The use of Bayesian analysis to detect recent changes in phenological events throughout the year

The most recent assessment report of the Intergovernmental Panel on Climate Change (IPCC, 2001. Climate Change 2001-The Scientific Basis. Contribution of Working Group I to the Third assessment Report of the Intergovernmental Panel on Climate Change. Houghton, J.T., Ding, Y., Griggs, D.J., Noguer, M., van der Linden, P.J., Dai, X., Maskell, K., Johnson. (CA Eds.), Cambridge University Press, UK.) predicted further increases in global mean temperature as well as in climate variability and extreme events. The latter changes imply an increased risk of more abrupt and non-linear changes in many ecosystems. Phenology is one of the main bio-indicators of climate change impacts on ecosystems. In order to analyse observed phenological changes accurately, we used the Bayesian approach for phenological time series analysis developed by Dose and Menzel (2004) including the model comparison option. This option offers new opportunities to analyse and quantify changes in phenological time series. Our comprehensive phenological data set consisted of long-term observational records from the 1951 to 2000 period across central Europe. We analysed the data as constant (mean onset date), linear (constant trend over time) and one change point models. The one change point model involves the selection of two linear segments which match at a particular time ("broken stick"). The break is estimated by an examination of all possible breaks weighted by their respective matching point probability. The change point model provided the best description of the data from all seasons of the year. High probabilities for this specific model reveal Europe-wide non-linear changes in phenology. The dominance of the one change point model was most pronounced for phases in summer to late autumn. Between forest trees, fruit trees and herbaceous plants, there were no significant differences in the preferred model. For different phenological stages of Avena sativa and Aesculus hippocastanum, we observed an increasing probability of the one change point model through the year. Increasing model probabilities, especially at the end of the growing season, indicate abrupt and fast changes in phenology. A detailed analysis for 11 stations in Switzerland (1959-1999) revealed that the maximum matching point probabilities of leaf unfolding records were concentrated in the mid 1980s, whereas leaf colouring displayed a more heterogeneous pattern. The strength of the assessed trends in Switzerland differed by altitude. Only at a few places did the phenological time series exhibit a constant or linear course. The frequency of maximum matching point probabilities in the 1980s indicate that most changes in our analysed data sets occurred simultaneously. The advantages of the Bayesian approach are discussed.