SEARCH

SEARCH BY CITATION

Keywords:

  • SWAT;
  • NEXRAD;
  • TRMM 3B42;
  • precipitation data;
  • rain gauge data;
  • watershed;
  • rangeland;
  • time series analysis

Abstract

  1. Top of page
  2. Abstract
  3. Introduction
  4. Study Area
  5. Methodology
  6. Model Selection
  7. Data Preprocessing
  8. Model Set-up and Calibration
  9. Sensitivity Test
  10. Results
  11. Discussion
  12. Summary and conclusions
  13. Acknowledgments
  14. Literature Cited

Abstract:  Both ground rain gauge and remotely sensed precipitation (Next Generation Weather Radar – NEXRAD Stage III) data have been used to support spatially distributed hydrological modeling. This study is unique in that it utilizes and compares the performance of National Weather Service (NWS) rain gauge, NEXRAD Stage III, and Tropical Rainfall Measurement Mission (TRMM) 3B42 (Version 6) data for the hydrological modeling of the Middle Nueces River Watershed in South Texas and Middle Rio Grande Watershed in South Texas and northern Mexico. The hydrologic model chosen for this study is the Soil and Water Assessment Tool (SWAT), which is a comprehensive, physical-based tool that models watershed hydrology and water quality within stream reaches. Minor adjustments to selected model parameters were applied to make parameter values more realistic based on results from previous studies. In both watersheds, NEXRAD Stage III data yields results with low mass balance error between simulated and actual streamflow (±13%) and high monthly Nash-Sutcliffe efficiency coefficients (NS > 0.60) for both calibration (July 1, 2003 to December 31, 2006) and validation (2007) periods. In the Middle Rio Grande Watershed NEXRAD Stage III data also yield robust daily results (time averaged over a three-day period) with NS values of (0.60-0.88). TRMM 3B42 data generate simulations for the Middle Rio Grande Watershed of variable qualtiy (MBE = +13 to −16%; NS = 0.38-0.94; RMSE = 0.07-0.65), but greatly overestimates streamflow during the calibration period in the Middle Nueces Watershed. During the calibration period use of NWS rain gauge data does not generate acceptable simulations in both watersheds. Significantly, our study is the first to successfully demonstrate the utility of satellite-estimated precipitation (TRMM 3B42) in supporting hydrologic modeling with SWAT; thereby, potentially extending the realm (between 50°N and 50°S) where remotely sensed precipitation data can support hydrologic modeling outside of regions that have modern, ground-based radar networks (i.e., much of the third world).


Introduction

  1. Top of page
  2. Abstract
  3. Introduction
  4. Study Area
  5. Methodology
  6. Model Selection
  7. Data Preprocessing
  8. Model Set-up and Calibration
  9. Sensitivity Test
  10. Results
  11. Discussion
  12. Summary and conclusions
  13. Acknowledgments
  14. Literature Cited

There is much debate over how precipitation can be most accurately measured and there is no clear consensus on which method yields the best results, especially in the context of providing input for hydrological modeling. Traditionally, precipitation quantification has been based on rain gauge measurements. In the last decade remote-sensing technologies such as ground-based National Weather Service (NWS) Next Generation Weather Radar (NEXRAD; e.g., Di Luzio and Arnold, 2004; Moon et al., 2004; Kalin and Hantush, 2006) and satellite platforms, most notably the Tropical Rainfall Measurement Mission (TRMM; e.g., Milewski et al., 2007), have been increasingly utilized to support spatially distributed hydrologic modeling. Each of these approaches to the measurement of precipitation have potential issues and drawbacks. For example, tipping bucket rain gauge data are used to adjust raw NEXRAD data for local biases through a Kalfam filter algorithm (Smith and Krajewski, 1991) in the development of the NEXRAD Stage III precipitation product that provides precipitation estimates for the specific region covered by a NWS River Forecast Center. However, rain gauge measurements can exhibit significant bias, particularly in high wind conditions that generate aerodynamic lift localized over the mouth of the gauge (e.g., Sevruk, 1989). Additionally, tipping bucket rain gauges, used by the NWS to obtain hourly precipitation data, consistently underperform during intense precipitation events (e.g., Duchon and Essenberg, 2001; Tokay et al., 2003). Finally, the question of how representative rain gauge data, with gauges separated by tens of kilometers, can be in terms of providing a representative estimate of precipitation at the watershed scale remains unanswered (Rodrigrez-Iturbe and Mejia, 1974; Morrissey et al., 1995). A major limitation that impedes accurate modeling of the hydrology in many regions of the world is the lack of precipitation data that are representative at the watershed scale. There is a clear need for realistic spatially distributed precipitation estimates to support hydrological modeling applications.

Remote-sensing technologies potentially provide a solution to developing spatially distributed estimates of precipitation at the watershed scale. While NEXRAD Stage III precipitation data provide a reasonable depiction of the spatial extent of precipitation, there are significant problems with assessing precipitation intensities on a pixel-by-pixel basis and matching specific radar grid cells with point (rain gauge) measurements (e.g., Zawadski, 1975; Krajewski, 1987; Kitchen and Blackall, 1992; Smith et al., 1996; Ciach and Krajewski, 1999; Krajewski and Smith, 2002). However, NEXRAD has been able to provide reasonable estimates of average precipitation at the watershed scale in areas with good radar coverage (e.g., Smith et al., 1996; Bedient et al., 2000; Xie et al., 2005). Obviously, localities with limited radar coverage, such as those near the edge of the coverage envelopes of the radar systems present in a region, will have erroneous NEXRAD precipitation estimates (Harrison et al., 2000). Satellite-based radar missions, such as TRMM or the planned Global Precipitation Mission (GPM), Huffman et al. (2007), do not have the line-of-sight limitations of ground-based radar systems. Today, TRMM estimates of precipitation (3B42 product) are available at both fine spatial (0.25° × 0.25°) and temporal (three hours) scales (Huffman et al., 2007). However, coverage is not continuous, with significant gaps in the current availability of three-hourly, high-quality, precipitation estimates from satellite-based microwave sensors; this is a problem that will be addressed by the future GPM. Therefore, no precipitation measurement is without inherent biases.

Our approach is to utilize and compare the performance of NWS rain gauge, NEXRAD Stage III, and TRMM 3B42 data in the hydrological modeling of the Middle Nueces River Watershed in South Texas and Middle Rio Grande Watershed in South Texas and northern Mexico. The hydrologic model chosen for this study is the Soil and Water Assessment Tool (SWAT), which is freely available for download and is a comprehensive tool that besides modeling watershed hydrology can simulate water quality as well as nutrient and sediment loading within stream reaches (Arnold and Fohrer, 2005; Gassman et al., 2007). The results of this study can be directly linked to the status of regional water resources and can serve as a foundation for the characterization of the overall ecological health in the region. The approaches and methodologies developed in this project have a potential for adoption in other regions of the world that have limited ground-based precipitation measurements and could benefit from remotely sensed precipitation data. In this paper, we will describe the procedures used to set-up and calibrate the SWAT model rain gauge, NEXRAD Stage III, and TRMM 3B42 data. Analysis of input precipitation data and simulated streamflow results using three types of precipitation data are provided. Finally, included is a discussion of the overall significance of this study both in terms of highlighting the ability of SWAT to simulate variable climatic conditions (wet versus dry periods) and the utility of remotely sensed precipitation data in supporting semi-distributed hydrological modeling.

Study Area

  1. Top of page
  2. Abstract
  3. Introduction
  4. Study Area
  5. Methodology
  6. Model Selection
  7. Data Preprocessing
  8. Model Set-up and Calibration
  9. Sensitivity Test
  10. Results
  11. Discussion
  12. Summary and conclusions
  13. Acknowledgments
  14. Literature Cited

The hydrometeorological patterns of the South Texas region have been described in numerous publications, but perhaps most recently in a comprehensive manner as part of the Water For Texas – 2002 statewide water planning project (Texas Water Development Board, 2002; Rio Grande Regional Water Planning Group, 2001). This region ranges from subtropical sub-humid (near Gulf coast) to subtropical steppe. Prevailing winds carry moisture north from the Gulf of Mexico to the eastern side of the region (subtropical sub-humid) throughout much of the year, resulting in hot and humid summers and mild winters. The eastern side of the study area has annual rainfall averages of about 700 mm; the July maximum temperatures range from about 36-37°C and January minimum temperatures range from about 4-9°C. Annual lake evaporation rates in the coastal zones of the South Texas region range from 1,000 to 1,100 mm (Rio Grande Regional Water Planning Group, 2001). The western side of the study region (subtropical steppe) is more strongly dominated by the North American Monsoon (e.g., Higgins et al., 1999); summers are hotter than along the coast, and months of maximum precipitation are usually June through September. Average precipitation ranges from about 381-559 mm and lake evaporation from 1,524 to 1,626 mm. For example, Texas temperatures can reach 45°C in July in Laredo. In regions of northern Mexico, adjacent to the Rio Grande, climatic conditions are similar to those in South Texas with a subtropical steppe climatic regime.

The Middle Nueces River was chosen for this study because of its location; straddling the transition between the subtropical sub-humid to subtropical steppe climatic regimes described above. Additionally, the Nueces River empties into Corpus Christi Bay, which is an ecologically sensitive estuary where release inputs from upstream derived freshwater have to be managed carefully. Consequently, developing a better understanding of the hydrology of the Middle Nueces Watershed can aid in effective water resources planning and management in the region. The Middle Nueces River Watershed has United States Geological Survey (USGS) stream gauges at both the inlet (Cotulla, Texas; Id No. 08194000) and outlet (Tilden, Texas; Id No. 08914500) of the watershed (Figure 1a). This watershed occupies an area of 7,720 km2. Terrain in the watershed consists of subdued ridges surrounding the central valley of the Nueces River (Figure 1b). Elevation generally increases from east to west; however, maximum elevation occurs in the extreme southern side of the watershed (280 m); with a minimum elevation of 66 m at the watershed outlet. The predominant land cover consists of rangeland, most of which is covered by scrub and brush (RNGB – 78%) with subdominant grassy rangeland (RNGE – 18%). Minor areas with agricultural (AGRL – 2%) and various urban, forest, and water (total = 1%) land covers are also present (Figure 1c). The dominant activity that impacts the landscape in the region is ranching. Soils in the Middle Nueces Basin are defined based on State Soil Geographic Database (STATSGO) series from the United States Department of Agriculture (USDA, 1994), which are highly variable across the watershed. Major soils present in Middle Nueces Basin are described below and are shown in Figure 1d. In the east, the sandy and silty loam soils of the Aguilares (TX 004) and Houla (TX 234) series dominate with a moderate to very high hydraulic conductivity (1-10s mm/h). In the south-central Middle Nueces Basin the Montell series (TX 349) and in the north-central part of the basin the Cotulla series (TX 115) are present and both of these soils are dominated by clay and have low hydraulic conductivity (< 1 mm/h). To the west, the sandy soils of the Duval series (TX 149) dominate with a very high hydraulic conductivity (16-73 mm/h). Finally, the main channel and floodplain of the Nueces River is underlain by the clay-rich Coquat series (TX 114) with a low hydraulic conductivity (0.26-0.38 mm/h) and upstream tributaries are associated with the sandy Brundage series (TX 078) with a much higher hydraulic conductivity (16-61 mm/h).

image

Figure 1.  Middle Nueces Watershed: (a) Watershed Map that Shows 25 Subbasins and Related Hydrographic Features. Indicated is the spatial distribution of NWS precipitation and temperature stations (triangle) and centroid of TRMM grid cells (squares). Inset map showing location of the Middle Nueces Watershed in Texas. (b) Digital Elevation Map. (c) Land Use/Land Cover Map With Hydrography. (d) Soil Map (STATSGO) With Hydrography. Major soils include the Aguilares (TX 004), Brundage (TX 078), Coquat (TX 114), Cotulla (TX 115), Duval (TX 149), Houla (TX 234), and Montell series (TX 349), which are defined in USDA (1994).

Download figure to PowerPoint

The Middle Rio Grande River Watershed (Figure 2) is a region of rich cultural and environmental diversity, but with increasing levels of environmental stress in recent years from a rapidly expanding human population. The City of Laredo, Texas, for example, grew by nearly 60,000 (more than 30%) between 1990 and 2000. The entire Rio Grande River Watershed population increased during that same period by more than 87,000, or 25%. The population in this region is expected to be more than double over the next 50 years, from about 1.3 million people to about 3.1 million in the year 2050. The population in the Mexico portion of the Rio Grande Watershed is expected to increase from about 1.8 million to about 3.7 million by the year 2020 (Rio Grande Regional Water Planning Group, 2001). This rapid population growth has already resulted in critical regional environmental problems related to water supply, wastewater treatment, point and nonpoint source pollution, and general degradation of the natural environment.

image

Figure 2.  Middle Rio Grande Watershed: (a) Watershed Map That Shows 25 Subbasins and Related Hydrographic Features. Indicated is the spatial distribution of NWS precipitation and temperature stations (triangle) and centroid of TRMM grid cells (squares). Inset map showing location of the Middle Rio Grande Watershed in Texas and Northern Mexico. (b) Digital Elevation Map With Hydrography. (c) Land Use/Land Cover Map. (d) Soil Map (U.S. – STATSGO/Mexico – 2006 WRB for Soil Resources) With Hydrography. On the U.S. side of the basin major soils include the Copita (TX 113), Maverick (TX 320), and Rio Grande series (TX 471), which are defined in USDA (1994). In Mexico dominate soils are userdefined based on the 2006 World Reference Base (WRB) for Soil Resources (International Union of Soil Sciences, 2006) and include Calcisols and Regosols (MEX 15, 16, 21) and Leptosol (MEX 4).

Download figure to PowerPoint

The Middle Rio Grande River Watershed has International Boundary and Water Commission (IBWC) stream gauges on the Rio Grande River at both the inlet (El Indio, Texas and Villa Guerrero, Coahuila; Id No. 08-4587) and outlet (Laredo, Texas and Nuevo Laredo, Tamaulipas; Id No. 08-4590) of the watershed (Figure 2a). This watershed occupies an area of 8,905 km2. Terrain in the watershed consists of the foothills of the Sierra Madre Oriental range to the northwest (820 m) with lower elevations in the valley along the Rio Grande River (108 m) in the middle of the watershed (Figure 2b). Like the Middle Nueces Watershed the predominant land cover consists of rangeland, most of which is covered by scrub and brush (RNGB – 71%) with subdominant grassy rangeland (RNGE – 22%). Minor areas with agricultural (AGRL − 4%) and various urban, forest, and water (total = 2%) land covers are also present (Figure 2c). Note that land use/land cover classifications are contiguous across the international boundary. Soil type varies across the basin; below is a brief discussion of the major soils present in the Middle Rio Grande Basin as depicted in Figure 2d. On the U.S. side of the watershed soils are defined based on the STATSGO database (USDA, 1994). Dominant soils on the U.S. side of the watershed include the Maverick (clay/silt dominated with low hydraulic conductivity < 1 mm/h; TX 320) and Copita (sand dominated with higher hydraulic conductivity = 17-290 mm/h; TX 113) series. The immediate valley of the Rio Grande River is underlain by the Rio Grande series (TX 471), which is dominated by silt and has a moderately low hydraulic conductivity (4-6 mm/h). For Mexican soils, polygons from the earlier 1974 Food and Agriculture Organization of the United Nations Educational, Scientific and Cultural Organization (FAO/UNESCO) soil map were reclassified based on the newer 2006 World Reference Base (WRB) for Soil Resources (International Union of Soil Sciences, 2006) and for each unique combination of soils within a polygon an alpha-numeric designation (e.g., MEX 4) was assigned. Calcisols and Regosols soils that tend to be sandy and have high hydraulic conductivities dominate the Mexican side of the watershed (MEX 15, 16, 21). To the north adjacent to the foothills of the Sierra Madre Oriental, Leptosol soils dominate and have higher clay content and low measured hydraulic conductivities (e.g., MEX 4).

Methodology

  1. Top of page
  2. Abstract
  3. Introduction
  4. Study Area
  5. Methodology
  6. Model Selection
  7. Data Preprocessing
  8. Model Set-up and Calibration
  9. Sensitivity Test
  10. Results
  11. Discussion
  12. Summary and conclusions
  13. Acknowledgments
  14. Literature Cited

Table 1 summarizes the data sources and formats utilized in this study. Core data from the U.S. associated with this project were downloaded using the application Better Assessment in Science Integrating Point and Nonpoint Sources (BASINS 3.1; USEPA 2007). BASINS has a graphic user interface that allows one to graphically select a watershed (Middle Nueces Watershed, Hydrologic Unit Code (HUC) – 12110105; Middle Rio Grande Watershed – HUC 13080002) and interactively select data sources for download using the Data Download Tool. Downloaded core data includes a digital elevation model [USGS 300 Meter Resolution, 1-Degree Digital Elevation Model (DEM) for the Continental United States; USGS, 1995], hydrography layer (National Hydrography Dataset, high resolution, 1:100,000; USGS, 1999), soil layer [STATSGO, 1:250:000; United States Department of Agriculture (USDA, 1994)], land use layer (Land use/land cover; Geographic Information Retrieval and Analysis System; GIRAS dataset, 1:250,000; USGS, 1986), and USGS stream gauge locations. Additionally, both USGS surface runoff data (USGS, 2008) and IBWC surface runoff data (IBWC, 2008) were downloaded as tab-separated text files.

Table 1.   Data Type and Format Used in Study.
Data TypeData Source
Soil – U.S.BASINS Data Download Tool
Soil – MexicoIGENI Website
Land use – U.S.BASINS Data Download Tool
Land use – MexicoIGENI Website
Elevation – U.S.BASINS Data Download Tool
Elevation – MexicoIGENI Website
Hydrography – U.S.BASINS Data Download Tool
Hydrography – MexicoIGENI Website
Streamflow – NuecesUSGS Website
Streamflow – Rio GrandeIBWC Website
TemperatureNOAA National Climate Data Center Website
NWS Rain GaugeNOAA National Climate Data Center Website
TRMM 3B42Giovanni Website
NEXRAD 12ZNOAA Southern Division Precipitation Analysis Website
NEXRAD 05-06ZWest Gulf Coast River Forecast Center

The middle Rio Grande Watershed is a bi-national watershed occupying both the U.S. and Mexico. Data sources from Mexico were obtained from the Instituto Nacional de EstadÌsticas, GeografÌa, e Informática website (INEGI, 2008). Specific data layers obtained from INEGI included DEM (1:50,000), vegetation type (1:1,000,000; land use/land cover), and soil type (1:1,000,000). Data integration of U.S. and Mexican DEMs was facilitated in ArcGIS 9.1 utilizing the mosaic tool combining these raster layers seamlessly together. On the U.S. side, relatively low-resolution DEM, land use, and soil (STATSGO) inputs were selected that correspond most closely to the data resolution from Mexico facilitated seamless integration of data for the bi-national Rio Grande Watershed.

Daily temperature and rain gauge precipitation data were downloaded from the U.S. National Climate Data Center (NCDC, 2008). Di Luzio and Arnold (2004) indicate that the use of temperature data can significantly improve hydrologic simulations. All stations in proximity to the Nueces and Rio Grande Watersheds were examined for completeness during the period of study (1998-2006) and only stations that had limited or no missing values were selected (Figures. 1 and 2). The SWAT website (Soil and Water Assessment Tool, 2008) provides a downloadable applet (Extract), which facilitates the conversion of NCDC temperature and precipitation data into a dbase IV file format with daily data in units of °C (for temperature) and millimeters (for precipitation), which can be imported into the SWAT model.

The satellite-estimated precipitation data used in this study is the TRMM 3B42 (Version 6, Archived – Research Quality) product, which is a fine spatial (0.25° × 0.25°) and temporal (three hours) resolution product discussed in detail by Huffman et al. (2007). All available satellites (both microwave and infrared-based) were used to estimate precipitation in association with the current TRMM mission which derived its data from the TRMM Microwave Imager (TMI) and Precipitation Radar (PR). The TRMM Combined Instrument (TCI) dataset reflects a combination of data from both TMI and PR. There has been a move towards using a combination of orbital sensors to determine satellite-estimated precipitation as reflected by the planning for the future GPM mission (Huffman et al., 2007). When data from the TRMM mission are combined with data from multiple satellite platforms, the composite dataset is described as Tropical Rainfall Measuring Mission Multi-Satellite Precipitation Analysis (TMPA). TMPA estimated precipitation is calculated in four distinct stages as described by Huffman et al. (2007):

  • 1
    Microwave precipitation data are calibrated and combined.
  • 2
    Infrared precipitation data are created using calibrated microwave precipitation.
  • 3
    Microwave and infrared estimates are combined.
  • 4
    Ground rain gauge data are added (only for the archived version of the product).

However, this product is not perfect because satellite coverage is not continuous, with significant gaps in the current availability of three-hourly active microwave precipitation estimates, which provides the highest quality data for the merged TRMM 3B42 product. TRMM 3B42 data were acquired using the Goddard Earth Sciences Data and Information Services Center (GES-DISC) through the GES_DISC Interactive Online Visualization ANd aNalysis Infrastructure (Giovanni; National Aeronautics and Space Administration; NASA, 2008) in a flat text format. Data for each TRMM grid cell (Middle Nueces Watershed– 27.50-29.00°N; 98.50-100. 00°W; Figure 1a; Middle Rio Grande Watershed – 27.75-28.50°N; 99.75-100. 75°W; Figure 2a), for the period of 1998-2007, were organized using Visual Basic and written to a digital text file using the beginning date of the time series as a header and precipitation values (mm) sequentially listed for each date below the header in one column. Data were downloaded on a daily basis with three different daily definitions with a day that begins at 00Z (Greenwich Mean Time midnight), 06Z (local midnight), or 12Z (early morning, time of measurement for daily rain gauge data) to evaluate the impact of this definition on model results. Daily NWS rain gauge data are not collected at local midnight but at 5-6 am local time (12Z) unlike daily USGS surface runoff data, which have a day beginning at local midnight. Consequently, an examination of the impact on how different definitions of a day affect model results derived using TRMM precipitation data is warranted. The accuracy of data in all TRMM input files was confirmed by comparison of annual precipitation calculated in these files with results from a query to Giovanni for the same annual period.

The third source of daily precipitation data used in this study was NEXRAD Stage III data. Two types of NEXRAD data were obtained in which the beginning of a day was defined as 12Z or 05-06Z. NEXRAD 12Z data are easily downloaded through the National Oceanographic and Atmospheric Agency (NOAA) Southern Division Precipitation Analysis website (NOAA, 2008) in a zipped shapefile format. Data were processed with ArcGIS 8.3 with the centroid of each NEXRAD grid cell represented by a point in both latitude/longtitude and HRAP (Hydrologic Rainfall Analysis Project) coordinates. Processing in ArcGIS 8.3 and Visual Basic combined all NEXRAD points (total 46-53 per TRMM grid cell) from each of the TRMM grid cells with the final product of a single digital file consisting of the daily average of NEXRAD precipitation values within each TRMM grid cell. Similarly, NEXRAD 05-06Z data (in which a day was based on local midnight) were obtained from the West Gulf Coast River Forecast Center as XMRG files and these files were organized and processed using C++ (see Xie et al., 2005) and Visual Basic producing a single file for each TRMM grid cell containing the daily average of NEXRAD precipitation values. Finally, all NEXRAD 05-06Z and 12Z data files were cross-checked for accuracy.

Model Selection

  1. Top of page
  2. Abstract
  3. Introduction
  4. Study Area
  5. Methodology
  6. Model Selection
  7. Data Preprocessing
  8. Model Set-up and Calibration
  9. Sensitivity Test
  10. Results
  11. Discussion
  12. Summary and conclusions
  13. Acknowledgments
  14. Literature Cited

The computational abilities of modern desktop computers can support geographic information systems (GIS) in tandem with complex deterministic hydrologic models (see Ogden et al., 2001) using spatially distributed precipitation products, such as NEXRAD and TRMM data, as the primary input variables. Many watershed-scale models have been developed over the last few decades including Agricultural Nonpoint Source Pollution Model (AGNPS; Young et al., 1989), Areal Nonpoint Source Watershed Environmental Response Simulation (ANWSERS; Beasley et al., 1980), Simulator for Water Resources in Rural Watersheds (SWRRB; Arnold et al., 1990), and SWAT (Arnold et al., 2005; Gassman et al., 2007). SWRRB and SWAT are continuous as opposed to single event hydrologic models and therefore facilitate time series analysis. In the U.S., SWAT, a semi-distributed hydrologic model, has been developed into an integrated watershed management tool and is supported by a robust array of GIS data sources (DEMs, land use, soil type, hydrography). The SWAT model was selected for this study due to its ability to simulate the hydrology of large watersheds through physically based parameterization (Arnold et al., 2005) and demonstrated global applications (Gassman et al., 2007). SWAT has been incorporated into a BASINS interface (Di Luzio et al., 2002) to support analysis of water quality by state regulatory agencies. SWAT supports not only physical hydrologic modeling but also characterizes sediment, nutrient, and pesticide transport supporting quantification of Total Maximum Daily Loads within specific stream reaches. SWAT has been successfully validated as a hydrologic and water quality modeling tool at the watershed scale (Gassman et al., 2007). A major advantage of SWAT in the current effort is its capability to import precipitation data from numerous discrete locations like NEXRAD and TRMM grid cells. While, to date, there are no studies that have used TRMM data to support SWAT modeling there have been a number of studies that have successfully utilized NEXRAD Stage III precipitation data with SWAT at the watershed scale (Di Luzio and Arnold, 2004; Moon et al., 2004; Kalin and Hantush, 2006). Finally, SWAT 2003, the version of the model used in this study, has built-in tools for calibration and model sensitivity analysis supporting objective comparison between modeled and observed results.

Data Preprocessing

  1. Top of page
  2. Abstract
  3. Introduction
  4. Study Area
  5. Methodology
  6. Model Selection
  7. Data Preprocessing
  8. Model Set-up and Calibration
  9. Sensitivity Test
  10. Results
  11. Discussion
  12. Summary and conclusions
  13. Acknowledgments
  14. Literature Cited

Geographic information systems data for the SWAT model were preprocessed by two separate functions [watershed delineation and determination of hydrologic response units (HRUs)]. The automatic watershed delineation tool was selected, which allows the user to set-up and preprocess the DEM for modeling, define the threshold area for the subbasins in the SWAT model, modify outlet and inlet definitions for the watershed, and define the main watershed outlet. Additional options used in this study include burning (superimposing) the digitized network of streams (NHD hydrography layers) on to the DEM to ensure reach continuity and preprocessing the DEM to remove sinks. The next major function in the automatic watershed delineation tool is an option to set a threshold area for the suggested area of subbasins within the watershed (Middle Nueces – 17,500 ha; Middle Rio Grande – 20,000 ha). The program suggested threshold area provides a good trade-off between generating a detailed spatial delineation of watershed processes and software/hardware limitations in dealing with such a complex model. Above a reasonable level of subbasins, streamflow is not seriously affected by subbasin size (Gassman et al., 2007) and therefore there is no reason to increase the number of subbasins excessively. Additionally, FitzHugh and MacKay (2000) also recommended that the user remove small spurious subbasins during the watershed delineation. The net result of our delineation is the definition of 25 subbasins that were used in subsequent modeling of both watersheds.

The Landuse and Soil Layer definition tool merges land use and soil data within each subbasin to determine HRUs. A HRU is a unique combination of land use and soil layers that SWAT subsequently uses in defining the hydrologic characteristics of the subbasins within a watershed. Final delineation of HRUs for this study was accomplished using the HRU Distribution tool. This tool allows thresholds to be established for both land use and soil type within a given subbasin, which prevents creation of HRUs that are smaller than the thresholds and eliminates their incorporation in the spatially weighted average of subbasin hydrologic parameters. This tool prevents the creation of numerous tiny (<< 1%) HRUs and results in increased model computational efficiency. In this study, a threshold for land use of 10% and soil type of 0% was applied, consistent with the premise that soil type is more important than land use in defining the hydrologic character of a subbasin (Di Luzio and Arnold, 2004).

For modeling of the Middle Rio Grande Watershed a special customized soil database was developed combining the SSURGO database from the U.S. with information from the WRB for soil resources in Mexico. From the WRB database the average values of the following parameters (sand, silt, clay, rock fragment, available soil moisture, bulk density) were obtained for each soil horizon for each unique soil combination in a Mexico soil polygon. For permeability, the average composition of sand, silt, and clay was used to determine sediment type and hydraulic conductivity (Clapp and Hornberger, 1978) within each unique soil polygon.

Model Set-up and Calibration

  1. Top of page
  2. Abstract
  3. Introduction
  4. Study Area
  5. Methodology
  6. Model Selection
  7. Data Preprocessing
  8. Model Set-up and Calibration
  9. Sensitivity Test
  10. Results
  11. Discussion
  12. Summary and conclusions
  13. Acknowledgments
  14. Literature Cited

A series of six SWAT simulations, identical except for type of daily precipitation input (NWS Rain Gauge, TRMM 3B42 00Z, TRMM 3B42 06Z, TRMM 3B42 12Z, NEXRAD 06Z, NEXRAD 12Z) for the Middle Nueces Watershed and three simulations (NWS Rain Gauge, TRMM 3B42 06Z, NEXRAD 06Z) for the Middle Rio Grande Watershed were completed. Simulations were conducted from January 1, 1998 to December 31, 2007 (November 30, 2007 for Rio Grande Watershed). Precipitation data from January 1, 1998 to June 30, 2003 were used to initialize the model (warm-up period) with the calibration period spanning the period from July 1, 2003 to December 31, 2006. The warm-up period includes nearly two years (August 24, 2001-June 30, 2003) of TRMM data that postdates the boost in orbit of the TRMM satellite, which occurred during August 2001. The change in orbit affected the swath geometry of high quality microwave precipitation quantification from the TRMM satellite. Consequently, a warm-up period with a significant period (nearly two years) of TRMM 3B42 data reflecting the new orbital configuration was selected. The validation period selected spans one full year (2007) in the Middle Nueces Watershed; 11 months, January-November 2007, in the Middle Rio Grande Watershed. Both calibration and validation periods have significant periods (duration of at least four months) with both dry and wet climatic conditions.

Adjustments to selected parameters in the SWAT model databases (Table 2) were made to more realistically represent the values of the parameters in the study area based on previous studies. Surface Runoff Lag Coefficient (SURLAG) affects the timing of modeled surface runoff and our selected value is appropriate for larger watersheds that may have responded more slowly to precipitation events (Neitsch et al., 2002) and provided the best temporal fit of preliminary modeled surface runoff with measured surface runoff at the watershed outlet. Adjusted Manning values for overland flow, tributary channels, and the main channel were consistent with values previously used and recommended for rangeland settings (Hanratty and Stefan, 1998; Neitsch et al., 2002). The maximum canopy interception (CANMX) was determined based on a similar study in the nearby Texas Hill County (Afinowicz et al., 2005). The hydraulic conductivity of tributary and main channel alluvium (CH_1, CH_2) were determined based on the average hydraulic conductivity of soils in the STATSGO soil database that underlie these features. The Baseflow Alpha Factor (ALPHA_BF) was calculated based on results from a baseflow filter program available from the SWAT website (Arnold et al., 1995; Arnold and Allen, 1999). The Soil Evaporation Compensation Factor (ENSO) and major groundwater parameters (GWQMN, GW_REVAP, REEVAPMN) were also adjusted to achieve a better fit between actual and simulated streamflow, especially during dry periods.

Table 2.   SWAT Model Parameters.
Parameter TypeModel ParameterVariableParameter Values
BasinSurface runoff lag coefficientSURLAG0.50
HRUSoil evaporation compensation factorENSO0.01
HRUManning’s “n” value for overland flowOV_N0.60
HRUMaximum canopy interceptionCANMX7.0 mm
HRUAverage slope steepnessSLOPENo adjustment
SubbasinHydraulic conductivity of tributary channel Alluvium – Nueces & Rio GrandeCH_K1 20.0 mm/h 100.0 mm/h
SubbasinManning’s “n” value for tributary channelCH_N(1)0.10
RoutingManning’s “n” value for main channel – Nueces & Rio GrandeCH_N(2)0.05 0.014
RoutingHydraulic conductivity of main channel alluvium – Nueces & Rio GrandeCH_K20.27 mm/h 1.0 mm/h
GroundwaterBaseflow alpha factor – Nueces & Rio GrandeALPHA_BF0.0352 0.0167
GroundwaterThreshold depth of water in the shallow aquifer required for return flow to occurGWQMN5000 mm
GroundwaterGroundwater “Reevap” coefficientGW_REVAP0.20
GroundwaterThreshold depth of water in the shallow aquifer for “Reevap” or percolation to the deep aquifer to occurREEVAPMN0

Key options used in the SWAT model included: (1) an estimation of the surface runoff using the NRCS runoff curve method approach with crack flow not active (Soil Conservation Service Engineering Division, 1986), (2) calculation of daily potential evapotranspiration values using the Priestly-Taylor method (Neitsch et al., 2002), and (3) variable storage method (Williams, 1969).

Sensitivity Test

  1. Top of page
  2. Abstract
  3. Introduction
  4. Study Area
  5. Methodology
  6. Model Selection
  7. Data Preprocessing
  8. Model Set-up and Calibration
  9. Sensitivity Test
  10. Results
  11. Discussion
  12. Summary and conclusions
  13. Acknowledgments
  14. Literature Cited

An advantage of SWAT is its incorporation of a number of tools for calibration and sensitivity analysis that allow the user to objectively compare model results with actual data (Van Grinsven, 2002). A total of 27 model parameters were adjusted in each sensitivity test using the Latin Hypecube (LH) One-factor-At-a-Time (OAT) method (for details, see Van Grinsven and Meixner, 2003).

Results

  1. Top of page
  2. Abstract
  3. Introduction
  4. Study Area
  5. Methodology
  6. Model Selection
  7. Data Preprocessing
  8. Model Set-up and Calibration
  9. Sensitivity Test
  10. Results
  11. Discussion
  12. Summary and conclusions
  13. Acknowledgments
  14. Literature Cited

Comparison of Precipitation Data

To better understand how input precipitation can impact model results a comparison of the three precipitation types that have been used in subsequent SWAT modeling were compared. For the Middle Nueces Watershed TRMM 3B42 and NEXRAD Stage III data were compared at a monthly time scale within each TRMM grid cell; comparisons were not made with official NWS Rain Gauge data because the gauges were not located in the watershed (Figure 3a). For the middle Nueces Watershed, there was a moderately robust linear correlation between TRMM 3B42 and NEXRAD Stage III data (R2 = 0.564-0.821) in all TRMM grids and a positive bias towards TRMM 3B42 data indicated by a positive y-intercept (9-15 mm) in all TRMM grid cell comparisons. Total rainfall recorded in the Middle Nueces Basin was biased toward TRMM 3B42 data in nine out of eleven grid cell comparisons. The bias of TRMM 3B42 data ranges from −2 to 17% (July 2003-December 2007). These results were consistent with previous results (e.g., Bocchiola, 2007) that document a positive bias for TRMM (PR and TMI) over NEXRAD precipitation estimates.

image

Figure 3.  Scatterplot of Monthly Precipitation Comparing NEXRAD Stage III and TRMM 3B42 Data for (a) Middle Nueces Basin, Grid Cell (28.25°N-28.50°N, 99.00°W-99.25°W) and (b) Middle Rio Grande Basin, Grid Cell (27.50°N-27.75°N, 99.75°W-99.50°W), Which is the Same Grid Cell in Which NWS Rain Gauge from Laredo is Located. Regression equation and R2 values for each scatterplot are indicated.

Download figure to PowerPoint

Additionally, data were analyzed based on wet versus dry periods. A wet period was defined as a period with at least two months with a watershed-wide average of NEXRAD stage III precipitation above 75 mm for a minimum period duration of four months. For the Middle Nueces Basin, this definition resulted in a total of four wet (July-November 2003; April-October 2004; September-December 2006; May-August 2007) and four dry periods (December 2003-March 2004; November 2004-August 2006; January-April 2007; September-December 2007). During wet periods, the correlation between TRMM 3B42 and NEXRAD Stage III data were variable (R2 = 0.361-0.732) with a positive y-intercept (13-29 mm) indicating a significant positive bias for TRMM 3B42 data. There was a slightly better correlation between TRMM 3B42 and NEXRAD Stage III data during dry periods (R2 = 0.600-0.875); however, a positive y-intercept (0-9 mm) indicated a positive bias for TRMM 3B42 data.

For the Middle Rio Grande Basin there was a similar trend, as observed in the Middle Nueces Basin, between TRMM 3B42 and average NEXRAD Stage III precipitation data within each TRMM grid cell (R2 = 0.493-0.821) with a positive bias towards TRMM 3B42 data as indicated by a positive y-intercept (5-14 mm) in all TRMM grid cell comparisons (Figure 3b). Total rainfall recorded in the Middle Rio Grande Basin was greater for ten out of eleven grid cell comparisons; bias for TRMM 3B42 data ranges from −2 to 14% (July 2003-November 2007). Additionally, precipitation data from wet (July 2003-December 2004; May-October 2007) and dry periods (January 2005-April 2007; November 2007) were examined. During both wet and dry periods correlations were more variable (R2 = 0.333-0.806; 0.322-0.800; respectively); however, during wet periods the y-intercept (10-34 mm) was greater than compared with dry periods (0-10 mm). Comparison between NEXRAD Stage III data (averaged for TRMM grid cell 27.50°N, 99.75°W) and NWS rain gauge from Laredo, Texas, the only rain gauge within the examined watershed, yielded a positive bias (total rainfall +15%) toward rain gauge data (R2 = 0.703; y-intercept = 5 mm), which was similar to previous comparisons between rain gauge and NEXRAD data (Johnson et al., 1999; Wang et al., 2000; Stellman et al., 2001; Neary et al., 2004). Likewise, comparison of TRMM 3B42 (grid cell – 27.50°N, 99.75°W) and Laredo rain gauge data (R2 = 0.653; y-intercept = 6 mm) yielded a positive bias (total rainfall +10%) for rain gauge data.

Streamflow Sensitivity Results

Results of the sensitivity test are broadly consistent within each watershed and are presented in Figure 4. Only major parameters that affect hydrologic results are discussed below. In the Middle Nueces Watershed the four most sensitive parameters are consistent for each simulation (CN, ENSO, SOL_AWC, SURLAG) and account for over 90% of the variability in model results. In the Middle Rio Grande Watershed over 99% of model variability is accounted for by the top four most sensitive parameters (CN, ENSO, SOL_AWC, Sol_z). These results validate the approach to manual calibration as discussed above. Most notable was that robust simulations in all watersheds were obtained without any ad hoc adjustments to either CN or SOL_AWC. It must be noted that calibration was obtained using the minimum permissible values for both ENSO and SURLAG (Van Liew et al., 2007) and no further adjustment of these parameters were possible. Finally, soil depth (Sol_z) in the U.S. was well defined by the STATSGO soil database and in Mexico by the World Reference Base (WRB) for soil resources so no further adjustment of this parameter was warranted.

image

Figure 4.  Sensitivity of Streamflow to Various SWAT Parameters for (a) Middle Nueces Watershed and (b) Middle Rio Grande Watershed. For both watersheds: NWS rain gauge data (squares), NEXRAD 05-06Z (triangles), and TRMM 06Z (circles).

Download figure to PowerPoint

Quantification of Streamflow Results

Three measures to quantify the goodness-of-fit of simulated streamflow relative to observed streamflow: mass balance error (MBE); Nash-Sutcliffe efficiency coefficients (NS; Nash and Sutcliffe, 1970); and root mean square error (RMSE) as defined below:

  • image(1)
  • image(2)
  • image(3)

where Qobs,a is the average observed streamflow. Additionally, Qsim,i and Qobs,i are the simulated and observed streamflow at the ith observation, respectively and N is the number of observations. Acceptable simulations have streamflow that are within 25% (MBE) of actual streamflow values with NS values >0.50 and RMSE values <0.70 (Moriasi et al., 2007). Marginal simulations can have NS values as low as 0.36 (Motovilov et al., 1999). Negative NS values indicate that simulated data performs worst than if the average of the observed streamflow was utilized for correlation. Legate and McCabe (1999) have demonstrated that the NS coefficient of efficiency is a superior measure of goodness-of-fit for comparison of hydrological time series data sets.

Calibration Period Results

For the Middle Nueces Basin the above parameter adjustments resulted in a reasonable manual calibration during the calibration period (July 2003-December 2006) of modeled streamflow values using only NEXRAD Stage III precipitation data at a monthly time scale (Table 3). The frequency of simulated monthly discharge as a function of exceedance probability (based on value of 1.00) is plotted in Figure 5 for NWS Rain Gauge, TRMM 3B42 06Z, and NEXRAD 05-06Z data. The lower exceedance probability the less frequent (and more extreme) the precipitation value. The simulation with NEXRAD 05-06Z data best matches observed streamflow data; although the NEXRAD 05-06Z simulation underestimates discharge associated with months with the greatest discharge (Figure 5). Both the NEXRAD 05-06Z and 12Z simulations yield acceptable values at a monthly time scale (NS = 0.62-0.63; RMSE = 0.38 to 0.40; MBE +9 to +13%; Figure 6). All TRMM 3B42 models (00Z, 06Z, and 12Z) yield similar results with marginal monthly NS (0.44-0.46), acceptable RMSE (0.30-0.32), but high MBEs of +67 to +71% (Table 3). Likewise NWS Rain Gauge data produces a simulation with low monthly NS (0.19) but acceptable RMSE (0.39) with a high MBE (+117%; Table 3).

Table 3.   SWAT Calibration and Validation Results for Middle Nueces Basin.
Precipitation TypeSurface Runoff Mass Balance Error Calibration/ValidationSurface Runoff Monthly NS Calibration/ValidationSurface Runoff Monthly RMSE Calibration/Validation
NWS Rain Gauge117%/61%0.19/0.690.39/0.16
TRMM 3B42 00Z70%/−18%0.46/0.610.30/0.48
TRMM 3B42 06Z71%/−20%0.44/0.600.31/0.50
TRMM 3B42 12Z67%/−19%0.46/0.600.32/0.50
NEXRAD 05-06Z9%/−8%0.63/0.700.40/0.40
NEXRAD 12Z13%/−17%0.62/0.690.38/0.38
Wet Periods
 NWS Rain Gauge61%/16%0.47/0.820.29/0.13
 TRMM 3B42 06Z26%/−37%0.61/0.600.28/0.50
 NEXRAD 05-06Z10%/−23%0.65/0.700.39/0.37
Dry Periods
 NWS Rain Gauge301%/413%−2.24/−6.016.55/9.22
 TRMM 3B42 06Z220%/107%−1.00/0.473.73/0.76
 NEXRAD 05-06Z72%/112%0.41/0.451.39/0.75
image

Figure 5.  Exceedance Probability (July 2003-December 2007) of Simulated (labeled gray lines – NEXRAD 05-06Z, TRMM 3B42, NWS Rain Gauge) and Observed (black line) Monthly Streamflow in the Middle Nueces Basin.

Download figure to PowerPoint

image

Figure 6.  Observed Monthly Streamflow (black) and Simulated Streamflow (gray) Based on SWAT Modeling for Calibration Period (July 2003-December 2006) and Validation Period (2007) for the Middle Nueces Watershed Using NEXRAD 05-06Z Precipitation Data. Wet and dry periods are indicated. Additionally, inset plotting simulated (labeled gray dashed lines – NEXRAD 05-06Z, TRMM 3B42, NWS Rain Gauge) and observed (black line) streamflow during the wet validation period (May-August 2007) is included.

Download figure to PowerPoint

Analysis of streamflow based on dry versus wet periods (as defined above for the Middle Nueces Basin) yields additional insights. SWAT model performance exhibits superior results during wet periods when compared with dry period response (Table 3). NEXRAD 05-06Z data yields a simulation with results for wet times during the calibration period (NS = 0.65; RMSE = 0.39; MBE +10%; Table 4; Figure 6). Simulations with TRMM 3B42 06Z (MBE = +26%) and NWS Rain Gauge (MBE = +61%) overestimate streamflow during the wet calibration times but to a lesser extent when compared with results from the entire calibration period (Table 3). All simulations from dry times during the calibration period consistently over-predict streamflow (Table 3).

Table 4.   SWAT Calibration and Validation Results for Rio Grande Basin.
Precipitation TypeSurface Runoff Mass Balance Error Calibration/ValidationSurface Runoff Monthly NS Calibration/ValidationSurface Runoff Monthly RMSE Calibration/Validation
NWS Rain Gauge26%/−20%0.30/0.980.58/0.03
TRMM 3B42 06Z20%/−15%0.38/0.940.65/0.07
NEXRAD 05-06Z6%/−13%0.90/0.950.06/0.06
 Surface Runoff Daily NS Calibration/ValidationSurface Runoff Daily RMSE Calibration/Validation 
 NWS Rain Gauge0.41/0.890.43/0.16 
 TRMM 3B42 06Z0.41/0.870.39/0.20 
 NEXRAD 05-06Z0.60/0.880.43/0.17 
Wet Periods
 NWS Rain Gauge45%/6%0.07/0.960.61/0.03
 TRMM 3B42 06Z44%/−18%0.11/0.940.69/0.07
 NEXRAD 05-06Z11%/−15%0.91/0.950.05/0.06
Dry Periods
 NWS Rain Gauge13%/5%0.70/0.960.28/0.05
 TRMM 3B42 06Z3%/−2%0.89/0.950.09/0.05
 NEXRAD 05-06Z3%/−4%0.89/0.940.09/0.06

For the Middle Rio Grande Basin, the above parameter adjustments resulted in an excellent manual calibration during the calibration period (July 2003-December 2006) using NEXRAD Stage III precipitation data at a monthly time scale (Table 4). The frequency of simulated monthly discharge is plotted in Figure 7 for NWS Rain Gauge, TRMM 3B42 06Z, and NEXRAD 05-06Z data with the best match with observed streamflow data obtained with NEXRAD 05-06Z data in the Middle Nueces Basin. Both NEXRAD 05-06Z and TRMM 3B42 06Z data produce simulations with acceptable results. The model run-based NEXRAD 05-06Z data yields total simulated streamflow that are +6% (MBE) of actual streamflow values and very high monthly NS (0.89) and daily NS (0.60) coefficients as well as low monthly RMSE (0.06) and daily RMSE (0.43) values (Table 4; Figure 8). These results are significant because SWAT is designed for monthly and annual use (Arnold et al., 1998). Note that daily data are not precisely daily, but are time-averaged over a three-day period including both the day before and after the date simulated. Nonetheless, high NS coefficients for NEXRAD model results at a sub-weekly time scale are deemed encouraging. Simulation based on TRMM 3B42 06Z data is marginally acceptable with MBE (+20%), monthly NS (0.38), and RMSE (0.65) values (Figure 9). Conversely, NWS Rain Gauge data do not produce an acceptable simulation with a low monthly NS coefficient (0.30) and high MBE (+26%).

image

Figure 7.  Exceedance Probability (July 2003-November 2007) of Simulated (labeled gray lines – NEXRAD 05-06Z, TRMM 3B42, NWS Rain Gauge) and Observed (black line) Monthly Streamflow in the Middle Rio Grande Basin.

Download figure to PowerPoint

image

Figure 8.  Observed Daily Streamflow (black) and Simulated Streamflow (gray) Based on SWAT Modeling for the Middle Rio Grande Watershed Using NEXRAD 05-06Z Precipitation Data. Wet and dry periods are indicated.

Download figure to PowerPoint

image

Figure 9.  Observed Monthly Streamflow (black) and Simulated Streamflow (gray) Based on SWAT Modeling for the Middle Rio Grande Watershed Using TRMM 06Z Precipitation Data. Wet and dry periods are indicated.

Download figure to PowerPoint

Analysis of streamflow based on dry versus wet periods (as defined above for the middle Rio Grande Watershed) yields additional insights. For NEXRAD 05-06Z data an excellent simulation is generated for both wet and dry periods (Table 4; NS = 0.89-0.94; RMSE = 0.06-0.09; MBE = +3-11%). Conversely, simulations with TRMM 3B42 06Z and NWS Rain Gauge data consistently over-predict streamflow during wet periods and yield acceptable results during dry periods (Table 4).

Validation Period Results

Results from the validation period (2007) were generally better than those from the preceding calibration period. In the Middle Nueces Watershed, both NEXRAD data (both 05-06Z and 12Z) and TRMM 3B42 (00Z, 06Z, 12Z) yielded acceptable simulations. During the validation period SWAT model performance exhibited results during wet periods with acceptable simulations for both NEXRAD 05-06Z (NS = 0.70; RMSE = 0.37; MBE −23%; Table 3; Figure 6) and NWS Rain Gauge data (NS = 0.82; RMSE = 0.13; MBE +16%; Table 3). During dry times of the validation period simulations consistently overestimated streamflow for all types on input precipitation (Table 4). A major discrepancy during the validation period was the failure of all simulations to reproduce peak discharge during July 2007 (inset in Figure 6). In the Rio Grande Watershed excellent results were obtained for all input precipitation types (Table 4).

Discussion

  1. Top of page
  2. Abstract
  3. Introduction
  4. Study Area
  5. Methodology
  6. Model Selection
  7. Data Preprocessing
  8. Model Set-up and Calibration
  9. Sensitivity Test
  10. Results
  11. Discussion
  12. Summary and conclusions
  13. Acknowledgments
  14. Literature Cited

Daily Definition Impact on Model Results

In the Middle Nueces Watershed an examination of whether the definition of the day for input precipitation data affected model results was completed by comparing NEXRAD (05-06Z, 12Z) and TRMM 3B42 (00Z, 06Z, and 12Z) data. In both cases, simulated results for NEXRAD and TRMM 3B42 datasets were internally consistent with a variation of <10% MBE and <0.1 differences in monthly/daily NS when comparing NEXRAD 05-06Z and 12Z as well as all three TRMM data types with each other (Table 3). At the daily time scale an examination of both raw (nontime averaged) and time averaged datasets were completed, and in both instances, results were consistent in terms of low internal variability between simulations for both NEXRAD and TRMM 3B42 data sets; although, time averaging consistently resulted in more robust daily NS coefficients. These results were significant because SWAT was not specifically designed to support daily modeling. The results of this study indicate that SWAT is capable of producing robust results at fine time scales and the nuance associated with how hourly precipitation data were organized to generate daily totals does not significantly impact model performance at a sub-weekly scale.

SWAT Model Performance

Some interesting trends were revealed by examining data based on wet and dry climatic periods. In the Middle Nueces Basin simulated streamflow grossly overestimated actual streamflow during dry climatic periods and yielded more accurate results for all precipitation types during wet periods (Table 3). Previous studies have indicated that SWAT performs better under more humid climatic conditions (Van Liew et al., 2005, 2007) and SWAT has problems with accurately accounting for soil moisture storage and water loss through infiltration and evapotranspiration especially during dry climatic periods (Feyereisen et al., 2007). Sheridan and Shirmohammadi (1986) suggested that developing customized CN for extremely dry periods when soil moisture values approach the wilting point can significantly improve model performances by increasing soil infiltration and decreasing streamflow. Superior performance during wet periods was attributed to physical conditions more consistent with those for which SWAT was designed to simulate (e.g., Kannan et al., 2006). However, a documented problem of SWAT is the performance during extreme high streamflow events (e.g., Binger, 1996; Chu and Shirmohammadi, 2004) in which SWAT tends to grossly underestimate streamflow. Utilization of three input precipitation sets was revealing in this situation as both NEXRAD and TRMM simulations grossly underestimate streamflow for July 2007 (inset in Figure 6). Conversely, NWS Rain Gauge data yielded better results as documented by the high correlation during the four month wet period (May-August 2007) between observed and simulated streamflow (Table 3).

Results for the Middle Rio Grande Basin are significantly different, poor performance during wet versus dry climatic period (for TRMM 3B42 and NWS Rain Gauge), opposite of Middle Nueces Basin results and can be related to differences in inlet discharge between the two basins. Discharge in the Middle Rio Grande Basin was influenced greatly from upstream releases of water from Amstidad Reservoir. For the Middle Rio Grande Basin, during the calibration period, inlet discharge accounted for 94% of the streamflow in the basin as opposed to the Middle Nueces Basin in which inlet discharge is 50% of total streamflow. Therefore, in the Middle Nueces Basin, streamflow was more strongly controlled by physical processes present throughout the basin whereas in the Middle Rio Grande Basin processes associated with the channel have a much stronger impact on outlet streamflow. During dry climatic periods streamflow was strongly influenced by discharge contributions from the inlet and physical processes within the watershed as a whole have a limited impact on streamflow at the outlet. Therefore, acceptable simulations are obtained with all precipitation types during dry climatic periods in the Middle Rio Grande Basin (Figure 9). During the wet climatic period (July 2003-December 2004) associated with the calibration period both TRMM 3B42 and NWS Rain Gauge data overestimated the contribution of water contributed to the channel by the tributary subbasins (Table 4; Figure 9); whereas, NEXRAD data yielded simulations that closely matched observed streamflow (Table 4). The validation period has a six-month wet climatic period (May-October 2007) that was preceded by over two years of dry climatic conditions. This was unlike the wet period during the calibration period (July 2003-December 2004), which was preceded by an exceptionally wet year (2002). During the wet calibration period positive biases in TRMM 3B42 and NWS Rain Gauge data yielded streamflow simulations that overestimated observed values (Table 4; Figure 9). Conversely, during the wet validation period low antecedent moisture status in the basin biases all model results toward decreased simulated streamflow values (Table 4; Figure 9). These factors explain the better performance of TRMM 3B42 and NWS Rain Gauge data in generating simulated streamflow during the validation versus the calibration period (Table 4; Figure 9). Previous studies (Chu et al., 2004; El-Nasr et al., 2005; Jha et al., 2006) have documented superior results during the validation period as opposed to the calibration period. Finally, NEXRAD-based simulations for both calibration and validation periods with both wet and dry conditions produced excellent results in the Middle Rio Grande Basin (Figure 8).

Utility of Remotely Sensed Precipitation Data for Supporting Hydrological Modeling

Hydrologic variability is especially extreme in semi-arid climatic regimes such as the Middle Rio Grande and Nueces River Watersheds. The controlling physical processes that generate hydrologic variability are complex and interconnected in ways not well understood in the context of the region’s landscape especially at finer time-scales (daily or even hourly). Variability exists at a number of spatial and temporal scales. Therefore, at the watershed scale, hydrologic modeling must be spatially distributed. A major limitation that has in the past impeded accurate modeling of the hydrology of the Middle Rio Grande and Nueces River Basins was the lack of precipitation data that are representative at the watershed scale, which is a problem in many regions of the world.

This study demonstrates that remotely sensed precipitation data can effectively support spatially distributed hydrological modeling at a large watershed scale. A number of previous studies have documented the ability of SWAT to effectively model watershed processes with NEXRAD Stage III data (see Gassman et al., 2007) with daily/monthly NS coefficients in excess of 0.36 documented in numerous studies. This study is the first to successfully demonstrate the utility of satellite-estimated precipitation (TRMM 3B42) in supporting hydrologic modeling with SWAT; thereby, potentially extending the realm (between 50°N and 50°S) where remotely sensed precipitation data can support hydrologic modeling outside of regions that have modern, ground-based radar networks (i.e., much of the third world). In this study, TRMM 3B42 performs slightly better than sparsely distributed rain gauge data.

This study emphasizes that while both NEXRAD Stage III and TRMM 3B42 data provided at least satisfactory simulation results in the watersheds examined in this study, the results do not necessarily imply that remotely sensed precipitation data will always either (1) outperform rain gauge data, or (2) provide good results for every watershed. Clearly, the quality of remotely sensed data can vary on a spatial basis and affect modeling results for specific storm events. For example, NEXRAD Stage III data can be potentially degraded by a number of errors including incomplete ground clutter suppression, anomalous beam propagation, bright bands, beam overshooting at the edge of the radar coverage envelop, returns from nonweather echoes such as birds and insects, difficulty in detecting low intensity precipitation; to name a few (e.g., Harrison et al., 2000). Likewise, TRMM 3B42 data suffer from limited spatial and temporal coverage of high quality microwave data, a high detection limit (0.7 mm/h) for detecting low intensity precipitation, and data retrieval problems especially over semi-arid land (Ferraro et al., 1998; Tian and Peters-Lidard, 2007); most calibration efforts associated with TRMM products has focused on oceanic regions (Huffman et al., 2007). Despite these potential pitfalls the fact that both NEXRAD Stage III and TRMM 3B42 data can be used to produce reasonable simulations even at a monthly time scale should be deemed as encouraging. This study does not necessarily negate the utility of rain gauge data and it has been demonstrated that a sufficiently dense network of rain gauges can support hydrological modeling that outperforms results obtained using NEXRAD Stage III (Neary et al., 2004). However, particularly in regions with limited ground-based precipitation monitoring, both NEXRAD Stage III and especially TRMM 3B42 data, throughout most of the third world, should be seriously considered as a potentially viable source of precipitation data for hydrological simulations.

Summary and conclusions

  1. Top of page
  2. Abstract
  3. Introduction
  4. Study Area
  5. Methodology
  6. Model Selection
  7. Data Preprocessing
  8. Model Set-up and Calibration
  9. Sensitivity Test
  10. Results
  11. Discussion
  12. Summary and conclusions
  13. Acknowledgments
  14. Literature Cited

Precipitation data (NEXRAD Stage III, TRMM 3B42, NWS Rain Gauge) and streamflow simulations generated by SWAT were examined from two basins in south Texas and northern Mexico (Middle Nueces and Middle Rio Grande). Both TRMM 3B42 and NWS Rain Gauge data exhibited significant positive bias compared with NEXRAD Stage III data. Generally, SWAT simulations with NEXRAD Stage III data yielded superior streamflow results at a monthly time scale compared with observed values than model results generated using TRMM 3B42 or NWS Rain Gauge data. There were distinct differences in model response during wet and dry climatic periods that can be related to physical processes within each basin. In the Middle Rio Grande Basin SWAT simulations generated using NEXRAD Stage III data also produced good results (NS = 0.60) at a daily time step, which was considered particularly encouraging in light of the fact the SWAT was originally designed to generate model output at a monthly and annual time scale. However, problems in simulating the extreme discharge event (July 2007) in the Middle Nueces Basin using NEXRAD Stage III and TRMM 3B42 perhaps indicate that additional work is needed to improve remotely sensed quantification of extreme precipitation events. While simulations with NEXRAD Stage III data were superior, TRMM 3B42-based simulations were acceptable and comparable to those based on the sparse rain gauge data in the region. These results suggest that TRMM 3B42 data can provide a source of precipitation input data for hydrological modeling at a monthly time scale in regions with limited ground-based precipitation monitoring and low resolution data sources such as STATSGO soil data.

Acknowledgments

  1. Top of page
  2. Abstract
  3. Introduction
  4. Study Area
  5. Methodology
  6. Model Selection
  7. Data Preprocessing
  8. Model Set-up and Calibration
  9. Sensitivity Test
  10. Results
  11. Discussion
  12. Summary and conclusions
  13. Acknowledgments
  14. Literature Cited

We acknowledge the support of the NASA Precipitation Science program through award # NA17AE2924 (Dr. Ramesh Kakar). The assistance of research assistant (Arturo Diaz) is greatly appreciated. In addition, the data used in this study were acquired using the GES-DISC Interactive Online Visualization and Analysis Infrastructure (Giovanni) as part of NASA’s Goddard Earth Science (GES) Data and Information Services Center (DISC). Additionally, we gratefully acknowledge the assistance of the West Gulf Coast River Forecast Center in providing NEXRAD Stage III data for this project.

Literature Cited

  1. Top of page
  2. Abstract
  3. Introduction
  4. Study Area
  5. Methodology
  6. Model Selection
  7. Data Preprocessing
  8. Model Set-up and Calibration
  9. Sensitivity Test
  10. Results
  11. Discussion
  12. Summary and conclusions
  13. Acknowledgments
  14. Literature Cited
  • Afinowicz, J.D., C.L. Munster, and B.P. Wilcox, 2005. Modeling Effects of Brush Management on the Rangeland Water Budget: Edwards Plateau, Texas. Journal of the American Water Resources Association (JAWRA) 41(1):181-193.
  • Arnold, J.G. and P.M. Allen, 1999. Automated Methods for Estimating Baseflow and Groundwater Recharge From Stream Flow Records. Journal of the American Water Resources Association (JAWRA) 35(2):411-424.
  • Arnold, J.G., P.M. Allen, R. Muttiah, and G. Bernhardt, 1995. Automated Base Flow Separation and Recession Analysis Techniques. Groundwater 33:1001-1018.
  • Arnold, J.G. and N. Fohrer, 2005. SWAT2000: Current Capabilities and Research Opportunities in Applied Watershed Modeling. Hydrologic Process 19(3):563-572.
  • Arnold, J.G., K.N. Potter, K.W. King, and P.M. Allen, 2005. Estimation of Soil Cracking and the Effect on Surface Runoff in a Texas Blackland Prairie Watershed. Hydrologic Process 19(3):589-603.
  • Arnold, J.G., R. Srinivasan, R.S. Muttiah, and P.M. Allen, 1998. Large-Area Hydrologic Modeling and Asssessment: Part I. Model Development. Journal of the American Water Resources Association (JAWRA) 34(1):73-89.
  • Arnold, J.G., J.R. Williams, R.H. Griggs, and N.B. Sammons, 1990. SWRRBWQ – A Bassin Scale Model for Assessing Management Impacts on Water Quality. USDA Agricultural Research Service, Washington, D.C.
  • Beasley, D.B., L.F. Huggins, and E.J. Monke, 1980. ANSWERS: A Model for Watershed Planning. Transactions of the ASAE 23:938-944.
  • Bedient, P.B., B.C. Hoblit, D.C. Gladwell, and B.E. Vieux, 2000. NEXRAD Radar for\Flood Prediction in Houston. Journal of Hydrologic Engineering 5(3):269-277.
  • Binger, R.L., 1996. Runoff Simulated From Goodwin Creek Watershed Using SWAT. Transactions of the ASABE 39(1):85-90.
  • Bocchiola, D., 2007. Use of Scale Recursive Estimation for Assimilation of Precipitation Data From TRMM (PR and TMI) and NEXRAD. Advances in Water Resources 30:2354-2372.
  • Chu, T.W. and A. Shirmohammadi, 2004. Evaluation of the SWAT Model’s Hydrology Component in the Piedmont Physiographic Region of Maryland. Transactions of the ASAE 47(4):1057-1073.
  • Chu, T.W., A. Shirmohammadi, H. Montas, and A. Sadeghi, 2004. Evaluation of the SWAT Model’s Sediment and Nutrient Components in the Piedmont Physiographic Region of Maryland. Transactions of the ASAE 47(5):1523-1538.
  • Ciach, G.J. and W.F. Krajewski, 1999. Radar-Rain Gauge Comparison Under Observational Uncertainties. Journal of Applied Meteorology 38:1519-1525.
  • Clapp, R.B. and G.M. Hornberger, 1978. Empirical Equations for Some Soil Hydraulic Properties. Water Resources Research 14:601-604.
  • Di Luzio, M. and J.G. Arnold, 2004. Formulation of a Hydrid Calibration Approach for a Physically Based Distributed Model With NEXRAD Data Input. Journal of Hydrology 298:136-154.
  • Di Luzio, M., R. Srinivasan, and J.G. Arnold, 2002. Integration of Watershed Tools and SWAT Model Into BASINS. Journal of the American Water Resources Association (JAWRA) 38(4):1127-1141.
  • Duchon, C.E. and G.R. Essenberg, 2001. Comparative Rainfall Observations From Pit and Aboveground Rain Gauges With and Without Wind Shield. Water Resources Research 37:3253-3263.
  • El-Nasr, A.A., J.G. Arnold, J. Feyen, and J. Berlamont, 2005. Modelling the Hydrology of a Catchment Using Distributed and a Semi-Distributed Model. Hydrological Processes 19:573-587.
  • Ferraro, R.R., E.A. Smith, W. Berg, and G.J. Huffman, 1998. A Screening Methodology for Passive Microwave Precipitation Retrieval Algorithms. Journal of Atmospheric Science 55:1583-1600.
  • Feyereisen, G.W., T.C. Strickland, D.D. Bosch, and D.G. Sullivan, 2007. Evaluation of SWAT Manual Calibration Sensitivity in the Little River Watershed. Transactions of the ASABE 50(3):843-855.
  • FitzHugh, T.W. and D.S. MacKay, 2000. Impacts of Input Parameters Spatial Aggregation on an Agricultural Nonpoint Source Pollution Model. Journal of Hydrology 236:35-53.
  • Gassman, P.W., M.R. Reyes, C.H. Green, and J.G. Arnold, 2007. The Soil and Water Assessment Tool: Historical Development, Applications, and Future Research Directions. Transactions of the ASABE 50(4):1211-1250.
  • Hanratty, M.P. and H.G. Stefan, 1998. Simulating Climate Change Effects in a Minnesota Agricultural Watershed. Journal of Environmental Quality 27:1524-1532.
  • Harrison, D.L., S.J. Driscoll, and M. Kitchen, 2000. Improving Precipitation Estimates From Weather Radar Using Quality Control and Correlation Techniques. Meteorological Applications 6:135-144.
  • Higgins, R., Y. Chen, and A. Douglas, 1999. Interannual Variability of the North American Warm Season Precipitation Regime. Journal of Climate 12:653-680.
  • Huffman, G.J., R.F. Adler, D.T. Bolvin, G. Gu, E.J. Nelkin, K. Bowman, E.F. Stocker, and D. Wolff, 2007. The TRMM Multi-Satellite Precipitation Analysis (TMPA): Quasi-Global, Multi-Year, Combined-Sensor Precipitation Estimates at Fine Scales. Journal of Hydrometeorology 8(1):38-55.
  • IBWC (International Boundary Water Comission), 2008. Historical Streamflow Data: Rio Grande Basin. International Boundary and Water Commission, El Paso, Texas. http://www.ibwc.state.gov/Water_Data/histflo1.htm, accessed December 1, 2007.
  • INEGI, 2008. Main Webpage. Instituto Nacional de EstadÌsticas, GeografÌa, e Informática, Mexico City, Mexico. http://www.inegi.gob.mx/inegi/default.aspx, accessed July 1, 2007.
  • International Union of Soil Sciences, 2006. World Reference Base for Soil Resources 2006: A Framework for International Classification, Correlation, and Communication. World Soil Resources Report No. 103, p. 128. Food and Agriculture Organization of the United Nations, Rome.
  • Jha, M., J.G. Arnold, P.W. Gassman, F. Giorgi, and R. Gu, 2006. Climate Change Sensitivity Assessment on Upper Mississippi River Basin Streamflows Using SWAT. Journal of the American Water Resources Association (JAWRA) 42(4):997-1015.
  • Johnson, D., M. Smith, V. Koren, and B. Finnerty, 1999. Comparing Mean Areal Precipitation Estimates From NEXRAD and Rain Gauge Networks. Journal of Hydrologic Engineering 4(2):117-124.
  • Kalin, L. and M.M. Hantush, 2006. Hydrologic Modeling of an Eastern Pennsylvania Watershed With NEXRAD and Rain Gauge Data. Journal of Hydrologic Engineering 11(6):555-569.
  • Kannan, N., S.M. White, F. Worrall, and M.J. Whelan, 2006. Sensitivity Analysis and Identification of the Best Evapotranspiration and Runoff Options for Hydrologic Modeling in SWAT-2000. Jounral of Hydrology 332:456-466.
  • Kitchen, M. and R.M. Blackall, 1992. Representativeness Errors in Comparison Between Radar and Gauge Measurements of Rainfall. Journal of Hydrology 134:13-33.
  • Krajewski, W.F., 1987. Co-Kriging of Radar-Rainfall and Rain Gauge Data. Journal of Geophysical Research 92:9571-9580.
  • Krajewski, W.F. and J.A. Smith, 2002. Radar Hydrology: Rainfall Estimation. Advances in Water Resources 25:1387-1394.
  • Legate, D.R. and G.J. McCabe Jr, 1999. Evaluating the Use of “Goodness-of-Fit” Measures in Hydrologic and Hydroclimate Model Validation. Water Resources Research 35(1):233-241.
  • Milewski, A., M. Sultan, E. Yan, A. Abdeldayem, A. Geilil, S. Markondiah Jayaprakash, and C. Jones, 2007. Remote Sensing Solutions for Continuous Rainfall Runoff Modeling. 2007 AGU Joint Assembly. H51C-05. American Geophysical Union, Washington, D.C.
  • Moon, J., K. Srinivasan, and J.H. Jacobs, 2004. Stream Flow Estimation Using Spatially Distributed Rainfall in the Trinity River Basin, Texas. Transactions of the ASAE 47:1445-1451.
  • Moriasi, D.N., J.G. Arnold, M.W. Van Liew, R.L. Bingner, R.D. Harmel, and T.L. Veith, 2007. Model Evaluation Guidelines for Systematic Quantification of Accuracy in Watershed Simulations. Transactions of the ASABE 50(3):885-900.
  • Morrissey, M.L., J.A. Maliekal, J.S. Greene, and J. Wang, 1995. The Uncertainity of Simple Spatial Averages Using Rain Gauge Networks. Water Resources Research 31(8):2011-2017.
  • Motovilov, Y.G., L. Gottschalk, K. Engeland, and A. Rodhe, 1999. Validation of Distributed Hydrological Model Against Spatial Observations. Agricultural and Forest Meteorology 98:257-277.
  • NASA (National Aeronautics and Space Administration), 2008. GES-DISC Interactive Online Visualization ANd ANalysis Infrastructure (Giovanni). National Aeronautics and Space Administration, Washington, D.C http://disc2.nascom.nasa.gov/Giovani/tovas/TRMM_V6.3B42.2.shtml, accessed December 1, 2007.
  • Nash, J.E. and J.V. Sutcliffe, 1970. River Flow Forecasting Through Conceptual Models. Part 1: Discussion of Principles. Journal of Hydrology, 10:282-290.
  • NCDC, 2008. Climate Station Locator. National Atmospheric and Oceanic Agency, Ashville, North Carolina. http://www.ncdc.noaa.gov/oa/climate/stationlocator.html, accessed March 1, 2008.
  • Neary, V.S., E. Habib, and M. Fleming, 2004. Hydrologic Modeling With NEXRAD Precipitation in Middle Tennessee. Journal of Hydrologic Engineering 9(5):339-349.
  • Neitsch, S.L., J.G. Arnold, J.R. Kiniry, R. Srinivasan, and J.R. Williams, 2002. Soil and Water Assessment Tool User’s Manual. Texas Water Resource Institute Report TR-192, Texas Water Resource Institute, College Station, Texas, p. 455.
  • NOAA (National Oceanographic and Atmospheric Agency), 2008. Southern Division Precipitation Analysis Website. National Atmospheric and Oceanic Agency, Fort Worth, Texas. http://www.srh.noaa.gov/rfcshare/p_download/p_download.php, accessed December 2, 2007.
  • Ogden, F.L., J.F.L. Garbrecht, P.A. Debarry, and A.R. Maidment, 2001. GIS and Distributed Watershed Models II: Modules, Interfaces, and Models. Journal of Hydrologic Engineering 6:515-523.
  • Rio Grande Regional Water Planning Group, 2001. Rio Grande Regional Water Plan, Lower Rio Grande Valley Development Council, McAllen, Texas, http://riograndewaterplan.org/waterplanOLD.php, accessed March 15, 2006.
  • Rodrigrez-Iturbe, I., J.M. Mejia, 1974. The Design of Rainfall Networks in Time and Space. Water Resources Research 10:713-728.
  • Sevruk, B., 1989. Reliability of Precipitation Measurements. Proceedings of WMO/IAHS/ETH Workshop on Precipitation Measurements. World Meteorological Organization, St. Moritz, Switzerland, pp. 13-19.
  • Sheridan, J.M. and A. Shirmohammadi, 1986. Application of Curve Number Procedure on Coastal Plain Watersheds. ASAE. Paper No. 862505. ASAE, St. Joseph, Michigan.
  • Smith, J.A. and W.F. Krajewski, 1991. Estimation of the Mean Field Bias of Radar Rainfall Estimates. Journal of Applied Meteorology 30:397-412.
  • Smith, J.A., D.J. Seo, M.L. Baeck, and M.D. Hudlow, 1996. An Intercomparison Study of NEXRAD Precipitation Estimates. Water Resources Research 32:2035-2045.
  • Soil Conservation Service Engineering Division, 1986. Urban Hydrology for Small Watersheds. U.S. Department of Agriculture, Washington, D.C., Technical Release 55.
  • Soil and Water Assessment Tool, 2008, Main Website. Texas A&M University, College Station, Texas, http://www.brc.tamus.edu/swat/, accessed January 17, 2007.
  • Stellman, K.M., H.E. Fuelberg, R. Garza, and M. Mullusky, 2001. An Estimation of Radar- and Rain Gauge-Derived Mean Areal Precipitation Ovedr Georgia Watersheds. Weather Forecasting 16(1):133-144.
  • Texas Water Development Board, 2002, Water for Texas - 2002. State of Texas. Austin, Texas. http://www.twdb.state.tx.us/publications/reports/StateWater_Plan/2002/FinalWaterPlan2002.asp, accessed July 24, 2006.
  • Tian, Y. and C.D. Peters-Lidard, 2007. Systematic Anomalies Over Inland Water Bodies in Satellite-Based Precipitation Estimates. Geophysical Research Letters 34. Paper No. L14493.
  • Tokay, A., D.B. Wolff, K.R. Wolff, and P. Bashor, 2003. Rain Gauge and Disdrometer Measurements During Keys Area Microphysics Project (KAMP). Journal of Atmospheric & Oceanic Technology 20:1460-1477.
  • USDA (United States Department of Agriculture), 1994. State Soil Data Use Information. Soil Conservation Service. U.S. Department of Agriculture, Washington, D.C. http://www.ftw.nrcs.usda.gov/stat_data.html, accessed February 1, 2007.
  • USEPA (United States Environmental Protection Agency), 2007. Better Assessment Science Intergrating Point and Non-Point Sources. U.S. Environmental Protection Agency, Washington, D.C. http://www.epa.gov/waterscience/basins/, accessed February 1, 2007.
  • USGS (U.S. Geological Survey), 1986. Land Use and Land Cover Digital Data From 1:250,000-and 1:100,000-Scale Maps. Data Users Guide 4. U.S. Geological Survey, Reston, Virginia. http://edc2.usgs.gov/geodata/LULC/LULCDataUsersGuide.pdf, accessed February 1, 2007.
  • USGS (U.S. Geological Survey), 1995, Metadata for 1-Degree Digital Elevation Models. U.S. Geological Survey, Reston, Virginia. http://nsdi.usgs.gov/products/dem.html, accessed February 1, 2007.
  • USGS (U.S. Geological Survey), 1999. National Hydrography Dataset. U.S. Geological Survey, Reston, Virginia. http://nhd.usgs.gov, accessed February 1, 2007.
  • USGS (U.S. Geological Survey), 2008. Texas Water Data Website. U.S. Geological Survey, Reston, Virginia. http://waterdata.usgs.gov/tx/nwis/dv?referred_module=sw, accessed February 1, 2007.
  • Van Grinsven, A., 2002. Development Towards Integrated Water Quality Modeling for River Basins. Publication No. 40, Department of Hydrology and Hydralulic Engineering, Vrije Universiteit Brussel.
  • Van Grinsven, A. and T. Meixner, 2003. Sensitivity, Optimisation and Uncertainty Analysis for the Model Parameters of SWAT. Proceedings of the Second International SWAT Conference. Bari, Italy. TWRI Technical Report 266. pp. 162-167.
  • Van Liew, M.W., J.G. Arnold, and D.D. Bosch, 2005. Problems and Potential of Autocalibrating a Hydrologic Model. Transactions of the ASABE 48(3):1025-1040.
  • Van Liew, M.W., T.L. Vieth, D.D. Bosch, and J.G. Arnold, 2007. Suitability of SWAT for the Conservation Effects Assessment Project: Comparison on USDA Agricultural Research Service Watersheds. Journal of Hydrologic Engineering 12(2):173-189.
  • Wang, D., M.B. Smith, Z. Zhang, S. Reed, and V.I. Koren, 2000. Statistical Comparison of Mean Areal Precipitation Estimates From WSR-88D, Operational and Historical Gage Networks. Proceedings, 15th Conference on Hydrology. 80th American Meteorological Society Meeting, Boston.
  • Williams, J.R., 1969. Flood Routing With Variable Travel Time or Variable Storage Coefficients. Transactions of the ASAE 12(1):100-103.
  • Xie, H., X. Zhou, E.R. Vivoni, J.M.H. Hendrick, and E.E. Small, 2005. GIS-Based NEXRAD Stage III Precipitation Database: Automated Approaches for Data Processing and Visualization. Computers and Geoscience 31:65-76.
  • Young, R.A., C.A. Onstad, D.D. Bosch, and W.P. Anderson, 1989. AGNPS: A Nonpoint- Source Pollution Model for Evaluating Agricultural Watersheds. Journal of Soil and Water Conservation, 44(2):168-173.
  • Zawadski, I., 1975. On Radar-Raingauge Comparison. Journal of Applied Meteorology 14:1430-1436.