A MODEL FOR REAL-TIME QUANTITATIVE RAINFALL FORECASTING USING REMOTE-SENSING .2. CASE-STUDIES

The performance of a real-time physically based rainfall forecasting m odel is examined using radar, satellite, and ground station data for a region of Oklahoma. Model formulation is described in an accompanying paper (French and Krajewski, this issue). Spatially distributed radar reflectivity observations are coupled with model physics and uncertai nty analysis through (1) linearization of model dynamics and (2) a Kal man filter formulation. Operationally available remote sensing observa tions from radar and satellite, and surface meteorologic stations defi ne boundary conditions of the two-dimensional rainfall model. The spat ially distributed rainfall is represented by a two-dimensional field o f cloud columns, and model physics define the evolution of vertically integrated liquid water content (the model state) in space and time. R ainfall forecasts are evaluated using least squares criteria such as m ean error of forecasted rainfall intensity, root mean square error of forecasted rainfall intensity, and correlation coefficient between spa tially distributed forecasted and observed rainfall rates. The model p erforms well compared with two alternative real-time forecasting strat egies: persistence and advection forecasting.