3pm at NORCE (building F, floor 4, end of hallway).
Talk by Dean Oliver (NORCE).
Abstract
A recent work involving a hybrid formulation of the iterative ensemble smoother for parameter estimation in cases where the relationship between model parameters and observations is highly nonlinear and the posterior distribution is potentially multimodel. The “hybrid” in the title refers to the hybrid computation of derivatives for composite functions. We take advantage of ensemble methods to avoid the need for adjoints, but use analytical derivatives where possible. I used the method to compute derivatives for a recent paper onn data assimilation in a non-centered hierarchical model. Yuming Ba and I are using it with a more complex facies-like geology to compute weights.