Full-Coverage Mapping of Daily High-Resolution Xco2 Across China from 2015 to 2020 by Fusing Oco-2 and Carbontracker Data Using Deep Learning Techniques

Abstract

Carbon neutrality has gained significant attention in the international community, and column-average dry-air mole fraction carbon dioxide (XCO2) with high spatio-temporal resolution can help to analyze the process of carbon neutrality and adjust measures and policies accordingly. Satellite-derived XCO2 exhibits substantial spatio-temporal gaps due to the effects of orbit and cloud cover.Furthermore, the spatial resolution of XCO2 from CarbonTracker is insufficient to meet the monitoring requirements at the present fine scale.We developed a deep learning-based spatio-temporal model called DeepCarbon that extracts spatio-temporal features from multiple data sources related to atmospheric transport, carbon emissions, and carbon sinks, enabling the evaluation of XCO2 at a fine scale.Additionally, we fused XCO2 from the Orbiting Carbon Observatory-2 (OCO-2) and CarbonTracker CT2022 at a spatial resolution of 0.1° to obtain training labels with wider coverage and more samples, serving as the fitting labels for DeepCarbon.We generated daily 0.1° XCO2 maps with complete coverage for China from 2015 to 2020 and analyzed the temporal trends in XCO2 growth throughout the study period.The validation results using the fused data showed R2 and RMSE values of 0.97 and 0.74ppm, respectively, while the average R2 and RMSE values for validation at four ground observation sites were 0.86 and 2.67ppm, respectively.The experimental results confirm that DeepCarbon effectively addresses gaps in satellite observations and streamlines the computation process of CarbonTracker, enabling the provision of high-resolution XCO2 data in an end-to-end manner.Full-coverage and high-resolution XCO2 helps us understand carbon dynamics and trace the origins of carbon emissions and sinks.

Keywords: Full-coverage mapping, Daily high-resolution XCO2, OCO-2, CarbonTracker, Deep learning