2022-04-06[Event Report]Prof. Craig H. Bishop: The 87th Data Assimilation Seminar (April 6, 2022)
Following a collaborative research MOU made between RIKEN and the University of Melbourne, Prof. Craig Bishop gave a seminar talk on his latest work with the Bounded Variable Ensemble Transform Filter (BVETF), which is a nonlinear and non-Gaussian extension of the LETKF. Clouds, precipitation, ice, and aerosols are major sources of model errors when it comes to weather and climate predictions. Due to their non-Gaussian and nonlinear behaviors, observations of these variables have been studiously avoided in the data assimilation systems that assume Gaussian PDFs. Inspired by the concept of serial EnKF, the Bounded Variable Serial Ensemble Filter (BVSEF) was introduced to 1) handle bounded observations by incorporating nonlinear PDF of the prior and the observation likelihood and 2) account for the consistency between observation to state variable update via nonlinear iterations. In an idealized tropical cyclone assimilation experiment, the BVETF led to dramatic improvement over the EnKF performance where Gaussian assumption was used with linear regression mapping. Finally, in order to take the advantage of the computational scalability of the LETKF, BVSEF was incorporated into the LETKF via the new BVETF.
For other seminars, please see the following page:
