OVERVIEW OF THE IMPROVEMENTS MADE ON THE EMPIRICAL DETERMINATION OF THE SEA STATE BIAS CORRECTION
S. Labroue, P. Gaspar, J. Dorandeu, F. Mertz, N. Tran, O.Z. ZanifeP. Vincent, N. Picot, P. Femenias
Empirical Estimation of the SSB correction
Why empirical estimation?Theoretical EM bias models not accurate enough. An accurate empirical estimation isbased on 3 issues.
• Regression method=> empirical models based on correlatives measured by the altimeter (SWH and Wind speed)=> 2 kinds of regression methods
– Parametric based on an a priori formulation – Nonparametric estimation free of any assumption on the data
• Data accuracy=> orbit improvements=> geophysical corrections (Ocean tide models, MOG2D correction…)
• Good knowledge of residual SSH errors
Regression Method - Parametric models
Parametric models are generally fitted on SSH differences at crossovers.
TOPEX BM3 (NASA correction 1992)
a*SWH + b*SWH*U + c*SWH*U2
TOPEX BM4 (Gaspar et al. 1994)
a*SWH + b*SWH*U + c*SWH*U2 = d*SWH2
Regression Method - Parametric limitations
• Simulation of a BM4 model with TOPEX data.
• Regression at crossover differences with a BM3 model
=> Error between the estimated SSB and the simulated one, due to the a priori formulation (Gaspar et al. 1998)
=> Development of nonparametric methods
Regression Method - Several methods without a priori model
•Direct method (Vandemark et al. 2002) Direct SLAsAverage of SLA= SSH – MSS per bin of (U,SWH).Long time series needed.
•Hybrid method (Scharoo et al. 2004) Direct SLAsAverage of SLA= SSH – MSS per bin of (U,SWH).Merge with a parametric model and smoothing to extendthe SSB model where no data are available.
•Nonparametric method (NP) (Gaspar et al. 2002) SSH differences or SLAsLocally linear estimator.Kernel smoothing (weights) adapted to the data density.
=> 2 approaches with the NP technique : SSH differences / Direct SLAs
Direct SLAs
• The direct approach assimilates all the residual SSH errors.
• The MSS is equivalent to a mean profile for a given track/mission (Topex, ERS2, GFO).
What is the influence of the residual errors for each mission used in the MSS calculation?
Direct estimate derived on Topex A is the only one in agreement with other methods.
• The SSB derived with the direct method appears to be correlated with oceanic variability.
Regression Method - NP method - Direct vs Difference approaches
SSH Differences (Crossover or Collinear)
• Forming differences at the same location cancels all geographically correlated errors.
• Self consistent : use of the measurements.
• The SSH differences are more likely to cancel oceanic variability (crossovers or repeat cycles differences).
Regression Method - NP method - Direct vs Difference approaches
-3 cm2 +3 cm2
Variance gain on SLA, Jason, Cycles 1-90Variance(SSB_Direct) - Variance(SSB_Collinear)
The direct approach assimilates all the residual SSH errors.
The SSB derived with the direct method appears to be correlated with oceanic variability.
Orbit error (gravity field models)
-5 mm +5 mm
Data accuracy - MOG2D correction
Jason SSB difference, MOG2D correction - IB correction
3 cm
Data accuracy - Orbit
Orbit improvements (gravity field model…)
EnviSat crossovers
Impact on SSB estimation
EnviSat crossovers
-5 cm +5 cm
Good knowledge of residual SSH errors
Effect of a time tag bias of 150 microsec on the estimation SSB.
=> SWH gradient of 1.5 cm
This error only affects the crossover SSH.
Such an error on the data orientates the choice of the method towards the collinear approach.
Topex A Topex B GFO
Jason ERS2EnviSat
Higher SSB
Higher SSB
On board tracker
Ground retracking
18 cm15 cm 34 cm
30 cm 30 cm 42 cm
SSB Results with the NP method - Topex comparison
Topex A – Topex B SSB
NP estimates
Topex A – Topex B SSB
BM4 Chambers et al 2003
SSB Results with the NP method
GFO - Topex B SSB
= -2% SWH
Jason - Topex B SSB
= -1.5% SWH
SSB Results with the NP method
ERS2 – EnviSat SSB
= -1.5% SWH
EnviSat – Jason SSB
No difference
Quality assessment of the SSB correction
IB data, Topex A MOG2D data, Topex A
)(),(21
)()()(
SSBVarSSBSSHCov
SSBVarSSHVarHSSVarD
0 2 0 2
SSB Results with the NP method - C-band and S-band SSB
Ku-band SSB – C-band SSB, Jason Ku-band SSB – S-band SSB, EnviSat
Empirical SSB estimates in Ku-band, C-band and S-band confirms that the SSB increases asthe frequency decreases.
10 cm 13 cm
Conclusions
• NP technique with kernel smoothing has demonstrated to be an efficient tool for the empirical SSB estimation.
• Important improvement have been made on the IB correction which has an impact on the retrieved SSB accuracy.
• Ku-band SSB have been compared between 5 altimeters. They still exhibit differences in the SSB magnitude and (U,SWH) variations which are not fully understood.
• Results have been obtained in C-band and S-band and should help in modelling the frequency dependence of the SSB.
Top Related