Precise spatial and temporal soil moisture is one of the limiting variables affecting agricultural decision-making in semi-arid regions. Despite the availability of various soil moisture monitoring products, their utility at the field scale is limited due to their relatively coarse spatial resolutions. Our study aims to predict surface soil moisture (SSM) at a much finer spatial resolution (10 m) using the synthetic aperture radar (SAR) Sentinel-1 C-band and harmonized Landsat Sentinel (HLS) data sets, with the help of in-situ soil moisture measurements (0-30 cm) from crop (maize, wheat, millets and fallow) fields located in the semi-arid locale of Akron, CO, USA. Four machine learning (ML) models: support vector machine (SVM), random forest (RF), gradient boosting machine (GBM), and K-nearest neighbor (KNN), were evaluated based on their performance metrics. Our findings indicated that the GBM demonstrated superior accuracy when using multi-source data with R2 of 0.72, RMSE of 0.025 cm3/ cm3, and RRMSE of 11.9 % for unseen verification data. NDVI, EVI, and incident angle were influential variables for SSM variability in dry crop lands. The Results also indicated that ML models captured SSM variability at different crops stages, with high accuracy in planting and late-season for all crops and relatively poor performance in mid-season for wheat. This suggests that the GBM model holds significant promise for field-scale SSM prediction in heterogeneous crop fields and different crop stages in semi-arid regions, offering a valuable tool for enhancing agricultural water management and hydrological modeling in the face of global water scarcity challenges.