A new threshold selection method for peak over for non-stationary time series

In the context of global climate change, human activities dramatically damage the consistency of hydrological time series. Peak Over Threshold (POT) series have become an alternative to the traditional Annual Maximum series, but it is still underutilized due to its complexity. Most literature about POT tended to employ only one threshold regardless of the non-stationarity of the whole series. Obviously, it is unwise to ignore the fact that our hydrological time series may no longer be a stationary stochastic process. Hence, in this paper, we take the daily runoff time series of the Yichang gauge station on the Yangtze River in China as an example, and try to shed light on the selection of the threshold provided non-stationarity of our time series. The Mann-Kendall test is applied to detect the change points; then, we gave different thresholds according to the change points to the sub-series. Comparing the goodness-of-fit of the series with one and several thresholds, it clearly investigates the series that employs different thresholds performs much better than that just fixes one threshold during the selection of the peak events.