Water quality evaluation in Liaoning Province large reservoirs: a new method integrating random forest-TOPSIS and Monte Carlo simulation

Abstract

This study introduces a novel integrated model aimed at enhancing the accuracy and efficiency of water quality assessments in four major reservoirs of Liaoning Province, China. The model integrates the technique for order preference by similarity to ideal solution with Monte Carlo simulation and employs the random forest method for weight allocation. Utilizing monthly water quality data, the model generates normally distributed datasets that are processed through the TOPSIS model, incorporating RF-derived weights and a membership function, for a comprehensive evaluation. Validation of the model demonstrated a predictive accuracy rate exceeding 83.87%, outperforming other assessment methods such as the analytic hierarchy process, criteria importance through intercriteria correlation, the evidential weighting method, and the COV method. The MCS significantly reduced uncertainties linked to multiple indicators, thereby enhancing the reliability of the assessments. In 2023, the model provided monthly assessments that closely matched the actual water quality conditions, with the four reservoirs exhibiting water quality levels of Grade II, Grade II, Grade III, and Grade II, respectively. A global sensitivity analysis identified chemical oxygen demand (COD), biochemical oxygen demand (BOD₅), total phosphorus (TP), and potassium permanganate index (COD_Mn) as critical determinants of water quality. The study further confirmed the model’s robustness by outlining its optimal assessment accuracy within a 5% error margin under normal distribution.