Short-term explicit forecasting and nowcasting of convection systems with WRF using a hybrid radar data assimilation approach

Forecasting and nowcasting convective storms are important and inherently challenging. At present, convection nowcasting (1 – 2 hours) is essentially based on simple extrapolation using radar measurements. Physics-based mesoscale numerical weather prediction (NWP) models often capture large-scale features of convection, but are greatly limited in forecasting the timing, locations and structures of high-impact convective cores. This is because of our a) very limited understanding of convective initiation, b) imperfect representation of various atmospheric physical processes, and c) inability to describe convective features in the initial conditions. In this paper we present a hybrid continuous data-assimilation (DA) scheme that combines variational and Newtonian-relaxation approaches, and uses radar and other observations as input to the Weather Research and Forecasting (WRF) model for explicit (convection-resolving) nowcasting and forecasting. This system encompasses three-data assimilation algorithms in the WRF modeling framework: WRF-3DVAR, WRF Grid-Nudging and WRF Observation-Nudging. WRF-3DVAR is employed to intermittently assimilate radar radial winds and hydrometeor reflectivities. These analyses are then nudged into the WRF model between analysis times, using the Grid-Nudging approach. This hybrid DA approach addresses the issue of “model spin-up” and “data rejection” that occurs during the first few hours of model forecasts with typical 3DVAR-based radar data assimilation systems that are used for short-term forecasting of convection. The third component of the system is the WRF station-based Observation-Nudging DA. The Observation-Nudging scheme allows us to effectively assimilate temperatures, winds and moisture that are measured by diverse platforms at regular and irregular times and locations. This scheme has been proven to be effective for mesoscale weather analysis and forecasting, where the background error covariance cannot be estimated properly, and where the available observations are not sufficient to depict the mesoscale circulations. Observation-Nudging allows the full-physics NWP model to fill in the data gaps by spreading observation information in space and time during the DA process. Retrospective forecasting experiments were conducted for the 10 – 16 June 2002 IHOP convection episode, to illustrate the capabilities of this modeling system.