mMHhhmmmbmh 
HM 
We selected 7 steady sites (Table 1), expected to be unvarying seasonally as well as from year to year. 
Table 1 - Positions of the targets 
Latitude 
Longitude 
Surface type 
Argentina 
37IFS 
66.U W 
Frame 
Australia 
20.0 S 
132.0 E 
Sand desert 
Brazil 
0.0 
59.0 W 
Forest 
China 
40.0 N 
85.0 E 
Sand desert 
Kalahari 
23.6 S 
24.0 E 
Stone desert 
Libya 
24.3 N 
13.0 E 
Sand desert 
USÄ 
32.5 N 
106.2 W 
Sand desert 
The signal extracted from the original GVI product was calibrated in visible and near infrared reflectances, using the first 
post-launch available coefficients (from Kaufman and Holben, 1993) and correction for atmospheric effects where 
applied using the SMAC method (Rahman and Dedieu, 1994) with climatological values for water vapor (Oort, 1983) 
and ozone content (London et al., 1976) and a constant aerosol optical depth of 0.05 at 550 nm. Surface temperature was 
derived through a split window formula (Kerr et al., 1993). The mean and standard deviation of those three values were 
also computed over 3x3 pixel windows around selected target. These last quantities were used to filter the signal 
according to a method described in Cabot and Dedieu 1993a. 
It can be seen on figure 1 that the signal we obtain after all these preprocessing is much smoother than the original signal 
but still shows variations of about 10%, partly due to bidirectional effects. 
METEOSAT measurements were extracted from the so-called ISCCP B2 archive, which provide visible and thermal 
infrared channel digital counts from the METEOSAT satellites, subsampled to a spatial resolution of 30 km. One image 
is available every 3 hours. We focused here on METE0SAT4, which entered in operation in June 1989 and is still in 
operation. Digital counts were extracted from this data set over the libyan desert, for two hours (11.30 and 14.30 UT), 
calibrated in reflectances using an fixed value of the calibration factor (Kriebel and Amann, 1991), corrected for 
atmospheric effects, in the same way as NOAA/AVHRR measurements, and filtered for cloud and large optical depth 
contamination. The filtering procedure is described in Dedieu 1993. Filtered reflectances, plotted on figure 2, still show a 
large dispersion, but are principally organized in two main clusters corresponding to the two hours of acquisition. 
0 200 400 600 800 1000 1200 
Day tinea launch 
Figure 1. Surface reflectance over Australian site. 
VIS: unfiltered - filtered, NIR:.. unfiltered -- filtered of the number of day since launch, over Libyan desert 
3. METHOD 
3.1 Gain drift monitoring 
After accounting for atmospheric contamination, one has to consider bidirectional effects. In this study, Rahman et al 
(1993) model was used. The basic assumption is that, if we are able to normalize the reflectances acquired throughout the 
whole period to a reference geometry, since this reflectance should be constant we should see in this normalized signal 
only an evolution due to the gain drift of the instrument. 
In order to achieve this normalization, we retrieved, through a least square computation, parameters of the model, 
performing the inversion on a sliding window. With these retrieved parameters, we computed reflectances for a reference 
geometry (Sim at zenith and nadir viewing). Since the reflectance of the surface is assumed to be constant, and according 
to the following expression for the reflectance: 
na(DN-DN 0 ) 
( 1 ) 
E o cos0 s (d 0 /d) : 
42 
where p is the surface r 
atmospheric solar irradi 
distance, a the gain os 
assumptions that i) atrr 
variations of the calibral 
If we compute a reflects 
we then can write: 
where (0) represents th 
procedure to the two ser 
We assumed that the off: 
GVI data. Kaufman and 
1990. 
3.2 Inter calibration 
To achieve intercalibra 
reflectances. This is doi 
1991, in a METEOSA1 
simulated digital counts 
where a,, and a M rep 
and a 2 are the weights ti 
Eq,, Eo 2 and Eq are the ex 
are the computed digital 
4. RESULTS AND DIS 
In what follows, we com 
value for the calibration 
of the ratio of the calibr; 
the factor of the day of i 
used a calibrated spectn 
operated in order to get 
transformed in satellite 1 
year a desertic area in N 
geometric conditions, ii 
measurements, correctec 
sensor with calibration p 
Arizona (UAZ). 
4.1 Gain drift monitorinj 
The different sites, i 
present in almost all 
authors. Second, son 
a remaining noise d 
atmosphere over the 
can see on the graph 
shortly after tire eru 
beginning of the peri 
Comparison between 
with the highest refl< 
they show very simi