Extreme event return times in long-term memory processes near 1/f

Abstract. The distribution of extreme event return times and their correlations are analyzed in observed and simulated long-term memory (LTM) time series with 1/f power spectra. The analysis is based on tropical temperature and mixing ratio (specific humidity) time series from TOGA COARE with 1 min resolution and an approximate 1/f power spectrum. Extreme events are determined by Peak-Over-Threshold (POT) crossing. The Weibull distribution represents a reasonable fit to the return time distributions while the power-law predicted by the stretched exponential for 1/f deviates considerably.

For a comparison and an analysis of the return time predictability, a very long simulated time series with an approximate 1/f spectrum is produced by a fractionally differenced (FD) process. This simulated data confirms the Weibull distribution (a power law can be excluded). The return time sequences show distinctly weaker long-term correlations than the original time series (correlation exponent γ≈0.56).