This spatial dataset supports the results presented in the scientific article “Bridging soil mapping units and digital soil mapping through probability-driven machine learning: an uncertainty-aware framework for soil class prediction in Antarctica”, currently under review in Catena. It includes spatial predictions of soil classes, final soil mapping unit–structured soil maps , and intermediate outputs generated within the modeling framework The files are organized as follows: probability_maps_per_soil_class.zip: The first set of spatial data generated consists of 19 raster files (.TIF), representing continuous probability maps for each soil class across the study area. These maps correspond to median probabilities derived from 100 model runs, with values ranging from 0 to 1. probable_soil_classes.zip: This archive contains five raster files (.TIF) representing the ranked soil classes at each pixel, ranging from the most probable to the fifth most probable soil class. Each raster contains discrete values from 1 to 19, where these numbers represent different soil classes, as listed below: Anthorthents / Anhytrubels Aquiturbels Aquorthels Cryopsamments Dystrogelepts / Humigelepts Fibrists / Saprists Gelaquents / Psammaquents Gelaquepts / Petraquepts Gelifluvents Gelorthents Haplogelepts Haplohemists / Cryofibrists Haplorthels Haploturbels Psammorthels Psammoturbels Rocky Outcrops Umbrothels / Umbriturbels Vitrigelands probability_probable_soil_classes.zip: Five raster files (.TIF) representing the probability values associated with the ranked soil classes from the most probable to the fifth most probable soil maps. final_maps.zip: Vector files in SHP format (polygons) containing the final soil maps structured by mapping units, generated through the integration of the intermediate maps described above (see methodology in the forthcoming article). The files include a complete attribute table with soil class names, corresponding codes (1–19), and polygon areas. Mapping units may contain from one up to five soil classes (components). The file soil_mapping_raw_with_probs also includes the mean probabilities and standard deviation for each component. The file soil_mapping_units_aggregated represents the dissolved mapping units, while files ending in _1000, _2500, and _5000 correspond to generalized mapping units for these scales, based on the USDA minimum mappable area concept. number_classes_per_mapping_unit.zip: At last, this archive contains a single raster file (.TIF) with discrete numerical values ranging from 1 to 5, representing the number of components (soil classes) within each mapping unit.