GLAMEPS: Grand Limited Area Model Ensemble Prediction System Plans and ideas for a contribution to a

1
GLAMEPSGrand Limited Area Model Ensemble
Prediction SystemPlans and ideas for a
contribution to a European-wide contribution to
TIGGE LAM

• Trond Iversen
• met.no Univ. Of Oslo
• HIRLAM MG-member

2
The GLAMEPS objective
is in real time to provide to all HIRLAM and
ALADIN partner countries an operational,
quantitative basis for forecasting probabilities
of weather events in Europe up to 60 hours in
advance to the benefit of highly specified as
well as general applications, including risks of
high-impact weather.
3
Expectations from SR ensemble prediction

• How certain is todays weather forecast?
• What are the risks of high-impact events?
• Forcasted risk probability x potential damage
  (vulnerability)
• Predictability of free flows decreases with
  decreasing scales
• i.e. higher resolution increases the need for
  information about spread and the timing of spread
  saturation
• Predictability of forced flows may be longer than
  for free flows
• i.e. it can give benefits to separate
  unpredictable features from
• those strongly influenced by surface contrasts
• e.g. topography, coast-lines, land-use, etc.

4
Aspects to consider

• Operational aspects
• Constructing initial and lateral boundary
  perturbations
• Each LAM-version uses its own data-assimilation
• Imported global eps-members
• Calculate model-specific perturbations (SVs,
  ETKF, SLAF,)
• Lower boundary data perturbations
• Stochastic perturbations
• Switch surface schemes
• Model perturbations
• Switching models (e.g. Aladin and Hirlam)
• Switching physical packages (e.g. Straco, RKKF,
  ECMWF-physics)
• Stochastic perturbations (EC Cellular automata.)
• Forcing Singular Vectors
• EPS-calibration and probabilistic validation
• Post-processing, graphical presentation, products
• Further downscaling to meso- and convective scales

5
Ideas for GLAMEPS

• An array of LAM-EPS models or model versions
• Each partner runs a unique model version
• Partners who run the same model version,
• use different lower boundary data,
• or different ways of producing initial and
  lateral boundary perturbations
• Each partner runs between 5 and 21 ensemble
  members based on initial and lateral boundary
  perturbations
• (one control pairs of symmetric initial
  perturbations)
• Grid resolution
• Now 20km, later 10km or finer, 40 levels,
  identical in all model versions
• Forecast range
• 60h (shorter?) - starting daily from 12 UT
  (should be more frequent?)
• A common pan-European integration domain
• Or alternatively a minimum common overlap

6
A proposed commondomain

• Basic (lat,lon) (68N,40W)
• Size
• 0.2 deg 301 x 381
• Lower left corner i -40, j -260
• upper right corner i 260, j 120
• - 0.1 deg 601 x 761
• Lower left corner i -80, j -520
• upper right corner i 520, j 240

7
AlternativelyProposed minimum common overlap of
domains

• Basic (lat,lon) (68N,40W)
• Size
• 0.2 deg 211 x 233
• Lower left corner i 30, j -136
• upper right corner i 240, j 96
• - 0.1 deg 421 x 465
• Lower left corner i 60, j -272
• upper right corner i 480, j 192

8
Quality objective

• To operationally produce ensemble forecasts with
• a spread reflecting known uncertainties in data
  and model
• a satisfactory spread-skill relationship
  (calibration) and
• a better probabilistic skill than the operational
  ECMWF EPS
• for
• the chosen forecast range of 60 hours
  (negotiable)
• our common target domain and
• weather events of our particular interest
• (probabilistic skill parameters).

9
First Phasea GLAMEPS laboratory at ECMWF

• To select a small set of model versions which are
  equally valid but significantly different,
• 3 different models
• ALADIN, HIRLAM-STRACO, HIRLAM RK-KF (or
  ECMWF-physics)
• To construct initial/lateral boundary
  perturbations
• ECMWF TEPS / EPS (build on met.no LAMEPS)
• Could be adjusted by defining e.g 3 different
  targets
• Ensemble calibration (partly build on INM SREPS)
• spread-skill-relation, ensemble size
• Probabilistic estimation (e.g. BMA),
• Predictability of the day
• Quality Reliability, BSS, ROC, Value,

10
Operational

• To set up a first phase suite tested at ECMWF
  (SPNOGEPS)
• Over 1-2 years, this suite should gradually
  become distributed to partners, and run in RT by
  the end of 2008
• NB The success of GLAMEPS relies critically on
  dedicated partners for this! And we will most
  probably need dunding for supporting staff at
  ECMWF.
• To use ECMWF for data exchange in RT
• (NB Selected data set for forecasting,
• re. TIGGE-LAM recommendations)
• To use ECMWF software to provide a default set of
  graphical presentations

11
Further RD in parallel

• Through research to gradually increase ensemble
  size and error-sources
• Include lower boundary perturbations and other
  types of model perturbations
• vary model coefficients (all)
• Targeted Forcing SVs or Forcing Sensitivities
  (KNMI, met.no),
• weak 4D-Var perturbed tendencies (KNMI?)?
• Stochastic physics (DMI)
• Include alternative initial/lateral boundary
  perturbations
• ETKF generalized breeding (SMHI),
• HIRLAM and ALADIN LAM SVs (KNMI, SMHI, HMS),
• Pdf-estimation, presentation, validation etc.
• BMA, (KNMI, INM)
• Products
• Validation

12
Tiziana PaccagnellaTIGGE-LAM output and formats

• The starting point is the list of parameters
  defined by the TIGGE WG. TIGGE-LAM can of course
  add or withdraw data to this list .
• Recommended Fields must be produced with the
  same unit and keeping the same philosophy as
  regards cumulating and averaging periods (e.g.
  precipitation must be cumulated from the forecast
  start). The homogeneity between data from Global
  and LAM ensemble will make easier (feasible)
  verification and comparison.

13
TIGGE-WG single level fields (GRIB-2)

14
TIGGE-WG single level fields cont. (GRIB-2)

15
TIGGE-WG upper air fields

• 5 parameters on 8 pressure levels 1000, 925,
  850, 700, 500, 300, 250 and 200 hPa.
• Geopotential height on 50 hPa as well.
• 41 fields in all.

16
TIGGE-WG - formats

• GRIB2 agreed by all partners
• Units, names of fields, accumulation periods,
  etc.. Will be identical for all data providers
• Grids
• Data Providers will be asked to provide data on
  grids of their own choosing, which are as close
  as possible to the native grid employed to carry
  out the predictions
• Data Providers should ensure that appropriate
  software is available to the Data Centres to
  enable users to interpolate data to
  latitude/longitude grids and locations of their
  choosing
• Data Providers should ensure that when revisions
  to their systems are made the interpolation
  software will still work.

17
ECMWF and GLAMEPS

• Operationally produce alternative intial/lateral
  boundary perturbations
• TEPS / EPS
• Refinement of TEPS (higher resolution, shorter
  OT, 2-3 targets)
• Data exchange central in RT operation
• A selected set of data from TIGGE-list copied to
  ECMWF in RT each LAM-EPS.
• At an agreed time, all partners can download the
  set of GLAMEPS members.
• Archiving
• Archiving EPS and TEPS for use by GLAMEPS
• Archiving GLAMEPS raw data and products
• Use software developed at ECMWF for
• Selected probabilistic products,
• Probabilistic verification and validation
• Calibrate and validate the entire GLAMEPS
• Develop and maintain
• Prototype codes and scripts for downloading by
  partners,
• Testing and quasi-operationalization in research
  mode,
• Further co-operate with ECMWF staff,
  scientifically and operationally.

18
Thank you for your attention
Mother of Pearl clouds over Oslo, January 2002
19
Discussions

• Basic ideas and expectations to a GLAMEPS, and to
  a wider SRNWP LAM-EPS
• selecting large ensemble sizes vs. grid
  resolution and model sophistication.
• Is there an upper limit to the ensemble size
  beyond which little new meaningful information is
  gained?
• If yes, how is this diagnosed, and do we know
  why?
• What kind of products? Is estimating pdf-s
  over-ambitious?
• choices and priorities for
• ways to perturb initial conditions (SV, ETKF,
  SLAF)
• perturbing model parameter, model physics
  tendencies, or running multi-models
• selecting a common model domain (or a common
  domain of interest)

20
Discussions II

• running eps with completely different models and
  model versions
• how to calibrate?
• Can useful information be deduced about model
  quality?
• And present forecast quality?
• storing data at ecmwf for further downloading by
  members practical solutions
• implementing TIGGE rules for data storage, and
  other aspects related to data and formats,
• such as combining results from different grids
  and model levels (PEPS experience?)
• presentation of probabilistic results, using
  available software from ecmwf or elsewhere
• probabilistic validation, using available
  software
• eps-calibration a potentially huge task given
  the different types of perturbations.