Changes in probability of sequences, exit time distribution and dynamical entropy in the Potsdam temperature record

Investigations of the temporal change of the probability of defined sequenc es and dynamical entropy (Nicolis et al., 1997) are presented using the dai ly maximum and minimum of the air temperature of the Potsdam time series, w hich spans the 102 years from 1893-1994. To reduce the number of parameters a nonhierarchical cluster analysis algorithm (Steinhausen and Langer, 1977 ) was used. This algorithm assigns one of five clusters each day, classifie d as very cold, cold, temperate, warm or hot. According to the algorithm, e ach day is described by only one of the symbols vc, c, t, Mt or h. Subseque ntly a time series analysis on this series of symbols was performed. The ba sic analysis length was defined by a 30 year window, which was shifted over the entire time series thus yielding 73 analysis sections. The quantities vary from window to window. Their variations were used for the detection of climate change. One of the major findings was the increased persistence of typical temperature characteristics. Within the windows we investigate rel atively long-term correlations which extend over many days. The results sho w the time series to be markedly non-Markov.