Retrieval, Spatiotemporal Characteristics, and Regional Differentiation Analysis of Potassium Permanganate Index in the Dongjiang River Basin Based on Sentinel-2 Satellite and Machine Learning Models

The Dongjiang River Basin (DRB) is a crucial water source for the Guangdong-Hong Kong-Macao Greater Bay Area (GBA), and its water quality is vital for ensuring regional ecological security and sustainable development. The Potassium Permanganate Index (COD<sub xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">Mn</sub>) is one of the key indicators for assessing water quality in the DRB. However, traditional COD<sub xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">Mn</sub> monitoring methods have limitations in inland waters, making it difficult to comprehensively reveal the spatiotemporal dynamics of water quality in the basin. To address this issue effectively, this study utilized multi-band data from Sentinel-2 imagery and water body type information to construct machine learningbased remote sensing retrieval models for CODMn in the DRB. Through comparative analysis, the optimal retrieval model was identified, demonstrating significant accuracy (R² = 0.71, RMSE = 0.8 mg/L, MAPE = 23.28%). Based on the optimal model, highresolution spatiotemporal distribution products of COD<sub xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">Mn</sub> in the DRB from 2017 to 2024 were generated. The results indicate the COD<sub xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">Mn</sub> in some segments of the midstream and the delta are relatively higher, which is closely related to agricultural activities impacts in the midstream, as well as industrial wastewater discharge pollution in the delta. Temporally, the COD<sub xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">Mn</sub> in most areas of the DRB reach their lowest values in winter, a phenomenon that differs from the water quality characteristics of other basins. This study highlights the significant application value of machine learning methods and water body type attributes in improving COD<sub xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">Mn</sub> remote sensing retrieval, reveals the unique water quality characteristics of the DRB, and emphasizes the importance of region-specific management strategies within the basin. The findings provide a theoretical foundation and data support for the scientific management and protection of the DRB.