On the Comparison of Observed and GCM Simulated Climate Ensembles

Abstract

Within the sampled climate ensemble framework for describing the climate, objective univariate statistical tests are presented which permit a straightforward assessment of the extent to which observed and GCM simulated climates agree or differ with respect to various first- and second-moment measures (i.e., ensemble averages and standard deviations) of the climate. As an example of this approach, the vertically averaged transient beat flux, is considered as a basic climate element and ensemble averages and standard deviations of this climate element are objectively compared fox. the same number of samples from a global data set assembled by A. Oort of GFDL and from sets of 2.5 and 5° resolution realizations with a GCM developed several years ago at NCAR. It is found that the degree of agreement between observed and simulated climate is highly dependent on the geographical location, the statistical moment used as the climate measure and whether zonally averaged or grid-point values constitute the measure. Also, for there is insufficient evidence to conclude that the use of a model with finer spatial resolution results in a substantially improved climate simulation. These same objective tests can and should be applied as a powerful diagnostic tool for model validation with other sampled climate ensembles either generated by other models or assembled from future observing systems.