State-Space Analysis of Model Error: A Probabilistic Parameter Estimation Framework with Spatial Analysis of Variance.

摘要

An over-arching goal in prediction science is to objectively improve numerical models of nature. Meeting that goal requires objective quantification of deficiencies in our models. The structural differences between a numerical model and a true system are difficult to ascertain in the presence of multiple sources of error. Numerical weather prediction (NWP) is subject to temporally and spatially varying error, resulting from both imperfect atmospheric models and the chaotic growth of initial-condition (IC) error. The aim of our work is to provide new methods that begin to systematically disentangle the model inadequacy signal from the initial condition error signal.