=Paper=
{{Paper
|id=Vol-1565/bmaw2015_paper3
|storemode=property
|title=None
|pdfUrl=https://ceur-ws.org/Vol-1565/bmaw2015_paper3.pdf
|volume=Vol-1565
}}
==None==
https://ceur-ws.org/Vol-1565/bmaw2015_paper3.pdf
A Tool for Visualising the output of a DBN for fog forecasting
(Abstract only)
T. Boneh, X. Zhang, A.E. Nicholson and K.B. Korb
Faculty of Information Technology, Monash University, Australia
Abstract the remainder of the forecast period (until mid-
day the following morning). It does not provide
any way to predict the times of fog onset or clear-
Fog events occur at Melbourne Airport, Aus- ance, which is of particular interest to the avia-
tralia, approximately 12 times each year. Un- tion companies, as this will allow them to adjust
forecast events are costly to the aviation industry, flight schedules and additional fuel loads. We
cause disruption and are a safety risk. Thus, there have developed an initial prototype DBN which
is a need to improve operational fog forecasting. includes an explicit representation the fog status
However, fog events are difficult to forecast due over the forecast period. More specifically, it in-
to the complexity of the physical processes and cludes 5 gweatherh variables, plus the length of
the impact of local geography and weather ele- night, over the 8 time-slices (3 hourly forecast
ments. times, starting at 12 midday). When building
Bayesian networks (BNs) are a probabilistic rea- this prototype, we quickly found that it was dif-
soning tool widely used for prediction, diagno- ficult for both the BN knowledge engineer (au-
sis and risk assessment in a range of application thor Boneh) and our fog domain experts, to in-
domains. Several BNs for probabilistic weather spect and understand the behaviour of the DBN,
prediction have been previously reported, but to as its use was simulated over the 24 hr forecast
date none have included an explicit forecast de- cycle. This motivated the development of our fog
cision component and none have been used for DBN visualisation tool for understanding and ex-
operational weather forecasting. A Bayesian De- ploring the output of the DBN. The was devel-
cision Network (Bayesian Objective Fog Fore- oped using D3 [2], a JavaScript library for ma-
cast Information Network; BOFFIN) has been nipulating documents based on data within a web
developed for fog forecasting at Melbourne Air- browser, which uses a combination of HTML,
port based on 34 years of data (1972-2005). Pa- SVG, and CSS. The original template of the tool
rameters were calibrated to ensure that the net- was Matthew Weberfs “block” [3], which we’ve
work had equivalent or better performance to modified in a number of ways. The tool supports
prior operational forecast methods, which lead both the knowledge engineering process and the
to its adoption as an operational decision sup- use of the resultant DBN by forecasters.
port tool. The operational use of the network
by forecasters over an 8 year period (2006-2013)
Acknowledgements
has been evaluated [1], showing significantly im-
proved forecasting accuracy by the forecasters This work has been supported by ARC grant number
using the network, as compared with previous LP120100301. The authors would like to thank Tim Dwyer
years. BOFFIN-Melbourne has been accepted and other members of the Monash Visualisation group for
by forecasters due to its skill, visualisation and their assistance with the design and construction of the
explanation facilities, and because it offers fore- DBN visualisation tool.
casters control over inputs where a predictor is
considered unreliable.
References
However the static BN model now in operational
use has no explicit representation of time and [1] T. Boneh, G.T. Weymouth, P. Newham, R. Potts,
only forecasts whether or not a fog will occur for J. Bally, A.E. Nicholson, and K.B. Korb.
12
[2] Mike Bostock. D3 data-driven documents
d3.js[software]. http://d3js.org, 2015.
[3] Matthew Weber. D3 block #5645518.
http://bl.ocks.org/Matthew-Weber/5645518,
2015.
13