1080 
date 
A\ A] A 3 
06/21/87 (transmitted) 
03/20/87 (calibration) 
09/20/87 (calibration) 
0.504 0.361 0.525 
0.510 0.363 0.563 
0.496 0.353 0.554 
Table 1: Calibration coefficient At of the HRV1 multispectral bands XS k. Parameter values transmitted at 
overpass time and values calibrated on March 20 and on September 20, 1987 (SPOT (1987)) 
b) Geometric correction 
code 5S originally developed by Tanre et id. (1986, 1990) and modified by Teillet and Santer (1991), look-up 
tables were generated to correct the apparent reflectances to ground values. The atmospheric parameters were 
set as follows: subarctic summer, continental aerosol model, visibility 23 km, terrain elevation 720 ma.s.l. . 
d) Correlation between ground reflectance p and dust area proportion / 
dust) is sensed by the HRV instrument as if it would have an average reflectance p. The proportion / of the 
ground area covered by the dust particles (in the following denoted as "area proportion”) and the reflectance 
p are assumed to be linearly correlated: 
radiation properties of vegetation in general (e.g. Guyot (1989) or with agricultural or forest canopies. As an 
example, Fig. 2 shows the function p v (A) typical for vegetation in general as used by the radiation transfer 
code 5S (Tanre et al. (1986)). 
results are shown in Fig. 2 as well. 
Bearing this typical reflectance data in mind, it is evident that the contrast between vegetation and 
dust is low for wavelengths below the well known red edge of the vegetation curve, i.e. the step in reflectance 
at about 0.7 pm. This effect is particularly adverse at low values of digital counts (typically 20 — 30 in our 
scene) where the digital resolution is poor. Thus, we expect that the most promising channel is channel XS3 
where the effective center wavelength is greater than the red edge wavelength and the reflectance of vegetation 
is relatively high. This assumption is supported by an error estimation. 
In practice, the remotely sensed ground reflectance will hardly agree exactly with the values given in 
the literature, due to variations of the tundra vegetation properties, non-lambertian reflectance of the ground 
cover, inaccurate calibration coefficients, incomplete meteorological data for the atmospheric corrections and 
so on. Thus, the reflectances of pure dust, pj, and pure vegetation, p., were determined on the basis of two 
typical training areas visible on the satellite image: a large parking area (for dust) and an area sufficiently 
far from the road (for pure vegetation). The vegetation reflectance of the area of interest, however, shows a 
spatial variability. Accurate values of the area proportion / can only be derived if the spatial distribution 
of the reflectance for unpolluted conditions is known. Obviously, this is impossible. Hence, we used a single 
value for p„ for the entire domain, being aware that this lack of information introduces an additional error. 
e) Correlation between dust area proportion / and dust load ti 
Let us assume the dust particles as spheres of radius r and bulk density p f , hence having a cross section 
The image was rotated and shifted using ground control points in order to match the model domain. 
c) Atmospheric correction 
The apparent reflectance p' may substantially differ from the actual value p at ground level due to gaseous 
absorption and molecular and aerosol scattering of radiation in the atmosphere. Using the radiation transfer 
A fully vegetated surface (reflectance p„) partly coated with dust (reflectance pj of a surface fully covered by 
( 2 ) 
In order to compute /, the reflectances p„ and pd at the effective center wavelengths A e must be 
known. Data for p„ of low tundra vegetation is difficult to be found. Most of the references deal with the 
As regards the dust reflectance pd (A), a sample of the actual dust was analyzed by Frey (1992). The 
A p = ir s and a volume V p = | rr 3 . The area proportion / covered by t particles per m 3 is 
( 3 )