| Summary: In this study, temperature data which are collected from VLBI/GPS co-located sites of two different regions in Europe Matera (Italy) and Wettzell (Germany), have been used. Since the data collected at sites are in the form of time series, time series analyses and conventional k0-sigma outlier detection have been implemented for these data sets in advance of using goodness of fit tests. The series have been decomposed to trend, signal (seasonal) and residual components. Linear trend model has been applied for the trend components. Models which consist of sine and cosine components, have been applied for signal part. Coefficients of signal and trend have been estimated by using linear least square estimation method. Thus by decomposing trend and signal effect from orginal time series, residual components of time series have been acquired. For testing goodness of fit of normal disdribution, ten goodness of fit tests have been implemented, i.e Pearson χ2 test as χ2 based tests; Kolmogorov-Smirnov, Lilliefors and Anderson-Darling test as frequentist tests; Shapiro-Wilk or Shapiro-Francia tests as regression-corelation tests; D’Agostino, Jarque-Bera, skewness and kurtosis test as moment based tests. Sufficiency of time series analyses have been analyzed with respect to the results of goodness of fit test aided by using such hypothesis, “data distribute normally” and “data do not distribute normally”, on the data which are theoretically assumed to distribute normally. Some of graphical methods have been utilized in order to support results of goodness of fit tests. By taking into consideration characters of implemented goodness of fit tests, some proposals concerning the application of goodness of fit tests have been put forward. |
