Alison Fowler's Home Page : Alison Fowler : UoR, Dept Of Meteorology

Alison Fowler

Observation system simulation experiments in highly non-linear large-dimensional geophysical models.

Traditional methods of data assimilation are based on the assumption that the error statistics of the observations and prior can both be described as Gaussian. These methods of data assimilation have proven to be a powerful tool in reducing the errors in numerical weather predictions (NWP). However non-Gaussian/non-linear data assimilation is becoming of increasing interest in the Geosciences. This is due, in part, to the models and the observation operators becoming more non-linear which can potentially lead to highly non-Gaussian errors in the data. The description of the error statistics is an essential part of data assimilation, as it gives a robust way of deciding how much weight to give to the observations and prior information respectively.

It is of interest to understand how these new methods of non-linear data assimilation affect the influence observations have on the analysis. Such information may be used to:

  • improve the efficiency of the assimilation by removing expensive observations with a comparatively small impact
  • highlight erroneous observations or assumed statistics;
  • improve the accuracy of the analysis by adding observations which should theoretically have a high impact.

Within this study, I have been concentrating on three different metrics used to define the observation impact: the analysis sensitivity to the observations, relative entropy and mutual information. The analysis can be defined as the mean of the posterior distribution given by Bayes' theorem. These measures each give a very different interpretation of the observation impact. In Gaussian data assimilation it is known that these three measures tend to be in good agreement, with accurate independent observations which give information about the largest part of the state space having the greatest impact. However when the error statistics, described by the prior and likelihood, are no longer Gaussian this agreement breaks down.

More information.