Multimodel Comparative Assessment and Optimal Geodetector-Driven Analysis of Aeolian Desertification in the Horqin Sandy Land

The Horqin Sandy Land (HSL), a critical ecological barrier in northern China and core area of the Three-North Shelterbelt Project, requires precise monitoring of aeolian desertification (AD) dynamics to maintain ecological stability in densely populated northern and northeastern China. This study innovatively integrated multi-source remote sensing parameters via Google Earth Engine to systematically compare three feature space models (Albedo-NDVI, Albedo-MSAVI, Albedo-KNDVI) and five machine learning algorithms for AD monitoring, while applying the Optimal Parameter Geographic Detector (OPGD) to quantify natural-anthropogenic driving mechanisms. High-resolution AD classification (5-year intervals, 2000–2020) revealed spatiotemporal dynamics, with machine learning models consistently outperforming feature space models. Random Forest (RF) achieved the highest accuracy (97.7%) and Kappa coefficient (0.967), effectively distinguishing non-desertified areas from extreme desertification. Temporal analysis showed a significant AD reversal: total desertified area declined to 30,166.35 km² by 2020, driven by ecological policies and climate amelioration. OPGD analysis indicated soil texture as the strongest single driver (q > 0.48), but factor interactions (e.g., SD ∩︀ LU) exerted greater explanatory power, highlighting compound effects of natural and anthropogenic factors. This study provides a robust framework for high-precision AD monitoring in semi-arid agro-pastoral zones, supporting global efforts toward zero land degradation.