Data Assimilation Meetings at Reading

Date Meeting type Speakers
9 Jan 2013 Informal DARC meeting Joanne Pocock (University of Reading)
Estimation of observation error covariances using an ensemble transform Kalman filter (pdf)

Data Assimilation is the incorporation of observational data into a numerical model to produce a model state which accurately describes the observed reality. Statistics of the errors associated with the observations and model state are included in the data assimilation scheme in the observation and background error covariance matrices. Observation errors have contributions from a number of sources including instrument and preprocessing errors, errors from the incorrect specification of the observation operator and representativity errors. Errors of representativity are errors that arise when the observations can resolve scales that the model cannot. Little is known about representativity errors, and consequently they are currently not correctly included in assimilation schemes.
In this presentation we combine an ensemble filter with the Desroziers diagnostic, a method which provides an estimate of the observation error covariance matrix, to allow us to calculate a time dependent estimate of representativity error that is then used in the assimilation scheme. Using this method we were able to estimate a time varying observation error matrix that when included in the assimilation scheme improved the analysis.

Page navigation