Multi-scale Comparison of Stage Iv Nexrad (mpe) and Gauge Precipitation Data for Watershed Modeling

Watershed hydrologic and fate-andtransport models are widely used to forecast water quantity and quality responses to alternative land use and climate change scenarios. The ability of such tools to forecast changes in ecosystem services with reasonable accuracy depends on calibrating reliable simulations of streamflow, which in turn require accurate climatic forcing data. Precipitation is widely acknowledged to be the largest source of uncertainty in watershed modeling. Most watershed models are designed to easily incorporate publiclyavailable precipitation data from rain gauges (e.g., data provided by the National Climatic Data Center), but several additional data products from ground-based radar and satellite-based sensors are now available and can potentially be used to generate more precise, spatially-explicit precipitation estimates. Here, we investigate whether the use of higher-resolution Multisensor Precipitation Estimator (MPE, also known as Stage IV NEXRAD) data can improve the accuracy of daily streamflow simulations using the Soil and Water Assessment Tool (SWAT) watershed hydrology model. Simulated vs. observed streamflow and model calibrations are compared for two Piedmont sub-basins of the Neuse River in North Carolina (21 and 203 km watershed area) for an 8 year simulation period (January 1, 2002 to August 31, 2010). MPE simulations led to more accurate simulations of daily streamflow magnitude and frequency measures than gauge data, and differences were more pronounced in the smaller watershed. Compared with USGS-observed flows, MPE simulations produced R values of 0.64 and 0.54 for the larger and smaller watershed, respectively, while gauge data produced R values of 0.19 in both watersheds. NashSutcliffe Efficiency and other goodness-of-fit indices also showed much better simulations associated with MPE data. Additionally, the temporal structure of MPEsimulated streamflows more closely approximated that of the observed streamflows. These results are likely extendable to the Piedmont of the broader southeastern U.S. Ongoing research on this topic investigates additional spatial and temporal scales, as well as additional precipitation data types. INTRODUCTION AND BACKGROUND Simulation of streamflow, sediment, and dissolved constituents requires climatic forcing data. Temperature can be reasonably estimated for hydrologic modeling from a sparse network of stations within and surrounding the study watershed (Attorre et al., 2007). However, accurate representation of precipitation spatial and temporal variability from available resources has proven to be a challenge for hydrologic modeling. Failure to incorporate such variability potentially introduces large amounts of uncertainty to hydrologic and fate-andtransport modeling efforts (Jordan, 2000; Andréassian et al., 2001; Schuurmans and Bierkens, 2007; Villarni et al., 2008). There are two predominant approaches to evaluating precipitation data sources. In the first approach, interpolations or area averages from a network of precipitation gauges is treated as the “actual” precipitation as a basis of comparison for other sources of precipitation data (e.g., sparser gauge networks, radar, satellite). In the absence of a very dense network of rain gauges, it is inappropriate to treat any rainfall data source as the actual precipitation, because of the known uncertainty of all available data types (Schuurmans and Bierkens, 2007; Villarni et al., 2009; Habib et al., 2009; Mandapaka et al., 2009). Because of this, many studies have adopted an alternative approach using streamflow simulations from a watershed model as an independent assessment of the precipitation data accuracy (e.g., Borga, 2002; Su et al., 2008; Schuurmans and Bierkens, 2007; Tobin and Bennett, 2007; Starks and Moriasi, 2009). In this approach, the disparities between streamflow simulations are attributed to differences in precipitation data accuracy. This technique is especially appropriate for studies whose objective is to evaluate the potential of various precipitation data sources as forcing data for hydrologic modeling and is employed in this study. Recent studies evaluating and comparing precipitation data sources have shown mixed results, suggesting that there is no universally optimal precipitation data source for hydrologic modeling. Hossain et al. (2004)