Deriving the distribution of carbon dioxide surface fluxes using concentratons at a tall and an Eulerian transport model

This study seeks to estimate the regional distribution of carbon dioxide sources and sinks on a regional scale. A region is thereby defined as an area with spatial dimensions of a few hundred kilometers. It thus forms a link between local (~ 1 km) flux measurements using an EC tower and estimates obtained using large-scale (global or continental) inversions. This is useful, as bottom-up estimates of the distribution of sources and sinks that are based on aggregation of surface flux measurements yields estimates of the European sink of carbon of about 240 Tg a-1, while estimates of top-down approaches that result from large-scale inversions, give strengths of the European carbon sink of about 140 Tg a-1, almost half as low (Janssens et al. 2003). However, both approaches are prone to considerable errors and assumptions have to be made. Investigation of these errors and evaluation of these assumptions, requires investigation of the carbon budget on regional scales.

To obtain the regional distribution of sources and sinks, Bayes' rule is used. This implies that it is acknowledged that different data sources are available that all contain valuable information. And that the optimal estimate of the regional distribution of sources and sinks is the weighted average of estimates from these data sources. Hereby, the weights are functions of the uncertainties in the different data sources that are being used. Here, three data sources are identified, namely 1) surface fluxes of carbon dioxide by inverting a regional transport model fed by measurements of the carbon concentrations taken at different heights of a tall tower in Cabauw (52N, 5E), the Netherlands, 2) estimates of the regional distribution of carbon dioxide fluxes obtained using the Land Surface Scheme of the coupled biosphere/atmosphere regional model RAMS-GEMTM, and 3) estimates of the anthropogenic emissions in the region around the Cabauw tower.

The setup as described above requires forward runs with a meso-scale atmospheric model. In this study the rams-4.3.0 model is used. This model is developed at the Colorado State University to study meso-scale processes. This model integrates the full non-hydrostatic equation of motion and continuity equation, and solves tendency equations for the potential temperature, total water mixing ration, while for this study carbon dioxide is added as an extra scalar. To calculate the surface fluxes of carbon dioxide, the GEMTM model (Eastman et al. 2001) is used. This scheme is an extension of the original Land Surface Scheme of RAMS accommodating the uptake and emission of carbon dioxide by the vegetation and thus the computing of biogenic surface fluxes. Anthropogenic emissions are taken from the EDGAR3.2 data base (Olivier et al. 2001) and are prescribed at the lower boundaries of the atmospheric part of RAMS-4.3.0. Boundary conditions are taken from the ECMWF ERA40 re-analysis, except for carbon dioxide for which a no-gradient boundary conditions is prescribed.

To implement the technique described above, aggregation of the surface fluxes is required. In this study, the fluxes are aggregated with respect to biome or land surface type. Therefore, rather than optimizing the fluxes themselves, scale factors are optimized. Surface fluxes within a biome then follow as the product of scale factors that varies according to biomes and a base function that varies temporally and spatially according to meteorological conditions. The total distribution of sources and sinks follows as the sum of the products of scale factors and base functions.

In this presentation, results will be shown for the application of this method for two days of the RECAB campaign, held in the central Netherlands. First, the setup described above is run forward yielding a set of a priori surface fluxes calculated by the RAMS-gemtm model whereby it is assumed that the a priori values of the scale factors are zero. Then, the modeled concentrations are compared to the observed concentrations and depending on the measurement error and estimates of the modeling error, estimates of the error in the a priori fluxes, the scale factors are optimized. Concentration measurements are taken from the Cabauw tower in the central Netherlands (52N, 5E).

It appears that assuming the proposed technique is powerful, in case the model can be assumed to be perfect. In this case, the concentration measurements put a strong constraint on the surface fluxes of the major biomes. Assuming that the uncertainty in the a priori values for the scale factors are 20 %, inclusion of measurements of the carbon dioxide concentration gives reductions of the uncertainty to 2 % for grassland and mixed/crop vegetation. Interestingly, the measurements taken at 20 m and 40 m reflect the local vegetation type, while for concentrations taken higher up constraint is strongest for the dominant vegetation type in the entire 900 km x 900 km domain. Results appear however to be sensitive to model error, most notably errors in the turbulent scheme of the RAMS-gemtm model. In the presentation these errors will be discussed and their impact on the invesion setup evaluated.