An effective alternative for predicting coastal floodplain inundation by considering rainfall, storm surge, and downstream topographic characteristics

Abstract

The extent of coastal flooding is influenced by many factors such as the topography of the low-lying land, tidal level, rainfall pattern, inflow discharge collected from the upstream drainage area, etc. This study establishes a new methodology of effectively predicting the flooding process in coastal areas, and which is achieved by combining the recurrent neural network (RNN) model with the detailed analysis of different hydrological and geomorphological factors. The novelty of this study is to apply the topographic wetness index (TWI) of each grid to classify all inputs into multiple classes for separative training to improve the overall accuracy of flooding simulations. A numerical inundation model based on hydrodynamic equations was applied to investigate the behavior of coastal flooding in the temporal and spatial domain under a variety of model settings with different hydrologic conditions and it was utilized to generate the target inundation depths for the training of the RNN model. The relevance between the downstream topography, tidal level, rainfall intensity, and the spatial distribution of flooding in coastal areas is explored via the use of machine learning (ML) techniques. The focus of this study is to evaluate the proposed alternative method that allows for improving the efficiency and stability of forecasting coastal floods caused by rains and storm surges due to the approaching tropical cyclones. The method developed in this study is promising to replace the numerical inundation model to reinforce the model's stability and computational efficiency.