Ensemble of Data Assimilations - EDA

An Ensemble of Data Assimilations (EDA) is an ensemble of independent 4D-Var data assimilations where the main analysis error sources (observation errors, model errors, and boundary condition errors) are represented by perturbing the related quantities - respectively observations, forecast model, and soil moisture + sea-surface temperature + sea ice, according to their estimated accuracy.  The EDA analyses are generated by randomly perturbing the observations and, across appropriate length scales, the sea surface temperature, sea ice and soil moisture fields.  The observations are assumed unbiased (once any dynamic bias correction has been applied), with observation errors assumed to have a normal distribution.  Model error is simulated using the Stochastically Perturbed Parameterisation Tendencies scheme (SPPT).  The same SPPT configuration is used in EDA as in ENS.  Differences between pairs of analyses (and forecast) fields have the statistical characteristics of analysis (and forecast) error. 

Fig5.1.7: An idealized schematic showing how the 12h assimilation window used by 4D-Var (left part of the diagram) modifies the initial trajectories of the EDA members (in blue) to reflect the information from the assimilated observations (black dots with error bars).  The analysis trajectories (in green) have taken into account the new observations and thus are confined within a narrower ensemble.  Assimilating the new observations reduces the spread.  Also a bias has been corrected by reducing the magnitude of some of the largest values in the original ensemble.

At the end of the assimilation window the EDA is used to provide:

• background error information for the following deterministic analysis update.
• the initial perturbations, around the control analysis, for the next ensemble forecasts
  

The advantages of the EDA system are:

• it provides perturbations more widely and uniformly over the globe than SVs, with larger amplitudes over the tropics.
• it attempts to explicitly model analysis uncertainty, incorporating observation coverage and accuracy.

A disadvantage of the current EDA system is:

• the variability of the perturbations does not grow sufficiently through the forecast (perturbations are currently under-dispersive).

Additional Sources of Information

(Note: In older material there may be references to issues that have subsequently been addressed)

• Read more about quantifying forecast uncertainty.
• Read more on EDA.