Author(s): Go Eun Jang; Hye Ji Han; Ji Won Seo; Yong Gyun Park
Linked Author(s): Il Won Seo
Keywords: Deep Learning LSTM Flooding water height water quality
Abstract: This study aims to apply and evaluate the effectiveness of an LSTM (Long Short-Term Memory) -based deep learning model to predict water level and quality fluctuations in the upper Yeongsan River region in Korea, addressing the frequent flood damage and water quality incidents caused by recent climate change-induced heavy rainfall and rapid urbanization. Using hourly water level and quality data collected from 2017 to 2023, the study focuses on predicting water levels based on flood warning thresholds and water quality deterioration due to non-point source pollution inflow. Water level data was collected from the Geungnakgyo station of the Yeongsan River Flood Control Center, while water quality data was gathered from an automatic monitoring network at the Seochanggyo station, with any missing data points corrected using linear interpolation. In this study, the accuracy of predictions was enhanced by optimizing the LSTM model’s hyperparameters, and flood warning levels were set as key predictors to detect anomalies. The results demonstrate that the LSTM model effectively predicts water level and quality changes in the upper Yeongsan River, suggesting its potential contribution to future flood prevention and water quality management efforts.
Year: 2025