Flood Susceptibility Mapping of Maharashtra State by Impact Analysis of Climate Change using Machine Learning

Abstract

Flood is one of the prominent climate-induced disasters (CIDs), which causes huge dam ages, financial losses, and casualties every year across the world. Moreover, the intensities and damages of floods are prone to change due to future climate scenarios. In order to analyze the impact of climate change on flood patterns in different regions of the world, researchers apply machine learning-based models to learn from past data and simulate the scenarios of floods in the future. A historical dataset of the daily precipitation, minimum and maximum temperatures, and occurrence of flood events from 2001 to 2021 in all the districts of Maharashtra state in India has been collected from the India Meteorological Department and prepared for this work. In this paper, first, we derived several important factors to influence flood patterns as parameters. Then we considered those parameters to build the machine learning models such as Artificial Neural Networks (ANN), Light Gradient-Boosting Machines (LightGBM), and Least Squares Support Vector Machines (LSSVM) for estimating the approximate number of occurrences of floods till 2100 on different shared socioeconomic pathways (SSP) scenarios. Based on our simulation exper iments for data analytics, we observed that LightGBM performed the best in the validation phase giving an F1 score of 0.895 and a ROC-AUC score of 0.863. Moreover, we also used LightGBM to simulate future scenarios in Maharashtra state. This work introduces a novel approach to predicting CIDs by leveraging data from past disasters, global climate models, and climate change measurements to provide a perspective on predictions of the CID, e.g., floods in this case. This study can be utilized to predict the impact of climate change on floods and subsequently help the local government bodies and disaster management author ities to plan and prepare accordingly.