Answers are grounded in this paper and cite their sources.
Machine learning analysis reveals critical wind, rainfall, and grazing thresholds driving mountain sandy patch dynamics, highlighting key drivers of dryland aeolian degradation.
Mountain sandy patches are typical indicators of aeolian degradation in arid and semi-arid zones; however, few studies have systematically analyzed their static spatial morphological features and statistical correlations with environmental variables. Taking the Sarbulak River Basin in the Ili River Valley of Xinjiang as the study area, this study extracts multiple morphological metrics of mountain sandy patches from high-resolution UAV orthophotos and adopts the XGBoost-SHAP framework combined with correlation analysis to quantitatively analyze patch morphological traits and their statistical links with environmental predictors. The main results are as follows: (1) Elongated geometry dominates mountain sandy patches with diverse auxiliary shapes, and the average major axis of all patches reaches 16 m. Every pair of morphological indicators shows significant positive correlations at p < 0.01 level. (2) The model’s relative predictive importance varies markedly across predictors. Wind speed ranks first with a normalized SHAP contribution of 34.7%, followed by precipitation (18.7%), NDVI (13.0%), and grazing intensity (9.0%). The four predictors jointly account for over 75% of total predictive signals and constitute a wind–water–vegetation–grazing statistical association system. All predictors show obvious nonlinear responses to mountain sandy patch occurrence with distinct statistical thresholds. (3) Strong combined statistical correlations exist between wind speed, precipitation, NDVI, temperature, elevation, and grazing intensity, and multi-variable combinations correspond to a higher probability of large-scale sandy patches. This paper summarizes key threshold intervals derived from SHAP dependence curves: patches tend to expand when wind speed ranges from 2.10 to 2.15 m/s; precipitation below 219.7 mm presents negative correlations with patch distribution; NDVI within 0.17–0.29 corresponds to positive marginal associations with sandy patch occurrence; grazing intensity exceeding 3.60 SU/ha matches frequent patch enlargement; and areas above 645.9 m elevation display higher patch prevalence.
No takes yet. Share an insight, caveat, or question.
Song et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: