Performance Evaluation of a Substituted Topography-based Model To Forecast Rainfall and tide-induced Lowland Flooding

Abstract

A variety of factors, including rainfall distribution, downstream tide levels, upstream contributing areas, and terrain characteristics, can influence the extent of flooding disasters. The objective of this study is to develop a hybrid model that integrates the analysis of hydrological and geomorphological factors in the catchment with machine learning algorithms, thereby providing efficient flooding information while addressing the issue of numerical instability. Various environmental factors are examined to determine the model inputs necessary for forecasting the spatial distribution of inundation depths using the proposed AI-based hybrid model. A key contribution of the proposed model is the use of informative indices for preprocessing large volumes of input data prior to model training, thereby enhancing the accuracy of inundation depth forecasts. Additionally, a classification algorithm, the Self-Organizing Map (SOM) network, is adopted for preprocessing input data, emphasizing the physical significance of the methodology. To ensure reliable inundation-depth data during the model training phase, a numerical integration model based on theoretical governing equations for floodplain simulations in lowland areas is also applied. The proposed methodology offers an alternative approach for real-time coastal flooding simulation and forecasting, with advantages in efficiency, stability, and predictive accuracy.