Slide 1
ECMWF report - WGNE 2011 Slide 1, ©ECMWF
ECMWF Report 2010-2011
Jean-Noël Thépaut ECMWF
October 2011
and many colleagues from the Research Department
Slide 2
Forecasting system upgrades 2010-11 Cycle 36R4 – 11 November 2010
5-species prognostic cloud scheme
SEKF soil moisture and OI snow analysis Convection entrainment, all-sky assimilation refinement, 2 outer loops in
early-delivery 4D-Var, etc.
Cycle 37R2 – 18 May 2011 Reduced AMSU-A radiance observation errors
Use of EDA variances in 4D-Var background error formulation
Cloud scheme modification, GPSRO obs operator improvement, etc.
Cycle 37R3 – November 2011 Aircraft temperature bias correction
Combined UV-IR ozone observation assimilation Cloud scheme & surface roughness modification, assimilation of Stage-IV,
stratospheric model error cycling , convection detrainment, etc.
Outlook ECMWF report - WGNE 2011 Slide 2, ©ECMWF
November 2010 IFS cycle 36r4
Selected contents"
• Prognostic rain and snow with more comprehensive cloud microphysics!
• EKF for soil moisture analysis!
• New snow analysis (O-I)!
• Enhancement of all-sky radiance assimilation!
Slide 4
New prognostic cloud microphysics scheme
WATER VAPOUR
CLOUD Liquid/Ice
PRECIP Rain/Snow
Evaporation CLOUD FRACTION CLOUD
FRACTION
Old Cloud Scheme New Cloud Scheme
• 2 prognostic cloud variables + w.v. • Ice/water diagnostic Fn(T)
• Diagnostic precipitation
• 5 prognostic cloud variables + water vapour • Ice and water now independent
• More physically based, greater realism • Significant change to degrees of freedom
• Change to water cycle balances in the model • More than double the lines of “cloud” code!
Slide 4, ©ECMWF ECMWF report - WGNE 2011
Slide 5
New prognostic cloud microphysics Representation of mixed phase
• One of the most significant changes in the new scheme is the physical representation of the mixed phase.
• Old scheme: Single prognostic condensate and diagnostic fn(T) split between ice and liquid cloud
• New scheme: Prognostic liquid and prognostic ice and any fraction of supercooled liquid water allowed for a given T (determined by sources+sinks)
PDF of liquid water fraction of cloud for
the diagnostic mixed phase scheme
(dashed line) and the prognostic ice/
liquid scheme (shading)
Slide 5, ©ECMWF ECMWF report - WGNE 2011
Slide 6
Tem
pera
ture
Ice Water Content (g m-3)
-80
-60
-40
-20
0 106 105 104 103 102 101 100
-80
-60
-40
-20
0
-80
-60
-40
-20
0 106 105 104 103 102 101 100 106 105 104 103 102 101 100
Ice Water Content (g m-3) Ice Water Content (g m-3)
CloudSat/CALIPSO observations
ECMWF old scheme without snow
ECMWF new scheme with snow
New scheme with prognostic ice and snow allows much higher ice water contents (seen by the radiation scheme)
Relative frequency of occurrence of ice/snow for NH mid-latitudes in June 2006: ECMWF model vs. Cloudsat/Calipso retrievals
CY36R4 highlights: New cloud scheme
Slide 6, ©ECMWF ECMWF report - WGNE 2011
Slide 7
Ice water content zonal cross-sections (1 year average)
• Previous scheme had unrealistic peaks of ice water content in 0 to -23°C temperature range.
• New scheme reduces cloud ice water content in closer agreement with CloudSat derived estimate.
• Careful comparison required at lower altitudes due to limitations of CloudSat observations.
Previous scheme
New scheme
mg m-3
CloudSat (non-convective, non-precipitating derived ice water)
Slide 7, ©ECMWF ECMWF report - WGNE 2011
Slide 8
Mod
el le
vel Cy37r3 super-cooled liquid (red) and ice (green)
52°S 63°S
Cy37r2 super-cooled liquid (red) and ice (green)
Mod
el le
vel
Southern Ocean SLW layer from CALIPSO
Ice
SLW
50S 60S
(4) Super-cooled liquid water Super-cooled liquid water
• Super-cooled liquid water (SLW) cloud frequently occurs in atmosphere down to -30°C and below (as seen in aircraft obs, lidar etc.)
• Fine balance between turbulent production of water droplets, nucleation of ice, deposition growth and fallout.
• New cloud scheme represents microphysical processes in mixed-phase cloud rather than a diagnostic.
• Cy36r4/Cy37r2 had less SLW, Cy37r3 increases SLW, particularly at cloud top (as often observed).
SLW layer (blue)
Slide 8, ©ECMWF ECMWF report - WGNE 2011
Slide 9
Land surface data assimilation evolution
1999 2004 2009 2010/2011
Optimum Interpolation (OI)
screen level analysis
Douville et al. (2000) Mahfouf et al. (2000)
Soil moisture analysis based on Temperature and relative
humidity analysis
Revised snow analysis
Drusch et al. (2004) Cressman snow depth
analysis using SYNOP data improved by using NOAA / NSEDIS Snow cover extend
data
Recent developments/implementations: • SEKF (Simplified Extended Kalman Filter) surface analysis • Use of active microwave data: ASCAT soil moisture product • Use of passive microwave SMOS Brightness Temperature product • New snow analysis and use of NOAA/NESDIS 4km snow cover product
Structure Surface Analysis
OI snow analysis and high resolution NESDIS data (4km)
SEKF Soil Moisture analysis Simplified Extended Kalman Filter
METOP-ASCAT SMOS
Slide 9, ©ECMWF ECMWF report - WGNE 2011
Slide 10
ECMWF report - WGNE 2011
CY36R4 highlights: SEKF soil moisture analysis
ECMWF land surface analysis: Snow depth SYNOP
Snow cover NESDIS (AVHRR, SSM/I, sta?ons)
2 m T SYNOP
2 m RH SYNOP
Soil moisture SYNOP (T2m, RH2m) → ASCAT → SMOS → SMAP
Soil temperature SYNOP (T2m)
Land and atmospheric analyses are performed separately indirect coupling
OI: computa?onally cheap
simple error covariance formula?on
only based on screen-‐level observa?ons, fixed rela?onship
EKF: computa?onally expensive (currently w/ finite differences for Jacobians)
physically based, i.e. uses land surface model
open for using more diverse observa?ons (at correct ?me)
(SEKF: B is sta?c) Slide 10, ©ECMWF
Slide 11
EKF soil moisture analysis
- Dynamical estimates of the Jacobian Matrix that quantify accurately the
physical relationship between observations and soil moisture
- Flexible to account for the land surface model H-TESSEL evolution
- Makes it possible to combine different sources of information
- Possible to investigate the use of new generation of satellite data:
- Active microwave (C-Band MetOp/ASCAT)
- Passive microwave (L-band SMOS, SMAP)
SYNOP ASCAT SMOS
Slide 11, ©ECMWF ECMWF report - WGNE 2011
Slide 12
Impact on 2-meter Temperature Forecasts
Global mean RMS (against SYNOP)
T2m 48h FC errors (OI-SEKF) EKF improves T2m
- EKF consistently improves SM & T2m - Makes it possible to assimilate satellite
data to analyse soil moisture Impr
oved
de
grad
ed
Slide 12, ©ECMWF ECMWF report - WGNE 2011
Slide 13
CY36R4 highlights: SEKF soil moisture analysis
36R4 – 11/2010
as 32R3 – 11/2007 as ERA-Interim
colder than ERA-I analysis warmer than ERA-I analysis
Mean annual 2-metre temperature errors from 13-month model integrations
Slide 13, ©ECMWF ECMWF report - WGNE 2011
Slide 14
A new snow analysis Snow analysis uses SYNOP snow depth data and
NOAA/NESDIS IMS snow cover
2010 implementation: - New Snow analysis based on the Optimum
Interpolation with Brasnett 1999 structure functions
- A new IMS 4km snow cover product to replace the 24km product
- Improved QC (monitoring, Blacklisting)
2011: - Assimilate additional snow data From Sweden (New Report Type)
Slide 14, ©ECMWF ECMWF report - WGNE 2011
Slide 15
ECMWF report - WGNE 2011 Slide 15
Upgrades to all-sky assimilation at 36r4
All-sky assimilation of microwave imager radiances over ocean (SSMIS, TMI, AMSRE): Improves analysed water vapour, cloud and precipitation.
Direct 4D-Var assimilation was implemented in March 2009 in cycle 35r2, replacing the previous 1D+4D-Var approach.
Initially, large observation errors were applied, giving relatively weak observational control over oceanic water vapour.
36r4 upgrade substantially increases weight of microwave-imager observations, due to: Decreased observation errors, with errors now dependent on the
mean cloud amount in observations and model – a “symmetric error” approach.
More observations getting through an improved quality control that can distinguish real information (e.g. missing cloud) from instrument problems.
Slide 16
ECMWF report - WGNE 2011 Slide 16
Forecast fits to assimilated AMSU-B: Standard deviation of FG departures
Upper-tropospheric humidity
Mid-tropospheric humidity
Lower-tropospheric humidity
36r4 36r2 No all-sky Experimental: even smaller obs errors
Slide 17
Normalized anomaly correlation difference - z500 CY36R4 vs CY36R2
SEEPS: impact of CY36R4
Slide 17, ©ECMWF ECMWF report - WGNE 2011
May 2011 IFS cycle 37R2
Selected contents"
• Increased weight to AMSU-A data!
• Direct use of EDA in 4D-Var!
• Retuning of new physics!
• GRIB-2 for model level fields!
Slide 19
CY37R2 highlights: AMSU-A data
Comprehensive analysis of observation error covariances (spatial, spectral) for IR and MW sounders:
evaluate current data thinning, Different estimates of σo
• produce covariances for 4D-Var.
Spatial error correlation AMSU-A ch. 9
Slide 19, ©ECMWF ECMWF report - WGNE 2011
Slide 20
CY37R2 highlights: EDA variances for 4D-‐Var
Analysis Xib(tk)
y+εio
Boundary pert. i & SPPT
Xia(tk) Forecast Xib(tk+1)
ε1m
InBalize EPS Produce σb for B in 4D-‐Var
10-‐member 4D-‐Var ensemble (T399 w/ T95/T159):
Std. dev. of vor?city ML64 (~ 500 hPa): old new
Slide 20, ©ECMWF ECMWF report - WGNE 2011
Similar strategy to that of Meteo-‐France (Raynaud, Berre and Desroziers, 2008)
Slide 21
Slide 21
CY37R2 highlights: EDA variances for 4D-‐Var
Tropical cyclone Aere
MSLP:
MSLP variances:
MSLP increments:
old New ECMWF report - WGNE 2011
Slide 22
EDA – flow dependent co-variances (based on 20 members EDA)
Background error correlaBon length scale and pmsl
km
Slide 22, ©ECMWF ECMWF report - WGNE 2011
Slide 23
Normalized anomaly correla?on difference -‐ z500 CY36R4 vs CY36R2 CY37R2 vs CY36R4
Slide 23, ©ECMWF ECMWF report - WGNE 2011
Slide 24
Overall performance: CY37R2-‐CY36R4
Europe
N. Hem.
Tropics
S. Hem.
Generally very posiBve – mostly range day 1-‐5 (AMSU-‐A, EDA)
IFS cycle 37R3 (November 2011)
Selected contents"
Bias correction of aircraft temperature observations
Activation of stratospheric model error cycling (weak constraint 4DVAR)
Improved ozone by activation of combined IR radiance and derived TCO analysis
Entrainment/detrainment of convection
Supersaturation and deposition rate for clouds Active assimilation of accumulated rainfall from NEXRAD"
Slide 26
Aircraft temperature bias correction Based on the variational bias correction scheme developed at ECMWF
Each aircraft is bias corrected individually using a constant predictor
First predictor: constant temperature correction. Second predictor: function of the vertical aircraft velocity (dp/dt) to account for ascend/descend bias conditions
aircraft R/S R/S
Slide 26, ©ECMWF ECMWF report - WGNE 2011
Slide 27
ECMWF report - WGNE 2011 Slide 27, ©ECMWF
Slide 28
ECMWF report - WGNE 2011 Slide 28, ©ECMWF
Slide 29
Mixed-Phase Cloud Motivation for recent work (input for CY37R3)
• Cold temperature bias reported for IFS over Scandinavia in early January. Investigation showed correlation with regions of low-level mixed-phase cloud (cloud with super-cooled liquid water and ice)
• Previous cloud scheme (36R3 and before) had liquid everywhere (by definition) between 0 and -23°C.
• New cloud scheme (36R4 and 37R1/37R2) has similar liquid water, except in weakly forced situations (where sink due to deposition greater than source due to condensation).
Slide 29, ©ECMWF ECMWF report - WGNE 2011
Slide 30
CY36R4 CY37R3
T2m 60h f/
c error
TCLW 60h f/
c
CY37R3 highlights: Revised cloud scheme
Sodankyla
Slide 31
Ceilometer observations
Sodankyla/Finland:
<CY36R4 CY36R4 CY37R3
T2m 60h f/c error
clw/ciw 60h f/c
CY37R3 highlights: Revised cloud scheme
Slide 32
CY37R3 highlights: O3 analysis
• Combined assimilaBon of ozone-‐sensiBve observaBons from UV (SBUV, SCIAMACHY, OMI) and IR (HIRS, AIRS, IASI):
→ more consistent ozone analysis, parBcularly in absence of daylight
AN – MLS (SBUV) AN – MLS (SBUV+IR)
(12/07-11/08/2011)
Slide 33
Normalized anomaly correlation difference - z500 CY36R4 vs CY36R2 CY37R2 vs CY36R4 CY37R3 vs CY37R2
Slide 33, ©ECMWF ECMWF report - WGNE 2011
Slide 34
Outlook - 2012 Cycle 38 – October 2011:
Common cycle with Météo-France
Seasonal Forecasting System 4 – November 2011 Higher resolution, updated model cycle, more members New NEMO ocean model
Larger hindcast data (15 members for 30 years)
Cycle 38R1 – spring 2012: Revised L91 background error covariance statistics
Aliasing noise removal in spectral transforms
Lagged 10+10 member EDA EDA mean instead of control as reference for EPS initial perturbations
Cycle 38R2 – autumn 2012: Vertical resolution upgrade L91 → L137 for high-resolution forecast
model and DA-system. ECMWF report - WGNE 2011 Slide 34, ©ECMWF
Slide 35
S4 shows higher skill then S3 over Europe, a very difficult area, as indicated by 2mT reliability for m2-‐4 hindcasts for JJA (30 years, 1981-‐2010). Note that S3 has 11*T159L40 members, and S4 has 15*T255L91 members.
New seasonal system-4 (S4, Nov 2011): reliability over EU
S3 S4
Slide 36
Vertical resolution increase from 91 to 137 levels (1)
• Still a lot of work to do, in particular background error covariances generation • Implementation: Early 2012 Slide 36, ©ECMWF ECMWF report - WGNE 2011
Slide 37
Outlook – continued More hybrid EDA/long-window weak constraint 4D-Var
Require fundamental algorithmic changes
Seamless EDA/EPS-monthly Enhanced cloud analysis Numerical experimentation into the “grey zone”
Fast Legendre transforms at T3999
Physics-dynamics coupling for NH
Increased horizontal resolution T2047 by 2015
IFS maintenance and optimisation Continuous Observation Preprocessing
Object Oriented Prediction System Modularity, flexibility for new algorithmic developments
ECMWF report - WGNE 2011 Slide 37, ©ECMWF
Slide 38
Long-window, weak-constraint 4D-Var Results based on a two-layer quasi-geostrophic model indicates that increasing the length of the analysis window is beneficial, despite the lack of realism in the model error representation.
Slide 39
Cloud assimilaBon increment (500hPa) from a single cloud
observaBon (red circle) in a very dry area. NoBce the asymmetry of the increments caused by the flow-‐dependent background error
variances.
Cloud assimilaBon increment (500hPa) from a single cloud
observaBon (red circle) in a nearly saturated area (Colour=rh_bg,
black=qc_incrE6, white=q_incrE6). The q increments come from a balance relaBonship with qc.
Cloud analysis: Add new control variable: cloud condensate
Slide 39, ©ECMWF ECMWF report - WGNE 2011
Slide 40
Reanalysis Highlights ERA-interim has been
extended back to 1979
ERA-CLIM has been kicked-off in January 2011 (3 year project funded under EU FP-7)
ECMWF report - WGNE 2011 Slide 40, ©ECMWF
Slide 41
Work on “old observations” in ERA-CLIM
ECMWF report - WGNE 2011 Slide 41, ©ECMWF
Feedback from Marco Matricardi, explaining why the new coefficients would be so much befer
“there are three main differences to the 2001 SSU coefficient files:
1) The new coefficients have been computed using accurate line-‐by-‐line computaBons and the most recent spectroscopic data available at the Bme. 2)A more realisBc value of the cell pressure (which is now plahorm dependent). 3)The CO2 profile can be varied to reflect actual concentraBons (although I do not know how you have addressed this in your computaBons).
It is difficult to disentangle all contribuBons. What I can say is that the transmikances used to derive in the 2001 coefficient files were based on a 20 years old spectroscopy and were performed using a parameterized model rather that accurate line-‐by-‐line computaBons. I expect this to have a rather large impact on the computaBons (more than the contribuBons from 2) and 3). “
Impact of more recent RTTOV on obs-‐guess departures within ERA-‐interim : SSU channel 1
Slide 42
ERA-CLIM pilot reanalyses
What Period Resolution Ens When Vol
ERA-Int Interim reanalysis 1989-NRT T255L60 1 ongoing 33 Tb
ERA-P0 AMIP ensemble 1900-2011 T159L91 10 Jun 2011 (9M)
ERA-P1 EDA using sfc obs only 1900-2011 T159L91 10 Sep 2011 (15M) 655 Tb
ERA-S1 Land surface using ERA-P1 1900-2011 T799 1 Sep 2012 (9M) 77 Tb
ERA-P2 Reanalysis using all obs 2 early decades T511L91 1 Sep 2012 (9M) 180 Tb
ERA-E2 As ERA-P2 but with
SST/sea-ice perturbations 2 early decades T159L91 10 Jan 2013 (9M) 180 Tb
ERA-P3 To replace ERA-Interim 1979-NRT T511L91 1 Jan 2012 (24M+) 234 Tb
ERA-20C 20th-century reanalysis 1900-NRT T511L91 1 2014 (36M+) 1062 Tb
ER
A-C
LIM
ECMWF report - WGNE 2011
Slide 42, ©ECMWF
Slide 43
Thank You
ECMWF report - WGNE 2011 Slide 43, ©ECMWF