Evaluating the effects of image filtering in short-term radar rainfall forecasting for hydrological applications

Abstract

Radar rainfall nowcasting at short lead times has important hydrometeorological applications in the fields of weather prediction and flood forecasting. The predictability of rainfall events can vary significantly with scale as smaller storm features are less predictable than the storm envelope motion. As a result, various techniques have been developed for filtering a radar image and deriving short-term forecasts from the more predictable, larger storm scales. In this study, the effects of image filtering on radar nowcasting performance using the Storm Tracker Nowcasting Model (STNM) are evaluated. Radar rainfall nowcasts are evaluated for three storms exhibiting varying degrees of organisation over the Arkansas-Red River basin. In each case, it is found that the nowcast skill decreases with the forecast lead time, increases with the verification area used around a forecast location, and decreases with higher rainfall thresholds. Furthermore, it is demonstrated that a set of properly tuned filtering nowcasts are superior to simple 'persistence' and slightly better than 'uniform advection'. At the scale of a large hydrologic basin (∼ 6000 km2), filter-based nowcasting is shown to capture the temporal variation in rainfall amount and its spatial distribution based on a set of catchment-based metrics. Finally, a method for relating changes in nowcasting skill to errors associated with storm dynamics not captured by image filtering techniques is evaluated.