A Modified Forward Search Approach Applied to Time Series Analysis

Studying the behavior of some slow phenomena, such as crustal or continental deformations, it is necessary to consider time series data, with a sufficient numerousness that depends on their characteristics. Time series analysis is a delicate step; in order to obtain a correct significance and validity it is necessary to use an internal coherent outlier free dataset. It is necessary to analyze the data in order to find any possible outliers and to verify inner coherence. Some interesting features about outlier removal, research of zero degree unknown discontinuities and their evaluations can be provided applying the forward search method. In this case, a modified version has been developed, starting from the more general FS technique. The method has been applied to artificial data which are created in order to know the exact entity of the introduced discontinuities with the aim of simulating a time series solution that originates from GPS permanent stations networks. Simulated data were analyzed considering a period of 100 epochs and with a repeatability of 100. Experiments with different values of the ratio between the imposed jump in the series and its noise level were performed with the aim of firstly defining a percentage of positive results based on where the jumps are found and how good their entity is evaluated, and secondly in order to calculate the repeatability. The algorithms have been implemented in a automatic procedure, developed in R language. This work shows some of the obtained results and gives a statistical interpretation.