Loïk Berre, Météo-France/CNRS (CNRM)
Simulation of error cycling and covariance filtering using ensemble data assimilation and innovations
Abstract:
Ensemble data assimilation methods based on observation and model
perturbations are powerful approaches in order to simulate
observation and model error contributions to background errors
and to the associated error cycling. This relies partly on the use
of innovation-based diagnostics for the estimation of space-
and/or time-averaged estimates of observation and background
error covariances. The ensemble approach can be seen as
a way to provide associated space- and time-dependent background
error covariance estimates. As will be shown formally and experimentally,
innovation-based estimates of background error covariances
can also be compared with ensemble data assimilation estimates,
in order to deduce some model error estimates.
Another major aspect is the application of spatial filtering
methods to ensemble estimates of covariances. This will be
shown to be important not only for correlations, but also
for variances. Filtering methods based on either localisation
or convolution-like approaches will be discussed in particular.
Four-dimensional aspects of covariance filtering will also
be tackled possibly during the talk.