An ensemble strategy for high-resolution regional model forecasts

Summary An ensemble methodology is proposed for very high-resolution regional model Quantitative Precipitation Forecasts. To facilitate a systematic study, the model and the boundary conditions are assumed to be perfect. The generation of perturbations is derived on the basis that the largest errors in precipitation forecasts at very high resolutions arise from miss-specified diabatic heat sources and sinks which feedback erroneously to the grid scale variables in the initial state. This methodology is tested in a Proxy Observed System Simulation Experiment (POSSE) involving an intense cyclone over eastern Canada. The perturbations of wind and temperature in this ensemble strategy are obtained as normalized coefficients of a Combined Empirical Orthogonal Function analysis of the difference fields between the control and the diabatically initialized model runs. These perturbations are added to and subtracted from the control initial state to obtain a set of two perturbed initial states. Several such perturbed initial states are obtained from initializing observed rain rates at different times close to the time of the analysis. The results from the POSSE reveal that the Quantitative Precipitation Forecast of the ensemble mean outperforms the control model run. Received August 8, 2000/Revised November 1, 2000