Mapeamento digital do carbono orgânico do solo em escala de bacia hidrográfica no Sul do Rio Grande do Sul

Abstract

Concerns about soil quality has been growing every year, as it has an influence on water quality, food security and energy production. Soil attributes at farm scale with high spatial resolution are still scarce. Among them, soil organic carbon (SOC) stands out, as it can serve as an indicator of fertility, soil quality, as well as a quality indicator for the environment. A key point in farm scale mapping is its spatial resolution. Smaller mapped areas need a higher spatial resolution, capable of representing their characteristics in a more detailed way, which are lost as the resolution decreases. Digital Soil Mapping (DSM) has been shown to be very effective for mapping soil properties in areas with these characteristics. This study carried out the DSM of the SOC, using Random Forest (RF), in 4 Study Areas (SA) of approximately 30 ha called SA-A, SA-B, SA-C and SA-D, with distinct characteristics of location and reliefs, in the Epaminondas Stream Hydrographic Basin (ESHB) in the south of Rio Grande do Sul - Brazil. The model accuracy and the uncertainty of the predicted maps were estimated at 3 different spatial resolutions (1 m, 5 m and 10 m). The accuracy of the models trained with covariates with a spatial resolution of 1 m was higher than the other resolutions, with an R2 of 0.67 for SA-A, R2 of 0.83 for SA-B, R2 of 0.88 for SA-D and R2 of 0.68 for SA-D. The maps that presented the smallest uncertainties were, in general, maps with a spatial resolution of 1 m, with emphasis on the SA-C and SA-D, ranging between 0.08 and 0.46 and 0.14 and 0.74 respectively. The SA-C also showed the lowest coefficients of variation (CV), 17.02% for the 1 m spatial resolution, 20.19% for the 5 m resolution and 22.83% for the 10 m resolution.