Deciphering Phosphorus Fractions in Sediments Using Structural Equation Modelling and Explainable Machine Learning in the Saigon River

ABSTRACT

Phosphorus (P) enrichment in urban rivers and canals is a major driver of eutrophication, yet its spatial and seasonal behavior remains poorly resolved in tropical monsoon systems. This study analyzed P fractions and associated metals in 48 surface sediment samples from 13 river and 11 canal sites during rainy and dry seasons. The results showed that iron‐bound inorganic (Fe‐Pi) and organic P (Fe‐Po) dominated sediment pools, with higher concentration in canals and during the dry season ( p  < 0.05), while aluminium‐bound (Al‐Pi) and soluble P concentrations increased in the rainy season. Structural equation modelling (SEM) revealed contrasting phosphorus transformation pathways, with Fe–Pi acting as the dominant regulator in river sediments and Al–Pi together with soluble P exerting greater influence in canal sediments. Composite indices indicated substantial P enrichment across all site–season combinations, with DPS ranging from 295.90% to 323.49% and PSR from 2.96 to 3.23. The ERI remained relatively low (0.19–0.21), whereas PSC was negative throughout (−25.75 to −28.46 mol kg ⁻¹ ), indicating limited P storage capacity and potential P mobility in the sediments. Using eXtreme Gradient Boosting (XGBoost) coupled with SHapley Additive exPlanations (SHAP) interpretation, key environmental controls included cation exchange capacity (CEC), temperature, pH, salinity and grain size. Cross‐validated models achieved robust predictive performance, with the lowest prediction error for ERI (RMSE = 0.047). By integrating P fractionation, process‐based indices, SEM and interpretable machine learning, this study provides a transferable framework for assessing P loading and eutrophication risk in tropical urban river systems.