Temporal changes are clearly detected by analysis of their development and
respective change point probabilities. The most important aspects of the method
are a rigorous treatment of uncertainties, e.g. of trend (days/year) or functional
behaviour, and the possibility of prediction of missing and future data with associated
uncertainties. Figure 4.12 displays the results for the analysis of a 50-year
record of flowering of lilac at Gr¨unenplan, Germany. The one change point model
is preferred by 97.9%, the linear model by 1.6% and the constant model by 0.5%
likelihood. The analysis of change point probabilities for the one change point
model reveals a clear maximum in the first half of the 1980s. The average functional
behaviour is a steady, modest delay of flowering till the mid 1980s and then
a sharp advance. The resulting trends reach −1.19 days/year, clearly different from
zero