Inclusion of Wave Model Data in Sea State Bias Correction Refinement

1
Inclusion of Wave Model Data in Sea State Bias
Correction Refinement D. Vandemark, N. Tran, H.
Feng, B. Chapron, B. Beckley, T. Moore Ocean
Surface Topography Science Team, Hobart March
2007
SGT, Inc.
2
APPROACH
Driving Assumption information on wave
steepness from global wave model can be
integrated with altimeter Hs and U10 to improve
routine sea state range corrections TRACK
1 Nonparametric global SSB solutions using 2
input variables SLA averaging method Inputs are
Hs, family of alternatives Tran et al., 2006
JGR - methods and 1st results TRACK 2 Three step
clustering approach Partition measurements using
fuzzy clustering Develop multi-class SSB
solutions Combine to give single global result
3
UPDATED DATA and 2nd PASS RESULTS
Ocean surface wave model at UNH WaveWatch III
(H. Tolman/NCEP Feng et al., 2006) 2000-2006
complete, 6 hourly time step, 1 x 1 deg
grid Composite altimeter wave model data sets
at UNH many new correction field updates
TOPEX 2000-2003 - latest NASA/GSFC
pathfinder Jason-1 2002-2005 - GDRa, GDRb
correction versions Key new corrections have SSB
implications MOG2D and ITRF2005 orbits Do we
see impacts? How robust/significant are these
two SSB approaches?
4
Track 1 update Global two parameter NP solutions

• 1- Recomputation of alternate models with
  residual sea surface heights that take into
  account latest geophysical correction versions
  and on longer time period (1-year dataset at the
  time instead of 1/10 of the 2002 year dataset)
• - altimeter retracking (GDRa MLE3 1st order
  Brown model)
• - JMR wet troposphere (GDRb calibration
  parameters from cycles 1-115)
• - dry troposphere (GDRb from ECMWF atmospheric
  pressures and model for S1 and S2 atmospheric
  tides)
• - Mog2D ocean model (GDRb model forced by ECMWF
  atmospheric pressures after removing S1 and S2
  atmospheric tides)
• - Tide (GDRb GOT00.2 S1 ocean tide)
• - Solid Earth tide
• - pole tide
• - dual-frequency ionospheric correction
• CLS01 mss (GDRb)
• 2- Consolidation of the OSTST06 (Venice) results
  with models developed on 1-year period.
• 3- Analysis of 2002, 2003 and 2004 results are
  presented as follows
• ? datasets (2002) (2003) (2004)
• Models ?
• (2002) ?
• (2003) ?

5
Jason-1 SSB Skill Map for H_swell vs Tm vs U_alt
(benchmark is 3.8Hs) H_swell dark best
performer
2002 SSB models, 2002 data results
2004 SSB models, 2004 data results
6
Jason-1 SSB Skill Map for PseudoWaveAge(?) vs
InverseWaveAge vs U_alt (benchmark is 3.8Hs)
? b(Hs/U2)0.62 note ? uses no wave model
data ? dark best performer
2002 SSB models, 2002 data results
2004 SSB models, 2004 data results
7

• Summary on Two Input NP work
• Most results follow Tran et al., 2006
• Relatively stable and positive performance
  observed with swell height (H_swell) and Hs in
  the tropics
• Results with mean wave period (Tm) parameter are
  fluctuating, its contribution to improved SSB is
  less compelling in these analyses
• Pseudo wave age shows surprisingly improved
  performance (over Venice results) compared to the
  usual parameterization with U_alt. Possible
  reason results are now from full year model
• Development of global NP 3-parameter SSB models
  are on-going

8
Track 2 Objective classification of wave
steepness using WW3 and altimeter data
older seas low sea state
mixed seas low sea state

• Method assigns all ocean data samples to
  dynamic provinces
• Useful for dealing with transient and noisy
  wind-wave process, imperfect wave model
  information
• This illustration applies to sea state bias
  work, 6 classes

young seas low sea state
mixed seas mod sea state
steep seas mod. sea state
high seas
9
Clustering Results 6 classes
Clustering input variables Variable Data
source v1 Hs T/P , Jason1 v2 DHHsea/Hs WW3
model v3 DSmssl/msst Alt. WW3 where mssl
(2p )4 /(g2 m4) (m4 accel. variance ) msst
Reff / ?-Ku (where Reff0.45) Rationale
combination of empiricism, theory, and
pragmatism. Effectively using Pseudo Wave Age
(v2) and rms slope (v3)
Resulting classification yields Low wave height
classes (1-3) 1 swell-dominated 2 mixed
sea 3 young seas Moderate wave height classes
(4-5) 4 older seas/mix 5 young seas
(steep) Extreme wave height (6) 6 high seas
10
CLUSTERING APPROACH ISSUES
1. Partition measurements using fuzzy
clustering Can we develop a single classifying
algorithm for full mission partitioning? For
multiple missions? 2. Develop multi-class SSB
solutions Do multi-year analyses show stability
in results? Do Topex and Jason altimeters yield
similar results? physics (EM bias) vs. sensor
engineering. Do NP solutions help to clean up
class-based results? Is there a MOG2D
impact? 3. Combine to give single global
result What is the improvement level? How
systematic is the improvement level in spatial
domain?
11
Multi-year stability in solution- TOPEX
2000
2001
2002
Top panel shows the global direct SSB
mapping on H10 and Hs for 2000, 2001, and 2002
Low panel shows the 200 samples of
data samples for each of the 6 classes in the 2D
domain. from TOPEX-WW3ecmwf ((NASA-GSFC
Pathfinder datasets 1/10 of the total points)
12
2000
2001
2002
Figure 7. Hard partition (max membership)
class-specific direct SSB maps on U10 and Hs
domain for 2000, 2001, and 2002. from TOPEX
WW3-ecmwf (NASA-GSFC Pathfinder datasets 1/10 of
the total points) (200 samples in a cell)
13
2000
2001
2002
Figure 8. Anomaly between the global and
class-specific SSB models for each of the six
classes for the hard partition (max membership)
class-specific direct SSB maps on U10 and Hs
domain for 2000, 2001, and 2002. from TOPEX
WW3-ecmwf (NASA-GSFC Pathfinder datasets 1/10 of
the total points) (200 samples in a cell)
14
Multi-year stability in solution, Jason-1
2002
2003
2004
Anomaly between the global and class-specific
SSB models for each of the six classes for the
hard partition (max membership) class-specific
direct SSB maps on U10 and Hs domain for 2002,
2003, and 2004. from Jason-gdraWW3-ecmwf by
CLS collocation ( 5,000,000 samples were randomly
selected from 17,000,000 total samples).
15
Radar-to-radar stability in SSB anomaly an EM
bias test
2002 TOPEX
2002 Jason-1
16
CORRECTION EFFECTS- MOG2D
Result MOG2D impact apparent in global SSB
algorithm but not in SSB residual between global
model and class-based Conclusion Not a strong
correlation between ocean gravity wave-induced
and HF Barotropic corrections
2000 TOPEX w/o MOG2D
2000 TOPEX with MOG2D
17
Now creating NP solutions for the class-based
data at CLS Jason-1 2004 Anomaly between the
global and class-specific SSB models
Class 1
Class 2
Class 3
Class 4
Class 5
Class 6
18
Jason-1 2004 results using clustering-based NP
SSB solution (6 classes)

• Systematic improvements at all latitudes and
  most regions in the spatial benchmark at right
• Not optimized yet so results will improve

19
SUMMARY

• Both approaches showing measurable improvement
  in global SSB estimation increased at
  equatorial and southern ocean locations
• Stability of the methods appears solid across
  wave model runs, altimeter systems, and altimeter
  correction changes
• Optimization and combination of these two
  approaches is ongoing and they blend well
• Expected improvement level to be documented in
  coming months crossovers present metrics
  (20-30)
• Operational solution appears feasible and
  obtainable

20
We acknowledge the groundbreaking efforts and the
always fruitful collaborations with our
colleagues Roman Glazman and Tony
Elfouhaily They are fondly remembered and
sorely missed.
21
(No Transcript)
22
Blended Retrieval
Retrieval of constituent C is weighted sum
of retrievals from class-specific algorithms,
Ci where weights are based on memberships, fi
23
Membership Function

• Mahalanobis distance
• Zi2 (Rrs - mi)t ?i-1 (Rrs -
  mi)
• Rrs satellite pixel reflectance vector
• mi ith class mean vector
• ?i ith class covariance matrix
• Compute chi-square probability
• fi 1 Fn(Zi2)
• for n degrees of freedom.

24
SSB modeling for each class vs. global avg.
- An illustration of the potential for improved
sea surface estimates through combination of the
6 class-based SSB models (RED positive skill
in cm2)