182 
the measurement. To retrieve canopy biophysical characteristics from the corresponding radiometric response, 
the radiative transfer models have to be inverted. 
Inverting a model consists in determining the variables of the model describing canopy 
architecture and the optical properties of the elements by minimizing the distance between the measured and 
the simulated canopy reflectance. In plant canopy studies, efforts have been mainly concentrated on the use of 
the directional reflectance measurements. Goel and Thompson (1984a, b) or Deering et al. (1992) retrieved the 
leaf area index and less accurately the mean leaf angle by inverting respectively the SAIL (Verhoef, 1984, 
1985) and TRIM (Goel, 1988) models on directional data. However, the knowledge of the leaf reflectance and 
transmittance, soil reflectance, and fraction of diffuse skylight was requested. Assuming canopy structure as 
known, Otterman (1987) inverted a geometrical model to estimate the leaf reflectance for several wavelength 
bands and zenith view angles. Later, the same author tried to infer also information about the leaf orientation 
(Otterman, 1990). He found difficulties in separating leaf reflectance from leaf orientation or leaf area index. 
Pinty et al. (1990, 1991), Kuusk (1991) estimated the leaf optical properties as well as the spatial distribution of 
scatterers in the canopy by inverting analytical models of bidirectional reflectance. All these studies have been 
performed using data acquired at ground level because such directional measurements cannot be easily 
collected at the same date by current satellite sensors. Multitemporal data may be used but this implies that the 
target does not change from one measurement to another and that we are capable of correcting the atmospheric 
disturbances with great accuracy. 
The use of the spectral variation rather than the directionnal one was concurrent to the 
development of high spectral resolution sensors A spectral matching technique has been applied by Gao and 
Goetz (1990) with a very simple model to the 1.5-1.74 pm region. These authors asserted that the vegetation 
spectrum in this spectral region was driven by liquid water and dry vegetation spectral components. Schmuck et 
al. (1993) described the vegetation reflectance with a Kubelka-Munk formula containing the chlorophyll and 
water specific absorption coefficients. This model was mainly used to evaluate the magnitude of the 1.7 pm 
residuals which are thought to be a signature of canopy biochemical composition. In a theoretical study based 
on simulated spectra, Jacquemoud (1993) tried to evaluate the potential of high spectral resolution data to 
estimate canopy biophysical properties such as leaf mesophyll structure, chlorophyll concentration, water 
equivalent thickness, leaf area index, and mean leaf inclination angle by inversion of the PROSPECT 
(Jacquemoud and Baret, 1990) leaf optical properties model coupled with the SAIL (Verhoef, 1984, 1985) 
canopy reflectance model. The potentials and limits of such an approach have been discussed: chlorophylls and 
water seemed to be attainable with a good accuracy but difficulties were encountered in separating the leaf area 
index from the leaf orientation. Kuusk (1993) was the first to attempt to invert his fast canopy reflectance 
(FCR) model both on directional and spectral data. Results obtained on soybean and com crops are very 
promising. They offer new prospects to interpret remote sensing measurements acquired by current satellites. 
It appears that inverting a canopy reflectance model from remote sensing data is not such an 
easy task. An operational use of this approach requires a compromise between simple models that cannot take 
into account the complexity of canopy architecture, and complex models whose inversion requires many 
parameters to be adjusted. It is very time consuming and most often leads to unstable solutions. The acquisition 
of a set of reflectance data large enough that describes the bidirectional and spectral features of canopies is not 
yet posible from current sensors. Thus, the radiometric information to invert models is often restricted. The 
measurement of the spectral variation of canopy reflectance at a given time and location is, from a 
technological point of view, the easiest way to get a radiometric information from a satellite or airborne 
platform. That is the reason why this paper will mainly focuss on the use of the spectral inforrmation. 
We will evaluate the possibility of infering some canopy biophysical variables by invertion of 
a simple radiative transfer model using canopy reflectance spectra acquired in the field with a high spectral 
resolution radiometer. One of the basic critics about the use of high spectral resolution over the whole optical 
domain is the high degree of redundancy observed between the many narrow and adjacent bands available. This 
was analysed statistically in several papers by Price (1990, 1992) who concluded that a limited set of wavebands 
was enough to synthetise with a very good accuracy all the spectral information. In this section, we will adress 
the problem in a different way, by comparing the results of the inversion process performed with the full 
spectral information to what is obtained with only a selection of broad bands. Looking for the optimal number 
and position of bands necessary to achieve the best retrieval performances of some canopy characteristics, will 
certainly be the best way to answer the question. However, this requires more work which is beyond the aim of 
this paper. Here, we will simply compare the performances of the inversion when using a more restricted 
number of wavebands such as the 6 broad bands of Landsat Thematic Mapper (TM). We will first briefly 
describe the data set on which this study is based. Then, we will justify the choice of the models and the 
inversion procedure used.