Statistical Analysis and Forecasting Model for Monthly Mean Temperature in Colombo

Abstract

Colombo is an industrial city in the western province of Sri Lanka which has a high population density. Also, it is one of the cities which is warming faster due to industrial processes and lack of forestry. According to the Koppen climate classification, Colombo has a tropical rainforest climate and the city consists of a geography based on a mix of land and water. The climate of Colombo is fairly temperate all throughout the year and consists of intra-year seasonal and cyclic temperature patterns. The scope of this research is to develop a time-based regression model to forecast monthly mean temperature in Colombo with respect to the most related weather component. In` this study, monthly mean temperature anomalies(in Celsius) of Colombo city for the period of 1986 to 2008 were analyzed and furthermore monthly mean temperatures for the year 2009 were forecasted using the estimated model. Data were collected from the weather station in Colombo attached to the department of meteorology. An intra year (within the year) seasonal pattern was observed in January, February, May-July, November and December. The weather components average wind (Km/h) and average humidity (%) indicated higher correlations to the monthly mean temperature than average rainfall (mm) and cloud (oktas). Stepwise regression was performed to find the most accurate model by using the stationary set of series. The most accurate model consisted of average wind and 11 numbers of dummy variables generated due to the seasonality. The significance of each predictor and indicator for the final model were checked. The Gaussian correlated residuals were modelled using linear autoregressive model AR (2) derived from Box Jenkins methodology. The constant and the coefficients of final estimated model for forecasting monthly mean temperature were 27.1, 0.0893, 0.388, 1.04, 1.33, 1.47, 1.08, 0.838, 0.874, 0.812, and 0.374. Accordingly, the fitted regression model was accepted as the best model with R2 of 58.5 %.