Preparing coastal erosion vulnerability index applying deep learning techniques in Odisha state of India

Abstract

With the changing climatic and anthropogenic conditions, the natural ecosystem especially the coastal zones is at great risk. The ocean is engulfing the land through the process of coastal erosion and this is becoming a great threat to coastal communities by forcing them to relocate from their homes and destroying their livelihoods. This study has tried to use machine learning algorithms for the first time to find the probability of the vulnerability associated with this hazard along the coast of Odisha state of India using a total of 32 factors involving environmental and socio-economic conditions. A total of 2500 locations have been used to create support vector machine (SVM), random forest (RF), shallow neural network (SNN), deep neural network (DNN), and convolutional neural network (CNN) models. Various accuracy metrics have been calculated which showed the RF model outperformed all with an accuracy score of 0.96. This is followed by CNN (0.93), DNN (0.91), SVM (0.88), and SNN (0.88). Further, to find the impact of all the factors in the model sensitivity analysis has been performed. Factor importance analysis by RF has been performed at state and district levels to understand the influence of various parameters in this disaster. This novel method will broaden the approach which we use to analyze this calamity and serve as an aid in the decision-making process of the concerned authorities.