POLDER 2 IN-FLIGHT CALIBRATION AND PERFORMANCES

During the POLDER-2 Calibration phase, ended at the end of 2003, a large set of vicarious methods, mainly historically developed for POLDER-1 and improved, has been used to characterize in-flight radiometric and geometric performances.

Radiometric performances

The POLDER instrument has no internal calibration source for in-flight on board calibration. However calibration is ensured by a calibration approach based on both pre-flight modeling and observation of specific references of Earth targets. These targets present well-known reflectance signatures: glitter, desert areas, Rayleigh scattering, clouds...

1/ Absolute calibration

POLDER 2 absolute calibration is achieved through an absolute calibration of shorter spectral bands (443nm Polarized , 443nm, 490nm, 565nm, and 670nm) based on the observation of Rayleigh scattering over well-characterized oceanic sites. This absolute calibration is then transferred to other longer wavelengths through inter-band calibration using the specular reflection of the sun over the ocean (sunglint). The absolute calibration derived through these two nominal methods has been validated using inter-band calibration over white convective clouds and cross-calibration with other space sensors (POLDER-1 and MODIS) over stable and well-characterized desert sites. The Table 1. presents all results derived from the various methods.

	443P	443	490	565	670	763	765	865	910
Pre-flight	1	1	1	1	1	1	1	1	1
Rayleigh	0.901	0.902	0.970	0.990	0.995
Glitter	0.927	0.928	0.960	0.962	-	0.987	0.987	0.980
Clouds	0.948	0.947	0.971	0.965	-		0.981	0.980
Deserts	0.949	0.951	0.985	0.984	0.995		0.982	0.978
Ak	0.937	0.938	0.970	0.990	0.995	0.987	0.987	0.980	0.980
Table 1. In-flight absolute calibration coefficient (referring to the pre-fight values) for the 4 main calibration methods. The last line represents calibration coefficients considered for the level-1 data processing.

It clearly appears that the calibration coefficient derived for 443nm depends on the observed target while a very good consistency between the different methods is observed within 2-2.8% for all other wavelengths. This spectral dependence for 443 has not yet been explained and is still under investigation.

Important remark: The adopted calibration coefficient is optimized at 443 for spectrally flat observations such as glitter or cloud and consequently, an adjustment using the calibration coefficient derived over Rayleigh acquisitions is required for ocean color or aerosol detection over ocean applications. Consequently, for such applications, level-1 TOA reflectances at 443P and 443NP have to be adapted and multiplied by a factor of 1.040 for a better accuracy.

A light multi-temporal decrease of the radiometric sensitivity of about 0.1-0.2% per month (depending on the spectral band) has been clearly detected by all the methods and corrected into the level-1 data-processing.

The final in-flight accuracy is estimated to about 2% for the absolute calibration and 1% for the inter-band calibration for all the spectral bands (except for the 443 nm spectral band for which this accuracy can be considered only if the absolute calibration is adapted to the spectral behavior of the observed target as mentioned above).

2/ Other radiometric parameters

Other radiometric parameters are necessary to optimally process the level-1 data. In particular, multi-angular and in-polarization characterization, which are the main interests of POLDER instruments, must be in-flight checked and possibly adjusted. The following parameters, defined in [1], have been checked using various methods based on the observation of natural targets, the accuracy of results being of course dependent on the method:

low frequency multi-angular response: no evolution detected using absolute and inter-band methods- checked within 2%
medium frequency multi-angular response: light evolution detected and corrected for 443 spectral bands using various natural target (desert, Rayleigh, Antarctica...) - no evolution detected for other spectral bands - checked within 1%
high-frequency multi-angular response: improved in-flight estimation using statistical acquisitions over bright targets - accuracy better than 0.2%
polarization sensitivity of the wide field-of-view optic: no evolution detected over un-polarized targets - checked for 443P, 670P and 865P within 0.5%
transmission of polarizers: light evolution of about 1% detected at 443P, no evolution detected for 670 and 865 - checked within 0.5%
non-linearity of the instrumental response: no-evolution detected over desert sites - checked within 0.4%

Parameters for which evolution has been detected have been updated and corrected on the level-1 data processing.

Geometrical performances

1/ Geometric calibration

The geometric calibration consists in verifying and possibly adjusting the rotation matrix between imaging frames defined by the optics and the CCD matrix, used to geometrically resample images on the POLDER level-1 grid. Correlation between two images from the same orbit viewing a unique point at the Earth surface, were performed to assess biases when using the initial matrix derived from the pre-flight calibration. Various orbits were used at different date and cycle-orbit to derive a precise updated estimation of the in-flight geometric calibration. It was necessary to derive two sets of calibration depending on the use (or not) of GPS to restitute the satellite's attitude. Angular biases, referring to pre-flight initial calibration, are :

Bias in micro-radian	with GPS	with no GPS
roll	+1517	+1777
pitch	+1093	+608
yaw	+862	+1205
Table 2. In-flight correction of the geometric calibration.

Note that one micro-radian roughly corresponds to 1 meter at the ground surface. No significant evolution correlated with date and latitude was detected. Some marginal orbits during which GPS switched on or switched off, may have some slightly degraded performances. This updated calibration is necessary regarding to the geometric performances (see following section) that are sensitively impacted.

2/ Geometric performances

The geometric performance is described through 5 indicators:

the absolute localization accuracy: it represents the distance between the estimated position of a given point on the level-1 POLDER grid and its reel position. The localization performance was estimated, using Ground Control Points and after geometric re-calibration of the instrument, at 0.64 pixel Max, while the specification requires 1 pixel Max.
the multi-polarization registration accuracy: it represents the radius of the circle containing the 3 measurements corresponding to acquisitions through the 3 orientations of the polarizer for a given spectral band (among 443, 670, and 865). The specification requires a radius of 0.05 pixel Max. The multi-polarization mean performance was estimated through a correlation algorithm at 0.05 pixel at 443, 0.04 pixel at 670, and 0.025 pixel at 865. Consequently, regarding to the specification criteria (Max), the specification is always met at 670 and 865, and is met for satellite incident angles under 35° at 443 (i.e. up to viewing angle of 40°).
the multi-spectral registration accuracy: it represents the radius of the circle containing the 9 measurements corresponding to acquisitions through the 9 spectral bands (one turn of the filter rotating wheel). The specification requires a radius of 0.10 pixel Max. The multi-spectral performance was estimated through a correlation algorithm at less than 0.10 pixel Max and the specification is met.
the multi-angular registration accuracy: it represents the radius of the circle containing the (up to) 13 measurements corresponding to acquisitions of the 13 viewing directions (13 turns of the rotating wheels). The specification requires a radius of 0.10 pixel RMS. The multi-spectral performance, was estimated through a correlation algorithm and after geometric re-calibration of the instrument, at less than 0.10 pixel RMS and the specification is met.
the multi-temporal registration accuracy: it represents the radius of the circle containing all measurements acquired during one month of acquisitions. The specification requires a radius of 0.20 pixel RMS. Due to its definition, this total multi-temporal performance includes a multi-angular part and a purely multi-temporal part. Only the multi-temporal part was estimated after geometric re-calibration of the instrument at less then 0.07 pixel RMS. The total multi-temporal performance was consequently not strictly estimated but we can consider that this specification should be met.

Conclusion

Considering the image quality performances achieved, the calibration review panel decided in December 2003 to authorize the start of POLDER2 level 1 distribution to the users. The systematic processing of the 7 months of data available was achieved in February, 2004.

References

[1] : Hagolle O., P. Goloub, P.-Y. Deschamps, H. Cosnefroy, X. Briottet, T. Bailleul, J.M.Nicolas, F. Parol, B. Lafrance, and M. Herman, "Results of POLDER In-Flight Calibration," IEEE Transactions on Geoscience and Remote Sensing, vol. 37, pp. 1550-1566, 1999.