Answers are grounded in this paper and cite their sources.
We analysed the county-level comprehensive water quality index (CWQI) for January–September of each year from 2019 to 2023 across 122 county-level units in Hunan Province. Spatial autocorrelation, dual-channel state-sequence clustering using partitioning around medoids and XGBoost–SHAP were used to characterise spatial patterns, trajectories and nonlinear associations. Mean CWQI fell from 3.4971 in 2019 to 3.2434 in 2022 and rose to 3.3435 in 2023. Spatial clustering weakened and then strengthened. Four trajectory types were identified: Favourable Consolidated, Medium-level Improving, Fluctuating Rebound and Persistent Pressure. The 2023 hold-out test for previously observed counties yielded an R2 of 0.7061. County-grouped validation yielded a mean R2 of 0.0900, indicating limited transferability to counties excluded from training. Slope had the highest relative SHAP importance (32.5%), and slope and secondary industry share formed the strongest SHAP interaction pair in each of the three years examined. The results can inform monitoring priorities within the study area and observation window.
No takes yet. Share an insight, caveat, or question.
Xiang et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: