Evaluating Design Flood Levels Using a Machine Learning-based Costal Compound Flood Level Model

Abstract

Coastal floods, a type of compound disaster, results from the interactions between rainfall and tides. However, in Korea, the design flood level for coastal areas typically relies on the average water level according to the tide, ignoring this interaction. This study aimed to address this gap by developing a machine learning-based compound flood level model that incorporates the interactions between rainfall and tides. It also evaluated the adequacy of the design water level in the Taehwa River basin, a representative tidal river. Before developing the model, Discrete Wavelet Transformation (DWT) was used to identify the components of the compound flood level. The DWT results indicated that the compound flood level comprised three elements: rainfall-runoff, tides, and noise. A Long Short-Term Memory (LSTM) model was then created using these elements as input to estimate the compound flood level for a 200-year frequency, which is the design frequency of the Taehwa River. The compound flood level calculated by the model was 6.15 m, which is higher than the established design flood level of the Taehwa river (5.79 m). This discrepancy indicates that the current flood measurements for the Taehwa River are inadequate in accounting for compound flooding. Based on these findings, we recommend incorporating the compound flood concept when calculating the design water levels in coastal areas.