Dominik Jacques;
Weiguang Chang;
Seung-Jong Baek;
Thomas Milewski;
Luc Fillion
Environment and Climate Change Canada
Kao-Shen Chung
National Central University, Jhongli, Taiwan
Harold Ritchie
Environmental Numerical Prediction Research Section, Environment and Climate Change Canada,
The 8th EnKF Data Assimilation Workshop
The influence of model resolution on background covariances
Poster:
dominik_jacques_outlined_text.pdf
A 2.5 km, hourly-cycled, EnKF system was tested on the Canadian West Coast for the purpose of improving coastal forecasts. This assimilation system was nested within two other EnKF systems running at resolutions of 15 and 50 km respectively. As such, this setup offered an opportunity to study the influence of model resolution on the covariances/correlations estimated by the three different systems. It is found that increasing a model's resolution generally shrinks the correlation lengths being estimated. At high resolution, a denser observational network is therefore needed to avoid discontinuous "bullseye" analysis increments. Also, it is showed how the covariances estimated between different variables often have maxima located some distance away from the reference point at which they are estimated. This last finding is interesting as it points out avenues for improving localization strategies. In this respect, the statistical concept of "forecasting efficiency", which measures the capacity of a random variable to reduce the uncertainty of another random variable through correlation, is discussed.
A 2.5 km, hourly-cycled, EnKF system was tested on the Canadian West Coast for the purpose of improving coastal forecasts. This assimilation system was nested within two other EnKF systems running at resolutions of 15 and 50 km respectively. As such, this setup offered an opportunity to study the influence of model resolution on the covariances/correlations estimated by the three different systems. It is found that increasing a model's resolution generally shrinks the correlation lengths being estimated. At high resolution, a denser observational network is therefore needed to avoid discontinuous "bullseye" analysis increments. Also, it is showed how the covariances estimated between different variables often have maxima located some distance away from the reference point at which they are estimated. This last finding is interesting as it points out avenues for improving localization strategies. In this respect, the statistical concept of "forecasting efficiency", which measures the capacity of a random variable to reduce the uncertainty of another random variable through correlation, is discussed.
Thanks to McGill's Atmospheric and Oceanic Sciences department for hosting this web-site.