Date of Award
Doctor of Philosophy
Geological and Atmospheric Sciences
Herbert T. David
This study presents a statistical technique to partition numerical model forecast mean squared error into model deficiency components and components due to random errors in the initial conditions. In addition to the partitioning technique, this study presents a procedure to evaluate if each component's contribution to the total error, in the presence of the residual (unpartitioned) error, is negligible. The application of the partitioning techniques in an operational environment is discussed briefly;The partitioning technique and component evaluation procedure were applied to a hierarchy of global spectral models. The hierarchy consisted of a barotropic model, a "classic" two-layer baroclinic model, an "improved" two-layer baroclinic model, and a two-layer linear balance model. The models were initialized with a sample of data generated with Monte-Carlo techniques; details are presented in an appendix;The application demonstrated the relative ease in which partitioning and evaluations may be applied to model forecasts. From the analyses, the forecast time at which model differences were no longer "negligible", and how this time varied with latitude, was evident. The effectiveness of "incremental" model improvements, represented by the difference between the classical and "improved" baroclinic models were also evaluated and discussed.
Digital Repository @ Iowa State University, http://lib.dr.iastate.edu/
David Giles McDonald
McDonald, David Giles, "Partitioning forecast errors in numerical weather prediction models " (1993). Retrospective Theses and Dissertations. 10250.