Similarity of 'fast-growing perturbations' and an illustrative experiment with ensemble forecasting
School of Earth Sciences, University of Melbourne, Parkville, Victoria, Australia
We address an aspect of the role that ensemble forecasting may play in five-day forecasts both globally and over the Australian region, A central question that arises in this topic is how to choose the most useful perturbations from which to run the ensemble.
The philosophy adopted here is that the 'best' perturbations are those that are representative of the analysis errors and that project onto growing synoptic modes. Such perturbation modes are found using a method designed to 'breed' a perturbation that is representative of errors introduced in an analysis cycle. These fast-growing modes (FGMs) are discussed in terms of both synoptic variability and within a theoretical framework of the dynamics of initial uncertainty, The use of pattern correlation and empirical orthogonal function (EOF) analysis to compare FGMs in an ensemble is seen to give an indication of the number of regional modes sampled in the FGM ensemble, In this paper we also wish to quantify the sensitivity of the structure of the FGMs to the nature of the 'seeding perturbations' and to the synoptic patterns obtained during the period of their generation, Statistically significant improvements have been seen as a reduction in root mean squared forecast error of three per cent globally and four per cent in the Australian region by using an averaged, ensemble 'forecast' with only two FGMs as member perturbations.
Citation: Noone, D., and I. Simmonds, 1998: Similarity of 'fast-growing perturbations' and an illustrative experiment with ensemble forecasting. Australian Meteorological Magazine, 47, 2-19.