Bayesian network probability model for weather prediction

Abstract

Weather forecasting is important for various areas. In this paper, weather forecasting system is presented based on Bayesian network (BN) model. Bayesian networks, or belief networks, show conditional probability and causality relationships between variables. This work applies BN to model the spatial dependencies among the different meteorological variables for weather (rainfall and temperature) prediction over Myanmar. In this work, regional and global weather data which are contributing to rainfall prediction of Myanmar are used for rainfall prediction. Then, inference ability of BN approximate inference algorithm in weather prediction is analyzed with experiments over independent test data sets. For model training and testing, collected historical records of weather stations between 1990 and 2006 are used. We report prediction accuracy of our model with empirical results.