An extrapolative-statistical algorithm for predicting 0-1 hour rainfall within a radar umbrella has been implemented within the Advanced Weather Interactive Processing System (AWIPS). The algorithm produces probabilities that rainfall will exceed 0.1, 0.25, 0.5, and 0.75 inch, and a forecast of the rainfall amount category (< 0.1, 0.1-0.24, 0.25-0.49, 0.5-0.74, and 0.75 inch). These forecasts are for each box of a 4-km grid covering the umbrella to a distance of 180 km from the radar. A separate set of forecasts giving the probability that rainfall will exceed one inch in the vicinity of a convective storm is also produced. Input to the algorithm includes base reflectivity and vertically-integrated liquid (VIL) graphic products, and output from the WSR-88D Storm Track Algorithm.
The algorithm was developed by applying the Model Output Statistics approach to a simple extrapolation model for radar echo fields. Extrapolation forecasts of reflectivity and VIL were prepared for a sample of 113 historical cases. Statistical predictors such as rainfall accumulation and time-averaged VIL were derived from the forecasted radar fields. These predictors were collated with WSR-88D Stage III radar/gauge rainfall analyses observed during the valid period. Statistical regression was then used to find the optimum combination of predictors for specifying the probabilities of the various rainfall thresholds. These regression equations are applied to real-time extrapolation forecasts, yielding rainfall probability fields that accurately reflect radar echo intensity, shape, and motion. The categorical forecast field is derived by applying thresholds to the probability values.
In an operational setting, the extrapolation velocity is taken from the mean storm cell velocity output by the Storm Track Algorithm during convective events, and is derived by lag-correlation pattern matching between recent reflectivity image pairs in stratiform rain events. The 4-km gridded output is displayed as contoured probability isopleths and isohyets, while the storm-cell-based heavy rainfall probability is displayed with other cell characteristics in a user-requested popup box.
In a verification study of independent cases, we found that the 0.1-inch probabilities, when converted to yes/no forecasts at a threshold of 30%, yielded a probability of detection (POD) of 0.79 and a false alarm ratio (FAR) of 0.47. For the 0.75-inch probabilities, a yes/no threshold of 10% yielded a POD of 0.59 and FAR of 0.78. For the five-level categorical forecasts, 23% of the forecasts for 0.1 inch or more were verified in the correct category, and 79% were verified within one category of the correct one