Background Error Statistics for Assimilation of Atmospheric CO2

Recent improvements in the CO2 observational density have spurred the development and application of data assimilation systems for extracting information about global CO2 distributions from available observations. A novel application that has been pursued at the European Centre for Medium-Range Weather Forecasts (ECMWF), as part of the Monitoring Atmospheric Composition and Climate (MACC) project, is to use a state-of-the-art 4DVAR system to assimilate CO2 observations, along with meteorological variables to obtain a consistent estimate of atmospheric CO2 concentrations. Global CO2 fields generated in this way enhance the observational database, because the data assimilation procedure uses physical and dynamical laws, along with the available observations, to constrain the analysis. As in any data assimilation framework, the background error covariance matrix plays the critical role of filtering the observed information and propagating it to nearby grid points and levels of the assimilating model. For atmospheric CO2 assimilation, however, the errors in the background are not only impacted by the uncertainties in the CO2 transport but also by the spatial and temporal variability of the carbon exchange at the Earth surface. The background errors cannot be prescribed via traditional forecast-based methods as these fail to account for the uncertainties in the carbon emissions and uptake, resulting in an overall underestimation of the errors. We present a unique approach for characterizing the background error statistics whereby the differences between two CO2 model concentrations are used as a proxy for the statistics of the background errors. The resulting error statistics - 1) vary regionally and seasonally to better capture the changing degree of variability in the background CO2 field, 2) are independent of the observation density, and 3) have a discernible impact on the analysis estimates by allowing observations to adjust predictions over a larger area. In this presentation, we will show the results of assimilating observations from the Greenhouse gases Observing SATellite (GOSAT) into the ECMWF-4DVAR system using both the new error statistics and those based on a traditional forecast-based technique ('standard' NMC method). We will also present the validation results against independent CO2 observations for a couple of representative months (January and June 2010). Results confirm that the performance of the data assimilation system significantly improves in the summer (winter) over the Northern (Southern) Hemisphere, when significant variability and uncertainty in the fluxes are present, and are correctly accounted for in the background error statistics. The proposed approach is also relevant for other trace gas assimilation applications, especially ones in which the background errors are influenced by both atmospheric transport and the emission patterns.