|Title||Ensemble of Data Assimilations applied to the CAMS' greenhouse gases analysis|
|Publication Type||Technical memorandum|
|Year of Publication||2016|
|Secondary Title||ECMWF Technical Memorandum|
|Authors||Massart, S, Bonavita, M|
The Ensemble of Data Assimilations (EDA) is currently the method used at ECMWF to estimate the background error statistics for the meteorological analysis. It particularly allows to have flowdependent background errors which is beneficial for ECMWF’s operational deterministic meteorological analysis.
The Copernicus Atmosphere Monitoring Service (CAMS) provides an atmospheric tracers analysis based on Composition-IFS (C-IFS). Distinctively from ECMWF’s operational deterministic meteorological analysis, CAMS analysis uses climatological estimates of background error. We want to investigate whether using EDA-based flow-dependent background errors for atmospheric tracers could be beneficial for the CAMS analysis as it is for ECMWF’s operational analysis.
The first step before using EDA-based background errors in the CAMS analysis is to design the EDA for atmospheric tracers. This document describes the Composition-EDA (C-EDA), a modified version of ECMWF’s EDA that accounts for the sources of uncertainties associated with atmospheric tracers. The focus is on two species of the anthropogenic long-lived greenhouse gas family: carbon dioxide (CO2) and methane (CH4).
A default C-EDA experiment with only perturbations of the meteorological parameters showed that the spread of the ensemble is larger at the surface than in the troposphere for both CO2 and CH4. Moreover the ensemble standard deviation has a large variability in time at the surface probably associated with the time variation of the boundary layer. It also has strong spatial variations at the surface and in the troposphere probably associated with the spatial distribution of the surface fluxes. This provides further motivation for implementing fully flow-dependent background errors in the analysis of atmospheric tracers.
The spread of the ensemble is estimated to be too low when comparing to the ensemble mean background error. We demonstrate the adding directly a perturbation on the surface fluxes helps to increase the ensemble spread. Nevertheless, it is found that the amplitude of the surface fluxes perturbation used in this document could be too large according to a diagnostic based on the ensemble mean background error. This emphasizes that the perturbation of the surface fluxes has to be implemented in the C-EDA experiments, but this requires further developments