H11G-0978:
Impact of Spatial Scale on Calibration and Model Output for a Grid-based SWAT Model

Monday, 15 December 2014
Garett Pignotti1, Vamsi Krishna Vema2, Hendrik Rathjens3, Cibin Raj1, Younggu Her1, Indrajeet Chaubey1,4 and Melba M Crawford5, (1)Purdue University, Agricultural and Biological Engineering, West Lafayette, IN, United States, (2)Indian Institute of Technology Madras, Chennai, India, (3)University of Kiel, Department of Geography, Kiel, Germany, (4)Purdue University, Earth, Atmospheric, and Planetary Sciences, West Lafayette, IN, United States, (5)Purdue University, Lyles School of Civil Engineering, West Lafayette, IN, United States
Abstract:
The traditional implementation of the Soil and Water Assessment Tool (SWAT) model utilizes common landscape characteristics known as hydrologic response units (HRUs). Discretization into HRUs provides a simple, computationally efficient framework for simulation, but also represents a significant limitation of the model as spatial connectivity between HRUs is ignored. SWATgrid, a newly developed, distributed version of SWAT, provides modified landscape routing via a grid, overcoming these limitations. However, the current implementation of SWATgrid has significant computational overhead, which effectively precludes traditional calibration and limits the total number of grid cells in a given modeling scenario. Moreover, as SWATgrid is a relatively new modeling approach, it remains largely untested with little understanding of the impact of spatial resolution on model output. The objective of this study was to determine the effects of user-defined input resolution on SWATgrid predictions in the Upper Cedar Creek Watershed (near Auburn, IN, USA). Original input data, nominally at 30 m resolution, was rescaled for a range of resolutions between 30 and 4,000 m. A 30 m traditional SWAT model was developed as the baseline for model comparison. Monthly calibration was performed, and the calibrated parameter set was then transferred to all other SWAT and SWATgrid models to focus the effects of resolution on prediction uncertainty relative to the baseline. Model output was evaluated with respect to stream flow at the outlet and water quality parameters. Additionally, output of SWATgrid models were compared to output of traditional SWAT models at each resolution, utilizing the same scaled input data. A secondary objective considered the effect of scale on calibrated parameter values, where each standard SWAT model was calibrated independently, and parameters were transferred to SWATgrid models at equivalent scales. For each model, computational requirements were evaluated. While the definition and choice of input data resolution widely depends on the specific application and area of interest, this study will serve as a general guide in identifying an optimal range of input spatial resolution for SWATgrid applications.