An Ensemble Machine Learning Model to Estimate Urban Water Quality Parameters Using Unmanned Aerial Vehicle Multispectral Imagery

Abstract

Urban reservoirs contribute significantly to human survival and ecological balance. Machine learning-based remote sensing techniques for monitoring water quality parameters (WQPs) have gained increasing prominence in recent years. However, these techniques still face challenges such as inadequate band selection, weak machine learning model performance, and the limited retrieval of non-optical active parameters (NOAPs). This study focuses on an urban reservoir, utilizing unmanned aerial vehicle (UAV) multispectral remote sensing and ensemble machine learning (EML) methods to monitor optically active parameters (OAPs, including Chla and SD) and non-optically active parameters (including CODMn, TN, and TP), exploring spatial and temporal variations of WQPs. A framework of Feature Combination and Genetic Algorithm (FC-GA) is developed for feature band selection, along with two frameworks of EML models for WQP estimation. Results indicate FC-GA’s superiority over popular methods such as the Pearson correlation coefficient and recursive feature elimination, achieving higher performance with no multicollinearity between bands. The EML model demonstrates superior estimation capabilities for WQPs like Chla, SD, CODMn, and TP, with an R2 of 0.72–0.86 and an MRE of 7.57–42.06%. Notably, the EML model exhibits greater accuracy in estimating OAPs (MRE ≤ 19.35%) compared to NOAPs (MRE ≤ 42.06%). Furthermore, spatial and temporal distributions of WQPs reveal nitrogen and phosphorus nutrient pollution in the upstream head and downstream tail of the reservoir due to human activities. TP, TN, and Chla are lower in the dry season than in the rainy season, while clarity and CODMn are higher in the dry season than in the rainy season. This study proposes a novel approach to water quality monitoring, aiding in the identification of potential pollution sources and ecological management.

Keywords

• UAV remote sensing
• optically and non-optically active parameters
• genetic algorithm
• ensemble machine learning