The dataset was used to estimate the relevant range of spatial scales with multi-scale contextual spatial modelling. The modelled soil properties were cation exchange capacity, pH, and water content at field capacity. The soil quality indicator data was modelled and predicted with partial least squares regression models based on NIR and MIR spectroscopy (Pangaea DOI (doi:10.1594/PANGAEA.938522): “Soil spectroscopy data from 130 soil profiles in Lora del Rio, Andalusia, Spain”). The soil samples were taken in an area of 1000 km² around Lora del Rio, Andalusia, Spain, in the Sierra Morena mountain range (Palaeozoic granite, gneiss, and slate), at the Guadalquivir river flood plain (Pleistocene marl, calcarenite, coarse sand, and Holocene sands and loams), and southern tertiary terraces (coarse gravel and cobble with sands and loams). Present soil types according to USDA Soil Taxonomy are Alfisols, Entisols, Inceptisols, and Vertisols. The basis for the multi-scale terrain analysis was a digital terrain model by the Centro Nacional de Information Geográfica (CNIG) of the Spanish government. The digital terrain model was published under the CC-BY 4.0 license via the Centro de Descargas del CNIG (IGN; doi:10.7419/162.09.2020) with the title Digital Terrain Model - DTM05 (EPSG: 25830) and last accessed on March, 31st 2020. The study area is covered by the MTN50 map sheets 0941, 0942, 0963, 0964, 0985, and 0986. The multi-scale contextual spatial modelling and the derivation of the scaled terrain covariates was based on the Gaussian pyramid (doi:10.1016/j.geoderma.2017.09.015 and doi:10.1038/s41598-018-33516-6) and the estimation of the relevant range of scales was based on exhaustive additive and subtractive machine learning sequences (doi:10.1038/s41598-019-51395-3). The models were trained with the multi-scale terrain covariates at each soil profile location extracted from the digital terrain model derivatives. For each soil depth of the soil dataset (0-10, 10-20, 20-30, 40-60, and 70-100 cm) two model sequences (additive and subtractive) were trained.

Supplement to: Rentschler, Tobias; Bartelheim, Martin; Behrens, Thorsten; Díaz-Zorita Bonilla, Marta; Teuber, Sandra; Scholten, Thomas; Schmidt, Karsten (2022): Contextual spatial modelling in the horizontal and vertical domains. Scientific Reports, 12(1)