Prediction of Dam Reservoir Level and Downstream River Level as Influenced by Discharge Based on A Machine Learning Method

The current study developed a method for the prediction of inflow, storage levels, and discharge flow rates from dams for extreme flood disaster prevention management. In recent years, because of the damage from frequent large floods all over Japan, the prediction of water storage levels and discharge flow rates to be utilized for effective dam management has become critical. Three dams on Japan's northern island of Hokkaido were targeted in this study and are used for the River Basin Disaster Resilience and Sustainability by All concept. First, using Elastic Net, a sparse modeling method capable of identifying relationships between data even from small amounts of information, we predicted the inflow volume for dams that had experienced cases in which it was required to engage disaster prevention management during extreme flooding. Subsequently, the water storage level was estimated based on the predicted inflow and the discharge based on dam operation regulations. Furthermore, the predicted discharge flow rate was shown to be effective for predicting the water level of rivers downstream from the dam. In summary, it is considered that the proposed method can be utilized for preemptively assessing discharge and recognizing the effect of the discharge on the downstream area.