Spatial analysis of flood susceptibility in Coastal area of Pakistan using machine learning models and SAR imagery

Abstract

Flooding is one of the most important and challenging natural catastrophes to anticipate, and it is getting more intense and frequent. The coastal areas in Pakistan, like Karachi, are highly vulnerable to flooding, especially during the monsoon rains, which cause immense environmental and socioeconomic damage. A massive flood badly destroyed the study area in 2022. We examined the flood susceptibility in the coastal area of Pakistan using various machine learning algorithms such as Extreme Gradient Boosting, Random Forest, and K Nearest Neighbor. Flood points were identified and validated using Landsat data, Google Earth, and news sources to generate a flood inventory map. A total of 262 flood spots were selected and randomly divided into 70% for training and 30% for validation. Susceptibility maps were validated using area under the receiver operating characteristic (ROC) curve and confusion matrix. In this research, remote sensing data was utilized to validate flood-prone areas using the Sentinel Application Platform for remote sensing image evaluation. The RF model achieved outstanding classification accuracy with an area under the curve (AUC) value of 0.983, accuracy of 0.950, kappa value of 0.900, specificity of 0.992, and sensitivity of 0.902. The research is valuable since the suggested models are being evaluated for the first time in the coastal area of Pakistan to measure flood vulnerability. The flood risk map assists coastal area planners and regulatory agencies in managing and mitigating flood events. Despite its simplicity, the approach used in this study exhibits high precision, making it applicable for expert knowledge-based flood mapping in other regions.