Zonation of flood prone areas by an integrated framework of a hydrodynamic model and ANN

Abstract

Limited flood zoning regulations and lack of flood control response units in developing countries make flood problems more severe. This study presents a new framework for categorizing a floodplain into critical risk zones by considering hydraulic and topographical aspects related to flood zoning. The framework was developed by integrating output of the MIKE Hydro River Model with an artificial neural network (ANN) technique which was explored in the lower part of Damodar river basin (Jharkhand, India). A total of nine flood causing factors were selected in three layers of ANN architecture which were optimized by a grid search technique. A confusion matrix was employed to check the unevenness and disproportionality in datasets from which were calculated F1 score values for low (0.815), moderate (0.731), high (0.818) and critical (0.64) zones with best overall accuracy of 75.06%. The results were presented in a GIS environment which shows the model correctly predicted 16, 38, 54 and 24 sites under critical, high risk, moderate risk and low risk zones respectively. Elevation and distance from the river were the most sensitive parameters. Further, this study contributes towards flood susceptibility mapping thereby supporting hydrologists in the course of action and decisions for combating floods in watersheds.

Keywords

• ann
• data driven model
• flood zoning
• hydrodynamic model
• machine learning