Digital Soil Quality Index Mapping in Semi-Arid Vertisols: The Combination of Multi-Temporal Spectral Indices and Machine Learning

In semi-arid regions, soil quality evaluation is challenging due to spatial heterogeneity and sparse field data. This research developed a digital soil mapping (DSM) framework for Vertisols in the Blue Nile region, Sudan (38150 km2), using spectral indices derived from Landsat, topographic attributes, and drought indices. A Soil Quality Index (SQI) was derived from 14 physicochemical properties using PCA-based indicator selection and non-linear scoring functions. Three models were compared: (1) SCORPAN-based Random Forest, (2) Ordinary Kriging, and (3) a weighted ensemble (One Out All Out). SQI ranged from 0.27 to 0.95 (mean 0.56 ± 0.18), positively correlated with vegetation indices (NDVI: r = 0.76; EVI: r = 0.77, NPP: r = 0.83) and negatively associated with drought indices (DI: r = -0.68; SPEI: r = -0.71, and LST: r = -0.75). The most accurate was the One Out All Out (R2 = 0.91, Kappa = 0.96), followed by SCORPAN (R2 = 0.82), for SQI predicted. SQI was controlled by the drought index, SPEI, land surface temperature, and potential evapotranspiration. The drought indices, the indices drought index, SPEI, land surface temperature, and potential evapotranspiration were the most important factors controlling soil quality degradation. Climate projections (CMIP6, SSP5-8.5, 2024-2100) indicate that 34% of high-quality soils may be threatened by increased aridity. This framework provides a reliable baseline for soil quality monitoring in data-limited semi-arid regions.