Digital Soil Mapping Dataset — Northern Kazakhstan Steppe Zone (2022–2023)

Dataset for "Digital Soil Mapping of the Steppe Zone in Northern Kazakhstan:Predicting Agrochemical Properties of Soils Using Multimodal Satellite Data andMachine and Deep Learning Techniques" (Agriculture 2026, 16(11), 1239,DOI: 10.3390/agriculture16111239). Contents--------Derived features and image patches for predicting six agrochemical soilproperties (pH, SOC, NO₃, P₂O₅, K₂O, S) across the Northern Kazakhstan steppe(20 farms, 81 fields, 1085 soil samples, 2022–2023) from multimodal satellitedata (Sentinel-2, Landsat-8, Sentinel-1 SAR, SRTM terrain, ERA5-Land climate). - data/features/master_dataset.csv — canonical modelling table (1085 × 530).- data/features/selected/ — per-target selected feature lists used by the models.- data/features/README.md — data dictionary (start here).- data/patches*/ — image patches for the CNN/ConvNeXt models.- data/<source>/ — intermediate per-source feature CSVs. Raw laboratory soil measurements are commercially sensitive and are aggregatedinto the derived features provided here. Code----Analysis code: https://github.com/vel5id/science_SOC_predicting(archived: DOI 10.5281/zenodo.19496423). The repository's data_loader readsdata/features/master_dataset.csv; with this archive it reproduces the publishedresults out of the box. See the "Version" block above for what changed in this release.