1. Strawman Plan for Inter-Calibration of Solar Channels
Dave Doelling
Tim Hewison

2. GSICS visible product Incoming solar irradiance spectrum
CEOS uses Thuillier_2002
Satellite visible channel spectral response and solar constants based on incoming solar
Gain and offset to calibrate the visible channels
Do we want to work in reflectance or radiance space?
Apply correction as Rad = gain*(C-Co) or Ref = Refo*gain + offset ?
Provide correction of calibration?
Where Refo is the nominal gain, provided on GSICS page
As day since launch function, or periodic gain updates
Use operational space offset as offset
Uncertainty of gain correction with respect to reference
Based on a common absolute calibration
MODIS-Terra for 2000 to present and historical?

3. Reference Sensor
No current reference visible hyper-spectral sensor available, similar to IASI for IR
CLARREO launch ~2018, VIS ~100km footprint, IR ~25km
GOME & SCHIAMACHY absolute calibration uncertain to ~4%
Reference sensor to have solar diffuser or well-established calibration
MODIS (well characterized and known), ATSR, VIRS
Reference to have as many common channels as possible
Future NPOESS/METOP missions to continue with well-calibrated sensor, VIIRS similar to MODIS
Need to establish the difference in absolute calibration when reference sensors are replaced

4. Methodology
Use either the GEO-MODIS ray-match or invariant target as the primary approach and validate with the secondary
GEO-MODIS is easy to implement
Will not work for historical satellites
Absolute calibration will change with reference
Invariant Target Approach
Will work for historical satellites and not tied to reference sensor, once absolute calibration has been established
The reference satellite, which is common to all GEOs will monitor invariant target consistency
Both methods still require spectral response corrections
From model or hyper-spectral or multi-spectral observations

5. GEO-MODIS ray-match Approach
Regress ray-matched coincident and collocated reflectances from GEO and reference
Make sure the regression is linear
Traceability to reference
Monitor gain over time
Apply correction factor to take out spectral differences
Model or empirical regressions

6. Ray-matching to reference sensor
Ray-match coincident GEO counts, radiances or reflectances and MODIS radiances or reflectances averaged over a 502 km ocean grid near the sub-satellite point (±15°lat by ±20°lon area)
Use GEO provided space offset
Perform monthly regressions to derive monthly gains
Compute timeline trends from monthly gains

7. Apply spectral correction
Approach: Model (MODTRAN) reflectance spectra is determined by surface reflectance and atmospheric absorption and is based on LUT for every individual ray-matched coincident reflectance pair
Fokke et al. 2009
• Redo monthly regression to derive monthly gain with spectral correction

8. Invariant Target Approach
Identify invariant targets consistently
DCC, deserts, moon, clear oceans
Determine scene identification thresholds
For DCC, T<205K, sVIS<2% etc, limit auxiliary input
Develop robust statistics
Build bidirectional model
From observations during stable time period
Or Radiative Transfer Modeled
Identify angular ranges that are predictable
Indentify and build bidirectional model from reference sensor, using same approach as GEO
Normalize the GEO to the reference bidirectional model
Apply the spectral response difference from model or SCIAMACHY spectra

9. Normalize the GEO DCC with MODIS
Indentify DCC between GEO and MODIS consistently
Build an observed bidirectional reflectance model with DCC pixel reflectance
Normalize the GEO DCC reflectances to MODIS
MET9, 0.65µm, one month
Terra-MODIS, 0.65µm
Both approaches used T<205°K and sT<1°K and sVIS<2%
Approach 1
Approach #1: Regress corresponding angular bins
Approach #2: Develop bidirectional model with ray-matched GEO/MODIS calibration
Approach: #3 Normalize theoretical model onto observed model
MET9
MODIS
Regress reflectances of corresponding bins and apply coefficients to all bins of MET9

10. ( ) S S rmodi rmodj i k rVIRSi,k rj,k
Normalize theoretical model onto observed model to develop complete reference model
VIRS (observed, scans to 45° VZA)
DISORT
DISORT+VIRS
Use all available angular bin ratios and observed reflectances to estimate unfilled bins ( )
_______
rmodi
rmodj i k
S S
MVIRS bins
N bin footprints
rVIRSi,k
rj,k =
/MN
Loeb et al, 2002

11. Correct for sensor spectral response differences
• Use SCIAMACHY spectral response to estimate the spectral correction from MODIS to MET8
• Similar approach for a modeled spectra

12. Invariant target monthly gain approach
Predicted MET9 Radiance, Wm-2sr-1um-1
DCC
Deserts
Clear Ocean
Offset
Again track over time

13. Desert Target Approach
Deserts are a fixed land target, that are optimally Lambertian
Have only one VZA in the GEO domain
Have only a limited SZA with reference satellite in sun-synch orbit
may need a common bidirectional model from VIRS, POLDER or MISR data to transfer absolute calibration between reference sensor
MODIS bi-weekly surface bidirectional reflectance product?
Deserts have unique spectral responses and should base results on several deserts for robustness
May need to use a SCIAMACHY (30x120km footprint) or database spectral response
Normalize to MODIS standard channels, or first with multi-channel satellites, if using bidirectional model from POLDER or MISR
Desert spectral response can then be used to correct for spectral response differences between sensors
Bidirectional spectral effects can be taken out with models

14. Monitor Stability over time
Update calibration coefficients over time if necessary
Both approaches need to evaluated for uncertainty in the gain derivation with respect to the reference satellite
Reference absolute satellite calibration stated in GSICS
Invariant target uncertainty: temporal target invariance, bidirectional models, and spectral correction, stratospheric aerosols
GEO-MODIS regression uncertainty: standard error and spectral correction
All Invariant targets and GEO-MODIS calibration can monitor relative calibration independently
Desert site stability can be established through the reference satellite, two GEOs, or surface measurements
DCC are well suited for GEO since their orbits do not degrade
Moon looks are available on GEOs operational and can be used if characterized properly with phase angle, etc

15. Way Forward Agree this strategy
Identify specialists working on each method
Deserts, DCCs, Ocean, Moon, Ray-matching collocations...
Coordinate presentations from GSICS members
to cover each method at next meeting in Toulouse
Agree to combine methods after period of research
e.g. by Sept 2010
Decide a method of combining the methods
to check consistency and
identify most robust for each application
Identify products for solar channel inter-calibration
Develop prototype products
e.g. by June 2011