Water Quality Source Apportionment of Qingyi River Based on the APCS‐MLR Model Optimized by Random Forest

The accurate apportionment of pollution sources is crucial for effective river water quality management. In this study, we analyzed the water quality status of Qingyi River and identified and quantified potential sources affecting water quality changes. Although the APCS‐MLR model is widely used for source apportionment, its high sensitivity to outliers and reliance on linear assumptions limit its accuracy in rivers with unstable water quality dynamics, often resulting in a high proportion of unidentified sources. To address this gap, this study introduces an approach that integrates the feature importance of random forest (RF) to optimize the APCS‐MLR model, enhancing its ability to capture nonlinear relationships, handle outliers, and improve source quantification. This approach was applied to the Qingyi River, which has good overall water quality but exhibits spatial variability and instability in some parameters. The results demonstrated the validity of our method: the coefficient of determination ( R 2 ) of the optimized model increased from 0.617 to 0.695, indicating a superior goodness‐of‐fit. Concurrently, the average contribution rate of unidentified sources decreased from 38.54% to 16.97%, confirming the model's enhanced ability to resolve pollution sources. The final apportionment ranked the sources as industrial sewage (48.73%), natural factors (17.69%), unidentified sources (16.97%), agricultural runoff (10.06%), and landfill leachate (6.55%). These findings provide a more reliable basis for crafting targeted protection strategies for the Qingyi River. Meanwhile, this study offers a novel and insightful methodology for advancing source apportionment techniques in the field of water environment research.