Resent upgrades to ECMWFs forecasting system Erik Andersson

1
Resent upgrades to ECMWFs forecasting
systemErik Andersson ColleaguesData
Assimilation Section,ECMWF, Reading, U.K.New
forecast productsEarly DeliveryRecent
performanceOne or two current research
topicsPlans for 2004/2005
2
ECMWF forecast products Recent additions

• 3-hourly processing (T511 0-gt72h) introduced on
  24 October 2000
• New parameters - Cloud liquid water and ice
  contents (13 February 2001)
• Potential Vorticity and related parameters (22
  March 2001)
• Wind gust at 10m and surface downwards radiation
  fluxes (19 Sept 2001)
• Multi-analysis products (12 Oct 2001)
• Potential Vorticity on pressure levels (26 March
  2002)
• Internet Dissemination (5 Dec. 2002)
• 00 UTC model available including EPS (25 March
  2003)
• CAPE, Freak waves, UV-B, photo-synthetically
  active radiation (7 October 2003)
• Extreme Forecast Index (EFI) (20 October 2003)
• SST seasonal forecasts

3
Extreme Forecast Index (EFI)
4
Evolution of the number of products

• Daily total
• Total reaches 1.1 M per day 11.4 GB per day
• Was 86k in 1995
• Was 300k in 2000
• Internet distribution81 411 per day4.4 GB per
  day

5
Recent extensions to Web products

• GTS products (incl. EPS probabilities)
  http//www.ecmwf.int/products/forecasts/d/charts/d
  eterministic/world/ http//www.ecmwf.int/products/
  forecasts/d/charts/eps/probabilities/world/
• Data monitoring (Monthly Monitoring Report, GUAN,
  satellites) http//www.ecmwf.int/products/forecast
  s/monitoring/mmr http//www.ecmwf.int/products/for
  ecasts/d/charts/monitoring/guan
  http//www.ecmwf.int/products/forecasts/d/charts/m
  onitoring/radiances/
• Documentation (Tech. Memo 373 on the EFI, 388,
  389 and 393 on EPS verification Research Dept.
  Papers) http//www.ecmwf.int/publications/library/
  ecpublications/techmemos/tm00.html
• Users Meeting presentations http//www.ecmwf.int/
  publications/member_states_meetings

6
European Time Zones (summer)
7
The current operational schedule for the 00Z cycle
8
The Early Delivery System
9
The early delivery schedule for the 00Z cycle
10
Anomaly Correlation of 500 hPa Height, NH and SH
11
AIREP U Wind Departure Statistics
Standard deviation and bias for the departures of
used AIREP u-winds in the Northern Hemisphere
extra-tropics from the background (solid) and the
analysis (dotted), accumulated over the period
2004/02/01 00Z 2004/02/20 12Z. The top diagram
compares the shifted window 12 hour 4D-Var
(black) and the 28r1 e-suite(red). The bottom
diagram compares the early delivery 6 hour 4D-Var
with the 12 hour 4D-Var e-suite.
12
Cut-off Time
Anomaly correlation of 500 hPA geopotential
height forecast error for the Northern Hemisphere
extra tropics (top) and Southern Hemisphere
extra tropics (bottom) for the forecasts from 00Z
early delivery analyses with an observation
cut-off time of 04Z red
06Z blue 08Z
green and 26r3 esuite brown averaged over 23
days from 2003/09/01 00Z
13
TEMP Data Coverage
14
PILOT Data Coverage
15
AIREP Data Coverage
16
SYNOP/SHIP Data Coverage
17
DRIBU Data Coverage
18
ATOVS Data Coverage
19
Eumetsat Atovs Retransmission Service
20
Winter 2003/04 vs other centres (Europe)
21
WMO/CBS scores (Europe, wind vs RS)
22
Summer 2003 vs previous summers
23
2m-temperature verification (T511)
24
Daily rainfallgt10mm ETS scores

• t 42

• t 66

• t 90

• t 114

25
Data assimilationaccuracy Verification of
short-range FC against near-surface winds over
ocean.
26
ECMWF Snow Analysis Revision (uses NESDIS snow
product)
NWS National Operational Hydrologic Remote
Sensing Center
27
Polar WV winds from MODIS
Source P. Menzel, 2003
28
Forecast imagery versus observed
Model T and q
Observation Satellite Radiance
Model Radiance
H
compare
29
Number of observational data used in the ECMWF
assimilation system (with AIRS)
3.6 millions
millions
30
(No Transcript)
31
IBM High Performance Computer (HPC)IBM
performance relative to VPP systems
5
4
Phase 3 Regatta H 3000 processors Federation
Switch
3
Phase 2
2
Phase 1
1
Fujitsu
2002
2003
2004
2006
2005
Performance on ECMWF codes (Fujitsu 400 GF
sustained)
32
The ensemble system

• Weekly running of the monthly forecast
• Develop a seamless ensemble forecast system
  across a range of timescale, e.g.
• T399(50 km) to D7 twice daily
• continue once a day at T255(80 km) to D15
• continue twice a week at T159(123 km) to D30, in
  coupled mode
• Moist singular vector
• Ensemble data assimilation

33
Impact of resolution increase in the EPS
Increasing the resolution from TL255 to TL399
would reduce the gap between the TL511 and EPS
control forecasts, and improve EPS performance
(34 cases, cy23r4 and 25r3).
34
Impact of a resolution increase on precipitation

• A resolution increase from TL255 to TL399 would
  have a large impact on the quality of
  12h-accumulated precipitation 12h
  predictability gain for 5 and 20 mm thresholds on
  average over NH (results based on 34 cases,
  cycles 23r4 and 25r5).

35
ECMWF participation in the TOST

• Deployment of targeted observations requires
  decision about 30 hours in advance
• Impact of targeted observations is to remain for
  48-72 hours. Targeting thus requires reliable
  100-hour forecasts
• ECMWF provides a set of products to the TOST
• Pre-defined SV-based information for various
  areas and lead times
• Evaluation on demand of the information content
  of various observations scenarios
• ECMWF displays on the web and archives products
  of the TOST partners

36
Sensitive area computations for targetting
Sensitivity area computed with vertical
integration of the moist total energy
37
Candidates for upgrades in summer/autumn 2004

• Early Delivery System (1st week July)
• Use of latest release of RTTOV (RTTOV-8)
• Use good radiosonde humidity above 300 hPa
• Monitoring and then assimilation of MSG and SSMIS
  radiance data. Monitoring of ground-based GPS
  data.
• New PBL scheme including boundary layer clouds
• Revisions to cloud scheme
• Seasonal LAI and monthly MODIS surface albedo
  climatologies

38
Current plans for end 2004 and 2005

• Regional variation of background error
  covariances (wavelet-Jb) in data assimilation
• Bias correction for SYNOP and METAR surface
  pressure data
• Higher vertical resolution
• L91 for 4D-Var and deterministic
  forecast
• L65 for EPS, seasonal and monthly
  forecasts
• Assimilation of precipitation data and then cloud
  data
• T799 horizontal resolution in deterministic
  forecast and outer loops of 4D-Var with T255
  inner loops (T319 for SVs?)
• Preparation for METOP (European polar orbiter)

39
TEMP U Wind Departure Statistics
Standard deviation and bias for the departures of
used TEMP u-winds in the Northern Hemisphere
extra-tropics from the background (solid) and the
analysis (dotted), accumulated over the period
2004/02/01 00Z 2004/02/20 12Z. The top diagram
compares the shifted window 12 hour 4D-Var
(black) and the 28r1 e-suite(red). The bottom
diagram compares the early delivery 6 hour 4D-Var
with the 12 hour 4D-Var e-suite.
40
BUFR2ODB (April 2003)

• Significant technical changes to observation
  handling
• Allowing more satellite data to be ingested
• Multi-level parallelism
• Facilitating diagnostics by giving easier access
  to sub-ensembles of data
• A major effort on ODB documentation is currently
  done in collaboration with Meteo-France and the
  Hungarian Met. Service

41
Total column ozone Ozone hole
16 Sept. 2003
O-suite

E-suite (26r3)
TOMS
42
Forecast imagery

• Issues
• Underestimation of cloud ice
• Poor description of convection diurnal cycle over
  land
• Horizontal resolution
• Future
• MSG
• Geostationary belt
• Update of cloud optical properties

43
Forecasted imagery IR

• 42-hour forecast vs. observed

Different grey scales
44
Forecasted imagery WV

• 42-hour forecast vs. observed

Different grey scales
45
The observation operator

• Observations are not made at model grid points
• Satellites often measure radiances, NOT
  temperature and humidity
• We calculate a model radiance estimate of the
  observation to enable comparison.
• This is done with the observation operator H.
• H may be a simple interpolation from model grid
  to observation location
• H may possibly perform additional complex
  transformations of model variables to radiance
  space for satellite data

Model T,u,v,q
Model radiance
Observation Satellite Radiance
compare
H
46
The ECMWF forecast model

• The model has more than 20 million grid points
• Spacing of grid points 40 km
• 60 levels from surface to 65 km
• Temperature, wind, humidity and ozone are
  specified at each point.
• This is done in 15 minute time-steps.
• Number of computations required to make a ten-day
  forecast 188,000,000,000,000

47
Numerical Aspects

• Developments in the semi-Lagrangian scheme
• Numerical instabilities still arise by
  interaction between SL and the momentum equation
• Conservation needs to be improved
• More general and efficient treatment of the
  generic grid-point fields
• Preparations for horizontal and vertical
  resolution increase
• Work on the upper boundary condition

48
Noise in stratospheric forecasts
H12 Z10 from 28-12-2002 at 12z
Level 11 D
49
SETTLS with LLSI at both departure and arrivalin
the vertical trajectory computation
H12 Z10 from 28-12-2002 at 12z
Level 11 D
50
Physical Aspects

• Diurnal cycle of convection
• New PBL scheme including clouds
• Progress in TL and AD physics
• Stochastics Physics
• New orographic drag scheme
• Evaluation of the McICA approach for cloud
  treatment in radiation
• New assimilation method for soil moisture (ELDAS)

51
Diurnal Cycle of Convection(1a) Hovmoeller
Diagrams for February 1999 from 40-day forecasts
at T159 (25R125R4/26R1), and comparison with
TRMM (available for 2002)
25R1
Observations show maximum rainfall intensity for
Africa and South America at 15 LST. Cycle 25R1
produces maximum precip at 9 LST, compared to 12
LST with the new cycle.
25R4
52
Has the N. America problem gone away? Impact of
revising the triggering of convection (courtesy
P. Bechtold)
53
Predictability and Diagnostics

• Future configurations of the EPS
• Moist singular vectors
• Observation targeting (Atlantic THORPEX
  Observation System Test, TOST) in Oct/Nov 2003.
• New diagnostics of forecast system performance
  (evolution along trajectories, PV scores, etc..)

54
Seasonal Forecasting

• El Niño 2002-2003 prediction
• DEMETER results demonstrate the benefits of a
  multi-model approach. A operational multi-model
  system with UKMO will soon exist, and will be
  joined by Meteo-France in 2004
• OPA currently evaluated, OPA-VAR soon available
  (cooperation with CERFACS)
• Comparison of ocean assimilations within the EU
  project ENACT
• Work has started to couple the wave model to the
  ocean model

55

• January 2002 Seasonal Forecast System 2
• The Coupled model
• Atmosphere (IFS-23r4) Ocean (HOPE 1x1)
  
• Ocean Analysis
• OI of subsurface T
• 5-member ensemble of ocean analysis
• Ensemble BURST MODE
• 40 forecasts on 1st of month to 6 months
• The Calibration Period
• ERA15 1987-2001

56
(No Transcript)
57
ECMWF forecasts (CLIVAR)
58
DEMETER MULTI MODEL (Not real time, but 7
coupled models, 9 member ensembles based on 3
ocean analyses.)

• ECMWF has pioneered the multi-model ensemble
  approach, which accounts for the two main sources
  of error, model formulation and initial
  conditions.
• The problem of model error is addressed by
  applying different models and hence increasing
  the reliability of the probability forecast.
• The error in initial conditions is tackled by
  running these models from sets of initial
  conditions.

59
ensemble size versus multimodel
From DEMETER
60
Monthly Forecasting System

• Coupled ocean-atmosphere integrations a
  51-member ensemble is integrated for 32 days
  every 2 weeks.
• Atmospheric component IFS with the latest
  operational cycle and with a TL159L40 resolution.
  Initial conditions ERA40 or operational
  analysis. Perturbations singular vectors
  stochastic physics.
  
• Oceanic component HOPE (from Max Plank
  Institute) with a zonal resolution of 1.4 degrees
  and 29 vertical levels. Same perturbations as in
  seasonal forecasting.

Verification - The monthly forecasting system
is running in real-time since 27 March 2002 - 30
cases have been verified.
61
Last summer heat wave 2m-tm anomaly3-9 August
2003
VERIFICATION
Forecast 30/07 DAYS 5-11
Forecast 25/07 DAYS 12-18
Forecast 16/07DAYS 19-25
62
Ocean Waves

• Monitoring of Envisat RA-2 has revealed a 4 bias
• Assimilation of RA-2 introduced in 26r3 but not
  activated
• Effect of unresolved bathymetry
• Towards prediction of freak waves Benjamin-Feir
  index is now a regular output of the model,
  evaluation has started. Collaboration in progress
  with group in Trondheim

63
Benjamin-Feir Index
64
ERA-40

• Production of analyses was completed in April
  2003
• There us already much use of data, in Europe,
  via MARS and national data centres
• 2.5o data are available on a public data server
  (http//data.ecmwf.int/data)
• Data will be supplied to NCAR for UCAR and other
  US use
• Observations have been supplied to JMA for JRA-25

65
Conclusions from ERA-40 General synoptic
quality of analyses

• Best for most recent years
• Quite good throughout for northern hemisphere
  troposphere and lower stratosphere
• Poor in southern hemisphere in early years
• Some improvement in early 1970s
• Big improvement in 1979

66
Trends and inter-annual variability

• Global temperature trends reasonably well
  captured from surface to lower stratosphere
• Caution needed when looking at regional trends
• Clear improvement on ERA-15

67
Hydrology over tropical oceans

• Excessive tropical rainfall in satellite era
• increases as more data are assimilated
• compounded by misinterpreting effects of Pinatubo
  aerosols
• Water vapour
• model biased dry compared with IR and MW data
• analyses biased moist compared with SSM/I
  retrievals
• Problems are
• model and data biases
• inhomogeneity of data coverage
• how to distribute analysis increments in
  horizontal and vertical

68
Stratosphere

• Several problems with temperature biases
• due to model biases and difficulties in radiance
  assimilation
• worse in ERA-40 3D-Var than in operational 4D-Var
• Too-strong Brewer-Dobson circulation
• seen in humidity, and ozone when no data is
  assimilated
• worse when model temperature biases are corrected
  by radiance assimilation
• QBO and SAO handled well

69
Plans beyond ERA-40

• An interim reanalysis
• Start next year, T159L60, latest version of
  forecasting system
• Run from 1991 (tbd) onwards, continued in close
  to real time
• Baseline for ongoing developments (e.g.
  constituent analysis)
• Experimentation
• Observing system experiments
• High-resolution 4D-Var analyses for specific
  cases
• To validate new versions of forecasting system
  
• Development of observational data base and
  processing software
• An extensive new reanalysis in 2008 or beyond