The estimated predictability of seasonal mean Australian surface temperature

Abstract

The seasonal mean of a climate variable is affected by processes with timescales from less than seasonal to interannual or longer. In this paper, the seasonal mean is conceptualised as consisting of intraseasonal and slowly varying (longer than a season) components. The slow component of the seasonal mean is related to slowly varying internal and external processes which are potentially predictable, and is therefore itself regarded as potentially predictable. An analysis of variance method, which separates the interannual variance of the seasonal mean of these components, is applied to Australian surface maximum and minimum temperature. The potential predictability is defined as the percentage of the interannual variance of the seasonal mean that is due to the slow component. Using data from the Australian Water Availability Project dataset for the period 1958-2007, it is found that there is high estimated potential predictability (over 50 per cent) for surface maximum and minimum temperature for northern Australia (north of 25°S) in most of the 12 three-month seasons. In contrast, there are regions of southern Australia (south of 25°S) with low estimated potential predictability (under 30 per cent) in many seasons. The results given here provide guidance for where and when longrange forecasts of seasonal mean temperature variables are most likely to be skilful. IntroductionForecasts of seasonal mean Australian surface temperature have been produced using both statistical (e.g. Jones 1998) and dynamical forecast schemes (e.g. Hudson et al. 2011). Changes in surface temperature are also often used to assess the impact of climate variability and climate change on Australian society and the environment (CSIRO 2007). It is well known that there is a strong relationship between the Southern Oscillation Index (SOI) and Australian surface temperature (e.g. Coughlan 1979, Halpert and Ropelewski 1992, Jones 1999, Jones and Trewin 2000a, and sea surface temperature (SST) has been successfully used as a predictor in statistical forecasts of Australian seasonal mean temperature (Fawcett et al. 2005). The skill of seasonal forecasts can be verified (e.g. Fawcett et al. 2005, Fawcett 2008, Fawcett and Stone 2010. However, an alternative question can be proposed: to what extent is the seasonal mean temperature potentially predictable for any given season and region? Leith (1973) introduced the concept that the seasonal mean of a climate variable can be considered as a statistical random variable with 'signal' and 'noise' components. Within a threemonth season, the 'noise' component changes on timescales from daily to intraseasonal. Thus, it is more likely to be related to meteorological phenomena that vary significantly within the season. The 'noise' component has been referred to as the intraseasonal component (e.g. Frederiksen and Zheng 2007). The 'signal' is regarded as being constant within the season. Therefore it is more likely to arise from slowly varying (season or longer) boundary or external forcings on the climate s...