Singular spectrum analysis : with application in meteorology

Abstract

Singular Spectrum Analysis (SSA) is a more recent technique in time series analysis, finding its origins in 1980s papers by Broomhead and King. It combines classical time series analysis, multivariate statistics, multivariate geometry, signal processing, linear algebra and non-linear dynamics. SSA is not widely known among statisticians but it is a standard tool in meteorology and climatology. It is a non-parametric method that can be applied to any time series. The aim of SSA is to decompose the original series into a sum of interpretable components such as trend, oscillatory components and noise. The properties of these components can then be studied separately and future values can be forecasted. The main aims of this dissertation are the following: that of delving into the theory on which SSA is constructed and assessing its performance when applied to data from meteorology. The dataset which shall be analyzed is a two-year time series consisting of the maximum air temperature recorded in Malta at the Meteorological Office, based at the Malta International Airport.