Multi-source uncertainty quantification of water quality index models based on two-dimensional Monte Carlo simulation

Abstract

This study addresses the subjectivity in weight determination of traditional Water Quality Index (WQI) models and their limited consideration of input uncertainty. A framework integrating machine learning methods and two-dimensional Monte Carlo simulation (2D MCS) is proposed for water quality uncertainty analysis, using the robustness monitoring data in Hanjiang River Basin, China, as a case study. Algorithms like Random Forest (RF), XGBoost, Recursive Feature Elimination (RFE), and Support Vector Machines (SVM) are used to assess parameter importance, select key indices, and assign weights for an improved WQI model. Four uncertainty scenarios are evaluated, and a two-layer Monte Carlo simulation examines the model's stability. Results show that the optimized WQI model is more robust and adaptable, especially under combined uncertainty, highlighting the risk of synergistic effects.