Multi-model Bayesian assessment of climate change in the northern annular mode

A typical question in climate change analysis is whether a certain observed climate characteristic, like a pronounced anomaly or an interdecadal trend, is an indicator of anthropogenic climate change or still in the range of natural variability. Many climatic features are described by one-dimensional index time series, like for instance the global mean temperature or circulation indices. Here, we present a Bayesian classification approach applied to the time series of the northern annular mode (NAM), which is the leading mode of Northern Hemisphere climate variability. After a pronounced negative phase during the 1950s and 1960s, the observed NAM index reveals a distinct positive trend, which is also simulated by various climate model simulations under enhanced greenhouse conditions. The objective of this study is to decide whether the observed temporal evolution of the NAM may be an indicator of global warming. Given a set of prior probabilities for disturbed and undisturbed climate scenarios, the Bayesian decision theorem decides whether the observed NAM trend is classified in a control climate, a greenhouse-gas plus sulphate aerosol climate or a purely greenhouse-gas induced climate as derived from multi-model ensemble simulations. The three climate scenarios are well separated from each other in terms of the 30-year NAM trends. The multi-model ensembles contain a weak but statistically significant climate change signal in the form of an intensification of the NAM. The Bayesian classification suggests that the greenhouse-gas scenario is the most probable explanation for the observed NAM trend since 1960, even if a high prior probability is assigned to the control climate. However, there are still large uncertainties in this classification result because some periods at the end of the 19th century and during the “warm” 1920s are also classified in an anthropogenic climate, although natural forcings are likely responsible for this early NAM intensification. This demonstrates a basic shortcoming of the Bayesian decision theorem when it is based on one-dimensional index time series like the NAM index.