Multi-Step Discharge Forecasting in The Upper Yom Basin

Abstract

The upper Yom River basin often experiences flooding every year due to its mountainous terrain and steep slopes, coupled with the lack of water storage facilities in the upstream area. Discharge forecasting is one of the measures for preventing and mitigating the damage caused by floods that affect human lives and property. This study developed a 24-hour runoff forecasting model for the upper Yom River basin using Recurrent Neural Networks (RNNs), Long Short-Term Memory networks (LSTMs), and Gated Recurrent Units (GRUs), combined with the application of watershed response time parameters. Data from 8 rainfall stations and 4 runoff stations were used as inputs to train the models, with 60% of the data allocated for training, 20% for model validation and 20% for model testing. The study found that the GRU model can most accurately predict water flow rates 24 hours in advance. The GRU model achieved r, RMSE and NSE values of 0.9649, 19.7187, and 0.9283, respectively, outperforming the LSTM and RNN models. The study demonstrates that the model can effectively learn and adapt to changing data patterns, making it suitable for water management decision-making and flood warning systems in flood-prone areas with high risk.