Title: Operational forecasting at ECMWF: Science, Components and Products
1Operational forecasting at ECMWF Science,
Components and Products
- Ervin Zsoter
- ECMWF, Meteorological Operations Section
- ervin.zsoter_at_ecmwf.int
With contributions from Renate Hagedorn, David
Richardson, Antonio Garcia Mendez, Gerald van der
Grijn, Lars Isaksen and others
2Outline
- ECMWF as an operational and research centre
- EMOS ECMWF Meteorological Observational System
- Quality control at ECMWF
- Important characteristics of the ECMWFs
operational analysis and forecasting system - ECMWF 4D-VAR data assimilation system
- Model computational characteristics
- Model performance
- Some applications
- Different forecast products
3ECMWF as an organisation
ECMWF is an independent international
organization, supported by 18 member states and 8
co-operating states
Convention establishing ECMWF entered into force
on 1st Nov 1975
Co-operating states
Co-operating organisations
4ECMWF Budget 2006
Main Revenue 2006 Member Statescontributions 27
,460,600 Co-operating Statescontributions
425,100 Other Revenue 1,454,600 Total
29,340,300
Main Expenditure 2006 Staff 12,961,900 Leaving
Allowances Pensions 1,807,500 ComputerExpendit
ure 11,785,900 Buildings 1,858,000 Supplies 927
,000 Total 29,340,300
5Objectives of the centre
- Development of global models and data
assimilation systems for the dynamics,
thermodynamics and composition of the Earths
fluid envelope and interacting parts of the
Earth-system - Preparation and distribution of medium-range
weather forecasts - Scientific and technical research directed
towards improving the quality of these forecasts - Collection and storage of appropriate
meteorological data - Make available research results and data to
Member States - Provision of supercomputer resources to Member
States - Assistance to WMO programmes
- Advanced NWP training
6Principal Goal
- Maintain the current, rapid rate of improvement
of its global, medium-range weather forecasting
products, with particular effort on early
warnings of severe weather events.
- Impressive improvement in the quality of the NWP
- 2-3 days over 15-20 years
7Operational activities at ECMWF
- Observations
- Acquisition / Pre-processing / Quality control /
Bias correction - Data assimilation
- Dynamical fit to observations
- Forecasts
- Product dissemination and archiving
- Verification
- Operational / pre-operational validation
- Data Monitoring
8Data sources for the ECMWF Meteorological
Operational System (EMOS)
Number of observed data assimilated in 24 hours
13th February 2006
9Conventional observations used
BUOY MSL Pressure, Wind-10m
SYNOP/METAR/SHIP
MSL Pressure, 10m-wind, 2m-Rel.Hum.
PILOT/Profilers Wind
TEMP Land - ASAP - Dropsonde
Wind, Temperature, Spec. Humidity
Aircraft Wind, Temperature
10(No Transcript)
11Positive trend in the number of Radiosondes
reaching the upper Startosphere
12Manual obs
Automatic obs
1328 satellite data sources used in the operational
ECMWF analysis
DMSP SSM/I
NOAA AMSUA/B HIRS, AQUA AIRS
SCATTEROMETERS
GEOS
TERRA / AQUA MODIS
OZONE
14Satellite data important
- Satellite measurements are increasingly
important - Global coverage (often only source of
observations over ocean and remote land) - High spatial and temporal resolution
- Decrease in conventional observing networks
(fewer radiosonde stations) - But satellite data are not easy to use
- Satellites do not measure the model variables
(temperature, wind, humidity) - They measure radiances, so
- either use derived products (e.g. cloud motion
and scatterometer winds) - or calculate model radiances and compare with
observations - Recent developments in data assimilation are
designed to improve the use of satellite data - Variational data assimilation can use radiance
data directly - Added model levels in upper stratosphere allow
use of additional satellite data - 4D-Var use observations at appropriate time
- Increased resolution more in agreement with
resolution of measurements
15Example Tropical cyclone Bonnie near Florida
satellite data complement conventional data
L. Isaksen Assimilation of ERS-1 and ERS-2
scatterometer winds in ERA-40 ECMWF ERA-40
proceedings 2002
16Large increase in number of observations used
- Especially number of satellite data increases
- A scientific and technical challenge
17Observations for one 12h 4D-Var cycle
0900-2100UTC 26 March 2006
Screened
Assimilated
Synop 389.000 (0.49) Aircraft 362.000 (0.46) Dribu 20.000 (0.03) Temp 135.000 (0.17) Pilot 108.000 (0.14) AMVs 2.811.000 (3.56) Radiance data 74.825.000 (94.81) Scat 269.000 (0.34) TOTAL 78.918.000 (100.00) Synop 60.000 (1.84) Aircraft 179.000 (5.50) Dribu 5.600 (0.17) Temp 67.000 (2.06) Pilot 48.000 (1.48) AMVs 127.000 (3.90) Radiance data 2.646.000 (81.34) Scat 122.000 (3.75) TOTAL 3.253.000 (100.00)
99 of screened data is from satellites
86 of assimilated data from satellites
18Observations Quality control - Analysis
Data extraction
- Blacklist
- Data skipped due to systematic bad performance
or due to different considerations (e.g. data
being assessed in passive mode) - Departures and flags available for further
assessment
- Check out duplicate reports
- Ship tracks check
- Hydrostatic check
- Thinning
- Skipped data to avoid Over sampling
- Even so departures from FG and ANA are generated
and usage flags also
- 4DVAR QC
- Rejections
- Used data ? increments
ANALYSIS
19Observations Quality control - Analysis
Data input
Data assimilation
- Raw observation
- Departures (FG AN)
- Flags (data used, thinned, rejected)
- Feedback files (BUFR)
- ODB
Monthly BUFR files for different Obs types
Long term statistics
20Data Monitoring (Procedures)
- The basic information is included in the feedback
files or ODB (feedback from the assimilation
scheme) - The statistics are normally computed by comparing
the observations with a FG (6 or 12 hours
forecast) - Model independent statistics should be used also
? Co-locations - But the quality of those forecasts is not the
same everywhere ? no fixed criteria should be
applied when assessing data quality
21Blacklists
- The idea behind the blacklist usage is to remove
from the system observations with a systematic
bad performance. A blacklisted observation is
considered as passive data in the data
assimilation - The blacklist at ECMWF is flexible enough to
consider partial blacklisting depending on - Parameters, areas, atmospheric layers, time
cycles - And of course different observation types.
- MetOps Data Monitoring elaborates a proposal to
update the blacklist which then is discussed with
HMOS and HDA. In cases with heavy changes
sensitivity experiments are carried out before
implementing the new blacklist
22Blacklists
Quality problems in Asia Russia
23Blacklists
Quality problems in Africa and southern Asia
24Whats the benefit of using a blacklist?
All data
Ob-FG
Ob-AN
Used data Blacklist and 4DVAR QC applied
NH
Reduced random deviation
25Whats the benefit of using a blacklist?
All data considered
26Whats the benefit of using a blacklist?
Blacklist applied
27Whats the benefit of using a blacklist?
Blacklist plus 4DVAR Quality Control applied
28Sites directional setting changed by 6.5 degrees
29Example for data monitoring SYNOP pressure bias
correction
Strong biases related to wrong station heights in
the catalogue
30Example for data monitoring SYNOP pressure bias
correction
Bias correction applied
31Pressure bias correction at ECMWF
- Applied to Synop, Ship, Buoy and Metar data when
needed - OI and Kalman filter schemes run in parallel
- OI is used for the corrections although Kalman
filtering can be switch on on request - The scheme is not applied when
- The difference between the station height and the
model orography is larger then 200 hPa - The observation is RDB flagged
- The history of the station is not long enough
32Adaptive bias correction scheme for surface
pressure data
- Time series of Original Ps departure, Ps bias
estimate and Corrected Ps departure for station
82353 (Dec-Apr 05) - Once the sample size (30) was reached (Warm-up
period) bias correction kicked in - Long-term bias of about -3hPa was recognised and
corrected for - Station height is thought to be correct and real
reason for bias is unknown - If not bias corrected the station was just
surviving the First Guess check but to be
rejected by the analysis check - When bias corrected, the station survived all the
checks and was successfully used in the analysis
Warm-up
3hPa Bias
33Example for data monitoring SYNOP pressure bias
correction
34Example for data monitoring SYNOP pressure bias
correction
35Example for data monitoring SYNOP pressure bias
correction
36ECMWFs operational analysis and forecasting
system
The comprehensive earth-system model developed at
ECMWF forms the basis for all the data
assimilation and forecasting activities. All the
main applications required are available through
one integrated computer software system (a set of
computer programs written in Fortran) called the
Integrated Forecast System or IFS
- Numerical scheme
- Spectral model - TL799L91 (799 waves around a
great circle on the globe, 91 hybrid vertical
levels 0-80 km (0.01 hPa)) - Semi-Lagrangian time scheme
- 12 minutes timestep
- Prognostic variables
- wind, temperature, humidity, cloud fraction and
water/ice content, pressure at surface
grid-points, ozone - Grid
- Gaussian grid for physical processes, 25 km,
76,757,590 grid points (843,490 on the surface)
37Spectral and grid point representations
- ECMWF model uses both spectral and grid point
representations - Most upper air model variables (wind,
temperature) are stored as spectral fields - Horizontal derivatives of these variables are
calculated in spectral space - Surface variables and upper air humidity are
stored in grid point space - Dynamical tendencies and physical
parameterizations are calculated in grid point
space - Resolution is the same in physical (grid point)
and spectral space - Grid
- Gaussian grid for physical processes, 25 km,
76,757,590 grid points (843,490 / level) - Gaussian grid is regular in latitude, almost
regular in longitude - On regular grid (same number of points on each
latitude row) points get closer together nearer
the poles - Reduced Gaussian grid keeps distance between
points nearly constant over globe
38Operational model levels
39Operational model grid (reduced Gaussian)
40Model approximations orography and spatial
resolution
High spatial resolution is needed to impose
accurate boundary conditions. For example, the
representation of the orography becomes more
realistic with increased horizontal resolution.
T799 orography, grid spacing 25 km
T255 orography, grid spacing 80 km
41ECMWF model 10m wind (T799, 25 km)
42Katrina (2005 Aug) 90h forecasts - T511 versus
T799
Central pressure 909hPa, 785mm/24h rain
T799
Central pressure 940hPa, 448mm/24h rain
T511
43Hurricane Gordon T799 forecast
AN
30 hrs
78 hrs
126 hrs
44Limitation Model grid box still large
45Physical processes in the ECMWF model
46Data assimilation for weather prediction
The FORECAST is computed on a grid over the
globe. The meteorological OBSERVATIONS can be
taken at any location in the grid. The computer
models prediction of the atmosphere is compared
against the available observations, in near real
time
A short-range forecast provides an estimate of
the atmosphere that is compared with the
observations. The two kinds of information are
combined to form a corrected atmospheric state
the analysis. Corrections are computed and
applied twice per day. This process is called
Data Assimilation.
474D-Var Data assimilation
Analysis values
Background values
Observations
12-hour forecast
Model variables, e.g. temperature
True state of the atmosphere
12 UTC 13 March
00 UTC 13 March
00 UTC 14 March
12 UTC 14 March
Time
48A few 4D-Var Characteristics
All observations within a 12-hour period
(3,300,000) are used simultaneously in one
global (iterative) estimation problem
- Observation minus model differences are computed
at the observation time using the full forecast
model at T799 (25 km) resolution - 4D-Var finds the 12-hour forecast evolution that
optimally fits the available observations. A
linearized forecast model is used in the
minimization process based on the adjoint method
(2 minimisation loops T95/T255) - It does so by adjusting surface pressure, the
upper-air fields of temperature, wind, specific
humidity and ozone - The analysis vector consists of 30,000,000
elements at T255 resolution (80 km)
9z 12z 15z 18z
21z
49ECMWF 4D-Var procedure
- Use all data in a 12-hour window (0900-2100 UTC
for 1200 UTC analysis) - Group observations into ½ hour time slots
- Run the T799 (25km) high resolution forecast from
the previous analysis and compute observation-
model differences - Adjust the model fields at the start of
assimilation window (0900 UTC) so the 12-hour
forecast better fits the observations. This is
an iterative process using a lower resolution
linearized model T255 (80 km) and its adjoint
model - Rerun the T799 high resolution model from the
modified (improved) initial state and calculate
new observation departures - The 3-4 loop in repeated twice to produce a good
high resolution estimate of the atmospheric state
the result is the ECMWF analysis
50Multi-incremental quadratic 4D-Var at ECMWF
T799L91
T95L91 T255L91
51Analysis increments 1st and 2nd minimization
Temperature level 60 (10metre). 0.2K contours
(blue is negative red is positive)
2nd minimization Additional T159 T increment
1st minimization T95 T increments
Most of the increment is formed at the lower
resolution with smaller additions and corrections
obtained at the higher resolution.
52Forecast versus observations
12-hour forecast temperature change
Correction, as a result of data assimilation
The corrections are 10 times smaller than the
12-hour forecast temperature change
53Tropical cyclone LILI - Impact of NSCAT data in
4D-Var
First guess MSL pressure
First guess MSL pressure
Analysis MSL pressure
MSL pressure Analysis increments
NSCAT analysis
No SCAT analysis
S.M. Leidner, L. Isaksen and R.S. Hoffman Impact
of NSCAT Winds on Tropical Cyclones in the ECMWF
4DVAR assimilation system Mon. Wea. Rev.
131,1,3-26 (2003)
544D-Var is using more a-synoptic data than 3D-Var
4D-Var SYNOP Screening
3D-Var is like 4D-Var without the time dimension.
The analysis is performed at synoptic times only
(0000, 0600, 1200 and 1800 UTC). Mostly only data
valid a synoptic time is used. The 12 hour
forecast evolution is NOT an integral part of the
analysis.
4D-Var is using more data from frequently
reporting stations. The plots show the use of
SYNOP surface pressure observations. Column
height gives the number of observations
available, while the shaded part displays those
actually used in the assimilation.
3D-Var SYNOP Screening
554D-Var versus 3D-Var and Optimum Interpolation
- 4D-Var is comparing observations with background
model fields at the correct time - 4D-Var can use observations from frequently
reporting stations - The dynamics and physics of the forecast model in
an integral part of 4D-Var, so observations are
used in a meteorologically more consistent way - 4D-Var combines observations at different times
during the 4D-Var window in a way that reduces
analysis error - 4D-Var propagates information horizontally and
vertically in a meteorologically more consistent
way - More complex needs linearized perturbation
forecast model and its adjoint to solve the cost
function minimization problem efficiently
564D-Var versus 3D-Var performance
N. HEM
12h T511 3DVAR
6h T319 4DVAR
6h T319 3DVAR
S. HEM
57Computational cost of the model
- Higher horizontal resolution
- 31 more vertical levels
- 12 min timestep instead of 15 min (T511)
- Altogether 4 times more floating point operations
are required to complete a 10-day forecast than
with the T511 version - 1.700.000.000.000.000 operations
58Operational schedule for 0000UTC cycleEarly
delivery suite introduced June 2004
59Supercomputer performance at ECMWF
1978-2003 Mflops/s Peak performance
(G.-R. Hoffman)
602004/2006 A significant Performance increase
2006
2002
IBM p690 2 x 960 processors
IBM p575 2 x 2400 processors
Peak performance 7.6 Gflops per processor
Peak performance 5.2 Gflops per processor
Switch 350 Mbytes/s
Switch 2000 Mbytes/s
8 processors per shared memory node
16 processors per shared memory node
(Deborah Salmond/Sami Saarinen )
61T511 1-day Forecast on IBM
CPUs
624D-Var T799/T95/T255 with 91 levels on present
ECMWF IBM system
512
768
1024
1536
63Time series Z500 N Hemisphere against analysis
64Time series Z500 N Hemisphere against
radiosondes
65ECMWF Re-analysis project (ERA)
- Main objective is to promote the use of global
analysis of the state of the atmosphere, land and
surface conditions over the period - ERA-15 1979 1993
- ERA-40 1957 - 2002
- T159L60
- 3DVAR
- ERA interim 1989-
- T255L91
- 12h 4DVAR
66Different application of the ECMWF products
67Link with limited-area ensemble systems
- Over Europe, there are 4 operational Limited-area
EPSs (SRNWP-PEPS, COSMO-LEPS, NORLAMEPS and
PEACE) that produce daily 81 forecasts with
horizontal resolution ranging from 7 to 120 km,
and with forecast length ranging from 30 to 120
hours. 8 further centres (Met Office, INM, DMI,
HMS, MeteoSwiss, SAR, PIED-SE) are developing and
testing LEPSs. Studies have shown that compared
to global EPSs, limited-area EPSs are better able
to predict small-scale, local phenomena. - - Boundary conditions from the global ECMWF model
68Hydrological application - EFAS the European
Flood Alert System
- EFAS is a forecasting tool designed to give
early-warnings for European rivers with
catchments in excess of 2000 km. A
pre-operational prototype is under testing at the
Joint Research Center (JRC, Ispra). The system
uses meteorological inputs from DWD (forecasts up
to 7 days), ECMWF (high-resolution and ensemble
forecasts up to 10 days) and aims to provide
single and probabilistic predictions.
69ECMWF operational system 2006 Forecast Products
- Data assimilation (4D-VAR)
- Four-dimensional variational data assimilation
based on T799 (25 km) / T255 (80 km) / T95
(200 km) horizontal resolution and 91-level
vertical resolution (4 times a day) - Medium-range atmospheric global model
- High resolution deterministic T799 (25 km)
91-level high resolution model for single
deterministic forecast, twice a day up to 10 days - Ensemble T399 (50 km) 62-level model for
50-member ensemble forecasts, twice a day up to
15 days - Coupled ocean wave model (WAM cycle4)
- 2 versions global and regional (European Shelf
Mediterranean) - Numerical scheme irregular lat/lon grid, 40 km
spacing spectrum with 30 frequencies and 24
directions - Coupling wind forcing of waves every 15 minutes,
two way interaction of winds and waves - Extreme sea state forecasts freak waves
- Wave model forecast results can be used as a tool
to diagnose problems in the atmospheric model - Monthly forecast system
- Seasonal forecast system
70Global forecasts (deterministic, fields)
Mean Sea Level Pressure Rain (06-18UTC)
500 hPa height and 850 hPa temperature
Base Fri. 10/03/06 00UTC, Valid Tue 14/03/06 12UTC
71Global forecasts (deterministic, fields)
T2m and 30m-winds
Cloud cover (high, medium, low)
Base Fri. 10/03/06 00UTC, Valid Tue 14/03/06 12UTC
72ECMWF deterministic Ocean wave forecasts
- Global forecast to ten days from 00 and 12 UTC at
50 km resolution
AFRICA !!!!
- European waters forecast to five days from 00 and
12 UTC at 25 km resolution
73The Ensemble Prediction System (EPS)
- A Stochastic Medium-range model (EPS)
- Spans the unstable sub-space of initial
conditions with a Gaussian samples of 50 T42
singular vectors 5 per tropical target (TC) - Runs with stochastic perturbations of physical
tendencies - TL399/L62 range 15 days
- Schedule twice per day
- 00UTC (all products available before 1000UTC)
- 12UTC (all products available before 2200UTC)
- Ensemble Forecasting (Thursday afternoon)
74EPS forecasts time series (EPSgram)
EPSgram for Pretoria Base Friday 27/10/06 00UTC
75EPS forecasts (field probabilities)
Probability of 10m wind speed more than 10 m/s
Probability of precipitation more than 1mm in 24
hours
Base Fri. 10/03/06 00UTC, Valid Tue 14/03/06 12UTC
76EPS forecasts (post-processed products)
Extreme forecast index for 2m temperature Base
Fri. 10/03/06 00UTC, Valid Tue 14/03/06 12UTC
77Katrina forecasts (days from landfall)
4 days before landfall
78Monthly forecasting
- Coupled atmosphere / ocean model
- Atm. T159 (125 km) 62 vertical levels (same
model as oper) - Ocean 29-level, 0.3 equator - 1 mid-latitudes
- 51 member ensemble
- Runs once a week up to 32 days
- Compared to 5 forecasts for same day over last 12
years - 60 member ensemble
- Results interpreted in terms of anomalies and
probabilities - For example probability that 2m temperature
averaged over day 12 to 18 is in the
upper/middle/lower tercile - Products become available every Thursday at 22UTC
79Monthly forecast
Probability that 2m temperature is in the upper
tercile (third) of the climate distribution -
warmer than normal Base Thu 26-10-2006. Valid
days 5-11 (30-10 to 0511)
80Monthly forecast
Probability that 2m temperature is in the upper
tercile (third) of the climate distribution -
warmer than normal Base Thu 19-10-2006. Valid
days 12-18 (30-10 to 0511)
81Monthly forecast
Probability that 2m temperature is in the upper
tercile (third) of the climate distribution -
warmer than normal Base Thu 12-10-2006. Valid
days 19-25 (30-10 to 0511)
82Monthly forecast
Probability that precipitation is in the upper
tercile (third) of the climate distribution
more wet than normal Base Thu 26-10-2006. Valid
days 5-11 (30-10 to 0511)
83Monthly forecast
Probability that precipitation is in the upper
tercile (third) of the climate distribution
more wet than normal Base Thu 19-10-2006. Valid
days 12-18 (30-10 to 0511)
84Monthly forecast
Probability that precipitation is in the upper
tercile (third) of the climate distribution
more wet than normal Base Thu 12-10-2006. Valid
days 19-25 (30-10 to 0511)
85Monthly forecast performance over the Northern
Extratropics
ROC area of probability of 2-metre temperature in
upper third of climate range
Monthly Forecast
Monthly Forecast
Persistence of day 5-11
Persistence of day 5-18
Day 12-18
Day 19-32
86ECMWF seasonal forecast System 2
- 6-month ensemble produced each month with coupled
atmosphere-ocean forecast system - ECMWF atmospheric model (same cycle as used for
ERA) 200 km (T95), 40 levels - HOPE ocean model 29 levels 1x1 mid-latitudes,
increased latitudinal resolution to 0.3 at
equator - 40 ensemble members 5 ocean analyses, 40 SST
perturbations stochastic physics - Products issued as anomalies relative to model
climatology (15 years of ensemble re-forecasts) - Initial date 1st of each month forecasts issued
15th of month
87ECMWF seasonal forecast System 2
Anomaly of 2m temperature Base 01-10-2006. Valid
December-January-February (DJF)
88ECMWF seasonal forecast System 2
Anomaly of precipitation Base 01-10-2006. Valid
December-January-February (DJF)
89ECMWF seasonal forecast System 2
90Seasonal forecasting EUROSIP multi-model ensemble
- Three models running at ECMWF
- ECMWF System 2
- Met Office HADCM3 model, Met Office ocean
analyses - Meteo-France Arpege/Climat, Mercator ocean
analyses - Spain Germany may join
- Unified system
- All data in ECMWF operational archive
- Common operational schedule (products released at
12UTC on the 15th of each month) - Common products being developed
- EUROSIP appears to be better than the individual
systems
91EUROSHIP seasonal forecast
EUROSHIP anomaly of 2m temperature Base
01-10-2006. Valid December-January-February (DJF)
92EUROSHIP seasonal forecast
EUROSHIP anomaly of precipitation Base
01-10-2006. Valid December-January-February (DJF)
93ECMWF Products for WMO members
- A new web page has been created as a single entry
point for all services to WMO members - www.ecmwf.int/about/wmo_nmhs_access/index.html
- Council enhanced in July 2006 the product set
available to all WMO members - For several parameters, an extension of the
forecast range from day 7 to day 10 - Global products from the EPS in support of high
impact weather - Site-specific forecasts at selected locations,
targeting synoptic stations in developing
countries, especially the least developed ones
94ECMWF Products for GTS dissemination
Param level steps
Z 500 G 0,24,48,72,96,120,144,168, 192, 216, 240
T 850 G 0,24,48,72,96,120,144,168
u,v 850, 700, 500, 200 G 0,24,48,72,96,120,144,168
Rh 850, 700 G 0,24,48,72,96,120,144,168
MSLP G 0,24,48,72,96,120,144,168
Div 700 T 0,24,48,72,96,120,144
Vort 700 T 0,24,48,72,96,120,144
95ECMWF Products for ACMAD
Param level steps
u,v 925 0,24,48,72,96,120,144,168
div 925, 200 0,24,48,72,96,120,144,168
2m T 0,06,12,24,30,36,42,48,54,60,66, 72,78,84,90,96,102,108,114,120
10m u,v 0,06,12,24,30,36,42,48,54,60,66, 72,78,84,90,96,102,108,114,120
precip 0,06,12,24,30,36,42,48,54,60,66, 72,78,84,90,96,102,108,114,120
96ECMWF Web site (www.ecmwf.int)