Fog events occur at Melbourne Airport, Australia, approximately 12 times each year. Unforecast events are costly to the aviation industry, cause disruption and are a safety risk. Thus, there is a need to improve operational fog forecasting. However, fog events are difficult to forecast due to the complexity of the physical processes and the impact of local geography and weather elements.
Bayesian networks (BNs) are a probabilistic reasoning tool widely used for prediction, diagnosis and risk assessment in a range of application domains. Several BNs for probabilistic weather prediction have been previously reported, but to date none have included an explicit forecast decision component and none have been used for operational weather forecasting. A Bayesian Decision Network (Bayesian Objective Fog Forecast Information Network; BOFFIN) has been developed for fog forecasting at Melbourne Airport based on 34 years of data (1972-2005). Parameters were calibrated to ensure that the network had equivalent or better performance to prior operational forecast methods, which lead to its adoption as an operational decision support tool. The operational use of the network by forecasters over an 8 year period (2006-2013) has been evaluated [1], showing significantly improved forecasting accuracy by the forecasters using the network, as compared with previous years. BOFFIN-Melbourne has been accepted by forecasters due to its skill, visualisation and explanation facilities, and because it offers forecasters control over inputs where a predictor is considered unreliable.
However the static BN model now in operational
use has no explicit representation of time and
only forecasts whether or not a fog will occur for
the remainder of the forecast period (until
midday the following morning). It does not provide
any way to predict the times of fog onset or
clearance, which is of particular interest to the
aviation companies, as this will allow them to adjust
flight schedules and additional fuel loads. We
have developed an initial prototype DBN which
includes an explicit representation the fog status
over the forecast period. More specifically, it
includes 5 gweatherh variables, plus the length of
night, over the 8 time-slices (3 hourly forecast
times, starting at 12 midday). When building
this prototype, we quickly found that it was
difficult for both the BN knowledge engineer
(author Boneh) and our fog domain experts, to
inspect and understand the behaviour of the DBN,
as its use was simulated over the 24 hr forecast
cycle. This motivated the development of our fog
DBN visualisation tool for understanding and
exploring the output of the DBN. The was
developed using D3 [
Acknowledgements This work has been supported by ARC grant number LP120100301. The authors would like to thank Tim Dwyer and other members of the Monash Visualisation group for their assistance with the design and construction of the DBN visualisation tool. [1] T. Boneh, G.T. Weymouth, P. Newham, R. Potts, J. Bally, A.E. Nicholson, and K.B. Korb.