INTEGRATING NASA DATA AND MODEL PRODUCTS IN HSPF AND SWAT HYDROLOGIC SIMULATION MODELS

Rapid Prototyping Experiment: INTEGRATING NASA DATA AND MODEL PRODUCTS IN HSPF AND SWAT HYDROLOGIC SIMULATION MODELS Charles G. O Hara 1, Vladimir J. Alarcon 1, Ted Engman 2, Joe Nigro 2, David Toll 2, Roland Viger 3 1 GeoResources Institute, Mississippi State University 2 NASA Goddard Space Center 3 US Geological Survey

Major Objectives Evaluate NASA Existing Data, Simulated Future Data Streams, and Model (LIS) Data Products to Provide Enhancements to Watershed Modeling Programs/DSTs HSPF (Hydrologic Simulation Program - FORTRAN SWAT (Soil and Water Assessment Tool) Study area Two main river catchments in Saint Louis Bay watershed at the Mississippi Gulf Coast were the focus of most of the experiments One experiment focuses on the Luxapallila watershed (located in Northeast Mississippi).

Standard Dataset Inputs for Watershed Modeling Dataset Provider Limitations Topography DEM: 300 m resolution, NED: 30 m resolution Land use, land cover GIRAS: 400 m resolution, NLCD: 30 m resolution USGS Depending on the size of the watershed under study DEM could result a coarse approximation to actual relief USGS Both datasets are outdated. The most current dataset is NLCD-2001, based in land use information collected during the 1990's NCDC Several stations have incomplete timeseries Precipitation Gage station records at hourly, daily frequencies Potential Evapotranspiration Calculated internally in watershed models. Simple algorithms.

NASA Datasets Evaluated for Watershed Modeling Satellite/Model Sensor System Operator Shuttle Radar Topography Mission RADAR: two radar antennas located in the shuttle's payload bay, and the other on the end of a 60-meter mast NASA, NGA Product SRTM 30-meter resolution digital elevation model TERRA/AQUA Moderate Resolution Spectroradiometer (MODIS) NASA MODIS 12Q1 land use land cover, 1000 m resolution LIS NASA GODDARD Precipitation and Evapotranspiration time-series

Methodology Several experiments were performed to assess the effectiveness of NASA datasets for watershed hydrology modeling. The exploration was done using individual NASA products as well as combining inputs in factorial fashion with other datasets used in watershed modeling. Experiments can be grouped into the following types: topography, land use, land use and topography, forcings, and land use dynamics.

Topography Experiment Watershed delineation: hydrologic division of a watershed into sub-watersheds Topography is used as the primary reference. Shuttle Radar Topography Mission (SRTM) data are used to delineate selected regions of the Saint Louis Bay watershed (Mississippi). Topographical indicators were computed. Comparison of topographical indicators obtained using SRTM against those obtained using: Interferometric Synthetic Aperture Radar (IFSAR), National Elevation Dataset (NED) and USGS digital elevation models (DEM). Science Question: How does SRTM perform for watershed delineation?

Topography Experiment A) USGS-DEM B) IFSAR C) SRTM - 1 Crabgrass Creek White Cypress Crabgrass Creek White Cypress Crabgrass Creek White Cypress Dead Tiger Dead Tiger Dead Tiger E) SRTM - 2 F) NED G) STUDY AREA White Cypress White Cypress Crabgrass Creek Crabgrass Creek Dead Tiger Dead Tiger

12000.00 Results: SRTM, NED and USGS-DEM produce equivalent, smooth and continuous demarcation of sub-basins. The experiment determined that the use of SRTM provides equivalent delineation results and topographical parameterized information. Area and river length measurements are highly correlated for all data sets excepts IFSAR for area measurements. 10000.00 8000.00 6000.00 4000.00 2000.00 0.00 Hickory Creek White Cypress Creek Catahoula Creek Crane Pond Branch Jourdan River Crabgrass Creek Dead Tiger Creek Jourdan River Sub-basin USGS-DEM IFSAR SRTM-1 SRTM-2 A R E A (ha) 30000.00 25000.00 20000.00 15000.00 10000.00 5000.00 0.00 Hickory Creek White Cypress Creek Catahoula Creek Crane Pond Branch Jourdan River Crabgrass Creek Dead Tiger Creek Jourdan River Topography Experiment Sub-basin RIVER LENGTH (m)

Land Use Experiment This experiment explored the impact of land use/land cover (LU/LC) data quality to simulate watershed processes at hill slope scale and at the watershed outlet of the Luxapallila Creek basin, located in Alabama and Mississippi. Simulated values of flow and sediments were obtained after swapping three land use databases. Science Questions: Does land use dataset swapping affects HSPF s stream flow and sediment estimations? How does MODIS MOD12Q 1 LU/LC dataset perform?

Land Use Experiment

Land Use Experiment After swapping land use datasets, the HSPF model estimations did not show substantial changes on the water balance components (evapo-transpiration, total runoff, and deep groundwater) and stream flow. At the hill slope scale, the LU/LC datasets generated noticeable differences in simulated values of total fraction of sediments when swapping LU/LC datasets. The HSPF simulation that used NLCD as LU/LC data produced sediment fraction values 8.7% smaller than sediment fraction values calculated using GIRAS. Conversely, MODIS produced values 8.6% bigger. In sediment modeling, agricultural land was the main source of sediment export (i.e., between 70% and 80% of total sediments). Larger differences in agricultural areas were calculated: -18.9% and 10% for NLCD and MODIS, respectively.

Land Use and Topography (Combined) Experiment This experiment explored the effects of swapping several topographical and land use datasets of different spatial resolution and scale in hydrological estimations of stream flow. Science Question: Is the combination of MODIS and SRTM useful for hydrological modeling? A factorial design with several different LULC and topography datasets was performed (see Table 3). Twelve concurrent scenarios of Topographical/LULC cases were generated for HSPF ingestion. Science Question: Can we use advanced approaches to data exploration in modeling to determine the best combinations of input datasets for a give hydrologic modeling activity? Science Question: Do geography, scale, and temporal considerations mean there is no ideal universal set of rules for what data works best?

Land Use and Topography (Combined) Experiment TOPOGRAPHICAL datasets A USGS-DEM (300m) B NED (30 m) LANDUSE datasets C SRTM (30 m) D IFSAR (5 m) NLCD USGS-GIRAS BASE CASE MODIS

Land Use and Topography (Combined) Experiment Jourdan River Watershed Nash-Sutcliff (NS) model fit efficiency Coefficient of determination (R^2) 0.75 MODIS (1000 m) NLCD (30 m) IFSAR (5m) SRTM (30m) NED (30m) DEM (300m) 0.745 0.74 0.735 0.73 0.725 0.72 0.745-0.75 0.74-0.745 0.735-0.74 0.73-0.735 0.725-0.73 0.72-0.725 MODIS (1000 m) GIRAS (400 m) NLCD (30 m) IFSAR (5m) SRTM (30m) NED (30m) DEM (300m) 0.8 0.795 0.79 0.785 0.78 0.775 0.77 0.765 0.76 0.795-0.8 0.79-0.795 0.785-0.79 0.78-0.785 0.775-0.78 0.77-0.775 0.765-0.77 0.76-0.765 Wolf River Watershed MODIS (1000 m) GIRAS (400 m) NLCD (30 m) Nash-Sutcliff, NS, model fit efficiency IFSAR (5m) SRTM (30m) NED (30m) DEM (300m) 0.83 0.82 0.81 0.8 0.79 0.78 0.77 0.76 0.82-0.83 0.81-0.82 0.8-0.81 0.79-0.8 0.78-0.79 0.77-0.78 MODIS (1000 m) Coefficient of determination (R^2) 0.76-0.77 DEM GIRAS (400 m) NED (300m) SRTM (30m) NLCD (30 m) IFSAR (30m) (5m) 0.84 0.835 0.83 0.825 0.82 0.815 0.81 0.835-0.84 0.83-0.835 0.825-0.83 0.82-0.825 0.815-0.82 0.81-0.815 The combination of moderate resolution topographical datasets (such as SRTM, 30 m) and low resolution land use datasets (such as MODIS, 1000 m) produce good statistical fit between simulated and measured stream flow hydrographs. Model fit coefficients (R 2 and NS) for the MODIS-SRTM combination range between 0.73 and 0.81 (perfect fit is 1.00).

Land Cover Dynamics and Forcings Experiment MODIS 12Q1 land use products from 2001 up to 2004 were introduced in a calibrated HSPF model application for Wolf River watershed. Meteorological data only existed for the period 1970 to 1996. Precipitation and evapo-transpiration time-series from 1996 to the present for the area were produced by NASA-LIS (GSFC). NASA-LIS data was re-formatted to update the original Wolf watershed HSPF application (that previously run from 1970 to 1996) up to 2007. The introduction of yearly land use information generated four different HSPF models for the Wolf River watershed. Each of the models was run from 1997 to 2006. Simulated daily hydrographs at the outlet (Landon Station) were compared against measured hydrographs at the same outlet (USGS gage station 02481510). Model fit efficiencies were evaluated for each combination of HSPF model for the Wolf River Watershed using LIS Data and the time-series Land Cover Dynamics Data. Science Question: Can Land Cover Dynamics be used in conjunction with model-derived meteorological data to fill gaps in data for hydrologic simulation? Science Question: In many areas where detailed land cover or meteorological stations are not available or time-series data are interrupted, what data and methods may be used to create reliable models?

Land Cover Dynamics and Forcings Experiment NASA: MODIS MOD12Q1 Urban 2001 2002 Agricultural Rangeland Forest Water Wetlands Barren land NASA-LIS: Time-series LANDUSE/LANDCOVER PRECIPITATION EVAPO-TRANSPIRATION 2003 2004 BASE CASE

Land Cover Dynamics and Forcings Experiment Hydrographs for models using 2001 and 2004 land use MODIS datasets (for brevity, output for years 2002 and 2003 is not shown). Hydrographs exhibit only minor differences. Scatter-plots of observed vs. simulated daily stream flow rates are also very similar. Correlation for 2001 and 2004 models are R= 0.843 and R=0.845 respectively (corresponding to a common R 2 =0.71).

Forcings Experiment (SWAT & LIS) Although the Land Cover Dynamics Experiment incorporated NASA-LIS forcings (precipitation and evapo-transpiration) in calibrated HSPF models, the models were calibrated and validated with measured gage station data from USGS and NCDC. Science Question: In the absence of ground station meteorological data can NASA-LIS data be utilized to generate (independently) a calibrated watershed model? In this experiment, that validity is assessed by introducing NASA-LIS precipitation and potential evapo-transpiration data into a SWAT model application for the Wolf River watershed.

Forcings Experiment This figure presents the SWAT model for the Wolf River Watershed. The NLDAS grid is also shown.

Forcings Experiment The calibration and validation of the SWAT application to Wolf River Watershed is equivalent to the calibration of a watershed model for the same location using HSPF. A correlation coefficient of 0.79, a coefficient of determination R 2 =0.62 and a model fit efficiency of NS= 0.61 were achieved. This experiment support the valid use of NASA-LIS precipitation and potential evapo-transpiration data for generating a fully calibrated watershed model in the absence of weather-gage-station data.

Geoprocessing Experiment While there are several interfaces designed to handle geo-processing for providing parameterized information to HSPF and SWAT, an experiment was performed to assess an alternative geoprocessing tool: GEOLEM. The Geospatial Object Library for Environmental Modeling (GEOLEM) was used to pre-process topographical and land use datasets for generation of topographical and land use parameters for (HSPF). The export of those tables to the HSPF input file (UCI file) for creation of a new HSPF project is performed outside from the GEOLEM framework but those and other enhancements are part of current efforts of the core developers (USGS) and will be reported by that research team.

Geoprocessing Experiment Standard GIS geoprocessing software (BASINS) and GEOLEM output for geographical parameters is compared. For brevity, only slope and land use area parameter values are shown in the comparison. Slope ratio 0.07 0.06 0.05 0.04 0.03 0.02 BASINS/NED GEOLEM/NED GEOLEM/DEM GEOLEM/SRTM GEOLEM/IFSAR GEOLEM and BASINS produce exactly the same values of slope using NED and land use area. Therefore, the customized GEOLEM framework is equivalent to standard methodologies for obtaining parameterized geographical data. Results for other datasets show the usefulness of the new system in rapidly obtaining the same geographical parameters from different sources. Area km2 0.01 0 0 5 10 15 20 25 30 35 40 Sub-basin 30000 BASIN/GIRAS GEOLEM/NLCD GEOLEM/GIRAS GEOLEM/MODIS 25000 20000 15000 10000 5000 0 5 7 9 11 13 15 17 19 21 Sub-basin

Conclusions The Topography Experiment: Determined that the use of SRTM provides equivalent delineation results and topographical parameterized information for input into hydrological models. The Land Use Experiment: Showed the usefulness of MODIS 12Q1 for providing insight in the modeling of flow and sediments in an inland watershed using HSPF. The Combined Land Use and Topography Experiment: The combination of moderate resolution topographical datasets (such as SRTM, 30 m) and low resolution land use datasets (such as MODIS, 1000 m) produce good statistical fit between HSPF-simulated and measured stream flow hydrographs. Model fit coefficients (R 2 and NS) for the MODIS-SRTM combination range between 0.73 and 0.81 (perfect fit is 1.00). Therefore, MODIS 12Q1 and SRTM datasets can be used successfully for calibration and validation of coastal watersheds such as Wolf and Jourdan rivers catchments.

Conclusions Land Cover Dynamics Experiment: When hydrographs are compared for the whole period of simulation (1997-2006) model fit efficiency for the four models was calculated as NS=0.61 and coefficient of determination resulted in R 2 =0.71. Long-term simulation using MODIS 12Q1 land use products and NASA-LIS forcings provide good results. Forcings Experiment: The calibration and validation of the SWAT application to Wolf River Watershed was shown to be equivalent to the calibration of a watershed model for the same location using HSPF. A correlation coefficient of 0.79, a coefficient of determination R 2 =0.62 and a model fit efficiency of NS= 0.61 were achieved. This shows the validity of using NASA-LIS precipitation and potential evapo-transpiration data for generating a fully calibrated SWAT watershed model without requiring gage station data. Geoprocessing Experiment: GEOLEM and BASINS produce exactly the same values of slope and land use area. Therefore, the customized GEOLEM framework is equivalent to standard methodologies for obtaining parameterized geographical data.

Final Comments Gaps in meeting DST needs: The only limitation identified in these experiments is the spatial resolution of the MODIS 12Q1 dataset (1000 m) for short-span hydrological simulation. However, when long periods of hydrological simulation are required, the land use information contained by MODIS MOD 12Q1 seems to be equivalent to finer datasets. Recommendations NASA should consider the feasibility of generating land use datasets of finer spatial resolution than 1000 m. A 500-m resolution dataset would impact greatly on the use of this type of NASA product. This could be tested with the up-coming VIIRS implementation. Next Steps Introduce improved precipitation data in current HSPF and SWAT model applications. Expand the land use experiment to SWAT. Assess the combined effects of topographical and land use in SWAT simulations. Expand the exploration to water quality simulations (sediments, nutrients). Apply the know-how developed in the experiments to simulate watersheds abroad. Expand the experiments detailed in this report to watersheds that cover bigger geographical areas.

Papers and Proceedings Published Publications Alarcon, V. J., O Hara, C.G., Viger, R., Shrestha, B., and Mali, P., 2007c. Using and interoperable geoprocessing system for hydrological simulation. In: Computation in Modern Science and Engineering (Simos, T. E. and Maroulis, G., Eds.). American Institute of Physics, 1137-1140. Diaz, J.N., Alarcon, V.J., Duan, Z., Tagert, M. L., McAnally, W. H., Martin, J. L. and C.G. O Hara, 2008. Impacts of land use characterization in modeling hydrology and sediments for the Luxapallila Creek watershed, Alabama/Mississippi. Transactions ASAE. (Accepted). Alarcon, V. J. and O Hara, G. G., 2007a. Advanced Techniques for Watershed Visualization. Proceedings Applied Imagery and Pattern Recognition Conference, AIPR 2006, p 30. Alarcon, V. J. Zwaag, J, Moorhead, R.J., 2007b. Estimation of Estuary Phytoplankton Using a Web-Based Tool for Visualization of Hyper-Spectral Images, Proceedings Applied Imagery and Pattern Recognition, AIPR 2006, p 25. Alarcon, V. J. and O Hara, G. G., 2006. Using IFSAR and SRTM elevation data for watershed delineation of a coastal watershed. Proceedings American Society Photogrammetry and Remote Sensing (ASPRS-MAPPS) 2006 Fall Conference. San Antonio, Texas, November 6-10, 2006. Alarcon, V. J., O'Hara, C. G., McAnally, W., Martin, J., Diaz, J., Duan, Z., 2006.a HSPF-estimated flowrate sensitivity to topographical parameter values for three watersheds in Mississippi. Proceedings AWRA 2006 Spring Specialty Conference GIS & Water Resources IV, Houston, Texas, May 8-10, 2006. Alarcon, V. J., O'Hara, C. G., McAnally, W., Martin, J., Diaz, J., Duan, Z., 2006b. Influence of elevation dataset on watershed delineation of three catchments in Mississippi. Proceedings AWRA 2006 Spring Specialty Conference GIS & Water Resources IV, Houston, Texas, May 8-10, 2006. Posters Alarcon, V.J., C. G. O'Hara, R.Viger, B. Shrestha1, P. Mali1, D. Toll, T. Engman, 2007. Interoperable Geoprocessing for Rapid Prototyping of Landuse/Landcover, Topographical and Meteorological Datasets for Hydrological Simulation. AGU Fall 2007 Conference, December 10-14. San Francisco, CA. Alarcon, V.J., O' Hara, C. G., Viger, R., Toll, D., Diaz, J., Tagert, M. L., McAnally, W., Martin, 2006. Rapid Prototyping Capabilities for Evaluating Current and Future NASA Data in Multi-Scale Sensitivity Analysis in Watershed Hydrology Modeling. American Geophysical Union (AGU) Fall 2006 Meeting. 11-15 December 2006, San Francisco, California

Questions? THANK YOU! Chuck O Hara cgohara@gri.msstate.edu