Predictions of saltwater intrusion in the Changjiang Estuary: Integrating Machine learning methods with FVCOM

Saltwater intrusion is a typical hydrological phenomenon in estuaries that has a significant impact on daily life and is difficult to predict. While FVCOM has been instrumental in simulating sea surface salinity, recognizing its potential errors in dynamic estuarial domains is crucial. This study pioneered the integration of machine learning techniques, including Bagged Regression Trees (BRT), Artificial Neural Networks (ANN), and Long Short-Term Memory (LSTM) with FVCOM to provide deeper insights into the underlying processes and enhance the simulation of saltwater intrusion dynamics in the Changjiang Estuary. The BRT-FVCOM model determined Changjiang River discharge as the primary natural factor influencing salinity estimates, with the significance of wind and water level varying across stations but remaining important factors. Notably, ANN-FVCOM models demonstrated a substantial reduction in root mean square error of 53.8%, 55.8%, and 50.0% at the Baozhen, Shidongkou, and Santiaogang stations, respectively. LSTM-FVCOM models showed slightly higher error reductions and smoother predictions but required more computational resources. This study introduces a novel framework that enhances numerical model accuracy and provides insights into complex environmental processes, with potential applications extending to various coastal and ocean systems where improving the accuracy of numerical simulations is important.