Flood Susceptibility Modelling by Advanced Convolutional Neural Networks (CNN) in the foothills of Southern Western Ghats, Kerala, India

Abstract

Abstract The intensity and frequency of extreme events have increased significantly in the past few years due to climate change, leading to more severe and devastating floods worldwide. In India, Kerala state has witnessed the most catastrophic floods of the century in the past five years. Thus, accurate flood susceptibility models are required for effective risk assessment and disaster management. In the present study, Machine Learning-based flood susceptibility models are developed for one of the severely affected districts, Kottayam, in the foothills of the Southern Western Ghats of Kerala state in India. The performance of SVM, tree-based XGBOOST, and Deep-Learning CNN models have been evaluated in flood susceptibility modelling. The performance of candidate models is evaluated using the Area Under the Curve of the Receiver Operating Characteristic curve (AUC-ROC). The models are validated using Overall accuracy, Precision, Recall, Specificity, and F1- score. CNN model outperformed SVM and XGBOOST. The AUC - ROC for SVM, XGBOOST, and CNN is 0.96, 0.97, and 0.99, respectively. The flood susceptibility model developed in the present study will be helpful in better disaster preparedness and the development of tailored flood mitigation plans, which would eventually reduce the impact of floods in the coming years.