Statistical Modelling of Maximum and Minimum Temperature in Thiruvananthapuram District

Abstract

Time series analysis on maximum and minimum temperature using ARIMA model was done separately with the monthly mean temperature for the period from 1985-86 to 2013-14. The best fit ARIMA model for forecasting was selected based on different selection criteria viz., AIC (Akaike Information Criterion), BIC (Bayesian Information Criterion), Hannan-Quinn etc. In case of maximum temperature, ARIMA (101) (111) was the best fit one whereas ARIMA (011) (011) was best fit for minimum temperature. Mean absolute error for the forecasted maximum and minimum temperature were only 0.034 and 0.14, respectively, which showed the forecasted value and original values differ least. I. INTRODUCTIONClimate change is the most important topics which is the greatest ecological, economic and social challenge of our time. Weather parameters such as temperature, rainfall, relative humidity etc. could be mainly useful for making decision and its prediction could be used for risk management. Time series analysis method have been applied to a large number of practical problems including modeling and predicting the future data according to the past. The aim of this study is to modeling and prediction of maximum and minimum temperature of Trivandrum district in Kerala based on Box & Jenkins ARIMA model. Time-series models are better suited for predicting response to precipitation than temperature, whereas panel or cross-section models are better suited for temperature. (Lobell and Burke, 2010). A study by Unnikrishnan et al.(2018) on forecasting weather parameters showed that ARIMA (011) (011) is the most commonly fitted ARIMA model for seasonal parameters. Similar study by Murat et al. (2018) on forecasting meteorological time series data found that ARIMA models is best fit for air temperature studies. In the study by El-Mallah and Elsharkawy (2016) on time series modelling and short term forecast of yearly temperature disclosed that the quadratic ARIMA model and linear ARIMA model had the best overall performance in making short-term forecasts of yearly total temperature in Libya. Muhammet (2012) used the ARIMA method to predict the temperature and rainfall in Afyonkarahisar Province, Turkey, until the year 2025, and found an increase in temperature according to the quadratic and linear trend models. Khedhiri (2014) studied the statistical properties of time series temperature data in Canada for the period 1913-2013 and determined a seasonal ARIMA model for the series to predict future temperature records as ARIMA modelling is found to be good fit for time series air temperature studies.