Seminar: Martin Jung

Global atmospheric CO₂ inversion results have been providing key estimates of net carbon flux variations between atmosphere, land, and ocean. These results seem robust at large spatial scales but are uncertain locally due to spatially compensating errors arising from atmospheric transport uncertainty, observation density and other factors. Inversions using satellite-based CO₂ benefit from much larger observation density relative to in situ CO₂ data and promise improved capabilities in localizing carbon fluxes. However, it remains unclear which regions and at which spatial granularity atmospheric inversion results are robust and useful for policy relevant budgeting, process interpretation, or as data constraint for global ecosystem model evaluation or calibration.

To address this question we developed a pattern recognition methodology to delineate regions with optimal robustness based on the ensemble of atmospheric inversions of the OCO2-MIP project. The employed optimization procedure balances systematic differences of carbon flux patterns between regions and uncertainties within regions. The recursive partitioning of regions delivers a hierarchical tree-like structure of nested regions that are beneficial for interpretation and analysis. Due to the factorial design of OCO-2-MIP we can address the following key questions: 1) How do these regions look compared to the traditionally used TRANSCOM regions? 2) For which regions does the inclusion of satellite based CO₂ data lead to large changes in net carbon flux estimates? 3) For which do we find the largest differences between prior and posterior flux estimates? 4) For which regions do we find systematic differences to the data-driven FLUXCOM-X-BASE product?