Transcript Agent-based Microburst Detection from Weather Radar

Intelligent Agents in the
Australian Bureau of
Meteorology
Sandy Dance and Mal Gorman
Introduction
•
•
•
•
•
•
•
About the Bureau of Meteorology
Project to improve forecast process
Alerts
Agents in Bureau
TAF alert pilot project
Research issues
The future
New Bureau building in March
2004, 700 Collins St, Docklands.
Forecast “Database”
A project to enhance the forecasting process,
involving:
•
•
•
•
•
Machine-readable forecasts in database
Forecaster “personal digital assistant” (PDA)
Automatic alerting
Multi “view” product generation
Integration of existing systems
Forecast DB – stage 1
radar
satellite
products
AWS
model
interfaces
db1
db2
db3
Intelligent Alerts Goals
• Forecaster PDA
• Alerts from inconsistency between
Forecast / Guidance / Observations
• Weather element alerts, eg temp
• Severe weather event alerts, eg hail
Forecaster PDA
•
•
•
•
•
•
•
Manage alerts
Sanity check for forecasts (“deviates from climate”)
Arrival alerts (ie, latest model, satellite images)
‘elephant stamps’ for successful unusual forecasts
Automatic text generation for various forecast types
Graphical editing of numerical forecast
Control of alerting through media such as SMS, email,
phone
Consistency Alerts
Inter-comparison between:
• Forecasts and observations (verification),
• Observations and guidance,
• Guidance and forecasts.
(guidance = numerical atmospheric model)
Severe weather alerts
• Storm alerts from radar
• Microburst from radar
• Tornado from radar
• Hail from radar
• Lightning from radar and GPATS
• Fronts from satellite
….this is not exhaustive!
Forecast DB - with agents
radar
Microburst
detector
Cold front
satellite
forecast
?? alert
front
detector
warning
AWS
??
detector
Storm track
special
model
??
detector
db1
db2
???
db3
…and again in more detail.
An example of an agent –based detector:
microburst detection
Reflectivity output showing detected microbursts
(see www.bom.gov.au/weather/radar/ for more radar)
Exploratory pilot project
To trial an end-to-end system employing Jack agents to alert on
discrepancies between aviation forecasts and observations.
• Inputs: TAF (forecast) and AWS (observation) data from decoders
• Passed by TCP/IP and Jacob to Jack agent network
• An agent handles subscription to data of interest by other agents
• A monitoring agent issues alerts upon discrepancies between TAF and
AWS data
• GUI subscribes to alerts and displays them under control of forecaster.
Conducted in collaboration with RMIT Agents Group and Agent Oriented
Software Pty Ltd.
A typical TAF
TAF YMML 122218Z 0024
24006KT 9999 FEW025 BKN030
FM02 18015KT 9999 SCT040
FM17 25006KT 9999 BKN025
T 15 19 20 16 Q 1028 1026 1025 1026
A typical AWS
Alerting agent pilot
Data flow view of pilot agent network.
Research issues raised
The wish list from the Bureau, plus experience from the pilot
project, highlight our requirements for a large scale Bureau
agent network. These include:
• Self-describing data
• Service description
• Service lookup
• Failure handling
• Dynamic quality-of-service management
These are research issues that will be dealt with in a possible
ARC Linkage grant in association with RMIT and AOS.
Self-describing data
We require a data representation that:
• Allows agents to interpret data from elsewhere sensibly
• Allows reasoning about data
• Allows translation between related concepts.
Could use our in-house metadata-rich tree-table-xml.
Or more generally, an object model that can represent rich
agent-oriented semantics and ontologies with data.
A research question!
Service Description
• Services will need to be advertised and searched.
• Must allow efficient reasoning about services,
• Must express the data provided, the transformations made,
and the quality of the data and service.
Could use technologies like DAML+OIL*, or extensions or
alternatives to these. Again an open research question.
* DARPA agent markup language, ontology inference language
Service lookup
Agents will need to seek data sources upon startup, as well
as continuously during operation.
• Must allow new services to compete with old
• Handle data source failure or removal by seeking
alternatives
• Handle vastly different temporal characteristics of data
sources
The future
• Extend the pilot to more stations, datatypes, forecast types,
alerting scenarios.
• Merge with forecaster GUI under development
• Incorporate severe weather detectors into the network.
• Pursue research issues to give us agents that can find and
talk to each other – possible ARC Linkage grant!
• Gradually infiltrate agents throughout the Bureau.