Online analysis of Amazon's soils through reflectance spectroscopy and cloud computing can support policies and the sustainable development

Autor(es)

Novais, Jean Jesus Macedo ; Melo, Borges Marfrann Dias ; Neves Junior, Afranio Ferreira ; Lima, Raimundo Humberto Cavalcante ; de Souza, Renato Epifanio ; Melo, Valdinar Ferreira ; do Amaral, Eufran Ferreira ; Tziolas, Nikolaos ; Dematte, JoseA. M.

• Tipo de documento: Artigo Científico
• Fonte: Journal of Environmental Management; v. 375, p. 11-pg., 2025-01-22.

Resumo

Analyzing soil in large and remote areas such as the Amazon River Basin (ARB) is unviable when it is entirely performed by wet labs using traditional methods due to the scarcity of labs and the significant workforce requirements, increasing costs, time, and waste. Remote sensing, combined with cloud computing, enhances soil analysis by modeling soil from spectral data and overcoming the limitations of traditional methods. We verified the potential of soil spectroscopy in conjunction with cloud-based computing to predict soil organic carbon (SOC) and particle size (sand, silt, and clay) content from the Amazon region. To this end, we request physicochemical attribute values determined by wet laboratory analyses of 211 soil samples from the ARB. These samples were submitted to spectroscopy Vis-NIR-SWIR in the laboratory. Two approaches modeled the soil attributes: M-I) cloud-computing-based using the Brazilian Soil Spectral Service (BraSpecS) platform, and M-II) computing-based in an offline environment using R programming language. Both methods used the Cubist machine learning algorithm for modeling. The coefficient of determination (R2), mean absolute error (MAE) and root mean squared error (RMSE) served as criteria for performance assessment. The soil attributes prediction was highly consistent, considering the measured and predicted by both approaches M-I and M-II. The M-II outperformed the M-I in predicting both particle size and SOC. For clay content, the offline model achieved an R2 of 0.85, with an MAE of 86.16 g kg-1 and RMSE of 111.73 g kg-1, while the online model had an R2 of 0.70, MAE of 111.73 g kg-1, and RMSE of 144.19 g kg-1. For SOC, the offline model also showed better performance, with an R2 of 0.81, MAE of 3.42 g kg-1, and RMSE of 4.57 g kg-1, compared to an R2 of 0.72, MAE of 3.66 g kg-1, and RMSE of 5.53 g kg-1 for the M-I. Both modeling methods demonstrated the power of reflectance spectroscopy and cloud computing to survey soils in remote and large areas such as ARB. The synergetic use of these techniques can support policies and sustainable development. (AU)

• Processo FAPESP: 21/05129-8 - Qualidade dos solos do Brasil via geotecnologias: mapeamento, interpretação e aplicações agrícolas/ambientais: um legado para a sociedade
• Beneficiário: José Alexandre Melo Demattê
• Modalidade de apoio: Auxílio à Pesquisa - Temático
• Processo FAPESP: 22/14935-0 - Levantamento de dados e avaliação detalhada da qualidade do solo em áreas piloto: bases para a validação do mapa de qualidade dos solos do Brasil
• Beneficiário: Jean de Jesus Macedo Novais
• Modalidade de apoio: Bolsas no Brasil - Pós-Doutorado