Statistical methods
Statistical methods combine information from multiple dynamical models
and correct for systematic model deficiencies. Two multi-model
combination methods are used; one is Bayesian (Rajagopalan
et al. 2002 ; A.
W. Robertson et al. 2004), and the other is a canonical variate
technique (Mason
and Mimmack 2002). Both methods estimate an optimum weighting of
the individual AGCM predictions for a given season and location, based
on the past performance of seasonal simulations. This procedure
improves the forecast reliability. Multivariate spatial corrections
identify model patterns in past simulations that are related to
observed patterns temperature and precipitation and then replace model
patterns with observed ones. These corrections are usually applied on a
regional basis.