Bayesian Network Probability Model for Weather Prediction

Abstract

Bayesian networks, or belief networks, show conditional probability and causality relationships between variables. Weather forecasting is important for various areas. In this paper, weather forecasting system is presented based on Bayesian network (BN) model. This work applies BN to model the spatial dependencies among the different meteorological variables for weather prediction (rainfall and temperature) over Myanmar. In this work, regional and global weather data which are contributing to rainfall prediction of Myanmar are used. Then, inference ability of BN approximate inference algorithm in rainfall 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. Prediction accuracy of the model is reported with empirical results.