Application of GIS and Logistic Regression for Flood Susceptibility Mapping in Nilwala River Basin, Sri Lanka

Abstract

Flood susceptibility mapping is a crucial element in flood management. This study aimed to investigate the applicability of logistic regression (LR) and GIS tools to produce a flood susceptibility map for Nilwala river basin in Sri Lanka. Four conditioning factors: elevation, slope angle, distance from river and land-use, were selected as flood conditioning factors. Flood inventory database consisted of 205 flood and 199 non-flood locations; out of these, 70% was selected as the training dataset, and the remaining 30% was taken as the testing dataset. The flood susceptibility map was developed in ArcGIS 10.5, based on the LR coefficients derived in R statistical language. The four flood conditioning factors were statistically significant at P < 0.05. The Receiver Operating Characteristics curve method was used to validate the model. Area under the Success Rate Curve and Prediction Rate Curve were 86% and 87%, respectively. The Area Under the Curves values reveal the model's excellent compatibility and predictability. Therefore, a high degree of confidence can be placed on the model to identify flood vulnerable areas in Nilwala basin. Hence, the developed susceptibility map might be a vital decision-making tool for water managers.