Plastic film mulching technology (PFMT) is vital for enhancing agricultural production in arid and semiarid regions. However, its causes environmental issues, particularly plastic film residues that threaten soil health. Accurately monitoring the spatiotemporal distribution of mulched farmland is essential for sustainable agriculture and ecological protection. This study focused on the core area of the Beijing-Tianjin Sandstorm Source Region (BTSSR), where PFMT is widely applied. Multi-source remote sensing data from Sentinel-1, Sentinel-2, and Landsat 8 on the Google Earth Engine platform were used to extract spectral, texture, index, backscattering, and temperature features of plastic film-mulched farmland (PFMF) and non-mulched farmland (NMF). The differences in time-series characteristics were analyzed, and classification was conducted across various time windows to determine the optimal period. Four machine learning algorithms-Random Forest (RF), Support Vector Machine (SVM), Classification and Regression Tree (CART), and Gradient Boosting Tree (GBT)-were used for classification. The combination of spectral, index, and radar backscattering features achieved the highest accuracy, with RF performing best (95.37 % overall accuracy, Kappa = 0.91), followed by GBT (94.92, Kappa = 0.90). SVM and CART had lower accuracy (87.53 % and 91.78 %, respectively). The proposed method identified PFMF covering 33.51 % of the cultivated area in the BTSSR. Strengthening plastic film recycling through improved policies and technologies is crucial for sustainable development.