Transcript An oceanographic assessment of the GOCE

1
2
3
5
An oceanographic assessment of
the GOCE geoid models accuracy
S.
Mulet1,
Rio1,
Knudsen2,
3
M-H.
P.
F. Siegesmund3, R. Bingham4, 4
O. Andersen2, D. Stammer3,
J. Benveniste5 and the GUT consortium
3
Overwiew
• Assessment of satellite only geoid model: GOCE R5 by
computation of MDT, mean current and comparison with
unfiltered and filtered drifters
• Assessment of satellite only geoid model though
assimilation in numerical model
• Toward higher resolution
– Combined geoid model (with altimetry)
– Combined MDT (with oceanic in-situ data)
Computation of geodetic MDT
from GOCE
Geoid height DIR5
Spatial scale ~ 100km
Mean Sea Surface MSS CNES-CLS11
Spatial scale ~ 10km
(m)
Omission and
commission errors
- Difficulty in
subduction areas and
at the boundary
(land/sea)
 need to be filtered
= Mean Dynamic Topography (MDT)
(cm)
Mean Dynamic Topography (MDT)
Filtering MDT at different scales
MDT from GOCE (DIR5)
filtered with a Gaussian
filter
80 km
Filtering MDT at different scales
MDT from GOCE (DIR5)
filtered with a Gaussian
filter
80 km
100 km
Filtering MDT at different scales
MDT from GOCE (DIR5)
filtered with a Gaussian
filter
80 km
100 km
125 km
Filtering MDT at different scales
MDT from GOCE (DIR5)
filtered with a Gaussian
filter
80 km
100 km
125 km
200 km
Compute associated mean geostrophic
currents
Mean currents from
GOCE (DIR5)
80 km
Compute associated mean geostrophic
currents
Mean currents from
GOCE (DIR5)
80 km
100 km
Compute associated mean geostrophic
currents
Mean currents from
GOCE (DIR5)
80 km
100 km
125 km
Compute associated mean geostrophic
currents
Mean currents from
GOCE (DIR5)
80 km
100 km
125 km
200 km
Comparison with independant
estimate: unfiltered drifters
80 km
Velocity estimate from
surface drifters processed
to match the physical
content of the mean
geostrophic currents
Standard deviation of the
difference with drifters (cm/s)
zonal
Optimal resolution
between 100 and
125 km
100 km
meridional
125 km
80 100 125 150
200 km
200 (km)
Improvement of geoid models: from
GRACE to GOCE R5
zonal
Standard deviation of
the difference with
drifters (cm/s)
meridional
R5
80 100 125 150
200 (km)
Improvement of geoid models: from
GRACE to GOCE R5
zonal
Standard deviation of
the difference with
drifters (cm/s)
meridional
GRACE
R5
80 100 125 150
200 (km)
Improvement of geoid models: from
GRACE to GOCE R5
zonal
Standard deviation of
the difference with
drifters (cm/s)
meridional
R2
GRACE
R5
80 100 125 150
200 (km)
Improvement of geoid models: from
GRACE to GOCE R5
zonal
Standard deviation of
the difference with
drifters (cm/s)
meridional
R2
GRACE
R3
R5
80 100 125 150
200 (km)
Improvement of geoid models: from
GRACE to GOCE R5
zonal
Standard deviation of
the difference with
drifters (cm/s)
meridional
R2
GRACE
R3
R4
R5
80 100 125 150
200 (km)
Improvement of geoid models: from
GRACE to GOCE R5
Zonal (Tx)
Standard deviation of
 Comparison to unfiltering drifters
the difference with
drifters (cm/s)
 Huge improvement of GOCE over GRACE
 Improvement of the differente releases (GOCE
models go closer and closer to the 100 km
resolution !)
 Optimal resolution of mean velocities
computed from GOCE = 100-125 km
and
 At 100 km (GOCE targeted resolution)
StD ~ 7.5 cm/s (include omission and
commission error)
! This include:
- ~3 cm/s on drifter velocity estimate
- an ~1 cm on MSS
Note also that
- velocities (gradient of the height) are much more
sensible that height
- we use gaussian filter that is not an ideal filter
Meridional (Ty)
R2
GRACE
R3
R4
R5
80 100 125 150
200 (km)
Comparison with independant
estimate: filtered drifters
 Comparison to filtering drifters
estimation of commision error only !!
zonal
Standard deviation of
the difference with
filtered drifters (cm/s)
 TIM and DIR5: similar results
 At 100km StD ~ 5cm/s
 Error impacted by geoid
but also MSS, filtering, drifter !
 Error > error only due to GOCE !!
meridional
DIR4
DIR5
TIM5
80 100 125 150
200 (km)
Regional comparison with independant
estimate: filtered drifters
125 km
Regional Comparison
Regional comparison with independant
estimate: filtered drifters
125 km
Assessment though assimilation
in numerical model
Assessment though assimilation in
numerical model
GECCO Setup
MITgcm-Model (PE-eqn. C-grid)
Resolution:1o horizontal, 23 layers
Region: global, 80oS-80oN
Includes sea ice model
Closed boundaries towards the Arctic
KPP mixed layer model
Gent & McWilliams eddy param.
Adjoint assimilation method
Assimilated data:
- SLA from altimetry
- surface S and T
- S,T profiles
- MDT
Control parameters
•
•
Salinity and temperature initial state
Atmospheric state near ocean
surface:
 Wind vector
 Temperature
 Specific humidity
 Precipitation
 Downward longwave radiation
[m]
MDT (DTU10 – GOCO01s) = assimilated MDT
 Iteration 0: no assimilation
 GECCOref: reference synthesis without MDT assimilation
(45 iterations)
 GECCOmdt: synthesis with MDT assimilation (GOCO01s;
89 iterations)
Stop criterion: cost function reduction per iteration step < 0.5%
Siegismund et al. (2014)
Assessment though assimilation in
numerical model
GECCOref
GECCOmdt
Major contributions to the cost function (normalized)
 In comparison to a reference synthesis without
MDT assimilation but otherwise same
configuration, the cost function for all major
components significantly reduces when the MDT
is added as constraint.
 Especially, model-data residuals for SST and MDT
fields strongly reduce
[K]
SST model-data residuals [K]
Toward higher resolution
Comparison filtered drifters
What about combined geoid model ??
 GOCE better than combined geoid model
between 100-200 km !!
 GOCE improved combined geoid model !
 Scales < 100 km have to be estimated with
other data:
•
•
Combined geoid model (altimetry)
Combined MDT (oceanic in-situ data)
Combined geoid models
Combination models including altimetric marine gravity combine
geoid and mean sea surface at observation equation level:
• Augment the series of spherical harmonic functions, hereby
reducing the unmodelled parts of the geoid,
• Reduce errors due to inhomogeniety and anisotropy
Signal and error degree variances
10
-14
10
-16
10
-18
10
-20
10
-22
10
-24
EGM2008 model
EGM2008 errors
GOCO03s errors
0
50
100
150
Harmonic degree
200
250
Eigen-6c minus Eigen-6s
Combined geoid models
Comparing two combination models EGM 2008 and Eigen-6c:
10
-14
10
-16
10
-18
10
-20
10
-22
EGM2008
Eigen-6c - EGM2008
GRACE
GRACE
0
50
Mean Sea Surface
------- GOCE -------------
100
150
200
MSS
250
MSS – Geoid model:
---------------- MDT ----------------------------- ----- 0 ----->> easier filtering
Andersen and
Knudsen, 2014
Combined geoid models
An example using DTU13MSS with Dir-r5
(upper/read) and with Eigen-6C3 (lower/blue):
• MSS-geoid (left)
• Power spectrum (below)
• 2D power spectrum (right)
• Most signal >0.6 cy/deg (≈ d/o 220)
have been removed.
New model - DTU13MDT:
Similar to DTU12MDT updated with
• DTU13MSS
• Eigen-6C3
Improved mainly in the Arctic and in
the equatorial region..
20 year reference period
Consistent with the new
AVISO altimetry reference period.
Geostrophic Currents (DTU13MDT):
Other way to compute HR MDT, combination
with oceanic in-situ data  CNES-CLS13 MDT
Rio et al., 2014
Geodetic MDT
MDT=MSS CNES-CLS11 – Geoid (DIR4)
filtering
Large scale
MDT=First guess
Synthetic Method
The short scales of the MDT (and
corresponding geostrophic currents)
are estimated by combining
altimetric anomalies and in-situ data
Multivariate Objective Analysis
High resolution MDT
The CNES-CLS13 MDT
Have a look at the poster:
3
CONCLUSION
• Optimal resolution of GOCE MDT = 100-125 km (due to
geoid model but also to MSS and filtering method)
• GOCE geoid model helps to improved combined geoid
model between 100 and 200 km
• Positive impact for assimilation in numerical model
• Conclusion !