nces, using the first 
teric effects where 
vapor (Oort, 1983) 
ce temperature was 
e three values were 
to filter the signal 
a the original signal 
visible and thermal 
: 30 km. One image 
1989 and is still in 
.30 and 14.30 UT), 
991), corrected for 
large optical depth 
igure 2, still show a 
: acquisition. 
600 700 
ice as a function 
Libyan desert 
;tudy, Rahman el al 
uired throughout the 
is normalized signal 
leters of the model, 
ances for a reference 
istant and according 
0) 
where p is the surface reflectance. DN is the digital count as could be measured by the satellite, E,, represents the exo- 
atmospheric solar irradiance in the considered wavelength band, d/do the ratio of the distance Earth-Sun to the mean 
distance, a the gain of the sensor and DN 0 the offset of the sensor. All this procedure relies on the underlying 
assumptions that i) atmospheric corrections allow to overcome completely the effect of the atmosphere and ii) the 
variations of the calibration factor remain small enough to consider atmospheric effects as linear. 
If we compute a reflectance in a reference geometry, simulate equivalent digital counts using first day calibration factor, 
we then can write: 
<x(J) _ DN(0 )-DN o (2) 
a(0) “ DN (J) -DN 0 
where (0) represents the value of any parameter on the launching day and (J) the value at day J. We applied this 
procedure to the two sensors independently, AVHRR channels 1 and 2 and METEOSAT shortwaves. 
We assumed that the offset of the AVHRR was constant throughout the period since there is no way to retrieve it from the 
GVI data. Kaufman and Holben (1993) estimated offset change from 41.0 to 33.9 between launching date and September 
1990. 
3.2 Inter calibration 
To achieve intercalibration, we have to be able to make comparable the AVHRR reflectances and METEOSAT 
reflectances. This is done by combining the AVHRR visible and near infrared reflectances, according to Arino el al., 
1991, in a METEOSAT equivalent reflectance. After we computed normalized reflectances for the two sensors, and 
simulated digital counts, we then can write: 
_*o 
DN 
a ] (DN, -DN 0 ) a 2 (DN 2 -DN 0 ) 
a, ^ + aj 
*01 
*02 
(3) 
where a„ ctj and a M represent the calibration factor for AVHRR channels 1 and 2, assumed perfect, and METEOSAT, a, 
and a 2 are the weights to apply to the AVHRR reflectances and depending on the spectral responses of the two sensors, 
Eo,, Eto and Eo are the exo-atmospheric solar irradiance for AVHRR channel 1 and 2 and METEOSAT, DN,, DN 2 and DN 
are the computed digital counts for AVHRR channels 1 and 2 and METEOSAT respectively. 
4. RESULTS AND DISCUSSION 
In what follows, we compared our results with those of various authors. But since we have no way to retrieve an absolute 
value for the calibration factor, the comparison is done only in term of drift, that is, we compared the temporal evolution 
of the ratio of the calibration factor at day J and calibration factor at day 0. For each author, we used for factor at day 0 
the factor of the day of the first calibration experiment held after launch. The different sources are: i) Abel el al., 1993, 
used a calibrated spectroradiometer flown over White Sands, New Mexico, which is our USA site. The radiometer was 
operated in order to get the same viewing and illumination geometry as NOAA-AVHRR, computed scene radiance was 
transformed in satellite level radiance using the LOWTRAN-7 code, ii) Kaufman and Holben, 1993. monitored twice a 
year a desertic area in North Africa, asstiming its reflectance remains stable, and paying particular attention to matchin g 
geometric conditions, iii) Che el al., 1991, also used the White Sands site but derived surface reflectance from SPOT 
measurements, corrected for atmospheric contamination. They present two sets of results obtained by calibrating SPOT 
sensor with calibration parameters given by French Centre National dEtudes Spatiales (CNES) and by the University of 
Arizona (UAZ). 
4.1 Gain drift monitoring 
The different sites, even if they show very different evolutions of the gain, exhibit some general features which are 
present in almost all the results (Figure 3). First, the amplitude of the variations falls wi thin the range observed by other 
authors. Second, seme of the sites show a cyclical behavior. Whether this cycle is really a characteristic of the sensor or 
a remaining noise due to atmosphere is hard to decide, principally because of the lack of information regar ding the 
atmosphere over the various sites. Over certain sites, such as USA, this cycle is annual and the most important feature we 
can see on the graphs (Figure 4). On this site, we can also notice the sharp diverging at the end of the period, starting 
shortly after the eruption of Mt Pinatubo. At last certain sites, such as Libyan desert, show good agreement at the 
beginning of the period but tend to diverge towards the end (Figure 5). 
Comparison between the different sites was also done and particularly between China and Libya, which are the two sites 
with the highest reflectances. The calibration factor drift of these two sites is plotted on figure 6, and it is obvious that 
they show very similar behavior. When examining these results, one should keep in min d that these two sites are very 
43