Hurst’s Memory for Chaotic, Tree Ring, and SOI Series

Hurst’s memory that roots in early work of the British hydrologist H.E. Hurst remains an open problem in stochastic hydrology. Today, the Hurst analysis is widely used for the hydrological studies for the memory and characteristics of time series and many methodologies have been developed for the analysis. So, there are many different techniques for the estimation of the Hurst exponent (H). However, the techniques can produce different characteristics for the persistence of a time series each other. This study uses several techniques such as adjusted range, rescaled range (RR) analysis, modified rescaled range (MRR) analysis, 1/f power spectral density analysis, Maximum Likelihood Estimation (MLE), detrended fluctuations analysis (DFA), and aggregated variance time (AVT) method for the Hurst exponent estimation. The generated time series from chaos and stochastic systems are analyzed for the comparative study of the techniques. Then, this study discusses the advantages and disadvantages of the techniques and also the limitations of them. We found that DFA is the most appropriate technique for the Hurst exponent estimation for both the short term memory and long term memory. We analyze the SOI (Southern Oscillations Index) and 6 tree-ring series for USA sites by means of DFA and the BDS statistic is used for nonlinearity test of the series. From the results, we found that SOI series is nonlinear time series which has a long term memory of H = 0.92. Contrary to earlier work, all the tree ring series are not random from our analysis. A certain tree ring series show a long term memory of H = 0.97 and nonlinear property. Therefore, we can say that the SOI series has the properties of long memory and nonlinearity and tree ring series could also show long memory and non-linearity.

Forecasting Short Time Series with Missing Data by Means of Energy Associated to Series

In this work an algorithm to predict short times series with missing data by means energy associated of series using artificial neural networks (ANN) is presented. In order to give the prediction one step ahead, a comparison between this and previous work that involves a similar approach to test short time series with uncertainties on their data, indicates that a linear smoothing is a well approximation in order to employ a method for uncompleted datasets. Moreover, in function of the long- or short-term stochastic dependence of the short time series considered, the training process modifies the number of patterns and iterations in the topology according to a heuristic law, where the Hurst parameter H is related with the short times series, of which they are considered as a path of the fractional Brownian motion. The results are evaluated on high roughness time series from solutions of the Mackey-Glass Equation (MG) and cumulative monthly historical rainfall data from San Agustin, Cordoba. A comparison with ANN nonlinear filters is shown in order to see a better performance of the outcomes when the information is taken from geographical point observation.