Development of the Method for Flood Water Level Forecasting and Flood Damage Warning Using an AI-based Model

This study reviewed the applicability of AI-based models to predict flood water level and evaluate damage in small rivers with short arrival times. The Namyangju-si (Jingwan Bridge) watershed, where most warnings have occurred, was selected as target study. Rainfall data from 2008 2020 were collected for watershed. A total 40 rainfall events identified when 1m or higher June September, corresponding season. Additionally, forecasting performed using such deep neural network (DNN), long term memory (LSTM), storage function models. Predictive power evaluation revealed DNN model displayed lowest normalized root mean square error (NRMSE) a value 0.06. concludes that there are issues existing warning heavy rain standards due variability, correlation occurrence caused by rain, application consistent nationwide. To solve this issue, cause classified risk assessment criteria established linking data. develop an optimal classification prediction based on criteria, two applied: XGBoost random forest model. Evaluation predictive F1-score 0.92, indicating excellent power. Based presented herein, technique results can be used basic disaster managers’ decision-making.