Abstract: To address the spatiotemporal limitations of conventional water quality monitoring in Lake Chaohu, as well as the deficiency of existing remote sensing retrieval studies on the lake—which have focused mainly on single parameters such as chlorophyll-a and lack multi-feature-based retrieval of non-optically active water quality parameters—this study is based on Landsat 8 OLI imagery and synchronously monitored water quality data. Four non-optically active water quality parameters were selected as retrieval targets: permanganate index (COD_Mn), electrical conductivity (EC), dissolved oxygen (DO), and ammonia nitrogen (NH₃-N). Five categories of remote sensing features were extracted, namely band reflectance, 10 vegetation indices, 8 texture features, and the Tasseled Cap Transformation and Principal Component Analysis. For the statistical regression models, sensitive features were screened by Pearson correlation analysis (|r| ≥ 0.40) before model construction; the Random Forest model directly employed all five feature categories and was trained and validated using repeated 5-fold cross-validation. The results show that the Random Forest model achieved higher coefficients of determination (R²) than the statistical regression models, with cross-validated R²_cv values of 0.65~0.76 for the four parameters, demonstrating stronger nonlinear fitting capability. The different remote sensing features exhibited distinctly different indicative capacities for the four parameters: vegetation indices performed best for COD_Mn retrieval, texture features showed a stronger capability for indicating DO, reflectance was most sensitive to EC, and the Tasseled Cap Transformation yielded the best results for NH₃-N. The inversion results reveal that the spatial distribution of COD_Mn was relatively uniform in Lake Chaohu during winter, whereas EC showed a weak decreasing gradient from north to south; DO showed a pattern of higher values in the west and lower values in the east, and high NH₃-N values were concentrated in the northwestern part of the lake. This study demonstrates that combining multiple remote sensing features with the Random Forest model is feasible for retrieving non-optically active water quality parameters in Lake Chaohu during winter, and the results can provide spatially explicit references for winter water. quality inspection and pollution source tracing in the lake.