Monitoring water quality parameters using multi-source data-driven machine learning models

Systematic monitoring of water quality parameters in aquatic environments was a critical task for environmental protection and water resource management. Remote sensing technology, as an effective monitoring tool, provided real-time water quality data. Currently, most research primarily relied on re...

Full description

Saved in:
Bibliographic Details
Main Authors: Yubo Zhao, Mo Chen, Jinyu He, Yanping Ma
Format: Article
Language:English
Published: Taylor & Francis Group 2025-12-01
Series:Engineering Applications of Computational Fluid Mechanics
Subjects:
Online Access:https://www.tandfonline.com/doi/10.1080/19942060.2025.2509658
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:Systematic monitoring of water quality parameters in aquatic environments was a critical task for environmental protection and water resource management. Remote sensing technology, as an effective monitoring tool, provided real-time water quality data. Currently, most research primarily relied on reflectance analysis of remote sensing data, often overlooking the impact of environmental factors on aquatic environments. This study integrated field data, multispectral imagery, meteorological data, and hydrological data to invert the water quality conditions of aquatic environments. A machine learning model (CNN-RF) was developed to estimate three water quality parameters (TP, DO, COD), and its performance was comprehensively evaluated using four error indices (R², MAE, MSE, MAPE). The results indicated that, compared to models using only spectral reflectance as features, the inclusion of environmental factors significantly enhanced the accuracy of the inversion of the three water quality parameters. For the watershed, the R² values for TP, DO, and COD increased by 0.34, 0.43, and 0.42, respectively. The MAE was less than 0.85, the RMSE was less than 1, and the MAEP was less than 15.65%. For the lake, the R² values for TP across four months were 0.75, 0.66, 0.67, and 0.73; for DO, they were 0.73, 0.79, 0.73, and 0.64; and for COD, they were 0.73, 0.62, 0.66, and 0.61, all exceeding 0.6. The MAE was less than 0.85, the RMSE was less than 0.95, and the MAEP was less than 12.6%. Both the watershed and lake models met the inversion requirements. Additionally, the SHAP method was employed to quantitatively analyze the contribution of environmental factors and spectral bands to the model inversion. Therefore, the findings confirmed the importance of environmental variables in water quality inversion and provided a theoretical basis for the optimization and application of future water quality monitoring systems.
ISSN:1994-2060
1997-003X