Prediction of organic carbon contents in Brazilian soils through proximal sensing

• Grant number: 24/18138-3
• Support Opportunities: Scholarships in Brazil - Program to Stimulate Scientific Vocations
• Start date: January 20, 2025
• End date: March 11, 2025
• Field of knowledge: Agronomical Sciences - Agronomy - Soil Science
• Principal Investigator: Sérgio Henrique Godinho Silva
• Grantee: Carlos Henrique de Andrade
• Host Institution: Escola de Ciências Agrárias. Universidade Federal de Lavras (UFLA). Ministério da Educação (Brasil). Lavras , SP, Brazil

Abstract

Soil is the world's main reservoir of carbon (C). In light of climate change, the dynamics of C forms in soil and the role of C in greenhouse gases, especially in the forms of CO2 and CH4, are drawing increasing attention on this chemical element. Measuring soil organic C (SOC) content generally requires various laboratory analyses, which involve chemical reagents, a range of equipment, and the handling of generated waste. Proximal sensors, such as portable X-ray fluorescence spectrometers (pXRF), have been successful in generating variables (the content of chemical elements in samples) that, combined with machine learning algorithms, can estimate C content and other soil attributes. However, studies using pXRF in this context and with a large number of contrasting soil samples from different Brazilian regions are still rare. Thus, this project aims to test different machine learning algorithms and a database of over 3,000 Brazilian soil samples for predicting SOC. At least three machine learning algorithms (Cubist, Random Forest, and Support Vector Machine) will be tested on 70% of the data, leaving the remaining 30% for model validation. This database is already available for use. It is expected that this project will generate models capable of predicting the organic C content of Brazilian soils with high accuracy based on pXRF data. (AU)