Water quality monitoring using remote sensing and PCA-enhanced machine learning models

Abstract

Successful water quality monitoring is critical to environmental sustainability and public health. Conventional methods of monitoring, based on laboratory analysis and manual sampling, are labour-intensive, costly, and geographically limited. In this research, the use of Remote Sensing (RS) and Machine Learning (ML) algorithms is explored to improve water quality forecasting, with a focus on the role of Principal Component Analysis (PCA) in optimising model performance. Through satellite-based multispectral images, a year (2022) of the Federal Ministry of Water Resources databases' key water quality parameters, including turbidity and chlorophyll-a concentration, were estimated using ML algorithms, i.e., Support Vector Machine (SVM), Random Forest (RF), and Decision Tree (DT). Following PCA-based dimensionality reduction, SVM maintained optimal classification performance, Random Forest retained strong predictive capability with minimal accuracy reduction, while Decision Tree performance declined, indicating greater sensitivity to reduced feature representation. These findings emphasise the importance of ML model selection according to the requirements of the targeted applications, particularly in real-time and resource-constrained systems. Future studies can investigate hybrid dimensionality reduction methods to further enhance model performance and extend real-world applications in water resource management.