Australian GFE

NexGenFWS – past, present and future
Gary Weymouth, Beth Ebert, Tennessee
Leeuwenburg, Michael Foley, Phil Riley,
Nathan Faggian, Brianna Laugher, Ben Hu,
James Sofra, Tim Hume, Ying Zhao, John
Bally and others
The Centre for Australian Weather and Climate Research
A partnership between CSIRO and the Bureau of Meteorology
NexGenFWS value chain & outcomes
Probabilities
Observations
Forecaster intervention
Graphical and mobile forecasts
every 6 km
Post-processor –
corrected guidance
and uncertainties
650 automated location
based 7 day forecasts
Graphical Forecast Editor
Derive grids – automation with
support for expert intervention
Local and
overseas NWP
Gridded downstream
applications
NexGenFWS history
2014
2012
2013
2010
2011
Pilot 2008
GFE is hybrid gridded / point / area system
• Gridded
•
•
•
•
Very flexible
Naturally supports production of derived fields and downstream use of data
Natural match with gridded guidance
Natural support of probabilistic forecasts
• Point
• Site forecasts match obs well – important for credibility of Bureau service
• Area
• District forecasts utilise areal means, with automated 'local effects' to describe
significant variations across a region. Similarly for hazards like areas of storms
• Area QNH forecasts for aviation
• Some complexity in graphical representation of areal forecasts
• System-relative
• Aspects of TC forecasts
• Consistency
• Some complexity in sampling and maintaining area – grid consistency in text
• Combination of gridded and point forecasts works but has some complexity
Handling TCs in GFE – TC Iggy:
TCModule official track
GFE TC grids
Wind grids
Marine grids
de Maria
Text Formatters
hurricane possible
storm possible
Wind prob grids
34/48/64 kt
gales possible
North West Cape to Carnarvon
Winds: Above 48 knots possible
depending on movement and
development of Tropical Cyclone
Iggy.
First cut forecast
Automated routine forecast in minutes, baseline for verification
Ensemble and probabilistic guidance
Mean and spread
information (OCF),
PDFs
Forecaster
tuning?
Ensemble
members – may
include ensemble
means
Public (probability)
forecasts through
GFE
Bias
correction,
downscaling
Chance of
events
Fields derived from
ensemble members
Operational Consensus Forecasts (OCF)
• Weighted bias-corrected mean of several NWP models
• Point and grid versions
• Mainly surface level variables (T, Td, wind, precip, POP,
cloud)
Error in
hourly T
forecasts
Benefit of guidance (NWP) post-processing
Adds considerable value at low relative cost
Bureau of
Meteorology's
OCF performance
Novak et al. (2013) found QPF calibration gave equivalent skill gain to ~13 years of
GFS development
Foundations of forecast process streamlining
• Focus
forecaster
effort
• Focus
developer
effort
• Demonstrate
value
Verify
Optimal
forecaster
interventions
& efficiency
Alert
Improve
guidance &
automate
• Baseline for
verification –
‘first cut
forecast’
• Efficiency
• Allow focus
on highimpact value
add
• Focus on highimpact weather
• Fewer misses in
less time
• Safer automation
Using hi-res Rapid Update Cycle NWP
• Combine high resolution structure of RUC with largerscale bias-corrected multi-model consensus
• Stitch into GFE
Gridded Operational
Consensus Forecast
Bias-corrected RUC
• Forecast Demonstration Project in Sydney Oct-Dec 2014
Automated fire indices, difference grids, alerts
Grass Fire Danger Index
NexGenFWS summary
NexGenFWS vision achieved:
Timely and accurate forecasts where you are, when you need them.
• Massive gains to services,
efficiency, automation, standard
processes
• Effective path to operations
• Modern dissemination channels
• Downstream systems support
• Operational probabilistic forecasts
and guidance
• High-resolution forecasts
• Foundation for future
Future focus
• Automation, verification supporting efficiency,
particularly for routine forecasts
• Surge capacity where a forecast office needs help due
to severe weather
• More ensemble guidance and probabilities requiring
less editing, supporting ensemble-based services
• Hazard forecasts including probabilistic forecasts, alerts
Merci!
The end
Bureau and international context, trends and
developments – Bureau and community resilience
Routine forecast
automation
NWP – post-processing
adds great value to
NWP investments.
Continuous guidance
improvement. RUC and
ensembles.
Forecasts and
warnings more
timely, more often
GFE & VW: core
parts of NWP
value chain
Severe wx
focus, surge
capacity
Impact
forecasts
Expanded
forecaster
communicati
on role
New channels
and products
GFE is a solid platform to support forecast
automation
• Current GFE forecaster workload would be impossible without the
automation already used every shift
• automation has already enabled massive service increase.
• More automation possible and needed
• GFE as automation platform:
•
•
•
•
•
developers and forecasters use the same system
forecaster input where automation doesn’t stack up
brilliant way to capture business knowledge and rules
good development platform including visualisation
Forecaster intervention support is a small
part of total NexGen cost
• already does model post-processing and
product generation
US GFE & Australian GFE
US GFE
- Original foundation for Australian
GFE
- >100 WFOs each with SOO to
undertake developments
Australian GFE
- 7 RFCs covering huge domain
- Significant capability,
infrastructure and automation
improvements
- Centrally-managed innovation
using regional contributions
- Encapsulates Bureau business
logic
Australian NexGen and UK VW
Feature
Australian GFE UK VW
Grid/point
Grid (and 650 points)
Point text forecasts,
gridded NWP processing
Domain
Huge
Small
Automation
Auto + forecaster
edits
Fully auto/ fully manual
tendency
Handles TC
Yes
No ?
Graphics
GPU acceleration
CPU rendering
Forecast
database
Yes
No
Grid editing
Strong
Limited (metmorph being
further developed)
Visualisation
OK
Advanced
Layers
Simple
Advanced
Alerting
Limited
Strong
NWP postproc
Limited
Strong