Figure 2 : Odd ISM spectral responses of pine canopies at different ages (5/11/36) and crops (maize). Two
detectors around 2 pm did not work correctly.
* Atmospheric corrections :
The use of the 5S atmospheric model (Tanrd et al., 1990) showed clearly that atmospheric effects in the middle
infrared should not be neglected. Atmospheric influences on vegetation reflectances amounted up to 6 % in the
[1500nm - 1700nm] band and up to 10% in the [2100nm - 2300nm], depending on the wavelength, the
reflectance of the target and the reflectance of the neighbourhood of this target. Moreover, these effects
amounted up to 80% (dense dark vegetation) in the visible and 50% (dark surface) in the near infrared
(Zagolski, 1994). Consequently, all AVIRIS images were corrected with a specifically designed methodology
(Zagolski, 1994), that is based on the inversion of the atmospheric model 5S, through an iterative approach
using the Gauss Seidel principle. For all spectral bands, convergence was attained after less than 5 iterations.
The influence of the environment of each pixel is taken into account. It is determined through circular
neighbourhoods the radii (10 to 60) of which are variable with wavelength and atmospheric conditions.
The aerosol content of the atmosphere is directly derived from spectral bands in the blue and red regions. This
methodology, derived from Kaufman and Sendra (1988), relies on the reflectance of dense dark vegetation.
However, no hypothesis is made concerning their values. They are computed with an iterative approach. The
aerosol optical depth displayed an important spatial variation (0.16 to 0.26) throughout the AVIRIS channel 7
(460nm); its mean value was 0.21. The associated absolute reflectance variation was 1.5%.
Figure 3 shows raw and atmospherically corrected AVIRIS reflectance spectra of some land units. Opposite
(NIR) and additional (MIR) effects of canopy structure and LAI appear clearly. NIR reflectance of trees
decreases with their age; i.e. structure effects dominate the effect of LAI increases. In the MIR region the
decrease of reflectance with age is even stronger because LAI and structure increases have similar effects. This
is clearly shown with bare soil, pine (lyear), i.e. low cover regrowth, and grass. The reflectances of these land
units, with similar structures, display opposite variations in the NIR and MIR regions. Moreover, parcels with
23 years old trees, i.e. maximum structure effect, display the smaller reflectances in the NIR and MIR regions.
II.3 Field sampling
A total of 117 samples, mainly pine needles, were collected in the field in 13 parcels throughout the study
area, within one week of AVIRIS data acquisition. Each sample consisted of several needles from different
canopy positions, from top to bottom. The samples were refrigerated in the field prior to laboratory spectral
and chemical analyses of lignin, nitrogen and cellulose content. Because all pine trees within one parcel had
the same age and had grown under the same conditions, it was assumed that the average values of 9 samples
collected from a parcel were representative for this parcel. Chemical analyses validated this assumption: on the
average, the range of the concentrations of chemical compounds, per parcel, was less than 4% of the mean
concentration value of this parcel. Moreover, it was less than 10% of the range variability between all parcels.
Sampling was based on the assumption that the wide range of ages (0 to 50 years) and growing conditions
between the parcels would correspond to variations of nitrogen, lignin and cellulose content large enough for
authorising statistical analyses. This was verified (Table 1): chemical concentrations in the various parcels
have a range variability that is larger than 50%. The strong negative correlation (r=-0.71%) between nitrogen
concentration and the age of the sampled pine trees is explained by the fact that needles of young trees have
larger nitrogen content than older trees.
263
