Urban Fine-Grained Water Quality Monitoring Based on Stacked Machine Learning Approach

Abstract

Urban rivers play a crucial role in human environments and urban development. The use of drone remote sensing for fine-scale water quality inversion is vital for water environment monitoring and protection. However, given the complex and variable nature of water quality environments, the accuracy of most current water quality inversion studies is limited by the constraints of single machine learning models. To address this issue, we have constructed three stacked ensemble models. Among them, the most effective model is Stacked-RF(ST-RF). It uses Random Forest (RF) as the meta-model and combines linear regression models (including Least Absolute Shrinkage and Selection Operator (LASSO) and Multiple Linear Regression (MLR)) and the ensemble of three advanced machine learning techniques (Gradient Boosting, CatBoost, and Adaboost) as base models. By adopting the optimal weight scheme, individual base learners are combined, avoiding the prediction bias of individual base learners caused by differences in specifications and prediction accuracy. The performance of the ensemble model and five basic machine learning methods were evaluated. A new water body index is also proposed for water quality inversion. The results show that the ST-RF model with Random Forest as the meta-learner reached determination coefficients ( $R^{2}$ ) of 0.823, 0.691, and 0.647 for chlorophyll-a (Chl-a), total nitrogen (TN), and total phosphorus (TP), respectively. Compared with the commonly used single machine learning models, the method proposed in this paper has better performance. This research has important practical significance for urban water resource management and ecological environment protection.

Keywords

• Urban rivers
• water quality monitoring
• machine learning
• stacked model