Analyzing the key drivers of pollutant transport in semi-enclosed bays: Insights based on numerical model and interpretable machine learning.

Due to the confined topography and limited water exchange, the pollution transport in semi-enclosed bays is a complex process driven by the interaction of tidal action, upstream river flow, and wind forcing. Traditional numerical models and machine learning models struggle to effectively analyze the key drivers of pollutant transport in such environments. This study employed a numerical model (Environmental Fluid Dynamics Code, EFDC) to reveal the transport characteristics of pollutant in semi-enclosed bays, and adopted interpretable machine learning (XGBoost + SHAP model) to analyze the potential key environmental drivers of pollutant horizontal physical transport and their interactions. Using the China's Sanmen Bay as a case study, results demonstrate that the short-term tidal fluctuations predominantly control pollutant transport in the inner bay regions. Flood tides promote pollutant accumulation in the bay. Long term spring tide can promote the outward transport of pollutant. Upstream river flow dominates pollutant transport in the bay mouth regions. High flow markedly increases the outward transport efficiency, with the highest contribution value reaching 56.1 %. The pollution discharged in open bay mouth will transport towards the open sea with the upstream river flow. Wind forcing amplifies tidal action on pollutant transport, particularly during low water period when the water body is more susceptible to wind forcing. At this time, the pollutants discharged will flow back and accumulate inside the bay. These findings support science-based pollution control in semi-enclosed bays.