Chaos, Potential Predictability and Model Validation of Climate Variations: Computational Techniques
by
Carsten S. Frederiksen
Bureau of Meteorology Research Centre
Coauthors: Dr Xiaogu Zheng (National Institute of Water and Atmospheric Research)

In this paper, we describe computational techniques which make it possible to separate interannual climate variability of seasonal means into chaotic and potentially predictable components. Based on analysis of variance and frequency analysis of daily time series, the techniques are applicable to both observed data sets and ensembles of multidecadal simulations using atmospheric general circulation models forced by observed sea surface temperatures and different initial conditions. Our formulation treats a seasonal mean of daily values as comprising three parts:

a forced component (due to slowly varying external boundary forcing, e.g. sea surface temperature); a low frequency internal, or chaotic, component (induced by slowly varying internal dynamics); a weather noise chaotic component (a response to rapidly varying day-to-day weather events). The forced component is potentially predictable at the long range because the forcings themselves are potentially predictable. The extent to which this component is correctly simulated is a traditional problem of model validation. Here, we also provide, as a measure of model validation, an estimation procedure for the correlation between the forced components for simulated and observed seasonal means.

Date received: July 11, 1999