Physically-based data assimilation

Abstract

Ideally, a validation and assimilation scheme should maintain the physicalprinciples embodied in the model and be able to evaluate and assimilate lowerdimensional features (e.g., discontinuities) contained within a bulksimulation, even when these features are not directly observed or representedby model variables. We present such a scheme and suggest its potential toresolve or alleviate some outstanding problems that stem from making andapplying required, yet often non-physical, assumptions and procedures incommon operational data assimilation. As proof of concept, we use a sea-icemodel with remotely sensed observations of leads in a one-step assimilationcycle. Using the new scheme in a sixteen day simulation experiment introducesmodel skill (against persistence) several days earlier than in the controlrun, improves the overall model skill and delays its drop off at later stagesof the simulation. The potential and requirements to extendthis scheme to different applications, and to both empirical and statisticalmultivariate and full cycle data assimilation schemes, are discussed. © 2010 Author(s).