Runoff Prediction Using Deep Learning Model in The Phetchaburi

Abstract

Flooding in Mueang District, Phetchaburi Province, is often caused by heavy rainfall in the area, combined with water discharge from the Kaeng Krachan Reservoir. This results in rising water levels in the Phetchaburi River, leading to overflow and flooding in economic zones and residential areas. Flood forecasting is therefore a crucial measure for issuing early warnings and preparing response plans to minimize damage to lives and property. This study applied deep learning models using Recurrent Neural Networks (RNNs) to predict water flows 24 hours in advance at the B.10 runoff station. The case study was divided into two scenarios for predicting water discharge: Case 1 involved Multi-Step forecasting, while Case 2 focused on Single-Step forecasting. The results indicated that during the peak flooding season, the model was able to forecast maximum water discharge values closely aligned with measured values, due to the application of the timing of peak water flows. However, the statistical indices used to evaluate the model's performance in both scenarios showed little difference. Specifically, when comparing forecasts made 6, 12, and 24 hours in advance, the Nash-Sutcliffe’s Efficiency (NSE) values for Case 1 were 0.989, 0.960, and 0.860, while for Case 2, they were 0.984, 0.961, and 0.813. The results highlight the effectiveness of the deep learning model, demonstrating its potential application in forecasting water flows in advance for effective water management planning during flooding situations.