Real-Time Monitoring of Dissolved Oxygen Using a Novel Ground-Based Hyperspectral Proximal Sensing System

Abstract

High-frequency and high-precision dissolved oxygen (DO) monitoring is essential for lake health assessment, but it is limited by equipment and methods. This study developed a novel ground-based hyperspectral proximal sensing system (GHPSs) combined with machine learning methods for continuous monitoring of DO with an observation interval of 20 s. Five machine learning and deep learning models were calibrated and validated to estimate DO based on four combination scenarios of the GHPSs reflectance of 420–830 nm, chlorophyll-a (Chl-a), and water temperature (WTR). The results showed that a support vector machine model was preferred for DO estimation with satisfactory accuracy (R² = 0.84, mean absolute percentage error = 0.12, and root mean square error = 1.15 mg/L) based on 11,131 in situ measurements. Multitemporal changes of DO concentration in Lake Taihu were obtained from October 2021 to December 2023 by applying the model, which suggested that the surface of Lake Taihu was hypoxic in 2.1% out of 754 days. Finally, the potential significance of monitoring real-time DO dynamics was elaborated under global warming. This study highlights the effectiveness, accuracy, and high frequency of novel GHPSs in real-time DO monitoring, which is crucial for predicting and early warning of lake pollution.