characterization of scattered celestial signals in smos observations over the ocean

Space Reflecto, November 4th -5th 2013, Plouzané


• an Earth Explorer mission (ESA) for Soil Moisture and Ocean Salinity

• Sun synchronous orbit, 3-days repeat cycle, 32.5º tilt

• MIRAS payload : a polarimetric and interferometric radiometer at L-band, 72 receivers

SMOS

Field of View

Concept


Retrieving salinity


Forward model


The problem

Spatio-temporal systematic biases in SMOS salinities

sun-synchronous orbit

the orbital plane makes one complete rotation relative to the stars per year.

the instrument views roughly the same portion of the sky in successive orbits.


Spatio-temporal systematic biases in Aquarius salinities

The problem


GEOMETRICAL OPTICS GALACTIC MODEL

Fitted galactic model

High-frequency limit of the Kirchhoff approximation for scattering cross sections:

Express in terms of slope probability distribution:

Assume Gaussian slope pdf. Then fit the slope variance (MSS):


SMOSSMOS descending passes example descending passes exampleSSS bias using the SSS bias using the old old (pre-launch) galatic model (pre-launch) galatic model

GO model impact


SMOSSMOS descending passes example descending passes exampleSSS bias using the SSS bias using the newnew galatic model galatic model

GO model impact


GO model impact

AQUARIUSAQUARIUS inner beam (ascending passes only) inner beam (ascending passes only)

AquariusAquarius galactic reflection model galactic reflection model


AQUARIUSAQUARIUS inner beam (ascending passes only) inner beam (ascending passes only)

GO model impact

SMOSSMOS ascending pass fit galactic reflection model ascending pass fit galactic reflection model


DescendingAscending

GO model adjustment to SMOS data

Geometrical optics fitted to the data - slope variance :

•increases with wind speed (expected)

•changes with incidence angle (unexpected)

•changes with pass orientation (unexpected)


1 - Systematic underestimation at high incidence angle

Imperfect model physics (GO, Gaussian slope pdf ?)

Remaining issues with the GO mss fit

the fitted mss depends on incidence angle and pass orientation


Remaining issues with the GO mss fit

2 - Potential phasing between biases and model errors

Need to uncouple bias estimation and model improvement tasks

Estimate biases from expected low model errors

dataset

Update galactic model using a corrected dataset


Faraday :

Dielectric :

Roughness :

Atmosphere :

Direct sun :

Radiometric noise :

Celestial reflections :

• use 1st Stokes parameter

• cancel contribution using Klein-Swift model• latitude band with low SST seasonal variations

• thin interval of wind speed values• thin bands of latitude (sea state)

• model for emission/attenuation contributions

• remove sun alias and sun tails (0.05)

• average over 18-days and longitude [-180º -100º]

• ?

Variability source Reduction approach

1 - phasing between TB biases and model errors

Variability reduction strategy


1st Stokes TB time series : an example

centered(zero mean)

Galactic(KA)



Faraday :

Dielectric :

Roughness :

Atmosphere :

Direct sun :

Radiometric noise :

Celestial reflections :

• use 1st Stokes parameter

• cancel contribution using Klein-Swift model• latitude band with low SST seasonal variations

• thin interval of wind speed values• thin bands of latitude (sea state)

• model for emission/attenuation contributions

• remove sun alias and sun tails (0.05)

• average over 18-days and longitude [-180º -100º]

• select xi/eta points with low annual variations (as predicted by the KA)

Variability source Reduction approach

Variability reduction strategy



highly consistent throughout the field of view

Temporal bias estimates



2 - Underestimation at high incidence angle

55º incidence angle



A fully-empirical correction of the bistatic coefficients



55º incidence, 8 m/s dataGOmodified GO



55º incidence, 4 m/s

55º incidence, 12 m/sdataGOmodified GO


Non gaussian slope distribution

The model is still physically inconsistent :

• the slope variance is still incidence angle dependent

• the slope variance is still a filtered value (the surface slope pdf is frequency-

dependent)

Preliminary results for a more physically-based approach …

• Use a non gaussian pdf which variance is based on Cox and Munk’s results


Non gaussian slope distribution

35º incidence, 8 m/sdataGO

non Gaussian GO



using Cox and Munk’s results

mss = 0.044

for all incidence angles


Summary

• Accurate modeling of the scattered celestial signal is of importance for salinity retrieval at L-band

• The pre-launch Kirchhoff scattering model strongly underpredicts the scattered brightness near the galactic plane and overpredicts the brightness away from the galactic plane under most surface roughness conditions

• A gaussian GO model fitted to SMOS residuals significantly improves the scattering description near the specular lobe BUT remaining inaccuracy at high incidence angle AND physically inconsistent slope variance values

• A methodology was developed to characterize separately TB biases and sea surface scattered celestial signal consistent SMOS celestial residuals

• It is shown that problems come from off-specular contributions modeling

• Preliminary results suggest that most inconsistencies may be removed when accounting for non-gaussian slope statistics

characterization of scattered celestial signals in smos observations over the ocean

Documents