Dados do Trabalho

Título

SPATIAL-TEMPORAL MAPPING OF CARBON STOCK USING BARE SOIL REFLECTANCE DATA

Resumo

The soil organic carbon (SOC) is related to climate change mitigation and food security. There was an exponential increase in publications of content/stock SOC mapping by Digital Soil Mapping and remote sensing. However, few studies aim to retrieve the spatial-temporal dynamic of SOC. We aimed to test 3 strategies for SOC stock spatial-temporal mapping based on bare soil reflectance data and machine learning.  The study was conducted in a 184 ha sugarcane farm in Rafard, São Paulo, Brazil.  It was performed a 100x100 grid sampling in 1998 with exact coordinates at 0-20 cm. Thus, a new sampling was made in 2022 in the same locations. It resulted in 119 observation points in 1998 and 2023 years and 42 only for 1998. A synthetic soil image (SySI) was obtained from Landsat 4, 5, 8 and 9 using the GEOS3 method for 1997-1998 and 2021-2022 time periods. The area has detailed soil and geology maps, bare soil frequency (BSF) from GEOS3 and occurrence of erosion information. Three prediction models were built using the Cubist algorithm: global model= points and covariates of 1998 and 2022; 1998 model = points and covariates of 1998; 2023 model = points and covariates of 2023. The models were evaluated by cross-validation and by external validation using 42 points of 1998. The global model reach R2 = 0.37 and RMSE = 1.09 kg m2, while the 1998 model reach R2 = 0.44 and RMSE = 1.06 kg m2 and the 2023 model reach R2 = 0.22 and RMSE = 1.39 kg m2 for external validation. The use of a global model or a model of other period is a valid strategy for SOC stock spatial-temporal mapping when there is not sufficient data for all periods to be mapped, nevertheless, it causes accuracy decrease.  It was verified an increase of SOC stock in the most part of study area, that can be related with the BSF, geology, soil class and absence/occurrence of erosion. Retrieving the spatial-temporal variation of SOC stock is decisive to evaluate and planning environmental and agricultural practices.

Palavras-chave

pedometrics; soil health; digital soil mapping; soil quality.

Instituição financiadora

The São Paulo Research Foundation (FAPESP) n 2021/10063-6 and 2021/05129-8

Agradecimentos

The authors are grateful to the members of Geotechnologies in the Soil Science Group (GEOCIS) (https://esalqgeocis.wixsite.com/english)