A probabilistic pluvial flood warning model based on nest som using radar reflectivity data

The occurrence and intensity of extreme weather events have increased under climate change, making flooding disasters more likely during the flood season from May to November in Taiwan. The current early warning system for flooding disasters developed by the Water Resources Agency in Taiwan relies on the density of rain gauges, which limits its effectiveness. To improve this system, our research collected historical radar reflectivity and rainfall data in the flood-prone area at the Zhonghua village of Taipei City. An unsupervised neural network called the self-organizing map (Clayton and Emery) is applied to establish the relationship between radar reflectivity and rainfall observations, enabling the analysis of clustering vectors corresponding to pluvial flood disaster events. A Nest SOM-based pluvial flood warning model was proposed for identifying flooding hot zones and delivering probabilistic flood warning information. Based on radar reflectivity characteristics along with corresponding rainfall intensity and frequency, the proposed model was evaluated during extreme events to demonstrate its applicability and provide probabilistic warning information prior to flood disasters. Consequently, the model provides considerable practical value in enhancing flood disaster management.