Analyzing the spatial interactions between rainfall levels and flooding prediction in São Paulo

Rainfall is one of the primary triggers for many geological and hydrological natural disasters. While the geological events are related to mass movements in land collapse due to waterlogging, the hydrological ones are usually assigned to runoff or flooding. Studies in the literature propose predicting mass movement events as a function of accumulated rainfall levels recorded at distinct periods. According to these approaches, a two-dimensional rainfall levels feature space is segmented into the occurrence and non-occurrence decision regions by an empirical critical curve (CC). Although this scheme may easily be extended to other purposes and applications, studies in the literature need to discuss its use for flooding prediction. In light of this motivation, the present study is unfolded in (1) verifying that defining CCs in the rainfall levels feature space is a practical approach for flooding prediction and (2) analyzing how geospatial components interact with rainfall levels and flooding prediction. A database containing the rainfall levels recorded for flooding and non-flooding events in Sao Paulo city, Brazil, regarding the period 2015-2016, was considered in this study. The results indicate good accuracy for flooding prediction using only partial rain, which can be improved by adding physical characteristics of the flooding locations, demonstrating a direct correlation with spatial interactions, and rainfall levels.