Refining the parameterization of atmospheric particle settling velocity in seawater: Insights from interpretable...

The settling velocity of atmospheric particles in seawater is a key determinant of the ecological impact of atmospheric deposition on marine ecosystems, as it regulates particle residence time and bioavailability in the upper ocean. Settling velocity is primarily governed by particle diameter, shape, and density, however, accurate prediction remains challenging because of the heterogeneous morphology and composition of atmospheric particles. To address this challenge, we developed an interpretable Random Forest model trained on laboratory settling experiments. The model predicts particle settling velocity based on dimensionless diameter (D*), organic matter content (OM), and Corey shape factor (csf), achieving high accuracy (R² > 0.86) relative to theoretical formulations. SHapley Additive exPlanations (SHAP) analysis indicates that particle diameter is the dominant factor influencing dimensionless settling velocity (W*), contributing over 80 %. However, D* exerts weak, stable negative marginal impacts on W* when D* < 0.30 (salinity 0, 10) or D* < 0.25 (salinity 20, 30). Beyond these thresholds, its positive marginal contribution to W* increases markedly. Particle shape exerts a significant influence on W* only when D* exceeds the above thresholds across all four salinity conditions. Organic matter exerts a significant effect under high-salinity conditions. Building on the results, we refined the existing empirical formulation by re-fitting drag coefficient (C_D)-Reynolds number (R_e) relationships separately for low and high-salinity waters, reducing mean squared error (MSE) by 43-58 % while maintaining a high R² of 0.83-0.85. This refinement enables more accurate prediction of atmospheric particle residence time in the upper ocean.