Using the Ensemble Kalman Filter to estimate turbulence model parameters: Intermediate results

Simon Clement
Seminar Date: 
5. September 2018 - 12:30 - 13:00
Lecture room, Ground Floor, NERSC

Our ultimate goal is to compute the contextual model evidence in a 1D energy- length-scale turbulence closure model, and to use this model evidence to perform model selection.
In particular, contextual model evidence would be used to estimate parameters of the turbulence-closure and more generally to select the best parametrisation of the turbulence.
We use the EnKF-N to perform state estimation, and test its performance first by using a twin experiments setup with synthetic observations.
A number of aspects are studied such as the minimal ensemble size and the sensitivity to the observational network scenario.
We then moved to real observations and describe the issues and challenges related.
This seminar will present the intermediate results we got so far and the problems we encountered in particular with the non-negative and bounded variables.