Author(s): Sungjin Kim; Sewoong Chung

Linked Author(s): Se-Woong Chung

Keywords: CE-QUAL-W2; Data driven model; Hybrid model; Soyang Reservoir; Turbidity

Abstract: Mechanical models (MM), like CE-QUAL-W2 (W2), are pivotal for addressing water pollution issues and informing water management policies [1]. However, uncertainties, such as model parameters and input data, challenge their reliability [2]. This study introduces a novel hybrid model integrating W2 with various boundary condition prediction models to enhance turbidity predictions in Soyang Reservoir. W2 was calibrated using real-time turbidity data collected at the Soyang Dam from January to December 2020. Nine machine learning models, hydrological simulation program Fortran (HSPF), and load estimator (LOADEST) were employed for boundary condition prediction. Korea Water Resources Corporation (K-water) provided field data for training and testing, with flow rate and rainfall data used to predict inflow turbidity. Among the models, the neural network demonstrated superior performance compared to HSPF, linear regression, and decision tree models. The Long short-term memory (LSTM) model emerged as the most effective in predicting inflow turbidity. When applied as boundary condition data for W2, the LSTM model significantly improved turbidity predictions compared to existing linear regression equations. This study underscores the potential of neural network-based deep learning models to enhance MM performance in cases of insufficient boundary condition data, ultimately improving the accuracy of predictions and model reliability.

DOI: https://doi.org/10.3850/iahr-hic2483430201-19

Year: 2024