Joint Seminar: Nonlinear and Non-Gaussian Ensemble Data Assimilation for Earth System Applications
It is now a quarter of a century since the first application of an ensemble Kalman filter for atmospheric prediction. Since that time, many extensions to the ensemble Kalman filter have been proposed and algorithm variants that are fast and scalable have been developed. However, most extensions retain the underlying assumptions of the original Kalman filter: errors have a Gaussian distribution and there is a (noisy) linear relation between observations and forecast model state variables.
Ensemble (Kalman) filter assimilation algorithms can be described as two steps. Step 1 is computing ensemble increments for an observed variable given an observation and a description of its error. An efficient algorithm that allows the use of arbitrary distributions (not just Gaussians) for this step is described. Step 2 is computing increments for a model variable given the ensemble of observation increments. In ensemble Kalman filters, this is just a simple linear regression. However, doing the regression in a transformed space can remove this linearity constraint. Transforming with a probability integral transform followed by a probit transform (these terms correspond to very basic statistics), guarantees that the posterior ensembles for state variables share the advantages gained by the improved step 1. For example, if a state variable is bounded (like specific humidity or ozone concentration) then posterior ensembles will be bounded.
These new ensemble filter algorithms can lead to significant improvements in analyses and forecasts, especially for bounded quantities and for remote sensing observations that have complex relations to model variables. Results are presented for reanalysis applications in the troposphere with a global chemical general circulation model and to the deep atmosphere with a space weather model.
Date
11.06.2025
Time
13:30 h
Place
- Bundesstr. 53, room 022/023
- Seminar Room 022/023, Ground Floor, Bundesstrasse 53, 20146 Hamburg, Hamburg
Organizers
- Nedjeljka Žagar