Cross-Calibration of Optical Sensors

Desert cross calibration accuracy assessment
using Hyperion data and lab. sand samples
Patrice Henry, Bertrand Fougnie, Sophie Lacherade,
Philippe Gamet, Denis Blumstein - CNES
Thomas Colin - CS
GSICS GRWG Meeting – March 2011 – Patrice HENRY / CNES
Hyperion data (1)
■ Hyperion data over 5 desert sites provided by USGS
Site
Number of Spectra
Algeria 3
33
Algeria 5
3
Mauritania 1
17
Mauritania 2
17
Libya 4
107
Number of Hyperion data per site
GSICS GRWG Meeting – March 2011 – Patrice HENRY / CNES
Hyperion data (2)
10 Hyperion spectra over Algeria 3
GSICS GRWG Meeting – March 2011 – Patrice HENRY / CNES
Hyperion data (3)
10 Hyperion spectra over Libya 4
GSICS GRWG Meeting – March 2011 – Patrice HENRY / CNES
Simulation using Hyperion data (1)
■ TOA reflectance of different sensors (MODIS, MERIS, PARASOL,
VGT, ETM+) are simulated using Hyperion TOA data
where :
rBi = simulated Bi band TOA reflectance
rHyperion(l) = Hyperion TOA reflectance at wavelength l
SBi = Bi band spectral profile
ETOA(l) = solar irradiance at wavelength l
GSICS GRWG Meeting – March 2011 – Patrice HENRY / CNES
Simulation using Hyperion data (2)
Example of Hyperion TOA reflectance profile and simulated data
Algeria 3 – 26 Jan. 2007
GSICS GRWG Meeting – March 2011 – Patrice HENRY / CNES
Cross calibration method
■ TOA ref. sensor reflectance are
transfered to the ground
■ BOA reflectance are resampled to
compute simulated ‘to be
calibrated’ sensor reflectance
Atmospheric correction
to surface reflectance
r k ,measured( L1)
Sensor 2
Reference Sensor
■ ‘to be calibrated’ sensor reflectance
are uploaded at TOA level
■ Simulated and measured TOA
reflectance comparison gives cross
calibration between reference and
‘to be calibrated’ sensors
TOA
TOA
Ak 
rk ,measured( L1)
rk , predicted ( Method)
Comparison
TOA
Spectral resampling
SURFACE
Main sources of errors :
Pseudo sensor 2
surface reflectance
 Spectral resampling
 Atmospheric correction (twice)
 Acquisition geometry
GSICS GRWG Meeting – March 2011 – Patrice HENRY / CNES
Atmospheric simulation
to ToA reflectance
r k , predicted ( Method)
Sensor calibration vs itself
Libya 4 site (107 measurements)
interpolated value
 allows to assess atmospheric correction processing accuracy
(max. error : 0.5%)
GSICS GRWG Meeting – March 2011 – Patrice HENRY / CNES
Sensors cross calibration
■ Different sensors cross calibration performed :
 Aqua/MODIS vs MERIS
 MERIS vs Aqua/MODIS
 ETM+ vs Terra/MODIS
 VGT2 vs Parasol/POLDER
 Parasol/POLDER vs Aqua/MODIS
 Parasol/POLDER vs MERIS
■ Different cross calibration method tested :
 Same geometry (data pairs simulated with the same Hyperion data)
 Close geometry (data pairs from close geometry Hyperion pairs)
 Closest spectral band (direct band to band comparison no spline
interpolation)
 Omitted spectral bands to assess interpolation and extrapolation effect
GSICS GRWG Meeting – March 2011 – Patrice HENRY / CNES
Acquisition geometry error
■ Comparison of same geometry and close geometry calibration
Example of Aqua/MODIS vs MERIS
Same geometry
Close geometry
■ Very important increase of standard deviation (x2 to x10) but small
effect on the mean value (0.5% max.)
But viewing geometry is always the same (Hyperion geometry).
Discrepancies are only due to : sun angles, atmospheric correction, annual
variation of the site
GSICS GRWG Meeting – March 2011 – Patrice HENRY / CNES
Reflectance interpolation error
■ Comparison of spline interpolation and band to band calibration
Example of Landsat/ETM+ vs Terra/MODIS
Spline interpolation
Band to band
■ Increase of cross calibration unaccuracy
■ Increase of site to site discrepancy
Band to band calibration shall be limited to very similar bands (VGT2/VGT1,
Aqua/MODIS vs Terra/MODIS…)
GSICS GRWG Meeting – March 2011 – Patrice HENRY / CNES
Reflectance extrapolation error
■ Comparison of cross calibration with different set of reference band
Example of Aqua/MODIS vs MERIS
Without 412 nm as reference band
With 412 nm as reference band
■ Very important error due to extrapolation (> 20%)
Site reflectance profiles do not allow any extrapolation neither in the blue or
in the SWIR…
GSICS GRWG Meeting – March 2011 – Patrice HENRY / CNES
Sensitivity to atmospheric correction
■ Small change in the spectral response for atmospheric correction
coefficient computation
Parasol/POLDER vs MODIS
Parasol/POLDER vs MERIS
■ Important error induced to atmospheric correction with a
misknowledge of the spectral bands
GSICS GRWG Meeting – March 2011 – Patrice HENRY / CNES
Sensitivity to site reflectance profile
■ Comparison of simulation performed with Hyperion data and with
laboratory characterization of desert sand
Libya 4 profile after spline interpolation
Lab. sand profile after spline interpolation
■ No meaning for any interpolation when ‘high frequencies’ in the site
profile
GSICS GRWG Meeting – March 2011 – Patrice HENRY / CNES
Desert sand laboratory characterization
■ Spectral and directional characterization of sand samples from
Algeria 2, 3 and 4 performed in ONERA
■ Very smooth spectral profile
■ Need to apply atmospheric simulation to get pseudo TOA reflectance
GSICS GRWG Meeting – March 2011 – Patrice HENRY / CNES
Site reflectance impact (1)
■ Calibration comparison using Hyperion or lab. sand data
Example of Aqua/MODIS vs MERIS
Hyperion simulation
Lab. sand simulation
■ Small differences (1 to 2%) in the blue range
But…
GSICS GRWG Meeting – March 2011 – Patrice HENRY / CNES
Site reflectance impact (2)
■ Calibration comparison using Hyperion or lab. sand data
Example of MERIS vs Aqua/MODIS
Hyperion simulation
■ Large differences (> 6%) in the red range !
GSICS GRWG Meeting – March 2011 – Patrice HENRY / CNES
Lab. sand simulation
Hyperion spectral TOA profiles
■ Very similar high frequency shape of desert and Dome C profiles
■ Can be due to :
 Interband calibration unaccuracy
 Bad solar irradiance correction (some correlation in the blue range)
GSICS GRWG Meeting – March 2011 – Patrice HENRY / CNES
Impact of solar irradiance profile
■ Comparison between 2 solar irradiance profiles
Thuillier 2003 solar irradiance model
■ Some differences
 High frequency differences
 Differences in the blue range
GSICS GRWG Meeting – March 2011 – Patrice HENRY / CNES
Hyperion solar irradiance model
Perspectives
■ To perform simulation from Hyperion radiance profiles
 To contact USGS to acquire radiance profiles
 To compute TOA reflectance with different solar irradiance profiles
■ To perform simulation with other multispectral data
 To contact RAL (Dave Smith) to obtain Sciamachy data over same sites
■ To perform a complete error budget depending on
 Sensors to be cross calibrated
 Spectral range
 Objective : cross calibration or calibration monitoring
 …
GSICS GRWG Meeting – March 2011 – Patrice HENRY / CNES