Can machine learning bridge the gaps in satellite-based carbon dioxide monitoring? This study proposes a machine learning framework for fusing column-averaged dry-air mole fraction of CO2 (XCO2) retrievals from GOSAT and OCO-2 satellites, addressing limitations in spatiotemporal coverage and biases among different satellite retrievals. The best model achieved improved consistency of GOSAT retrievals by reducing 71.5% of the average monthly bias while using OCO-2 retrievals as a benchmark. Incorporating adjusted GOSAT retrievals into the OCO-2 data set increased observation coverage, enhancing yearly temporal coverage by 53.6%. The machine learning method maximizes satellite resources for a more robust carbon flux inversion. The study demonstrates the potential to enhance observation coverage and improve the accuracy of carbon cycle monitoring using satellite data.
Published in Geophysical Research Letters, this study is highly relevant to the journal's focus on groundbreaking research in Earth sciences. By improving the consistency and coverage of satellite-based CO2 retrievals, the research contributes to a more accurate understanding of the carbon cycle and its response to climate change, aligning with the journal's emphasis on advancing geophysical knowledge.