Author(s): Sewoong Chung; Hyunhan Kwon; Honglae Cho; Sungjin Kim; Dongmin Kim

Linked Author(s): Sungjin Kim, Dongmin Kim, Se-Woong Chung, Hyun-Han Kwon

Keywords: Extreme rainfall; Turbidity prediction; Resevoir management; Ecological stress index

Abstract: In monsoon regions such as the Republic of Korea, numerous multipurpose dams have been constructed to secure water resources, generate hydropower, and regulate floods. However, climate change has intensified extreme rainfall events, causing prolonged turbidity releases that impose severe stress on downstream ecosystems. To address this challenge, we developed an AI-enhanced turbidity prediction system integrating conventional numerical modeling with machine learning techniques. The system links three components: the extreme rainfall prediction model (MNSRP), the deep learning–based ensemble watershed model (DEWMOST), and an LSTM-supported two-dimensional turbidity model (CE -QUAL-W2). The framework was calibrated and validated using observed data from the North Han River basin, including the Soyanggang Dam, the largest and most turbidity-prone dam in Korea. Results showed that the AI-enhanced system substantially outperformed conventional methods in predicting inflow turbidity and reservoir turbidity dynamics. Applications to the 2006 turbidity event and synthetic scenarios generated by MNSRP revealed ecological stress index values exceeding 12, a threshold associated with lethal consequences for aquatic organisms. The findings highlight the urgent need for mitigation strategies such as soil erosion control and selective withdrawal. The proposed system provides a decision -support tool for proactive turbidity forecasting, ecological risk assessment, and adaptive reservoir management under climate-driven extremes.

DOI: https://doi.org/10.64697/iahr.proc.ise2026.abs.107

Year: 2026