BAYESIAN MODEL AVERAGING WITH MARKOV CHAIN MONTE CARLO FOR CALIBRATING TEMPERATURE FORECAST FROM COMBINATION OF TIME SERIES MODELS

Global warming is an important issue related to the climate and weather forecast. It is shown by significantly increasing the atmospheric temperature level. Hence, improving the forecast accuracy of temperature is an important issue. The forecast is commonly done by performing a deterministic forecast meaning that the system will generate a point forecast without taking into account the uncertainty induced by model specification as well as the nature behavior. Ensemble forecast has been introduced to overcome this problem and it has been implemented in many Ensemble Prediction Systems (EPS) over the world. A problem arises in some developing countries that unable to develop such EPS due to the system restrictions. This paper discusses the performance of combined forecasts generated from a class of time series model as an alternative of EPS. The models are calibrated using Bayesian Model Averaging (BMA) where the parameters are estimated by Markov Chain Monte Carlo (MCMC). The results show that the proposed procedure is capable to increase the reliability of the forecast.