Space-temporal mapping of soil organic carbon for Brazil

• Grant number: 25/00139-6
• Support Opportunities: Scholarships in Brazil - Post-Doctoral
• Start date: June 01, 2025
• End date: May 31, 2027
• Field of knowledge: Agronomical Sciences - Agronomy - Soil Science
• Principal Investigator: Carlos Eduardo Pellegrino Cerri
• Grantee: Cássio Marques Moquedace dos Santos
• Host Institution: CENTRO ESTUDOS CARBONO AGRICULTURA TROPICAL/USP
• Associated research grant: 21/10573-4 - Center for Carbon Research in Tropical Agriculture (CCARBON), AP.CEPID

Abstract

The present project aims to map the spatial (with 30m resolution) and temporal (for 10-year periods between 1984-2023) variation of soil organic carbon (SOC), for 0-20, 20-40, 40-60, 60-80 and 80-100 cm layers, for the Brazilian territory and evaluate the relationship with land use and cover and agricultural area expansion. For this, legacy soil data obtained from the Brazilian Soil Spectral Library (BSSL) and the Brazilian Free Repository for Open Soil Data (FEBR) will be used. To determine the temporal variations of the SOC, the environmental covariates will be divided into static and dynamic. As static covariates, 13 terrain attributes will be used from the digital elevation model of the Embedded Radar Topographic Mission (SRTM) with a 30m pixel and calculated using the Terrain Analysis package on the Google Earth Engine (GEE) platform called TAGEE. A synthetic soil image (SYSI) of bare soil will also be obtained by mining temporal data from the Landasat series (1984 to the present) using the GEOS3 method. As dynamic covariates, the following vegetation indices, calculated from Landsat images will be obtained: the normalized difference vegetation index (NDVI), the soil adjusted vegetation index (SAVI) and the enhancement vegetation index (EVI). Dynamic covariates are those that change over the time series, that is, mean values will be obtained for these indices in the periods and 10 years of interest. These covariates will also be calculated with the mean value of the entire Landsat series (1984-present).The Random Forest algorithm will be used to calibrate a unique model, containing all sampling points (after depth standardization) as the dependent variable and the static covariates, together with the average vegetation indices from 1984 to the present, as independent variables. This model will be used to predict the SOC of each period of interest, and for this, the average vegetation indices will be replaced by dynamic covariates (vegetation indices of each period) in the prediction. A validation set will be selected from the initial database for each 10-year period to assess the accuracy of the generated maps. Land use and land cover and land use change maps will be obtained for the period in order to assess the relationship between these and SOC. The digital maps obtained will be organized within a Geographic Information System on the Web (SIGWEB) for analysis and public consultation.