Mapping flood risk using a workflow including deep learning and MCDM– Application to southern Iran

Abstract

Floods - an important risk that threatens many people worldwide – affect both the environment and human-made structures, and can cause loss of agricultural activities and life, economic challenges such as the destruction of infrastructure. Therefore, spatial maps of flooding probability can be useful to identify regions with high risk, these can be used to mitigate its negative consequences. Here, we developed a methodology to map flood risk in a catchment in southern Iran by combining a hazard map produced by a bidirectional long short-term memory (bLSTM) deep learning (DL) model, and a flood vulnerability map produced by a complex proportional assessment (COPRAS) model as a multi-criteria decision making (MCDM) model. Different environmental variables as lithology, vegetation cover, land use were mapped spatially, and a GrootCV was employed to identifying the most important variables controlling flood risk. Among various variables explored as controls flood risk, the variables extracted from a digital elevation model (DEM) (e.g., topographic wetness index (TWI), river density, topographic position index (TPI), stream power index (SPI), slope, elevation and distance to river) were recognized as the most effective features controlling the flood risk. Finally, a bLSTM model was employed to map the flood hazard. Its performance was assessed by the cumulative gain and Kolmogorov Smirnov (KS) tests. To map flood vulnerability, seven socio-economic variables were mapped as key controls, and then, analytical hierarchy process (AHP) and COPRAS models were employed to determine the weights of variables to map flood vulnerability. Finally, a flood risk model was generated by integration of the bLSTM and COPRAS. The results revealed that 23.2 %, 27.7 %, 18.7 %, 15.8 % and 14.6 % of the total study area are classified as very low to very high risk classes, respectively. Overall, our methodology based on DL and MCDM can employ to map flood risk and another disasters (e.g., landslide, land subsidence, soil erosion, etc.) in different climatic regions worldwide.