Work

DDWQI: A novel water quality index based on data-driven approaches

Year: 2025

Type: article

Abstract: Faced with the ongoing deterioration of global water quality, scientific assessment of water quality is crucial for water resource classification management and pollution source tracing. To improve the adaptability and accuracy of water quality evaluation methods, this study builds on the traditional water quality index, proposes a system for constructing water quality index based on a data-driven mechanism, achieves scientific optimization in terms of indicator selection, weight allocation, and aggregation calculation, and constructs a novel water quality index. In the indicator selection stage, machine learning, SHapley Additive exPlanations (SHAP), and recursive feature elimination are combined to quantify and select water quality indicators. In terms of weight allocation, a SHAP-ROC method based on water quality indicator contribution was proposed to enhance the objectivity and stability of weight calculation. In aggregation calculations, the Normalized Weighted Harmonic Mean (NWHM) function is applied to reduce indicator eclipsing effect and improve accuracy. To validate the practical application of this method, an empirical analysis was conducted using a reservoir in a city in eastern China as a case study. The results indicate that this method can effectively identify key water quality indicators. The coefficient of variation of the constructed weighting system is 0.85, and the weighting fit R 2 is 0.86, significantly enhancing stability. In addition, the eclipsing effect of NWHM is 24.30%, better than traditional aggregation functions. The overall evaluation results demonstrate that the proposed method exhibits high reliability and applicability, offering an efficient and practical tool for water quality assessment in complex environments. • Apply Shapley additive explanations to extract water quality indicators. • Propose a novel SHAP-ROC approach to calculate weights of water quality index. • Utilize the harmonic mean function for aggregation to decrease the eclipsing.

Source: Ecological Indicators

DOI: 10.1016/j.ecolind.2025.113850