Dados do Trabalho

Título

ACCESSING THE WEATHERING DEGREE OF BRAZILIAN SOILS THROUGH EARTH OBSERVATION DATA MODELED BY MACHINE LEARNING

Resumo

Soil weathering (SW) is directly associated with the capacity of soils to produce ecosystemic services (ES). Thus, mapping the SW degree can assist in determining the potential of a given soil to make a particular ES. This study aims to map the SW using remote sensing and machine learning (ML) for the entire agriculture area of Brazil and correlate them with legacy pedological maps. We determined the degree of SW by the Ki index, a ratio between Si and Al contents in the clay fraction. We created SiO2 and Al2O3 maps to calculate Ki for the 0-20 cm layer, performing the ratio of both soil attributes. Random Forest ML algorithm modeled the oxide content through environmental covariates (EC), producing an oxide map. The soil reflectance, obtained from the GEOS3 algorithm on the Landsat time series and terrain attributes from the SRTM digital elevation model, have produced such EC. Moreover, we got the sample dataset containing oxide content from the FeBr open data repository. The determination coefficients reached 0.44 and 0.36, while the RMSE were 49.85 and 47.98 for the respective models. The most important EC in the oxides prediction was the SWIR2 band (2110 to 2290 nm). We obtained a Ki map that covers 40% of the Brazilian territory. On the one hand, the Ki map revealed the most weathered soils in the Central Brazil region, mainly in the Cerrado biome. We also observed intense weathering processes in the basaltic spill region and the soils in the Barreiras region in Bahia, where Oxisols predominate. On the other hand, we detected the youngest soils in the Pantanal biome region, in the northeastern region of Brazil, and in the Pampa biome, where soils such as Planosols, Vertisols, Neosols, and Chernosols predominate. Therefore, we demonstrated the potential of remote sensing techniques to generate a SW map in agricultures areas with high detail level (30m), which agreed with the pedological distribution of Brazil.

Palavras-chave

Remote sensing, Soil reflectance, Ecosystem services

Instituição financiadora

The State of São Paulo Research Foundation (FAPESP) 2021/05129-8

Agradecimentos

The authors thanks GeoCis Research Group for sampling and analyses.