Modeling Temperature Variation in Trincomalee District, Sri Lanka: Time Series Approach

Abstract

Sri Lanka has a warm and tropical climate throughout the year, with temperature variations depending on the region and the influence of monsoon winds. It is important to note that specific temperature patterns and changes can vary across different areas of Sri Lanka. The area selected for this study is Trincomalee district of Sri Lanka which is the hottest temperature measured from 1949 to 2018 and the annual and monthly average air temperature from 2013 to 2018 was collected from the Department of Census and Statistics. The main objective of this research was to develop a suitable model and forecast average temperature of the Trincomalee district. In this study, the p-value (8.292e-09) in Kruskal-Walli’s test identified the existence of seasonality. The p-value (0.01) in Augmented Dickey-Fuller (ADF) test and the p-value (0.1) in Kwiatkowski-Phillips-Schmidt-Shin (KPSS) test realized that the series was difference stationary. According to the results, Seasonal Autoregressive Integrated Moving Average (SARIMA) model was fitted to find the best model by using R programming and identified SARIMA (1,0,0) (2,1,1) [12] as the best fitted model, out of the 14 tentative models considering minimum AIC, AICc and BIC values. The p-value (=0.6559) of the Portmanteau test reveals that the residual time series are different from zero. Autocorrelation plot of residuals suggested that residuals have no correlation. Finally, the accuracy of the model was evaluated using Mean Absolute Percentage Error (MAPE) value. MAPE value of less than 10% reveals that our model is excellent.