Extrapolative-statistical forecasts of radar reflectivity

A common approach to short-range precipitation forecasting involves the extrapolation of radar reflectivity fields that ha ve been analyzed in digital form on a map grid. In an attempt to. refi.ne this basic technique, a large number of extrapolatlveforecasts were prepared and statistically correlated with actual radar observations at the valid time. The relationships determined in this manner can then be applied if.l interp~·etfng other extrapolative forecasts. This extrapolatIve-statIstIcal method implicitly accounts for echo decay and IIncertainties in the extrapolative process, and is analogOlls to the Model Output Statistics approach often used to prodllce forecasts of sensible weather from numerical weather prediction model output. This study was based on observations obtained with Radar Data Processor 1/ at Oklahoma City, Oklahoma, during the period /985-1990. A large sample of 30and 60-minute extrapolativeforecasts of zero-tilt reflectivity, vertically-integrated liqllid, and /8-dBZ echo tops was prepared. The extrapolation velocity was estimated from digital pattern matching within a sequence of reflectivity images collected before the forecast initial time, orfrom the 700-mb environmental wind vector. Linear regression was used to relate the areas of zero-tilt reflectivity 2: 40 dBZ to the extrapolative forecasts of reflectivity variables. The probability forecasts produced by the regression equations 11'1'/:1' then verified 011 independent data. Categorical (yeslno)./orecastsfor40-dBZ reflectivity, valid at 30 minutes wlt!7in ~I /2-/\117 square region, were producedfrom the probabtllt)' forecasts. A threshold of 35% yielded a probability of detect~OJ: of O.~ with a false alarm ratio of 0.4 and bias of 1.2. SU17t1ar s/oll scores were obtained for 60-minute forecasts \'{{Iid within 20-km square regions.