1081 
values transmitted at 
1987)) 
nodel domain. 
1 level due to gaseous 
the radiation transfer 
5 an ter (1991), look-up 
heric parameters were 
evation 720 m a.s.l. . 
trface fully covered by 
le proportion / of the 
”) and the reflectance 
( 2 ) 
velengths A e must be 
erences deal with the 
brest canopies. As an 
the radiation transfer 
l by Frey (1992). The 
tween vegetation and 
;he step in reflectance 
ically 20 - 30 in our 
îannel is channel XS3 
lectance of vegetation 
th the values given in 
ectance of the ground 
»heric corrections and 
d on the basis of two 
d an area sufficiently 
est, however, shows a 
e spatial distribution 
nee, we used a single 
an additional error. 
aving a cross section 
m 2 is 
(3) 
where A t is the ground area. The dust load u of these i particles, defined as the deposited dust mass per 
unit area, amounts to u = »'4f r 3 p p . Combining this equation with Eqn. (3) and neglecting the higher order 
terms of the Taylor expansion of the logarithms involved, the folllowing equation results: 
4 r 3 1 
U ~ ~3 Pp 7* ln(l - /) ’ (4) 
In order to calculate u according to Eqn. (4), the dust particle radius r must be given . In general, r 
is not constant for all particles, but follows a distribution function p(r). Hence, the cross section (depending 
on r 2 ) and the volume (depending on r 3 ) are to be replaced by their expectations. In Eqn. (4), r 2 and r 3 
have to be replaced by their expectations r 2 = r 2 p(r) dr and r 3 = r 3 p(r) dr as well. It is obvious 
from Eqn. (4) that u and / do no t depend on the average radius r but on the characteristic particle radius 
r c which is defined as r c = r 3 / r 2 . 
The particle size spectrum p(r) is known from the experiment only for a limited number of locations. 
Applying the above mentioned stochastic model, the spectrum was simulated for all pixels (x, y). In principle, 
these spectra could be used to calculate r e (x,y). However, the evaluation of the satellite-based dust distribu 
tion u(x, y) could be biased by using the particle size spectra delivered by the stochastic model, complicating 
an independent comparison of the two approaches. Thus, a simplified dispersion model has been devised to 
estimate an unbiased particle size distribution. Due to the influence of the particle’s size and the prevailing 
wind conditions on the deposition process, p(r) = p(r, x,y) turns out to be a function of the coordinates 
(*,!/)• 
4 Results 
Fig. 2 shows the SPOT reflectance values for vegetation and dust, corrected for atmospheric effects. For dust, 
the values in the XS1 and XS3 bands match satisfactorily well with the results measured in the laboratory 
(Frey (1992)). For vegetation, larger difference between these two bands and the typical values used by Tanre 
(1986) are expected. Conversely, there is a strong evidence from this figure as well as from Table 1 that for 
XS2 the reflectance for both, vegetation and dust is too high. The main reason for this effect is probably 
an incorrect calibration coefficient. Indeed, if we state that the on board gain of the HRV instrument is one 
step too high, (i.e. the absolute calibration coefficient Aj is by a factor of 1.3 too low), the reflectances of 
vegetation and dust as detected in the XS2 band decrease to reasonable values given in Fig. 2. 
On the basis of the particle size distribution, p(r, x, y), the characteristic radius r e (x,y) was computed 
for the whole model domain. In Fig. 3 this array is shown as a surface plot. r e increases from about 7 to 
30 pm for decreasing distances from the road. 
Inserting r e (x,y) in Eq. (4) and taking 2.6 y/cm 3 for the particle bulk density p p , the satellite-based 
dust load u(x,y) was computed. Fig. 4 shows ti(x, y) together with the dust load distribution u, <oe *(x,y) 
computed with the stochastic model. 
It is clearly visible that the dust shown on the satellite-based image is dispersed only in the vicinity of 
the road. The distribution shows a similar pattern as in the case of the dust load calculated with the stochastic 
model. However, some artificial features were found which are due to the non-uniform ground cover. The 
model presented in Sec. 3.2 states that the reflectance value p v is independent of the location (x, y). However, 
if the satellite-based reflectance of pure vegetation detected at a given pixel position (x, y) far from the road 
is greater than that value, erroneous values /(x,y) > 0 and v(x,y) > 0 will be found. It appears that such 
variations in p„(x,y) are correlated with orographic features influencing the type of vegetation. Conversely, 
at reflectance values lower than the threshold reflectance p v , meaningless negative dust loads will result. For 
a water surface sensed in band XS3 (e.g. the little lake located close to the highway), having a very low 
reflectance, /(x, y) > 1 is pretended. Finally, from Fig. 4 it is evident that the car frequency on the access 
road connecting Dalton Highway and Toolik Lake is much lower than on the main road. The vehicles are less 
heavy and slower than those traveling on the highway, thus propelling less dust. 
Since the simulation of the dust load u, t0 eh was performed only for 65 consecutive days, the absolute 
ranges of u, loe fc(x,y) and u(x , y) differ. H ence, for the co mparison of the distribu tion shapes, u(x, y) was 
rescaled by the factor ti(x,y)/u, <oe a(x,y). The quantities ti,«oefc(*.v) and ti(x,y) are the respective dust 
loads averaged over a strip of 400 m width along the road, omitting the values at points located on the road 
axis (see also Lamprecht (1993)). The average ratio was found to be 9.3 ± 0.1).