The Mongolian Plateau is one of the primary sources of dust storms globally, and these storms significantly impact the climate, ecology, and human health in East Asia and beyond. Accurately identifying the source areas of dust storms is crucial for improving disaster prevention, mitigation, and environmental governance. This article proposes an innovative geospatial heterogeneous ensemble learning model (sp-GSH_EL) that incorporates multiple spatial proximity measures. By combining Euclidean distance with great-circle distance, this model effectively integrates geographically weighted regression, geographically optimal similarity (GOS), and random forest (RF) models, taking into account spatial heterogeneity, global similarity, and nonlinear relationships. Using dust storm occurrence data from the Mongolian Plateau between 2000 and 2021, the study introduces a novel method for identifying dust storm sources by combining aerosol optical depth (AOD) and dust storm occurrence data. By setting an AOD threshold and combining the hourly change rate of AOD, the model accurately captures the dynamic changes of dust source points. These source and nonsource points are then used as training samples for dynamic identification of the susceptibility of dust storm source areas in the Mongolian Plateau. The results show that the overall accuracy of the sp-GSH_EL model was 0.941, with a precision of 0.929, recall rate of 0.954, F1-score of 0.941, MCC of 0.883, and AUC of 0.983, all outperforming traditional models. The study highlights that temperature, snow depth, and wind speed are key factors influencing dust storm source areas in the Mongolian Plateau. In addition, it was found that the main source areas of dust storms are located in the Gobi Desert and arid steppe regions in the central and southern parts of the Plateau, with a significant expansion of source areas between 2005 and 2010. The total area of high-risk and very high-risk zones increased from 3.4% to 9.9% .