Abstract: The escalating salinization and alkalization of arable soils represents a significant threat to the sustainable development of agriculture and environment. The assessment of salinization and alkalization can be facilitated by measuring crucial indicators including soil salinity content (SSC) and pH. The utilization of remote sensing technology could facilitate the effective and large-scale monitoring of soil salinity and alkalinity conditions. In this study, we selected the saline and alkaline farmland soil in the Hetao Plain as the research object, integrated measured soil salinity content (SSC), pH, and Sentinel-2 images (comprising six bands and 24 salinity indices), and incorporated environmental variables, soil physicochemical attributes, and synthetic aperture radar (SAR) data as mode-ling variables. Following the implementation of feature screening through the utilization of gradient boosting machine (GBM), we established the inverse models of SSC and pH based on six machine learning algorithms, including extreme gradient boosting (XGBoost), light gradient boosting machine (LightGBM), adaptive boosting (AdaBoost), category boosting (CatBoost), random forest (RF), and extreme random tree (ERT). We further visuali-zed the variable contributions by Shapley additive explanation (SHAP) interpretation, and realized the inverse mapping for the spatial distribution of salinity and alkalinity information. The results showed that the overall soil salinization and alkalization were at mild to moderate levles, with significant spatial heterogeneity between salinization and alkalization. The GBM algorithm could effectively reduce the model’s complexity by filtering the feature variables with a cumulative contribution of up to 90%. The contribution of different types of variables to the salinization and alkalization information varied significantly. The XGBoost and ERT models demonstrated optimal perfor-mance in the SSC and pH inversions, respectively, with model validation R² values of 0.925 and 0.818, respectively. The SHAP analysis revealed that the salinity index was the most significant variable that contributed 34.9% to the SSC and 34.2% to the pH, respectively. Soil physicochemical properties and topographic factors exhibited a range of 15.7% to 23.0% contributions. There were minimal contributions from climatic factors and radar data, and the least contribution from single band. The study could offer a scientific reference for the monitoring of soil salinization and alkalization, the selection of variables, and the decision-making process concerning agricultural enhancement in analogous regions.

Key words: soil salinization and alkalization, Sentinel-2 imagery, environmental variable, Shapley additive explanation, digital soil mapping