Time series analysis and forecasting with applications to climate science

Abstract

The current research study introduces the novel use of Singular Spectrum Analysis (SSA) as an effective method for conducting data analysis and predicting time series. The SSA method is recognised for its proficiency in evaluating time series that ex- hibit non-stationarity and non-linearity. It does this by decomposing such series into separate components which effectively represent trends and rhythmic patterns. The at- tractiveness of this approach is its simplicity of execution which only requires the use of a couple of parameters. The current research examines the daily temperature and humidity data collected from five meteorological stations located in Saudi Arabia for the period spanning from Jan- uary 2017 to December 2021. The findings indicate a statistically significant increase in temperature and humidity levels within the region of Jeddah, with a confidence level of 95%. Nevertheless, no notable changes were observed in the cities of Tabuk, Dammam or Abha. A statistical framework was devised to assess and contrast various methods used for SSA forecasting. The study conducts comparative studies to assess the accuracy of SSA forecasts in relation to established methods such as autoregressive integrated mov- ing average (ARIMA), and exponential smoothing state space (ETS). The results of the study suggest that ETS exhibits a comparative advantage because they offer more ef- fective forecasting approaches for daily temperature and humidity relative to the alter- native models. In summary, the findings of the current study illustrate the capacity of SSA as a viable method for data manipulation and a dependable instrument for predict- ing intricate time series. However, when compared to ETS, the latter provides much lower RMSE results, this could be due to the limited scope of the study used in the research, which only analysed daily data from five years, making the seasonality less clear. Furthermore, the current study makes a substantial contribution to the domain of time series analysis and forecasting by demonstrating its exceptional efficacy in terms of predicting daily temperature and humidity.