Analog Ensemble Data Assimilation in a Quasigeostrophic Coupled Model
Abstract
State of the art data assimilation methods use ensemble forecasts. Ensemble forecasts with high resolution models or coupled models incur computational costs that can be prohibitive for all but the most well-supported modeling centers. Faced with the high cost of these methods, but still wanting to use expensive, high-fidelity models, the practitioner is forced to scan down the list of available DA methods in search of one that will produce the best results while fitting within the computational budget. There is thus impetus to develop methods that may be less accurate than the state of the art, but that maximize computational efficiency. One approach is to use a fixed set of ensemble perturbations, so that a single expensive forecast can be augmented into an ensemble by simply adding the ensemble perturbations. This is how ensemble optimal interpolation (EnOI) works. An improvement to this approach is to find an ensemble of states from a catalog (e.g. a reanalysis) that are close to the forecast, and then re-center this ensemble of analogs on the single forecast to create an ensemble of similar states. These similar states are called analogs (despite the name, this is different from analog forecasting). Another approach is to construct synthetic analog model states instead of finding them in the catalog; this can be done using machine learning or more classical approaches. Four analog ensemble data assimilation methods, including two new ones, are compared here in the context of a coupled model of intermediate complexity: Q-GCM. Depending on the method and on the physical variable, analog methods can be up to 40% more accurate than EnOI.
- Publication:
-
AGU Fall Meeting Abstracts
- Pub Date:
- December 2022
- Bibcode:
- 2022AGUFMNG32A..07G