Monthly Rainfall Forecasting Using Bayesian Belief Networks

ABSTRACT: Bayesian Belief Networks (BBNs) provide an effective graphical model for factoring joint probability distributions under uncertainty. In this paper we introduce application of BBNs in weather forecasting. We work with a database of observations (monthly rainfall) in a network of 20 stations in Khorasan provinces (Iran), measured for the years 1985-2011 on a grid of approximately 600km resolution. Firstly, we analyze the efficiency of Tabu search algorithm to structural learning of BBN. In this step, a directed acyclic graph shows dependencies among weather stations in the area of study; we also use Netica software for parametric learning of BBNs. The comparisons show usefulness of proposed method as a probabilistic rainfall forecasting model.