261 
Forest Canopy Chemistry with High Spectral Resolution 
- An evaluation of the atmospheric influence - 
Gastellu-Etchegorry J.P. (I) , Zagolski F. (1) , Mougin E. (1) , Marty G. (l) , Dubuisson P. <2) 
(1) Centre d'Etude Spatiale Des Rayonnements 
C.N.R.S. - U.P.S., BP 4346, 31029 - Toulouse cdx. 
(2) Laboratoire d'Optique Atmosphérique 
C.N.R.S. - U.S.T.L., 59655 - Villeneuve d'Ascq cdx. 
Tel.: 61-55-61-30 ; Fax: 61-55-67-01 
Abstract: Canopy chemistry of a pine forest (The Landes, France) was analysed with airborne (AVIRIS, ISM) 
and laboratory MIR spectrometry. Simultaneously with airborne acquisition, foliar samples were collected for 
laboratory chemical and spectral analyses. Stepwise regressions led to predictive relationships between 
chemical concentrations (nitrogen, lignin, cellulose,....) and reflectances. Predicted and actual concentrations 
were well correlated (nitrogen, r=0.97; lignin, r=0.89; cellulose, r=0.83). Then, it was investigated the 
possibility to extrapolate the laboratory-determined predictive relationships to the remote sensing level. AVIRIS 
data were atmospherically corrected with a specifically designed iterative procedure that inverts the 5S 
atmospheric model. Adjacency effects are fully considered with variable circular neighbourhoods. The aerosol 
optical depth spatial distribution is retrieved with blue and red bands. MIR AVIRIS reflectances were poorly 
correlated with chemical concentrations. Predictive relationships led to larger correlations (nitrogen, 0.74; 
cellulose, 0.79; lignin, 0.55). In order to improve results two approaches that consider the canopy structure 
were used. (1) Chemical concentrations were weighted with local biomass and LAI; poorer results were 
obtained. (2) Foliar reflectances, derived from the inversion of two canopy reflectance models, were used 
instead of canopy reflectances. Results w-ere encouraging: lignin (0.74), nitrogen (0.70) and cellulose (0.69). It 
shows that airborne spectrometry may be an interesting tool for forest canopy chemistry analyses. 
Key words: spectrometry, chemistry, forest, AVIRIS, ISM. 
I. Introduction 
Knowledge of foliar chemical content is often essential for describing and modelling forest processes such as 
net primary productivity, using chlorophyll and nitrogen content (Vitousek, 1982), the rate of litter 
decomposition, using litter lignin to nitrogen ratio (Meentemeyer, 1978; Melilloetal., 1982), and the cycling 
and availability of nutrients such as nitrogen (B irk and Matson, 1987). Commonly, this information is obtained 
through chemical analyses. These analyses, however, are generally destructive, costly and time-consuming. 
Therefore, alternative methods, such as spectrometric measurements, are pursued. In particular, the [0.4|im - 
2.5pm] spectral region that is compatible with optical remote sensing is being studied. The presence of infrared 
absorption bands in target compounds is at the basis of this high spectral resolution technology. 
Today, the accuracy and repeatability of reflectance estimates of protein, lignin and starch content in dried 
plant materials are comparable to those obtained through wet laboratory methods (Curran, 1989). The question 
arises, however, whether the chemical composition of vegetation can be determined through remote sensing. 
Should this technique provide accurate estimates of foliar chemical content from space, it would be a powerful 
tool for describing and modelling ecosystem processes and nutrient cycling on a local to a global scale 
(Wessman, 1990). However, due to factors such as atmosphere, instrumentation and canopy structure, it may 
be difficult to link remotely acquired spectrometric information to canopy chemistry (Wessman et al., 1989). 
A methodology (Figure 1) was designed for assessing airborne spectrometry for determining the concentration 
of chemical compounds (lignin, nitrogen, cellulose) in a forest cover. It relies on spectrometric analyses at two 
levels: dried ground needles (i.e. laboratory analyses), and canopy (i.e. airborne data) with contributions from 
needles, canopy structure and soil. Objectives were two fold: (1) computation, with stepwise regression, of 
relationships between spectrometric information and concentrations of chemical compounds, and ( 2 ) analysing 
the "stability" of these relationships, when one goes from laboratory to remotely acquired information, with the 
aid of model simulation and biophysical information (Leaf Area Index, height, age, biomass).