Comparison of Spatial and Temporal Rainfall Characteristics of WRF-Simulated Precipitation to gauge and radar observations

Weather Research and Forecasting (WRF) meteorological data are used for US EPA air and water quality modeling applications, within the CMAQ modeling system to estimate wet deposition and to evaluate future climate and land-use scenarios. While it is not expected that hindcasting applications of WRF simulations match observed rainfall on a day-to-day or individual event basis, it is important that the overall spatio-temporal structure of precipitation events represents reality. It has been shown that contaminant fate-and-transport is strongly event-dependent, and the temporal structure of precipitation (and subsequent streamflows) is a major driver of instream flows relating to habitat suitability, contaminant fluxes, dilution, water supply, etc. The spatial and temporal variability of WRF-simulations in the North Carolina Piedmont and Coastal Plain was compared to two observed precipitation datasets, interpolated National Climate Data Center (NCDC) gauge and Multisensor Precipitation Estimate (MPE) radar data. NCDC data are point data comprised of rain-gauge observations, which we interpolated to the 12 km WRF grid using co-kriging (with elevation as the covariate). MPE data, also known as Stage IV NEXRAD, are Doppler-radar derived, HADS-adjusted rainfall estimates at a 4 km resolution, which we resampled to match the 12 km WRF grid. Variographic properties were used to compare spatial structure of rainfall across daily, monthly, seasonal, and annual rainfall totals during the five-year study period. In addition, the variography of a sample of storm events stratified by intensity and type (e.g., frontal , local convective, and tropical cyclones) was compared. Three-way ANOVA was used to compare variographic parameters across the datasets. Temporal structure was compared using partial autocorrelation functions and seasonal decompositions. Evaluation of modeled precipitation spatial and temporal structure, as compared with two observed datasets, allows insights into errors that propagate through stages of water quality, future scenarios, etc. when using WRF simulations. These results will guide future improvements to WRF simulations used within the EPA air and water modeling programs. These results also demonstrate differences in the spatio-temporal structure of the two observed precipitation datasets and provides guidance for use of gauge and radar precipitation data in hindcasting modeling efforts for understanding watershed function and process.