Information extraction of seasonal dissolved oxygen in urban water bodies based on machine learning using sentinel-2 imagery: An open access application in Baiyangdian Lake

Water bodies are crucial components of urban ecology. The development of rapid and timely water-quality assessment tools using easily measured variables is essential for the health management of urban water bodies. In this study, we focused on the dissolved oxygen (DO) of Baiyangdian Lake using 251 sets of empirically measured water quality data and corresponding Sentinel-2 satellite images. Nine machine learning algorithms were then used to develop a rapid detection algorithm for the spatial distribution of the DO concentration in Baiyangdian Lake. This study successfully applied these methods to invert the DO concentration in Baiyangdian Lake during spring, summer, and autumn. The results indicated that extra tree regression (ETR) provided the most accurate and stable results for inverting the DO concentration among the nine machine learning methods. In contrast, AdaBoost regression (ABR), Bayesian ridge regression (BRR), and support vector machines (SVM) exhibit relatively poor regression performance and lack sensitivity to DO concentrations. Moreover, the DO concentration in Baiyangdian Lake ranged from approximately 0 to 12 mg/L, with notable spatiotemporal variations. The highest overall DO concentration was observed in the spring, particularly in the southern region. The DO concentration significantly decreased during summer compared to that in spring, with higher values in the southwestern area and lower values in the northern region. The DO concentration reached its lowest value in autumn, with slightly higher values in the southern region. This study focused on the estimation and inversion of DO concentrations in the water bodies of Baiyangdian Lake. By introducing and comparing the performances of commonly used machine learning models, a rapid estimation of the DO concentration was achieved, thereby overcoming the limitations of traditional water quality monitoring methods in DO inversion. It not only intuitively explained the temporal and spatial variation patterns of DO concentration but also laid a foundation for further in-depth exploration of the interactions between DO and other water quality parameters.