Extreme streamflow time series analysis: trends, record length, and persistence

Trends can be detected in time series of extreme hydrological events. However, persistence and record length are often ignored in those analyses resulting in contradicting conclusions. The aim of this study is to evaluate their influence on trend detection in extreme streamflow time series. In this study, 108 time series of maximum and minimum streamflow in Brazil were analysed, with a minimum length of 60 years and an average of 76 years. Mann-Kendall (MK), Spearman's rho, and Pettitt statistical tests were applied to assess trends. Portmanteau and Hurst's autocorrelation tests were adopted to assess the persistence. Modifications of the MK test were used to remove the persistence effects. We found a strong persistence in the studied time series. Even after removing it, several time series remained non-stationary. Record length significantly affected the results of the analyses, with an increase in the number of trends according to the period analysed.

Automated summary

This study evaluates the influence of persistence and record length on trend detection in extreme streamflow time series. The findings show a strong persistence in the studied time series, and even after removing it, several time series remain non-stationary. Record length significantly affects the results of the analyses, with an increase in the number of trends according to the period analyzed.