,28. 
VHRR- 
adiance 
■vations 
version 
surface 
I albedo 
se from 
le solar 
rom the 
out the 
late the 
ange of 
osphere 
, 13, 17 
le most 
ss were 
2, 1991 
:tion of 
ged in a 
h angle, 
to June, 
•y out a 
ad-band 
oretical 
afficient 
usually 
nee into 
an since 
ow- and 
,3 to 2.5 
; against 
ects due 
:d in the 
i infinite 
;rve the 
incident 
i almost 
m of the 
ings are 
For the conversion between NOAA and METEOSAT signals, it would be possible to carry out 
different regressions between narrow- and broad-band albedos for each channel and surface type. A better 
approach consists in making up a multilinear regression between the narrow-band channels 1 and 2 of 
NOAA and the broad-band visible of METEOSAT, employing the data obtained from the simulation. This 
permits to find a linear combination of AVHRR-1 and -2 albedos to estimate the signal which METEOSAT 
would register independently of the surface type. Applied to TOA albedos, this linear combination is 
«met = 0.440 (Xavhrr-i + 0.529 oc A vhrr-2 
(la) 
(R 2 = 0.983 and standard error = 4.5%) 
Inclusion of a constant only improves the correlation by 1% and the regression gives 
«met = 0.425 cxavhrr-1 + 0.474 cx A vhrr-2 + 0.018 
(lb) 
(R 2 = 0.989 and standard error = 3.5%) 
The effect of the atmosphere may be eliminated by correlating NOAA and METEOSAT albedos at 
ground level. The result then gives 
«met = 0.459 (Xavhrr-i + 0.481 ocavhrr-2 
( 2 ) 
(R 2 = 0.999 and standard error = 1.2%) 
which shows an almost perfect correlation. 
As for the simulation at ground level of broad-band albedo (0.3-2.5 pm), higher precision can be 
reached by means of both NOAA channels rather than by the METEOSAT one. This is shown by testing the 
ability of each alternative with their respective filter combination to represent a square broad band (0.3-2.5 
pm) without considering atmospheric effects. Eq. (3) refers to METEOSAT and (4) to NOAA-11. 
«BROAD-BAND = 0.974 OtMET 
(3) 
(R 2 = 0.938 and standard error = 7.3%) 
«BROAD-BAND = 0.545 «AVHRR-1 + 0.320 «AVHRR-2 + 0.035 
(4) 
(R 2 = 0.982 and standard error = 4.4%) 
It would then be preferable to use NOAA data rather than METEOSAT's for calculating broad-band 
albedos at ground level. It has not yet been checked, however, whether the previous atmospheric correction 
of the images would contain a greater source of error for NOAA, due fundamentally to large nadir angles. 
The results of this simulation were applied to actual satellite images. A data set consisting of a 
METEOSAT-4 image and a NOAA-11 pass over the Iberian Peninsula was available for a practical test of 
the TOA NOAA-METEOSAT spectral conversion obtained through theory. The images corresponded to 
June 12,1991 approximately at 14:45 GMT for NOAA-11 and at 15:00 GMT for METEOSAT^. The three 
images were geometrically corrected to a polar stcreographic projection. The METEOSAT one was 
remapped using a polynomial fit with control points and resampled through cubic convolution to a latitude- 
longitude projection. Pixel length was upgraded to the same resolution as NOAA's. Every pixel was divided 
into four, and the digital count for every one was calculated by taking into account its former value and the 
near neighbouring pixels. The AVHRR channels 1 and 2 images were remapped by utilizing a refined 
orbital model with control points (Moreno and Melia, 1993a). The final result of all these spatial corrections 
was three matching images on a polar stereographic projection. 
Eq. (lb) for the TOA NOAA-METEOSAT conversion was applied for albedos of the AVHRR ch. 1 
and 2. This predicted METEOSAT albedo was plotted in the scattergram of Fig. 4 versus the actual albedo 
obtained from the METEOSAT image. The cluster of points in this figure distinguishes three groups. The 
cluster near the vortex corresponds to water surface (about 35% of the pixels), those close to the center are 
land surfaces (about 55% of the total), and the points in the halo coincide with cloud movements on the 
images (about 10%). In Fig. 5 now, where water surface and clouds have been eliminated, the different 
colours represent the deviation between the actual METEOSAT albedo and the predicted one via the 
AVHRR channels 1 and 2. The results of the comparison between both albedos show a rise in the errors for 
the METEOSAT albedo estimation due to the strong influence of anisotropic reflectance in the atmosphere. 
The image comparison reveals how the NOAA-derived albedos increase across the viewing track from west 
323