IMPACTS OF LAND USE CHARACTERIZATION IN MODELING HYDROLOGY AND SEDIMENTS FOR THE LUXAPALLILA CREEK WATERSHED, ALABAMA AND MISSISSIPPI J. N. Diaz‐Ramirez, V. J. Alarcon, Z. Duan, M. L. Tagert, W. H. McAnally, J. L. Martin, C. G. O'Hara
ABSTRACT. The Hydrological Simulation Program - Fortran (HSPF), interfaced with the Better Assessment Science Integrating Point and Nonpoint (BASINS), was used to evaluate the impact of land use (as characterized by different land use/land cover (LU/LC) datasets) on hydrology and sediment components of the Luxapallila Creek watershed. The 1,770 km2 watershed is located in Alabama and Mississippi. Simulation of the watershed processes were tested at the hillslope and at the watershed outlet for the period between 1985 and 2003. Three LU/LC databases were used: the Geographic Information Retrieval and Analysis System (GIRAS), the Moderate Resolution Imaging Spectroradiometer land cover product (MODIS MOD12Q1), and the National Land Cover Dataset (NLCD). The two main land use categories revealed by the three LU/LC databases were forest and agricultural lands. Whereas forest cover mechanisms were the main source of water loss in hydrology simulation, agricultural land was the main source of sediment export in sediment modeling. Land use datasets of coarser spatial resolution (MODIS and GIRAS) produced larger HSPF estimations for sediment fraction values than land use datasets identifying smaller percentages of those agricultural land cover classes (NLCD). Differences in agricultural land characterization among the land use datasets showed that sediment predictions were more sensitive than streamflow predictions to the scale and resolution of land use datasets. Choosing the right land use dataset will impact the modeling of sediments and, potentially, other water quality constituents that are related with agricultural activities. Keywords. HSPF, Hydrology, Land use, Sediments, Watershed modeling.
L
and use is an important factor in watershed modeling, especially when water quality or environmental man‐ agement is the focus of the modeling/ simulation ef‐ fort. Loss of farmland to urban sprawl impacts the environment in a variety of ways, including groundwater re‐ charge, water pollution, and stormwater drainage (Engel et al., 2003). Some studies have identified a relationship between land use and surface water quality. For example, Tong and Chen (2002) used a comprehensive approach to study this relationship in a local watershed in the East Fork Little Miami River basin, Ohio, using the Better Assessment Science Integrating Point and Nonpoint Sources (BASINS) software suite. Their study re‐ vealed that there was a significant relationship between land use and in‐stream water quality, especially for nitrogen, phosphorus, and fecal coliform. It also revealed that BASINS is a very useful
Submitted for review in August 2007 as manuscript number SW 7156; approved for publication by the Soil & Water Division of ASABE in December 2007. The authors are Jairo N. Diaz‐Ramirez, ASABE Member Engineer, Post‐Doctoral Associate, Department of Civil and Environmental Engineering, and Vladimir J. Alarcon, Assistant Research Professor, GeoResources Institute, Mississippi State University, Mississippi State, Mississippi; Zhiyong Duan, Engineer, CDM, New Orleans, Louisiana; Mary Love Tagert, Assistant Research Professor, Water Resources Research Institute, William McAnally, Associate Professor, and James L. Martin, Professor, Department of Civil and Environmental Engineering, and Charles G. O'Hara, Associate Research Professor, GeoResources Institute, Mississippi State University, Mississippi State Mississippi. Corresponding author: Jairo N. Diaz‐Ramirez, Department of Civil and Environmental Engineering, Mississippi State University, P.O. Box 9546, Mississippi State, MS 39762; phone: 662‐325‐9885; fax: 662‐325‐7189; e‐mail:
[email protected].
and reliable tool, capable of characterizing the flow and water quality conditions for the study area under different watershed scales. Whitehead et al. (2002) applied steady‐state and dynam‐ ic modeling studies to illustrate the impact on the water quality of the River Kennet (U.K.) due to land use change. Their results showed that there has been a significant shift in the nitrogen bal‐ ance in the River Kennet system caused by nonpoint‐source pollution from agriculture and point‐source discharges. BASINS is a multipurpose environmental analysis pro‐ gram developed by the U.S. Environmental Protection Agency (USEPA, 2001). It integrates a geographic informa‐ tion program (ArcView), national watershed and meteoro‐ logical databases, and modeling simulation routines into one package (USEPA, 2001). BASINS provides an interface that allows the user to manipulate and display geospatial data and interact with point and nonpoint source pollution models in a GIS‐based environment. The Hydrologic Simulation Program - Fortran (HSPF) model (Bicknell et al., 2001) computes the movement of wa‐ ter through a complete hydrologic cycle (rainfall, evapotran‐ spiration, runoff, infiltration, and flow through the ground) and the associated transport of constituents with that flow. It represents a watershed as a collection of land segments and channels (reaches). The land segments, either pervious or im‐ pervious, are connected to channel reaches, which can func‐ tion as either streams or reservoirs. Rainfall is computed over the entire watershed and runs off land segments and reaches. Pervious land segments also store water in the plant canopy, on the surface, and in the soil, from which it can percolate into groundwater or flow downslope as interflow. HSPF also com‐ putes the transport and kinetics of multiple water quality con‐
Transactions of the ASABE Vol. 51(1): 139-151
E 2008 American Society of Agricultural and Biological Engineers ISSN 0001-2351
139
stituents, including temperature, sediment, nutrients, and pesticides. As such, it presents a well designed package for modeling the hydrology and water quality of a watershed. A more complete description of its features and capabilities can be found in the HSPF user's manual (Bicknell et al., 2001). The version of HSPF that is packaged with Version 3.1 of BA‐ SINS can be run in the Windows environment (in stand‐alone mode) and is called WinHSPF. In this article, WinHSPF and HSPF will be used interchangeably. Employing different land use databases for simulating hy‐ drological phenomena can produce different simulation re‐ sults, affecting subsequent modeling and simulation of related hydrological or water quality processes. For example, hydrograph estimations were sensitive to swapping alterna‐ tive land cover maps (Endreny et al., 2003), showing effects ranging from 35% underestimation to 20% overestimation for peak flows. McAnally et al. (2005) detected differences in simulated values of total suspended sediments after using two land use databases when modeling hydrological pro‐ cesses for a watershed in Mississippi. Moglen and Beighley (2002) showed that temporal land use change (captured with two different land use maps) resulted in considerable peak discharge changes throughout the Watts Branch watershed, Maryland. Anantharaj et al. (2006) showed that changes in land cover altered surface fluxes (particularly the latent heat fluxes) when simulating sea breeze in the gulf coast area us‐ ing NASA's Moderate Resolution Imaging Spectroradiome‐ ter (MODIS) and USGS land cover data. While several studies have focused on the sensitivity of flow rate simula‐ tions to land use, not many have analyzed the sensitivity of water quality estimations (sediments, nutrients, dissolved oxygen, biochemical oxygen demand) to land use data quali‐ ty. Today's easy and fast access to updated land cover maps (e.g., MODIS) provides good opportunities to investigate the impact of land use data in modeling and simulation of wa‐ tershed processes. This research explores the impact of agricultural lands, as characterized by different land use/land cover (LU/LC) data‐
sets, on simulation of watershed processes at hillslope scale and at the watershed outlet of the Luxapallila Creek basin, lo‐ cated in Alabama and Mississippi. Simulated values of flow and sediments are obtained after swapping three land use da‐ tabases. The changes in simulated values are analyzed and compared. BASINS and WinHSPF are used to perform the spatial and hydrological estimations. The MODIS land use/ land cover data (MODIS MOD12Q1) provided by NASA, the Geographic Information Retrieval and Analysis System (GIRAS), and National Land Cover Data (NLCD) were used in the analysis. Although the GIRAS land use dataset is not the best‐quality land use map available, it is taken here as the benchmark for comparison in illustrating how higher‐ resolution data (NLCD) or more up‐to‐date data (MODIS) would impact BASINS/WinHSPF's flow and sediment es‐ timations.
METHODOLOGY BASINS/WinHSPF was used to calculate rainfall‐runoff, flow, and sediments for the Luxapallila Creek watershed. The watershed was initially modeled with a standard set of proce‐ dures and data to provide a base set of results. The system was then modeled with improved LU/LC inputs, and the results were compared with the base conditions. The GIRAS, NLCD, and MODIS LU/LC databases were swapped to eval‐ uate their effect on water balance and sediment components. For all model runs, the meteorological input data were the same. Goodness‐of‐fit coefficients, residual analysis, and graphical evaluation were used to assess the model results. STUDY AREA Luxapallila Creek is an upland, rural watershed in Ala‐ bama and Mississippi (fig. 1). Near the outlet, the watershed has a drainage area of 1,770 km2, an average basin slope of 2%, and average annual precipitation of 1,379 mm recorded at the Millport 2E weather station (1982‐2004 period).
Figure 1. Location of the Luxapallila Creek watershed.
140
TRANSACTIONS OF THE ASABE
LAND USE/LAND COVER DATA Three LU/LC databases were used in this research: BA‐ SINS GIRAS, NLCD, and NASA's MODIS MOD12Q1. Throughout this article, the MODIS MOD12Q1 dataset will be named MODIS, for brevity. The GIRAS LU/LC data, developed by USGS between 1977 and 1980, are provided in geographic coordinate units with a resolution of 0.0001 decimal degrees (USEPA, 1994). The original land use/land cover digital data were converted to ARC/INFO format by the USEPA. Although these datasets are useful for environmental assessment of land use patterns with respect to water quality analysis, growth management, and other types of environmental impact assessment, their use is limited because they are not current (USEPA, 1994). The 1992 NLCD set was produced by the Multi‐ Resolution Land Characteristics Consortium (MRLC), which consists of several federal agencies. The 1992 NLCD dataset was derived from Landsat - Thematic Mapper (TM) satellite data at a 30 m resolution and exists for the entire U.S. (USGS, 2004). A 21‐class modified Anderson land cover classification scheme was consistently applied to Landsat TM scenes for the entire country using an unsupervised clus‐ tering algorithm. The 1992 NLCD layer for the Mississippi region was obtained in .tif format and clipped to the Luxapal‐ lila watershed boundary. When the NLCD dataset was subset, an adequate buffer outside the watershed boundary was in‐ cluded in the subset to provide flexibility if a different wa‐ tershed delineation became necessary. BASINS requires that the LU/LC information be in a shapefile format; therefore, the subset raster files were converted to a polygon shapefile using ArcGIS's “raster to feature” option of the Spatial Ana‐ lyst extension. The “value” field in the image file, which rep‐
HSPF LU Classification
resents the LU/LC classification, was used to populate the “gridcode” field in the new shapefile. Finally, the resultant LU/LC shapefile was reprojected from Albers Conical Equal Area to Mississippi State Plane East. The MODIS Land Cover product (MOD12Q1) is publicly available in five types of land cover classifications with 1 km spatial resolution. The data are provided by NASA in integer‐ ized sinusoidal (ISIN) projection. For this research, MODIS' Land Cover Type 1, which includes 17 classes of land cover in the International Geosphere‐Biosphere Programme (IGBP) global vegetation classification scheme, was repro‐ jected and reclassified to meet the needs of the HSPF model for Luxapallila Creek. The original MODIS LU/LC data‐ cube (MOD12Q1.A2001001.h10v05.004.2004149203002. hdf) was obtained through NASA's Earth Observing System Data Gateway website. The metadata file associated with the data‐cube specifies a production date of 28 May 2004. The reprojection of the data‐cube to the State Plane, Mississippi East coordinate system was performed using the MODIS Re‐ projection Tool. During the reprojection phase, the original HDF format was converted to GEOTIFF for easier input of the data layer into ArcView. Then, the area of interest (Luxa‐ pallila watershed) was clipped using the BASINS shape‐file boundaries for Luxapallila. The clipped land use polygons were converted to shapefile format for introduction into BA‐ SINS. GIRAS, NLCD, and MODIS have different numbers of land use classes and terminology. For introduction into HSPF, those different LU/LC classifications were consolidated into six categories: forest land, urban or built‐up land, agricultural land, barren land, water, and wetlands. Table 1 and figure 2 show results of the LU/LC reclassification.
Table 1. Consolidation of GIRAS, NLCD, and MODIS LU/LC classification into HSPF land use classification. GIRAS LU/LC NLCD LU/LC MODIS LU/LC Classification Classification Classification
Forest land
Deciduous forest land Evergreen forest land Mixed forest land
Deciduous forest Evergreen forest Mixed forest
Evergreen needleleaf forest Evergreen broadleaf forest Deciduous needleleaf forest Deciduous broadleaf forest Mixed forests Woody savannas Closed shrublands Open shrublands
Agricultural land
Crop land and pasture Orchards, groves, vineyards, nurseries, and ornamental horticultural areas Confined feeding operations Other agricultural land
Pasture/hay Row crops Urban/recreational grasses
Savannas Grasslands Croplands Cropland/natural vegetation mosaic
Urban or built‐up land
Residential Commercial and services Industrial Transportation, communications, and utilities Industrial and commercial complexes Mixed urban or built‐up land Other urban or built‐up land
Low‐intensity residential High‐intensity residential, commercial, industrial, and transportation
Urban and built‐up
Wetlands
Forested wetland Nonforested wetland
Woody wetlands Emergent herbaceous wetlands
Permanent wetlands
Water
Stream and canals Lakes Reservoirs
Open water
Water
Barren land
Strip mines quarries and gravel pits Transitional areas
Bare rock/sand/clay Quarries/strip mines/gravel pits Transitional
Barren or sparsely vegetated
Vol. 51(1): 139-151
141
Figure 2. HSPF land use classification using NLCD, USGS‐GIRAS, and MODIS datasets.
BASINS/HSPF MODEL SETUP Spatial and climatic data, including topography, LU/LC, soil properties, reach characteristics, and detailed meteoro‐ logical data, were established using the BASINS/HSPF Arc‐ View interface. The topographic data used in the model setup was the U.S. Geological Survey (USGS) Digital Elevation Model (DEM). The DEM was used to delineate the watershed and subwatershed boundaries and generate the associated stream network (digitized streams). All geoprocessing opera‐ tions were performed using the toolkits provided by BA‐ SINS. During this process, land use areas and topographical parameters (overland plane slopes, streams slope and length, etc.) were summarized for export to HSPF's User Control In‐ put file. HSPF also requires a tabular characterization of streams geometry (FTABLE) with relationships among area, volume, and flow in a river cross‐section. These relationships are calculated by BASINS using the DEM and Manning's equation for steady uniform flow. Daily rainfall data were obtained from the National Weather Service (NWS) for the Sulligent and Millport 2E gauging stations (fig. 3). Hourly precipitation recorded at the Haleyville station was used to disaggregate the above cited stations. Hourly potential evapotranspiration, air tempera‐ ture, dew point, wind speed, solar radiation, evaporation, and cloud cover values were obtained from the Haleyville station. The weather database for the Haleyville station was down‐ loaded through the BASINS ArcView interface. BASINS' automatic delineation tool subdivided the Luxa‐ pallila watershed into ten subwatersheds or hydrologic response units (HRUs). Consequently, the channel network was divided into ten reaches. After delineation, the initial (not calibrated) HSPF model for Luxapallila was generated from within BA‐ SINS (fig. 4). The climatological database was processed inde‐ pendently using the WDMUtil software (also part of the BASINS suite) and then incorporated into the watershed data management file (.wdm) specific for Luxapallila. HYDROLOGY CALIBRATION The GIRAS land use dataset was used for flow calibration. Daily streamflow data, recorded at USGS gauging station
142
Figure 3. Location of Luxapallila water quality, streamflow, and weather stations.
02443500 at the outlet of the watershed (fig. 3), was compared to HSPF‐simulated streamflow at the same loca‐ tion. Table 2 shows the minimum, maximum, and coefficient of variation (CV) values for selected time periods of observed flow data. Hydrologic calibration was performed for the
TRANSACTIONS OF THE ASABE
Figure 4. Channel network conceptualized by HSPF. Table 2. Statistical characteristic of observed flow data. Maximum Minimum (m3 s‐1) (m3 s‐1) Period CV 1 Jan. 1985 to 30 Sept. 2003 1 Jan. 1985 to 31 Dec. 1993 1 Jan. 1994 to 30 Sept. 2003
1.7 2.0 1.5
897.6 897.6 549.3
0.7 0.7 1.0
period 1 January 1985 to 31 December 1993 because during this period flow data showed the highest CV value and widest max‐min range. Although an hourly time step was used for model calibration, results of the HSPF streamflow were pre‐ sented on a daily time step for comparison with the gauging station records. Flow calibration was performed by using an iterative procedure. Initial parameter values were chosen based on EPA BASINS Technical Note 6 (USEPA, 2000) ac‐ cording to the specific physiographic characteristics of Luxa‐ pallila watershed. HSPF's flow calibration guides were followed (USEPA, 2004). These guides provide a sequential strategy for determining which parameters are adjusted for flow calibration. Flow calibration consisted of adjusting the parameters that govern water balance, seasonal flows, and storm events. In this research, the HSPF parameters adjusted during cal‐ ibration were: lower zone nominal soil moisture storage (LZSN), infiltration capacity (INFILT), variable groundwa‐ ter recession (KVARY), base groundwater recession (AGWRC), fraction of groundwater inflow to deep recharge (DEEPFR), fraction of remaining evapotranspiration from baseflow (BASETP), fraction of remaining evapotranspira‐ tion from active groundwater (AGWETP), interception stor‐ age capacity (CEPSC), upper zone nominal soil moisture storage (UZSN), Manning's for overland flow (NSUR), inter‐ flow inflow parameter (INTFW), interflow recession param‐ eter (IRC), and lower zone evapotranspiration parameter (LZETP).
Vol. 51(1): 139-151
HYDROLOGY VALIDATION To increase the confidence in model simulation findings, a flow validation study was performed for the period 1 Janu‐ ary 1994 to 30 September 2003. Statistical characteristics of observed flow data for this period are showed in table 2. Dur‐ ing the validation period, the HSPF parameters were not ma‐ nipulated. The land use dataset and watershed delineation were the same as those used during the model calibration pe‐ riod. MODEL EVALUATION The generation and analysis of model simulation scenar‐ ios for watersheds (GenScn) software (Kittle et al., 2001) was used for evaluation of the HSPF outputs. Visual evaluation was performed using a scatterplot of observed and simulated streamflow data. The following numerical criteria were used to evaluate observed data versus simulated data by HSPF: the coefficient of determination (R2), the Nash‐Sutcliffe coeffi‐ cient (NS), and the relative error. The coefficient of determination (R2), which is the square of Pearson's product‐moment correlation coefficient, repre‐ sents the fraction of variability in y that can be explained by the variability in x. It ranges from zero to one, with higher val‐ ues indicating better agreement, and is given by: 2
⎧ ⎫ n ⎟ ⎟ − − (O j O)( S j S ) ⎟ ⎟ ⎟ ⎟ j =1 2 R =⎨ 0.5 0.5 ⎬ ⎤ ⎡n ⎤ ⎟ ⎟⎡n ⎟ ⎪ (O j − O) 2 ⎥ ⎪ ( S j − S ) 2 ⎥ ⎟ ⎥ ⎪ j =1 ⎥ ⎟ ⎟ ⎪⎣ j =1 ⎦ ⎣ ⎦ ⎭ ⎩
∑
∑
(1)
∑
where Oj is the observed streamflow at time step j, O is the average observed streamflow during the evaluation period, Sj is the simulated streamflow at time step j, and S is the average simulated streamflow at time step j.
143
The Nash‐Sutcliffe coefficient (NS) (Nash and Sutcliffe, 1970) represents the fraction of the variance in the measured data explained by the model. The NS ranges from minus in‐ finity to one. An NS value of one represents a perfect fit. The NS coefficient is considered one of the best statistical criteria for the evaluation of continuous‐hydrograph simulation pro‐ grams (Engelmann et al., 2002; Legates and McCabe, 1999). The NS is given by the following equation: n
∑ (O NS = 1 −
j
− S j )2
j =1 n
∑
(2) (O j − O j ) 2
j =1
The relative error for long‐term continuous simulation is given by the following equation: n
∑ Error (%) =
j =1
Oj − S j Oj n
* 100
(3)
A positive value in equation 3 implies that (on average) the model underpredicted flow, whereas a negative value im‐ plies overprediction of flow. SEDIMENT CALIBRATION HSPF's subroutines SEDMNT (production and removal of sediments) and SOLIDS (accumulation and removal of solids) simulate production/removal of sediments from per‐ vious and impervious areas, respectively. In this research, only parameters from SEDMNT were adjusted during cal‐ ibration due to the predominance of pervious areas in the Luxapallila Creek watershed. Sediment calibration was performed for the period 1 Janu‐ ary 1985 to 30 September 2003 following the BASINS/ WinHSPF sediment calibration guide (USEPA, 2006). The HSPF parameters included in the process were: management practice factor (SMPF), coefficient in the soil detachment equation (KRER), exponent in the soil detachment equation (JRER), daily reduction in the detached sediment (AFFIX), fraction land surface protected from rainfall (COVER), at‐ mospheric addition to sediment storage (NVSI), coefficient in soil matrix scour equation (KSER), exponent in soil matrix scour equation (JSER), coefficient in soil matrix scour equa‐ tion (KGER), and exponent in soil matrix scour equation (JGER). Calibration was accomplished on an annual basis by comparing simulated soil erosion rates to values reported in the literature (table 3). The erosion rates found by Grace (2004), Hairston et al. (1990), and Larson et al. (1985) were for nearly similar soil types as those found in the Luxapallila Creek watershed. Sediment validation was not performed in this project because of lack of observed data. SENSITIVITY OF CONSTITUENTS TO LAND USE DATABASE SWAPPING As stated previously in this study, the initial simulation scenario was set up using the GIRAS land use database. Streamflow and sediment model calibrations were subse‐ quently performed to provide a base set of results. Then, addi‐ tional simulation scenarios were conducted using improved land use datasets (NLCD and MODIS' MOD12Q1). The new
144
Table 3. Soil erosion rates observed in Alabama and elsewhere in the U.S. Soil Erosion Rate (ton ha‐1 year‐1)
Cover
Undisturbed forest land (Alabama) Agricultural land (Alabama) Cropland (Alabama, 1977) Natural (U.S.) Cultivated (U.S.) Barren (U.S.)
0.7 17.5 11.2 to 31.1 0.03 to 3.0 5.0 to 170.0 4.0 to 9.0
Source
Grace, 2004 Hairston et al., 1990 Larson et al., 1985 Morgan, 2005 Morgan, 2005 Morgan, 2005
simulation scenarios used the calibrated parameter values (obtained during calibration) and were run for streamflow and sediments. Simulated streamflow time‐series and total outflows of sediment (ROSED output in HSPF) were compared to those calculated with the base conditions. This comparison was performed at the watershed outlet. Sensitivity of simulations under different land use datasets was evaluated by calculating the mean, standard deviation, and coefficient of variation for each time series. Selected per‐ centiles (5th, 25th, 50th, 75th, and 95th) were estimated for each time series. Relative change analyses were also per‐ formed using the GIRAS dataset as a base line: Relative change = Constituent NLCD or MODIS − ConstituentGIRAS ConstituentGIRAS
× 100 (4)
where Constituent is the simulated streamflow (FLOW) or simulated sediment fraction (ROSED), and the subscripts de‐ note which land use database was used to obtain the simu‐ lated constituent. When generating a new HSPF model run (from within BASINS), the ArcView interface assigns “effective rainfall areas” according to the land use distribution specified by the land use database. Hence, the water balance in the watershed is affected by variations of land use acreage. To assess the ef‐ fects of land use database swapping in the water balance at the hillslope scale, relative changes on water balance compo‐ nents per land use class were also assessed using equation 4.
RESULTS LAND USE DATABASE CHANGES GIRAS' land use characterization of the Luxapallila wa‐ tershed attributes more than 90% land coverage to forest and agricultural lands (73% and 20%, respectively), with other land use categories (wetlands, urban, barren, etc.) accounting for the remaining 7% (table 4). Although NLCD and MODIS land use databases also show that more than 89% of the wa‐ tershed is covered by forest and agricultural land, percent dif‐ ferences exist in most land use classes, as shown in table 4. Most NLCD land use categories decrease in area (table 4). Only wetlands, barren land, and water show increments. Table 4 shows that forest and agricultural lands present low to moderate decreases in area (-0.5% and -18.9% respective‐ ly) and a substantial decrease in urban areas (-67.7%). How‐ ever, since urban areas account for less than 1% of the Luxapallila watershed area, this high relative change could not be expected to have big effects in the context of hydrolog‐ ical modeling and simulation. The same can be said about the
TRANSACTIONS OF THE ASABE
Table 4. Land use areas and relative change. Area (km2) and Percentage GIRAS
Relative Change (%)
HSPF Land Use Classification Forest land Agricultural land Urban or built‐up land (pervious) Urban or built‐up land (impervious) Wetlands Barren land Water
NLCD
MODIS
NLCD
MODIS
1291.3 347.5 9.3 9.3 109.2 1.8 1.8
73% 20%