Assessing NEXRAD P3 Data Effects on Stream-Flow Simulation Using SWAT Model in an Agricultural Watershed

Radar-derived next generation weather radar (NEXRAD) process 3 (P3) data offer high spatial resolution precipitation data that might improve stream-flow simulation accuracy using watershed models. The objective of this study was to assess the performance of spatially averaged subwatershed-specific P3 data on stream-flow simulations by using the soil and water assessment tool (SWAT) hydrologic model. This study was the first to evaluate the P3 product in watershed modeling. The SWAT model was chosen to simulate hydrological processes in North Fork Ninnescah Watershed ($2,400{\mathrm{km}}^2$) located in south central Kansas. A precipitation gauge (PG) station was created for each subwatershed by using an area-weighted average of P3 precipitation estimates for all $16\text{-}{\mathrm{km}}^2$ grid cells intersecting the subwatershed. Average subwatershed area ($19{\mathrm{km}}^2$) was selected to correspond roughly with NEXRAD cell area to minimize spatial aggregation of radar precipitation data. The SWAT model was calibrated with both P3 data and PG data from January 1, 2002 to December 31, 2008. The P3-calibrated model was validated by using both different precipitation source data for the same simulation period (2002–2008) and the same source data with an independent simulation period (1996–2001). The PG-calibrated model yielded slightly higher Nash-Sutcliffe efficiency ($E_{NS}$), both daily (0.40) and monthly (0.60), than the P3-calibrated model (daily: 0.35, monthly: 0.52); percent bias was very good ($<\pm 2\%$) for both P3- and PG-calibrated models at all time scales. Validation $E_{NS}$ results, however, were diminished using PG data with the P3-calibrated model (daily: 0.27, monthly: 0.52) but remained similar using P3 data with the PG-calibrated model (daily: 0.40, monthly: 0.52). The PG data appeared to demonstrate variable uncertainty among stations that was not evident for the P3 data, providing incentive for the use of P3 data. This study found evidence that the SWAT model, when run using more spatially representative precipitation data (in this case, P3), was less sensitive to minor variations in model calibration parameters. On the basis of this initial assessment of P3 data for hydrologic modeling, bias adjustment to reduce the effect of a small number of extreme P3 precipitation values may have improved results but was not required to produce watershed-outlet stream flow results similar to using PG data in this study.