Author(s): Mohammad Kamruzzaman; Simon Beecham; And Andrew V. Metcalfe
Keywords: Control chart; Hotelling T-square; Multivariate statistics; Holt-Winters method
Abstract: The aim of this study is to use a range of multivariate statistical methods to analyse the daily rainfall pattern changes at six rainfall stations in South Australia over the 50-year period from 1960 to 2010. The first analysis is based on the HoltWinters adaptive model (HW) for identifying changes in the underlying level, trend and seasonal effects. This is applied to monthly one day maxima, two consecutive day maxima and three consecutive day maxima. The HW algorithm is applied with the same smoothing parameter for trend, seasonal effect and level. Results for smoothing parameter values of 0.2, 0.15,0. 12,0. 10,0. 08 and 0.06 were applied for all stations. Based on the sum of squared one step ahead prediction errors (SS1PE), the optimal smoothing parameter values were 0.08 for Adelaide Airport, Coulta, Cowell and Mount Barker and 0.06 for Ceduna Amo and Willunga. These were applied to estimate the trend. The second analysis involved the application of a multivariate control chart based on Hotelling’s T-square, for providing evidence of changes in rainfall extremes. No significant evidence of change in the last 50 years was found. However, the analysis can be continually updated and used to monitor further trends at these and other sites.