Model Bias in a Continuously Cycled Assimilation System and Its Influence on Convection-Permitting Forecasts

Abstract

uring the spring 2011 season, a real-time continuously cycled ensemble data assimilation system using the Advanced Research version of the Weather Research and Forecasting Model (WRF) coupled with the Data Assimilation Research Testbed toolkit provided initial and boundary conditions for deterministic convection-permitting forecasts, also using WRF, over the eastern two-thirds of the conterminous United States (CONUS). In this study the authors evaluate the mesoscale assimilation system and the convection-permitting forecasts, at 15- and 3-km grid spacing, respectively. Experiments employing different physics options within the continuously cycled ensemble data assimilation system are shown to lead to differences in the mean mesoscale analysis characteristics. Convection-permitting forecasts with a fixed model configuration are initialized from these physics-varied analyses, as well as control runs from 0.5° Global Forecast System (GFS) analysis. Systematic bias in the analysis background influences the analysis fit to observations, and when this analysis initializes convection-permitting forecasts, the forecast skill is degraded as bias in the analysis background increases. Moreover, differences in mean error characteristics associated with each physical parameterization suite lead to unique errors of spatial, temporal, and intensity aspects of convection-permitting rainfall forecasts. Observation bias by platform type is also shown to impact the analysis quality.