Forecasting desertification in central Iran using an ensemble of machine learning models: a comprehensive modeling approach

Desertification involves processes that arise from both natural factors and human mismanagement. Therefore, it is essential to develop effective strategies for the quantitative assessment of desertification. In this study, the desertification status of the region under investigation was first examined using the MEDALUS model. Based on the results obtained from this model and a review of studies conducted by other researchers, seven remote sensing indices were selected for modeling. The methods SVM, GBM, GLM, and RF were employed to model desertification risk in Central Iran (Isfahan Province). Ultimately, to provide a comprehensive model, a weighted average of the four models was utilized in the SDM statistical package for modeling and predicting desertification. According to the results from the MEDALUS model, the most significant factors contributing to desertification in the study area were identified as management and policy criteria (score 1.61), vegetation (1.51), erosion (score 1.48), climate (1.34), and soil (score 1.26). Additionally, 41% of the area falls into the low and negligible desertification class, 3% into the moderate class, 18% into the severe class, and finally, 38% into the very severe desertification class. Statistical indicators revealed that in 2009, the SVM model performed better than other models with an AUC of 0.81, TSS of 0.86, and Kappa of 0.89; while in 2023, the RF model outperformed others with an AUC of 0.83, TSS of 0.7, and Kappa of 0.87. The results from the ensemble model indicate an increasing trend of desertification in eastern Isfahan Province. By examining the changes in desertification from 2009 to 2023, three key areas for desertification expansion were identified: Gavkhuni Wetland (southeast), Sagzi Plain (central), and Rig-e Boland Desert (north and northeast). Therefore, based on the results from the ensemble model as a comprehensive framework with minimal uncertainty, appropriate planning, optimal management, and corrective measures can be employed in affected areas to prevent further desertification processes.