Deficit irrigation (DI) is a crucial strategy for optimizing water use in arid and semi-arid agriculture, yet its success depends on accurately determining the crop yield response factor (K-y). This study introduces a novel, machine learning-assisted framework for large-scale estimation of sugarcane K-y using satellite-derived water stress indicators. fused Landsat 7/8/9 and MODIS data within the Google Earth Engine platform to generate high-resolution daily Crop Water Stress Index (CWSI) maps for the 2023 season in southern Iran. The Random Forest (RF) algorithm was applied to correct biases in land surface temperature (LST), achieving high accuracy (RMSE < 1.0 degrees C, nRMSE < 3%, rMBE approximate to 0%) before CWSI calculation. Ground measurements from 12 field points, including canopy temperature and yield, were used for calibration and validation. The satellite-based CWSI showed strong agreement with field data (RMSE = 0.05, nRMSE = 11%), with values ranging from 0.18 to 0.71 at dekadal scale. Using this CWSI, K-y was computed at dekadal, monthly, and seasonal scales, revealing substantial spatiotemporal variability (0.2-1.63) and an average seasonal K-y of 1.05. This value is lower than the FAO-66 default of 1.2, indicating that the standard coefficient may prompt over-irrigation without yield benefits. The analysis further identified early July as the period of peak water stress sensitivity, with K-y values exceeding 1.82. This ML-enhanced, satellite-based approach provides a robust tool for deriving spatially explicit K-y values, offering a significant advancement for precision irrigation planning and water resource management.