Sumathipala, N. S.Hewaarchchi, A. P.Dissanayake, D. M. P. V.2024-11-292024-11-292024Sumathipala N. S.; Hewaarchchi A. P.; Dissanayake D. M. P. V. (2024), The best-fitting models for weather data in Katunayaka, Sri Lanka, Proceedings of the International Conference on Applied and Pure Sciences (ICAPS 2024-Kelaniya) Volume 4, Faculty of Science, University of Kelaniya Sri Lanka. Page 202http://repository.kln.ac.lk/handle/123456789/28947Climate series patterns are affected by abrupt shifts induced due to the changes in observers, station relocation, and gauge replacements. Identifying these changes is vital in finding the best-fitted models to find the predictions. In this study, we analyze the weather data in Katunayaka, as Katunayaka is an important urban hub, where the main airport is located. This study explores the detection of structural changes in Temperature, Wind Speed, and Humidity for Katunayaka data through statistical analysis to find the best-fitted models for forecasting. Since the weather patterns are changed due to artificial factors, changepoint analysis is applied to identify abrupt shifts. A changepoint is a distributional shift or abrupt change in a time series data structure. Over the years, numerous changepoint detection methods have been proposed by researchers. The Pruned Exact Linear Time (PELT) method is one of these methods standing out for its speed and accuracy in identifying multiple changepoints in mean and variance. Due to the complexity of temporal variations in climate data, simple models often struggle to identify these shifts accurately. Climate time series also exhibit autocorrelation, and failing to account for this can lead to false detections. The objective of this paper is to fit the best models by considering changepoints to forecast weather patterns in Katunayaka, Sri Lanka, a major industrial area. For this study, temperature (°C), wind speed (km/h) data from 2007 to 2022, and humidity (%) data from 2007 to 2010 in Katunayake were used. The Pruned Exact Linear Time method was applied to detect changepoints in the mean and variance of the data. In this study there were no any changepoints detected for the temperature and humidity data but a mean changepoint was detected in wind data. A changepoint was found in March 2012 for the average wind speed for the given period. The fitted model without considering changepoints was ARIMA (2,0,1)(0,1,1) while the fitted model with considering changepoints structure was ARIMA (1,0,1)(2,1,1). The best model for forecasting average wind speed was ARIMA (1,0,1)(2,1,1) which is given under changepoint structure with Root Mean Square Error (RMSE) of 1.24 and Mean Absolute Percentage Error(MAPE) of 8.65. Then ARIMA (2,0,1)(0,1,1) model and ARIMA(0,0,0)(0,1,0) models were fitted to the temperature and humidity data respectively for the forecasting. The detected changepoint confirms a significant shift of the average wind speed. Considering the time of this shift occurs, the best-fitted model is built to predict the wind speed accurately.Changepoints, Humidity, PELT Method, Temperature, Wind