Nonlinear high-dimensional data assimilation Fully nonlinear data-assimilation are needed to get better predictions in high-dimensional environmental models (atmosphere, ocean etc.). However, the high dimension makes this very difficult NCEO is world-leading in this field and has made very good progress using so-called particle filters, giving NCEO a unique capability.
uncertainty in whole ensemble of model runs Particle filter x observation individual model run uncertainty in whole ensemble of model runs x x time Note the growing uncertainty between observations, and the contraction of the ensemble of model runs at observation times.
Standard particle filter Truth Standard particle filter Vorticity field evolution showing interacting eddies Left : Truth Right: ensemble mean of standard method, with data assimilation every 50 time steps. Results are not very good.
New Particle filter Truth Vorticity field evolution showing interacting eddies Left : Truth Right: ensemble mean of new method, with data assimilation every 50 time steps Results are much better!
Future The new method is now implemented and being tested in several high-dimensional environmental models, such as an ocean model and a climate model.