Michèle De La Chevrotière;
Peter Houtekamer
Meteorological Research Division, Environment Canada
The 8th EnKF Data Assimilation Workshop
Spread-skill relationships in a hybrid gain algorithm
Talk:
chevrotiere.pdf
A hybrid gain algorithm was recently developed at the Canadian Meteorological Centre in an effort to improve the representation of error sources in the existing ensemble data assimilation system. In comparison to the EnKF control, the new system was shown to increase the ensemble spread in model space. Here, the ensemble spread in observation space is compared against observation departures to estimate the skill of the new system at reproducing flow-dependent background errors that are consistent with observation departures.
The spread-skill relationships are analyzed for different families of observations. For radiance observations assimilated from the temperature-sensitive AMSU-A high peaking channels, the hybrid system produces smaller innovations than the control experiment with more consistent background errors. For AMSU-B humidity channel data, the hybrid system is too dispersive and thus predicts observation departures that are too large. These diagnostics can inform future recommendations on the design and calibration of anisotropic background errors.
A hybrid gain algorithm was recently developed at the Canadian Meteorological Centre in an effort to improve the representation of error sources in the existing ensemble data assimilation system. In comparison to the EnKF control, the new system was shown to increase the ensemble spread in model space. Here, the ensemble spread in observation space is compared against observation departures to estimate the skill of the new system at reproducing flow-dependent background errors that are consistent with observation departures.
The spread-skill relationships are analyzed for different families of observations. For radiance observations assimilated from the temperature-sensitive AMSU-A high peaking channels, the hybrid system produces smaller innovations than the control experiment with more consistent background errors. For AMSU-B humidity channel data, the hybrid system is too dispersive and thus predicts observation departures that are too large. These diagnostics can inform future recommendations on the design and calibration of anisotropic background errors.
Thanks to McGill's Atmospheric and Oceanic Sciences department for hosting this web-site.