STATISTICAL MACHINE LEARNING FOR ESTIMATING SOIL CO2 EMISSIONS IN AGRICULTURAL AREAS

• Grant number: 24/12296-6
• Support Opportunities: Scholarships in Brazil - Scientific Initiation
• Start date: February 01, 2025
• End date: January 31, 2026
• Field of knowledge: Agronomical Sciences - Agronomy - Soil Science
• Principal Investigator: Alan Rodrigo Panosso
• Grantee: Luis Felipe Trevelim
• Host Institution: Faculdade de Ciências Agrárias e Veterinárias (FCAV). Universidade Estadual Paulista (UNESP). Campus de Jaboticabal. Jaboticabal , SP, Brazil

Abstract

The concentration of greenhouse gases (GHG) in the atmosphere, such as carbon dioxide (CO2), has significantly increased due to anthropogenic sources. In Brazil, agricultural and forestry activities substantially contribute to CO2 emissions, mainly due to deforestation and the conversion of native forests. Previous studies have demonstrated that soil CO2 emission (FCO2) can be accurately modeled using a large number of environmental variables. However, the long-term conversion of native forests to agroecosystems is still poorly understood, especially in the Brazilian context. Thus, the central hypothesis is that land-use changes for agricultural purposes alter the chemical and physical attributes of the soil, inducing changes in CO2 emission. This project aims to investigate soil CO2 emissions (FCO2) in agricultural areas of the Cerrado biome, using statistical machine learning techniques to model FCO2 based on associated variables. The techniques used will include artificial neural networks (ANN), Adaptive Neuro-Fuzzy Inference System (ANFIS), support vector machines (SVM), random forest, and extreme gradient boosting. Generally, 70-80% of the observations will be used for model training, and 30-20% for validation. Model accuracy will be determined using Pearson correlation (r), coefficient of determination (R²), root mean square error (RMSE), mean error (ME), concordance index (d), confidence coefficient (c), and mean absolute percentage error (MAPE). It is expected that this approach will contribute to understanding the dynamics of CO2 flux in Brazilian soils and provide a basis for constructing more accurate spatiotemporal patterns, essential for developing effective climate change mitigation strategies.