Abstract: Water turbidity directly affects the visual perception of urban water bodies and serves as an important indicator of sensory pollution. Traditional measurement methods are costly, limited in scope, and lack representativeness. A turbidity inversion approach that integrates UAV-based multispectral remote sensing with ensemble learning for urban rivers is proposed. By selecting the spectral bands highly correlated with turbidity, three inversion mo- dels—linear regression, random forest, and extreme gradient boosting—are constructed. Comparative analysis indicated that the random forest model performed best (R²=0.744). After applying spectral correction, a Random Forest-based turbidity inversion model (RF-Turbidity Model, RF-TM) is developed and further validated for its generalization capability. Experiments conducted in Suzhou showed that: (1) The RF-TM achieved high accuracy (R²=0.790, RMSE=4.087, rRMSE=0.322, ARE=30.10%, CE=0.312); (2) The RF-TM model exhibited excellent performance in the generalization verification (R²=0.651), outperforming the linear regression model and the XGBoost model; (3) The model inversion results align with measured data, accurately depicting Suzhou urban rivers' turbidity spatial distribution. In summary, the RF-TM model with high accuracy and strong generalization enables efficient urban river turbidity inversion, supporting large-scale holistic monitoring of urban water sensory pollution.