Detectingoftrendsinlong-range,correlatedserieswith.docVIP

Detrending time series Stationarity Of course, we do not expect the future values of a data series to be identical to some past series of the existing observations. However, in many instances it may be very reasonable to assume that their statistical properties will be the same. The idea that past and future values of a time series will be similar statistically is an informal expression of what is called stationarity. Usually, the term stationarity is understood to mean weak stationarity or covariance stationarity. In this sense, stationarity implies that the mean and autocovariance function of the data series do not change through time. Different time slice of a stationary data series (for example, the data observed to date and the data to be observed in future) can be regarded as having the same underlying mean, variance, and covariances. Furthermore, the correlations in a stationary series are determined only by their separation in time (i.e., their lag), and not their absolute positions in time. Qualitatively, different portions of a stationary time series look alike statistically, even though the individual data values may be very different. Covariance stationarity is a less restrictive assumption than strict stationarity, which implies that the full joint distribution of the variables in time series does not change through time. Most methods for analyzing time series assume stationarity of the data. Many atmospheric processes are distinctly not stationary. Detrending Figure 1: Trend in the mean. December atmosphericConcentration (ppmv- parts per million by volume) at Mauna Loa, Hawai, 1958-2007. Source: /ftp/trends/co2/maunaloa.co2 Trend in a time series is a slow, gradual change in some property of the series over the whole interval under investigation. Trend is sometimes loosely defined as a long change in the mean (see Figure 1), but can also refer to change in other statistical properties. For example tree-ring s