Soil salinization poses critical threats to agricultural sustainability in arid regions, while traditional monitoring approaches lack efficiency and single remote sensing platforms encounter inherent spatiotemporal resolution trade-offs. This study developed a comprehensive daily-scale soil salinity monitoring framework by systematically evaluating four spatiotemporal fusion algorithms: Spatial and Temporal Adaptive Reflectance Fusion Model (STARFM), Unmixing-Based Data Fusion (UBDF), Fit-FC, and Unmixing-Based Spatiotemporal Image Fusion Based on the Self-Trained Random Forest Regression and Residual Compensation (RERC), integrating 5-cm resolution Unmanned Aerial Vehicle (UAV) multispectral imagery with daily Plantscope satellite data across high-standard farmland and saline-alkaline farmland throughout cotton phenological stages. Four machine learning algorithms (Random Forest, eXtreme Gradient Boosting, Support Vector Regression, Artificial Neural Network) were comparatively assessed for salinity inversion. RERC demonstrated superior fusion performance with Root Mean Square Error (RMSE) of 0.055 (high-standard farmland) and 0.024 (saline-alkaline farmland), Structural Similarity Index (SSIM) reaching 0.821 and 0.925, respectively. The optimal Random Forest-RERC framework achieved coefficient of determination (R2) = 0.94 during boll-opening stage, approaching actual UAV accuracy (R2 = 0.95). Generated daily salinity maps successfully captured fine-scale spatiotemporal dynamics, providing operational solutions for precision salinity management in arid agricultural systems.