Optimizing Precipitation Estimates Using Merged Observations and Model Output: A Case Study In The California Sierra Nevada Mountains

Abstract

We present a method based on optimal data assimilation techniques used today in weather analysis to blend together high-resolution precipitation forecasts with point observations at rain gages or other estimates (for instance, radar). This method is applied to several winter storms during 2005-2007 in the American River Basin (ARB) northeast of Sacramento, California. For these storms, several high-resolution (3km horizontal grid spacing) Weather Research and Forecasting (WRF) model runs were performed with different microphysical packages, and from these runs lagged ensemble precipitation forecasts were produced. Carefully screened rain gage observations over a small domain over the ARB were then merged with the mean ensemble forecasts to produce precipitation estimates optimized as described above. To assess these ‘optimum' fields, and compare to other forecasts and estimates, verification scores were derived using sets of gages withheld from each of several runs of the optimization procedure. For an extreme rainfall episode between 29 December 2005 and 1 January 2006, the optimal QPE analyses qualitatively resembled the WRF ensemble forecasts, and were less smooth than other gauge-only anlayses. Verification scores for this event revealed that the optimal QPE were superior to the other analyses. Finally, we discuss a potential application of these optimized QPE fields to operational monitoring of high-precipitation episodes in California and to decision making for water resource management.