A catastrophe identification method for rainfall time series coupled sequential Mann-Kendall algorithm and Bernaola Galvan algorithm: a case study of the Qinglong River watershed, China

Abstract

AbstractThe identification of rainfall catastrophe characteristics is important for rainfall consistency testing in hydrostatistical analysis. In this study, a new classification method (trend, mean and change-rate catastrophe) was proposed and applied to the Qinglong River watershed, Northern China. Two groups of algorithms were compared to obtain the optimal algorithm: the Cumulative-anomaly method and the Sequential Mann-Kendall (SQ-MK) algorithm, the Pettitt algorithm and the Bernaola-Galvan heuristic segmentation (B-G) algorithm. Its parameters were optimized and its robustness was tested. Results revealed that: (1) The SQ-MK algorithm was suitable for trend catastrophe and sensitive to the length of the time series. The most significant point of trend catastrophe in the Qinglong River watershed was in 2012. (2) The sensitivity of parameter P0 (Range value (R) = 4.333) in the B-G algorithm was greater than that of parameter l0 (R = 2.889). (3) The B-G algorithm was suitable for identifying mean catastrophes and insensitive to the length of the time series. In the Qinglong River watershed, mean catastrophe points were identified in 1997, 2002, 2004, 2006, 2009, 2012, and 2018. (4) There was no change-rate catastrophe point in the Qinglong River watershed. Trend catastrophe and mean catastrophe do not necessarily lead to change-rate catastrophe.