ºÚÁÏÍø

Abstract

Research focusing on regime shift in marine time series has increased in the last decade. Last year alone, there were 140 published papers and 5500 citations within the literature. One commonly used method to detect shifts in physical and ecological time series is the sequential t-test analysis of regime shift (STARS). This method has a convenient Visual Basic Application (for Excel) and therefore is widely used by marine ecologists. In this work, we analyse, using simulated data, the limitations and accuracy of the STARS method for identifying threshold points in time series. We synthesized two groups of time series generated with the program R, each one consisting of 1000 different random series containing known change points and magnitude values. The two groups are as follows: 1) 1000 random time series without autocorrelation, and, 2) 1000 random time series with incorporated autocorrelation and seasonality. Then, all-time series are analysed using the STARS method, utilizing a CRAN-package in R that replicates Rodionov�s program. The work is still in progress; however the first results indicate that there are inaccuracies in STARS in determining the exact timing of change points. The aim of this work is to provide researchers with useful indications on the limits this method for detecting regime shifts and to provide an R routine accessible for all researchers.

Biography

Email: luca.stirnimann3@gmail.com