Multi source Earth observation and machine learning framework for mapping land degradation vulnerability in salinity- and hydrothermal-stressed regions of the southern main Ethiopian Rift.

Soil erosion and land degradation are the main environmental challenges disturbing ecosystem services, agricultural productivity, and community livelihoods in Ethiopia. The Southern Main Ethiopian Rift is one of the regions experiencing high levels of soil erosion and land degradation, driven by active geological processes and anthropogenic activities. However, the contribution of these geological processes to salinization and their subsequent impact on soil erosion and land degradation in the region remains poorly understood and insufficiently studied. This study was conducted to assess and map the spatial patterns of soil erosion and land degradation vulnerability in the salinized and hydrothermally stressed region of the Southern Main Ethiopian Rift using integrated multi-source earth observation data and machine learning frameworks. Landsat, Sentinel-2, SRTM DEM, CHIRPS rainfall data, soil data, and field observations were integrated with Google Earth Engine and GIS platforms to analyze existing biological, chemical, and physical indicators of land degradation. Land degradation indicators like soil erosion, land use/land cover (LULC), soil-adjusted vegetation index (SAVI), Normalized Multi-Band Drought Index (NMDI), land surface temperature (LST), salinity, and soil erosion were computed using respective approaches and integrated to produce a land degradation vulnerability map of the study area. The results show that large portions of the study area are moderately to highly vulnerable to land degradation. The major driver of land degradation in the study area is salinization, which increases surface runoff and sediment removal through surface crusting. The annual soil loss values for the area range from 25 to 100 t ha-1 yr-1, with about 10.69% of the area under moderate to very severe erosion risk. Accordingly, the land degradation vulnerability map shows that 1.68%, 50.15%, 38.14%, 6.31%, and 3.72% of the study area fall into the very low, low, moderate, high, and very high vulnerability categories, respectively. The land degradation vulnerability (LDV) model demonstrated strong validation performance, with an overall accuracy of 92.34%, a Kappa coefficient of 0.87, an F1-score of 0.948, and an ROC-AUC of 0.901. These validation results demonstrate the accuracy and reliability of the LDV model for mapping land degradation vulnerability. The findings indicate that soil salinity, surface soil loss, and elevated land surface temperature, together with anthropogenic pressures, are the primary drivers of high land degradation risk in the study area.