1. Sébastien Wagner (1) Tom Stone (2), Gary Fowler (1), Tim Hewison (1)
Using lunar observations for calibrating the solar channels of the METEOSAT SEVIRI imager
Sébastien Wagner (1)
Tom Stone (2), Gary Fowler (1), Tim Hewison (1)
(1) EUMETSAT
(2) USGS

2. The Meteosat Second Generation SEVIRI instrument
 Launches
MSG1: launched on 28/08/2002
MSG2: launched on 21/12/2005
MSG3: summer 2012
MSG4: ?
15 min repeat cycle / 10 bits coding
 Sampling distance at SSP:
3 km for VIS06 / VIS08 / NIR16
1km for HRVIS / 4 solar channels
SEVIRI solar band calibration requirements:
10% absolute accuracy for NRT applications + 5% long-term stability
No on-board calibration system  Vicarious calibration only

3. The SEVIRI instrument solar bands
HRVIS
VIS06
VIS08
NIR16

4. Vicarious calibration
SEVIRI Solar Channel Calibration system  offline vicarious calibration
Reference = RTM simulations of Top-Of-Atmosphere radiances
Comparison with TOA measured signal
2 target types used for comparison
Desert bright targets (18 targets)
Dark sea targets (10 targets) (checking purposes)
What about the Moon?
Slide: 4

5. The Moon as a radiometric calibration source (courtesy T. Stone)
INTRINSIC PROPERTIES:
Exceptionally stable
Non-uniform appearance, varying hemisphere seen
(lunar librations)
Non-Lambertian reflectance
Smooth reflectance spectrum
Continuous and periodic changes in apparent brightness
(e.g. phase)  Can be characterized and modeled
UTILITY AS A CALIBRATION REFERENCE
Requires an analytic model with a continuous predictive capability
Stability of the Moon  Model valid for any time  Calibration reference = model
Source : PixHeaven.net / Wikipedia
To date, USGS has the only established lunar reference model, based on the ROLO observation program (more than 8 years of observations).

6. Development of the USGS lunar calibration reference (courtesy T. Stone)
Scope : Capture the cyclic brightness variations of the Moon sufficiently for modeling
Identified need: sample ~25% of the 18.6-year phase/libration cycle
More than 8 years of observations done at the Robotic Lunar Observatory (ROLO)
Site : USGS in Flagstaff, Arizona
Dedicated observatory, located at altitude 2143 m
Twin telescopes
23 VNIR bands, 350−950 nm
9 SWIR bands, 950−2450 nm
imaging systems — radiance
database of > Moon images
(phase coverage from eclipse to 90°)
More than star images, for nightly atmospheric transmission measurements

7. The USGS lunar irradiance model (courtesy T. Stone)
 ROLO lunar images spatially integrated to irradiance and converted to disk reflectance:
Empirical reflectance model — a function of geometric variables only:
By fitting the observations, the model parameters are estimated
Estimated uncertainty:
5-10% in absolute irradiance scale (due to measurement of Vega by the ROLO telescopes)
1% relative accuracy
ref: Astronomical Journal 129, (2005)

8. What about lunar observations with SEVIRI?
Fields of regard for Low Resolution Channels:
Level 1.0: raw measurement data + auxiliary data
Visible Moon (up to 5 observations in a row)
FOR covers a rectangle of 22 N/S and 18 E/W
Level 1.5: data rectified to an uniform grid
No visible Moon anymore
FOR = Earth disk
Lunar observations available in the 4 image corners
(more than 100 potential observations / year)
SEVIRI Level 1.5 image
SEVIRI Level 1.0 image (forward and backward scan)

9. Lunar calibration with SEVIRI
From L15 trailer (24/02/2008 – slot 14:27)
VIS 06:
Moon Min/Max Count = 50 / 170
Moon Mean Count = 86
Moon Std Dev = 27
VIS 08:
Moon Min/Max Count = 50 / 186
Moon Mean Count = 91
Moon Std Dev = 31
NIR16:
Moon Min/Max Count = 46/ 403
Moon Mean Count = 166
Moon Std Dev = 86

10. Sensor spectral responses...ROLO versus MSG1/SEVIRI

11. How to extract the lunar observations for SEVIRI images?
Information available in Lev Lev 1.5 images
Level 1.5 image headers/trailers:
Flag to indicate if the Moon is in the FOR
Statistics on the Moon counts (min / max / mean / std dev)
Level 1.0 images: raw images to be decoded and processed

12. How to extract the lunar observations from the SEVIRI L1.0 images?
Location of the Moon within the SEVIRI Level 1.0 image
Realignment of the Earth image
Location of the sub-satellite point using horizon detection
Location of the Moon with respect to the centre of the Earth
Realignment of the Moon image (to correct for the apparent motion of the Moon)
Location of the Moon centre using horizon detection
Extraction of the Moon imagettes
From a centre of the Moon  estimated radius of the Moon extraction circle (Moon radius + margin)
Scaling of the Level 1.0 counts to equivalent Level 1.5 radiances
Calculation of the equivalent Level 1.5 counts (correction for small potential instrument non-linearities)
Conversion from the equivalent Level 1.5 counts to radiance, using calibration coefficients from the Level 1.5 headers

13. Why is the Moon rectified?
Rectification
Corrected apparent motion
L1.0 image (here from GOES)
Fitted curve derive overall reflectance
Figures: courtesy T. Stone

14. Example of cut-off for 09/02/2004 – 14:30 UTC (SEVIRI / MSG1)
Extracted L1.0 counts
Space count + other (stray light, stars, etc.)
Moon count distribution
Mask
(edge fitting)
Space count sanity check
Offset setting is correct within 0.5 count
Potential of using such an approach to check the Point Spread Function (using the sharp edge of the Moon)

15. Current achievements and preliminary results
Automated extraction tool to build up a lunar observation dataset
Dataset generated for MSG1 (between 2004 and end of 2005) and MSG2 (from mid 2006 till early 2009):
Various phases / librations / positions in the SEVIRI Field Of Regard (potential for looking at scan-angle dependency of the calibration?)
Over 30 potential lunar observations for MSG1
Over 50 potential lunar observations for MSG2
Not all potential lunar observation yet available

16. Preliminary results from Meteosat 8 (MSG1)
Lunar calibration method and instrument are stable. SD <1% - consistent with expected performance of ROLO  BUT is it affected by seasonality?
As expected, radiances in the VIS06 and VIS08 are too low.  But absolute accuracy only “5-10%”
However, VIS08 performs better than VIS06 (Better inter-channel relative accuracy)  in contradiction with current findings with DCCs
Note: End at the end of 2005 after the start of Rapid Scan Service

17. Preliminary results for Meteosat 9 (MSG2)
Results for MSG2 consistent with the ones for MSG1
(VIS08 performs slightly better)

18. Conclusions and future work
What was achieved?
Lunar observation database for MSG1 and MSG2
Systematic extraction from the Level 1.0
MSG1/SEVIRI and MSG2/SEVIRI = very stable in time
Systematic underestimation for VIS06 and VIS08
Amplitude of the bias in disagreement with findings from other groups using DCCs (VIS08 performs better than VIS06)
What does remain?
Update the current database (systematic use of the Level 1.5 headers to check the availability of the Moon in the FOR)  VIS06 / VIS08 / NIR16 (MSG1 and MSG2)
Setup of a database for the HRVIS band (MSG1 and MSG2)
Investigate the availability of lunar observations during the Rapid Scan Service of MSG1
Analysis of the current results to understand the variations in the time series
What about Meteosat First Generation? (MET6 for instance)

19. What about the HRVIS band and the Rapid Scan Service?
Fields of regard for High Resolution Channel:
Level 1.0:
How many visible Moon ?
Limited FOR
Level 1.5:
No visible Moon anymore as for Low Resolution
Lunar observations = available but lower frequency + always on the East
SEVIRI Level 1.5 image
SEVIRI Level 1.0 image (forward and backward scan)