融合机器学习与地理探测器的绿洲盐渍化垂直特征解析OA

Soil salinization is a critical barrier constraining the sustainable development of oasis agriculture in ar-id regions.This study aims to reveal the vertical variation characteristics of soil salinity in the Aksu oasis and its principal controlling mechanisms.The research integrated machine learning algorithms with optimal parameter geospatial detectors and constructed soil total salt models at four depths(0-10 cm,10-20 cm,20-30 cm,and 30-50 cm)using Sentinel-2 imagery.Results showed that the XGBoost model achieved the highest prediction accu-racy(R2≥0.6,RMSE≤5.97 g·kg-1),with overall robust performance but slightly higher uncertainty in severely sa-linized areas.Spatially,salt concentration decreased with increasing depth and exhibited significant low values near rivers due to leaching effects during wet periods.Attribution analysis demonstrated that the driving mecha-nisms exhibited vertical stratification:surface layer(0-10 cm)salinity was dominated by bivariate synergistic en-hancement of human activities,soil,and climate as local factors;while the deeper layer(30-50 cm)was deter-mined by the coupled mechanism of"groundwater-climatic evaporation."This study elucidated the vertical differ-entiation mechanisms of salinization processes and provided scientific evidence for three-dimensional monitoring and prevention of oasis salinization.