ML and GIS-Based Approaches to Flood Prediction: A Comparative Study

Floods are one of the most devastating natural disasters and cause damage worth millions of dollars every year. The research on models predicting floods and advances in the methods applied have significantly reduced risk and loss of life and allowed better suggestions for policies to reduce harm to property due to floods. Empirical approaches use historical data to predict floods that might occur on the basis of statistics. Although this approach has received tons of attention in recent years thanks to advances in supporting technology, particularly within the areas of massive data and machine learning, it lacks in modeling generalized systems. The spatial distribution of the input features like meteorological conditions and other physical parameters can be captured by using physical models. Since the physical models are dynamic, they may overcome the empirical models which are less accurate and also lack generalizability. However, these models are data intensive, difficult to build, and require expertise in the domains. This paper reviews various hydrological modeling methods and predictive models and proposes a hybrid approach which combines the two approaches to predict floods in the vicinity of the river basins.