Description:  Research questions about the dynamics that drive extreme weather events in numerical weather prediction (NWP) models can often be posed as: “How is the forecast of this event sensitive to the model’s initial conditions? How can I modify the initial conditions in order to provoke a more/less intense event?” The adjoint of the NWP model can characterize this sensitivity, but the adjoint’s accuracy is limited by being linearized along the NWP model’s forecast trajectory, which makes it applicable only to how very small perturbations to the initial state may yield small changes to the forecast. Borrowing from incremental 4DVAR data assimilation, I will present a case study using an iterative method to successively perturb the initial conditions in regions of sensitivity. This method allows for much larger initial and final perturbations, as well as a relaxation of the adjoint model’s linearity constraint.

Location:

Click here to join the meeting