Operational use of scatterometer winds in Météo-France data assimilation system - Eumetsat NWP-SAF workshop on scatterometer DA, Darmstadt ...
←
→
Page content transcription
If your browser does not render page correctly, please read the page content below
Operational use of scatterometer winds in
Météo-France data assimilation system
Christophe Payan1,2, Hervé Bénichou1, Bruno Lacroix1
1 Météo-France, 2 CNRS
Eumetsat NWP-SAF workshop on scatterometer DA, Darmstadt, Germany, 14 April 2011
Data assimilation/NWP system: global model ARPEGE
Main characteristics:
• Streched global model, spectral, semi-implicit semi-Lagrangien temporal
scheme, T738C2.4L70, time step:600s
• 4DVAR Assimilation: 6 hr time window, T107/T224 (90 km Gaussian grid),
Time window: T ± 3hr, Analysis times (T):00,06,12,18 Z
2
Data assimilation/NWP system: regional model ALADIN
Main characteristics:
• regional model over Europe, spectral, semi-implicit semi-Lagrangien
temporal scheme, 7.5km of resolution, 70 levels, time step: 450s
• 3DVAR Assimilation: 6 hr time window, T199 (100 km Gaussian grid),
Time window: T ± 3hr, Analysis times (T):00,06,12,18 Z, coupling every 3
hours with ARPEGE
Overseas extensions since 11/2010 (IFS coupling, in red):
Aladin France
Antilles-Guyana
French Polynesia New-Caledonia
Aladin La Réunion
since 2006
3Data assimilation/NWP system: meso-scale model AROME
Main characteristics:
• meso-scale model (non-hydrostatic), spectral, semi-implicit semi-
Lagrangien temporal scheme, 2.5km of resolution, 60 levels, time step: 60s
• 3DVAR Assimilation: 3 hr time window, T359 (55 km Gaussian grid), Time
window: T ± 1.5hr, Analysis times (T):00,03,06,09,12,15,18,21 Z, coupling
each hour with ARPEGE
Operational domain Arome
(red square defines the first operational
model until 11/2010)
4Brief history of Scatterometer Winds use
QuikScat winds assimilated since 10/2004, in-house inversion with
QSCAT-1, resolution 50km, 2 most likely solutions on up to 4 considered.
ERS-2 winds assimilated since 09/2007, in-house inversion with
CMOD5.4, resolution 50km on a 25km-grid.
Ascat winds on Metop-2 assimilated since 02/2008, from Eumetsat
OSI-SAF (KNMI), with CMOD5, resolution 50km on a 25km-grid.
07/2008 onwards, the 4 solutions for QuikScat are considered.
02/2009 onwards, winds are considered as neutral winds (CMOD5.N for
C-Band, biais correction switch off for Ku-Band).
Loss of QuikScat in 11/2009.
Impact estimated in the frame of the GSM Arpège, with operational
use extended to the LAM models (Aladin and Arome).
5QuikScat quality control and use (since 10/2004)
01/09/2004 e-suite run 0UTC, long Cutoff, Indian Ocean zoom
Strict selection of QuikScat
observations, with a high rate
of rejection
Legende (rejection flag) :
(around 70%).
assimilated data
land contamination (model mask>0)
ice contamination (SST < 278K)
swath edge
rain contamination
dd(sol1)-dd(sol2)< 135°
Speed bias correction.
Choice of the ambiguous solution closest to the current model trajectory (2 loops).
Thinning equivalent to 100km.
Test showed a neutral impact on forecast scores (in 2004). 6Ers-2 winds assimilation (since 09/2007):
Regional coverage since 2002
(Northern Atlantic, Australia)
=>800/900 obs used a day on
average (5000/d for Synop-
ship, 1000/d buoys).
Vector difference (obs-first
guess) ~2.8m/s (#4.3m/s for
Synop-ship, 3.0m/s for buoys)
Speed bias around 0.0m/s.
100km thinning (1/16 obs).
Forecast impact: neutral.
7Metop-A winds assimilation (since 02/2008):
29 000 obs assimilated/day
on average.
Vector difference (obs-first
guess) ~2.2m/s.
Speed bias around 0.0 m/s.
100km thinning (1/16 obs).
Forecast impact: neutral.
Bootstrap test/TEMP (32
cases): Z positive impact at
10/100hPa in NH (Europa
and N.Am).
8EARS (EUMETSAT Advance Retransmission Service) - ASCAT:
R0: +13% R18: +22%
R6:
R12: +6%
+40%
EARS allows to reduce the ASCAT timeliness of 15’ to 45’ on the ascending
passes (Fast-dump extract => Northern hemisphere coverage only).
Production (KNMI) and quality equal to the global stream.
+17% data used per day on average on forecast runs, since 02/2008.
9Scatterometer winds assimilated
on a 6 hours assimilation window:
Almost a global coverage per 6h assimilation window of surface oceanic winds.
Loss of QuikScat since November 2009 (red swath).
10Metop versus QuikScat 2 solutions:
Diff Vector (Obs-Background)
Metop: 2.3m/s 2nd Quarter 2008 QuikScat 2sols: 2.7m/s
With the use of the 2 most likely solutions only, Quikscat winds
have higher differences than Ascat winds wrt the model
background in rainy/strong wind areas (ITCZ, baroclinic areas).
11QuikScat: 4 solutions instead of 2?
12Metop versus QuikScat 4 solutions (since 07/2008):
Diff Vector (Obs-Background)
3rd Quarter 2008 QuikScat 4sols: 2.3m/s
Metop: 2.2m/s
Differences between Ascat and QuikScat have disappeared
with 4 wind solutions for QuikScat.
Without losing information where differences to the
background have already been in agreement (and lower)!
Test showed nevertheless a neutral impact on forecasts until
4 days! 13Neutral Wind versus Real Wind?
• Geophysical Model Function: conditions of stability (CS) treated
implicitly
• true in mean but source of error for a singular observation
• in theory, U10 = GMF (σo,CS), in practice not possible
=> solution: U10N = GMFN (σo,CS=neutral)
P L
: State parameters at the last model level
U 10 N
:10m neutral wind observation operator (Geleyn 1987):
U 10 N
= U L [LOG ( z0 , z = 10m ) − COR (CS )] BD(U L , u * )
GMFN with
U L
model wind at the last level, ~17m
z0 roughness length
σ backscattered
LOG logarithmic function of zo and z
COR corrective term function of stability conditions
u * : friction velocity
BD drag coefficient function of U L and u *
u * = CS ( PL , PS )
P S
: Surface state parameters
14Neutral Wind versus Real Wind: impact?
• test of neutral wind in the global model Arpège, from
22/11/2008 to 08/01/2009.
• reference: E-suite Arpège (with a new scheme of turbulence
(Cuxart et al, 2000)). Previous operational scheme based on
Louis, 1979.
• for ERS-2 winds, home-made inversion with CMOD5.N and for
QuikScat winds, change in the speed bias correction.
• switch to a neutral product for Ascat winds from KNMI
(CMOD5.N used since 20/11/2009).
15Speed Bias (Obs-Background)
Metop: +0.3m/s QuikScat: +0.1m/s
22/11/ au 08/01
Neutral Wind versus Real Wind:
impact?
O-Suite
Metop: 0.0m/s E-Suite (Ref) QuikScat: -0.2m/s
Metop: -0.1m/s E-Suite+Neutral QuikScat: +0.1m/s
16Neutral Wind versus Real Wind: impact?
Neutral-wind operator impact is neutral for Ascat/E-suite.
Speed bias improved for QuikScat/O-suite+E-suite (ITCZ+Mid-
latitudes).
Forecast impact positive/E-suite on the first ranges of forecast,
on SOUTH20 (Bootstrap Test/its own analysis). +/- : confidence of 99%
Z, 48 cases ++/-- : confidence of 99.9%
s
nge SOUTH20 TROPICS NORTH20
a
R 0h 24h 96h
Fc
10hPa
1000hPa
Neutral with the other diagnosis.
Operational since 02/2009. 17Neutral Wind versus Real Wind: impact?
Impact confirmed by reduction of the analysis increments on
the MSLP (in blue):
Note the impact is more important in the Southern Hemisphere
because the land mass fraction is weaker.
18Summary on the 2 last changes
Equivalent quality between Ascat and QuikScat 4
solutions.
Neutral-wind operator improved speed bias of
QuikScat, is without effect on Ascat after turbulence
scheme change and in the end better agreement between
the Model and its Analysis.
Outlook
Quality control improvements (land, ice, current,…)
Tuning of observation errors, thinning,…
Failure of QuikScat since 11/2009, stop of ERS-2 in 2011?
Validation of Oscat on Indian satellite Oceansat-2, Eumetsat
OSI-SAF product, ongoing…
Other instruments (HY-2A?, Metop-B, ScatSat?, CFOsat…)?
19Ongoing: land quality control checking
Land contamination control, based on model land-sea mask (sea=0, land=1)
LSM=0.1
LSM=1
LSM=0
Lowest speed bias and vector difference values for Land-Sea Mask=0 (over sea,
no land fraction), for all instruments (here AMI, Ascat and Seawinds), in
agreement with operational QC (data rejected if LSM>0)
20Ongoing: ice quality control checking
Ice contamination control, based on model Sea-Surface Temperature (SST).
SST=273K
SST=278K
SST=271.5K
Sensitivity of error to SST (as ice indicator) depends of instrument. Suggesting
different SST thresholds for the Ice-QC, between 272 and 275K.
Current operational threshold, 278K, is too high! Fixed first with QuikScat data
only, with 2 solutions only, and giving in 2004 the best results in term of forecast
scores wrt to a lower threshold (273K). Will be revised (this year?).
21Outlook: surface current and errors
Scatterometer winds are relative winds!
Local wind > 0 Local wind > 0
Incident
wave
Backscatter
component Reflected
wave
Zero current Current = Local wind
Slight sea Calm sea
METEO FRANCE MONITORING STATISTICS : 01/04/2010 – 30/04/2010
Ascat Vector difference (obs-fg)
MEAN RMS=2.3m/s
Common patterns between Ascat Vector difference (left) and Mercator
analysis of surface current (right) ITCZ, Gulf Stream, Western South-Atlantic,
Cape Point, Japan’s East coast, Papua… 22You can also read
