Transcript Ensemble Forecasting

Ensemble Forecasting
of High-Impact Weather
Richard Swinbank
with thanks to various, mainly Met Office, colleagues
High-Impact Weather THORPEX follow-on project meeting,
Karlsruhe, March 2013
Ensemble forecasting of High-Impact
Weather
 Challenges of convective-scale ensembles
 Ensemble-based warnings & products
 Links with other post-THORPEX initiatives
Limits of Predictability
 Following Lorenz (1984), errors grow fastest at smaller
scales, eventually affecting largest scales.
 Leads to challenges in high-resolution forecasting – in
both making and using the predictions
 Since the predictability limit is shorter for small scales,
ensembles are key to high-resolution prediction.
An Ensemble-based future
 For data assimilation, as we focus on higher resolution
(convective scales), we cannot exploit Gaussian
assumptions about the behaviour of error statistics, so
need an ensemble-based approach.
 For short-range high resolution forecasting, ensemble
methods are needed to predict the risks of severe
weather at close to the model grid scale.
 For longer range global forecasts, ensemble methods
are required to estimate the risks of high-impact
weather and produce probabilistic forecasts beyond
the limits of deterministic predictability.
Challenges of convective-scale:
modelling
 Operational centres are now starting to introduce
convective-scale ensembles.
 Gives the potential to produce much more detailed
forecasting of storm systems, but…
 Grey zone – still cannot afford to truly resolve convective
processes, rather use “convection permitting” km-scale
resolutions.
 Limited to small, (sub?) national-scale domains.
 During life of the HIW project, look forward to <1km
grid scale and larger (regional) domain sizes.
Example: MOGREPS-UK system
 Currently run as a downscaling ensemble, initial and
boundary conditions driven by 33km MOGREPS-G
(NB. No intermediate regional ensemble).
 Challenges:
 Time to spin up small scales
 Use high-resolution analysis to initialise ensemble?
Ensemble Modelling challenges
 Representing uncertainties
 Initial condition uncertainties - in MOGREPS, currently from
MOGREPS-G, but should use ensemble DA.
 Model errors – what stochastic physics is appropriate for
convective scales?
 Surface uncertainties – how to represent uncertainties in soil
moisture, surface roughness, sea surface, etc.?
 Consistency with lateral boundary conditions – movie
from Warrant Tennant
Tropical Cyclones
 Potential for improved prediction of structure & intensity
using high resolution nested ensembles.
 High-resolution simulation, by Stu Webster (Met Office)
Challenges of convective-scale:
post-processing
 How to post-process when details are unreliable?
 Neighbourhood methods for displaying output at predictable scales
observed
forecast
Threshold exceeded where squares are blue
[thanks to Nigel Roberts]
Optimising smoothing for skill
MOGREPS-UK
Heavy Rainfall forecast
17-18Z Torrential >16mm/hour
Probability Torrential Rain >16mm/hour
CT 2012/06/28 03Z VT 17-18Z
17-18Z Heavy >4mm/hour
Probability Heavy Rain >4mm/hour
CT 2012/06/28 03Z VT17-18Z
Warnings based on ensembles:
EPS-W weather impact matrix
High
Likelihood
≥60%
Medium
≥40%
Low
≥20%
Very Low
≥1%
Very Low
Low
Medium
High
Example of
EPS-W wind
gust thresholds
used for the
“Highlands and
Islands”
Impact
≥70mph
≥80mph
≥90mph
• Likelihoods of low, medium and high impact weather are presented as
probability contour maps
• These are also combined to form overall warning colour maps…
© Crown copyright Met Office
Thanks to Rob Neal, Met Office
MOGREPS-UK example – yellow warning for gales
in Orkneys & Shetlands 14-15 Dec 2012
36hr forecast
© Crown copyright Met Office
30hr forecast
HIW project - links with other ensemble
forecasting initiatives
 A trio of complementary datasets:
 TIGGE project (global medium-range EPS), since October 2006.
 TIGGE-LAM project, limited area counterpart to TIGGE, will be an
additional resource for HIW project – European LAM-EPS data now
starting to be archived at ECMWF.
 Sub-seasonal to Seasonal archive to support S2S project – coming
soon.
 All planned to use similar GRIB2 format and conventions.
 A technical liaison group (representatives from data providers &
archive centres) could manage archive.
 Proposed “Predictability and Ensemble Forecasting” working
group, focusing on science of dynamics & predictability and
ensemble forecasting.
WWRP-THORPEX
GIFS-TIGGE
working group
PDP
working group
TIGGE-LAM
panel
TIGGE-LAM
dataset
TIGGE
dataset
Users
Predictability, dynamics, probabilistic forecasting
WWRP
HIW project
team
P&EF
expert team
WCRP
S2S project
team
Dataset
liaison group
TIGGE-LAM
dataset
TIGGE
dataset
S2S dataset
Users
Sub-seasonal to seasonal and polar predictability,
high-impact weather, probabilistic forecasting, RDPs, FDPs
Summary
 Convective-scale ensembles give new challenges and
opportunities
 Opportunities




More realistic simulation of severe storms
More detailed local forecasts
Better warnings of severe weather
Exploit TIGGE & TIGGE-LAM datasets for HIW research
 Challenges





Resolving convection?
Representing uncertainties – initial and model error
Balance between resolution, domain size & members
Presentation of small-scale information
Combine short-range detail & longer range warnings
Any Questions?