Pieter Houtekamer
Environment and Climate Change Canada
The 8th EnKF Data Assimilation Workshop
Use of the hybrid gain algorithm to sample analysis error
Talk:
houtekamer.pdf
Following positive results at ECMWF, the hybrid gain algorithm has been implemented at the Canadian Meteorological Centre. This algorithm recenters, at each analysis time, the ensemble of analyzed states around the mean of the 4D ensemble variational analysis and the mean ensemble Kalman filter (EnKF) analysis. In the Canadian implementation, half of the analyzed states are recentered around the variational analysis and the other half is left as is.
With the addition of a second assimilation system in the ensemble generation procedure, the ensemble spread increases. Consequently, it becomes possible to reduce the amplitude of the isotropic additive error component that is used to maintain a realistic level of spread in the ensemble system. This is desirable because the random isotropic error patterns cannot be associated with specifically known error sources.
With the hybrid gain algorithm, the EnKF inherits from the variational solver the ability to closely fit to a large number of observations.
The hybrid algorithm also recognizes that fairly independent algorithms, in which different assumptions have been made, can both add valuable information. As is commonly seen with multi-model ensembles, the multi-analysis approach leads to improved results.
Following positive results at ECMWF, the hybrid gain algorithm has been implemented at the Canadian Meteorological Centre. This algorithm recenters, at each analysis time, the ensemble of analyzed states around the mean of the 4D ensemble variational analysis and the mean ensemble Kalman filter (EnKF) analysis. In the Canadian implementation, half of the analyzed states are recentered around the variational analysis and the other half is left as is.
With the addition of a second assimilation system in the ensemble generation procedure, the ensemble spread increases. Consequently, it becomes possible to reduce the amplitude of the isotropic additive error component that is used to maintain a realistic level of spread in the ensemble system. This is desirable because the random isotropic error patterns cannot be associated with specifically known error sources.
With the hybrid gain algorithm, the EnKF inherits from the variational solver the ability to closely fit to a large number of observations.
The hybrid algorithm also recognizes that fairly independent algorithms, in which different assumptions have been made, can both add valuable information. As is commonly seen with multi-model ensembles, the multi-analysis approach leads to improved results.
Thanks to McGill's Atmospheric and Oceanic Sciences department for hosting this web-site.