- 1. CORRECTION OF SPATIAL ERRORS IN SMOS BRIGHTNESS TEMPERATURE
IMAGES L.Wu, I. Corbella, F. Torres, N. Duffo, M. Martn-Neira
2. SMOS Image Reconstruction Errors
- All visibility samples show residual systematic errors due to
non-perfect instrument calibration.
-
- PMS gain (residual Tph dependence)
- The Flat Target response, as a set of Visibilities, isprone to
the same errors.
- Uncertainties in antenna patterns are directly inherited by the
G-matrix elements.
- As a result: The reconstructed Brightness temperature shows
spatial distortion in the , plane
- This error is systematic and cannot be reduced by time
averaging.
3. Spatial errors model
- At each snapshot MIRAS delivers a Brightness Temperature imageT
B ( , ) .
- Spatial artifacts can be modeled as having
-
- Different gains at each grid point
-
- Different offsets at each grid point
- The first option is considered here as offset is highly
cancelled by the Flat Target Transformation
4. Estimating the spatial error
- Spatial error is estimated by carefully analyzing a large
number of measurements over a constant target (i.e. the
ocean).
- To minimize temporal and geophysical variations, a large number
of orbits at different dates are used.
- Ascending and descending orbits are considered.
- The final estimation is an image independentmaskto be applied
to all measurements.
5. Basics on mask estimation over the ocean Geometry (faraday)
rotation Polynomial regression Geometry (faraday) rotation The
final mask is computed on antenna frame. 6. Polynomial
regression
- Measured brightness temperature at different ( - ) is converted
to ground plane and arranged for equal incidence angle.
- Geophysical variability within the FOV of a single snapshot is
neglected.
- Spatial errors are computed as the difference between the
measured brightness temperature and its estimation by polynomial
regression.
7. Mask estimation
- Brightness temperature absolute errors at ground frame are
estimated by the regression
- Errors are transformed to instrument frame and converted to
relative
- The mask is computed from the estimated relative errors
as:
- Once the mask is available, the corrected TB is:
8. Faraday rotation angle
- Faraday rotation is stronger when the TEC (Total Electron
Content) of the atmosphere is larger
9. Faraday rotation correction
- Faraday rotation correction must be applied.
Difference between ascending and descending orbits, before
(left) and after (right) faraday rotation correction. 10. Final
Mask on antenna frame Mask is computed using several orbits over
the pacific ocean from 20 th ,Feb to 20 th ,Sep, 2010. 11. Cuts of
the mask X pol Y pol The spatial error is constrained to 2% in SMOS
AF-FoV 12. Mask average along track Mask average along track is not
zero mean producing artifacts along this direction. 13. Images over
the Indian ocean (01/12) applying the mask 14. Residual Spatial
errors Radiometric spatial errors (pixel bias) over ocean The mask
clearly reduces the residual error and randomizes its spatial
distribution. case Error budget v7.0 2.13 2.13 Before mask
correction 1.34 1.57 After mask correction 0.38 0.47 15. Corrected
brightness temperature H/V brightness temperature arranged for
equal incidence angle for a SMOS image over the Atlantic Ocean and
the Indian Ocean, before (top) and after (bottom) applying the mask
correction. 16. Effect on Level 1C brightness temperature
Horizontal polarization Vertical polarization (35 to 55 incidence)
17. Image cut at -25 latitude 18. Mask stability check (i) single
mask deviation from mean mask computed in March, May and July 2010
to show temporal stability. 19. Mask stability check (ii) the masks
computed using ascending orbits' data are more stable than using
descending orbits'. 20. Conclusions
- Systematic image reconstruction spatial errors have been
estimated from a large number of observations over the ocean.
- A method to correct for these errors has been developed, with
special application to improve salinity retrievals.
- Correction is as simple as multiplying the raw measurements at
level 1B by a constant, direction dependent mask.
- The mask computation is affected by Faraday rotation
correction.
- Results both in L1B and L1C data show that there is a reduction
of the artifacts.