1. SEVIRI Solar Channel Calibration system
Sébastien Wagner
Tim Hewison, Marianne Koenig, Yves Govaerts

2. Objectives and methodology SSCC methodology and algorithm
Outline
Context
Objectives and methodology
SSCC methodology and algorithm
SEVIRI calibration: current status
Impact of the calibration uncertainties
Future developments

3. Context: Meteosat observations...
Archived data since 1982
Keeps growing...
Meteosat imagers solar channel calibration device?
No on-board calibration system (MVIRI / SEVIRI)
On-board solar diffusers (FCI)
Various system set-ups:
MFG: 30 min repeat cycle / 6 bits (Met 1 to 3) or 8 bits (Met 4 to 7) coding / Sampling distance at SSP = 2.5 km / 1 solar channel
MSG: 15 min repeat cycle / 10 bits coding / Sampling distance at SSP = 3 km for VIS06 / VIS08 / NIR16 and 1km for HRVIS / 4 solar channels
MTG: 10 min repeat cycle / Sampling distance at SSP = 0.5 and 1.0 km/ 8 channels (2 non-window channels)
Various specifications for the solar channel calibration:
MVIRI = no spec
SEVIRI = 10% + 5% long-term stability
FCI = 5% + 2% long-term stability

4. Context... The particular case of SEVIRI
HRVIS
VIS06
VIS08
NIR16

5. Objectives and strategy
Near real time / real time applications  operational evaluation of the calibration coefficients
Long time series analysis / climate applications  re-analysis with consistent calibration for the various missions (31 years of archived data)
Strategy:
Vicarious calibration using reference RTM simulations and comparing with TOA radiances
2 target types (potentially 3):
Desert bright targets (18 targets)
Dark sea targets (10 targets)
Deep convective clouds

6. SEVIRI Solar Channel Calibration methodology
Initially for MSG/SEVIRI  Extended to Meteosat missions
Definition of the “reference”: simulated Top-Of-Atmosphere radiances
Evaluation of the “reference”: against well-calibrated polar-orbiting instruments (SeaWiFs, ATSR2, AATSR, VEGETATION, MERIS)
Advantages of using bright desert targets:
If well chosen  very stable
Many potential targets in the Meteosat FOV
Disadvantages of using bright desert targets
If not well chosen  seasonal variations
Difficulty to characterize the surface reflectance
Aerosol characterization

7. The SEVIRI Solar Channel Calibration algorithm
Meteosat Images
(Lev. 1.5)
Data accumulation (5 up to 10 days)
Radiances
Target identification
Pixel extraction
RTM calculation
Cloud mask +
Cloud analysis
K + K0 + target properties (surf + atm)
Cloud free scenes / admissible geometry and surface conditions
Calibration coefficients + associated errors
ECMWF (Analysis – 6h)
H2O total column content
Surface pressure
10m U/V wind
Quality control
Update Lev.1.5 headers
TOMS + AERONET climatology LUTs
O3 total column content
Aerosol AOD
Final calibration coefficients + associated errors + Quality indicator
Estimate of the sensor temporal drift + associated error

8. Example of results – MSG1
Impact of the nature of the signal on the use of some target types
September 2007
Need for quality check and error estimation in order to remove outliers and poor quality retrievals
Large differences in the amount of retrievals

9. SSCC references
Govaerts, Y. M. and M. Clerici (2004). "Evaluation of radiative transfer simulations over bright desert calibration sites." IEEE Transactions on Geoscience and Remote Sensing 42(1):
Govaerts, Y. M., M. Clerici, et al. (2004). "Operational Calibration of the Meteosat Radiometer VIS Band." IEEE Transactions on Geoscience and Remote Sensing 42(9):
Govaerts, Y. M., and Clerici, M. (2004). MSG-1/SEVIRI Solar Channels Calibration Commissioning Activity Report (EUMETSAT).

10. Current status – Meteosat 8 – Level 1.5 Headers

11. Current status – Meteosat 9 – Level 1.5 Headers

12. Current status SEVIRI / MSG 1 Difference Desert / Sea targets:
VIS06 = 8.16 %
VIS08 = 6.26%
NIR16= not quantifiable
HRVIS = 1.54%
SEVIRI / MSG 2
VIS06 = 7.74 %
VIS08 = 3.95 %
HRVIS = 1.38%

13. Impact of calibration uncertainties...
Retrieval of the aerosol optical depth using MSG/SEVIRI data from VIS06 / VIS08 / NIR16  Comparisons with AERONET
(Wagner, Govaerts, and Lattanzio, Joint retrieval of surface reflectance and aerosol optical depth from MSG/SEVIRI observations with an optimal estimation approach: 2. Implementation and evaluation, JGR, 2010)
Other sources of uncertainties in the retrieval scheme could also explain the differences
Non corrected data
Correction:
VIS06 = 9%
VIS08 = 6%
NIR16 = none
Corrected data
Mean RMSE
RMSE std dev
Mean Bias
Bias std dev
No correction
0.145
0.063
0.032
0.0480
Correction
0.055
0.015
0.0476

14. Future work and developments -1
In order to :
Meet the requirements on MTG/FCI: 5% accuracy (half the current requirement on MSG/SEVIRI)
Improve SEVIRI current calibration system
Re-visit the Meteosat archive
Need for reducing uncertainties and improving calibration accuracy
What is foreseen with SSCC ?
Re-assessment of the current system uncertainties
Use of MODIS / MISR data in the current system
Definition of more stable desert targets + characterization of the associated BRF
Improvement of the RTM
Improvement of the aerosol climate data sets, and use of an dust-aerosol mask to avoid dust events to be processed if not well detected
Re-visit the consistency analysis, taking the new BRF errors into account
Re-evaluation of the reference against reference instruments (MODIS, MISR, MERIS-like, ATSR-like, VEGETATION, PARASOL...)
For non-window channels: use of DCCs + homogeneous water clouds as targets

15. Future work and developments - 2
Sphericity:
Spherical 6S
Non-spherical LDA_NSP_MEDRAD
Aerosol load:
Aerosol Optical Thickness = 0.2
Aerosol Optical Thickness = 0.4

16. Future work and developments - 3
What about lunar calibration ?
With SEVIRI, information available in Lev Lev 1.5 images
Very stable surface properties
Reference = entire Moon radiance
Limited number of observations (up to 30min in the FOV, 3 up to 5 times / month)
 Useful for long-term drift but no use for operational calibration
Rectification needed to avoid double counting of the pixels due to lunar motion
Difficult to estimate accurately the surface reflectance (at pixel level and total)
Forward model: need to predict lunar reflectance according to the illumination/observation geometry
SEVIRI Level 1.0 image (forward and backward scan)

17. Thank you for the attention