A0166
Title: Modified Mann-Kendall trend test for higher-order non-linear autocorrelation using bispectral analysis
Authors: Syafrina Abdul Halim - Universiti Putra Malaysia (Malaysia) [presenting]
Nik Auni Danisha Kamaruzzaman - Universiti Putra Malaysia (Malaysia)
Abstract: A widely adopted statistical test to assess the significance of trends in climatic and hydrologic time series is the Mann-Kendall test due to its non-parametric nature. While it is effective for detecting trends in time series, its accuracy is compromised when the data exhibits autocorrelation. Although first-order autocorrelation has been studied extensively, higher-order non-linear autocorrelation, which is prevalent in real-world climatic and hydrologic time series, has not been adequately addressed. Therefore, the aim is to examine the effect of higher-order non-linear autocorrelation on the variance of the Mann-Kendall trend test statistic by incorporating higher-order spectral analysis, specifically the bispectrum. Time series data of extreme rainfall indices from stations in Malaysia are collected to compute the bispectrum, enabling the identification of quadratic interactions between frequency components. By incorporating the computed bispectrum, a theoretical variance of the Mann-Kendall test statistic is derived. Utilizing the modified variance of the Mann-Kendall trend test statistic, a modified non-parametric trend test tailored for higher-order non-linear autocorrelated data is proposed. The accuracy of the modified trend test is evaluated through a simulation study, where both the original and modified Mann-Kendall tests are applied to the time series data, and the results are compared using performance metrics.
