Temporal and spatial Bayesian network modelling for improved fog forecasting

Accurate fog forecasts can vastly improve public safety and the economics of air, road and marine transport. Despite advances in weather simulations, fog forecasting is still a subjective art perched upon the experience of the forecaster. Early efforts have seen a Bayesian network-based decision support system near completion for Perth airport, but its models are limited, ignoring crucial physical issues. We aim to develop new methods for creating spatiotemporal Bayesian decision networks that will improve fog forecast models, evaluated using new, robust techniques. The new methods will also help forecasters by giving informative discretisations of input weather data, employing machine learning and integrating economic and other costs.