Answers are grounded in this paper and cite their sources.
Three-panel summary: map of the BHH irrigation system with 15 monitoring stations; flowchart of the XGBoost-SHAP analytical pipeline (model calibration, SHAP importance and dependence); dendrogram and three-zone precision-management framework (protect, control, mitigate). Wastewater-impacted irrigation systems present complex, non-linear water quality dynamics complicating conventional monitoring. An integrated framework combining the WQI, Mann–Kendall analysis, hierarchical cluster analysis, and explainable machine learning (Random Forest, XGBoost, and LightGBM, with SHAP) was developed to characterize long-term degradation and identify dominant pollution drivers in the Bac Hung Hai. Results showed chronic, system-wide impairment, with a median WQI of 23 and a coefficient of variation exceeding 70%, indicating persistent 'Very Poor' to 'Poor' conditions. Mann–Kendall analysis detected improving trends at 33% of stations; however, a recovery peak in 2023 reversed by 2025. This recovery coincided with active management interventions (pumped augmentation, dredging, and discharge enforcement) rather than favorable hydrology, occurring in the driest year on record and indicating intervention-dependent improvement. Cluster analysis delineated three functional zones with distinct WQI distributions. XGBoost achieved the highest performance (R² = 0.934, RMSE = 6.60) and was the primary model. SHAP identified total coliforms as the dominant parameter, with a non-linear threshold near 5 × 10³ MPN/100 mL. Spatially structured anthropogenic loading, rather than hydroclimatic forcing, governs pollutant concentrations; DO exerted a restorative influence below approximately 2.3 mg/L. These findings support a transition from uniform to spatially differentiated, threshold-informed management.
No takes yet. Share an insight, caveat, or question.
Dung et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: