Flash-flood susceptibility modelling in a data-scarce region using a novel hybrid approach and trend analysis of precipitation

Recently, the incidence of heavy rainfall events and associated flash floods have encouraged us to investigate long-term trends in extreme rainfall and flash flood vulnerability mapping. Thus, in this study, a hybrid model was designed by integrating weight of evidence and Naive Bayes (WOE-NB) to identify areas in Uttarakhand prone to flash floods, and we compared its ability with that of AdaBoost. Furthermore, the significance of long-term rainfall trends was evaluated using Mann-Kendall, modified Mann-Kendall, and innovative trend analysis (ITA), and extreme rainfall events were examined for 51 years (1970-2020). Results showed the WOE-NB and AdaBoost had acceptable goodness of fit (area under the curve = 0.969 and 0.973, respectively). Moreover, ITA can identify some important patterns based on on-trend results that other tests cannot. The return period revealed about 97.54% of the flash floods were caused by normal rainfall, with 2.45% being caused by severely abnormal rainfall.