Effective observing network design is an important component of the data assimilation problem. The observing network is composed of both fixed and adaptive components, with the adaptive component usually being deployed with the goal of decreasing the forecast error for a particular event. The fact that we are dealing with a non-normal, time-dependent system adds considerable complexity to the problem. Evolved forecast errors don't necessarily look like the initial errors that spawned them, and the perturbations that grew rapidly yesterday are not necessarily the ones that will grow rapidly tomorrow. Because errors evolve structurally, and propagate faster than particular features (with the group speed, rather than the phase speed), model dynamics have been incorporated into adaptive observing techniques. Thus, adaptive methods have advanced beyond observing features of interest (such as when "Hurricane Hunters" observe in and around the hurricane itself), to include observations of upstream or precursor features. There are different methods for incorporating model dynamics into observing system design; including adjoint sensitivity-based techniques and ensemble transform Kalman filter techniques.
A desirable capability of observing system design is the quantitative prediction of the impact of the hypothetical observing network on the forecast error. As such, adaptive observing techniques have been extended beyond consideration of the model dynamics only, to include the characteristics of the data assimilation system itself. In this presentation, aspects of observing network design, with the primary focus on adaptive observing, will be presented in the context of both simple and operational systems. The impact of limiting assumptions (such as linear perturbation growth), will be discussed.
Presentation Files (Zip Archive)
Copyright. All Rights Reserved.