Retrospective-cost-based adaptive model refinement for the ionosphere and thermosphere



Mathematical models of physical phenomena are of critical importance in virtually all applications of science and technology. This paper addresses the problem of how to use data to improve the fidelity of a given model. We approach this problem using retrospective cost optimization, which uses data to recursively update an unknown subsystem interconnected to a known system. Applications of this technique are relevant to applications that depend on large-scale models based on first-principles physics, such as the global ionosphere–thermosphere model (GITM). Using GITM as the truth model, we demonstrate that measurements can be used to identify unknown physics. Specifically, we estimate static thermal conductivity parameters, as well as a dynamic cooling process. © 2011 Wiley Periodicals, Inc. Statistical Analysis and Data Mining 4: 446–458, 2011