SCIENCE OF SEASONAL CLIMATE PREDICTION

Statistical seasonal climate forecasting in Australia: An historical overview
by Roger C Stone, Queensland Department of Primary Industries and Fisheries

Dr Roger Stone holds the positions of science leader of the Climate and Systems Technologies research unit within the Queensland Department of Primary Industries and Fisheries and also Associate Professor in Climatology at the University of Southern Queensland. He is also active in a number of WMO Commissions, notably as a Rapporteur within the Commission for Climatology and as the leader of two ‘expert teams’ within the Commission for Agricultural Meteorology. He holds a PhD from the University
of Queensland.

Scientists at the Bureau of Meteorology initiated statistical climate forecasts as early as 1910 by applying Darwin pressure (now known to be linked to the El Niño - Southern Oscillation phenomenon) to provide a prediction of southern Australian rainfall. Although further experimental monthly forecasts were prepared by the Bureau (based on patterns of anticyclonicity), it was the remarkable increase in understanding of the mechanistic linkages between the Southern Oscillation and El Niño in the late 1960s and subsequent validation of earlier empirical analyses that led to the establishment of more scientifically acceptable seasonal climate forecast systems in Australia and elsewhere. Simple linear, lagged, relationships between the Southern Oscillation Index (SOI) and rainfall formed the basis of further developments in statistical climate forecasts, whether by using multiple linear regression-based systems or by applying slightly more sophisticated approaches such as principal component analysis, cluster analysis, or discriminant analysis to identify more subtle patterns of SOI activity (and rainfall patterns) over time as predictors in such schemes. Extension of such approaches, using empirical orthogonal functions of both predictors and predictands, applied to sea-surface temperature data formed a natural scientific progression from the earlier statistical attempts and form the basis of most of the currently applied systems in Australia. It has been particularly important to conduct independent verification in real time analyses and cross-validation methods to identify any potential for ‘artificial skill’, especially where a high number of predictors appear to provide an apparent increase in forecast skill but which, in fact, lead to a degradation of the forecast system (eg. Nicholls, 1997). The need for more thorough understanding of the underlying mechanisms responsible for variation in climate patterns and also climate predictors has highlighted the value of coupled general circulation models (CGCMs) where, for example, a 100-year integration of a CGCM on NINO4 and all-Australian rainfall produced a correlation coefficient of -0.45 compared to the observed value of -0.53, thereby providing validity for those approaches that apply such systems in any statistical climate forecast application (eg. Power et al. 2005).

References