Integrating the Oasis Cooling Effect into a Multidimensional STGP Feature Cube for Cropland Recognition in Xinjiang (2015–2024)

Monitoring cropland dynamics in arid regions is critical for balancing food security with water scarcity constraints. However, distinguishing fragmented agricultural oases from spectrally similar desert vegetation remains a persistent challenge due to spectral confusion and landscape heterogeneity. To address these challenges, this study developed the STGP-OCE feature cube on the Google Earth Engine platform (GEE) by integrating the Oasis Cooling Effect (OCE) into the commonly used STGP (Spectral, Textural, Geomorphic, and Phenological) feature space, coupled with the XGBoost ensemble model. Through ablation experiments and feature importance analysis, we quantified the feature construction mechanism for arid regions. Oasis Cooling Intensity emerged as the most influential variable (Gain score: 0.315), demonstrating that the thermal signature of continuous anthropogenic irrigation serves as a robust thermodynamic proxy to resolve the spectral ambiguity between crops and drought-tolerant desert vegetation. By hierarchically coupling this thermal indicator with textural features to suppress fragmentation noise, topographic constraints to filter non-arable terrain, and phenological trajectories, the STGP-OCE feature cube achieved an Overall Accuracy of 95.12% and a Precision of 94.95%, significantly outperforming models built on lower-dimensional cubes as well as existing global land cover products. We generated a 10 m annual cropland dataset for Xinjiang, China, revealing a substantial 32.9% expansion (19,360 km2) from 2015 to 2024, mainly occurring in vulnerable oasis–desert transition zones and coinciding with reported reclamation activities. These highlight the continuous agricultural encroachment into desert margins, while the proposed STGP-OCE cube provides a reliable methodology for high-precision cropland monitoring in arid regions.