WORKSHOP ON SHORTRANGE ENSEMBLE PREDICTION USING LIMITEDAREA MODELS

1
Limited-Area Ensemble Prediction the ARPA-SMR
LEPS system Stefano Tibaldi, Tiziana
Paccagnella, Chiara Marsigli, Fabrizio Nerozzi,
Andrea Montani ARPA-SMR With contributions
from F.Molteni, R.Buizza and H.Hersbach, all at
ECMWF, at some time
2
OUTLINE

• The need for the regionalisation of scenarios
• The LEPS approach
• Methodology
• Some case studies
• Statistical evaluation
• COSMO-LEPS
• Concluding remarks

3
THE NEED FOR REGIONALISATION OF SCENARIOS (1)

• Despite the recent increase of computer power
  resources, which have allowed the development of
  more and more sophisticated NWP models, the
  accurate forecast of extreme weather conditions,
  especially when related to intense and localised
  precipitation structures, is still difficult.
• This limitation is due, among other reasons,
  to the inherent low degree of deterministic
  predictability associated to this kind of
  phenomena.

4
THE NEED FOR REGIONALISATION OF SCENARIOS (2)

• Global-model ensemble systems have been shown
  to be important tools to tackle the
  predictability problem beyond day 3.
• Operational ensemble systems are usually run
  at a coarser resolution with respect to single
  deterministic model integrations for obvious
  economy reasons.
• The EPS skill in producing quantitative
  forecasts of intense and localised events in the
  short- and early-medium-range is still limited,
  although growing.

5
THE LEPS APPROACH

• The main purpose of the LEPS project is to
  produce a system capable of providing the
  forecaster some probabilistic guidance to
  identify the possible occurrence of severe
  weather conditions in the time range
• late-short-range (gt48h) up to
  early-medium-range (120h).

6
THE LEPS APPROACH
LEPS is designed to combine in a single
system the (supposed) ability of a global
ensemble prediction system to generate an
exhaustive set of large-scale evolution scenarios
(through an adequate sampling of the phase-space
in the neighbourhood of the best available
initial conditions) With the (supposed)
capability of a LAM of detailing atmospheric
phenomena on the local scales, particularly the
precipitation field, in regions with complex
orography
7
The B.F. (Brute Force) APPROACH

• The obvious solution one LAM integration
• for each global EPS member
• All the information from the global EPS is
  retained
• BUT
• it is hardly feasible on an operational basis
  (at least
• at ARPA-SMR)

8
B.F. REQUIRED RESOURCES (evaluated on COSMO-LEPS
configuration)

• For each LAM run
• Computer time
• 120 hours of LAM integration
• 306 x 258 x 32 grid points
• 13 hours cpu on vpp5000
• Data volume (IC and BCs)
• 0.9 GB for each member

51 runs 663 hours cpu! 46 GB!
9
THE LEPS APPROACH

• The LAM is nested in only a limited number of
• members selected from the global EPS, the
• Representative Members
• Some of the information from global EPS is
  lost
• BUT
• the operation becomes feasible on an
  operational basis

10
Most Representative Member

• one per cluster
• choice is based on selected 3D fields has to
  be the closest to the mean of its own cluster
  AND the most distant to the other clusters means
• 5 runs instead of 51, 102 or 153!!

11
LEPS Limited area Ensemble Prediction System
EPS and ensemble size reduction
Cluster members chosen as representative members
(RMs)
12
LEPS Limited area Ensemble Prediction System
LAM scenario
LAM scenario
LAM scenario
13
The basic idea was to leave the task of exploring
the phase space to the global EPS, while the LAM
has to zoom in the forecast, producing adequately
intense local phenomena (e.g. precipitation
maxima),but.
14
Precip outliers 91 ECMWF EPS DJF 1999/2000
15
QUESTIONS Are we adequately sampling the space
of possibilities (i. e. the phase space around
the initial conditions)?
16
Spread EPS started at 14 May 1999, 00 UTC
The same 700hPa isoline plotted for the 51
Members at 72
The same 700hPa isoline plotted for the 51
Members at 120
17
Spread vs error
Mean square error/mean square spread
18
How can we improve our exploration of the phase
space, increasing the available number of EPS
forecasts? By resurrecting the old concept of
time-lagged ensemble forecast! The
Super-Ensemble!
19
Precip outliers 91 ECMWF EPS DJF 1999/2000
20
LEPS super-ensemble
21
Summary of the LEPS methodology
Super ensemble 2 global ensembles starting 5/3
days before the verification time
102 members (50 1)2
Hierarchical Cluster Analysis method Complete
Linkage area Southern Europe fields 4 variables
at 4 levels (3D cluster) number of clusters
fixed to 5
5 clusters

• Representative Member Selection
• one per cluster
• base on the nearest (3D fields) to the mean of
  its own cluster AND the most distant to the other
  clusters means

5 representative members (RMs)
5 LAMBO integrations nested on 5 RMs LEPS -
Limited-area (High Resolution) Ensemble
Prediction System
22
LEPSSOME CASE STUDIES
23
Soverato 8, 9, 10 - Sep - 2000
24
(No Transcript)
25
(No Transcript)
26
(No Transcript)
27
(No Transcript)
28
(No Transcript)
29
(No Transcript)
30
Soverato flood
31
Soverato (Calabria) flood
91
ECMWF proxy of 24h cumulated precipitation
10/09/2000 at 00Z
32
Between 9 and 10 September, rainfall peaks above
300 mm in 24 hours were recorded close to the
village of Soverato this causing landslides,
great disruption and losses oflife.
8 9 10
8 9 10
33
Soverato (Calabria) flood
Ensemble T255
P gt 20 mm/ 24 h
P gt 50 mm/ 24 h
P gt 100 mm/ 24 h
100
30
ECMWF probability maps for 24h cumulated
precipitation exceeding P threshold at 60h
34
Soverato (Calabria) flood
LEPS 5 LAMBO runs driven by the 5 RMs selected
from 153 members of the Super-Ensemble TL255
75
42
58
47
LEPS probability maps for 24h cumulated
precipitation exceeding P threshold at 60h
35
Soverato (Calabria) flood BF vs LEPS
Brute-force approach 51 LAMBO runs on ensemble
EPS TL255L40 (51 members)
LEPS 5 weighted LAMBO runs on ensemble EPS
TL255L40
P gt 50 mm / 24 h
25
54
Probability maps for 24h cumulated precipitation
exceeding threshold at 60h
36
MAP IOP2b 19, 20 - Sep - 1999
37
(No Transcript)
38
(No Transcript)
39
(No Transcript)
40
(No Transcript)
41
MAP IOP 2B, 20-21 September 1999
24 hours observed precipitation from 06 to 06 UTC
42
MAP IOP 2B
ECMWF EPS forecasts 51 members
P gt 20mm/24h
P gt 50mm/24h
98
ECMWF probability maps for 24h cumulated
precipitation exceeding threshold at 66 hours
43
MAP IOP 2B
LEPS 5 LAMBO runs at 20 km on Super-Ensemble
TEPS TL159
P gt 20mm/24h
P gt 50mm/24h
LEPS probability maps for 24h cumulated
precipitation exceeding threshold at 66 hours
44
MAP IOP 2b BF vs LEPS
LEPS 5 weighted LAMBO runs at 20 km on ensemble
EPS TL159
Brute-force approach 51 LAMBO runs on TEPS TL159
(51 members)
P gt 50mm/24h
P gt 50mm/24h
Probability maps for 24h cumulated precipitation
exceeding threshold at 66 hours