Valid for version v2023
The Jena CarboScope atmospheric CO2 inversion estimates the spatio-temporal CO2 exchange fluxes between surface and atmosphere, based on high-precision measurements of atmospheric CO2 mole fractions from various stations around the globe. It uses individual measurements from various sampling networks, without smoothing or gap filling. The surface CO2 fluxes are linked to the atmospheric mole fractions by the TM3 atmospheric transport model (currently run on 5° longitude × about 4° latitude resolution) driven by NCEP reanalysis (Kalney et al., ). The flux estimates are found from the condition that they lead to the best match between the modelled atmospheric CO2 mole fractions and the measurements (by minimizing a quadratic cost function). Fluxes are estimated at the grid-scale resolution (about 2.5° longitude × 2° latitude) to reduce aggregation errors.
As the estimation solely based on station data would be underdetermined, the CarboScope atmospheric CO2 inversion further uses Bayesian a-priori constraints. Fossil-fuel CO2 emissions are prescribed by the GridFED inventory (Jones et al.). In the base set-up, the ocean-atmosphere CO2 exchange is prescribed from an estimate based on SOCAT surface-ocean pCO2 data (CarboScope pCO2 interpolation). Thus, the base set-up uses the atmospheric CO2 signals to estimate the land-atmosphere CO2 flux (interpreted as Net Biome Exchange, NBE) only. Further, the CarboScope atmospheric CO2 inversion imposes Bayesian a-priori correlations, smoothing the estimated NBE field on scales smaller than about 1 week and about 1,600 km (in longitude direction) or about 800 km (latitude), respectively. The a-priori uncertainty of NBE is spatially weighted according to vegetation density from SYNMAP (Jung et al., ).
The cost function minimization is performed by a Conjugate Gradient algorithm with re-orthonormalization. A single minimization is done for the entire estimation period plus spin-up and spin-down periods.
In each yearly release, CarboScope produces and provides several variants intended for different purposes, in particular ``standard inversions'' over different time periods and the ``NBE-T inversion''.
The ``standard inversion'' is intended to provide interannual variations of NBE as seen by the atmospheric CO2 measurements. Therefore, it involves explicit degrees of freedom for NBE variations on all time scales longer than the above-mentioned correlation time.
In order to prevent spurious NBE variations due to changes in the data constraint when station records start or end, the ``standard inversion'' uses a limited set of measurement stations selected to completely cover the respective estimation period of the run. Since the rising number of available stations creates a trade-off between the achievable length of the estimation period and the number of stations covering it completely, the ``standard inversion'' is run for differently large station sets providing NBE variations over differently long time periods. That way, better-constrained but shorter inversions can be used to assess the results of longer but less-constrained inversions which regard to the temporal features being investigated.
The ``standard inversion'' estimates NBE anomalies around an a-priori mean seasonal cycle, taken from the ``NBE-T inversion'' described below. The advantage of the mean seasonal cycle from the ``NBE-T inversion'' is that it is constrained by many more measurement stations than the limited set of the ``standard inversion''. Using many stations for the mean seasonal cycle is possible because it is not affected by changes in the data constraint over time.
The ``NBE-T inversion'' (formerly imprecisely called ``NEE-T inversion'') ............