Upload
cory-hill
View
228
Download
0
Embed Size (px)
Citation preview
Use of Satellite data at ECMWF © ECMWF
JCSDA seminar 20/10/2014
Satellite Data Assimilation at ECMWFStephen English
with special thanks to:Tony McNally, Niels Bormann, Alan Geer, Marco Matricardi, Sean Healy, Cristina Lupu, Marta Janisková, Michael Rennie, Massimo Bonavita, Lars Isaksen, Mike Fisher, Richard Engelen, Peter Bauer, David Richardson, Thomas Laiden and Erland Källén.
Use of Satellite data at ECMWF © ECMWF
• General introduction to ECMWF
• Satellite data assimilation overview at ECMWF
• Data assimilation strategy at ECMWF
• Recent highlights in satellite DA at ECMWF
• Current impact – recent ECMWF OSEs
What to expect in this talk
Use of Satellite data at ECMWF © ECMWF
General introduction
Use of Satellite data at ECMWF © ECMWF~8-10km~210km ~63km ~39km ~16km
Evolution of ECMWF forecast skill E. Källén
Use of Satellite data at ECMWF © ECMWF
Forecast error growth500hPa height, NH
Forecast lead time (days)
RM
S e
rror
(m
)199020002010
1.5 day gain
2 day gain
2.5 day gain
E. Källén
Use of Satellite data at ECMWF © ECMWF
Skill gain relative to ERA-Interim (day 5)
Skill relative to ERA-I at day 5. Verification against analysis for 500 hPa geopotential (Z500), 850 hPa temperature (T850), mean sea level pressure (MSLP) and 2 m temperature (T2M_AN), using RMSE as a metric; against SYNOP for 2 m temperature (T2M), 10 m wind speed (V10), and total cloud cover (TCC), using error standard deviation.
T. Haiden
Use of Satellite data at ECMWF © ECMWF
Recent implementations:• November 2010 (36r4): New cloud scheme, SEKF soil moisture analysis,
SKEB• June 2011 (37r2): AMSU-A obs. error, EDA variances in 4D-Var• November 2011 (37r3): Rev. cloud scheme, aircraft b/c, NEXRAD assimilation• May 2012 (38r1): New Jb, EDA-filtering, clouds/convection• June 2013 (38r2): HRES/4DVAR/EDA L91 → L137• November 2013 (40r1): ENS L62 → L91, ocean day-0, 25M EDA,
GWD/diffusion
Note: Satellite introductions often between cycles e.g. first use of China’s FY3B satellite in September 2014.
Future implementations• December 2014 (40r3): 4DVAR 3xT255, 2x weekly 11M reforecasts, new
surface fields and trace gas climatologies, clouds & convection improvement (e.g. supercooled water), lake model (FLAKE), “all-sky” MHS, 2D RO operator
• Autumn 2015 (41r2): Horizontal resolution → 8 or 10km
System updatesP. Bauer
Use of Satellite data at ECMWF © ECMWF
ECWMF Satellite Data Assimilation
Use of Satellite data at ECMWF © ECMWF
5.5 more instruments per year
Around 80 satellite instruments are processed in the operational IFS. Sounders (microwave and
infrared) are the most critical observations for accurate NWP and must be properly maintained.
Use of Satellite data at ECMWF © ECMWF
M Janiskova S Di MicheleNew technologies in space: Cloud radar and lidar
Radar Lidar
O
B
A
Observed (O), Background (B) and Analysis (A) for spaceborne radar (Cloudsat) and lidar (Calipso) data. ECMWF short range forecast
captures 2D cloud structures seen in the observations.
New technology can fill gaps in the Global Observing System. Many years of sustained effort are needed to develop the capability to fully understand and make use of highly innovative new observations in a data assimilation system.
Use of Satellite data at ECMWF © ECMWF
Clear-sky MHS impact All-sky MHS impact
T+72 RMS difference normalised by RMS of control
-4% -2% 0% 2% 4%
Improved assimilation: “all-sky” microwave soundingModelling cloud effects extract more benefit (blue colours) from MHS
than trying to screen out cloud-affected data.
All sky assimilation of microwave imagers improves wind and humidity fields. SSMIS and MHS humidity sounders are being moved to all-sky, and infrared humidity observations may follow.
A. Geer
Use of Satellite data at ECMWF © ECMWF
Water vapour in the presence of cloud - 183±1
StartAssimilation window
UTHincrement (200-500 hPa mean RH)
Zonal windincrementat 400 hPa
Time of observation
Humidity reduction at observation time generated by changes in wind (and other dynamical variables) 1000km away, 9h earlier!
A. Geer
Use of Satellite data at ECMWF © ECMWF
ECMWF Data Assimilation
Use of Satellite data at ECMWF © ECMWF
Ensemble of data assimilation
• 25 members of 2 inner-loop 4D-Var’s at T95/159 L137, T399 outer lops• Perturbations from observations, SST, SPPT; noise filtering, scaling
Use of Satellite data at ECMWF © ECMWF
DA future strategy at ECMWF
Hybrid EDA & 4D-VarLong window 4D-var with weak constraint.
- 24h weak constraint beats 24h strong constraint.
- More work needed on weak constraint to beat current operational 12h strong constraint.
Scalability of 4D-var is an issue – but we anticipate solvable, at least for the time being (time-parallel saddlepoint formulation).
Theoretically the longer the window the better, but devil is in the detail (formulating model error!).
Use of Satellite data at ECMWF © ECMWF
• EDA is a very skilful system, but it is expensive. How well do cheaper alternatives to error cycling compare?
• An EnKF-based error cycling system without perturbed observations could be a cheaper alternative
Other options being investigated
• One among many options: If we have EnKF and 4D-var we can blend the Kalman gain – we call this “hybrid-gain”
xa=xb +(αKEnKF + (1- α)K4DVar)(y-Hx)
M. Bonavita
Use of Satellite data at ECMWF © ECMWF
Recent satellite DA highlights
Use of Satellite data at ECMWF © ECMWF
Recent highlights and progress in Satellite DA
ECMWF funded (6 people) All sky MHS assimilation
(2014, CY40R3) Metop + S-NPP instruments
(2014) IVER, ODB-IFS, cycle testing
(2014, CY40R3) High profile case studies and
OSEs e.g Haiyan, Sandy (2014) Improved understanding of
observation impact diagnostics (2014)
Observation error correlations for hyperspectral instruments (2015)
All sky IR assimilation (2016)
Externally funded (12 people)• Evaluation of Chinese FY3A/B/C (2014)
• Skin temperature analysis (2014, 40R2)
• RTTOV-11 (2014, 40R2+3)
• Improved MW emissivities (2014, CY40R3)
• Improved use of AMVs (2014, CY40R3)
• 2D RO operator (2014, CY40R3)
• MW observation error model (2015)
• Improved aerosol detection for IASI, CrIS and AIRS (2015)
• Improved MW imager assimilation e.g. cold air outbreaks (2015)
• PC assimilation demonstrated for cloud-free scenes (2016+)
• New ideas to exploit correlated error, rather than just allow for ir (2016+)
Use of Satellite data at ECMWF © ECMWF
surface
ray path
Improvement of the 2D approach in 40R3
The outer loop uses 31 profiles to describe the 1200 km “occultation plane”. 7 profiles used for inner loop.
Interpolate 2D information to the ray path
Tangent height of the ray-path determined by the impact parameter provided with the observation, .
S. Healy
Use of Satellite data at ECMWF © ECMWF
Impact of tangent point drift (37R2) and the 40R3 2D operator
S. Healy
Use of Satellite data at ECMWF © ECMWF
Aeolus wind product statusLaunch mid 2016?
- 2014: L2B processing facility: prototype → operational
- Summer 2014: ESA ground Segment testing started
M. Rennie
Use of Satellite data at ECMWF © ECMWF
Humidity
Better background fit to other observations over the tropics, especially for humidity-sensitive observations:
Temperature
Updated R for IASI and AIRS(new σO + correlations vs operational σO without correlations)
N. Bormann
Use of Satellite data at ECMWF © ECMWF
Updated R for IASI and AIRS (2)
S.Hem. N.Hem.
Normalised RMSE differences over
6 months: Aug-Oct 2013; Jan-Mar 2014,
(vs own analysis)
New worse
New better
Forecast day
Tropics
New worse
New better
Forecast day Forecast day Forecast day
Forecast day
N. Bormann
Use of Satellite data at ECMWF © ECMWF
Fastem – development history
Dielectric properties
Surface roughness
Directional anisotropy
Whitecapping
As full model
Fast fit to full model
As full model
As full model
Full model Fast model
RTTOV, CRTM etc.
Integration into RTM
Fastem-1: English and Hewison 1998: Created for 20-90 AMSU-A.Fastem-2: Deblonde and English 2002: Extend to MW imagers.Fastem-3: English 2007: Extend to polarimetric imagers.Fastem-4: Liu et al. 2011: Extend to 1-20 GHz and 90-200 GHz.Fastem-5: Liu et al. 2012: Fixed some weaknesses identified in Fastem-4 by users.Fastem-6: Kazumori and English 2014: Fixed anisotropy model in Fastem-5
V1 V2 V3 V4 V5 V6
Use of Satellite data at ECMWF © ECMWF
Fastem-6: more accurate ocean emissivity for RTTOV-11/CY40R3 + 2 QJ papersUse of the Ocean Surface Wind Direction Signal in Microwave Radiance Assimilation
Masahiro Kazumori and Stephen J. English
Asymmetric features of oceanic microwave brightness temperature in high surface wind speed condition
Masahiro Kazumori, Akira Shibata, and Stephen J. English
June 20 to October 3
= Fastem-5 = Fastem-6
M. Kazumori (JMA)
Use of Satellite data at ECMWF © ECMWF
Recent OSEs
Use of Satellite data at ECMWF © ECMWF
Whilst the impact of a single instrument may appear small, when we test multiple instruments together, like
all the Metop-B instruments, the benefit is clear.
Metop-B IASI only
A. McNally
S Healy
Metop-B IASI+AMSU-A +MHS+GRAS+ASCAT
Long term gains in forecast skill can arise from many small changes. We should not always expect to get measurable
medium range positive impact from adding new observations.
-2%
to +
2%
-2%
to +
2%
Use of Satellite data at ECMWF © ECMWF
OSE 500z (NH-24h)
NO SCAT
NO GEO
NO MWI
NO GPS
NO IRS
NO MWS
NO CONV
0 0.05 0.1 0.15 0.2 0.25 0.3
T. McNally
Use of Satellite data at ECMWF © ECMWF
OSE 500z (SH-24h)
NO SCAT
NO GEO
NO MWI
NO IRS
NO GPS
NO MWS
NO CONV
-0.02 0 0.02 0.04 0.06 0.08 0.1 0.12 0.14 0.16
T. McNally
Use of Satellite data at ECMWF © ECMWF
72hr Tropics 200hPa VW
NO SCAT
NO IRS
NO GPS
NO MWS
NO CONV
NO GEO
0 0.005 0.01 0.015 0.02 0.025 0.03
Series1
T. McNally
Use of Satellite data at ECMWF © ECMWF
OSE 500z (NH-24h) v FSO (NH-24h)
NO SCAT
NO GEO
NO MWI
NO GPS
NO IRS
NO MWS
NO CONV
0 0.05 0.1 0.15 0.2 0.25 0.3
T. McNally
Use of Satellite data at ECMWF © ECMWF
Do results of OSE and FSO disagree ?
RMSE(IASI) minus RMSE(NO-IASI)RMSE(IASI*) minus RMSE(NO-IASI)
C. Lupu
Use of Satellite data at ECMWF © ECMWF
Summary and thoughtsForecast skill continues to improve rapidly.By 2016 90+ satellite instruments processed operationally – a peak?Most satellite DA is achieved through collaboration (e.g. NWPSAF,
CMA-ECMWF partnership for FY3, Horizon-2020, Copernicus…).
New instruments: Radar, lidar, L-band, Limb sounders all demonstrating impact. Who will pay for an operational service?
Handling cloud and rain affected satellite data will become mainstream – potential being realised, this is the way forward.
System is robust to losing ANY observation type, even the most important. But note in situ data remains important.
ECMWF still thinks 4D-var has a bright future, but we are evaluating other approaches.
Use of Satellite data at ECMWF © ECMWF
Use of Satellite data at ECMWF © ECMWF
Atmospheric CompositionIFS code has been extended with full chemistry, aerosols, and
greenhouse gases -> “C-IFS”Strong reliance on ageing satellite systems: ENVISAT (failed), EOS
AURA (OMI & MLS), MOPITT, MODISMetop, Sentinel-5p, -4 and -5 therefore very important for
sustaining and improving current capabilitiesFrequently retrievals are assimilated
Use of Averaging Kernel in data assimilation removes sensitivity to prior information used in retrievals
Global mean CO analysis profiles
With AK
Without AK
R. Engelen
Use of Satellite data at ECMWF © ECMWF
P de Rosnay, J Muñoz Sabater, C AlbergelLand Surface
Snow cover preprocessing and 24 to 4 km resolution improvement
NH 1000 hPa Z
SMOS brightness temperature assimilation: soil moisture increments
Root zone soil moisture analysis from EKF assimilation of ASCAT
Marine, land surface and atmospheric composition also need high quality satellite observations though lack of mature tools means
retrievals are often assimilated: however we now know better how to assimilate retrievals.
Use of Satellite data at ECMWF © ECMWF
24h weak vs 24h strong24h weak-constraint (red) significantly better than 24h strong-constraint in NH.Small improvement also in SH.Verification is against own analyses.
Mike Fisher
Use of Satellite data at ECMWF © ECMWF
24h weak vs 12h strong24h weak-constraint (red) remains worse than 12h strong-constraint in NH.Similar scores in SH (although some degradation at short range).Verification is against own analyses.
M. Fisher
Use of Satellite data at ECMWF © ECMWF
T399 EnKF (100 member)T95/T159/T399 4DVar with static BT399 Hybrid Gain EnKF (100 member)
Z500 hPa AC - NHem Z500 hPa AC - SHem
M. Bonavita
Use of Satellite data at ECMWF © ECMWF
2D impact at short-range (winter experiment)
NH SH
Clear improvement in GPS-RO statistics of ~ 5 % in troposphere.
More importantly, the 2D framework enables us to investigate further improvements include more physics.
S. Healy
Use of Satellite data at ECMWF © ECMWF
Winds assimilated in an ECMWF cycle
• Very uneven distribution• AMV coverage good in tropics, but
obs errors large• Stratosphere poorly sampled
log10(number obs per area)
M. Rennie
Use of Satellite data at ECMWF © ECMWF
Forecast rms errors for the 850 hPa relative humidity in the Tropics
Verification against radiosondes: Temperature in the Tropics
RADPC_CORR
RADPC_CORR
Assimilation of PC scores derived from 305 IASI channels
M. Matricardi
Use of Satellite data at ECMWF © ECMWF
Hurricane Sandy
Use of Satellite data at ECMWF © ECMWF
Hurricane SandyT. McNally
Use of Satellite data at ECMWF © ECMWF
Hurricane SandyT. McNally
Use of Satellite data at ECMWF © ECMWF
Analysis differences that led to failed (NO –LEO SAT) forecast
Control minus NO-LEO SAT MSLP 2012102500z (after 5 days of data denial)
T. McNally