Estimation of Xco 2 in Rice and Sugarcane Crops with an Ensemble of Wofost and Random Forest

Abstract

The concentration of carbon dioxide (CO2) in the air is one of the main factors affecting climate change, which can impact climate zoning and, consequently, agricultural production, affecting food security and the quality of life of people, especially the most vulnerable. Therefore, it is important to monitor and estimate this quantity in agricultural areas. Thus, this study used productivity data (1984-2022) from rice (Oryza sativa) and sugarcane (Saccharum officinarum) areas, from the main producing regions of Brazil, under the hypothesis that the daily XCO2 could be estimated from the ensemble of the WOFOST and Random Forest (RF) models, using remote sensing data as a source of measured XCO2. The results confirmed the hypothesis, with high accuracy (R² > 0.75) for the models under sugarcane and rice cultivation, respectively. The daily total gross assimilation (DTGA) has a negative correlation with the calculated CO2 measured in the atmosphere (ρ=- 0.45, p-value<0.001). The DTGA profile did not change throughout the historical series. Key factors influencing XCO2 estimates were identified, including water stress and vegetation parameters for sugarcane, and radiation and leaf area for rice. This work contributed to the understanding that climate impacts can exert on agroecosystems, with a focus on XCO2.

Keywords: Machine learning, Remote sensing, climate change, Modeling, Python