COMPUTATIONAL MODELING OF ATMOSPHERIC CO2 and CH4 CONCENTRATION IN CENTRAL BRAZIL

• Grant number: 24/12884-5
• Support Opportunities: Scholarships in Brazil - Scientific Initiation
• Start date: December 01, 2024
• End date: November 30, 2025
• Field of knowledge: Agronomical Sciences - Agronomy - Soil Science
• Principal Investigator: Alan Rodrigo Panosso
• Grantee: Pedro Henrique Marucio de Oliveira
• Host Institution: Faculdade de Ciências Agrárias e Veterinárias (FCAV). Universidade Estadual Paulista (UNESP). Campus de Jaboticabal. Jaboticabal , SP, Brazil

Abstract

Global climate change is a growing concern worldwide, mainly due to the increasing concentration of greenhouse gases (GHGs) in the atmosphere, especially carbon dioxide (CO2) and methane (CH4). In order to implement effective measures to reduce GHG emissions, it is essential to understand the dynamics of these gases in the atmosphere and to establish relationships with other variables related to the soil-plant-atmosphere system. In this context, several efforts have been made to improve GHG monitoring techniques at global and regional scales. This proposal aims to describe the spatio-temporal variability of atmospheric CO2 and CH4 concentrations in areas of central Brazil, and to identify sources and potential sinks of these gases from 2015 to 2023. Data on atmospheric concentrations of CO2 (XCO2) and CH4 (XCH4) will be obtained from the GOSAT and OCO-2 orbital sensors.In addition, climate variable data will be obtained from the National Aeronautics and Space Administration (NASA) platform.For all GHG emitting sectors, data will be obtained from the reports of the Climate TRACE platform, a non-profit coalition able to track and provide information on global greenhouse gas emissions. Data collection will be systematised to reduce differences in spatial resolution of remote sensing data, and then the global trend of XCO2 and XCH4 will be removed. Spatial variability and interpolation in unsampled locations will be performed using geostatistical techniques. Statistical machine learning tools and algorithms implemented in R will be used for data exploration. The techniques used will include artificial neural networks, random forest and extreme gradient boosting. In general, 70-80% of the observations will be used for model training and 30-20% for validation. Model accuracy will be determined using Pearson's 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). This approach is expected to contribute to improve the understanding of the dynamics of greenhouse gases in the atmosphere and their interactions with climatic and land use variables in Central Brazil. It will provide important insights for the formulation of public policies aimed at mitigating GHG emissions and adapting to climate change.