Retrieval of Dissolved Oxygen Concentrations in Fishponds in the Guangdong–Hong Kong–Macao Greater Bay Area Using Satellite Imagery and Machine Learning

Dissolved oxygen (DO) is a fundamental water quality parameter that directly determines aquaculture productivity. China contributes 57% of the global aquaculture production, with the Guangdong–Hong Kong–Macao Greater Bay Area (GBA) serving as a key contributor. However, this region faces significant environmental challenges due to increasing intensive stocking densities and outdated management practices, while also grappling with the systematic monitoring limitations of large-scale operations. To address these challenges, in this study, a random forest-based model was developed for DO concentration retrieval (R2 = 0.82) using Landsat 8/9 OLI imagery. The Lindeman, Merenda, and Gold (LMG) algorithm was applied to field data collected from four cities—Foshan, Hong Kong, Huizhou, and Zhongshan—to identify key environmental drivers to the changes in DO concentration in these cities. This study also employed satellite imagery from multiple periods to analyze the spatiotemporal distribution and trends of DO concentrations over the past decade, aiming to enhance understanding of DO variability. The results indicate that the average DO concentration in fishponds across the GBA was 7.44 mg/L with a statistically insignificant upward trend. Spatially, the DO levels remained slightly lower than those in other waters. The primary environmental factor influencing DO variations was the pH levels, while the relationship between natural factors such as the temperature and DO concentration was significantly hidden by aquaculture management practices. The further analysis of fishpond water quality parameters across land uses revealed that fishponds with lower DO concentrations (7.293 mg/L) are often located in areas with intensive human intervention, particularly in highly urbanized regions. The approach proposed in this study provides an operational method for large-scale DO monitoring in aquaculture systems, enabling the qualification of anthropogenic influences on water quality dynamics. It also offers scalable solutions for the development of adaptive management strategies, thereby supporting the sustainable management of aquaculture environments.