Multi-source data driven spatial changes of FFCO2 emissions balance and mitigation potential of different land use in China

Estimates of fossil-fuel carbon dioxide (FFCO₂) emissions in China are contingent on large uncertainty, and currently have enormous discrepancies among different inventories/datasets. The uncertainty is most attributed to underlying causes: only a few actual measurements and consumption data, and the statistical methods to confine the spatial resolution of FFCO₂. In this study, an attempt is made to assess the heterogeneities and uncertainty associated with spatial distributions of emissions in six gridded FFCO₂ inventories/datasets, which are compared at a 0.25 × 0.25 degree resolution.

We extract signals of urban CO₂ emissions with a Deep Learning (DL) & Deep Reinforcement Learning (DRL) modeling framework from the existing new generation of satellite (OCO-2/GOSAT-2/TROPOMI) observations of atmospheric column CO₂ (XCO₂). We then use the results as a proxy to further estimate of the FFCO₂ uncertainty. Subsequently, the estimated FFCO₂ uncertainty is included in an up-to-date multivariate spatial statistic to analyze China’s spatiotemporal FFCO₂ emissions balance, with a specific consideration made for the mitigation potential of different land-use types.

We find an interconnected system between the spatial FFCO₂ emissions distribution and two diverse factors being the most important: urbanization and either croplands (rainfed, irrigated, and post-flooding) or native vegetation. We have determined that wettability in croplands or the increase in native vegetation have an association with the decrease of FFCO₂ emissions. Ongoing work addresses the potential impacts of this FFCO₂ uncertainty on flux inversions.