrected and subset composite image of 2003 SFRW....................40 3-6 Radiometric response of TM channel......................................................................42 ix

PAGE 10

3-7 Conversion of raw DN values to at-satellite radiance..............................................45 3-8 Spatial model used in the conversion of DN to reflectance values..........................47 3-9 Conversion of at-satellite radiance values to at-satellite reflectance values............48 3-10 Paths of satellite radiance.........................................................................................49 3-11 Shifting of origins during DOS................................................................................52 3-12 Application of DOS method.....................................................................................54 3-13 Cloud mask...............................................................................................................55 3-14 IFOV of satellite system...........................................................................................57 3-15 Road mask................................................................................................................60 3-16 Road masked SFRW................................................................................................61 3-17 Working of the MDM classifier...............................................................................64 3-18 Steps in supervised classification of image data......................................................65 3-19 Mean plot of land cover classes for the summer scene............................................67 3-20 Mean plot of land cover classes for the winter scene...............................................68 3-21 Transformed divergence distance values for the winter scene.................................69 3-22 Transformed divergence distance values for the summer scene..............................70 3-23 Land cover map of 1990 SFRW...............................................................................72 3-24 Land cover map of 2000 SFRW...............................................................................73 3-25 Land cover map of 2003 SFRW...............................................................................74 3-26 Percent distribution of land cover classes in 1990...................................................78 3-27 Percent distribution of land cover classes in 2000...................................................79 3-28 Percent distribution of land cover classes in 2003...................................................80 3-29 Trajectories of land cover change in SFRW............................................................85 3-30 Percent distribution of land cover change categories...............................................86 3-31 Percent distribution of soil orders in NAA change class.........................................87 x

PAGE 11

3-32 Percent distribution of geological classes in NAA change class..............................87 3-33 Percent distribution of soil orders in NNA change class.........................................88 3-34 Percent distribution of geological classes in NNA change class..............................88 3-35 Percentage of LC shifts between classes between 1990 and 2000...........................93 3-36 Estimated total N inputs in Columbia County.........................................................97 3-37 Estimated total N inputs in Suwannee County.........................................................98 3-38 Estimated total N inputs in Alachua County............................................................99 3-39 Estimated total N inputs in Gilchrist County.........................................................101 4-1 Nitrogen cycle........................................................................................................104 4-2 Design of sampling site selection protocol............................................................110 4-3 Reclassified LULC map.........................................................................................112 4.4 SSURGO database schema....................................................................................113 4-5 Soil orders in SFRW..............................................................................................114 4-6 Selected sampling sites using the protocol.............................................................119 4-7 Approved sites for the fall 2003 sampling event....................................................121 4-8 Composite soil sampling scheme for each sampling location................................123 4-9 Histogram of nitrate-nitrogen values in layer 1.....................................................125 4-10 Histogram of nitrate-nitrogen values in layer 2.....................................................126 4-11 Histogram of nitrate-nitrogen values in layer 3.....................................................126 4-12 Histogram of nitrate-nitrogen values in layer 4.....................................................127 4-13 Histogram of average nitrate-nitrogen value in each profile..................................127 4-14 Normal Q-Q plot of the nitrate-nitrogen values in layer 1.....................................129 4-15 Normal Q-Q plot of the nitrate-nitrogen values in layer 2.....................................129 4-16 Normal Q-Q plot of the nitrate-nitrogen values in layer 3.....................................130 4-17 Normal Q-Q plot of the nitrate-nitrogen values in layer 4.....................................130 xi

PAGE 12

4-18 Normal Q-Q plot of the average nitrate-nitrogen value at each site......................131 4-19 Mean Voronoi polygons for nitrate-nitrogen values in layer 1..............................132 4-20 Cluster voronoi polygons for nitrate-nitrogen values in layer 1............................133 4-21 Cluster voronoi polygons for nitrate-nitrogen values in layer 2............................133 4-22 Cluster voronoi polygons for nitrate-nitrogen values in layer 3............................134 4-23 Cluster voronoi polygons for nitrate-nitrogen values in layer 4............................134 4-24 Cluster voronoi polygons for the average nitrate-nitrogen value...........................135 4-25 Optimizing the power value...................................................................................138 4-26 Search neighborhood..............................................................................................139 4-27 Eight sector neighborhood interpolator..................................................................140 4-28 Fitting of piece wise curves in RBF.......................................................................141 4-29 Interpolated surface of nitrate-nitrogen concentrations across the SFRW in layer 1.....................................................................................................................145 4-30 Interpolated surface of nitrate-nitrogen concentrations across the SFRW in layer 2.....................................................................................................................147 4-31 Interpolated surface of nitrate-nitrogen concentrations across the SFRW in layer 3.....................................................................................................................149 4-32 Interpolated surface of nitrate-nitrogen concentrations across the SFRW in layer 4.....................................................................................................................151 4-33 Interpolated surface of the profile averages of nitrate-nitrogen concentrations across the SFRW....................................................................................................153 4-34 Illustration of the pixel based prediction technique...............................................153 4-35 Pixel based prediction of nitrate-nitrogen values in SFRW...................................155 5-1 Nutrient loadings by reaches/basins in the Suwannee River Basin.......................163 5-2 Mean nitrate-nitrogen concentration of the Floridian aquifer ...............................165 5-3 Location of surface water quality monitoring stations...........................................169 5-4 Temporal variation of nitrate-nitrogen concentrations in stations.........................170 xii

PAGE 13

5-5 Temporal variation of nitrate-nitrogen concentrations...........................................170 5-6 Spatial variation of long-term average nitrate-nitrogen concentrations.................171 5-7 Illustration of the upslope drainage area delineation.............................................172 5-8 Eight point pour model...........................................................................................173 5-9 Computation of flow direction grid........................................................................174 5-10 Flow accumulation grid for SFRW........................................................................175 5-11 Stream network......................................................................................................176 5-12 Delineated sub-watershed boundary......................................................................177 5-13 Location of three sub-watersheds...........................................................................178 5-14 Nitrate-nitrogen concentrations measured at ICH010C1.......................................179 5-15 Attributes of the sub-watershed draining into ICH010C1.....................................179 5-16 Nitrate-nitrogen concentrations measured at NEW007C1.....................................182 5-17 Attributes of the sub-watershed draining into NEW007C1...................................183 5-18 Nitrate-nitrogen concentrations measured at SFR02C1.........................................186 5-19 Attributes of the sub-watershed draining into SFR02C1.......................................186 5-20 Location of surface water monitoring stations in SRWMD...................................191 xiii

PAGE 14

Abstract of Thesis Presented to the Graduate School of the University of Florida in Partial Fulfillment of the Requirements for the Degree of Master of Science GEO-SPATIAL ASSESSMENT OF THE IMPACT OF LAND COVER DYNAMICS AND DISTRIBUTION OF LAND RESOURCES ON SOIL AND WATER QUALITY IN THE SANTA FE RIVER WATERSHED By Aarthy Sabesan December 2004 Chair: Sabine Grunwald Major Department: Soil and Water Science Since the early 90s, surface and ground water monitoring showed elevated nitrogen in the Santa Fe River Watershed (SFRW). The cause of degrading water quality in the SFRW has been attributed to non-point source pollution. The overall goal of this study was to gain a better understanding of soil and water nitrate conditions within the watershed and how it relates to spatially-distributed watershed characteristics. A holistic geo-spatial modeling approach was adopted integrating spatial layers of land resources (geology, soils, land use/land cover, topography, hydrographic features, etc.) in a Geographic Information System (GIS). Landsat TM and ETM+ satellite imagery from 1990, 2000, and 2003 was used to quantify the land cover shifts in the SFRW using a multi-temporal change detection analysis. A GIS-based stratified random sampling design was developed to target soil sampling locations spatially-distributed across the watershed and analyze soil samples collected at four different depths (0-30, 30-60, 60xiv

PAGE 15

120, 120-180 cm) for nitrate-nitrogen. Numerous geo-spatial interpolation techniques were tested and compared to identify the method producing the lowest prediction error estimating soil nitrate-nitrogen across the watershed. A spatially-explicit analysis was used to investigate relationships between soil nitrogen, land cover, land resource data and nitrate-nitrogen observed in surface water monitoring stations. Results indicate that multiple factors contribute to elevated nitrogen found in soils and water. Karst terrain, soil material, and agricultural and urban land uses pose the greatest risk for nitrate leaching. In addition the geographic position and spatial distribution of land resource factors and spatial interrelationships between factors influence nitrogen levels observed in soils and surface water. Understanding the interrelationships between land cover/land use, soils, geology, topography and other factors in a spatially-explicit context supports ongoing efforts to improve the water quality in the SFRW. xv

PAGE 16

CHAPTER 1 INTRODUCTION Increases in the concentration of nitrate-nitrogen in drinking water systems are the subject of concern for many residents in the Suwannee River Water Management District (SRWMD). More than 60% of the impaired rivers in the U.S are polluted due to excessive inputs of nitrogen and phosphorus (Carpenter et al., 1998). The Safe Drinking Water Act (SDWA) was initiated to protect and maintain the nation’s drinking water supply. This law enforces actions and many regulations to protect drinking water and its source – rivers, lakes, reservoirs, springs and ground water wells. The SDWA has authorized the U.S. Environmental Protection Agency (USEPA) to establish national standards of both natural and man-induced contaminant levels in drinking water. These enforceable values of allowable concentration values are referred to as the Maximum Concentration Limits (MCL). According to the USEPA, the maximum allowable value of nitrate-nitrogen level in drinking water supply is 10 mg/l. The USEPA also assists with state agencies and public water suppliers to set up multiple barriers to prevent water pollution. The barriers include source water protection, treatment, distribution system integrity and public information (USEPA, 1999). Ground water and surface water quality are monitored in the SRWMD by a network of monitoring wells distributed across its area. During February 1995 (high flow conditions) concentrations of nitrate plus nitrite ranged from 0.05 to 0.38 mg/l. In June 1995 (low flow conditions), they ranged form 0.07 to 1.05 mg/l. Highest concentration was observed in Branford (Raulston et al., 1998) which lies within the SFRW. In water 1

PAGE 17

2 year 1998 the Santa Fe Reach 2 showed an annual nitrate-nitrogen load of 1,130 tons/year, accounting for a total of 15.9% of the nitrate-nitrogen load into the Gulf of Mexico and the Santa Fe Reach 1 showed an annual nitrate-nitrogen load of 65.6 tons/year, accounting for an additional total of 0.9% of the nitrate-nitrogen load the Suwannee delivers into the Gulf of Mexico (SRWMD, 1998). Of the 2,676 tons of nitrate-nitrogen that were transported to the Gulf of Mexico through tributaries of the Suwannee River basin, the Santa Fe accounted for 593 tons of nitrate-nitrogen in water year 2000 (SRWMD, 2000). In water year 2001, a total of 3,067 tons of nitrate-nitrogen was transported to the Gulf of Mexico from the Aucilla, Econfina, Fenholloway, Steinhatchee, Suwannee, and Waccasassa Rivers. Of the 3,067 tons of nitrate-nitrogen, the Suwannee River Basin accounted for 2,999 tons of nitrate-nitrogen. The contributions of the Santa Fe reach 2 to the above were 15.8%. The location of the Santa Fe reach 1 and 2 in the SRWMD is shown in Figure 1-1. Based on the surface water quality report by the SRWMD for 2003, surface water measurements were recorded at 67 stations in the basin. In water year 2002, around 3,012 tons of nitrate-nitrogen were transported to the Gulf of Mexico from the Aucilla, Econfina, Fenholloway, Steinhatchee, Suwannee, and Waccasassa Rivers. Among these river basins, the Suwannee river basin accounted for about 2,971 tons of nitrate-nitrogen. Within the Suwannee River basin, the Santa Fe River Reach 2 accounted for 19.6% of the annual nitrate-nitrogen load delivered to the Gulf by the Suwannee River, but covers only 5.7% of the total basin area. This implies a 25% increase in nitrate-nitrogen transport from the Santa Fe Reach 2 to the Gulf of Mexico in 2001-2003. Over two decades of monitoring the water quality of major rivers of the Suwannee River Basin has indicated a statistically significant (at 95% confidence

PAGE 18

3 interval) increasing trend in the concentrations of nitrate-nitrogen (Ham and Hatzell, 1996; SRWMD, 1998). Figure 1-1. Location of the SFRW within the SRWMD. Figure 1-2. Contributions of SFRW to nitrate-nitrogen loads into the Gulf of Mexico.

PAGE 19

4 Ground water is monitored by a network of 97 monitor wells distributed over the SRWMD. The Suwannee and Lafayette counties have consistently exhibited elevated nitrate-nitrogen concentrations. The area around the Santa Fe reach 2 is increasingly impacted by nutrients under low flow conditions when the river receives a large quantity of groundwater flow through springs and seeps in the riverbed. Also, some site specific studies, conducted by the Department of Health, have identified wells with concentrations of nitrate-nitrogen above the MCL (SRWMD, 2003). Along the Suwannee and other river corridors, the high degree of interaction between surface water and ground water questions the reliability of ground water as an untreated drinking water source. During the high river stages and floods, springs and seeps reverse flow and river water enters the aquifer (Raulston et al., 1998). This occurs in Branford within the SFRW. As mentioned, elevated nitrate-nitrogen concentrations are prominent in this region. Elevated nitrate-nitrogen has been recorded in the Middle Suwannee River Watershed. In this area, the Floridian aquifer is unconfined, allowing water soluble containments to leach into the aquifer. Springs in the Middle Suwannee River Watershed have nitrate-nitrogen concentrations ranging from 1.2 to 19.2 mg/l. Groundwater from the watershed flows towards the Suwannee River and is affecting surface water via springs and seeps in the riverbed (Raulston et al., 1998). The Water Quality Assurance Act and the Total Maximum Daily Load (TMDL) programs were initiated to detect and predict the contamination of the nation’s ground water resources, to manage non-point source pollutants and determine the effectiveness of using non-point source controls. Pollution prevention requires a clear understanding of the impacts of land use and water quality at a watershed level. Many efforts have been

PAGE 20

5 documented in promoting the watershed protection approach (USEPA, 1993). Although studies of water flow have been conducted at a watershed level for a long time (Gaebrecht, 1991), most water quality studies focus on areas in the vicinity of pollution sources. A limited number of studies examine the relationship between spatial distributions of land use and water quality at a watershed level. Most planning agencies and local authorities do not have resources to collect dense high quality land use and water quality data, thereby constraining the development of detailed water quality protection plans (Wang and Yin., 1997) .Since the early 90s, the use of Geographical Information System (GIS) to analyze spatial data to support pollution modeling and evaluation has increased (Gallimore and Xiang, 1991; Morse et al, 1994). Studies have demonstrated that digital data can be used in watershed and environmental studies (Hamlett et al., 1992; White et al., 1992). Digital Elevation Models (DEM) data were used to delineate waterways, watershed boundary and catchments for each specific water quality monitoring station (Moore et al., 1991; Vieux, 1991). Basin characteristics such as land use/land cover, slope and soil attributes affect water quality by regulating transport of sediment and chemical concentrations. Among these characteristics, land use/land cover can be manipulated to improve water quality. These land use/land cover types can serve as nutrient detention media or as nutrient transformers as dissolved or suspended nutrients move towards the stream (Basnyat et al., 2000). From a land cover perspective, agricultural activities have been identified as major sources of non-point source pollutant (Viessman and Hammer., 1993). Mattikalli and Richards (1996) determined the relationship between land use and water quality by employing an export-coefficient model for the River Glen watershed, UK. The results showed a high

PAGE 21

6 correlation between high nitrate concentrations and agricultural land uses. Similar studies conducted by Kauppi (1984) also indicate a strong correlation between nitrate concentrations and agricultural extensification. 1.1 Research Significance More than 90% of the Floridian residents rely on ground water for drinking water supplies. Therefore controlling non-point source pollution is critical. The identification of non-point source pollution problem areas is often difficult because of the spatial, distributed nature of the processes involved that change through time (Engel et al., 1993). Combining site-specific observations with GIS and remote sensing techniques have the potential to improve predictions of environmental properties at a watershed-scale. This study adopts a holistic approach towards watershed management in the SFRW: (i) to characterize the geographic position and spatial distribution of land cover/land use and other landscape characteristics, (ii) to document land cover/land use change, (iii) to understand spatial relationships between watershed characteristics (e.g., soils, parent material, topography, land cover) and nitrate-nitrogen. 1.2 Hypotheses The hypotheses were: Land cover shifts extending the agricultural area in the SFRW have occurred between 1990 and 2003 The spatial distribution of soil nitrate-nitrogen is variable across the SFRW depending mainly on soil and land use/land cover types Spatially distributed patterns of land resources and land cover dynamics are useful proxies providing information about nitrogen levels in soils and surface water 1.3 Objectives The specific objectives of the study were to:

PAGE 22

7 Characterize the land cover dynamics in the SFRW from 1990 to 2003 Quantify the spatial distribution of soil nitrate-nitrogen across the SFRW Investigate spatial relationships between watershed characteristics and soil and surface water quality

PAGE 23

CHAPTER 2 WATERSHED CHARACTERISTICS 2.1 Introduction Located in north-central Florida, the SFRW encompasses all or parts of seven counties in the State. Figure 2-1 shows the location of the watershed in Florida. Parts of Suwannee, Gilchrist, Columbia, Union, Bradford, Alachua, Baker and Clay counties constitute the 3,585 km 2 extend of the watershed. Administratively, the watershed lies within the Suwannee River Water Management District (SRWMD). High Springs, Alachua, Starke, Lake City and small portions of Gainesville are the prominent cities in the SFRW. The watershed boundary, the constituting counties and the prominent cities in the SFRW are shown in Figure 2-2. 2.2 Population The area comprising the SRWMD is one of the least populated regions in the State. The rapid increase in population in many parts of Florida did not have a significant effect on the population growth in north-central Florida. Even though most parts of the rural areas did not show population growth as rapid as in south-east Florida, development has increased in Alachua, Gilchrist and Columbia counties. In the SFRW, most of the population density is concentrated around cities like Lake City, Alachua, High Springs and Starke as well as the northern and western edge of Gainesville. The population rise from 1980 to 1995 in the SRWMD is shown in Figure 2-3. 8

PAGE 24

9 Figure 2-1. Location of the SFRW in the State of Florida.

PAGE 25

10 Figure 2-2. Counties and prominent city limits in the SFRW. The population density is quite low in the eastern part of the watershed which is characterized predominantly by pine plantations and land owned by timber corporations. Growth and development along the regions’ rivers has been limited, due in large part to floodplain management ordinances, land use plans and land acquisition programs at the state, regional and local levels (Raulston et al., 1998). From Figure 2-3, a steady increase in population between the two years can be seen. Population density within the SFRW in 2000, derived from U.S. Census Bureau data is shown in Figure 2-4. From the figure, it can be inferred that the northern and western wedges of Gainesville, Lake City and High Springs have maximum population density.

PAGE 26

11 Figure 2-3. Population growth in the SRWMD (Source: Raulston et al., 1998). Figure 2-4. Population density in SFRW.

PAGE 27

12 2.3 Climate The climate for most part of the SRWMD is considered sub-tropical. This region can be overall characterized as having hot humid summers and generally mild winters. In summer the temperature ranges from 20C to 23C and in winter it ranges from 4C to 10C. The precipitation for the SFRW is around 140 cm to about 152 cm (Raulston et al., 1998). 2.4 Topography In north-central Florida, and particularly within the SRWMD, the topography can be considered subdued. The elevation in the SFRW ranges from around 1.5 m around the confluence of Suwannee and Santa Fe rivers to a maximum of 90 m. The elevation range across the watershed is shown in Figure 2-5. Despite the fact that the most part of the watershed is predominantly flat with 0 to 2% slope, the topography varies from sloping (0 to 5% slopes) to highly sloped areas (5 to 28%) in particular along the Cody Scarp. The variation of slope across the watershed is shown in Figure 2-6. 2.5 Physiographic Regions The Northern Highlands and the Gulf Coastal Lowlands are the predominant physiographic regions in the SFRW. These highlands are considered to be remnants of the once continuous residual highland. The Northern Highlands are predominantly broad upland area with moderate relief, gentle to steep slopes, and incised streams (White, 1970). The Gulf Coastal Lowlands have a similar topography and consist of a series of Pleistocene surfaces and shorelines with limestone at or near the land surface (Raulston et al., 1998). Figure 2-7 shows theses physiographic divisions. Hence, the presence of highly permeable sands and unconfined karstic geology is a characteristic of this region. The deposition of rock and sediment features below 3 meters of the surface is represented

PAGE 28

13 by the environmental geology layer in Figure 2-8. The line that separates the Northern Highlands from the Gulf Coastal Lowlands is called the Cody Scarp. According to Puri and Vernon (1964), the Cody Scarp is termed as “the most persistent topographic break in the state”. Figure 2-5. Elevation across the SFRW. 2.6 Hydro-Geologic Framework The surficial aquifer system, the intermediate confining aquifer and the upper Floridian aquifer are the major hydro-geological units in the middle Suwannee River Basin (Katz and Bohlke, 2000). The surficial aquifer system is comprised of undifferentiated sands and clay of post Miocene age.

PAGE 29

14 Figure 2-6. Slope across the SFRW. Figure 2-7. Physiographic regions in the SRWMD.

PAGE 30

15 Figure 2-8. Environmental geology in the SFRW. They range in thickness from 3m to around 10m, but in the eastern most part of the basin, they range from 15m to 18m. The intermediate confining unit is composed of silicate sediments of Miocene age (Scott, 1997).The upper Floridian aquifer system is comprised of limestone and dolomitedeposits from the Eocene age (Katz and Bohlke, 2000). The physiographic features direct the hydro-geological characteristics of the region. As mentioned above, the Northern Highlands and the Gulf Coastal Lowlands are characteristic of the SFRW. Surface water features are prominent in the Northern Highlands because of the clayey nature of the intermediate confining aquifer. The clayey sediments in the intermediate confining aquifer underlying the surficial aquifer are

PAGE 31

16 impervious to the vertical flow of water and cause the water to accumulate in surface soils. The Gulf Coastal Lowlands are however characterized by karstic features. A karstic topography constitutes features like caves, sinkholes, springs and other openings caused by intense solution activity. Interaction between surface and ground water is highly pronounced in these regions and can have a profound effect on water quality. Unique problems can arise in protecting water quality in karst areas. This is primarily due to the direct and rapid transport of recharge through conduits to the subsurface and through resurgence by springs (Katz and DeHan, 1996). Figure 2-9 illustrates this phenomenon. Figure 2-9. Cross-section of a karst section in the Suwannee River Basin. (Source: Katz and DeHan, 1996). According to Upchurch (2002), numerous streams originate on the clastic sediments of the highlands and discharge through moderately to deeply incised valleys in the lowlands. Sinkholes, and swallow holes are common physiographic features found at the bounds of the scarp. The Cody Scarp is a region of active erosion of the northern physiographic province sediments. Sinkholes are prominent at the toe of the scarp and also within stream valleys in the upland physiographic province (White, 1970). In this

PAGE 32

17 transition zone, the Santa Fe River flows completely underground, becomes part of the ground water system and reemerges about 5 km downstream. The U. S. Environmental Protection Agency (USEPA) and the National Water Well Association (NWWA) developed the DRASTIC index for mapping potential aquifer vulnerability for ground water contamination. The DRASTIC index is a cumulative index computed from seven hydro-geological parameters, which are Depth to water Net recharge Aquifer media Soil media Topography Impact of the vadose zone Hydraulic conductivity Depending on the soil-landscape characteristics of the region, all these parameters are assigned appropriate weights. The DRASTIC index is a sum of all the weighted scores. High DRASTIC scores in the range of 180-220 indicate high vulnerability for ground water contamination. The DRASTIC index of the upper Floridian aquifer for the SFRW is shown in Figure 2-10. The western part of the SFRW is characterized by high DRASTIC index value, suggesting that this region is highly vulnerable to groundwater contamination through the upper Floridian aquifer. 2.7 Soils Dominant soils in the watershed are Paleudults, Paleaquults, Alaquods, and Quartzipsamments, with lesser but important occurrences of Haplosaprists, Alorthods, and Paleudalfs. Of the dozens of soil series found across the SFRW, some of the more exemplary and prevalent are Blanton (loamy, siliceous, semiactive, thermic Grossarenic Paleudults); Pelham (loamy, siliceous, subactive, thermic Arenic Paleaquults); Mascotte

PAGE 33

18 (sandy, siliceous, thermic Ultic Alaquods); Penney (thermic, uncoated Typic Quartzipsamments); Pamlico (sandy or sandy-skeletal, siliceous, dysic, thermic Terric Haplosaprists); Newnan (sandy, siliceous, hyperthermic Oxyaquic Alorthods); and Otela (loamy, siliceous, semiactive, thermic Grossarenic Paleudalfs) (Dearstyne et al., 1991; Weatherspoon et al., 1992). Figure 2-10. DRASTIC Index values for the SFRW. The various soil orders in the SFRW are shown in Figure 2-11. Soil data were obtained from the Soil Survey Geographic (SSURGO) database. These data were developed by the Natural Resource Conservation Service (NRCS), USDA at a scale of 1: 24,000. Ultisols, Spodosols and Entisols are the dominant soil orders present in the SFRW.

PAGE 34

19 Entisols are immature soils usually associated with sediment deposits from weathering or erosion. Entisols are characterized by the presence of low organic matter and high sand content. They lack a confining soil horizon (NOAA, 1996). Also owing to the shallow water table in north-central Florida, these soils are highly vulnerable to nutrient leaching to the sub-surface systems (Paramasivam et al., 2001). Ultisols are formed in regions that have high rainfall and relatively low evapo-transpiration rates. Ultisols are abundant in the tropics and they are fine textured. The warm temperature, high moisture content in these soils and the texture enhances the leaching capacity of these soils. Spodosols have a high occurrence in the SFRW. The occurrences of Spodosols are usually associated with coniferous vegetation or any other vegetation that can supply sesquioxide (Fe and Al-oxide) mobilizing organic compounds. Spodosols have a distinct spodic horizon that is formed by mobilizing and illuviation of organic matter and sesquioxide from O, A and E horizons. These processes are collectively termed podzolization. Based on characteristics of the spodic horizon, they may be extremely friable, cemented, nodular or placic. In Florida, the A and E horizons of Spodosols are quite sandy and they have poor sorption characteristics. Also the high water table in-between the Bh and A horizons, enhance horizontal leaching of nutrients (Mansell et al., 1991). The Spodosols of Florida that receive significant loadings of animal manure are especially prone to subsurface leaching (Nair and Graetz, 2002). 2.8 Land Use The SFRW is a “mixed land use” watershed. The land uses range from croplands in the western part to timber lands in the eastern part of the watershed. The extent of agricultural land use/land cover (LULC) class, especially in improved pasture and row

PAGE 35

20 crops, is extensive in Columbia, Suwannee and Alachua counties. The row crop types in the watershed include corn, peanuts, soybeans, strawberries, blue berries, water melons, pecans and tobacco plantations. As mentioned, the eastern part of the watershed is predominantly covered by pine plantations and land owned by timber corporations. Pine plantations have the highest areal extend in this region and cover around 30% of the watershed. Native land cover classes in the watershed include upland hardwood forest, mixed forest and Pine plantations. They constitute around 47% of the watershed. Wetlands are also a predominant land cover and constitute around 11% of the area. Lakes and other water bodies are distributed widely in the watershed. Lake Santa Fe, Lake Alto, Hampton Lake, Lake Sampson, Lake Crosby, Lake Rowell, Lake Butler and Alligator Lake are the prominent water bodies in the SFRW. Urban areas appear sporadically throughout the watershed. The 1995 LULC data for the SFRW generated by the SRWMD and the SJRWMD is shown in Figure 2-12. This Figure is an aggregation of the Level 1 LULC classes.

PAGE 36

21 Figure 2-11. Soil orders in SFRW.

PAGE 37

22 Figure 2-12. 1995 LULC map of the SFRW.

PAGE 38

CHAPTER 3 LANDCOVER DYNAMICS IN THE SFRW 3.1 Definitions of Land Use and Land Cover In the analysis of land use and land cover dynamics, it is first necessary to comprehend the meaning of terms, land use and land cover. Despite the fact that they are not synonymous terms, they are often used in association with the other. Land cover is the biophysical state of the earth’s surface (Turner et al., 1994). According to Moser (1996), “The term originally referred to the type of vegetation that covered the land surface, but has broadened subsequently to include human structures, such as soils, biodiversity and surfaces and groundwater”. On the other hand, land use can be described as the purpose of the piece of land. Land use involves both the manner in which the biophysical attributes of the land are manipulated and the intent underlying that manipulation – the purpose for which the land is used (Turner et al., 1994). From a theoretical perspective, the differences between land use and land cover are distinguishable. However, this difference is not straight forward in practice. This results in confusion and complication in the analysis of either one of them (Briassoulis, 2000). Based purely on the spectral characteristics of the various bands, satellite images enable the identification of different land cover classes. Identification of land use types is more robust and time consuming as it may involve the use of aerial photographs and photo-interpretation techniques, reference data and other forms of auxiliary information. 23

PAGE 39

24 3.2 Impact of Land Cover Change in the SFRW According to Briassoulis (2000), “Changes in the use of land occurring at various spatial levels and within various time periods are the material expressions of environmental and human dynamics and of their interactions which are mediated by land.” ‘Change’ is often perceived as the differences in areal extend of a given land cover class over a defined temporal interval. The study LULC change is important in an environmental context, as Kates et al. (1990) state that the lands of the earth bear the most imprints of the anthropogenic actions. Anthropogenic use of the land is generally perceived as a static process but it is an extremely dynamic one (Lindgren, 1985). 3.3 Significance Man induced impacts on LULC changes have increased manifolds in the last 300 years and an example of this manifestation is seen in the water quality observed in the SRWMD. Studies on the ground water conditions in the SRWMD indicate an increasing trend in the nitrate-nitrogen concentrations. The middle reaches of the Suwannee River is increasingly being impacted by nutrients (Raulston et al., 1998). During February 1995 (high flow conditions) concentrations of nitrate plus nitrite ranged from 0.05 to 0.38 mg/l. In June 1995 (low flow conditions), they ranged form 0.07 to 1.05 mg/l. Highest concentration was observed in Branford (Raulston et al., 1998) which lies within the SFRW. Over two decades of monitoring the water quality of major rivers of the Suwannee River Basin indicated a statistically significant (at 95% confidence interval) increasing trend in the concentrations of nitrate-nitrogen (Ham and Hatzell, 1996; SRWMD, 1998). Contamination of the groundwater in the middle reaches of the Suwannee River and, in the SFRW, is believed to be due to agricultural activities, poultry farming, dairies and other sources. These LULC types are considered prominent sources

PAGE 40

25 of non-point source pollutants leaching into the ground water. Non-point source pollutants are a cause of concern in pollution prevention, both in the surface and subsurface soil and water resources, primarily because of the inability to trace them to a point location (Chowdary et al., 2001). This is primarily due to the spatially distributed nature of water contamination from non-point source pollutants (Engel et al., 1993). Previous studies by Perry and Vanderklein (1996) and by Viessman and Hammer (1993) indicated a correlation between pollution loading and LULC category. Hence, a multi-temporal change detection analysis is proposed to describe long term trends or directions of LULC change in the SFRW that possibly causes the leaching of nitrate-nitrogen into the ground water system. Thus, understanding these trends has the potential for improving water quality with proper land use management practices (Perry and Vanderklein, 1996; Viessman and Hammer, 1993). 3.4 Objectives The specific objectives of this study were to Identify recent changes within the land cover classes from 1990 to 2003 Quantify the areal extend of these changes Assess trend or nature of changes within land cover classes Describe spatial patterns of change 3.5 Materials 3.5.1 Landsat Imageries The Landsat program was initiated by the National Aeronautics and Space Administration (NASA) and the U.S. Department of Interior to monitor earth’s resources using optical spectrum. The SFRW falls within the Path 17, Row 39 of Landsat orbital tracks. Landsat TM (Thematic Mapper) and Landsat ETM+ (Enhanced Thematic Mapper +) images

PAGE 41

26 pertaining to this location were obtained from the NASA data archives, Department of Geography, University of Florida. Images were obtained for August 26 th 1990 (Landsat TM), August 13 th 2000 and February 13 th 2003 (Landsat ETM+). Specific documentation related to these images can be found in Appendix A. Landsat TM and ETM+ satellites operate in seven spectral bands. Bands and their spectral location are given in Table 3-1. Table 3.1. Spectral locations of Landsat bands Landsat Band Wavelength (m) Spectral location 1 0.45-0.52 Blue 2 0.52-0.60 Green 3 0.63-0.69 Red 4 0.76-0.90 Near-infrared 5 1.55-1.75 Mid-infrared 6 10.4-12.5 Thermal infrared 7 2.08-2.35 Mid-infrared 3.5.2 Software The following software was used for much of the analysis. Leica Geosystem’s (Atlanta, GA) ERDAS-Imagine 8.5 software was used for most of the remote sensing / image processing analysis. ArcGIS Desktop 8.3 – developed by Environmental System Research Institute (ESRI), Redlands, CA was used for spatial analysis and display. Trimble’s GPS pathfinder office and Terrasync software – developed by Trimble Navigation Limited, Sunnyvale, CA were used in field data collection, management and transfer of data to desktop systems. 3.5.3 Auxiliary Data Various auxiliary spatial data were used to guide the classification process. This data was downloaded from the Florida Geographic Data Library (FGDL). They are: SFRW boundary obtained from SRWMD. Road network for the SFRW, developed by the U.S. Census Bureau.

PAGE 42

27 National Wetland Inventory polygons, developed by the U.S. Department of Fisheries and Wildlife. Geographic Approach to Planning (GAP) analysis land cover data layers, developed by the Florida Cooperative Fish and Wildlife Research Unit. Digital Ortho Quarter Quad’s for the SFRW, developed by the Florida Department of Environmental Protection (FDEP)’s Bureau of Land Management. 3.5.4 Field Sampling Instruments The GPS Pathfinder instrument comprising of the GPS Pathfinder Pro XRS receiver, and the GIS TSCe field device developed by Trimble Navigation Limited, Sunnyvale, CA were used. 3.6 Methods 3.6.1 Design of Classification Scheme Maps categorize the earth’s surface features. Map categories are specified by the project’s classification scheme. The most common requirements in a remote sensing analysis are the need to classify an image into land cover or object classes (Schott, 1997). Classification schemes are the means of organizing spatial information in an orderly and logical way (Cowardin et al., 1979). Initially, LULC were not collected based on any common reference systems. Hence, there were different collection methods and classification schemes that affected the interoperability of the surveyed data. These datasets have very little in common and could not be merged or aggregated. In an effort to improve on this situation, the U.S. Geological Survey (USGS) developed a standardized common reference system for classifying LULC data obtained by various means of remote sensing techniques. The main objective was to be able to use detailed land cover studies at local and regional levels and aggregate them upwards to state and

PAGE 43

28 national levels (Lindgren, 1985). According to Anderson et al. (1976), the USGS land use and land cover classification system was developed based on the following criteria: The minimum level of interpretation accuracy in the identification of land use and land cover categories from remote sensor data should be at least 85 percent. The accuracy of interpretation for the several categories should be about equal. Repeatable or repetitive results should be obtainable form one interpreter to another and from one time of sensing to another. The classification system should be applicable over an extensive area. The categorization should permit vegetation and other types of land cover to be used as surrogates for activity. The classification system should be suitable for use with remote sensor data obtained at different times of the year. Effective use of subcategories that can be obtained from ground surveys or from the use of larger scale or enhanced remote sensor data should be possible. Aggregation of categories should be possible. Comparison with future land use data should be possible. Multiple uses of land should be recognized when possible. The USGS designed a LULC classification scheme. This system was designed such that four levels of land cover data can be represented. Data in level II, III and IV can be aggregated to represent level I data. Such a multilevel–multi-tier system was developed because different degrees of spatial detail can be obtained from different remote sensing products (Lillesand and Kiefer, 1994). Level I data can be effectively obtained from Landsat series of satellites. Level II can be derived by using high – altitude data and air photo interpretation. The various image interpretation formats for the LULC levels are shown in Table 3-2.

PAGE 44

29 These classification systems were developed to provide a uniform framework and to enable interoperability of datasets to satisfy a majority of users. It should be understood that there is no one ideal classification system for LULC mapping. According to Anderson (1971), the design of the classification scheme is a highly subjective process and depends on the objectives set by the analyst. It is hence very unlikely to develop a single ideal classification system. The Florida Land Use, Cover and Forms Classification System (FLUCCS) was developed by the Department of the Surveying and Mapping Office of the Florida Department of Transportation to accommodate specific LULC features pertaining to Florida. The FLUCCS system is based on the USGS LULC classification system and is shown in Appendix C. The FLUCCS system was taken as a reference for land cover classification in this study. The SRWMD and the SJRWMD had land cover maps pertaining to the study area. Since these maps were relatively old they were used as a reference rather than to characterize current land use and land cover conditions. They were used to obtain an approximate extent of LULC in the watershed. For the project, sixteen LULC classes were initially identified and grouped based on their aerial extent and on their potential of non-point source pollutant/ nutrient loading to the sub-surface. Table 3-2. Required image interpretation formats for various Anderson level LULC classes. LULC classification level Required format for interpretation Level 1 All Landsat series Level 2 Aerial photographs, Landsat TM, ETM+, SPOT Level 3 Medium scale aerial photos Level 4 Large scale aerial photos

PAGE 45

30 The sixteen LULC classes are shown in Figure 3-1. Grouping of these land cover classes that best suited interpretation from Landsat images was conducted. The aggregated LULC classes that make up the final land cover classes considered in the analysis are shown in Table 3-3 3.6.2 Ground Truth Data Collection Adequate ground truth information for these land cover classes did not exists. Hence, new ground truth data for both training set development and for accuracy assessment were collected. A random sampling strategy was adopted. According to Congalton (1991), a total of 50 Ground Control Points (GCPs) for each land cover class are recommended for accurate classification. Ground truth data were collected by using GPS Pathfinder instrument comprising of the GPS Pathfinder Pro XRS receiver, and the GIS TSCe field device. A total of 487 points were collected during December 2003 and January 2004. During the ground truth data collection, it was insured that the variability of land cover classes were taken into consideration. Hence, the data points were also collected at certain level 2 classes also. The data points were collected targeting the spatial extend of the watershed. The location of these points are shown in Figure 3-2. Some prominent LULC classes in the watershed are shown in Figure 3-3. Agricultural class encompassing row crops and improved pastures. pine plantation, upland forests, forest and non-forest wetlands, rangelands, water and urban land cover classes were targeted.

PAGE 46

31 Figure 3-1. LULC classes in 1995 SFRW.

PAGE 47

32 Table 3-3. Aggregated LULC classes to create final land cover classes. Final land cover class Aggregated classes Pine plantations Coniferous pine Upland forest Upland coniferous forests Upland hardwood forest Upland mixed forest Agriculture Cropland and pastureland Tree crops Feeding operations Nurseries and vineyards Specialty farms Rangeland Rangeland Unimproved pastures Woodland pastures Urban Residential, low density Residential, med. density Residential, high density Commercial and services Industrial Institutional Recreational Open Land Transportation Communications Utilities Wetlands Wetland hardwood forests Wetland coniferous forest Wetland forested mixed Vegetated non-forested wetlands Non-vegetated wetlands

PAGE 48

33 Figure 3-2. Location of ground control points. The collected ground truth data were split into training and test datasets. Approximately 60% of the dataset for each land cover class was categorized into training datasets. The listing of the test and training datasets are given in Appendix B. 3.6.3 Image Analysis 3.6.3.1 Preprocessing The perfect remote sensing system is not yet developed. Errors creep in during the data acquisition stage and degrade quality of recorded reflectance values as earth’s land and water surface features are amazingly complex systems and they do not lend themselves to be recorded by relatively simple remote sensing devices that have spatial, spectral temporal and radiometric constraints.

PAGE 49

34 (A) (B) Figure 3-3. Pictures of land cover classes in SFRW. (A) Improved pasture (B) Row crops (C) Pine plantations (D) Upland forest (E) Rangeland and (F) Wetland.

PAGE 50

35 (C) (D)

PAGE 51

36 (E) (F)

PAGE 52

37 Radiometric and geometric errors are the most common types of errors encountered in remotely sensed imagery (Jensen, 1996).These errors will cause a less accurate analysis. Image rectification and restoration processes aim in creating a more faithful representation of the earth surface features by rectifying images for these errors. The process typically involves processing of raw image data to correct for geometric distortions, to calibrate the data radiometrically and to eliminate noise present in the data (Lillesand and Kiefer, 1994). These processes are collectively termed as pre-processing operations as they typically precede any further analysis of image. 3.6.3.1.1 Geometric correction A number of spatially, temporally and spectrally varying factors have an influence on characteristics of acquired image and registration accuracy. The position and attitude of a sensor are usually known to some precision. Predicted image pixel location based on knowledge of platform parameters may result in an absolute location error. Also, uncertainty in platform attitude will cause geometric distortion of image data. Depending on stability of the platform and on accuracy of estimate of the platform ephemeris and attitude, these systematic errors can be removed by the use of tie points. However, non-systematic errors result in residual errors in the image location and, hence, a final step of refined image registration is necessary (Rignot et al., 1991). The process of image registration may be divided into two steps. The first step is to select registration control points with a suitable distribution and then to measure the corresponding image coordinates. In the second step, a suitable mapping function is chosen after which a co-ordinate transformation and image resampling are performed (Chen and Lee, 1992). All the Landsat images used for the project were geo-referenced and corrected for both systematic errors. However, images were not co-registered with each other.

PAGE 53

38 Registration between a reference image and its counterpart, a second remotely sensed image, is a necessity in many image analysis tasks such as change detection (Swain and Davis, 1985). According to Chen and Lee (1992), “Image to image registration is the translation and rotation alignment process by which two images of like geometry and of the same geographic area are positioned coincident with respect to one another so that corresponding elements of the same ground area appear in the same place on the registered images”. The most recent August 2003 image was taken as the base image and other images were registered to the base image using image to image registration. The first step is to determine the geometric relationship between pixel locations in the reference image and the distorted image. These errors/ distortions are mostly caused due to differences in altitude or attitude of sensors. Geometric errors are usually unsystematic and are best removed by identifying GCP’s in the original imagery and on the reference map and then mathematically by modeling the geometric distortions (Jensen, 1996). As a first step, geometric projection is specified for both the input and reference image. The images were projected to Universal Traverse Mercator (UTM) under Zone 17. Geographic Reference System (GRS) 1980 and North American Datum (NAD) 83 were the specified spheroid and datum, respectively. More than 15 GCP’s were identified in the reference and input image. As a rule, GCP’s are points that are clearly identifiable in both images. These typically include road intersections, buildings and others. Such techniques usually require a polynomial equation be fit through the GCP’s to model the distortions present in the input image. A second order polynomial

PAGE 54

39 transformation was used to rectify the input images to the August 2003 base image. A total root mean square (RMS) value of less than 0.5 pixels was ensured. A pixel in the rectified output image often requires a brightness value from the input pixel grid that does not fall neatly on a row and column co-ordinate. When this occurs, the brightness value to be assigned to the new rectified pixel is determined by the process of intensity interpolation (Jensen, 1996). The “nearest neighborhood” resampling algorithm was used to create the final co-registered images of 1990 and 2000. According to this algorithm, the Digital Number (DN) value in the new rectified image is based on the DN value of the closest pixel in the input image. The SWIPE function in ERDAS enables visual comparison of data. The swipe function carried out between the 2000 corrected image and the 2003 reference image is shown in Figure 3-4. The pixels fall within 0.5 pixels from each other. Figure 3-4. Illustration of the accuracy of the geometric correction process. The 2000 image on the left, superimposed over the 2003 images on the right hand side.

PAGE 55

40 3.6.3.1.2 Image subset For more effective computation and detailed study, the SFRW was clipped out from the imagery. The geometrically corrected images of the study area for 2003 are shown in Figure 3-5. 3.6.3.1.3 Radiometric correction Remote sensors basically record the intensities of absorption, reflection and emission of electromagnetic energy by various earth surface features. The signals are further analyzed to get a grasp of the conditions being investigated. Owing to the travel distance between the earth and the sensor through the atmosphere, these signals are subjected to scattering, absorption effects of gases, aerosols and others. Figure 3-5. The geometrically corrected and subset composite (band 4, 3 and 2 RGB) image of 2003 SFRW.

PAGE 56

41 The bands in the Landsat series have been selected to avoid atmospheric absorption in most cases. However, the effects caused by aerosols are difficult to characterize because of their variation in space and time (Kaufman and Sendra, 1988). This effect is thus a severe limitation for the radiometric normalization of satellite data (Liang et al., 1997). Atmospheric effects must be corrected in any study which involves the comparison of imagery through time (Dave and Bernstein, 1982). Correction techniques that address visible and infrared image radiometry are necessary for using remotely sensed data to address earth-science related problems from mapping geophysical surface features (Peddle and Franklin, 1993) to ecological and environmental monitoring (Hall et al., 1991). According to Duggin and Robinove (1990), failure to account for pre-processing corrections (both geometric and radiometric correction) may result in inaccurate image analysis and consequently inaccurate classification thematic maps. Sensor induced effects and scene induced effects are the two major sources of radiometric errors in remotely sensed data (Teillet, 1986). Sensor induced effects are induced due to the technical aspects of the remote sensing sensor system. Problems in the calibration of detectors, platforms and system stability, filtering, etc. can cause sensor induced radiometric errors in the data. The influence of topography, atmosphere, viewing angle, adjacency effect, position of sun and the reflectance properties of the objects induce scene related errors to the data (Meyer et al., 1993). The geometrically corrected images of the SFRW are treated for both sensor induced and scene induced errors to get a more faithful representation of the actual trends in the watershed, rather than differences in sensor or atmospheric conditions.

PAGE 57

42 DN is the standard format in which Landsat data are presented to the users. The DN values are quite consistent with the image and between different spectral bands and hence can be used “as is” for most single data analysis. However, before a multi-temporal image analysis is carried out, the DN values have to be transformed to standard values (Franklin and Giles., 1993). These standard values are computed from the raw DN values in a two step process. First, the DN values are converted to at-satellite radiance values. Secondly, the at-satellite radiance values are converted to standard at-satellite reflectance values. On board detectors or sensors are calibrated to produce a linear response to spectral radiance. Figure 3-6 shows a radiometric response of an individual TM channel. All the bands in the Landsat series satellites have their own radiometric response functions. These functions are monitored using internal calibrators (IC). Typically, the IC consists of three lamps that provide radiance levels to the detector in each band. The detector/ channel bias is also recorded by the IC (Vogelmann et al., 2001). Figure 3-6. Radiometric response of TM channel.

PAGE 58

43 The absolute radiance output of the calibration sources is known from pre-launch calibration and is assumed to be stable over the life of the sensor (Lillesand and Kiefer, 1994). These are published for civilian satellites following absolute calibration tests over terrestrial targets. These coefficients are stored in the image data tape headers and are updated regularly by the various satellite operations groups. These IC values are used to establish the relationship between radiance values incident on the detectors and the DN value measured by constructing the radiometric response function. Based on Figure 3-6, the relationship is established as below, L = “Gain” *L+ “Bias” Equation 1 where L is the calculated spectral radiance at the satellite’s aperture, measured in Watts/ (meter squared * ster* m). L is the calibrated pixel value in DN. Gain is the slope of the response function, measured in Watts/ (meter squared * ster* m) Bias is the intercept of the response function. Gain and Bias are the absolute calibration coefficients and are obtained from the image header file. Also, they can be derived from the lower (LMIN) and upper limit (LMAX) of the spectral radiance for a specific band. LMIN is the spectral radiance associated with a DN value of Zero. LMAX is the maximum radiance required to generate the maximum DN. That is, LMAX represents the radiance at which the channel saturates. The range from LMIN to LMAX is the dynamic range of the channel

PAGE 59

44 (Lilliesand and Keifer, 1994). The Gains and Bias are computed using the following equations Gain = (LMAX – LMIN) / 255 Equation 2 Bias = LMIN Equation 3 The Gain/Bias values associated with the images from 1990, 2000 and 2003 are given in Appendix A. The flowchart in Figure 3-7 shows the steps involved in the conversion of raw DN values to at-satellite radiance values. A spatial model was built using the “Model Maker” option in ERDAS-Imagine model. Figure 3-8 shows a snap shot of the model used for the conversion. In Figure 3-8, the input image is the 2003 Landsat ETM image of the SFRW. The input image is split into its constituting bands. In this case, six bands excluding the thermal band are used. The Gains and Bias values from the header file are used to define the function. The function definition boxes are shown as circles in Figure 3-8(A). The function definition dialogue box incorporating Equation (1) is shown in Figure 3-8 (B). The results of the computation for each band are initially stored in temporary rasters. The temporary rasters are denoted as ‘n_memory’ in Figure 3-8(A). The temporary rasters are stacked to create the final raster of the radiance values. Multi-temporal image analysis often involves the analysis of images taken at different time periods and seasons. To enable an effective comparison of these images, sun-elevation correction and earth-sun distance correction are necessary. Different solar illumination angles result due to the seasonal variation of the sun relative to the earth.

PAGE 60

45 Figure 3-7. Conversion of raw DN values to at-satellite radiance.

PAGE 61

46 To normalize the effects of the varying sun elevation angle on the same scene taken at different temporal intervals, each pixel in the scene is divided by the cosine of the solar elevation angle at the particular time and location of imaging. Also, the correction can be applied by computing the sun’s angle from the Zenith (90) (Lillesand and Kiefer, 1994). Also, depending of the time of the year, the distance between the earth and sun varies. To normalize the effect of the varying earth-sun distance, earth-sun distance correction is applied to the datasets.The at-satellite reflectance values are calculated by accounting for the earth-satellite distance and the sun-elevation differences in the calculated at-satellite radiance values. The equation is as below R = ( * L *D 2) / (ESUN * Cos ) Equation 4 where is a constant of value 3.14 R is the calculated at-satellite reflectance value L is the at-satellite radiance measurements D is the earth-sun distance in astronomical units ESUN is the mean solar exoatmospheric irradiances and is the solar zenith angle in degree The value of D can be approximately determined using the following equation D = 1 + 0.0167 sin [2 (d – 93.5)/365] Equation 5 where d is the day number of the year. The difference in the spectra can be attributed to the corrected sensor characteristics.

PAGE 62

47 (A) (B) Figure 3-8. Spatial model used in the conversion of DN to reflectance values (A) Snapshot of the model created to convert DN values to radiance values. (B) The process definition box used of ERDAS-Imagine software used in the conversion.

PAGE 63

48 Figure 3-9. Conversion of at-satellite radiance values to at-satellite reflectance values.

PAGE 64

49 According to Eliason et al. (1981), “These apparent reflectance values are approximations of the physical surface albedo and can be used with more confidence than the original digital counts or at-satellite radiances in comparison of spectral data over time and between different sensor systems”. The flowchart in Figure 3-9 shows the steps involved in the conversion to at-satellite reflectance values. A similar model as in Figure 3-8 was used for the conversion. In an ideal remote sensing system, the amount of radiant energy reaching the sensor’s aperture would be the amount of energy emitted by objects in the satellites field of view. This is usually not the case. Other radiant energy enters this path from other sources. According to Jensen (1996) the various paths of radiance received at the satellite’s aperture are shown in Figure 3-10. Figure 3-10. Paths of satellite radiance

PAGE 65

50 The path that is readily perceived as the most common form of energy propagation is path 1. Here, the incident energy from the sun, passes through the atmosphere, gets reflected by the earth surface features and is captured at the satellite’s aperture as L T . Some of the incident energy from the sun are scattered into the satellite’s system even before reaching the earth’s surface. This is shown in path 2 and this energy contributes to the haze, air light, fog, thick cirrus clouds that are seen in many remotely sensed imagery and aerial photographs. This form of energy is often termed as “up-welled radiance”. In some cases, the up-welled radiance may completely overwhelm the flux reflected from the earth surface features. Path 3 refers to the energy that originates from the sun, gets scattered by the atmosphere, and gets reflected by the earth’s surface features to the satellite system. Energy in path 4 originates from the sun, travels through the atmosphere, gets reflected or scattered by the background objects such as terrain, water, building and are reflected back to the satellite’s aperture. Adjacency effects cause multiple bounce of photons. These energy pockets are reflected from the surrounding objects and scattered to the satellites’ field of view. This is shown by path 5. The many travel paths of this incident energy introduces path radiance (L P ) to the radiance energy returned from the study area (L T ). This causes the radiance recorded at the sensor (L S ) not to equal the L T . The total radiance at the satellite’s aperture is thus given by the equation L S = L T + L P (Wm -2 sr -1 ) Equation 6 The objective of atmospheric correction procedures is to minimize or remove the effect of path radiance (L P ). The propagation of electromagnetic energy from the sun, through the atmosphere is affected by atmospheric scattering and absorption.

PAGE 66

51 Atmospheric attenuation is a collective term for scattering and absorption effects that reduce the effect of the forward going wave (Rees, 1990). Scattering can be explained as an unpredictable redirection of energy in the atmosphere. The effects of Rayleigh scatter, Mie scatter and non-selective scatter are obvious in many remote sensing images. Rayleigh scattering occurs when the incoming solar radiation interact with molecules in the atmosphere that are much smaller in diameter than wavelength of the incoming solar radiation. The effect of this scatter is inversely proportional to the fourth power of wavelength. Hence, Rayleigh scatter affects shorter wavelengths and causes more scattering than the longer wavelengths. The presence of haze in an image is a manifestation of Rayleigh scatter. Mie scatter occurs when the incoming solar radiation interacts with atmospheric molecules that are the same size as that of the incoming radiation. Water vapor, dust and aerosols in the atmosphere cause Mie scatter. Unlike Rayleigh scatter, Mie scatter influences longer wavelengths. Non-selective scatter is when the atmospheric radiation interacts with molecules that are much larger than their diameter. Cloud and fog are seen in imagery because of this effect (Lillesand and Kiefer, 1994). Atmospheric absorption is the absorption of energy at given wavelengths. The presence of water vapor, carbon-dioxide and ozone in the atmosphere absorbs the incoming radiation at specific wavelengths. These particular wavelengths in which the atmosphere is transmissive of energy are called atmospheric windows. Atmospheric scattering and absorption have a profound impact on the incoming and the reflected radiation observed on air-borne sensors (Giles and Curran, 1995).

PAGE 67

52 As with Landsat series satellites, the bands were selected to avoid possible effects due to atmospheric absorption, hence the multiplicative effect from absorption is often neglected. The atmospheric scattering effect is hence the dominant atmospheric attenuation factor in Landsat (Song et al., 2001). To account for atmospheric effects in this analysis, the Dark Object Subtraction (DOS) technique was adopted. This approach involves the identification of objects in the image with zero or close to zero reflectivity. Water bodies are the common invariant ground object targets for DOS. Any measurement of radiance greater than zero over this dark object is attributed to atmospheric effects (Franklin and Giles, 1993). The correction process involves subtracting the constant from all pixels in all the spectral bands. The mean vector value of the clusters change, but, the variance – covariance matrix will remain the same. Such a correction is nothing but translating the origin in multidimensional space (Song et al., 2001).This is shown in Figure 3-11. 2 Band J 1 3 Band I Figure 3-11. Shifting of origins during DOS. Based on Chavez (1994), the dark object subtraction correction is given as below R DOS = R * (R DO )/ ((Cos (90-)*)/180) Equation 7

PAGE 68

53 where R DOS is the calculated value after dark object subtraction techniques R is the at-satellite reflectance calculated in the previous step R DO is the at-satellite reflectance value for a dark object in the scene is the solar zenith angle A spatial model using the ERDAS-Imagine’s ‘Model Maker’ option was used for correcting the images. A flowchart of the procedure is shown in Figure 3-12. 3.6.3.1.4 Noise removal The presence of clouds and their shadows can degrade or mask the actual radiometric content in the scene. These regions have to be masked before any iteration involving image statistics can be performed. The 2003 image of the SFRW was cloud-free; however the 2000 and 1990 images had a small amount of cloud coverage. Combinations of techniques were used to create the cloud mask for these two images. The natural clusters present in the dataset were determined and identified using unsupervised image classification. The unsupervised image classification is explained in detail in chapter 3.6.3.2.2. The clouds were picked up in this method. The shadows of these clouds were difficult to pick up as they were confused with the water body pixels. Unsupervised image classification with 10 classes was performed in 1990, 2000 and 2003 images. Each of the images was recoded to represent the just two classes, “Water” and “Other land cover”. The 1990 and 2000 images had the clouds shadow pixels represented as “Water”. All the three images were taken to an ArcGIS environment and reclassified.

PAGE 69

54 Figure 3-12. Application of DOS method.

PAGE 70

55 (A) (B) (C) Figure 3-13 Cloud mask. (A) Cloud and its shadow on the raw image. (B) Red areas show a mask of cloud and their shadows (C) Clouds and their shadow masked in the raw image for the 2000 year image. A raster calculator operation was done to combine all the three images into a single raster file. A “Water” mask was created by using only the pixels that did not change in all the three years. This “Water” mask was then used on the original images to cut out the water bodies. The objective of this process was to enable easier delineation of cloud shadow regions without confusing them for water bodies. An unsupervised image

PAGE 71

56 classification was again performed on all the “Water” masked images. Accurate delineation of cloud shadows was made possible using the above methodology. The layers representing cloud and cloud shadows were overlayed to create a single image file. The image file containing two classes of “Cloud” and “Cloud shadow” were recoded to get an image file with one class namely “Cloud and Cloud shadows”. This cloud mask was used to mask out clouded regions in the raw images of 1990 and 2000. Figure 3-13 shows the cloud masked images for the 2000 image. The unsupervised classification of images is discussed in detail below. 3.6.3.2 Image processing Classification is the process whereby an image is converted into some kind of thematic map. It can thus be termed as an attempt to replace each pixel in the image with an appropriate thematic class value. The choice is made on the basis of reflection measurements that are stored in the pixels of an image. The collection of measurements in one pixel is called measurement vector or feature vector. With M spectral bands the feature vector has M components and corresponds to a point in M-dimensional feature space. The task of classification is to assign a class label to each feature vector, which means to subdivide the feature space into partitions that correspond to classes. This task is achieved by pattern recognition (Stein et al., 1999). Spectral pattern recognition refers to the family of classification procedures / algorithms that utilizes this pixel-by-pixel spectral information as the basis for automated land cover classification (Lillesand and Kiefer, 1994). Over the years a variety of algorithms have been developed to aid automated land cover classification. Most commonly employed methods entail: Classifications using the supervised / unsupervised approach

PAGE 72

57 Hybrid methods using both supervised and unsupervised and also other ancillary data and Fuzzy classification methods. The choice of the algorithm depends on a variety of factors, such as, the presence of ground truth data, desired classification accuracy, presence of ancillary data, the analyst’s knowledge of the study area and others. The classification algorithms and approaches used in this study are discussed below. 3.6.3.2.1 Pre-classification scene stratification The Instantaneous Field Of View (IFOV) of a satellite can be explained as the interval in which the scanner “sees” the energy from earth surface features. In Figure 3-14, refers to the IFOV and D is what is referred to as the spatial resolution of a satellite and H refers to the altitude of the flight. Figure 3-14. IFOV of satellite system The IFOV of a sensor system records the reflected radiant energy from heterogeneous mixtures of biophysical materials such as soils, water and vegetation. One land cover classes gradually grades into the other land cover class without sharp heterogeneous boundaries. Hence, reality is actually very imprecise and heterogeneous

PAGE 73

58 (Lam, 1990). Unfortunately, we usually use very precise classical set theory to classify remotely sensed data into homogenous information classes, ignoring the imprecision (Jensen, 1996). The case of mixed pixels is a problem in the case of a supervised classification, where the accuracy of the classification depends on the selection of representative samples. Aside from the development of new classification algorithms, any number of methods might be used to improve classification accuracy. An obvious method for improving classification would be to consider a greater number of object attributes. This might include consideration of conventional image attributes, such as size, shape, pattern, image association (Colwell, 1960). Many of these object attributes cannot be obtained from digital image processing. A more immediate method would be to incorporate information about object attributes derived from ancillary data sources (Hutchinson, 1982). Stratification is the use of ancillary data prior to image classification which involves sub-setting the larger study area into smaller areas based on specific criteria. The subset images are analyzed separately. Statistically, the purpose of stratification is to increase the homogeneity of the data sets to be classified. From a practical stand point, stratification is employed for classification improvement either to divide a large study into smaller homogenous units, or to separate different features which are spectrally similar (Hutchinson, 1982). In north-central Florida, urban areas can be very heterogeneous. Because urban areas are easily confused with bare soil they can be classified more accurately if separated from rural areas (Reese et al., 2002). Ancillary data in the form of TIGER road line files, developed by the U.S. Bureau of Census were downloaded for the SFRW. Visual interpretation of the urban areas was achieved by overlaying the TIGER line files

PAGE 74

59 on the imagery. DOQQ’s for the SFRW were also downloaded to delineate the urban areas. Figure 3-15 shows the urban mask created using the TIGER road files. The urban mask was used to clip out the non–urban region of the SFRW. One of the advantages of stratifying the imagery before classification is the reduction of variation in the dataset. Training and classification can proceed independently on each stratum and finally the two may be merged in a final product. Confusion is thus avoided and accuracy improved (Gaydos and Newland, 1978). The urban–rural stratified images of 2003 are shown in Figure 3-16. 3.6.3.2.2 Unsupervised classification The unsupervised classification technique uses a clustering algorithm to find natural grouping or clusters in the dataset. There are several approaches and several different clustering algorithms to unsupervised classification. Most of these algorithms can be classified into two broad categories: partitioning and hierarchical methods. A partitioning method constructs K clusters. That is, it classifies the data into K groups. K is a user-specified number of classes. The hierarchical methods do not construct a single partition with K clusters but they deal with all values of K in the same run (Kaufman and Rousseeuw, 1989). Clustering algorithms used for the unsupervised classification of remotely sensed data generally vary according to the efficiency with which the clustering takes place. Ball and Hall (1965) proposed the Iterative Self Organizing Data Analysis Technique (ISODATA) clustering method. The ISODATA is a widely used and a simple clustering algorithm (Jain, 1989), which is a partitioning method of cluster analysis.

PAGE 75

60 (A) (B) Figure 3-15. Road mask. (A) Road mask created to mask out the urban regions in the SFRW. The green patches represent the road network in the watershed. (B) Zoomed in road network.

PAGE 76

61 (A) (B) Figure 3-16. Road masked SFRW. (A) Road features masked out from the raw image. (B) Zoomed in road mask.

PAGE 77

62 The key attributes of this method are discussed. The method starts with a clustering into a given number K of clusters. In the second step, outliers and very small clusters are eliminated; they are disregarded in the remainder of the method. Next, a lumping function is used (Kaufman and Rousseeuw, 1989). It is iterative as it makes a large number of passes through the remote sensing dataset until specified results are obtained (Jensen, 1996). This algorithm is simple to use as it require minimum input from the analyst. The parameters required for the clustering analysis using ISODATA are the following: N: The maximum number of clusters or classes to be identified. M: The number of iterations to be performed before the algorithm stops. T: The convergence threshold, which is the maximum percentage of pixels that are supposed to be unchanged between the iterations. The analyst has to specify the maximum number of classes for the analysis. The algorithm basically tries to locate the mean vector value for each class. On the first iteration of the ISODATA algorithm, the means of N clusters are arbitrarily determined. After each iteration, a new mean for each cluster is calculated, based on the actual spectral location of the pixels in the cluster. Then, these new means are used for defining clusters in the next iteration (Swain and Davis, 1985). The sample pixels are tentatively reassigned using the new class means, and the procedure repeats in this fashion until the class means no longer change. At this point, the tentative means are assumed to be good estimates of the class mean vectors. All pixels in this image are then assigned to a class using the Minimum Distance to Mean (MDM) classifier (Schott, 1997). The MDM

PAGE 78

63 classifier is a simple classification strategy in which the average spectral value for each class in all the bands is used to classify the remaining of the pixels in the scene. This decision rule is computationally simple and commonly used. It requires that the user provides the mean vectors for each class in each band from the training data (Jensen, 1996). The distance of a given point in DN-space is calculated from the centroid of each cluster, and the minimum distance is used to identify the required cluster (Rees, 1990). This is illustrated schematically in Figure 3-17. The classification of the pixel in X is based on the Euclidian distances D1, D2 and D3 to the mean centers at M1, M2 and M3. Based on this figure, the pixel in X will get classified into class 2. The ISODATA performs the classification in a similar fashion. The resultant classes will be indicative of the natural clusters in the data. They may or may not correspond to land cover or material classes. In some cases, classes are formed with no obvious common characteristic. Because of this limitation, unsupervised classification is often used as a preprocessor for other algorithm. In practice, it is often useful to run an unsupervised K means classifier with the initial estimate of K slightly greater than expected to “see” what types of spectral groupings naturally occur in the image (Schott, 1997). An unsupervised classification with a maximum of 20 to 30 classes was performed on the three images of the SFRW. As mentioned above, the objective of this analysis was to identify the natural clusters in the dataset and also to help identify better training sites to develop the spectral signatures for supervised classification. The resulting clusters did not specify any particular land cover class in the scene, but the natural clusters.

PAGE 79

64 Figure 3-17. Working of the MDM classifier. 3.6.3.2.3 Supervised classification The supervised classification approach is more controlled by the analyst. The analyst controls the categorization of the pixels into different land cover classes. To achieve this, the analyst creates numerical interpretation keys for each land cover class. A numerical interpretation key represents the spectral attributes for each land cover class in all the bands. During the classification process, the algorithms compares the pixels in the study area on a pixel-by-pixel basis to the spectral attributes in the numerical interpretation key and groups it to the category it most likely belongs to (Lillesand and Kiefer, 1994). Three distinct stages can be identified in the supervised approach. They are the training stage, classification stage and the output stage. The fundamental difference between the supervised and unsupervised approach is the order in which these stages are carried out by the analyst. In the supervised approach, the training stage is followed by the classification stage. In the unsupervised approach the classification of the image data to identify the natural clusters in the data set is performed first. The analyst then identifies these clusters to determine the appropriate land cover class.

PAGE 80

65 In the training stage, the analyst prepares a numerical interpretation key by defining training areas for each desired land cover class. These keys are also referred to as “signatures”. In the classification stage, a clustering algorithm compares the pixels in the image data to the signatures and categorizes them into a group it closely resembles. If the spectral attributes of a pixel are not close to the signatures developed by the analyst, it is classified as ‘Unknown’. After the pixel categorization process, the results of the analysis are presented to the user in the form of thematic maps or tables. These three steps are shown below. Figure 3-18. Steps in supervised classification of image data. 3.6.3.2.3.1 Training stage Owing to inherent differences in absorption spectra of earth’s surface features, different features manifest very different spectral emittance characteristics. These differences in the spectral response characteristics permit the distinction between different land cover types. The core of the supervised classification process lies in representing the desired land cover classes as sets of descriptive statistics that represent its overall spectral response pattern. These descriptive statistics or the numerical interpretation keys are commonly referred to as ‘spectral signatures’. The spectral

PAGE 81

66 signatures for the desired land cover categories are compiled in the training stage. The quality of the training process determines the success of the classification stage and therefore, the value of the information generated from the entire classification effort. The development of the signature files can be a subjective process. This will require the analyst to have a very good knowledge or reference data about the study area (Lillesand and Kiefer, 1994). The collected ground truth data were used to identify these training samples. Spectrally, the 2003 winter image was quite different from the 2000 and 1990 summer images. Hence, two sets of signature sets had to be developed for proper classification of the image data. ERDAS IMAGINE enables the identification of training samples using the ‘signature editor’ option. The training sites collected during the ground truth data collection were overlaid over the images to delineate the training area. There are a number of ways to develop these signatures sets. They can be developed using digitized polygons, user defined polygons, seed pixel definitions and using thematic raster layer. The user defined polygon approach is more effective that the other methods and hence was adopted for this study. The system then calculates statistics from the pixels to define the mean vector or centroid for each land cover class in all the bands. The unsupervised thematic layers were also used as a reference to define these signatures. The spectral signatures for the winter and summer scene are quite different for the agricultural classes. There were some regions within the agricultural classes that were cultivated and harvested in both the scenes. Hence, two different signature sets were developed to consider the spectral variability within agricultural classes. After classification, the two agricultural classes were merged into a single class. The two agricultural classes are

PAGE 82

67 denoted as ‘Agricultural 1’ and ‘Agricultural 2’ in the mean plots. The mean signature plots for the summer and winter scenes are shown in Figure 3-19 and Figure 3-20. Figure 3-19. Mean plot of land cover classes for the summer scene. Refinement of the training sets was conducted to ensure that the signatures developed indeed represent the reflectance from appropriate land cover classes. The main objectives of the signature refinement step are: To check the quality of spectral statistics developed for each land cover class. To make sure that the training areas selected to represent land cover classes are spectrally pure.

PAGE 83

68 To add/ delete/ recompile signatures to get the best possible seperability among land cover classes. Figure 3-20. Mean plot of land cover classes for the winter scene. The evaluation of signatures can be assessed using different strategies. The spectral signature seperability statistics and the feature space plots were used to evaluate the signature files used in this study. The signature seperability is a measure of the n-dimensional spectral Euclidean distance between the mean vectors of a pair of signatures. As the distance between the pairs of signatures increase, the possibility to make a better classification increases. One of the most commonly used statistical parameter in the

PAGE 84

69 determination of seperability is divergence. Divergence is defined as the covariance weighted distance between mean vectors of land cover classes (Lillesand and Kiefer, 1994). Transformed divergence was used for this study. The transformed divergence values are inversely proportional to the spectral distance between the pair of signatures. The values decrease exponentially with increasing distance between the pair of signatures. The transformed divergence values for the signatures sets developed for winter and summer scenes are shown in Figure 3-21 and 3-22. Typically, the transformed divergence values range from 0 to 2000. If the values are greater than 1900, then the pair of signatures are spectrally separated to produce accurate classification. If the values are between 1700 and 1900, the pair of classes are fairly separated. Values below 1700 indicate small spectral distance between the classes and hence accurate grouping between the classes is difficult (Jensen, 1996). Figure 3-21. Transformed divergence distance values for the winter scene.

PAGE 85

70 Figure 3-22. Transformed divergence distance values for the summer scene. 3.6.3.2.3.2 Classification stage The classification stage is the next step in the supervised classification approach. Using the numerical interpretation keys/signature files, a classification algorithm groups the pixels in the study area. The Gaussian Maximum Likelihood algorithm is the most accurate classifier in ERDAS-Imagine 8.5. One of the assumptions for the effective functioning of the maximum likelihood algorithm is the normal distribution in the bands of data. Based on the assumption, the algorithm evaluates the mean vector and covariance matrix to determine the distribution of response pattern for each class. Using all these parameters, the Probability Density Function (PDF) for the land cover classes are computed. The PDF’s are used to determine the land cover class for an unknown pixel by computing the probability of the pixel value belonging to all other categories (Lillesand and Kiefer, 1994). Using the ‘Clump, Sieve and Fill’ functions in ERDAS-Imagine, a post classification smoothing was used for the urban areas. The basic objective of this analysis

PAGE 86

71 was to fill the areas of less than four adjacent pixels of the same class with a class value of the neighboring pixel. 3.6.3.2.3.3 Output stage The final supervised classified thematic maps for 1990, 2000 and 2003 are shown in Figure 3-23, 3-24 and 3-25. 3.6.4 Accuracy Assessment Accuracy assessment is used to determine the quality of the information obtained from the classification process (Congalton, 1991). Using the collected ground truth data, the accuracy assessment for the classified thematic maps was conducted. One of the most common means of expressing classification accuracy is the preparation of a classification error matrix. According to Lillesand and Kiefer (1994), “Error matrices compare, on a category-by-category basis, the relationship between known reference data and the corresponding results of an automated classification”. The matrix stems from classifying the sampled training set pixels and listing the known cover types used for training versus the pixels actually classified into each land cover category by the classifier. Several characteristics of the classification process can be understood using the error matrix. The producer’s accuracy is a measure of the accuracy of classification of a particular class. The user’s accuracy, which is the measure of the probability that a pixel classified into a given class accurately represents that class on the ground, were determined (Lillesand and Kiefer, 1994). According to Congalton and Green (1999), the Kappa statistics coefficient is a tool to determine if the classified LULC map is relatively better than a map generated by randomly assigning labels to areas. The Kappa is powerful metrics because it takes into account all the elements of the error matrix.

PAGE 87

72 Figure 3-23. Land cover map of 1990 SFRW.

PAGE 88

73 Figure 3-24. Land cover map of 2000 SFRW.

PAGE 89

74 Figure 3-25. Land cover map of 2003 SFRW

PAGE 90

75 Table 3.4. Accuracy assessment report for 2000 classified image. ERROR MATRIX Reference Data Classified Data Pine plantations Wetland Upland Forest Agriculture Rangeland Urban Water Pine Plantations 22 1 1 2 0 0 0 Wetland 5 7 0 0 0 0 0 Upland Forest 1 6 10 0 0 0 0 Agriculture 1 0 0 41 12 4 0 Rangeland 0 0 0 2 7 0 0 Urban 1 0 1 2 0 16 0 Water 0 0 0 0 0 0 6 ACCURACY TOTALS Class Reference Totals Classified Totals Number Correct Producer’s Accuracy User’s Accuracy Pine Plantations 30 26 22 73.33% 84.62% Wetland 14 12 7 50.00% 58.33% Upland Forest 13 17 10 76.92% 58.82% Agriculture 48 58 41 85.42% 70.69% Rangeland 19 9 7 36.84% 77.78% Urban 20 20 16 80.00% 80.00% Water 6 6 6 100.00% 100.00% Total 150 148 109 Overall Classification Accuracy = 72.67% KAPPA STATISTICS Overall kappa statistics = 0.6572

PAGE 91

76 Table 3.5. Accuracy assessment report for 2003 classified image. ERROR MATRIX Reference Data Classified Data Pine plantations Wetland Upland Forest Agriculture Rangeland Urban Water Pine plantations 27 1 1 0 0 0 0 Wetland 1 14 3 0 0 0 0 Upland Forest 0 2 19 0 0 0 0 Agriculture 1 0 1 52 15 1 0 Rangeland 0 0 0 0 4 0 0 Urban 2 0 1 3 0 23 0 Water 0 0 0 0 0 0 6 ACCURACY TOTALS Class Reference Totals Classified Totals Number Correct Producer’s Accuracy User’s Accuracy Pine plantations 31 29 27 87.10% 93.10% Wetland 17 18 14 82.35% 77.78% Upland Forest 25 21 19 76.00% 90.48% Agriculture 55 70 52 94.55% 74.29% Rangeland 19 4 4 21.05% 100.00% Urban 24 29 23 95.83% 79.31% Water 6 6 6 100.00% 100.00% Total 177 177 145 Overall Classification Accuracy = 81.92% KAPPA STATISTICS Overall kappa statistics = 0.7729 The Kappa statistics coefficient range between 0 and 1. Zero indicates no agreement and 1 indicates complete agreement. Kappa values are usually grouped into 3

PAGE 92

77 categories. Kappa values that are greater than 0.80 indicate a strong agreement. Values between 0.40 and 0.80 show moderate agreement and values less than 0.40 represent poor agreement. The accuracy assessment reports for 2000 and 2003 images are shown in Table 3.4 and 3.5. The ERDAS-Imagine’s accuracy assessment tool was used for this purpose. 3.6.5 Change Detection Monitoring change is often perceived as the most common application of space-borne remote sensing techniques in the global, ecological and environmental change sectors (Roughgarden et al., 1991). Many researchers believe that the integration of remote sensing and environmental change analysis is important for ecologist and environmentalist to meet the challenges of the future (Luque, 2000). According to Macleod and Congalton (1998), change detection is a procedure to determine the change of a particular land cover class between two or more time periods. This is achieved by providing quantitative information on the spatial and temporal distribution of the land cover class. Change detection procedures are important tools for monitoring and managing natural resources and to determine natural and human induced changes in the landscape. Four important considerations to determine human induced effects through a change detection analysis are (1) determining the changes in the landscape (2) assessing the nature or trend of change (3) quantifying the areal extend of change and (4) determining the spatial pattern of change (Macleod and Congalton, 1998). The supervised classified thematic maps for 1990, 2000 and 2003 are shown in Figure 3-23, 3-24 and 3-25. The areal extend of land cover classes in 1990, 2000 and in 2003 are shown in Table 3-6, 3-7 and 3-8 and the percent distribution is shown in Figure 3-26, 3-27 and in 3-28. Table 3.9 compares the land covers extent (in hectares) across the

PAGE 93

78 three dates and also with the 1995 LULC map prepared by the SRWMD and the SJRWMD. Figure 3-26. Percent distribution of land cover classes in 1990. Table 3-6. Areal extent of land cover classes in 1990. Land cover Area in Hectares Percent distribution Pine plantations 105,723 29.45 Wetland 49,058 13.67 Upland Forest 32,047 8.93 Agriculture 84,478 23.53 Rangeland 54,195 15.09 Urban 19,805 5.52 Water 4,111 1.14 Noise 9,552 2.67

PAGE 94

79 Figure 3-27. Percent distribution of land cover classes in 2000 Table 3-7. Areal extent of land cover classes in 2000 Land cover Area in Hectares Percent distribution Pine plantations 97,952 27.29 Wetland 60,570 16.87 Upland Forest 35,885 10.00 Agriculture 115,899 32.29 Rangeland 19,538 5.44 Urban 20,238 5.64 Water 4,070 1.13 Noise 4,816 1.34

PAGE 95

80 Figure 3-28. Percent distribution of land cover classes in 2003. Table 3-8. Areal extent of land cover classes in 2003. Land cover Area in Hectares Percent distribution Pine plantations 83,536 23.27 Wetland 64,353 17.93 Upland Forest 37,517 10.45 Agriculture 133,861 37.29 Rangeland 15,486 4.32 Urban 20,238 5.64 Water 3,977 1.10

PAGE 96

81 Table 3.9. Comparison of the areal extent of land cover classes in percent distribution. Land cover class 1990 1995 2000 2003 Pine plantations 29.45 32.16 27.29 23.27 Wetland 13.67 16.23 16.87 17.93 Upland Forest 8.93 14.68 10.00 10.45 Agriculture 23.53 24.21 32.29 37.29 Rangeland 15.09 3.65 5.44 4.32 Urban 5.52 8.81 5.64 5.64 Water 1.14 0.26 1.13 1.10 Noise 2.67 1.34 3.6.5.1 Trajectories of land cover change Trajectories can be described as generic paths of change between factors that influence the dynamic nature of human-environment relationships and their effects. The factors that influence this relationship can be in the form of government policies, change in land owner, demographic issue. The analysis of trajectories of change is important to understand factors and to maintain sustainability in the human – environmental relationship (Kasperson et al., 1995). The specific trajectories of change within land cover classes were analyzed. The land cover shifts between 1990 and 2000 are shown in Table 3-10 and the land cover shifts between 2000 and 2003 are given in Table 3-11. Since agricultural land cover classes are considered to be prominent source of nitrate-nitrogen loading to the subsurface, the thematic images for 1990, 2000 and 2003 were individually reclassified to represent two classes, agricultural and non-agricultural land cover classes. Such a classification helps focus on the specific trajectories of change within land cover classes that have the potential to cause high nutrient loadings to the ground water system. After reclassifying the images, a ‘three data change image’ was created using image addition techniques.

PAGE 97

82 Table 3-10. Land cover shifts between 1990-2000. Table 3-11. Land cover shifts between 2000-2003.

PAGE 98

83 The change image represents land cover information of each pixel in the three years. Adapted from Mertens and Lambin (2000), the categories of change are described in Table 3-12. Table 3-12. Land cover categories of change. Category 1990 2000 2003 Land cover class 1 (NNN) Non-agricultural Non-agricultural Non-agricultural Stable non-agricultural class 2 (NNA) Non-agricultural Non-agricultural Agricultural Recent agricultural class 3 (NAN) Non-agricultural Agricultural Non-agricultural Instable or transient non-agricultural class 4 (NAA) Non-agricultural Agricultural Agricultural Permanent agricultural class 5 (ANN) Agricultural Non-agricultural Non-Agricultural Permanent non-agricultural class 6 (ANA) Agricultural Non-agricultural Agricultural Instable or transient agricultural class 7 (AAN) Agriculture Agriculture Non-agriculture Recent non-agricultural class 8 (AAA) Agricultural Agricultural Agricultural Stable agricultural class 9 (Other) Noise Noise Noise Null In the SFRW, these categories of change are shown in Figure 3-29. For example, a pixel that falls in the agricultural class in all the three years is considered a stable agricultural field and no factors have influenced the class value in this pixel. A pixel that falls under the non-agricultural class in 1990 and falls in the agricultural class in 2000 and again falls under the non-agricultural class in 2003 is an example of an instable or transient class. The land cover is constantly changing with response to external stimuli or factors. Particular classes of interest for this study are categories 2, 4 and 8. Category 2 represents pixels that are non-agricultural in 1990 and 2000 and have been recently converted to agricultural class. Category 4 represents the transition from non-agricultural

PAGE 99

84 class in 1990 to stable agricultural classes in 2000 and 2003. These categories represent the extent of agricultural expansion within the SFRW. The relative percentages of change of classes are given in Figure 3-30. The areal extent associated with the above mentioned change classes are given in Table 3-13. There are many inter-related processes that drive land use and land cover changes in a landscape. Generalizations of these change patterns and their effects are possible by relating the specifics of change to related physiographical and environmental characteristics. Some parts of the watershed can be more sensitive to LULC change than others. To get a better understanding about the effect of increasing agricultural activities in the watershed, the specific trajectories of land cover change (NAA and NNA) are compared to the land resource attributes of soils and geology. The percent distribution of soil orders and geology associated with the land cover shifts are shown in Figure 3-31, 3-32, 3-33 and 3-34. 3.7 Discussions and Conclusions 3.7.1 Accuracy of Image Classification The desired overall classification accuracy of 82% was achieved for the 2003 image and 72% for the 2000 image. One of the biggest problems encountered in classifying the winter scene was the spectral similarity between urban, rangeland and certain agricultural classes. As a result, overall classification accuracy and producer’s accuracy for each of these classes suffered in earlier trials. Pre-classification scene stratification of urban areas from non-urban areas improved producer’s accuracy of urban and rangeland classes and urban and agricultural classes, and hence overall accuracy of the classification improved. Producer’s accuracy for the 2003 classified image of 95.8% was achieved for the urban class, but the user’s accuracy was around 79.31%. The

PAGE 100

85 agricultural class had highest producer’s accuracy of 94.5% and user’s accuracy of 74.2% in the 2003 classified image. Figure 3-29. Trajectories of land cover change in the SFRW.

PAGE 101

86 Figure 3-30. Percent distribution of land cover change categories. Table3-13. Aerial extend of land cover change class Land cover change category Area in Hectares Percent distribution NNN 187,853 53.88 NNA 12,546 3.60 NAN 4,459 1.29 NAA 61,889 17.75 ANN 26,424 7.56 ANA 6,058 1.74 AAN 1,347 0.37 AAA 48,120 13.81

PAGE 102

87 Figure 3-31. Percent distribution of soil orders in NAA change class. Figure 3-32. Percent distribution of geological classes in NAA change class.

PAGE 103

88 Figure 3-33. Percent distribution of soil orders in NNA change class. Figure 3-34. Percent distribution of geological classes in NNA change class.

PAGE 104

89 Except for the wetland class, low producer’s accuracy was observed only in these two classes for the 2003 classified image, implying that only around 79.3% and 74.2% of the pixels that were classified under the class may actually represent the respective class on the ground. Rangeland classes are poorly classified in the 2003 image. Rangeland classes had low producer’s accuracy of 21.5%. Distinguishing agricultural fields from rangeland or urban areas is rather difficult using the winter scene, as a result of harvest and/or dry conditions. The spectral response pattern from the nearly bare soil in all these land cover classes is quite similar and this results in pixels getting misclassified among these three classes. Multi-seasonal images are essential to distinguish agricultural fields from other spectrally similar classes. This differentiation is quite difficult using a single data image (Reese et al., 2002). Multi-seasonal images from the same year of the SFRW would have helped to better differentiate rangeland and agricultural land cover classes. For the 2000 image, overall accuracy of the classification and accuracy of all land cover class, as reported in the error matrix, were not as high as that of the 2003 classified image. Agricultural land cover classes had producer’s accuracy of 85% and users accuracy of 710% for the 2000 classified image. Classifications of rangelands were slightly better with producer’s accuracy of around 37% and user’s accuracy of 78% (2000 classified image). Cropped agricultural fields were easier to distinguish from the rangeland classes in the summer image. This is primarily due to the fact that agricultural fields are usually cropped in summer and harvested before winter. The apparent red tone in the summer image and green tone in the winter image are due to this fact. Producer’s and user’s accuracy for the urban class were 80% and 80%, respectively.

PAGE 105

90 Unsupervised image clusters were used as a reference to develop signature sets for both seasons. Using clusters from respective unsupervised images improved classification accuracy by 7%. These clusters were particularly useful in differentiating classes of pine plantations and wetlands and pine plantations and upland forest. Pine plantations had producer’s accuracy of 87.10% and users accuracy of 93.10% for the 2003 classified image. Pine plantations were the best represented land cover class in the map. Wetland classes had producer’s accuracy of 82.3% and user’s accuracy of 77.78% for the 2003 classified image. For the summer scene (2000), the pine plantations had a producer’s accuracy of 74% and the user’s accuracy of 85%. The accuracy for wetland classification in the 2000 classified image were lower than anticipated given the effort to obtain homogenous reference data and the effort to create spectrally separable signatures. When wetland classes were misclassified, they were often confused with pine plantations or upland forest classes. One explanation may be due to the presence of forested wetlands in most part of the watershed. Forested wetlands are usually dominated by wooded vegetation and flooded bottom. Forested wetlands have unique mapping problems using remote sensing techniques. In most cases, wetlands are usually identified on the basis of visible hydrology, vegetation, and geography. The presence of a forest cover obscures these wetlands (Augusteijn and Warrender, 1998). Hence, the use of conventional methods of remote sensing is not effective in this category. Accurate delineation of forested wetlands requires the use of RADAR or other hyperspectral images. The use of multi-seasonal images of the same year and hyperspectral images would have definitely increased the overall accuracy of classification and also classification accuracy of each individual land cover class. Due to the unavailability of resources, these options were not

PAGE 106

91 applied. A Kappa coefficient of 0.7729 was achieved for the 2003 scene classification and 0.6572 for the 2000 scene. The lab generated land cover class statistics for 1990, 2000 and 2003 were compared to the 1995 LULC statistics in Table 3-9. The 1995 LULC maps were created by SRWMD and SJRWMD. The overall trend in changes within land use and land cover classes can be seen. However, the statistics for a few classes do not agree. One reason for this discrepancy can be attributed to the amount of noise present in the raw remote sensing images that inhibit the recording of reflectance form the surface features and in turn inhibits the land cover classification in the region. Also, the 1995 LULC maps were generated by using exhaustive ground truth information and aerial photographs providing higher quality output. Previous studies by Reese et al. (2002) have document land cover classification accuracy of 90.32% for Level 1 LULC classes. This study was carried out for four years using extensive ground truth data, aerial images and other ancillary data. Pre-classification scene stratification of urban and non-urban classes and wetland and upland classes were performed prior to a supervised classification of Landsat TM images. In this study, low producer’s accuracy was observed for upland mixed forest classes (50%) and shrubland (64%). The author states that the reason for the low accuracy rates of these two classes might be due to the spectral homogeneity of upland forests with pine and lack of ground truth data. Wetland classes had an accuracy of 84%. The author concludes that the pre-scene stratification and guided clustering techniques were helpful in classifying spectrally similar classes of urban and barren land and wetlands and uplands. There were some limitations to this study. They were:

PAGE 107

92 Presence of cloud/cloud shadow/ noise that obscured the reflectance from earth surface features in the 1990 and 2000 image. Non-availability of anniversary images that would have been helpful in developing accurate spectral descriptors for the land cover classes. The images used in this study represent extremes in hydrology (summer and winter) and agricultural practices (cultivation and harvest) and hence change detection errors introduced by these seasonal differences are expected. However, based on overall classification accuracy, classification accuracy for each land cover class and the Kappa coefficient, the accuracy assessment of the 2003 and 2000 land cover were the best achievable considering constraints due to resources, ground truth data and others. 3.7.2 Land Cover Change Analysis There is a constant change within the land cover classes in all three years. Forested land cover types (pine plantations and upland forests) occupy nearly 40% of the watershed area. The forested regions in the watershed did not change dramatically over the years. Upland forests have an almost constant distribution through the three time periods. The areal extent of pine plantations decreased by 5% from 1990 to 2003. In 1990, the wetlands occupied about 14% of the watershed area, the distribution increased to around 17% in 2000 and 2003. One explanation would be the extreme drought conditions in north-central Florida in 1990, which possibly lead to an under representation of wetlands in 1990. As can be seen in Figures 3-26, 3-27 and 3-28, rangelands decreased and agricultural lands increased dramatically over the last 13 years. Trajectory analysis enables analysis of pixel-by-pixel land cover change over the three dates. The pixel based land cover shifts between 1990 and 2000 are shown in Table 3-10. Some of the prominent shifts in these land cover classes are discussed. Based on this analysis, it can be seen that 9.35% of the rangelands in 1990 shifted to agricultural

PAGE 108

93 land cover classes in 2000. This shift is accompanied by a 2.5% shift from the agricultural class to rangelands. 7.35% of the pine plantations in 1990 shifted to agricultural land cover in 2000. However, 4.52% shifted from agricultural land cover to pine plantations. These shifts are schematically represented in Figure 335. Other land cover shifts included the conversion from rangelands to pine plantations which accounted for 3% of the total land cover shifts in the watershed. The wetland shifts to pine plantations and upland forests were 3.55% and 3.90% between 1990 to 2000, respectively. 2.40% of the pine plantations shifted to upland forests and 1.8% of the upland forests shifted to pine plantations between 1990 and 2000 (Table 3-10). Figure 3-35. Percentage of LC shifts between classes between 1990 and 2000. The pixel based land cover shifts between 2000 and 2003 are given in Table 3-11. The majority of the changes were associated with the wetland class. 6.24% and 3.40% of the wetlands in 2000 converted to pine plantations and upland forests, respectively. This was accompanied by 7.52% shift from pine plantations to wetlands and 3.72% shift from upland forests to wetlands. 2.10% of the rangelands shifted to agriculture between 2000 and 2003. Overall, the land cover shifts are more dramatic from 1990 to 2000 (10 year period) when compared to the time period 2000 to 2003 (3 year period).

PAGE 109

94 Analysis of the three data change image reveals landscape change is about 32% (NNN and AAA) of the watershed area. Categories 1 and 8 refer to stable classes within change within land cover classes. Stable agricultural and stable non-agricultural classes constitute 68% of the landscape. Stable non-agricultural classes are predominantly distributed along the eastern part of the watershed, which are mostly forested regions. Between 1990 and 2003 there was about 18% increase in agricultural land cover (NAA) and 8% conversion from the agricultural class to the non-agricultural class (ANN). Hence, agricultural expansion accounted for an additional 10% of the watershed area between 1990 and 2000. Between 2000 and 2003 there was 3.6% increase in agricultural land cover (NNA) and 1.74% decrease in agricultural classes (ANA). From the Figures 3-31, 3-32, 3-33 and 3-34, it can be seen that nearly 42% of the agricultural shifts between 1990 and 2000 have occurred in Ultisols and 18% of the shifts in Entisols. 45% of the shifts to agricultural classes between 2000 and 2003 have occurred on Ultisols and 15% on Entisols. Nearly 60% of the land cover shifts to agricultural classes have occurred in Ultisols and Entisols, which are highly permeable for soil nitrate-nitrogen leaching to ground water systems. Clay sands constitute the major part of the geology pertaining to these land cover shifts. However, 40% of the land cover shifts occur in limestones and fine sand. Limestones and fine sand geology are favaroble for nitrate-nitrogen leaching to ground water systems. The accuracy of trajectory analysis depends on the geometric registration of the three images. A RMS error of less than 0.5 pixels (15 m) was maintained for registering the 1990 and 2000 images with the 2003 image. This error might cause slight discrepancies in the pixel based change analysis. Also, distinguishing forested wetlands

PAGE 110

95 from pine plantations and upland forests are rather difficult, considering the time, effort and resources. One of the limitations in using pixel based trajectory analysis is that the accuracy of the analysis depends on the accuracy of classification of each individual pixel. Errors might have creped in during the classification of each image. Also, the presence of clouds and cloud shadows in the 1990 and 2000 image obscures the reflectance from earth’s surface features. This might have affected the classification of the land cover change class ‘NAA’. In Figure 3-29, discrete patches of NAA can be seen along the Gilchrist and Union counties. These patches are associated with noise pixels surrounding clouds and cloud shadows that have escaped the cloud/cloud shadow mask. However, change trajectory analysis helps in identifying the overall trend or shifts within land cover classes. Previous studies by Mertens and Lambin (2000) and Southworth et al. (2002) have studied forest dynamics with non-forest land cover classes and have concluded that change trajectory analysis helps in getting a better perspective of the dynamic interrelationships within land cover change classes over a time period. These LULC changes have the potential to have a significant impact on water and soil contamination in north-central Florida. Earlier studies by Katz et al. (1999) have indicated that Alachua and Suwannee counties had the greatest extent of cropped agricultural lands in the SRWMD and the total estimated nitrogen inputs (million kilograms per year) ranged from 10.3 to 3.9 for Alachua and 2.8 to 10.9 for Suwannee counties. Columbia county had an input range between 2.2 to 5.8 and 1.0 to 6.1 (million kilograms per year) in Gilchrist county between 1955 and 1997. Katz et al. (1999) studied the loadings from the four main sources of nitrate-nitrogen to ground water systems: agricultural fertilizers, animal wastes from dairy and poultry and rangeland operations,

PAGE 111

96 atmospheric deposition and septic tank effluent and documented the temporal variation of these loadings. The changes in loadings from these sources were compared to LULC shifts. The nitrogen inputs from LULC change for Columbia, Suwannee, Gilchrist and Alachua counties between 1957 and 1997 are shown in Figures 3-36, 3-37, 3-38 and 3-39. The total nitrate-nitrogen input from Columbia county increased from 2.2 to around 5.8 (millions of kilograms) between 1975. The rates have decreased from the late 70s to the mid 80s. Between 1990 and 1997, the total nitrogen inputs from Columbia county increased from nearly 4.0 to 4.3 (millions of kilograms). The use of fertilizers increased from 1955 to the late 70s and decreased until the early 80s. The fertilizer use increased from 2.5 to around 3.7 (millions of kilograms) between 1990 and 1997. Nitrogen inputs from atmospheric deposition accounted for nearly 60% in the late 50s and continued to decrease until the early 80s. The rates increased after the 90s and decreased in 1997. The relative contribution of animal wastes inputs varied from about 15% to 30%. Based on these estimations, it can be inferred that the most significant contributor to nitrate-nitrogen loads from Columbia county are agricultural fertilizers and their rates have increased by several folds between 1955 and 1997. This increase can be inferred as significant increase in agricultural land use. The contributions of poultry activities have decreased between 1990 and 1997. Beef cattle activities have increased slightly after 1990 but have maintained a constant pace after about 1991.

PAGE 112

97 Figure 3-36. Estimated total N inputs and relative percentage of total inputs of nitrogen from other sources during 1955-97 in Columbia County (Source: Katz et al., 1999). Based on Katz et al (1999), in Suwannee county, the total nitrogen input increased from 2.8 to nearly 10.9 (millions of kilograms) between 1955 and 1997. A decreasing trend was observed between 1980 to 1990 and an increasing trend after 1990. The total nitrate-nitrogen input has increased significantly from 6.2 (millions of kilograms) in the early 90s to an all time high of around 11.0 (millions of kilograms) in 1997. Prior to 1960, atmospheric deposition accounted for nearly 40% of the total loads. The relative contribution of nitrate-nitrogen from agricultural sources increased from around 23% in 1955 to nearly 60% in 1980 and decreased until the early 90s. A fairly significant increase in the contribution of agricultural fertilizers was observed between the early 90s and 1997, where the total inputs increased by around 3.8 (millions of kilograms). Relative contribution of nitrate-nitrogen from animal sources (dairy and poultry) has increased since 1980. Between 1955 and 1995 the contribution of nitrate-nitrogen from animal

PAGE 113

98 wastes (poultry, beef cattle and dairy) contributed about 21% to 42%. These rates have decreased after 1995. Figure 3-37. Estimated total N inputs and relative percentage of total inputs of nitrogen from other sources during 1955-97 in Suwannee County (Source: Katz et al., 1999). Based on these rates, it can be inferred that the most significant contributor to nitrate-nitrogen loads from Suwannee county were agricultural fertilizers and their rates have increased significantly between 1993 and 1995. This increase was possibly caused by an expansion in agricultural land use between 1990 and 1995. Yet another trend that was observed was the decline in the contributions of animal wastes from beef cattle stations between 1990 and 1995. This decrease can be inferred as decrease in rangeland and animal feeding operations in Suwannee county between 1990 and 1995.

PAGE 114

99 In Alachua county, the total nitrate-nitrogen input ranged from 3.9 to nearly 4.4 (millions of kilograms) between 1955 and l997 peaking with about 10.4 million of kg in 1977. The total nitrate-nitrogen input increased continuously from 1955 to the late 70s. These rates significantly decreased until the mid 80s and have increased slightly after that. In 1955, atmospheric deposition was the significant contributor to nitrate-nitrogen loads in Alachua county. These rates decreased continuously until 1980. The contribution of atmospheric deposition increased from around 1.0 to 4.2 (millions of kilograms) between 1980 and 1997. The rates of agricultural fertilizers increased continuously from the late 50s to 1980. Figure 3-38. Estimated total N inputs and relative percentage of total inputs of nitrogen from other sources during 1955-97 in Alachua County (Source: Katz et al., 1999). These rates have decreased from around 8.2 to 3.4 (millions of kilograms) between 1980 and 1997. The contribution of poultry activities decreased after 1980 and did not

PAGE 115

100 contribute after around 1988. The contributions from beef cattle stations and dairy farms increased slightly in the late 80s. The contributions of beef cattle stations increased from around 1.8 to around 2.8 (millions of kilograms) between 1988 to 1997. Based on these rates, it can be inferred that the most significant contributors to nitrate-nitrogen loads from Alachua county are atmospheric deposition and agricultural fertilizers. The contributions of agricultural fertilizers decreased between 1990 and 1997. This decrease is most likely related to a reduction in agricultural activities between 1990 and 1997 in Alachua county. In Gilchrist county, the total nitrate-nitrogen input ranged from 1.8 to nearly 11.0 (millions of kilograms) between 1955 and 1999 (Katz et al, 1999). The total nitrate-nitrogen inputs show an increasing trend between 1955 to 1999. Prior to 1960 the atmospheric deposition was the significant contributor to nitrate-nitrogen loads from the county. The rates have decreased significantly ever since. The contributions of agricultural fertilizers increased from around 2.2 to 7.4 (millions of kilograms) between 1955 and 1997. The contributions of dairy farms have increased significantly from 0.2 to 2.8 (millions of kilograms) between 1975 and 1997. The contributions of beef cattle stations have decreased continuously from 1975 and 1980 and have maintained a constant pace ever since. Based on these rates, it can be inferred that the most significant contributor to nitrate-nitrogen loads from Gilchrist county are agricultural fertilizers and their rates have increased significantly between the mid 70s and 1997. These findings from Katz et al. (1999) translate into a possible expansion in agricultural activities in Gilchrist county and/or intensification in agricultural land use. Yet another trend that was observed was the increase in the contributions of animal wastes from dairy farms

PAGE 116

101 between the mid 70s and 1995. This trend can be inferred as a significant increase in dairy farming and feeding operations in Gilchrist county between 1975 and 1997. Figure 3-39. Estimated total N inputs and relative percentage of total inputs of nitrogen from other sources during 1955-97 in Gilchrist County (Source: Katz et al., 1999). Suwannee, Gilchrist and Columbia counties constitute the western part of the SFRW. Based on the studies by Katz et al. (1999), agricultural activities have significantly in Suwannee, Gilchrist and Columbia counties in the last few years. By comparing these results with the results of the change trajectory analysis, it can be seen that change classes of NAA and NNA are more concentrated in these counties. Analyzing the temporal variation of nitrate-nitrogen sources in Suwannee, Gilchrist, Columbia and Alachua counties, it can be seen that the estimated total nitrate-nitrogen inputs closely match with the agricultural fertilizer sales. Also, these inputs from Suwannee, Gilchrist and Columbia counties have increased manifolds in recent years.

PAGE 117

102 In addition to fertilizer inputs, other soil-landscape attributes known to have a profound effect on soil nitrate-nitrogen leaching to ground water systems are depth from the soil surface to the saturated zone, net recharge, aquifer media, soil media, topography, impact of the vadose zone and hydraulic conductivity. The results of the land cover change analysis in context of the DRASTIC vulnerability index described in chapter 2 are intriguing (Fig. 2-11). The DRASTIC index includes only land resource characteristics but does not consider LULC as a weighting factor. Combining DRASTIC vulnerability classes with results from the land cover change detection analysis provides more detailed information about potential/expected contamination in the SFRW. Comparing Figure 3-38 with 2-11, it can be seen that change classes of NAA and NNA are more concentrated in the western part of the watershed which are characterized by high DRASTIC values, suggesting that this region is highly vulnerable to groundwater contamination through the upper Floridian aquifer. In addition, the land cover change analysis showed prominent shifts in agriculture in the western part coinciding with Ultisols and karst topography that both can lead to accelerated leaching of nitrate into the aquifer. Moderate agricultural shifts were observed in the eastern part of the watershed, which is predominantly characterized by the soil order of Spodosols and a clayey geologic layer. Spodosols often have a confining spodic horizon that can slow down the leaching of nitrate-nitrogen. The same it true for clay-rich geologic layers that delays the leaching of nutrients from the unsaturated zone into the aquifer.

PAGE 118

CHAPTER 4 ANALYSIS OF SOIL NITRATE-NITROGEN VALUES 4.1 Introduction Eutrophication is the most common form of surface water impairment in the U.S. This is primarily caused by excessive loads of nitrogen and phosphorous. Eutrophication accounts for nearly 60% of river reach impacts in the U.S and are considered the most widespread pollution issue in estuaries. Pollution resulting from excessive loading of nitrogen and phosphorous is primarily due to non-point sources. Organic nitrogen exists in the form of organic soil material. It occurs primarily in manures, sewage wastes and other decomposing material, such as plants. Inorganic nitrogen, which is primarily produced from minerals and mineralization of organic materials, is introduced into the system by agricultural fertilizers or through precipitation. Inorganic nitrogen is essential for plant growth as it supplies the necessary nutrients for plants. There are many forms of inorganic nitrogen. Ammonium (NH 4 + ), nitrate (NO 3 ), and nitrite (NO 2 ) are some of the forms of inorganic nitrogen found in soils. Some of the inorganic forms, such as ammonium, are retained in the soil while others forms, like nitrate and nitrite, are soluble and leach into ground water systems. Some of the other forms are converted to nitrogen and ammonia gas and escape to the atmosphere. Ammonium is a positively charged inorganic form of nitrogen, which is attracted to the negatively charged clay/silt particles and is hence retained in the soil. Nitrate-nitrogen and nitrite-nitrogen are negatively charged and are repelled by the clay/silt particles in the soil and are hence transported out of the soil surface to the ground water system.

PAGE 119

Microbes act on the escaped nitrogen and can be converted to other inorganic forms of nitrogen that are useful to the plant or escape to the atmosphere. The various steps in the nitrogen cycle are explained in Figure 4-1. Figure 4-1. Nitrogen cycle. The extent of leaching is primarily related to nutrient concentration, soil type and texture. Clayey soils can hold more water than sandy soils and hence the leaching process is increased in sandy soils. Leaching is common in karstic terrain, providing macro pore structure for accelerated leaching. Some of the primary sources of nitrate-nitrogen in soils are agricultural fertilizers, manures, urban run-off, septic leachate, atmospheric

PAGE 120

deposition and activities such as wetland conversion. Inputs from non-point source pollutants are considered the major sources for water pollution in the U.S (Carpenter et al., 1998). Increasing concentrations of nitrate-nitrogen has been observed in ground water, springs and surface water monitoring station in the SRWMD, particularly within the SFRW. 95% of the residents in Florida rely on ground water sources for primary drinking water. Non-point source pollutants pose a significant environmental challenge because of the inability to identify the problem area, but methodologies have to be developed to strategically target non-point source problem areas. Hence, the objective of this study is to design a protocol that would best address the spatial variability of nitrate-nitrogen, by taking into consideration LULC categories and soil characteristics in the watershed. The design of the soil sampling protocol is discussed in the following section. The results of soil sampling events conducted based on the protocol are discussed in the later sections. 4.2 Sampling Site Selection 4.2.1 Assumptions The following criteria were used for development of the sampling scheme: Allocation of the number of sampling sites was based on expected impact of LULC and soil combination on nutrient loading. Allotment of the number of sampling sites was based on the areal extent of the particular LULC-soils combination category. LULC classes that may have a significant impact on nitrate-nitrogen loading were targeted, irrespective of areal extent. The number of sampling sites for LULC-soils combination classes that were not expected to have a significant impact in nitrate-nitrogen loading were allotted sampling sites in-proportion to the areal extend of the category.

PAGE 121

Minimum sampling locations were allocated to classes that have a relatively small areal extent and were not expected to be a significant contributor to the elevated nitrogen loads. Random selections within the categories were adopted to ensure unbiased selection of sites across the SFRW. 4.2.2 Materials 4.2.2.1 Software The following softwares were used for the analysis. ArcGIS Desktop 8.3 (Environmental System Research Institute (ESRI), Redlands, CA) Microsoft Excel (Microsoft Corporation, Seattle, WA). 4.2.2.2 Spatial data Various spatial datasets were used to implement the sampling scheme. The data were downloaded from the Florida Geographic Data Library (FGDL). The following data layers were used. SFRW boundary obtained from SRWMD. Road network for the SFRW, developed by the U.S. Census Bureau. LULC data (1995) for the SFRW, developed by the SRWMD and the SJRWMD. Soil Survey Geographic (SSURGO) soil data layers, developed by the Natural Resource Conservation Service (NRCS), USDA. 4.2.3 Methods The methodology for design of the sampling site protocol is shown in Figure 4-2. 4.2.3.1 Data acquisition The boundary of the SFRW was obtained from the SRWMD. The land use/land cover layers were derived from the 1995 Landsat Thematic Mapper (TM) imagery. Landsat images have a spatial resolution of 30m. Landsat images were analyzed and classified into different LULC categories and verified using ground truth information to

PAGE 122

generate the final LULC maps. The final thematic maps were generated by SRWMD and SJRWMD. Soil data were obtained from the Soil Survey Geographic (SSURGO) database derived at a scale of 1: 24,000. These soil data were developed by the Natural Resource Conservation Service (NRCS). These datasets were prepared by employing extensive soil survey techniques, aerial photo interpretation and other techniques. Parts of Suwannee, Gilchrist, Columbia, Alachua, Bradford and Clay counties constitute the SFRW. The soil and LULC information for all these counties were archived. To ensure accurate representation of all geographic features, a common reference system was chosen for the data layers. Albers equal area projection was chosen for the analysis. The Albers projection system preserves true area of the entities represented. As a result some minor negligible distortions to the shape and angle result. Projection parameters were specified using the ArcToolbox option. The parameters of Albers projection are given in Table 4-1. 4.2.3.2 Processing of LULC and soil data The LULC data obtained for each constituting county in the SFRW were merged to append the county-wise LULC information into a single mosaic. The LULC data for the SFRW was then ‘clipped’ off from the mosaic image. The ‘Merge’ and the ‘Clip’ functions in ArcGIS’s ‘Geo-processing wizard’ were used for this purpose. The LULC data layers for the SFRW are vector data layers. Computationally raster formats are easier than vector data sets. The LULC datasets were converted to rasters using the ‘Spatial Analyst’ extension in ArcGIS. The resolution of the output raster was specified to be 30 m. This was to ensure consistency with the Landsat image.

PAGE 123

Table 4.1 Projection parameters Parameters Projection System Albers Equal area False Easting 400000.00 False Northing 0.0 Central Meridian -84.00 Standard_Parallel_1 24.00 Standard_Parallel_2 31.50 Central Parallel 24.00 Datum North American 1983 Prime Meridian 0 Based on earlier studies by Spalding and Exner (1993), agricultural LULC types are significant contributors to elevated NO 3 levels in groundwater. To better target agricultural LULC classes, a reclassification of the LULC thematic map was performed. Reclassification of rasters helps the user to better visualize regions of interest within the given data set. This allows the user to assign values of preference, sensitivity, priority or some similar criteria to the raster (ArcGIS 3-D Analyst Manual, 2002). The maps of the SFRW exhibit a wide spectrum of LULC classes. A listing of all classes is provided in Appendix C. Anderson’s ‘Level 1’ class for the SFRW has eight land cover types. They are: Urban and Built-up Agriculture Rangeland Upland Forests Water Wetlands Barren land Transportation, Communication and Utilities. As mentioned, LULC classes sensitive in terms of elevated nitrogen load are the agricultural class. As a first step, all the other ‘Level 2’ LULC classes were aggregated into the corresponding ‘Level 1’ class. For example, the ‘Level 2’ classes of ‘Fresh Water

PAGE 124

Marshes’, ‘Salt Water Marshes’, ‘Bay Swamps’ were classified into the ‘Level 1’ class of ‘Wetlands’. LULC classes of ‘Urban and Built up’ and ‘Transportation and Utilities’ were merged into a single ‘Urban’ class. After this first reclassification, the layout comprised the ‘Level 2’ classes for ‘Agriculture’ and ‘Level 1’ classes for the rest of the land use types. At this stage, the data set comprised 38 classes. The ‘Level 2’ agricultural classes were individually studied and classified into an appropriate LULC class or kept as a separate class. ‘Level 2’ classes of Sod farms, Aquaculture, Dairies, Forest Regeneration, Pine Plantations, Improved Pastures and Specialty Farms were not aggregated into a ‘Level 1’ class because of their expected influence to nitrate-nitrogen loadings A summary of the reclassification procedure is given in Table 4-2. The reclassified LULC image of the SFRW is shown in Figure 4-3. The SSURGO datasets provide high level of spatial and categorical detail for soil attributes. An illustration of the database schema adopted for the SSURGO datasets is shown in Figure 4-4. SSURGO data (spatial) are related to a Map Unit Interpretation Record (MUIR) attribute data base, which contains variables describing soil characteristics in a set of related tables. These variables include characteristics like soil pH, water capacity, salinity, depth to bedrock, soil taxonomy and other physical and chemical attributes. The many tables in the MUIR are related with each other using a key unique identifier. This key field usually refers to a soil map unit called the Map Unit Identifier (MUID). All the tables in the MUIR database are linked via the MUID. Most tables contain the MUID field and can be related to other tables using the same key attribute. This enables the analysis of specific variables of interest for each soil type (Bolstad, 2002). The key attributes are shown in ‘bold’ format in Figure 4-4.

PAGE 125

Figure 4-2. Design of sampling site selection protocol.

PAGE 126

Table 4-2 Reclassification of land use features New LULC category Previous LULC category Nursery Blueberry Nurseries and Vineyards Ornamental Nurseries Tree Nursery Feeding Operations Feeding Operations Cattle Feeding Operations Poultry Feeding Operations Tree Crops Tree Crops Fruit Orchards Citrus Groves Other Groves Peaches Row Crops Row Crops Field Crops Fallow Cropland Mixed Crops Rangeland Rangeland Unimproved Pastures Woodland Pastures Barren Land Barren Land Abandoned Tree Crops Horse Ferns Kennels Shade Ferns Old Field Other Land Use Rural

PAGE 127

Figure 4-3. Reclassified LULC map.

PAGE 128

Figure 4.4. SSURGO database schema. The SSURGO datasets were used to map the spatial distribution of various soil orders across the SFRW. Information regarding the soil composition is stored in the Components Table. The MUID field was used to link tables. The MUID in the Map Unit Table and the MUID in the Components Table shared a many-to-one relationship. This implies that many rows in the first attribute table will have to be joined to a single row in the second table. Commonly, in the case of a many-to-one relationship, a spatial join is employed. Thus, the shapefile containing the spatial geometry was joined to the non-spatial attribute table. The new attribute table contained component information of the soil polygon. The same procedure was employed for all the counties within the watershed. The merge and clip functions were used to derive the soil information for the SFRW. Using the ‘Spatial Analyst’ extension, the soil vector data was converted to a 30 m raster. In order to produce soil order maps, two sets of reclassification procedures were adopted. The first reclassification step consisted of two grouping operation

PAGE 129

Grouping of soil complexes into a common ‘Miscellaneous’ class. This was done to ensure that the sampling locations are positioned in locations of known attributes. Grouping of soil phase into a single group. For example, datasets described as ‘Penney 2-5% slope’, ‘Penney 5-8% slope’ were both grouped into a single group ‘Penney soil’.A soil classification database describing the characteristics and the horizon of each of the soil type was created. The soil characteristic helped in the identification of the soil order by the descriptive formative elements. The horizon provides greater details about the soil profiles encountered in the SFRW. The soil orders present in the study area were Entisols, Ultisols, Mollisols, Histosols, Inceptisols, Spodosols and Alfisols. Soil types were reclassified to group soil data in terms of the soil orders. Figure 4-5 shows the spatial occurrences of different soil orders in the study area. Figure 4-5. Soil orders in SFRW.

PAGE 130

4.2.3.3 Design of sampling site selection protocol As mentioned, the extent of nitrate-nitrogen leaching depends both on the LULC category and also on the type of soil. Hence, to target the effect of various soil types on each LULC category, corresponding soil orders for each LULC category were first determined. The LULC-soil combination layers were created using the ‘Raster Calculator’ operation in ArcGIS. The function used is given below. Con ([LULC] = = X i & [Soil_Order] < = Y j , [Soil_Order], ) Equation 8 where X i represents a specific LULC category. The value of i varies between 1 and 16. Y j represents a specific soils category. The value of j varies between 1 and 7. represents a dummy variable. The function returns the appropriate soil orders if the first two conditions are met. The value of the dummy variable is returned when the conditions are not met. This operation was performed on all the LULC categories. The results are shown in Table 4-3. The table indicates the number of pixels under each category. As mentioned, the resolution of the LANDSAT TM is 30 meters; hence each pixel represents an area of 900 square meters. The area under each LULC-soil combination was calculated as a percentage of the total area of the watershed. This percentage value is expressed within parenthesis in Table 4-3. The LULC–soil combinations that had area coverage of less than 1 % were not allocated any sample locations, unless they were considered to be significant contributors to nutrient loading. The percentage of the aerial extends for the LULC–soil combinations were recalculated again. The recalculated percentages are shown in Table 4-4. The allocations of the number of samples were based on these percentages.

PAGE 131

Table 4-3 Area under each LULC-Soil combination category. LULC category Entisols Ultisols Spodosols Mollisols Alfisols Histosols Inceptisols Total Cover of Land use type Pine plantations 124,07 3.8 378,096 11.8 465,50 14.6 65 0 6,957 0.21 567 0 1,900 0.05 977,232 Upland Forests 143,041 1 4.42 251,884 7.9 116,699 3.6 60 0 11,258 0.35 623 0.01 14,005 0.4 537,570 Wetland 14,381 0.45 94,829 2.9 113,643 3.5 648 0.02 2,619 0.08 74,022 2.3 11,107 0.34 311,249 Improved Pastures 99,717 3.1 313,2689 9.8 54,722 1.7 0 0 8,262 0.25 96 0 6,090 0.19 482,155 Rangeland 22,968 0.72 55,686 1.7 27,937 0.87 0 0 1,797 0.05 359 0.01 797 0.02 109,544 Urban 75,321 2.3 154,787 4.8 85,856 2.6 72 0 2,720 0.08 514 0 3,480 0.1 322,750 Forest Regeneration 29,576 0.92 74,669 2.3 91,398 2.8 0 0 1,600 0.05 216 0 1,427 0.04 200,837 Crops 56,141 1.7 91,448 2.8 42,132 1.3 26 0 1,757 0.05 38 0 1,683 0.05 193,225 Water 1,887 0.05 3,398 0.10 3,040 0.09 4 0 84 0 232 0 791 0 9,436 Tree Groves 384 0.01 5,741 0.18 2,341 0.07 0 0 49 0 0 0 0 0 8,515 116

PAGE 132

Table 4 3. (Cont) Nursery 792 0 553 0 229 0 0 0 3 0 3 0 20 0 1,600 Sod Farms 0 0 523 0.01 43 0 0 0 1 0 0 0 0 0 567 Specialty Farms 124 0 158 0 41 0 0 0 24 0 0 0 0 0 347 Feeding Operations 164 0 801 0 1024 0.03 0 0 7 0 0 0 0 0 1,996 Dairy 407 0 55 0 19 0 0 0 0 0 0 0 0 0 481 Aquaculture 10 0 65 0 455 0.01 0 0 0 0 0 0 0 0 530 Barren Land 7037 0.22 12085 0.37 4788 0.15 0 0 120 0. 1 0 147 0 24,178 Total Cover of individual Soil orders 576,017 1,438,046 1,009,947 875 37,258 78,622 41,447 3,182,212 117 1 Number of pixels 2 Area of LULC-soil combination class as percentage (rounded to the nearest decimal) of the total watershed area.

PAGE 133

118 Table 4-4. Recalculated LULC-soil percentages. LULC category Entisols Ultisols Spodosols Mollisols Alfisols Histosols Inceptisols Pine plantations 4.05 12.35 15.21 0 0 0 0 Upland Forests 4.67 8.234 3.81 0 0 0 0 Wetland 0.47 3.09 3.71 0 0 2.41 0 Improved Pastures 3.25 10.24 1.78 0 0 0 0 Rangeland 0.75 1.82 0.91 0 0 0 0 Urban 2.46 5.05 2.80 0 0 0 0 Forest Regeneration 0.96 2.44 2.98 0 0 0 0 Crops 1.83 2.98 1.37 0 0 0 0 Water 0 0 0 0 0 0 0 Tree Groves 0 0.18 0 0 0 0 0 Nursery 0 0 0 0 0 0 0 Sod Farms 0 0 0 0 0 0 0 Specialty Farms 0 0 0 0 0 0 0 Feeding Operations 0 0 0.03 0 0 0 0 Dairy 0.01 0 0 0 0 0 0 Aquaculture 0 0 0 0 0 0 0 Barren Land 0 0 0 0 0 0 0 Table 4-5. Allocation of number of sample points based on the percent distribution Percentage of area covered Number of samples 0-2 4 2-4 5 4-6 6 6-8 7 8-10 8 10-12 9 >12 10 A random number generator was used to select potential sampling location in ArcGIS. The first version of the sampling sites in the SFRW is shown in Figure 4-6. 4.3 Replacement Sites Before soil samples could be collected from the 150 sites, the land owners had to be contacted to get their permission to access their property.

PAGE 134

119 Table 4.6. Sample location allotted to each LULC-soil combination. Land Use type Entisols Ultisols Spodosols Histosols Total Pine plantations 6 10 10 0 26 Upland Forests 6 8 5 0 19 Wetland 4 5 5 5 19 Improved Pastures 5 9 4 0 18 Rangeland 5 4 5 0 14 Urban 5 6 5 0 16 Forest Regeneration 4 5 5 0 14 Crops 4 5 4 0 13 Tree Groves 0 4 0 0 4 Dairy 4 0 4 0 8 Total 43 56 47 5 151 Figure 4-6. Selected sampling sites using the protocol. The parcel numbers for all 150 sites were obtained from the parcel GIS maps. Using the parcel information, the online archives and CD appraiser databases, the

PAGE 135

120 addresses of the land property owners were obtained. Most of the records in the appraiser database did not have phone number associated with the contact information. Online telephone directories like were used to arrive upon the land owner’s complete contact information. A database in MS Access was created with the land owner information for each site. A multi-tier approach was adopted to contact and obtain permissions from land owners. As a first step, letters were mailed to private land owners in the database and permission forms were submitted to access state owned land. After about three weeks, the land owners who did not respond to letters were contacted by phone or email. Also, local county extension agents were contacted to help the project in getting permissions from land owners operating priority land use classes like crop land, rangeland and other feeding operations. Numerous land owners were concerned about rules and regulations which might result from this study. The fear that state agencies like the Department of Environmental Protection and the SRWMD would use the results to develop new regulations limiting their land use activity was high and hence there were many sites, out of the 150 sites, for which permission to access the property was denied. For each of the rejected sampling site, three other replacement sites with similar LULC – soils combination were randomly identified in the watershed. The whole process of contacting the land owners were repeated for each rejected sampling site. By the end of August 2003, permissions to access 101 sites in the watershed were obtained. Prior to soil sampling, each of these sites was visited. The geographic coordinates, field LULC information and digital photographs for each site were recorded. As a result of LULC change or due to misclassification, some of the digital data on

PAGE 136

121 LULC were inaccurate. In these cases, replacement sites were identified. The 101 sites that were approved for the September (Fall) 2003 sampling event are shown in Figure 4-7. Figure 4-7. Approved sites for the fall 2003 sampling event. 4.4 Field Sampling Soil samples were collected from the 101 sites in September 2003. Composite soil samples were collected at four depths of 0 to 30 cm (layer 1), 30 to 60 cm (layer 2), 60 to 120 cm (layer 3) and 120 to 180 cm (layer 4) for each sample location. A sampling protocol was developed to address the quality assurance process. 4.4.1 Materials The following field sampling equipments were used for the field sampling procedure:

PAGE 137

122 Augers Buckets – approximately 3 inch buckets Plastic bags Sturdy coolers Marking pens Clinometers for slope measurement GPS unit 4.4.2 Methods A sampling protocol was developed to ensure quality control and also to address issues of sample support. It is necessary for the soil samples collected at different depths and different location to have the same support. As mentioned, soil samples were collected at four depths at each point. The sampling support in the upper profile (0 to 30 cm) was smaller than in the lower profile (0-60 cm). For example, the sampling support for the upper profile was 30 cm and for the lower profile was 60 cm. Different supports in the upper and lower profile would result in different standard error of mean for nitrate-nitrogen measured in the upper and lower profile, respectively. To adjust the sampling support, composite soil samples were obtained with more samples in the upper profile and fewer samples in the lower profiles. The design of the composite soil sampling scheme at each site is given in Figure 4-8. For example, in the case of layer one; individual soil samples were collected at four to five locations around the GPS location. The soil samples from all the five locations were emptied into one bucket, mixed thoroughly by hand to create a composite mix. Approximately 400 g of the mix were taken in a plastic sampling bag and transferred to a sturdy ice cooler. The soil samples were stored in this fashion until they were taken to the laboratory for the analysis. Composite soils were collected for the layer one by considering the entire soil support. However, for layer 2, 3 and 4 the composite soil

PAGE 138

123 samples were collected from approximately the uncontaminated lower 2/3 rd in each bucket. Figure 4-8. Composite soil sampling scheme for each sampling location. In September 2003, all of the 101 soil sites were sampled. Layer 1 and layer 2 soil samples were obtained from all sites. However, only 89 samples were collected from layer 3 and 59 samples from layer 4. The sites not sampled in layer 3 and 4 were either sites with high clay content or sites with high water table that inhibited collecting samples at such depths. 4.5 Laboratory Analysis The laboratory analysis of collected soil samples were conducted at the Forest Soils Laboratory, Department of Soil and Water Science, University of Florida.

PAGE 139

124 The collected soil samples were analyzed for nitrate-nitrogen content. Ten grams of moist soil and 50 ml of 2 M KCL were added in a polypropylene bottle and shaken for about one hour. The solution was then filtered using a 45 micron filter paper and refrigerated until analysis. The extraction process was usually carried out the same day or the following morning of collecting the sample. The samples were analyzed on a Rapid Flow Analyzer (RFA) for nitrate-nitrogen content. The detection limit for this procedure was 0.02 g/g soil. A MS-Access database was created to store the analyzed nitrate-nitrogen values along with the coordinates of the sampling location. 4.6 Analysis of Soil Sampling Results Shapefiles corresponding to layers 1, 2, 3 and 4 were created from the MS-Access database. These shapefiles comprise the coordinates of the sample points measured in that particular layer, the corresponding nitrate-nitrogen value and the LULC-soil combination of that point. Average nitrate-nitrogen value at each soil profile was also determined, based on a depth-weighted profile average. Geostatistical analyses of the nitrate-nitrogen values were conducted using ArcGIS Geostatistical Analyst extension and SPSS for windows (SPSS Inc., Chicago, IL). 4.6.1 Exploratory Spatial Data Analysis Exploratory Spatial Data Analysis (ESDA) can be defined as a set of techniques to quantify, visualize and better understand the distribution of spatial data. It helps in identifying outliers and patterns of spatial association within the datasets. The ESDA is often considered part of the preliminary exploration of the raw data before any formal analysis can be performed on the data. ESDA typically involves examining the data by displaying histograms, box plots or scatter plots and by computing summary statistics.

PAGE 140

125 4.6.1.1 Histogram and summary statistics A histogram can be defined as a graph that shows the frequency distribution of a dataset. In many cases, the datasets exhibit normal distribution, that is, most of the values are centered on the mean value with decreasing values towards the ends, but such a distribution cannot be expected in the case of environmental pollution measurement, especially non-point source pollutants. Approximation to normal distribution is essential for most of the interpolation techniques. The histogram distribution of nitrate-nitrogen values (g/g soils) in layer 1, layer 2, layer 3 and layer 4 and the profile average are shown in Figure 4-9, 4-10, 4-11 , 4-12 and 4-13. The histogram plots were generated using SPSS for Windows. Coun t Figure 4-9. Histogram of nitrate-nitrogen values in layer 1.

PAGE 141

126 Coun t Figure 4-10. Histogram of nitrate-nitrogen values in layer 2. Coun t Figure 4-11. Histogram of nitrate-nitrogen values in layer 3.

PAGE 142

127 Coun t Figure 4-12. Histogram of nitrate-nitrogen values in layer 4. Coun t Figure 4-13. Histogram of average nitrate-nitrogen value in each profile.

PAGE 143

128 The summary of the descriptive statistics for the nitrate-nitrogen value distribution for all layers and the profile average are given in Table 4-7. Table 4-7. Summary statistics of nitrate-nitrogen values (g/g soil) Parameters Layer 1 Layer 2 Layer 3 Layer 4 Profile Average Number of sites 101 101 89 59 101 Minimum value 0 0 0 0 0 Maximum value 7.60 9.39 9.80 8.78 6.54 Mean 0.70 0.63 0.87 0.67 0.70 Median 0.27 0.32 0.28 0 0.22 Standard deviation 1.34 1.30 1.89 1.59 1.31 Skewness 3.23 4.05 2.90 3.87 2.72 Kurtosis 14.63 23.17 11.32 18.25 10.13 From the histogram plots and the skewness values in the above table, it can be seen that all histograms were positively skewed. In such cases, the median value is lower than the mean value and is a better measure of the central tendency. There is a high frequency of zero values in all the layers. Maximum concentrations of nitrate-nitrogen level were observed in layer 3 of the soil profile. 4.6.1.2 Normal Q-Q plot The normal QQ plot is yet another tool to compare the distribution of the dataset to a normal distribution. QQ stands for Quantile-Quantile plot. Two identical distributions will result in a straight line in the plot. The quantile of the dataset is plotted against the quantile of normal distribution. Hence, the closer the dataset approximates the normal distribution, the straighter is the line in the plot (Johnston et al., 2001). The normal Q-Q plots for the nitrate-nitrogen values in layer 1, 2, 3, 4 and the average value are shown in Figure 4-14, 4-15, 4-16, 4-17 and 4-18. The histogram analysis and normal Q-Q plots indicated that the distribution of nitrate-nitrogen values do not observe a normal distribution.

PAGE 144

129 Figure 4-14. Normal Q-Q plot of the nitrate-nitrogen values in layer 1. Figure 4-15. Normal Q-Q plot of the nitrate-nitrogen values in layer 2.

PAGE 145

130 Figure 4-16. Normal Q-Q plot of the nitrate-nitrogen values in layer 3. Figure 4-17. Normal Q-Q plot of the nitrate-nitrogen values in layer 4.

PAGE 146

131 Figure 4-18. Normal Q-Q plot of the average nitrate-nitrogen value at each site. 4.6.1.3 Voronoi polygons The Voronoi maps are constructed such that a set of polygons are formed around the sample point and each location within the polygon is closest to that sample point rather than to any other sample point in the study area. Each polygon is associated with neighborhood polygons. The neighborhood polygons are those that share the same border. Once these neighborhood polygons are defined, descriptive statistics can be calculated for each of the polygon based on the neighborhood values. Parameters such as mean, median, mode, standard deviation, entropy, cluster, Inter Quartile Range (IQR) and local influence can be calculated. Figure 4-19 shows the mean values calculated using Voronoi polygons for the nitrate-nitrogen value in layer 1. The mean values of each of these polygons are calculated based on the mean value of the neighboring polygons.

PAGE 147

132 Figure 4-19. Mean Voronoi polygons for nitrate-nitrogen values in layer 1. The cluster option is quite an important Voronoi statistic which can be used to find outliers in the dataset. All the cells in the study area are divided into five classes. If a cell class is quite different from the neighboring classes, it is given a gray color. Figure 4-20, 4-21, 4-22, 4-23 and 4-24 are the Cluster Voronoi polygons for layer 1, 2, 3, 4 and the average nitrate-nitrogen value. These gray polygons indicate nitrate-nitrogen values that are very different from the neighboring points. It’s not a necessity that all these gray polygons indicate an outlier; it can also indicate an erratic value.

PAGE 148

133 Figure 4-20. Cluster voronoi polygons for nitrate-nitrogen values in layer 1. Figure 4-21. Cluster voronoi polygons for nitrate-nitrogen values in layer 2.

PAGE 149

134 Figure 4-22. Cluster voronoi polygons for nitrate-nitrogen values in layer 3. Figure 4-23. Cluster voronoi polygons for nitrate-nitrogen values in layer 4.

PAGE 150

135 Figure 4-24. Cluster voronoi polygons for the average nitrate-nitrogen value. 4.6.2 Interpolation Based Prediction of Nitrate-Nitrogen Values in SFRW Interpolation is a widely used technique to generate continuous data from discrete observations. Thus, interpolation can be defined as a technique used in predicting the attribute values at unsampled locations from measurements made at locations within the same region. The necessity to adopt interpolation techniques arise from a number of applications where continuous data are required, but it is impossible to obtain such data directly. Interpolation can be carried out using various measures. Interpolation techniques can be grouped into two classes: Deterministic interpolators and Geostatistical interpolators. Deterministic interpolators create prediction surfaces based on the extent of similarity or the degree of smoothing. Geostatistical interpolators create continuous surfaces by quantifying the spatial autocorrelation in the dataset and also by accounting for the spatial configuration of the sample points around the prediction location (Johnston et al., 2001). Inverse Distance Weighted (IDW) and Radial Basis Function (RBF) are

PAGE 151

136 examples of deterministic interpolators. Kriging is the most commonly used geostatistical interpolator. However, the rationale behind all these interpolation techniques is based on the observation that points closer in space are more likely to be similar than points that are further apart (Burrough and McDonnell, 1998). The interpolation techniques are discussed in the following sections. Based on the method of operation, deterministic techniques are divided into two classes: global and local interpolators. Global interpolators use all the available input data to generate the final continuous surface. One major disadvantage of this method is that the short-range or local variations in the dataset are not significantly represented. Hence, these interpolators are mostly used for studying the dataset for any apparent global variations or trends across the study area rather than for direct interpolation. Local interpolators, on the other-hand, operate within small neighborhoods around the point to ensure that the interpolations are made from data points that are in the immediate neighborhood. This is to make sure that the short-range or local variations, that are inherent in any surface, are not dismissed as random noise. The local interpolators compute some average value within each neighborhood or window in which it operates and hence, the prediction surfaces created using these interpolators are usually smooth (Burrough and McDonnell, 1998). Inverse Distance Weighted and Radial Basis Functions are two local deterministic interpolation techniques offered by the ArcGIS’s Geostatistical Analyst. These two methods are discussed below.

PAGE 152

137 4.6.2.1 Inverse Distance Weighted (IDW) IDW is a deterministic interpolation technique in which the attributes of unsampled areas are estimated based on the extend of similarity between the nearest measured point. IDW is an exact interpolator, which means that the technique predicts values identical to the measured values. As mentioned, the basic assumption underlying IDW interpolators is that each measured point has an influence on an unmeasured location that will reduce with the distance from the point to be interpolated. Hence, the measured points close to the prediction location are weighed more than points that are farther away. The general formula for interpolation using IDW is given below. )(*)(SSioZiZ where i = 1 to N. Equation 9 )(oSZ is the predicted value of the attribute for the location S o i are the weights assigned N is the number of sample locations surrounding the prediction location Z(S i ) is the location of the observed attribute. The weights that are assigned to each measured location decrease with its distance from the prediction location. The equation for determining weights in IDW is give in Equation 10. i = d i0 –p/ d i0 –p where i = 1 to N Equation 10 where d i0 is the distance between the prediction location (S o ) and the measured location (S i )

PAGE 153

138 p is the assigned power. Thus, the power function and the search neighborhood can be considered as two important factors that govern the result of this interpolation process. The decay of the influence is denoted by terms of the power function. It can be inferred from the equation that as the d i0 value increase, the value of the power function decrease exponentially. In an interpolation technique, if a higher power value is used, only the points immediately around the prediction location are used and hence the interpolation will not include all the measurements that have an influence on the prediction location. On the other hand, if a low power value is specified, the interpolation includes points that do not have an influence on the prediction location. Hence, an optimal value has to be chosen so that a reliable prediction surface is created with low Root Mean Square Prediction Error (RMSPE). The Geostatistical Analyst has an option in which we can use a power value that has the minimum RMSPE. The “Optimize power value” function determines the optimal power value for the dataset and uses this value for the interpolation. This is shown in Figure 4-25. RMSPE Power Optimal value Figure 4-25. Optimizing the power value (Source: Adapted from Johnston et al., 2001).

PAGE 154

139 Setting an optimal search neighborhood is important in local deterministic interpolation techniques as they restrict the number of points considered to interpolate a prediction location. This is to ensure that points that are further away are not considered in the analysis and to speed up computation. Specifying the neighborhood shape enables dividing the neighborhood into specific sectors. The neighborhood size parameters apply to each sector in this case. In Figure 4-26, the five measured points are used to predict the attribute at the prediction location using the circular shaped (one sector) interpolator. In addition to one sector interpolator, there are other shapes as the eight sector interpolators, ellipse with four sectors and others. The eight sector interpolator is shown in Figure 4-27. Figure 4-26. Search neighborhood (Source: Adapted from Johnston et al., 2001). 4.6.2.2 Radial Basis Functions (RBF) RBF are exact interpolators. They are piece-wise functions and they are fitted to data points exactly, while at the same time ensuring that the joins between one part of the curve and other is continuous and not abrupt (Burrough and McDonnell, 1998). RBF can be conceptually related to fitting a rubber membrane through the measured points and as they are piece wise function, one part of the curve can be modified without affecting the

PAGE 155

140 other parts. In Figure 4-28, the piece wise curves fitted for each sample location can be seen. The smooth fit of these curves to ensure continuous curvature is also shown in the cross-section diagram in Figure 4-28. Figure 4-27. Eight sector neighborhood interpolator. Based on the type of fit of the curves, the RBF can be categorized into different functions. ArcGIS’s Geostatistical Analyst supports five different basis functions. They are Completely Regularized Spline (CRS), spline with tension, thin plate spline, multi-quadratic spline and inverse multi-quadratic spline. 4.6.2.3 Cross validation The validity of the interpolation techniques is assessed by cross validation tests. This is carried out by first removing the value of a known point and proceeding with the interpolation. After the results are obtained, the interpolated value of the left out point and its actual value are compared. This is carried out for all of the sample points. So it is

PAGE 156

141 essentially a ‘drop one and interpolate’ technique. The minimum estimation variance is termed as the prediction error of the interpolation process. Figure 4-28. Fitting of piece wise curves in RBF (Source: Adapted from Johnston et al., 2001). 4.6.2.4 Prediction surfaces of nitrate-nitrogen values in SFRW The histogram distribution of nitrate-nitrogen values in layer 1, 2, 3, 4 and the profile average are shown in Figure 4-9, 4-10, 4-11. 4-12 and in 4-13. The normal Q-Q plots for these datasets are shown in Figure 4-14, 4-15, 4-16, 4-17 and 4-18. As can be seen from these figures, none of the datasets exhibit a normal distribution. Nearly half the samples that were analyzed had nitrate-nitrogen values below the detectable limit of 0.02 PPM and were thus assigned zero values. The histograms are positively skewed in all the cases. Most of the geostatistical interpolation techniques are based on normality

PAGE 157

142 assumptions for either the observed data or for the transformed data. Usually when a dataset is constraint to positive values and shows skewness in the direction, a logarithmic transformation is used to normalize the distribution. However, most of the nitrate-nitrogen values measured had zero values and performing a logarithmic transformation for such datasets to approximate a Gaussian distribution is not possible. Hence, the stationarity assumptions would be violated. And also, in the case of non-point source pollution contamination of soil, the concept of intrinsic stationarity is difficult to assume. For these reasons, the validity of using stochastic interpolation techniques is highly doubtful. Hence, local deterministic interpolation methods were adopted to create prediction surfaces for nitrate-nitrogen concentrations across the SFRW. 4.6.2.4.1 Interpolation of nitrate-nitrogen concentration in Layer 1 The nitrate-nitrogen concentrations in layer 1 ranged from 0 to 7.6 g/g soil. A number of local deterministic interpolation techniques were tested. The radial basis function of ‘Spline with Tension’ resulted in the lowest prediction error. Table 4-8 is a summary of the resulting prediction errors using different interpolation techniques. Figure 4-29 shows the interpolated surface for nitrate-nitrogen concentrations in layer 1 based on spline with tension. 4.6.2.4.2 Interpolation of nitrate-nitrogen concentrations in Layer 2 The nitrate-nitrogen concentrations in layer 2 ranged from 0 to 9.39 g/g soil. A number of local deterministic interpolation techniques were tested. The radial basis function of ‘Spline with Tension’ resulted in the lowest prediction error. Table 4-9 is a summary of the resulting prediction errors using different interpolation techniques. Figure 4-30 shows the interpolated surface for nitrate-nitrogen concentrations in layer 2.

PAGE 158

143 4.6.2.4.3 Interpolation of nitrate-nitrogen concentrations in Layer 3 The nitrate-nitrogen concentrations in layer 3 range from 0 to 9.8 g/g soil. A number of local deterministic interpolation techniques were tested. IDW resulted in the lowest prediction error. Table 4-10 is a summary of the resulting prediction errors using different interpolation techniques. Figure 4-31 shows the interpolated surface for nitrate-nitrogen concentrations in layer 3. 4.6.2.4.4 Interpolation of nitrate-nitrogen concentrations in Layer 4 The nitrate-nitrogen concentrations in layer 4 ranged from 0 to 8.78 g/g soil. A number of local deterministic interpolation techniques were tested. IDW resulted in the lowest prediction error. Table 4-11 is a summary of the resulted prediction errors using different interpolation techniques. Figure 4-32 shows the interpolated surface for nitrate-nitrogen concentrations in layer 4. 4.6.2.4.5 Interpolation of the average nitrate-nitrogen values in each profile The average value of nitrate-nitrogen concentrations in all the layers ranged from 0 to 6.54 g/g soil. A number of local deterministic interpolation techniques were tested. ‘Spline with Tension’ interpolation technique resulted in the lowest prediction error. Table 4-12 is a summary of the resulting prediction errors using different interpolation techniques. Figure 4-33 shows the interpolated surface for nitrate-nitrogen concentrations in layer 4.

PAGE 159

144 Table 4-8. Summary of the resulting prediction errors using different interpolation techniques for layer 1 nitrate-nitrogen concentrations. Interpolation Technique Prediction Error IDW with power 2 and with a maximum of 15 neighborhood points and minimum of 10 points 1.83 IDW with optimized power 1 and with a maximum of 15 neighborhood points and minimum of 10 points 1.61 IDW with optimized power 1 and with a maximum of 15 neighborhood points and minimum of 0 points CRS with a maximum of 15 neighborhood points and minimum of 10 points and circle (sector)neighborhood interpolator 1.47 CRS with a maximum of 15 neighborhood points and minimum of 10 points and 8 sector neighborhood interpolator 1.46 Spline with tension with a maximum of 15 neighborhood points and minimum of 10 points and circle (sector)neighborhood interpolator 1.46 Spline with tension with a maximum of 15 neighborhood points and minimum of 10 points and 8 sector neighborhood interpolator 1.45 Multi-quadratic spline with a maximum of 15 neighborhood points and minimum of 10 points and circle (sector)neighborhood interpolator 1.95 Multi-quadratic spline with a maximum of 15 neighborhood points and minimum of 10 points 8 sector neighborhood interpolator 1.46 Thin plate spline with a maximum of 15 neighborhood points and minimum of 10 points and circle (sector)neighborhood interpolator 10.04 1.57

PAGE 160

145 Figure 4-29. Interpolated surface of nitrate-nitrogen concentrations across the SFRW in layer 1.

PAGE 161

146 Table 4-9. Summary of the resulting prediction errors using different interpolation techniques for layer 2 nitrate-nitrogen concentrations. Interpolation Technique Prediction Error IDW with power 2 and with a maximum of 15 neighborhood points and minimum of 10 points and circle (sector)neighborhood interpolator 1.69 IDW with optimized power 1 and with a maximum of 15 neighborhood points and minimum of 10 points and circle (sector)neighborhood interpolator 1.50 IDW with optimized power 1 and with a maximum of 8 neighborhood points and minimum of 8 points and 8 sector neighborhood interpolator 1.45 CRS with a maximum of 15 neighborhood points and minimum of 10 points and circle (sector)neighborhood interpolator 1.42 CRS with a maximum of 8 neighborhood points and minimum of 2 points and 8 sector neighborhood interpolator 1.38 Spline with tension with a maximum of 15 neighborhood points and minimum of 10 points and circle (sector)neighborhood interpolator 1.38 Spline with tension with a maximum of 15 neighborhood points and minimum of 10 points and 8 sector neighborhood interpolator 1.37 Spline with tension with a maximum of 15 neighborhood points and minimum of 10 points and 4 sector neighborhood interpolator 1.36 Multi-quadratic spline with a maximum of 15 neighborhood points and minimum of 10 points and circle (sector)neighborhood interpolator 1.78 Multi-quadratic spline with a maximum of 8 neighborhood points and minimum of 8 points and 8 sector neighborhood interpolator 1.78 Thin plate spline with a maximum of 15 neighborhood points and minimum of 10 points and circle (sector)neighborhood interpolator 13.2

PAGE 162

147 Figure 4-30. Interpolated surface of nitrate-nitrogen concentrations across the SFRW in layer 2.

PAGE 163

148 Table 4-10. Summary of the resulted prediction errors using different interpolation techniques for layer 3 nitrate-nitrogen concentrations. Interpolation Technique Prediction Error IDW with power 2 and with a maximum of 15 neighborhood points and minimum of 10 points and circle (sector)neighborhood interpolator 2.18 IDW with optimized power 1 and with a maximum of 15 neighborhood points and minimum of 10 points and circle (sector)neighborhood interpolator 1.99 IDW with optimized power 1 and with a maximum of 15 neighborhood points and minimum of 10 points and 8 sector neighborhood interpolator 1.90 CRS with a maximum of 15 neighborhood points and minimum of 10 points and circle (sector)neighborhood interpolator 1.97 CRS with a maximum of 8 neighborhood points and minimum of 8 points and 8 sector neighborhood interpolator 1.93 Spline with tension with a maximum of 15 neighborhood points and minimum of 10 points and circle (sector)neighborhood interpolator 1.97 Spline with tension with a maximum of 15 neighborhood points and minimum of 10 points and 8 sector neighborhood interpolator 1.93 Spline with tension with a maximum of 15 neighborhood points and minimum of 10 points and 4 sector neighborhood interpolator 1.95 Multi-quadratic spline with a maximum of 15 neighborhood points and minimum of 10 points and circle (sector)neighborhood interpolator 2.39 Multi-quadratic spline with a maximum of 8 neighborhood points and minimum of 8 points and 8 sector neighborhood interpolator 2.38 Thin plate spline with a maximum of 15 neighborhood points and minimum of 10 points and circle (sector)neighborhood interpolator 4.10

PAGE 164

149 Figure 4-31. Interpolated surface of nitrate-nitrogen concentrations across the SFRW in layer 3.

PAGE 165

150 Table 4-11. Summary of the resulting prediction errors using different interpolation techniques for layer 4 nitrate-nitrogen concentrations. Interpolation Technique Prediction Error IDW with power 2 and with a maximum of 15 neighborhood points and minimum of 10 points and circle (sector)neighborhood interpolator 1.64 IDW with optimized power 1 and with a maximum of 15 neighborhood points and minimum of 10 points and circle (sector)neighborhood interpolator 1.47 IDW with optimized power 1 and with a maximum of 15 neighborhood points and minimum of 10 points and 8 sector neighborhood interpolator 1.47 IDW with optimized power 1 and with a maximum of 8 neighborhood points and minimum of 0 points and 8 sector neighborhood interpolator 1.46 CRS with a maximum of 15 neighborhood points and minimum of 10 points and circle (sector)neighborhood interpolator 1.49 CRS with a maximum of 8 neighborhood points and minimum of 8 points and 8 sector neighborhood interpolator 1.48 Spline with tension with a maximum of 15 neighborhood points and minimum of 10 points and circle (sector)neighborhood interpolator 1.48 Spline with tension with a maximum of 15 neighborhood points and minimum of 10 points and 8 sector neighborhood interpolator 1.48 Spline with tension with a maximum of 15 neighborhood points and minimum of 0 points and 8 sector neighborhood interpolator 1.48 Spline with tension with a maximum of 10 neighborhood points and minimum of 0 points and 8 sector neighborhood interpolator 1.48 Multi-quadratic spline with a maximum of 15 neighborhood points and minimum of 10 points and circle (sector)neighborhood interpolator 1.75 Multi-quadratic spline with a maximum of 8 neighborhood points and minimum of 8 points and 8 sector neighborhood interpolator 1.75 Thin plate spline with a maximum of 15 neighborhood points and minimum of 10 points and circle (sector)neighborhood interpolator 1.75

PAGE 166

151 Figure 4-32. Interpolated surface of nitrate-nitrogen concentrations across the SFRW in layer 4.

PAGE 167

152 Table 4-12. Summary of the resulting prediction errors using different interpolation techniques for average nitrate-nitrogen concentrations. Interpolation Technique Prediction Error IDW with power 2 and with a maximum of 15 neighborhood points and minimum of 10 points and circle (sector)neighborhood interpolator 1.54 IDW with optimized power 1 and with a maximum of 15 neighborhood points and minimum of 10 points and circle (sector)neighborhood interpolator 1.40 IDW with optimized power 1 and with a maximum of 15 neighborhood points and minimum of 10 points and 8 sector neighborhood interpolator 1.37 IDW with optimized power 1 and with a maximum of 8 neighborhood points and minimum of 0 points and 8 sector neighborhood interpolator 1.37 CRS with a maximum of 15 neighborhood points and minimum of 10 points and circle (sector)neighborhood interpolator 1.31 Spline with tension with a maximum of 15 neighborhood points and minimum of 10 points and circle (sector)neighborhood interpolator 1.31 Spline with tension with a maximum of 8 neighborhood points and minimum of 0 points and 4 sector neighborhood interpolator 1.30 Spline with tension with a maximum of 8 neighborhood points and minimum of 0 points and 8 sector neighborhood interpolator 1.31 Multi-quadratic spline with a maximum of 15 neighborhood points and minimum of 10 points and circle (sector)neighborhood interpolator 1.63 Multi-quadratic spline with a maximum of 8 neighborhood points and minimum of 8 points and 4 sector neighborhood interpolator 1.63 Thin plate spline with a maximum of 15 neighborhood points and minimum of 10 points and circle (sector)neighborhood interpolator 10.04

PAGE 168

153 Figure 4-33. Interpolated surface of the profile averages of nitrate-nitrogen concentrations across the SFRW. 4.6.3 Pixel Based Prediction of Nitrate-Nitrogen Values The objective of this analysis was to transfer the observed nitrate-nitrogen values at a pixel to unmeasured locations based on LULC-soil combinations. This is illustrated in Figure 4-34. Figure 4-34. Illustration of the pixel based prediction technique (OPixel soil-N represents the observed soil nitrate-nitrogen value and PPixel soil-N represents the predicted soil nitrate-nitrogen value)

PAGE 169

154 Table 4-13. Average nitrate-nitrogen values for each LULC-soil combination for all depths (g/g soil). Entisols Ultisols Spodosols Histosols Pine plantations 0.05 1 (0.04) 2 353 0.33 (0.00) 20 0.26 (0.08) 39 Upland forest 0.21 (0.06) 28 0.36 (0.14) 20 2.40 (1.10) 6 Wetland 0.33 (0.13) 14 0.15 (0.14) 13 0.00 (0.00) 4 0.32 (0.11) 19 Improved pasture 0.51 (0.42) 11 1.80 (0.39) 22 1.65 (0.40) 9 Rangeland 1.52 (0.91) 12 2.04 (1.19) 6 Urban 0.17 (0.01) 8 0.20 (0.19) 15 0.06 (0.18) 8 Forest regeneration 1.01 (0.59) 4 0.00 (0.00) 2 0.29 (0.43) 6 Crops 2.70 (0.77) 7 2.60 (1.29) 8 0.56 (0.06) 4 Tree groves 0.09 (0.28) 4 6.06 (4.07) 4 Feedlot 2.75 (2.72) 4 1 : Mean nitrate-nitrogen value per LULC-soil combination class. 2 : Standard deviation (S.D.) associated with the mean value. 3 : Number of samples in the LULC-soil combination class. 0 values: Represent measurements below detection limit Empty fields: Represent no observations As mentioned, 101 sites were sampled in layer 1 and 2; 89 and 59 sites were sampled in layer 3 and 4. Average values of nitrate-nitrogen concentrations for each LULC-soil combination and at each depth were determined. The final nitrate-nitrogen values were created by averaging the nitrate-nitrogen values measured for each LULC

PAGE 170

155 soil combination across all the depths (OPixel soil-N ). These values are given in Table 4-13. The average value for each of the LULC-soil combination was assigned on a pixel-by-pixel basis (PPixel soil-N ) to the corresponding LULC-soil raster layers. The output of this procedure is shown in Figure 4-35. Figure 4-35. Pixel based prediction of nitrate-nitrogen values in SFRW. White pixels: Nitrate-nitrogen could not be predicted because no observations were made in the LULC-soil combinations. A depth-weighted average nitrate-nitrogen concentration was determined for the LULC class. The results of this analysis are shown in Table 4-14.

PAGE 171

156 Table 4-14 Average nitrate-nitrogen value for the LULC classes (g/g soil). LULC class N-No 3 (g/g soil). Pine plantations 0.21 1 (0.03) 2 943 Upland forest 0.78 (0.38) 54 Wetland 0.24 (0.06) 50 Improved pasture 1.17 (0.38) 54 Rangeland 1.58 (0.45) 22 Urban 0.14 (0.11) 31 Forest regeneration 0.83 (0.70) 12 Crops 1.95 (0.55) 19 Tree groves 3.07 (1.95) 8 Feedlot 2.75 (2.72) 4 1 Mean nitrate-nitrogen value per LULC class. 2 Standard deviation (S.D.) associated with the mean value. 3 Number of samples in the LULC class. 4.7 Discussions and Conclusions The extent of nitrate-nitrogen leaching to ground water systems is primarily related to the soil nitrate-nitrogen concentrations and geology. The soil nitrate-nitrogen levels are dependent on LU activities, LC and soil type. A quantitative analysis of the impact of

PAGE 172

157 both LULC and soil types on soil nitrate-nitrogen levels is hypothesized to improve our understanding on the spatial variation of soil nitrate-nitrogen levels across the SFRW. There is a high variability in the nitrate-nitrogen values measured across the watershed. Based on Table 4-14, it can be seen that the LULC class of tree groves is associated with the highest soil nitrate-nitrogen content of 3.07 g/g (S.D. 1.95 g/g). Improved pasture had an average soil nitrate-nitrogen concentration of 1.17 g/g (S.D. 0.38) and crops had an average value of 1.95 g/g (S.D. 0.55 g/g). The observed soil nitrate-nitrogen values were relatively low. Studies by Woodard et al. (2002) have documented nitrate-nitrogen measurements in the range of 157 g/g to 160 g/g for row crops in north-central Florida. Also, soil nitrate-nitrogen studies on pastures documented values in the range of 102 g/g to 154 g/g. Feedlots and rangeland measured average values of 2.75 g/g (S.D. 2.72 g/g) and 1.58 g/g (S.D. 0.45 g/g), respectively. Relatively low soil nitrate-nitrogen measurements were observed in pine plantations (0.21 g/g, S.D. 0.03 g/g), upland forest (0.78 g/g, S.D. 0.38 g/g), wetland (0.24 g/g, S.D. 0.06 g/g) and in urban (0.14 g/g, S.D. 0.11 g/g) classes. The results of the September 2003 sampling event reveled relatively low soil nitrate-nitrogen concentrations. LULC-soil combination layers were used to develop the site sampling protocol. Analyzing the results in Table 4-13 provides insight into specific LULC-soil combinations that are more sensitive to nitrogen inputs than the others. For example, high soil nitrate-nitrogen values were associated with crop-Entisols and crop-Ultisols combination categories rather than with the crop-Spodosols category. Relatively high values were associated with the tree grove-Spodosols combination class. Other LULC

PAGE 173

158 soil combinations categories of upland forest-Spodosols, improved pastures-Ultisols-Spodosols, rangeland-Ultisols-Spodosols, and feedlot-Ultisols exhibited higher soil nitrate-nitrogen content than other combination categories. Overall, measured soil nitrate-nitrogen values were very low across all LULC-soil categories which might be due to the season when observations were made. Typically, in September harvest of crops is close to completed. Higher soil nitrate-nitrogen values are expected in spring/early summer shortly after fertilizer applications. The interpolated surfaces showed fairly low concentration of soil nitrate-nitrogen in the south-eastern and south-western part of the watershed. High values were concentrated in regions north of Alachua. This region of the watershed is dominated by sandy soils. Hence, this region has a high risk of nitrate-nitrogen in the soil to leach to the ground water systems. Two different prediction methods were compared to best represent the nitrate-nitrogen concentrations across the SFRW. Using the deterministic interpolation techniques of IDW and RBF, prediction surfaces were generated to visualize the nitrate-nitrogen concentrations based on the basis of similarity of observations and smoothing within a local neighborhood. The local deterministic interpolation techniques are based on the assumption that points closer in space are more likely to be similar than points that are further apart. On the other hand, the pixel-based prediction methods assigned the same mean value observed within a specific LULC-soil category to all pixels that were not sampled in the same category. A shortcoming of the pixel-based prediction methods is that it does not consider spatial autocorrelation. One of the disadvantages of IDW interpolation methods are that they do not account for variations in LULC within the search neighborhood. For instance, consider

PAGE 174

159 that nitrate-nitrogen concentrations are measured in the five locations denoted in pink color in Figure 4-26 and they correspond to the land cover class of pine plantations. The nitrate-nitrogen value is to be predicted for the prediction location with agricultural land cover class denoted in yellow color. Using deterministic interpolation techniques in this scenario would not yield an optimal prediction attribute at the location. LULC can change within short distances and the concentration of soil nitrate-nitrogen is highly depended on the LULC–soils class. This is one of the disadvantages of interpolating nitrate-nitrogen concentrations across the watershed. Assigning nitrate-nitrogen values specific to each LULC-soil combination on a pixel-by-pixel basis provides a solution to the above discussed problem. However, the spatial auto-correlation values across the watershed are not considered. There are inherent advantages and disadvantages in using both methods. One advantage of the interpolated surfaces is that cross-validation provides measures of prediction expressing the uncertainty of predictions. The site selection protocol was helpful in targeting sites to characterize the spatial variability of soil nitrate-nitrogen across the watershed. The mapped soil nitrate-nitrogen concentrations helped better understand the impact of different LULC-soil combinations in the SFRW. The results of the September 2003 sampling event reveled relatively low soil nitrate-nitrogen concentrations. This analysis is the first attempt in quantifying the spatio-temporal variation of soil nitrate-nitrogen concentrations across the watershed. Five more soil sampling events will be conducted to address seasonal changes of soil nitrate-nitrogen across the SFRW. The results of this analysis will help better target LULC-soil categories that have a significant impact on soil nitrate-nitrogen content in the

PAGE 175

160 SFRW. Soil sampling events will be conducted at varying time periods to quantify the spatial and temporal dynamics of soil nitrate-nitrogen within the SFRW.

PAGE 176

CHAPTER 5 ANALYSIS OF SURFACE WATER QUALITY 5.1 Introduction Non-point source pollutants are the major source of surface and ground water pollution in the U.S today. These pollutant inputs have increased rapidly in the recent years, and this has resulted in the degradation of water quality in many rivers, lakes, coastal and ground water systems. More than 60% of the impaired rivers in U.S are polluted due to excessive inputs of nitrogen and phosphorus (Carpenter et al., 1998). The Safe Drinking Water Act (SDWA) was initiated to protect and maintain the nation’s drinking water supply. This law enforces actions and many regulations to protect drinking water and its source – rivers, lakes, reservoirs, springs and ground water wells. The SDWA has authorized the USEPA to establish national standards of both natural and man-induced contaminant levels in drinking water. These enforceable values of allowable concentration values are referred to as the MCL. According to USEPA, the maximum allowable value of nitrate-nitrogen level in drinking water supply is 10 mg/l. The USEPA also assists with state agencies and public water suppliers to set up multiple barriers to prevent water pollution. The barriers include source water protection, treatment, distribution system integrity and public information (USEPA, 1999). Ground water and surface water quality are monitored in the SRWMD by a network of monitoring wells distributed across its area. During February 1995 (high flow conditions) concentrations of nitrate plus nitrite ranged from 0.05 to 0.38 mg/l. In June 1995 (low flow conditions), they ranged form 0.07 to 1.05 mg/l. Highest concentration 161

PAGE 177

162 was observed in Branford (Raulston et al., 1998) which lies within the SFRW. In water year 1998 the Santa Fe Reach 2 showed an annual nitrate-nitrogen load of 1,130 tons/year, accounting for a total of 15.9% of the nitrate-nitrogen load into the Gulf of Mexico and the Santa Fe Reach 1 showed an annual nitrate-nitrogen load of 65.6 tons/year, accounting for an additional total of 0.9% of the nitrate-nitrogen load the Suwannee delivers into the Gulf of Mexico (SRWMD, 1998). Of the 2,676 tons of nitrate-nitrogen that were transported to the Gulf of Mexico through tributaries of the Suwannee River basin, the Santa Fe accounted for 593 tons of nitrate-nitrogen in water year 2000 (SRWMD, 2000). In water year 2001, a total of 3,067 tons of nitrate-nitrogen were transported to the Gulf of Mexico from the Aucilla, Econfina, Fenholloway, Steinhatchee, Suwannee, and Waccasassa Rivers. Of the 3,067 tons of nitrate-nitrogen, the Suwannee River Basin accounted for 2,999 tons of nitrate-nitrogen. The contributions of the Santa Fe reach 2 to the above were 15.8%. Based on the Surface water quality report by the SRWMD for 2003, surface water measurements were recorded at 67 stations in the basin. In water year 2002, around 3,012 tons of nitrate-nitrogen was transported to the Gulf of Mexico from the Aucilla, Econfina, Fenholloway, Steinhatchee, Suwannee, and Waccasassa Rivers. Among these river basins, the Suwannee river basin accounted for about 2,971 tons of nitrate-nitrogen. Within the Suwannee River basin, the Santa Fe River Reach 2 accounted for 19.6% of the annual nitrate-nitrogen load delivered to the Gulf by the Suwannee River, but covers only 5.7% of the total basin area. This implies a 25 % increase in nitrate-nitrogen transport from the Santa Fe Reach 2 to the Gulf of Mexico from 2001-2003. Over two decades of monitoring the water quality of major rivers of the Suwannee River Basin has indicated a

PAGE 178

163 statistically significant (at 95% confidence interval) increasing trend in the concentrations of nitrate-nitrogen (Ham and Hatzell, 1996; SRWMD, 1998). The nutrient loadings by reaches/basins in the Suwannee River Basin for the water year 2002 are shown in Figure 5-1. Figure 5-1. Nutrient loadings by reaches/basins in the Suwannee River Basin for the water year 2002 (SRWMD, 2002). Ground water is monitored by a network of 97 monitor wells distributed over the SRWMD. The Suwannee and Lafayette counties have consistently exhibited elevated nitrate-nitrogen concentrations. The area around the Santa Fe reach 2 is increasingly impacted by nutrients under low flow conditions when the river receives a large quantity of groundwater flow through springs and seeps in the riverbed. Also, some site specific studies, conducted by the Department of Health, have identified wells with concentrations of nitrate-nitrogen above the MCL (SRWMD, 2003). Along the Suwannee

PAGE 179

164 and other river corridors, the high degree of interaction between surface water and ground water poses risks to the aquifer that is used for drinking water extraction. During the high river stages and floods, springs and seeps reverse flow and river water enters the aquifer (Raulston et al., 1998). This occurs in Branford within the SFRW. As mentioned, elevated nitrate-nitrogen concentrations are prominent in this region. Elevated nitratenitrogen has been recorded in the Middle Suwannee River Watershed. In this area, the Floridian aquifer is unconfined, allowing water soluble containments to leach into the aquifer. Springs in the Middle Suwannee River Watershed showed nitrate-nitrogen concentrations ranging from 1.2 to 19.2 mg/l. Groundwater form the watershed flows towards the Suwannee River and is affecting surface water via springs and seeps in the riverbed (Raulston et al., 1998). The mean nitrate-nitrogen concentrations in the ground water systems in the SRWMD are shown in Figure 5-2. The Water Quality Assurance Act and the TMDL programs were initiated to detect and predict the contamination of the nation’s ground water resources, to manage non-point source pollutants and determine the effectiveness of using non-point source controls. Pollution prevention requires a clear understanding of the impacts of land use and water quality at a watershed level. In recent years, many efforts have been made in promoting the watershed protection approach (USEPA, 1993). Although studies of water flow have been conducted at a watershed level for a long time (Gaebrecht, 1991), most water quality studies focus on areas in the vicinity of pollution sources. A limited number of studies examine the relationship between spatial distributions of land use and water quality in a watershed-level. Since most planning agencies and local authorities do not have resources to extensively collect land use and

PAGE 180

165 water quality data in developing plans, a water quality component is missing (Wang and Yin, 1997). Figure 5-2. Mean nitrate-nitrogen concentration of the Floridian aquifer for the water year 2002 (Source: SRWMD, 2002). The use of geographical information systems to analyze digital data in pollution modeling and evaluation has increased recently (Gallimore and Xiang, 1991; Morse et al., 1994). Studies have demonstrated that digital data can be used in watershed and environmental studies (Hamlet et al., 1992; White et al., 1992). DEM data were used to delineate waterways, watershed boundaries and catchments for each specific water quality monitoring station (Moore et al., 1991; Vieux, 1991). Basin characteristics such as LULC, slope and soil attributes affect water quality by regulating sediment and chemical concentration. Among these characteristics, LULC can be manipulated to improve water quality. These LULC types can serve as nutrient detention media or as nutrient transformers as dissolved or suspended nutrients move towards the stream

PAGE 181

166 (Basnyat et al., 2000). From a land cover perspective, agricultural activities have been identified as major sources of non-point source pollutant (Viessman and Hammer, 1993). Mattikalli and Richards (1996) determined the relationship between land use and water quality by employing an export-coefficient model for the River Glen watershed, UK. The results showed a high correlation between high nitrate concentrations and agricultural land uses. Similar studies conducted by Kauppi (1984), also indicated a strong correlation between nitrate concentrations and agricultural extensification in the study area. 5.2 Objectives The specific objectives of the study were: To delineate sub-watersheds based on surface water monitoring stations. To characterize the geographic position and distribution of land resources and LC within each sub-watershed. To understand the spatial relationships between sub-watershed characteristics and surface water quality data. 5.3 Materials 5.3.1 Software The following softwares were used for the analysis: ArcGIS Desktop 8.3, developed by Environmental System Research Institute (ESRI), Redlands, CA. Arc View 3.2 – developed by ESRI, CA. Soil and Water Assessment Tool (SWAT) extension, developed by USDASoil and Water Research Laboratory, Texas. SPSS for Windows, developed by SPSS, Inc., Illinois. Microsoft Excel, developed by Microsoft Corporation, Seattle, WA.

PAGE 182

167 5.3.2 Water Quality Data Surface water quality reports were obtained for the time frame 1989 to 2003 from the SRWMD. 5.3.3 Attributes for the Sub-Watershed Characteristics Database Various auxiliary spatial data were used to develop the sub-watershed characteristics database. These attributes were expected to have a significant effect on nitrate-nitrogen loadings and were discussed in detail in Chapter 2. Land cover data for the year 1990, 2000, and 2003 developed at the GIS Research Laboratory, SWS department (chapter 3). Soil Orders, developed by USDA-NRCS. Environmental geology, developed by the FDEP DRASTIC index scores, developed by U.S.EPA and NWWA. Population density in the year 2000, developed by U.S. Census Bureau. Digital Elevation Model (DEM), developed by USGS-National Elevation Dataset (NED). Slope derived from DEM. Soil organic carbon, developed by USDA-NRCS. 5.4 Methods 5.4.1 Processing Surface Water Quality Data Surface water quality data were obtained from the SRWMD. These data were in the form of ASCII data files. Nitrate-nitrogen concentrations were measured at 16 stations across the SFRW. The data records were reformatted to categorize the measured nitrate-nitrogen concentrations based on the time period. Annual averages of these concentrations were calculated for each year. The station average values were computed from the yearly averages recorded at each station. The final ASCII file was converted to shapefiles for use in a GIS environment. The Albers Equal Area projection was the

PAGE 183

168 standard projection system adopted for this project. The projection parameters are given in Table 4-1. Descriptions of the station ID/names are given in Table 5-1. The location of the surface water quality monitoring stations is shown in Figure 5-3 Table 5-1 Description of station ID STATION ID DESCRIPTION ALT010C1 LAKE ALTHO AT WALDO ICH010C1 ICHETUCKNEE RIVER .2 MI NORTH OF BRIDGE LSF010C1 SANTA FE LAKE NEAR KEYSTONE HEIGHTS NEW007C1 NEW RIVER AT SR-125 NEW008C1 NEW RIVER AT SR-229 NEAR RAIFORD NEW009C1 NEW RIVER NEAR LAKE BUTLER AT SR-100 NEW010C1 NEW RIVER NEAR WORTHINGTON SPRINGS AT C-18 OLS010C1 OLUSTEE CREEK AT SR-18 SFR010C1 SANTA FE RIVER NEAR GRAHAM SFR020C1 SANTA FE RIVER NEAR BROOKER AT SR-231 SFR030C1 SANTA FE RIVER AT WORTHINGTON SPRING SFR040C1 SANTA FE RIVER AT OLENO ST PARK SFR050C1 SANTA FE RIVER AT US-441 BRIDGE SFR060C1 SANTA FE RIVER AT SR 47 NEAR FORT WHITE SFR070C1 SANTA FE RIVER NEAR HILDRETH AT US 129 SMR010C1 SAMPSON RIVER AT CR18 Surface water quality data at SFR020C1, SFR030C1, SFR040C1, SFR050C1, SFR060C1, SFR070C1 and ICH010C1 were measured for all years between 1989 and 2003. The temporal variation of nitrate-nitrogen concentrations at these stations are shown in Figure 5-4. The water quality was not measured on a continuous basis at other stations. The temporal variation of nitrate-nitrogen concentrations in these stations is shown in Figure 5-5.

PAGE 184

169 Figure 5-3. Location of surface water quality monitoring stations. The spatial variation of average nitrate-nitrogen concentrations measured at each station location across the SFRW can be seen in Figure 5-6. From Figure 5-4 and 5-6, it can be seen that high concentrations are observed in SFR060C1, SFR070C1 and ICH010C1 stations. 5.4.2 Delineating Upslope Drainage Area The recognition of the importance of non-point sources of pollution in the overall spectrum of pollutants has led to increased efforts over the last two decades to identify and quantify non-point source pollutant loads in the U.S. However, estimating pollution

PAGE 185

170 contributions from non-point sources is difficult because of the spatially distributed nature of these pollutants. Figure 5-4. Temporal variation of nitrate-nitrogen concentrations in the above listed stations. Figure 5-5. Temporal variation of nitrate-nitrogen concentrations in the above listed stations.

PAGE 186

171 Figure 5-6. Spatial variation of long-term average nitrate-nitrogen concentrations. Earlier studies by Newson (1992), Moore et al. (1991) and Vieux (1991) have indicated that watershed management is the intellectual basis for responding to these challenges. Watersheds can be described by the upslope area of water flowing towards a given outlet point or pour point. The outlet point is the point at which the water flows out of the area and is often the lowest point (elevation) in the boundary. Delineation of watershed boundaries involves segmenting the terrain to represent the contributing area for a particular outlet point. This is illustrated in Figure 5-7. The contributing area to the outlet point, shown here as a green point, is the upslope area that drains into this particular point. The contributing area is shown by the red polygon. Watershed

PAGE 187

172 boundaries are defined by the hydrology of the region, which in turn is defined by the relief or terrain characteristics. Figure 5-7. Illustration of the upslope drainage area delineation. The objective of this process was to delineate the contributing areas that flow towards the surface water monitoring stations, in an attempt to understand the spatial variability of nitrate-nitrogen concentrations as a function of the distribution of land resources. The steps involved in the watershed boundary delineation are discussed in the following sections. 5.4.2.1 Terrain analysis Terrain analysis includes the preprocessing of the DEM, the generation of flow direction and flow accumulation grids and the stream definition. For this analysis, a 7.5 minute USGS-DEM, with a spatial resolution of 30 m was used and SWAT’s Arc View interfaces were used. 5.4.2.1.1 Preprocessing of DEM Elevation data usually exists in the form of contour maps. DEM are created by interpolating the digital contour files. Hence, regular grids are created from irregularly

PAGE 188

173 spaced points (Clarke et al., 1982). As a result, errors in the form of spurious depressions may creep in during the DEM generation stage. Earlier studies by Band (1986), Matrz and DeJong (1989) have suggested correcting for these spurious depressions or sinks before further analysis. A sink fill operation was performed to correct for these inconsistencies. 5.4.2.1.2 Flow direction grid After the sinks were treated, the flow direction for each cell in the raster was determined. The computation of the flow direction is based on the eight–direction pour method. Accordingly, the water from a cell was allowed to flow in one neighboring cell, based on the direction of steepest descent. The eight-point pour model is illustrated in Figure 5-8. Directional descriptors were assigned to describe the flow of water, 1 for east, 2 for south-east, 4 for south and others. An illustration of the computation of the flow direction grid is shown in Figure 5-9. The grid values in 5-9(A) represent elevation values. Based on the observation that water flows from a point of higher elevation to a point of lower elevation, the directions of flow are calculated and shown in 5-9(B). Figure 5-8. Eight point pour model.

PAGE 189

174 (A) (B) Figure 5-9. Computation of flow direction grid. (A) DEM (B) assigning directional descriptors of the flow direction. 5.4.2.1.3 Flow accumulation grid The flow accumulation grid was computed from the flow direction grid. The flow accumulation in a cell was determined by accounting for the number of cells that drain into it. The flow accumulation grid for the SFRW is shown in Figure 5-10. 5.3.4.1.4 Stream definition The stream network is defined from the flow accumulation grid by the pixels that have a higher flow accumulation value than the user defined thresholds. The size of the delineated sub-watersheds depends on this specified threshold. The recommended threshold value for NED datasets are 5,000 pixels. For this analysis, the stream threshold of 6,500 was set to define the stream network. Figure 5-11 shows the result of this process. 5.4.2.1.5 Automatic delineation The surface water monitoring stations are marked as the outlet point or pour point for the sub-watershed delineation. For the delineation of the sub-watershed boundary, the outlet point or the location of the surface water monitoring station is selected and the cells contributing the flow to that point are marked as its sub-watershed (Jensen and

PAGE 190

175 Dominique, 1988). The delineated sub-watershed boundaries for the station points are shown in Figure 5-12. Figure 5-10. Flow accumulation grid for the SFRW. 5.4.3 Creating Sub-Watershed Characteristics Database The land resource attributes of land cover, soil orders, soil organic carbon, environmental geology, population, elevation, slope and DRASTIC index scores were extracted for each sub-watershed. All these attributes were converted to raster datasets for effective computation. The raster calculator operation in ArcGIS was used to extract sub-watershed level land resource information for each sub-watershed. These attributes were complied into a database.

PAGE 191

176 Figure 5-11. Stream network. 5.4.4 Analysis of Correlations A correlation analysis was performed between the average nitrate-nitrogen concentrations measured at the outlet points and the land resource attributes of the contributing region. The analysis was achieved using the ‘SPSS for Windows’ software. The results of the correlation analysis are shown in Table 5-2. 5.4.5 Comparison of Watershed Characteristics To get a better understanding of the spatial relationships between water quality data and the distribution of land resources, three sub-watersheds were chosen and studied individually. These sub-watersheds are shown in Figure 5-13.

PAGE 192

177 Figure 5-12. Delineated sub-watershed boundary. Table 5-2 Results of the correlation analysis (only significant correlations are shown).

PAGE 193

178 5.4.5.1 Sub-Watershed-1 (SW1) This sub-watershed is located in the north-western part of the SFRW. The outlet point is the Ichetucknee River monitoring station (ICH010C1). Relatively high nitrate-nitrogen concentrations were recorded at this station. The temporal variation of nitrate-nitrogen concentrations is shown in Figure 5-14. The land resource attributes of land cover, soil order and environmental geology were extracted for this sub-watershed. The relative percent distribution of different land cover categories in 1990, 2000 and 2003 are shown in Figure 5-15(A), (B) and (C), respectively, soil orders in Figure 5-15(D) and environmental geology in Figure 5-15(E). Figure 5-13. Location of the three sub-watersheds.

PAGE 194

179 Figure 5-14. Nitrate-nitrogen concentrations measured at ICH010C1. (A) Figure 5-15. Attributes of the sub-watershed draining into ICH010C1.(A) Percentage distribution of 1990 land cover classes (B) Percentage distribution of 2000 land cover classes (C) Percentage distribution of 2003 land cover classes (D) Percentage distribution of soil orders (E) Percentage distribution of geological classes.

PAGE 195

180 (B) (C)

PAGE 196

181 (D) (E)

PAGE 197

182 5.4.5.2 Sub-Watershed-2 (SW2) This sub-watershed is located in the north-eastern part of the SFRW. The outlet point is the New River at SR-125monitoring station (NEW007C1). Relatively low nitrate-nitrogen concentrations were recorded at this station. The temporal variation of nitrate-nitrogen concentrations is shown in Figure 5-16. The land resource attributes of land cover, soil order and environmental geology were extracted for this sub-watershed. The relative percent distribution of different land cover categories in 1990, 2000 and 2003 are shown in Figure 5-17(A), (B) and (C), respectively, soil orders in Figure 5-17(D) and environmental geology in Figure 5-17(E). Figure 5-16. Nitrate-nitrogen concentrations measured at NEW007C1.

PAGE 198

183 (A) (B) Figure 5-17. Attributes of the sub-watershed draining into NEW007C1.(A) Percentage distribution of 1990 land cover classes (B) Percentage distribution of 2000 land cover classes (C) Percentage distribution of 2003 land cover classes (D) Percentage distribution of soil orders (E) Percentage distribution of geological classes.

PAGE 199

184 (C) (D)

PAGE 200

185 (E) 5.4.5.3 Sub-Watershed-3 (SW3) This sub-watershed is located in the south-eastern part of the SFRW. The outlet point is the Santa Fe River near Brooker at the SR-231monitoring station (SFR02C1). Moderate nitrate-nitrogen concentrations were recorded at this station. The temporal variation of nitrate-nitrogen concentrations is shown in Figure 5-18. The land resource attributes of land cover, soil order and environmental geology were extracted for this sub-watershed. The relative percent distribution of different land cover categories in 1990, 2000 and 2003 are shown in Figure 5-19(A), (B) and (C), respectively, soil orders in Figure 5-19(D) and environmental geology in Figure 5-19(E).

PAGE 201

186 Figure 5-18. Nitrate-nitrogen concentrations measured at SFR02C1 (A) Figure 5-19 Attributes of the sub-watershed draining into SFR02C1.(A) Percentage distribution of 1990 land cover classes (B) Percentage distribution of 2000 land cover classes (C) Percentage distribution of 2003 land cover classes (D) Percentage distribution of soil orders (E) Percentage distribution of geological classes.

PAGE 202

187 (B) (C)

PAGE 203

188 (D) (E) 5.5 Discussion and Conclusions Over 40 years of monitoring surface water quality has showed increasing nitrate-nitrogen concentrations in the SRWMD. Studies on surface water quality data from the SRWMD had indicated that the MCL limit of 10 mg/l was exceeded in a few stations in the district. The surface water monitoring stations located along prominent springs in the

PAGE 204

189 SRWMD is shown in Figure 5-20. The nitrate-nitrogen concentrations recorded in these stations show high spatial variability. Highest nitrate-nitrogen concentrations were observed in the station ‘SUW718971’, located north of Branford (Figure 5-20). The water quality measurements indicated that the MCL limit for nitrate-nitrogen concentrations were violated in the last two years with values ranging from 10.8 to 14.0 mg/l in 2001, 15.4 to 17.2 mg/l in 2002 and 10.9 to 15.0 mg/l in 2003. High nitrate-nitrogen concentrations were also observed at Fannin springs in Levy county (Figure 5-20). The nitrate-nitrogen concentrations ranged from 4.4 to 4.8 mg/l in 2002 and 3.9 to 5.7 mg/l in 2003. Other significantly elevated nitrate-nitrogen concentrations were recorded at Ruth (2.5 to 5.1 mg/l), Suwannee Blue (2.2 to 3.7 mg/l), Lafayette Blue (2.2-2.8 mg/l), and Telford (1.5 to 2.5 mg/l). Relatively low nitrate-nitrogen concentrations were measured in the SFRW. The water quality measurements recorded at the Santa Fe River near Fort White (SFR060C1 in Figure 5-3) ranged between 0.5 to 0.9 mg/l. The Ichetucknee River monitoring station (ICH010C1 in Figure 5-3) recorded values in the range of 0.4 to 0.6 mg/l. The concentrations of nitrate-nitrogen recorded in surface water monitoring stations are inversely related to the discharge. For instance, high nitrate-nitrogen concentrations are observed during base-flow conditions and when the contributions from ground water are low. Similarly low concentrations are observed during high-flow conditions where the contribution from ground water is diluted from the discharge (Katz et al., 1999) Based on reports by SRWMD and USGS, the stream discharge recorded in some of the above mentioned stations are given in Table 5-3. The USGS stream discharge reports are given in Appendix D.

PAGE 205

190 High surface water concentrations of nitrate-nitrogen observed in SUW718971 and Fannin can be explained by relatively low stream discharges at the stations. According to Hornsby and Mattson (1998), the discharge of water into the Santa Fe and Suwannee rivers has contributed to their significant increase in nitrate-nitrogen loadings to the Gulf of Mexico. Similar studies by Pittman et al. (1997) indicated that nearly 90% of the increase in nitrate-nitrogen loading from these reaches was attributed to discharge from spring flow. The relatively low concentrations of nitrate-nitrogen observed at many surface water quality monitoring stations in the Santa Fe Reach 2, despite its contribution to the high N-loading to the Gulf of Mexico, is likely to be attributed to this phenomenon. Identifying the sources and processes that have an effect (increasing or decreasing) on nitrate-nitrogen concentration in ground water systems in the SFRW are required to reduce the high N-loadings to the Gulf of Mexico and to prevent further degradation of ground water quality in the watershed. High fertilizer application rates on the sandy Floridian soils combined with the high precipitation rates and a karst aquifer system causes enhanced leaching of nutrients and other agrochemicals to ground water systems in Florida. Hence, this study was an attempt to understand spatial interrelationships between the distribution of land resources and the variation in nitrate-nitrogen levels measured at surface water quality monitoring stations across the SFRW. The delineation of sub-watersheds was performed using a DEM and terrain analysis. Due to relatively flat terrain (e.g. flat swampy areas) it was difficult to delineate watershed boundaries with high accuracy. We assumed that surface watershed boundaries determine water flow to the drainage outlet at each subwatershed. This is a simplified

PAGE 206

191 assumption because flow pattern in the SFRW are complex including recharge, river seeps, springs, etc. Figure 5-20. Location of surface water monitoring stations (springs) in SRWMD (Source: Katz et al., 1999). Table 5-3 Stream discharge at monitoring stations Station Discharge (m 3 /sec) SUW718971 0.20 Fannin 3.08 Santa Fe River at Fort White* 22.45-80.67 Santa Fe River near Hildreth* 23.44-80.47 Santa Fe River at Branford* 57.01-456.18 * Long term range

PAGE 207

192 Despite these shortcomings the correlation analysis relating average nitrate-nitrogen concentrations measured at the sub-watershed outlet points and their sub-watershed land resource attributes showed some significant correlations. Maximum population, DRASTIC score and the mean elevation had significant correlations with the nitrate-nitrogen values at the 0.01 significance level. DRASTIC index scores had the highest correlation with nitrate-nitrogen concentrations. The DRASTIC index scores were generated by the U.S.EPA and the N.W.W.A to map potential zones for aquifer vulnerability to ground water contamination by considering various soil-landscape attributes. Maximum population also had a significant correlation with the nutrient concentrations. Increase in urbanization, though not very significant in the SFRW, has potential impact on the nitrate-nitrogen concentrations measured at the surface water stations. Agricultural classes had a positive correlation and pine plantations had a negative correlation with nitrate-nitrogen values. Based on the land cover change analysis in chapter3, agricultural classes have increased significantly over the last decade in the SFRW. Agricultural expansion can cause surface water nitrate-nitrogen levels to increase. Also, the nutrient values had a significant correlation with the percent of Ultisols and limestones. Ultisols and limestones have the potential to rapidly leach nitrate-nitrogen present in the soils to ground water systems. The results of the correlation analysis were helpful in understanding the key sub-watershed attributes that have an effect on nitrate-nitrogen concentrations in the watershed. In the next analysis, three sub-watersheds were individually studied to better understand these spatial relationships and factor combinations of land resources that have an effect (increasing or decreasing) on surface water nitrate-nitrogen concentrations.

PAGE 208

193 Relative to the other water quality measurement stations in the SFRW, the station at ICH010C1 measured high values of nitrate-nitrogen (0.40-0.68 mg/l). The sub-watershed attributes of the SB1 were analyzed. The results shown in Figure 5-17 indicate the percent distribution of sub-watershed attributes. In 1990, nearly 32% of the area was covered by agricultural lands, 25% by pine plantations and nearly 23% by rangeland. By 2000, the percent of agricultural lands increased by 15%. This is also seen with a significant decrease in rangelands from 23% to 8%. Pine plantations occupied 20% of the sub-watershed in 2000. This trend continued to 2003, where 53% of the sub-watershed was occupied by agricultural lands. This is also seen with a nearly 3% decrease in rangeland and pine plantations from the year 2000.These land cover shifts in this sub-watershed were compared to the temporal variation of total nitrate-nitrogen inputs from Columbia county (Figure 3-36). From the figure, it can be inferred that nitrate-nitrogen inputs from agricultural fertilizers contributed a significant portion to the total nitrate-nitrogen inputs. And their contribution has increased from 2.6 to 3.8 (millions of kilograms) between 1990 and 1997. Thus, this increase can be explained with the nearly 11% increase in agricultural land between 1990 and 2003 in SB1. Analyzing the distribution of soils in the sub-watershed, it can be seen that nearly 73% of the soils were Ultisols and nearly 25% of the underlying geology were limestones and fine sands. The expansion of agricultural lands on sandy Ultisols and on a karst/fine sand geology can cause accelerated leaching of nitrate-nitrogen from the soil to the ground water system and surface water translating into elevated NO 3 -N measured at this station. The water quality measurements at NEW007C1 were relatively low values (0.01-0.07 mg/l) compared to other surface water nitrate-nitrogen measurements in the SFRW.

PAGE 209

194 Analyzing the sub-watershed attributes of the contributing area reveals that nearly 46% of the sub-watershed was occupied by pine plantations in 1990. Wetlands occupied a predominant portion of the sub-watershed area. Agricultural lands constituted nearly 9% of the total area. By 2000, the percentage of area covered by pine plantations increased by 6% and agricultural lands increased by around 10%. Wetlands dropped by 19% during the same time period. By 2003, the percent of land covered by pine plantations decreased to 35%. Wetlands accounted for about 35% of the total area in 2003. Agricultural lands increased by only 1% from 2000 to 2003. This sub-watershed is located in parts of Bradford and Baker counties. These regions are predominantly occupied by forest wetlands, pine plantations and land owned by timber corporations. Distinguishing forested wetlands from pine plantations/upland forest is rather difficult using only Landsat images. Effective delineation of these classes is possible by using extensive ground truth data, passive remote sensing techniques like LIDAR mapping and other. These factors might have introduced some classification errors during the thematic map generation and this has resulted in the dynamic shifts between these two land cover classes in this sub-watershed. Agricultural lands have a positive correlation and pine plantations have a negative correlation with nitrate-nitrogen concentrations measured across the SFRW. The relatively high portions of pine plantations and low portions of agricultural lands combined with the high distribution of clayey Spodosols on a clayey geologic layer can be attributed to the relatively low concentrations of nitrate-nitrogen measured at this station. The water quality measurements at SFR02C1 have recorded relatively moderate levels of nitrate-nitrogen (0.05-0.15 mg/l) concentrations when compared to the other

PAGE 210

195 stations in the SFRW. Analyzing the sub-watershed attributes of the contributing area reveals that nearly 34% of the area was occupied by pine plantations and 18% by agricultural lands in 1990. Rangelands occupied a significant 13% of the sub-watershed area. By 2000, the percent coverage of agricultural lands increased to 32%. This is also seen with a 6% decrease in pine plantations. Rangelands have also decreased significantly over the last decade. By 2003, agricultural lands continued to increase to 38% and pine plantations and rangelands continued to decrease to 21% and 5.2%, respectively. The underlying geology in this region is predominantly fine sand. The trend in land cover shifts in this sub-watershed is similar to the trends observed in SW1. However, the concentrations of nitrate-nitrogen measured at this station are lower than the measurements at ICH010C1. The high percent of clayey Spodosols in this sub-watershed may have been attributed to the relatively low nitrate-nitrogen levels at this station. Overall, measured nitrate-nitrogen at surface water stations within the SFRW are low from 1989 to 2003. Therefore, the recently increasing nitrate-nitrogen loads calculated at the reach 2 and 1 of the SFRW can be attributed to increasing discharge. The correlation analysis and the study of individual sub-watersheds indicated that multiple factors contribute to elevated nitrogen concentrations found in soils and water. Karst terrain, soil material, and agricultural and urban land uses pose the greatest risk for nitrate leaching. In addition, the geographic position and spatial distribution of land resources as well as the spatial interrelationships between factors influence nitrogen levels observed in soils and surface water. The complexity of land resources and flow patterns within the SFRW confounded a concise interpretation of data. Despite increasing

PAGE 211

196 agricultural lands in the SFRW there has been minor documented effect on surface water quality. This might be due to dominant vertical flow patterns that leach nitrate-nitrogen through the sand-rich soils into the aquifer. The transport of nitrate-nitrogen via surface runoff and lateral flow paths are expected to be minor. The relatively low soil nitrate-nitrogen values measured in Sept. 2003 throughout the watershed were consistent with the relatively low nitrate-nitrogen concentrations measured in surface water.

PAGE 212

CHAPTER 6 SYNTHESIS The Suwannee River and its estuary have been designated as a National Wildlife Refuge, a State Aquatic Preserve and as an outstanding Florida water body. However, the rising levels of nitrate-nitrogen in the surface and ground water systems have been a growing concern since the early 90s. Elevated concentrations of nitrate-nitrogen can cause eutrophication of water bodies. Eutrophication is the natural aging process through which shallow water bodies are converted to dry land and is greatly accelerated by increased nitrogen inputs into the surface and ground water system. One of the most visible consequences of eutrophication is the increase in algal blooms. Studies by Hornsby and Mattson (1996) have documented the increase in periphyton biomass along the middle and lower reaches of the Suwannee River. This results in oxygen depletion, which has an adverse effect on aquatic ecosystems. Eutrophication plays a prominent role in the loss of aquatic biodiversity. There are serious health implications associated with elevated nitrate-nitrogen concentrations. High nitrate levels in drinking water can cause methemoglobinemia or “blue baby” sickness in infants less than 3 to 6 months of age. Other problems may include stomach ulcer and enlargement of the thyroid gland. Water quality studies in many places in the Suwannee River Basin have shown high levels of nitrate-nitrogen in the upper Floridian aquifer, which is the drinking water source in the area. The contribution of the SFRW to the increased nitrate-nitrogen loadings to the Gulf of Mexico has increased significantly over the last few years. 197

PAGE 213

198 Basin characteristics such as LULC, slope and soil attributes affect water quality by regulating transport of sediment and chemical concentrations. Earlier studies have indicated agricultural fertilizers and manure from animal operations as the source of nitrate-nitrogen in surface and ground water systems. The overall goal of this study was to gain a better understanding of soil and water quality and how it relates to spatially-distributed watershed characteristics in SFRW. Three research hypotheses were analyzed in the study. They were: 1) Land cover shifts favoring agricultural expansion have occurred in the SFRW from 1990 to 2003, 2) The spatial distribution of soil nitrate-nitrogen is variable across the SFRW depending on both soil and LULC types and 3) Spatially distributed patterns of land resources and land cover dynamics are useful proxies providing information about nitrogen levels in soils and surface water. These hypotheses were analyzed using the following specific objectives: 1) Characterize the land cover dynamics in the SFRW from 1990 to 2003, 2) Quantify the spatial distribution of soil nitrate-nitrogen across the SFRW and 3) Investigate spatial relationships between watershed characteristics and water quality. A multi-temporal image classification and change detection analysis was performed to determine land cover changes from 1990 to 2003. The results of the analysis indicated an increase in agricultural lands by 14% between 1990 and 2003. A change trajectory analysis was performed to determine land cover shifts within classes. The results revealed that conversions between rangeland and agricultural land and between pine plantations and agricultural lands are prominent. A three data change image was created to quantify and visualize the regions of agricultural expansion over the three dates. From this analysis, it was inferred that most of the shifts to agricultural land cover classes are

PAGE 214

199 prominent in the western regions of the watershed. Suwannee, Gilchrist and Columbia counties occupy the western part of the SFRW. The results of this analysis were compared to the temporal differences in agricultural fertilizer sales and other sources of nitrate-nitrogen in water systems for each of the counties. This comparison revealed a similar trend as determined in the remote sensing analysis. Estimated total nitrate-nitrogen values and agricultural fertilizer sales were significantly high in Suwannee, Columbia and Gilchrist counties. The rates are relatively low for Alachua county. Based on this study it can be concluded that land cover shifts favoring agricultural expansion have occurred in the SFRW. Also, these shifts are prominent in the western part of the watershed, which is characterized by sandy soils and karst topography. Agricultural expansion in these regions poses the risk of increasing N-loadings to the Gulf of Mexico. Developing soil sampling schemes by considering the LULC information alone would not consider the within class variability of soil nitrate-nitrogen leaching based on the soil type. In the analysis of soil nitrate-nitrogen values in the SFRW, the site selection protocol was helpful in understanding the importance of considering both LULC and soil data for mapping the spatial variability of nitrate-nitrogen, rather than using only LULC information. The observed soil nitrate-nitrogen values proved that the spatial distribution of soil nitrate-nitrogen is variable across the SFRW depending on both soil and LULC types and is thus best addressed by considering both types of information. The mapped soil nitrate-nitrogen concentrations helped better understand the impact of different LULC-soil combinations in the SFRW. The results of the September 2003 sampling event revealed relatively low soil nitrate-nitrogen concentrations. But this analysis is the first attempt in quantifying the spatio-temporal variation of soil nitrate-nitrogen

PAGE 215

200 concentrations across the watershed. The results of this analysis will help better target LULC-soil categories that have a significant impact on soil nitrate-nitrogen content in the SFRW. Soil sampling events will be conducted at five more time periods to quantify the spatial and temporal dynamics of soil nitrate-nitrogen within the SFRW. The surface water nitrate-nitrogen measurements were analyzed to determine spatial interrelationships between the surface water nitrate-nitrogen concentrations and the geographic distribution of land resources. From this analysis, it was concluded that multiple factors contributed to slightly elevated nitrate-nitrogen concentrations and among these factors, agricultural land cover classes, population (urban land cover classes), karst topography and Ultisols pose the risk for exporting nitrate-nitrogen out of the SFRW into the Gulf of Mexico. Factor combinations of these attributes can be used as proxies to infer nitrate-nitrogen concentrations in surface and potentially ground water. BMPs for nitrate-nitrogen are broadly defined as “economically sound, voluntary practices that are capable of minimizing nutrient contamination of surface and ground water”. The recommendations of BMPs are based on sound research. Some of the management practices include manure and nutrient management, disposal of hazardous wastes, pest management, irrigation recommendations and design and engineering aspects in enterprises like row crops, dairy, poultry and beef cattle farms, agroforestry approaches like alley cropping, forest farming, creating riparian forest buffers and silvopastures. The effectiveness of these BMPs in the SFRW has to be evaluated. Adopting these techniques are key to effectively reduce N-inputs to the Gulf of Mexico. Implementing BMPs require a truly multi-disciplinary approach that meshes together science, technology and social realities. This research points into that direction.

PAGE 216

201 Understanding these complex and interrelated processes that drive high N-loads into the Gulf of Mexico support ongoing efforts to preserve and improve water quality in the SFRW.

PAGE 217

APPENDIX A HEADER FILE OF LANDSAT IMAGES 1990 Landsat tm image NDF_REVISION=2.00; DATA_SET_TYPE=EDC_TM; PRODUCT_NUMBER=011001108000700009; PIXEL_FORMAT=BYTE; PIXEL_ORDER=NOT_INVERTED; BITS_PER_PIXEL=8; PIXELS_PER_LINE=6900; LINES_PER_DATA_FILE=6299; DATA_ORIENTATION=UPPER_LEFT/RIGHT; NUMBER_OF_DATA_FILES=7; DATA_FILE_INTERLEAVING=BSQ; TAPE_SPANNING_FLAG=1/1; START_LINE_NUMBER=1; START_DATA_FILE=1; LINES_PER_VOLUME=44093; BLOCKING_FACTOR=1; RECORD_SIZE=6900; UPPER_LEFT_CORNER=0831800.7355W,0311431.1966N,280937.667,3458703.426; UPPER_RIGHT_CORNER=0811603.1867W,0305651.8865N,474443.916,3423841.615; LOWER_RIGHT_CORNER=0813528.0995W,0292109.2196N,442619.060,3247192.484; LOWER_LEFT_CORNER=0833529.7592W,0293835.4101N,249112.811,3282054.294; REFERENCE_POINT=SCENE_CENTER; REFERENCE_POSITION=0822614.6402W,0301801.4467N,361778.364,3352947.955,3450.50,3150.00; REFERENCE_OFFSET=13.79,-0.72; ORIENTATION=10.212774; MAP_PROJECTION_NAME=UTM; USGS_PROJECTION_NUMBER=1; USGS_MAP_ZONE=17; USGS_PROJECTION_PARAMETERS=6378137.000000000000000,6356752.; HORIZONTAL_DATUM=NAD83; EARTH_ELLIPSOID_SEMI-MAJOR_AXIS=6378137.000; EARTH_ELLIPSOID_SEMI-MINOR_AXIS=6356752.314; EARTH_ELLIPSOID_ORIGIN_OFFSET=0.000,0.000,0.000; EARTH_ELLIPSOID_ROTATION_OFFSET=0.000000,0.000000,0.000000; PRODUCT_SIZE=FULL_SCENE; PIXEL_SPACING=28.5000,28.5000; 202

PAGE 218

203 PIXEL_SPACING_UNITS=METERS; RESAMPLING=CC; PROCESSING_DATE/TIME=2000-11-08T12:32:53; PROCESSING_SOFTWARE=NLAPS_4_1_6e4; NUMBER_OF_BANDS_IN_VOLUME=7; WRS=017/039; ACQUISITION_DATE/TIME=1990-08-26T15:21:05Z; SATELLITE=LANDSAT_5; SATELLITE_INSTRUMENT=TM; PROCESSING_LEVEL=08; SUN_ELEVATION=54.71; SUN_AZIMUTH=116.64; BAND1_NAME=TM_BAND_1; BAND1_FILENAME=LT5017039009023810.I1; BAND1_WAVELENGTHS=0.45,0.52; BAND1_RADIOMETRIC_GAINS/BIAS=0.6024314,-1.5200000; BAND2_NAME=TM_BAND_2; BAND2_FILENAME=LT5017039009023810.I2; BAND2_WAVELENGTHS=0.52,0.60; BAND2_RADIOMETRIC_GAINS/BIAS=1.1750981,-2.8399999; BAND3_NAME=TM_BAND_3; BAND3_FILENAME=LT5017039009023810.I3; BAND3_WAVELENGTHS=0.63,0.69; BAND3_RADIOMETRIC_GAINS/BIAS=0.8057647,-1.1700000; BAND4_NAME=TM_BAND_4; BAND4_FILENAME=LT5017039009023810.I4; BAND4_WAVELENGTHS=0.76,0.90; BAND4_RADIOMETRIC_GAINS/BIAS=0.8145490,-1.5100000; BAND5_NAME=TM_BAND_5; BAND5_FILENAME=LT5017039009023810.I5; BAND5_WAVELENGTHS=1.55,1.75; BAND5_RADIOMETRIC_GAINS/BIAS=0.1080784,-0.3700000; BAND6_NAME=TM_BAND_6; BAND6_FILENAME=LT5017039009023810.I6; BAND6_WAVELENGTHS=10.40,12.50; BAND6_RADIOMETRIC_GAINS/BIAS=0.0551584,1.2377996; BAND7_NAME=TM_BAND_7; BAND7_FILENAME=LT5017039009023810.I7; BAND7_WAVELENGTHS=2.08,2.35; BAND7_RADIOMETRIC_GAINS/BIAS=0.0569804,-0.1500000; END_OF_HDR; 2000 LANDSAT ETM+ NDF_REVISION=2.00; DATA_SET_TYPE=EDC_ETM+;

PAGE 219

204 PRODUCT_NUMBER=011020618016900003; PIXEL_FORMAT=BYTE; PIXEL_ORDER=NOT_INVERTED; BITS_PER_PIXEL=8; PIXELS_PER_LINE=6874; LINES_PER_DATA_FILE=6315; DATA_ORIENTATION=UPPER_LEFT/RIGHT; NUMBER_OF_DATA_FILES=6; DATA_FILE_INTERLEAVING=BSQ; TAPE_SPANNING_FLAG=1/1; START_LINE_NUMBER=1; START_DATA_FILE=1; LINES_PER_VOLUME=37890; BLOCKING_FACTOR=1; RECORD_SIZE=6874; UPPER_LEFT_CORNER=0831828.1126W,0311430.4305N,280212.744,3458694.948; UPPER_RIGHT_CORNER=0811657.5901W,0305657.7775N,473000.895,3424026.532; LOWER_RIGHT_CORNER=0813622.4476W,0292100.0696N,441152.157,3246918.372; LOWER_LEFT_CORNER=0833557.1913W,0293819.6914N,248364.006,3281586.788; REFERENCE_POINT=SCENE_CENTER; REFERENCE_POSITION=0822655.5898W,0301756.4055N,360682.451,3352806.660,3437.50,3158.00; REFERENCE_OFFSET=106.24,-21.66; ORIENTATION=10.194344; MAP_PROJECTION_NAME=UTM; USGS_PROJECTION_NUMBER=1; USGS_MAP_ZONE=17; USGS_PROJECTION_PARAMETERS=6378137.000000000000000,6356752.314140000400000; HORIZONTAL_DATUM=NAD83; EARTH_ELLIPSOID_SEMI-MAJOR_AXIS=6378137.000; EARTH_ELLIPSOID_SEMI-MINOR_AXIS=6356752.314; EARTH_ELLIPSOID_ORIGIN_OFFSET=0.000,0.000,0.000; EARTH_ELLIPSOID_ROTATION_OFFSET=0.000000,0.000000,0.000000; PRODUCT_SIZE=FULL_SCENE; PIXEL_SPACING=28.5000,28.5000; PIXEL_SPACING_UNITS=METERS; RESAMPLING=CC; PROCESSING_DATE/TIME=2002-06-19T20:55:00; PROCESSING_SOFTWARE=NLAPS_4_1_12e9; NUMBER_OF_BANDS_IN_VOLUME=6; WRS=017/039; ACQUISITION_DATE/TIME=2000-08-13T15:52:17Z; SATELLITE=LANDSAT_7; SATELLITE_INSTRUMENT=ETM+;

PAGE 220

205 PROCESSING_LEVEL=08; SUN_ELEVATION=61.73; SUN_AZIMUTH=118.08; BAND1_NAME=ETM+_BAND_1; BAND1_FILENAME=LE7017039000022650.I1; BAND1_WAVELENGTHS=0.45,0.52; BAND1_RADIOMETRIC_GAINS/BIAS=0.7756863,-6.1999969; BAND2_NAME=ETM+_BAND_2; BAND2_FILENAME=LE7017039000022650.I2; BAND2_WAVELENGTHS=0.52,0.60; BAND2_RADIOMETRIC_GAINS/BIAS=0.7956862,-6.3999939; BAND3_NAME=ETM+_BAND_3; BAND3_FILENAME=LE7017039000022650.I3; BAND3_WAVELENGTHS=0.63,0.69; BAND3_RADIOMETRIC_GAINS/BIAS=0.6192157,-5.0000000; BAND4_NAME=ETM+_BAND_4; BAND4_FILENAME=LE7017039000022650.I4; BAND4_WAVELENGTHS=0.76,0.90; BAND4_RADIOMETRIC_GAINS/BIAS=0.9654902,-5.1000061; BAND5_NAME=ETM+_BAND_5; BAND5_FILENAME=LE7017039000022650.I5; BAND5_WAVELENGTHS=1.55,1.75; BAND5_RADIOMETRIC_GAINS/BIAS=0.1257255,-0.9999981; BAND6_NAME=ETM+_BAND_7; BAND6_FILENAME=LE7017039000022650.I7; BAND6_WAVELENGTHS=2.08,2.35; BAND6_RADIOMETRIC_GAINS/BIAS=0.0437255,-0.3500004; END_OF_HDR; 2003 LANDSAT ETM+ REQ ID =080030530024900002 LOC =017/0390000 ACQUISITION DATE =20030211 SATELLITE =LANDSAT7 SENSOR =ETM+ SENSOR MODE =NORMAL LOOK ANGLE = 0.00 TYPE OF PROCESSING =SYSTEMATIC RESAMPLING =CC VOLUME #/# IN SET =01/01 PIXELS PER LINE = 3966 LINES PER BAND = 3741/ 3741 START LINE # = 1 BLOCKING FACTOR = 1 REC SIZE = 3966 PIXEL SIZE = 57.00 OUTPUT BITS PER PIXEL = 8 ACQUIRED BITS PER PIXEL = 8 BANDS PRESENT =LH FILENAME =L71017039_03920030211_B61.FSTFILENAME =L72017039_03920030211_B62.FST BIASES AND GAINS IN ASCENDING BAND NUMBER ORDER 0.000000000000000 0.066823534667492

PAGE 221

206 3.199999809265137 0.037058822810650 0.000000000000000 0.000000000000000 0.000000000000000 0.000000000000000 0.000000000000000 0.000000000000000 0.000000000000000 0.000000000000000 0.000000000000000 0.000000000000000 0.000000000000000 0.000000000000000 GEOMETRIC DATA MAP PROJECTION =UTM ELLIPSOID =GRS_80 DATUM =NAD83 USGS PROJECTION PARAMETERS = 6378137.000000000000000 6356752.314140000400000 0.000000000000000 0.000000000000000 0.000000000000000 0.000000000000000 0.000000000000000 0.000000000000000 0.000000000000000 0.000000000000000 0.000000000000000 0.000000000000000 0.000000000000000 0.000000000000000 0.000000000000000 USGS MAP ZONE = 17 UL = 0833805.9640W 311435.3349N 249047.250 3459543.750 UR = 0811543.4392W 311611.6359N 475052.250 3459543.750 LR = 0811525.1917W 292046.1162N 475052.250 3246363.750 LL = 0833502.7399W 291916.9234N 249047.250 3246363.750 CENTER = 0822535.7588W 301928.3843N 362850.687 3355611.333 1998 1824 OFFSET = -2029 ORIENTATION ANGLE = 0.00 SUN ELEVATION ANGLE =38.0 SUN AZIMUTH ANGLE =144.0

PAGE 222

APPENDIX B FLUCCS LULC CLASSES 1000 URBAN AND BUILT-UP 1100 Residential, Low Density Less than two dwelling units per acre. 1200 Residential, Med. Density Two to five dwelling units per acre. 1300 Residential, High Density 1400 Commercial and Services. Condominiums and Motels combined. 1410 retail sales and services 1420 wholesale sales and services 1430 professional services 1460 oil and gas storage: except those areas associated with industrial use or manufacturing. 1470 mixed commercial and services 1480 cemeteries 1490 commercial and services under construction, as per zoning. 1500 Industrial 1510 food processing 1520 timber processing. 1523 pulp and paper mills 1530 mineral processing 1540 oil and gas processing 1550 other light industry 1560 other heavy industrial 1561 ship building and repair 1562 prestressed concrete plants. 1563 metal fabrication plants 1600 Extractive 1610 strip mines 1611 clays 1612 peat 1613 heavy metals 1620 sand and gravel pits 1630 rock quarries 1632 limerock or dolomite 1633 phosphates 1634 heavy minerals 1640 oil and gas fields 1650 abandoned lands 1660 reclaimed lands 1670 holding ponds 1700 Institutional 207

PAGE 223

208 1730 military 1750 governmental 1800 Recreational 1810 swimming beach 1820 golf course 1830 race tracks 1840 marinas and fish camps 1850 parks and zoos 1870 stadiums: those facilities not associated with high schools, colleges, or universities 1900 Open Land 1920 inactive land with street pattern but without structures 2000 AGRICULTURE 2100 Cropland and Pastureland 2110 improved pastures 2120 unimproved pastures 2130 woodland pastures 2140 row crops 2141 potatoes and cabbage 2150 field crops 2160 mixed crops: used if crop type cannot be determined 2200 Tree Crops 2210 citrus groves 2240 abandoned tree crops 2300 Feeding Operations 2310 cattle feeding operations 2320 poultry feeding operations 2400 Nurseries and Vineyards 2410 tree nurseries 2430 ornamentals 2431 shade ferns 2432 hammonck ferns 2450 floriculture 2500 Specialty Farms 2510 horse farms 2520 dairies 2540 aquaculture 2600 Other Open Lands Rural 2610 fallow cropland 3000 RANGELAND 3100 Herbaceous 3200 Shrub and Brushland 3300 Mixed Rangeland 4000 UPLAND FORESTS 4100 Upland Coniferous Forests 4110 pine flatwoods

PAGE 224

209 4120 longleaf pine xeric oak 4130 sand pine 4200 Upland Hardwood Forest (4200 4399) 4210 xeric oak 4300 Upland mixed forest 4340 upland mixed coniferous/hardwood 4370 australian pine 4400 Tree Plantations 4410 coniferous pine 4430 forest regeneration 5000 WATER 5100 streams and waterways 5200 lakes 5300 reservoirs 5340 reservoirs less than 10 acres (4 hectares) which are dominant features 5400 bays and estuaries 5500 major springs 5600 slough waters 6000 WETLANDS 6100 wetland hardwood forests 6110 bay swamps 6120 mangrove swamps 6150 river/lake swamp (bottomland) 6170 mixed wetland hardwoods 6180 cabbage palm savanna 6200 wetland coniferous forest 6210 cypress 6220 forested depressional pine 6300 wetland forested mixed 6400 vegetated non-forested wetlands 6410 freshwater marshes 6420 saltwater marshes 6430 wet prairies 6440 emergent aquatic vegetation 6450 submergent aquatic vegetation 6460 mixed scrub-shrub wetland 6500 non-vegetated wetland 7000 BARREN LAND 7100 beaches other than swimming beaches 7200 sand other than beaches 7300 exposed rocks 7400 disturbed land 7410 rural land in transition without positive indicators of intended activity 7420 borrow areas

PAGE 225

210 7430 spoil areas 8000 TRANSPORTATION, COMMUNICATION AND UTILITIES 8100 Transportation 8110 airports 8120 railroads 8130 bus and truck terminals 8140 roads and highways 8150 port facilities 8160 canals and locks 8180 auto parking facilities when not directly related to other land uses 8190 Transportation Facilities Under Construction 8191 highways 8192 railroads 8193 airports 8194 port facilities 8200 Communications 8300 Utilities 8310 Electrical Power Facilities 8320 Electrical Power Transmission Lines 8330 Water Supply Plants 8340 Sewage Treatment Plants 8350 Solid Waste Disposal 8390 Utilities Under Construction

PAGE 226

APPENDIX C LOCATION OF GROUND CONTROL POINTS Coniferous pine -82.49341 29.87568 Coniferous pine -82.48573 29.88051 Coniferous pine -82.45217 29.86381 Coniferous pine -82.45341 29.86381 Coniferous pine -82.46800 29.82818 Coniferous pine -82.35337 29.74199 Coniferous pine -82.35483 29.74145 Coniferous pine -82.35780 29.74746 Coniferous pine -82.35454 29.75176 Coniferous pine -82.35526 29.75284 Coniferous pine -82.35681 29.75284 Coniferous pine -82.35852 29.75847 Coniferous pine -82.35761 29.75915 Coniferous pine -82.35176 29.76568 Coniferous pine -82.39065 29.84338 Coniferous pine -82.32121 29.89020 Coniferous pine -82.32121 29.88993 Coniferous pine -82.28229 29.88115 Coniferous pine -82.28211 29.88194 Coniferous pine -82.16567 29.89367 Coniferous pine -82.21200 29.93396 Coniferous pine -82.21356 29.93332 Coniferous pine -82.72751 29.96028 Coniferous pine -82.71490 30.13205 Coniferous pine -82.71552 30.13160 Coniferous pine -82.42638 29.94020 Coniferous pine -82.49376 29.98326 Coniferous pine -82.50721 29.98994 Coniferous pine -82.60366 29.95641 Coniferous pine -82.58882 30.00248 Coniferous pine -82.51595 30.00128 Coniferous pine -82.44456 30.01161 Coniferous pine -82.32068 30.05056 Coniferous pine -82.26007 30.09987 Coniferous pine -82.15653 30.20467 Coniferous pine -82.18644 30.14560 Coniferous pine -82.21684 30.07770 Coniferous pine -82.35062 30.11458 Coniferous pine -82.51652 29.80367 211

PAGE 227

212 Coniferous pine -82.64268 29.96428 Coniferous pine -82.64394 30.00005 Coniferous pine -82.64171 29.99950 Coniferous pine -82.62322 30.02191 Coniferous pine -82.68648 29.63514 Coniferous pine -82.76442 29.61413 Coniferous pine -82.78414 29.65450 Coniferous pine -82.70158 29.72620 Forest regeneration -82.42747 29.96858 Forest regeneration -82.58568 30.00310 Forest regeneration -82.28289 30.15588 Forest regeneration -82.15560 30.20529 Forest regeneration -82.49453 29.81129 Forest regeneration -82.49202 29.85535 Forest regeneration -82.49176 29.85535 Forest regeneration -82.58981 29.84023 Forest regeneration -82.57525 29.81811 Forest regeneration -82.61070 29.91433 Forest regeneration -82.35583 29.74746 Forest regeneration -82.38538 29.80857 Forest regeneration -82.25969 29.87288 Forest regeneration -82.57949 29.94434 Forest regeneration -82.64266 30.00061 Forest regeneration -82.60962 30.02645 Forest regeneration -82.59964 30.06993 Forest regeneration -82.80784 29.72925 Forest regeneration -82.67131 29.72624 Forest regeneration -82.65841 29.79490 Forest regeneration -82.89650 29.95319 Forest regeneration -82.71753 29.95492 Forest regeneration -82.74508 29.97078 Forest regeneration -82.71076 29.98886 Forest regeneration -82.71045 29.98963 Forest regeneration -82.71070 30.01793 Forest regeneration -82.71097 30.01621 Forest regeneration -82.71215 30.02185 Forest regeneration -82.76277 30.08454 Impoved pasture -82.64070 29.95618 Impoved pasture -82.62309 30.02082 Impoved pasture -82.76629 29.61006 Impoved pasture -82.73099 29.72679 Impoved pasture -82.72796 29.72378 Impoved pasture -82.68357 29.72423 Impoved pasture -82.63082 29.72727 Impoved pasture -82.66527 29.82087 Impoved pasture -82.43339 29.77045

PAGE 228

213 Impoved pasture -82.46228 29.88078 Impoved pasture -82.45097 29.86956 Improved pasture -82.33520 29.89394 Improved pasture -82.32564 29.88818 Improved pasture -82.23962 29.86200 Improved pasture -82.59203 29.95232 Improved pasture -82.62003 29.93354 Improved pasture -82.63030 29.93483 Improved pasture -82.63383 29.93388 Improved pasture -82.71272 29.94894 Improved pasture -82.74472 29.97172 Improved pasture -82.70964 30.01736 Improved pasture -82.71693 30.02242 Improved pasture -82.70841 30.03682 Improved pasture -82.64966 30.09330 Improved pasture -82.64876 30.09321 Improved pasture -82.64786 30.09329 Improved pasture -82.69229 30.15532 Improved pasture -82.41753 29.87626 Improved pasture -82.44626 29.98455 Improved pasture -82.51089 30.00316 Improved pasture -82.58222 30.00469 Improved pasture -82.32212 30.05063 Improved pasture -82.25949 30.05221 Improved pasture -82.34540 30.00563 Improved pasture -82.41647 29.87429 Improved pasture -82.61833 29.92928 Improved pasture -82.62945 29.93531 Rangeland -82.49402 29.81099 Rangeland -82.49299 29.81097 Rangeland -82.49263 29.82792 Rangeland -82.49140 29.85649 Rangeland -82.45183 29.88091 Rangeland -82.46076 29.88089 Rangeland -82.45908 29.83775 Rangeland -82.47353 29.81564 rangeland -82.44573 29.77673 rangeland -82.48196 29.78604 Rangeland -82.38426 29.79949 Rangeland -82.38606 29.81410 Rangeland -82.33738 29.87581 Rangeland -82.33586 29.88168 Rangeland -82.32119 29.88738 Rangeland -82.32149 29.88835 Rangeland -82.22526 29.86103 Rangeland -82.14355 29.90182

PAGE 229

214 Rangeland -82.59844 29.98865 Rangeland -82.59651 29.98821 Rangeland -82.59879 29.95165 Rangeland -82.59082 29.94595 Rangeland -82.58884 29.94644 Rangeland -82.60009 29.94711 Rangeland -82.63052 29.93386 Rangeland -82.76550 29.61101 Rangeland -82.68348 29.72624 Rangeland -82.65849 29.77105 Rangeland -82.66473 29.82186 Rangeland -82.76183 29.95058 Rangeland -82.71283 29.94536 Rangeland -82.74514 29.97191 Rangeland -82.55259 30.00464 Rangeland -82.58575 30.00492 Rangeland -82.59847 29.97801 Rangeland -82.40952 30.01834 Rangeland -82.24060 30.21575 Rangeland -82.34677 30.00590 Rangeland -82.72520 29.72476 Row crops -82.63063 29.93332 Row crops -82.45244 29.87960 Row crops -82.46148 29.87978 Row crops -82.47418 29.87996 Row crops -82.47426 29.88069 Row crops -82.45412 29.87996 Row crops -82.45197 29.87864 Row crops -82.45197 29.87881 Row crops -82.45209 29.87783 Row crops -82.46602 29.82773 Row crops -82.38529 29.79988 Row crops -82.36234 29.84025 Row crops -82.09230 29.99556 Row crops -82.09245 29.99643 Row crops -82.09313 29.99661 Row crops -82.09317 29.99620 Row crops -82.71452 29.94194 Row crops -82.71412 29.94893 Row crops -82.71382 29.94837 Row crops -82.71076 29.96275 Row crops -82.71111 29.96198 Row crops -82.71028 29.97466 Row crops -82.71050 29.97734 Row crops -82.70782 29.97821 Row crops -82.70918 29.98261

PAGE 230

215 Row crops -82.70959 29.98237 Row crops -82.70942 29.98240 Row crops -82.71105 30.02309 Row crops -82.71215 30.02393 Row crops -82.71242 30.02338 Row crops -82.70837 30.03328 Row crops -82.42453 29.94775 Row crops -82.28206 30.15585 Row crops -82.35663 29.97828 Row crops -82.72790 29.72592 Tree groves -82.61675 29.93553 Tree groves -82.72815 29.81436 Tree groves -82.93138 29.98843 Upland forest -82.73393 29.72577 Upland forest -82.49317 29.82154 Upland forest -82.49362 29.83605 Upland forest -82.49232 29.84415 Upland forest -82.49191 29.87520 Upland forest -82.47223 29.80703 upland forest -82.59209 29.83484 upland forest -82.58968 29.83901 upland forest -82.60730 29.86109 upland forest -82.60955 29.86104 upland forest -82.60956 29.86104 upland forest -82.59891 29.91857 upland forest -82.59835 29.91830 Upland forest -82.35439 29.74746 Upland forest -82.38242 29.79965 Upland forest -82.38611 29.80841 Upland forest -82.39395 29.82866 Upland forest -82.39454 29.82793 Upland forest -82.40476 29.83808 Upland forest -82.40568 29.83829 Upland forest -82.28141 29.88045 Upland forest -82.23947 29.86306 Upland forest -82.24029 29.86223 Upland forest -82.59989 29.95259 Upland forest -82.58050 29.94502 Upland forest -82.54030 29.94481 Upland forest -82.60061 29.94596 Upland forest -82.63383 29.93180 Upland forest -82.63906 29.93398 Upland forest -82.64632 29.94260 upland forest -82.64358 29.98161 upland forest -82.64157 29.98129 upland forest -82.62252 30.02089

PAGE 231

216 upland forest -82.60887 30.02645 upland forest -82.60070 30.07039 upland forest -82.60262 30.08994 upland forest -82.60096 30.08997 upland forest -82.63087 29.64565 upland forest -82.67344 29.65854 upland forest -82.68570 29.63543 upland forest -82.79819 29.63500 upland forest -82.67150 29.72481 upland forest -82.65775 29.75001 upland forest -82.72870 29.81349 upland forest -82.88027 29.95204 upland forest -82.93039 29.98777 upland forest -82.81085 29.95223 upland forest -82.76257 29.94921 Upland forest -82.74661 29.96064 Upland forest -82.74705 29.96032 Upland forest -82.70957 29.97920 Upland forest -82.42383 29.94079 Upland forest -82.34410 29.95049 Urban -82.43282 29.77240 Urban -82.51672 29.80249 Urban -82.59929 29.82793 Urban -82.59963 29.82812 Urban -82.60097 29.82879 Urban -82.60011 29.82857 Urban -82.59738 29.82759 Urban -82.61553 29.83742 Urban -82.61449 29.83742 Urban -82.61959 29.83676 Urban -82.59783 29.82537 Urban -82.59743 29.82535 Urban -82.35777 29.75832 Urban -82.11175 29.93733 Urban -82.11250 29.93779 Urban -82.11220 29.93861 Urban -82.11151 29.93934 Urban -82.11126 29.93966 Urban -82.11123 29.93908 Urban -82.10706 29.94324 Urban -82.10768 29.94440 Urban -82.10961 29.94415 Urban -82.10848 29.94368 Urban -82.10889 29.94380 Urban -82.10797 29.94325 Urban -82.11178 29.93812

PAGE 232

217 Urban -82.10813 29.94329 Urban -82.57750 29.97759 Urban -82.63895 30.18354 Urban -82.63947 30.18273 Urban -82.63847 30.18982 Urban -82.64163 30.18951 Urban -82.63903 30.19182 Urban -82.93003 29.96031 Urban -82.92916 29.95932 Urban -82.32394 30.02307 Urban -82.70784 30.05204847 Urban -82.70091 29.64088892 Urban -82.71486 29.67628342 Urban -82.34315 30.01952685 Urban -82.41392 29.71561778 Urban -82.72854 30.17253921 Urban -82.69160 30.14998075 Urban -82.67155 30.16656908 Urban -82.71689 29.81876065 Urban -82.74500 29.80007613 Urban -82.60046 29.83233798 Urban -82.05946 30.16519721 Urban -82.04379 30.01777961 Urban -82.07346 30.04567134 Urban -82.36044 29.99006312 Urban -82.19409 30.07037738 Urban -82.57356 30.1846428 Urban -82.43336 29.77173 Water -82.73982 29.86474 Water -82.07528 29.73867 Water -82.09303 29.77020036 Water -82.16994 29.8622367 Water -82.18868 29.92751839 Water -82.15777 29.94277715 Water -82.15961 29.91962947 Water -82.33783 30.03343109 Water -82.30062 30.12386629 Water -82.14565 29.7785862 Water -82.11698 29.78116865 Water -82.18029 29.93248 Water -82.07527 29.75439772 Water -82.06272 29.72224611 Wetland -82.74248 29.82604 Wetland -82.13849 29.86471 Wetland -82.14014 29.86392 Wetland -82.13960 29.86399

PAGE 233

218 Wetland -82.61697 30.16695 Wetland -82.61822 30.16804 Wetland -82.61459 30.16709 Wetland -82.61332 30.16660 Wetland -82.76577 29.62022 Wetland -82.76534 29.61455 Wetland -82.75327 29.72629 Wetland -82.79599 29.72933 Wetland -82.76579 29.79461 Wetland -82.70120 30.15361 Wetland -82.33820 30.02711 Wetland -82.29387 30.11810 Wetland -82.42743 29.91975 Wetland -82.41572 29.87605 Wetland -82.40511 29.85325 Wetland -82.35124 30.13554 Wetland -82.35215 30.13570 Wetland -82.35517 30.13023 Wetland -82.35613 30.11230 Wetland -82.45279 30.06821

PAGE 234

APPENDIX D U.S.G.S SURFACE WATER QUALITY MONITORING REPORTS This file contains Calendar Year Stream flow Statistics for USGS 02322500 SANTA FE RIVER NEAR FORT WHITE, FLA. YEAR FLOW (CFS) 1928 2269 1929 2101 1933 1237 1934 1497 1935 1399 1936 1303 1937 1776 1938 1398 1939 1293 1940 1188 1941 1748 1942 2191 1943 1025 1944 1681 1945 1389 1946 1951 1947 2034 1948 2847 1949 1937 1950 1759 1951 1217 1952 1104 1953 1729 1954 1354 1955 844 1956 808 1957 1141 1958 1438 1959 2670 1960 2315 1961 1665 1962 988 1963 1195 1964 2422 1965 2544 1966 2684 219

PAGE 235

220 1967 1595 1968 1454 1969 1582 1970 2446 1971 1409 1972 1971 1973 1880 1974 1216 1975 1231 1976 998 1977 1076 1978 1733 1979 1375 1980 1509 1981 898 1982 1393 1983 1892 1984 2070 1985 1629 1986 1487 1987 2081 1988 1902 1989 1019 1990 793 1991 1709 1992 1555 1993 1185 1994 1264 1995 1161 1996 1275 This file contains Calendar Year Stream flow Statistics for USGS 02320500 SUWANNEE RIVER AT BRANFORD, FLA. YEAR FLOW (CFS) 1932 5460 1933 7747 1934 3139 1935 4167 1936 5849 1937 8313 1938 3318 1939 5285 1940 3997 1941 3255 1942 7893 1943 3265

PAGE 236

221 1944 8854 1945 8321 1946 9278 1947 11530 1948 15129 1949 8426 1950 3958 1951 4471 1952 5743 1953 6902 1954 3933 1955 2013 1956 3103 1957 5809 1958 8936 1959 10610 1960 8971 1961 6360 1962 4514 1963 4807 1964 16110 1965 11920 1966 10560 1967 5659 1968 2390 1969 5320 1970 9572 1971 7086 1972 8034 1973 12709 1974 5391 1975 9056 1976 7562 1977 7352 1978 7125 1979 6212 1980 7187 1981 3212 1982 5422 1983 10040 1984 12430 1985 4571 1986 8075 1987 9897 1988 5704 1989 3251

PAGE 237

222 1990 4113 1991 13569 1992 7138 1993 6907 1994 10270 1995 5900 1996 4325 1997 7472 1998 12020 1999 2728 2000 2351 2001 3806 2002 2192

PAGE 238

LIST OF REFERENCES Anderson, J. R., 1971. Land use classification schemes used in selected recent geographic applications of remote sensing, Photogrammetric Engineering, 37(4):379. Anderson, J. R., Hardy, E. E., Roach, J. T. and Witmer, R. E., 1976. A land use and land cover classification system for use with remote sensor data, U.S. Geological Survey report: 964, United States Government Printing Office, Washington, DC. Augusteijn, M. F. and Warrender, C. E., 1998. Wetland classification using optical and radar data and neural network classification, International Journal of Remote Sensing, 19(8):1545-1560. Ball, G. H. and Hall, D. J., 1965. A novel method of data analysis and pattern classification, Stanford Research Institute report: NTIS AD 699616, Stanford, California. Band, L. E., 1986. Topographic partition of watersheds with digital elevation models, Water Resources Research, 22(1):15-24. Basnyat, P., Teeter, L. D., Lockaby, B. G. and Flynn, K. M., 2000. The use of remote sensing and GIS in watershed level analyses of non-point source pollution problems, Forest Ecology and Management, 128:65-73. Bolstad, P., 2002. GIS fundamentals: a first text on geographical information systems, Eider Press, White Bear Lake, Minnesota. Briassoulis, H., 2000. Analysis of land use change: theoretical and modeling approaches, Regional Research Institute, West Virginia University, p2-8. [Online book – http://www.rri.wvu.edu/WebBook/Briassoulis/contents.htm -Verified on July 10 th 2004]. Burrough, P. A. and McDonnell, R. A., 1998. Principles of geographical information systems, Oxford University Press, New York. Carpenter, S., Correll, D. L., Howarth, R. W., Sharpley, A. N. and Smith, V. H., 1998. Nonpoint pollution of surface waters with phosphrous and nitrogen, Issues in Ecology, 3:1-12. Chavez, P.S. Jr., 1994. Automatic detection of vegetation changes in the southwestern United States using remotely sensed images, Photogrammetric Engineering and Remote Sensing, 60(5):571-583. 223

PAGE 239

224 Chen, L. and Lee, L., 1992. Progressive generation of conceptual frame work for image registration, Photogrammetric Engineering and Remote Sensing, 58(9):1321-1328. Chowdary, V. M., Yatindranath, S., Kar, S. and Adiga, S., 2001. Assessment of non-point source pollution on watershed basis using remote sensing, GIS and AGNPS model, paper presented at the 22 nd Asian Conference on Remote Sensing, November 5-9, Singapore. Clarke, A. L., Gruen, A. and Loon, J. C., 1982. The application of contour data for generating high fidelity grid digital elevation models, Proceesings of Auto-Carto, 5:231-222. Colwell, R. N., (ed.), 1960. Manual of photographic interpretation, American Society of Photogrammetry. p867-868. Congalton, R. G., 1991. A review of assessing the accuracy of classifications of remotely sensed data, Remote Sensing of Environment, 37:35-46. Congalton, R. G. and Green, K., 1999. Assessing the accuracy of remotely sensed data: principles and practices, Lewis Publishers, New York. Cowardin, L. M., Carter, V., Golet, F. C. and LaRoe, E. T., 1979. Classification of Wetlands and Deepwater Habitats of the United States, U.S. Fish and Wildlife Service report: FWS/OBS/-79/31, United States Government Printing Office, Washington, DC. Dave, T. J. V. and Bernstein, R., 1982. Effect of terrain orientation and solar position on satellite-level luminance observations, Remote Sensing of Environment, 12:331-348. Dearstyne, D.A., Leach, D. E. and Sullivan, K. J., 1991. Soil survey of Bradford County, Florida, United States Government Printing Office, Washington, DC. Duggin, M. J. and Robinove, C. J., 1990. Assumptions implicit in remote sensing data acquisition and analysis, International Journal of Remote Sensing, 11(10):1669-1694. Eliason, P. T., Soderblom, L. A., and Chavex, P. S., 1981. Extraction of topographic and spectral albedo information from multi-spectral images, Photogrammetric Engineering and Remote Sensing, 48(11):1571-1579. Engel, B., Srinivasan, R. and Rewerts, C., 1993. A spatial decision support system for modeling and managing agricultural non-point source pollution, Environmental Modeling with GIS, Oxford University Press, New York. Franklin, S. E. and Wilson, B. A., 1991. Spatial and spectral classification methods in remote sensing, Computers & Geosciences, 17(8):1151-1172.

PAGE 240

225 Franklin, E. S. and Giles, T. P., 1993. Radiometric processing of aerial and satellite remote sensing imagery, Computers & Geosciences, 21(3):413-423. Gaebrecht, J., 1991. Effects of spatial accumulation of runoff on watershed response, Journal of Environmental Quality, 20:31-35. Gallimore, W. P. and Xiang, W. N., 1991. Integrating GIS with a modified universal soil loss equation to support watershed management, International Papers and Proceedings of Applied Geography Conferences, Binghamton, New York. p198-202. Gaydos, L. and Newland, W. L., 1978. Inventory of land use and land cover of the puget sound region using Landsat digital data, Journal of Research, 6:807-814. Giles, M. F. and Curran, P. J., 1995. Environmental remote sensing from regional to global Scales, John Wiley & Sons, New York. Hall, F. G., Strebel, D. E., Nickeson, J. E. and Goetz, S.J., 1991. Radiometric rectification: toward a common radiometric response among multidate, multisensor images, Remote Sensing of the Environment, 35(1):11-27. Ham, L. K. and Hatzell, H. H., 1996. Analysis of nutrients in the surface waters of the Georgia Florida Coastal Plain study unit, U.S. Geological Survey report: 96-4037, U.S. Geological Survey, Tallahassee. Hamlett, J. M., Miller, D. A., Day, R. L., Peterson, G. W., Baumer, G. M. and Russo, J., 1992. Statewide GIS-based ranking of watersheds for agricultural pollution prevention, Journal of soil water Conservation, 47:399-404. Hornsby, D. and Mattson, R., 1998, Surface water quality and biological monitoring network, U.S. Geological Survey Water-Resources Investigations Report WR-98-02, U.S. Geological Survey, Tallahassee. Hutchinson, F, C., 1982. Techniques for combining Landsat and ancillary data for digital classification improvement, Photogrammetric Engineering and Remote Sensing, 48 (1):123-130. Jain, A.L., 1989. Fundamentals of digital image processing, Prentice Hall, Upper Saddle River, New Jersey. Jensen, J. R., 1996. Introductory digital image processing-a remote sensing perspective, Prentice Hall, Upper Saddle River, New Jersey. Jensen, S. K. and Dominique, J. O., 1988. Extracting topographic structure from digital elevation data for geographic information system analysis, Photogrammetric Engineering and Remote Sensing, 54(11):1593-1600.

PAGE 241

226 Johnston, K., VerHoef, J. M., Krivoruchko, K. and Lucas, N., 2001. Using ArcGIS geostatistical analyst, ESRI Press, Redlands, California. Kasperson, J., Kasperson, R. and Turner, B., 1995. Regions at risk, United Nations University Press, Tokyo, Japan. Kates, R. W., B.L. Turner II, and Clark, W.C., 1990. The Great transformation. In B.L. Turner, II, W.C. Clark, R.W. Kates, J.E. Richards, J.T. Mathews, and W.B. Meyer (eds.).The earth as transformed by human action, Cambridge University Press, Cambridge.p1-17. Katz, G. B., Hornsby, D. H., Bohlke, J. K., Morkay, M. F., 1999. Sources and chronology of nitrate contamination in spring waters, Suwannee River Basin, Florida, Water-Resources Investigations report 99-4252, U.S. Geological Survey, Tallahassee. Katz, G. B and Bohlke, J. K., 2000. Monthly variability and possible sources of nitrate in groundwater beneath mixed agricultural land use in Suwannee and Lafayette Counties, Florida, U.S. Geological Survey report: 00-4219, U.S. Geological Survey, Tallahassee. Katz, G. B. and DeHan, R. S., 1996. The Suwannee River Basin Pilot Study: issues for Watershed Management in Florida. U.S. Geological Survey, Tallahassee. Kaufman, L. and Rousseeuw, J. P., 1989. Finding groups in data: an introduction to cluster analysis, John Wiley & Sons, Inc. New York. Kaufman, Y. J. and Sendra, C., 1988. Algorithm for automatic atmospheric corrections to visible and near-IR satellite imagery, International Journal of Remote Sensing, 9:1357-1381. Kauppi, L., 1984. Contribution of agricultural loading to the deterioration of surface waters in Finland, Public Water Research Institute report: 57, Finland. Lam, N. S., 1990. Description and measurement of Landsat TM image using fractals, Photogrammetric Engineering and Remote Sensing, 56(2):187-195. Liang, S., Fallah-Adl, H., Jaja, J., Kaufman, Y. J., and Townshend, J. R.G., 1997. An operational atmospheric correction algorithm for Landsat thematic mapper imagery over the land, Journal of Geophysics Research, 102(17):173-176. Lillesand, T. M., and Kiefer, R. W., 1994, Remote Sensing and Image Interpretation, John Wiley & Sons, Inc, New York. Lindgren, T, D., 1985. Land use planning and remote sensing, Martinus-Nijhoff publishers, Boston, MA.

PAGE 242

227 Luque, S. S., 2000. Evaluating temporal changes using multi-spectral scanner and thematic mapper data on the landscape of a natural reserve, International Journal of Remote Sensing, 21:25892611. Macleod, R. D., and Congalton, R. G., 1998. A quantitative comparison of change detection algorithms for monitoring eelgrass from remotely sensed data, Photogrammetric Engineering and Remote Sensing, 64:207-216. Maidment, D.R. (eds.), 2002. Arc Hydro:GIS for water resources, ESRI Press, Redlands, California. Mansell, R. S, Bloom, S. A, and Burgoa, B., 1991. Phosphorous transport with water flow in acid, sandy soils. In Jacob, B and Corapcioglu, M. Y. (eds.). Transport processes in porous media. Kluwer Academic Publications., Dordrecht, Netherlands.p71-314. Mattikalli, N. and Richards, K., 1996. Estimation of surface water quality changes in response to land use change: Application of the export coefficient model using remote sensing and geographical information system. Journal of Environmental Management, 48:263-282. Matrz, L. W. and DeJong, E., 1989. Catch: a FORTRAN program for measuring catchment area from digital elevation model, Computers and Geosciences, 14(5):627-640. Mertens, B., and Lambin, E. F., 2000. Land cover change trajectories in southern Cameroon, Annals of the Association of American Geographers, 90(3):467-495. Meyer, P., Itten, I. K., Kellenberger, T., Sandmeier, S. and Sandmeier, R., 1993. Radiometric corrections of topographically induced effects on Landsat TM data in an alpine environment, ISPRS Journal of Photogrammetry and Remote sensing, 48(4):17-28. Moore, I. D., Grayson, R. B., Ladson, A. R., 1991. Digital terrain modeling: a review of hydrological, geomorphologic and biological applications, Hydrological Processes, 5:3-30. Morse, G., Eatherall, A., Jenkins, A., 1994. Managing agricultural pollution using a linked geographical information system and non point source pollution model, Institute of Water Environment Management, 8:277-286. Moser, S.C., 1996. A partial instructional module on global and regional land use/cover change: assessing the data and searching for general relationships, Geojournal, 39(3):241-283. Nair, V. D and D.A.Graetz., 2002. Phosphorous saturation in spodosols impacted by manure, Journal of Environmental Quality, 31:1279-1285.

PAGE 243

228 National Oceanic and Atmospheric Association, 1996. Climatologically data, annual summary, Florida. National Climatic Data Center, Asheville, NC. Newson, M., 1992. Land, water and development: river basin systems and their sustainable management, Routledge, London. p351. Novak, K., 1992. Rectification of digital imagery, Photogrammetric Engineering and Remote Sensing, 58(3):339-344. Paramasivam, S., Alva, A. K, Fares, A. and Sajwan, K. S., 2001. Estimation of nitrate leaching in an entisol under optimum citrus production, Soil Science Society of America Journal, 65:914-921. Peddle, D. R. and Franklin, S.E., 1993. Classification of active layer depth using remote sensing and topographic evidence, Remote Sensing of the Environment, 44(1):67-80. Perry, J. and Vanderklein, E., 1996. Water quality management of natural resource, Blackwell Science, Inc, Cambridge, MA. Pittman, J.R., Hatzell, H. H., Oaksford, E. T., 1997. Spring contributions to water quantity and nitrate loads in the Suwannee River during base flow in July 1995: U.S. Geological Survey Water-Resources Investigations Report 97-4152, U.S. Geological Survey, Tallahassee. Puri, H. S. and Vernon, R. O., 1964. Summary of geology of Florida and a guidebook to the classic exposures, Florida Geological Survey Special Publication 5 (revised). Raulston M., C. Johnson, K. Webster, C. Purdy and Ceryak. R., 1998.Suwannee river water management district. Chapter 10. p194-213. In E.A. Fernald and E.D. Purdum (ed.) Water resources atlas of Florida, Florida state university, Tallahassee. Rees, W.G., 1990. Physical principles of remote sensing, Cambridge University Press, New York. Reese, M, H., Lillesand, M, T., Nagel, E, D., Stewart, S, J., Goldmann, A, R., Simmons, E, T., Chipman, W, J., Tessar, A, P., 2002. Statewide land cover derived from multi-seasonal Landsat TM data: a retrospective of the WISCLAND project, Remote sensing of Environment, 82:224-237. Rignot, R. A., Kowk, R., Curlander, J and Page, S., 1991. Automated multi-sensor registration: requirements and techniques, Photogrammetric Engineering and Remote Sensing, 57(8):1029-1038. Roughgarden, J., Running, S. W., Matson, P.A., 1991. What does remote sensing do for ecology?, Ecology, 72:1918-1922.

PAGE 244

229 Schott, R. J., 1997. Remote sensing: the Image Chain Approach, Oxford University Press, New York. Scott, T. M., 1997. Miocene to Holocene: history of Florida. In Randazzo, A. F and Jones, D. S. (eds.). The geology of Florida. 1997. University Press of Florida.p57-68. Song, C., Woodcock, E. C., Set, C. K., Lenney, P. M., Macomber, A. S., 2001. Classification and change detection using Landsat TM data: When and how to correct atmospheric effects, Remote Sensing of Environment, 75:230-244. Southworth, J., Nagendra, H., Tucker, C., 2002. Fragmentation of a landscape: incorporating landscape metrics into satellite analyses of land cover change, Landscape Research, 27(3):253-269. Spalding, R. F. and Exner, M. E., 1993. Occurrence of nitrate in groundwater-a review, Journal of Environmental Quality, 22:392-402. SRWMD (Suwannee River Water Management District), 1998. Surface water quality and biological monitoring – Annual report. U.S. Geological Survey, Tallahassee. SRWMD (Suwannee River Water Management District), 2002. Surface water quality and biological monitoring – Annual report. U.S. Geological Survey, Tallahassee. SRWMD (Suwannee River Water Management District), 2003. Surface water quality and biological monitoring – Annual report. U.S. Geological Survey, Tallahassee. Stein, A, Van der Meer, F.D, Gorte, B. (eds.), 1999. Spatial statistics for remote sensing, Kluwer Academic Publishers, Dordrecht, Netherlands. Steyaert, T., 1993. A perspective on the state of environmental simulation models: Environmental Modeling with GIS, Oxford University Press, New York. Swain, P. H and Davis, S. M., 1985. Remote sensing: the quantitative approach, Mc-Graw Hill, New York. Teillet, P. M., 1986. Image correction for radiometric effects in remote sensing, International Journal of Remote Sensing, 7(12):132-161. Turner, B. L. II. and Meyer, B. L., 1994. Global land use and land cover change: an overview. In W.B. Meyer and B.L. Turner II (eds.).Changes in land use and land cover: a global perspective, Cambridge University Press, Cambridge, MA.p3-10. Upchurch, S. B., 2002. Hydrogeochemistry of a karst escarpment. In Martin, J. B, Wicks, C. M, and Sasowsky. (eds.) Hydrogeology and biology of post-paleozoic carbonate aquifers. Karst Water Institute Publication, Charles Town, WV,p205-211.

PAGE 245

230 USEPA (US Environmental Protection Agency), 1993. The watershed protection approach-report: EPA 840-s-93-001, United States Government Printing Office, Washington, DC. USEPA (US Environmental Protection Agency), 1999. Understanding the safe drinking water act-report: EPA 810-F-99-008, United States Government Printing Office, Washington, DC. Viessman, W., Jr. and Hammer, M. J., 1993. Water supply and pollution control, Harper Collins College Publications, New York. Vieux, B. E., 1991. Geographical information system and non-point source water quality and quantity modeling, Hydrological Processes, 5:101-113. Vogelmann, E.T., Helder, D., Morfitt, R., Choate, J.M., Merchant, W.J., Bulley, H., 2001. Effects of Landsat 5 Thematic mapper and Landsat 7 Enhanced thematic mapper plus radiometric and geometric calibrations and correction on landscape characterization, Remote Sensing of Environment, 78:55-70. Weatherspoon, R. L., Cummings, E. and Wittstruck, W. H., 1992. Soil survey of Gilchrist County, Florida, U.S. Government Printing Office. Washington, D.C. Wang, X. and Yin, Z., 1997. Using GIS to assess the relationship between land use and water quality at a watershed level, Environmental International, 23:103-114. White, D. A., Smith, R.A., Price, C.V., Alexander, R.B.and Robinson, K.W., 1992. A spatial model to aggregate point-source and non-point source water quality data for large areas, Computers and Geosciences, 18:1055-1073. White, W. A., 1970. The geomorphology of Floridian peninsula, Florida Bureau of Geology Bulletin, 51. Woodard, R. K., French, C. E., Sweat, A. L., Graetz, D. A., Sollenberger, L. E., Macoon, B., Portier, M. P., Wade, L. B., Rymph, J. S., Prine, G. M., Van Horn, H. H., 2002. Nitrate removel and nitrate leaching from forage systems receiving dairy effluent, Journal of Environmental Quality, 31:1980-1992.

PAGE 246

BIOGRAPHICAL SKETCH Aarthy Sabesan was born in Coimbatore, India, to Dr. Sabesan Sivam and Dr. Usha Sabesan. She was raised in Tamil Nadu, India. She received her high school diploma in mathematics, physics, chemistry and biology in 1997. She enrolled at the College of Engineering (CEG), Anna University, for her bachelor’s degree. She graduated first class from the Department of Civil Engineering. Later, she moved to the U.S to pursue her graduate education. She enrolled for the Master of Science program at the University of Florida, Gainesville. Her interests are in using and learning new geographic techniques to effectively analyze environmental issues. 231