Abstract
Soil nutrients play a fundamental role in terrestrial ecosystems and are essential for understanding the effects of global changes. Carbon, nitrogen, and phosphorus are required in specific quantities by plants and are related to soil fertility. In the Caatinga, one of the largest and most diverse tropical dry forests in the world, there are still some studies that seek to understand the determinants of the spatial variability of organic carbon (OC), N, and P in the soil and, even fewer, those that explored the use of ML modeling. In this work, we predict the spatial variability of the properties of these elements at depths between 0 and 20 cm in this biome and evaluate the predictive capacity of environmental and geographic variables. We used the Random Forest model in Google Earth Engine to forecast maps with a spatial resolution of 30 m. The highest result was obtained for predicting P (LCCC of 0.32 and R2 of 0.25), followed by OC (LCCC of 0.25 and R2 of 0.17), N (LCCC of 0.21 and R2 of 0.12) and C/N ratio (LCCC of 0.14 and R2 of 0.10). The final maps showed good spatial consistency, with OC, N, C/N distributed according to climatic covariates, topographic data, and geographic regions (longitude and latitude). The P content varies mainly depending on the parent material in the soil. We highlight the relevance of ecotones, which recorded the highest average levels of C and N and C/N, demonstrating the importance of these areas for the maintenance and dynamics of these ecosystems.