کارایی چند روش متفاوت برای محاسبه ی نمایه ی هرست در سری های زمانی اقلیمی

Abstract

Fractional Gaussian noise (fGn) is an important and widely used self-similar process, which is mainly parametrized by its Hurst exponent (H) to specify its long-term persistence (LTP). Many researchers have proposed methods for estimating the Hurst exponent of fGn. But there is only a few researches that has compared different methods for different time series covering different length of records. In this paper, we have compared the performance of 7 different methods covering rescaled range (R/S), 3 different approaches of aggregated standard deviation method (ASD[0], ASD[rec], ASD[opt]), variance method (VAR), and 2 approaches of autocorrelation method ([1] and [2]). Seven different time series including Mashhad annual temperature (127 and 66 years), yearly minimal water levels at the Nile River (660 years), two global phenomena of North Atlantic Oscillation (NAO) (62 years) and two Pacific Decadal Oscillation (PDO) series (112 and 331 years), and concentration of atmospheric CO2 measured at Mauna Loa, Hawaii (55 years) were considered. The results showed that NAO and CO2 series do not have LTP (H<0.5). The VAR method failed to satisfy the scaling relationship, [1] and [2] failed to have good correlation, ASD[rec] failed to work under T66 and CO2 series, both ASD[rec] and ASD[opt] yielded inconsistently high H and standard deviation. R/S and aggregated standard deviation showed to be suitable methods for computing H. It was shown that H decreases as record length increases where includes historic data.