Estimating global anthropogenic carbon dioxide emissions using satellite observations and machine learning methods

Abstract

Several countries are working to reduce their anthropogenic CO₂ emissions to meet the goals of the Paris Agreement. However, evaluation of the carbon reduction efforts is hindered by the larger uncertainties in the currently available datasets. Therefore, it is imperative to explore new efficient and reliable methods to estimate carbon emissions accurately. This study proposed a novel method to estimate global gridded anthropogenic CO₂ emissions using satellite datasets. The methodology included the development and integration of two machine learning models, i.e., RXCO₂ (Reconstruct XCO₂) and REMI (Reconstruct EMIssion), to achieve the objective. RXCO₂ utilized the CO₂ products from the Copernicus Atmosphere Monitoring Service (CAMS) model and the Orbiting Carbon Observatory 2 (OCO-2) satellite to produce a daily global long-term regular gridded column-averaged dry-air model fraction of CO₂ (XCO₂) dataset with a spatial resolution of 1°. The predicted XCO₂ dataset was thoroughly validated against the ground-based and satellite-derived XCO₂ observations, and good consistency was observed among the datasets. Further, the XCO₂ anomalies were derived using the predicted XCO₂ dataset and were utilized in the second model (REMI) along with tropospheric NO₂ column, nighttime light, and population density to predict annual gridded anthropogenic CO₂ emissions at a spatial resolution of 1° for 2021 and 2022. The model achieved high accuracy with a coefficient of determination (R²) of 0.96 and a root mean squared error (RMSE) of 10^0.3 tons. The predicted results were comprehensively compared with the anthropogenic CO₂ emissions provided by established inventories and good agreement was observed among the datasets.