Colombian soil texture: building a spatial ensemble model

Abstract. Texture is a fundamental soil property for multiple applications in environmental and earth sciences. Knowing its spatial distribution allows better understanding of the response conditions to changes environment, such as land use. This paper describes technical development Colombia's first texture maps, obtained via ensemble national global digital mapping products. work compiles new database with 4203 profiles, which were harmonized at five standard depths (0–5, 5–15, 15–30, 30–60, 60–100 cm) standardized additive log ratio (ALR) transformation. A compilation 83 covariates was developed 1 km2 resolution. Ensemble machine learning (EML) algorithms (MACHISPLIN landmap) trained predict particle size fractions (PSFs) (clay, sand, silt), comparison SoilGrids (SG) products performed. Finally, function created identify smallest prediction errors between EML SG. Our results are effort build map silt fractions) based on Colombia. The showed that their accuracies very similar each depth, more accurate than largest improvement found layer (0–5 cm). predictions frequently selected PSF depth total area; however, SG when increasing some specific regions. final error study area sand presented higher absolute values clay fractions, specifically eastern Colombia potential tool provide information water-related applications, ecosystem services, agricultural crop modeling. However, future efforts need improve aspects treating abrupt unbalanced data. compiled (https://doi.org/10.6073/pasta/3f91778c2f6ad46c3cc70b61f02532db, Varón-Ramírez Araujo-Carrillo, 2022, https://doi.org/10.6073/pasta/d6c0bf5847aa40836b42dcc3e0ea874e, et al., 2022) insights solve aforementioned issues.