ABSTRACT
Environmental numerical weather prediction (NWP) is expanding its role in science and society, to provide new understanding of and more actionable information about our environment (e.g., Holt et al. 2006, Seaman et al. 2012). Better weather guidance helps decision makers save lives and increase personal safety, protect property, enhance economic efficiency, and strengthen national defense (e.g., Stauffer et al. 2007a, Deng et al. 2012). It is therefore critical to identify, understand, quantify, and reduce as many sources of uncertainty in environmental models as possible. Because systematic model error, or bias, can be reduced by post-processing techniques, model uncertainty is often defined as that part of the model error that cannot be corrected by such methods, (e.g., Eckel and Mass 2005). In actuality, it is very difficult, if not impossible, to remove all bias. Model uncertainty, therefore, is defined here in the broadest sense to include (1) model error, whether systematic or non-systematic, in the mean or variance of the model solution, and (2) variability or sensitivity of a model solution to observational data or internal attributes of the model itself.
