The second benefit of weather radar derives from an observation that the trajectory of storm dynamics (velocity, azimuth, and intensity) estimated from subsequent radar scans may contain predictive information. Thus a short-term (0--3 hours) prediction of the precipitation field may be informative. Such a prediction offers the potential for better localizing flash flood warnings and increasing their lead times, albeit at a lower diagnosticity.
In the 1980s research concentrated mostly on estimation of rainfall rates from radar reflectivities and on prediction of rainfall rates through projection of trends detected in a time series of reflectivity fields. Some of these efforts have been reviewed by Chen and Kavvas [1992] who introduced a radically new technique. Harnessing concepts from pattern recognition theory, the technique decomposes the radar image of the rain field into constituent polygonal contours, whose evolution in time and space is tracked through subsequent radar images and then projected into the future via an adaptive exponential smoothing scheme. By recomposing the projected contours at a desired lead time, a prediction of the rain field is obtained. The technique was validated for lead times up to 30 minutes on historical radar data from one storm with encouraging results.