Three-Dimensional Mapping of Soil Textural Fractions on the Grounds of IFTO, Pedro Afonso Campus, Tocantins, Brazil

Soil texture, determined by the proportions of sand, silt and clay, is a fundamental indicator of soil physical quality and an important attribute for soil classification. This study mapped the three-dimensional distribution of soil texture in a pasture area at the Instituto Federal do Tocantins (IFTO), Pedro Afonso Campus, Tocantins, Brazil. Forty soil samples were collected at three depth intervals (0–20, 20–40 and 40–60 cm). Spatial modelling combined Ordinary Kriging and the Random Forest machine learning algorithm using environmental covariates derived from a digital elevation model and drone imagery. Clay content increased with depth, from 14% in the surface layer to 22% at 40–60 cm, whereas sand content declined from 71% to 58%, indicating greater textural differentiation in the subsurface. Model performance was generally modest, with R² values ranging from -0.04 to 0.21. Ordinary Kriging achieved the highest predictive performance at 40–60 cm (R² = 0.21, RMSE = 5.58% and MSE = 4.86%), slightly outperforming Random Forest (R² = 0.18, RMSE = 5.72% and MSE = 4.93%). These results indicate that the spatial dependence of the data had a stronger influence on soil texture prediction than the environmental covariates incorporated into the machine learning model. Although predictive accuracy was limited, the integration of geostatistical and geospatial techniques successfully characterised the spatial variability of soil texture. The resulting three-dimensional maps provide valuable information for sustainable soil management, land-use planning and future precision agriculture applications, while establishing a robust baseline for further investigations of soil spatial variability.