As a vital component of the land-atmosphere interaction system, soil moisture plays an indispensable role. In arid and semi-arid regions, soil moisture serves as a key indicator of ecosystem vulnerability and also functions as an essential role in drought monitoring, climate research, agricultural water resource management, and land management. This study integrates radar and optical remote sensing data combined with intelligent algorithms based on machine learning to construct and evaluate soil moisture estimation frameworks across diverse data combinations. The Yutian Oasis was selected as the study area, and four hybrid models (ACO-RF, ACO-SVM, SSA-RF, and SSA-SVM) were developed by optimizing standalone random forest (RF) and support vector machine (SVM) models using ant colony optimization (ACO) and sparrow search algorithm (SSA). Three different combinations of input data sources were constructed based on GF-3 and Sentinel-2 data. A total of six models were employed to assess soil moisture throughout the study area under three different data source scenarios. The results indicated that compared to the other two single-source datasets, all models achieved the highest prediction accuracy when using the GF-3 + Sentinel-2 datasets. Specifically, the ACO-RF model exhibited superior results, with R 2 values in the test set improved by 9.59% and 5.26% compared to models using GF-3 or Sentinel-2 data alone, respectively. Across all models, the four hybrid models outperformed the standalone RF and SVM models. Among the hybrid models, ACO-RF demonstrated the best overall outcomes, achieving an R 2 of 0.80, RMSE of 3.07%, and RPD of 2.30. Therefore, integrating radar and optical data with intelligent algorithm-optimized machine learning strategies improves soil moisture estimation precision, offering significant support for sustainable oasis agriculture and land management in arid regions.