Bayesian analysis of long-term changes in phenology

The identification of changes in long-term phenological time-series is a prerequisite for the analysis and interpretation of phenological observations as bio-indicator of climate change. The new method for the analysis of phenological time series based on Bayesian concepts, recently presented by Dose and Menzel (2004), is employed to the analysis of different long term observational flowering records in Germany (1900-2003) as well as a 600 year record of cherry flowering in Kyoto, Japan. The onset of these phenological phases in relation to reported temperature variations is discussed, considering three different models for function estimation: The one change model is preferred in all data sets to the less sophisticated alternatives, the linear model and the constant model. Most of the time-series covering the 20th century at Geisenheim have a clear maximum of change point probability mid of the 1980s; the record from Kyoto spanning the last 6 centuries has a broad maximum at the beginning of the 20th century. After these maxima, the resulting rates of change indicate advancing onset of flowering, reaching – with a considerable uncertainty range - higher rates at Geisenheim (up to -1.7 days / year in 2003) compared to Kyoto (-0.1 days / year in 1998). Thus, our analysis allows to quantify these shifts in phenological phases, which turn out to be strongly dependent on spring temperature. Consequently, they mirror the marked temperature rise in the 20th century (Kyoto), especially its enhancement in the last 2-3 decades as reflected by the Geisenheim time series.