004
ust
1se
ind
ed
em
yll
et
an
ce
re-
car
the
the
est
nd
ed
lex
ed,
4)
International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part B7. Istanbul 2004
TSAVI=+ ldo PIRTAP se] =
((Pr+aprir —ab+0,08{1+a?))]
(hir p)
(O'buc- p)
s bina, * =>) )
TSARVI=+— la pha A, Pre—Orp x
lp hu - a pg, +0,08(1+a2, )|
ARVI=
With: Dbin 7 5, Pg D,
Pre = PrT ye; 7 £i]
f oc
Fs |o... = pu |
GEMI=m(1-0,25 pri)
R
With: 7 2(0 up; +15 Pme+0.501]
(Ohr + pP à+0,5)
L* Dp 60g —1,5 05
RDVI Um.
1) PPIR+ PR
Prir =]
MIRA.
Pr
ip e eu | : et Alv; — Dn .J
GARI =
= lo For T ven BU Pr Ji
Where:
pr: ground reflectance in the red channel,
Ppir : ground reflectance in the near infrared red channel,
pp: ground reflectance in the blue channel,
Pn: apparent reflectance in the blue channel,
pv: apparent reflectance in green channel,
Pr: apparent reflectance in the green channel,
Bron : apparent reflectance in the near infrared channel,
Dan : apparent reflectance in the red-blue channel,
Par: atmospheric reflectance in the red channel,
Ap: atmospheric reflectance in the blue channel,
Y : atmospheric self-correction factor,
(6)
(7)
(8)
(9)
(10)
(11)
(12)
a and b : slope and ordinate at the origin of the bare soil in the
red/NIR spectral space,
ap and by, : slope and ordinate at the origin of the bare soil in
the red-bleu/NIR apparent spectral space,
L : soil adjustment factor, equal to 0.5.
N
803
3. Results and Discussions
Although atmospheric ozone component absorb the
electromagnetic radiation in certain wavelengths, the design
and the conceptualization of vegetation indices never
considers this effect. Considering the channels of MODIS,
VEGETATION and MODIS sensors, the reflectance analysis
shows that the ozone absorption decreases the reflectance in
the near infrared in the same way as the water vapour
absorption. However, when the ozone concentration increases
in the atmosphere, the values of the vegetation indices
decrease. This effect cannot be neglected especially for the
indices derived from the broadband AVHRR sensor.
Considering different situations, different ozone concentration
and different vegetation indices, the figure 1 illustrate the
sensitivity of each vegetation index to the ozone absorption
and potential of MODIS sensor to normalize very well the
ozone absorption effect.
The obtained results show us that the MSR is most sensitive
index to the ozone absorption. Compared to ground truth, the
relative error is approximately 23% for AVHRR, 20 % for
VEGETATION and 16% for MODIS. Moreover, this index is
the most sensitive to the atmospheric diffusion and water
vapour absorption (Bannari er al, 2000). Consequently, the
MSR cannot be used for temporal change detection of the
forest cover and land use without atmospheric correction.
The EVI sensitivity to the ozone absorption is much lower in
comparison with its sensitivity to the water vapour absorption
or the aerosols diffusion (Asalhi, 2003). For an extreme ozone
concentration (UO3 = 0,959 g.cm3), the relative error on this
index is 19 % for VEGETATION and MODIS sensors.
Furthermore, the GARI behaviour varies largely according to
the value of the atmospheric self-correction coefficient and
the density of vegetation cover. This index is so resistant to
the ozone absorption when it is calculated with a self
correction coefficient equal to 0.5 and a very dense vegetation
cover, the error is approximately 7 %. However, when the
vegetation cover decreases, the relative error increases to 15%
for a bare soil.
When we consider a dense forest cover with an extreme ozone
concentration (UO3 — 0,959 g.cm3), the relative error on the
PVI reaches 15 % for AVHRR, 8 % for VEGETATION and it
does not exceed 4% for MODIS. Contrary to the GARI, when
the vegetation becomes sparse, the relative error decreases
drastically on the PVI. Thus, for a 50 % forest cover Sen
the error on the PVI reaches 8 % for AVHRR, 5 % for
VEGETATION, whereas the error remains lower than 1 % for
MODIS.
The MSAVI shows a very good resistance to the ozone
concentration variations particularly for MODIS sensor.
Indeed, considering the broadband AVHRR sensor, the error
on the MSAVI is 14% when the cover is very dense and is 7%
for a moderate density. If we consider these two cover rates,
the relative error on this index is 10 % and 5% for
VEGETATION. The spectral resolution of MODIS sensor
offers a better precision; the relative error is 6-% for a very
dense cover and 2 % for a fairly dense cover.
For the indices that adjusted to the bare soils (SAVI, TSAVI,
OSAVI and RDVI), they show the same sensitivity to the
ozone concentration variations. This sensitivity decreases
significantly when we use the MODIS spectral resolution.
Considering a very dense forest cover, for the TSAVI, the
