Integrating Remote Sensing and Artificial Intelligence based Techniques for Investigating Regional Flood Susceptibility to Improve Flood Mitigation Planning

Floods are extreme events that cause huge loss of lives and properties. The flood events are expected to be more intensified and recurrent in the future due to climate change. It is required to develop robust flood mitigation strategies under climate change to mitigate the flood risk especially in the basins with limited data or ungauged basins. However, flood mitigation planning requires a huge amount of in-situ data of pre and post flood events which is not possible in data-scarce or ungauged river basins and almost inaccessible in the impassable and high-altitude complex terrain. The availability and accessibility of remote sensing data provides accurate and precise information regarding pre and post flood events in these regions.  The critical review of published literature reveals that the concept of model regionalization could be the scalability would provide the robust strategies for planning flood mitigation under climate change especially in these regions which involves transfer of knowledge from gauged to data-scarce or ungauged basins. However, the inefficiency of conventional process-based models in regionalization of model has motivated the researchers to think about the Artificial Intelligence (AI) data-driven approach. The present study combines remote sensing with AI approach to investigate the scope of regional flood susceptibility model development. The model development utilizes the remote sensing derived flood affecting parameters (or indicators) such as terrain, morphological, metrological. It has been first developed in data-rich basins and then transferred its knowledge to data-scarce or ungauged basins. The remote sensing derived historical flood records were used to generate the ground control points for training (70%) and testing (30%) of the model. To accomplish the objectives of enquiring the scope of regional flood susceptible model, the present study has chosen the two smaller sub-basins, one from the Krishna River basins, Maharashtra and the other from the Lower Ganga basin of Bihar. The chooses sub-basin from the Krishna River basins has been used for model development and the sub-basin from Lower Ganga basin of Bihar has been considered to investigate the scalability of the developed model for the regional AI-based model for flood susceptibility. The results of statistics F-1 score and Receiver Operating Characteristic (ROC)-Area Under Curve (AUC) have shown good performance of the model during training and testing. It also shows good performance during the model scalability check that advocates developed model for regional flood susceptibility. However, it suggested to apply fine tuning for future improvement of the model.  It has been concluded that the integration of remote sensing with AI-based could help in the development of good regional flood susceptible model which could be beneficial for policymakers in evolving enhanced strategies for mitigating futuristics floods especially in the data-scarce or ungauged basins.

Keywords: Regionalization, remote sensing, Artificial Intelligence, data-scarce or ungauged, flood mitigation planning.