Dynamic Pluvial Flash Flooding Hazard Forecast Using Weather Radar Data

Abstract

Pluvial flash floods are among the most dangerous weather-triggered disasters, usually affecting watersheds smaller than 100 km2, with a short time to peak discharge (from a few minutes to a few hours) after causative rainfall. Several warning systems in the world try to use this time lag to predict the location, extent, intensity, and time of flash flooding. They are based on numerical hydrological models processing data collected by on-ground monitoring networks, weather radars, and precipitation nowcasting. However, there may be areas covered by weather radar data, in which the network of ground-based precipitation stations is not sufficiently developed or does not even exist (e.g., in an area covered by portable weather radar). We developed a method usable for designing an early warning system based on a different philosophy for such a situation. This method uses weather radar data as a 2D signal carrying information on the current precipitation distribution over the monitored area, and data on the watershed and drainage network in the area. The method transforms (concentrates) the 2D signal on precipitation distribution into a 1D signal carrying information on potential runoff distribution along the drainage network. For sections of watercourses where a significant increase in potential runoff can be expected (i.e., a significant increase of the 1D signal strength is detected), a warning against imminent flash floods can be possibly issued. The whole curve of the potential runoff development is not essential for issuing the alarm, but only the significant leading edge of the 1D signal is important. The advantage of this procedure is that results are obtained quickly and independent of any on-ground monitoring system; the disadvantage is that it does not provide the exact time of the onset of a flash flooding or its extent and intensity. The generated alert only warns that there is a higher flash flooding hazard in a specific section of the watercourse in the coming hours. The forecast is presented as a dynamic map of the flash flooding hazard distribution along the segments of watercourses. Relaying this hazard to segments of watercourses permits a substantial reduction in false alarms issued to not-endangered municipalities, which lie in safe areas far away from the watercourses. The method was tested at the local level (pluvial flash floods in two small regions of the Czech Republic) and the national level for rainfall episodes covering large areas in the Czech Republic. The conclusion was that the method is applicable at both levels. The results were compared mainly with data related to the Fire and Rescue Service interventions during floods. Finally, the increase in the reliability of hazard prediction using the information on soil saturation is demonstrated. The method is applicable in any region covered by a weather radar (e.g., a portable one), even if there are undeveloped networks of rain and hydrometric gauge stations. Further improvement could be achieved by processing more extended time series and using computational intelligence methods for classifying the degree of flash flooding hazard on individual sections of the watercourse network.