181 
MODEL INVERSION TO RETRIEVE CANOPY CHARACTERISTICS 
FROM HIGH SPECTRAL RESOLUTION DATA. 
F. Baret (1) and S. Jacquemoud (2), B. Andrieu (3), M. Danson (4), K. Jaggard (5). 
(1) INRA Bioclimatologie. BP 91. 84 143 Montfavet Cedex, France 
(2) JRC, IRS A/AT, TP 272, 21020 Ispra (Va), Italy 
(3) INRA Bioclimatologie, Thiverval Grignon 78 850, France 
(4) Dept. Geography, University of Salford. Manchester M5WT, U.K. 
(5) Broom's Bam Experimental Station, Higham, Burry St Edmunds, U.K. 
ABSTRACT 
The PROSPECT leaf optical properties and SAIL canopy reflectance radiative transfer models are coupled and 
inverted using high spectral resolution data gathered in a field experiment on sugar beet plots expressing large 
variations in leaf area index, chlorophyll concentration and soil background optical properties. In a first 
attempt the inversion consisted in retrieving six biophysical variables that give the best fit between the 
measured canopy reflectance spectra and the simulated ones. For most of the 96 plots, the inversions converge 
toward a solution that reproduces accurately the measured canopy reflectance spectra. However, the four 
structural parameters retrieved, (leaf mesophyll structure parameter, leaf area index, mean leaf inclination 
angle and hot spot parameter) indicate unstabibility. Conversely, the two biochemical composition variables 
retrieved (chlorophyll concentration and equivalent water thickness) are reasonably well estimated, except over 
very bright soils. In a second attempt, three of the four structural variables are assigned to a fixed value 
corresponding to the average observed in this experiment. Model inversions to retrieved the remaining 
structural variable (leaf area index) and the 2 biochemical composition variables show large improvements on 
the accuracy of leaf area index, but slightly poorer performances for leaf chlorophyll and water contents. Here 
again, the very bright soils lead to poor results. The compensations observed between leaf area index and leaf 
chlorophyll or water contents lead us to evaluate the performances of a more synthetic variable: the canopy 
chlorophyll or water content. Results show a reasonnable estimation of these quantities. We then compare the 
previous inversions performed on high spectral resolution data (188 narrow bands) to what could be obtained 
using only the 6 broad bands of Landsat Thematic Mapper. Very similar results were found between both 
alternatives, either for the spectra reconstruction capability or the canopy biophysical characteristics retrieval. 
This finding suggests very high degree of redundancy in the spectral information. These results are discussed 
with regards to the limits of this study characterized by a given canopy, the radiative transfer models used, the 
assumptions made on canopy structure and soil optical properties. Further, the minor absorption features 
potentially observable in the middle infrared domain were not explicitely taken into account. 
KEYWORDS: High spectral resolution. Reflectance, Model inversion, Sugar beet, Biophysical characteristics, 
1 - INTRODUCTION 
The last decade has been mainly characterized by the development and the intensive use of empirical or semi- 
empirical approaches to relate simple vegetation indices to single canopy biophysical characteristics such as the 
leaf area index or the fraction of photosynthetically active radiation absorbed by the canopy. Those vegetation 
indices are linear or non linear combinations of the reflectances or radiances observed in several broad 
wavelength bands. However, these empirical relationships are limited by the very little use of the physics of 
radiative transfer in canopies. This restricts the robustness and the portability of the relationships to targets and 
conditions very close to what prevailed during the experiments. Using reflectance models that incorporate our 
current understanding of the physical processes governing radiative transfer in canopies appears as a very 
attractive and alternative approach. However, radiative transfer models compute canopy reflectance as a 
function of canopy biophysical variables, soil reflectance and the variables chracterizing the configuration of 1 
1 Permanent Affiliation: LAMP/OPGC, Université Blaise Pascal, 63 177 Aubière, France