The escalating threat to ocean water quality places strain on essential marine water resources, fishery habitats, and ecosystems. Salinity is one of the key indicators of water quality, offering valuable information about the exchange between coastal seas, rivers, and watersheds. The main object of this research is to predict water salinity and identify its influencing factors in estuarine and coastal waters, taking the Neuse River Estuary (NRE) in North Carolina as an example. This study was conducted at 11 mid-river sampling stations and involved comparing three machine learning models: Random Forest, Multiple Linear Regression, and Multi-Output Regressor with Gradient Boosting Regression (MOGBR), to predict salinity at various depths. The input predictors to our prediction models include aggregated river discharge, aggregated sea level, and aggregated wind based on eight directions. By prioritizing the most significant predictors, we streamlined the model-building process and developed a hindcast system covering the years 1994 to 2024. The methodology was divided into two phases: the first phase involved feature engineering and model selection, identifying MOGBR as the most effective model for predicting salinity across multiple depths and stations. In the second phase, the selected model was applied to predict seasonal salinity variations, enabling a comparative analysis of ground truth inputs, actual measurements, and predicted values. Results showed that the MOGBR model effectively captured spatial and seasonal salinity trends, with improved R² values across depths and stations. These findings demonstrate the utility of machine learning models in advancing salinity prediction and supporting coastal water management efforts.

Recommended Citation: Gachloo M., Song Y., Gokce E., Ricanek K., Liu Q., (2024). Enhancing Salinity Prediction in the Neuse River Estuary via Machine Learning Models. UNCW MS CSIS Proceedings. V. 18 , N. 20 .