Informations et ressources scientifiques
sur le développement des zones arides et semi-arides

Accueil du site → Doctorat → Pays-Bas → 2008 → Hyperspectral remote sensing of vegetation parameters using statistical and physical models

Wageningen University (2008)

Hyperspectral remote sensing of vegetation parameters using statistical and physical models

Darvishzadeh, R.

Titre : Hyperspectral remote sensing of vegetation parameters using statistical and physical models

Auteur : Darvishzadeh, R.

Université de soutenance : Wageningen University

Grade : PhD thesis 2008

Accurate quantitative estimation of vegetation biochemical and biophysical characteristics is necessary for a large variety of agricultural, ecological, and meteorological applications. Remote sensing, because of its global coverage, repetitiveness, and non-destructive and relatively cheap characterization of land surfaces, has been recognized as a reliable method and a practical means of estimating various biophysical and biochemical vegetation variables. The advent of hyperspectral remote sensing has offered possibilities for measuring specific vegetation variables that were difficult to measure using conventional multi-spectral sensors. Utilizing hyperspectral measurements, we examined the performance of different statistical techniques such as univariate versus multivariate techniques for predicting biophysical and biochemical vegetation characteristics such as leaf area index (LAI) and chlorophyll content. The study further investigated and compared the performance of the statistical approach with that of the physical approach for mapping and predicting these vegetation characteristics. From the laboratory up to airborne levels, the investigation involved structurally different vegetation canopies and heterogeneous fields with different vegetation communities. It was concluded that the red edge inflection point (REIP) is not an appropriate variable to be considered for LAI estimations at canopy level, especially if several contrasting species are pooled together or a heterogeneous canopy is being investigated. However, it may be appropriate for single species. Throughout this study, the bands in the shortwave infrared (SWIR) region have appeared to make a sound contribution in terms of the strength of relationships between spectral reflectance and LAI. Considering that the SWIR bands were important in all three investigated levels and for most vegetation indices in this study, vegetation indices that do not include this spectral region may be less satisfactory for LAI estimation. The results suggest that, when using remote sensing vegetation indices for LAI estimation, not only is the choice of vegetation index of importance but also prior knowledge of plant architecture and soil background. Hence, some kind of landscape stratification is required before using hyperspectral imagery for large-scale mapping of biophysical vegetation variables. Furthermore, the study results highlight the significance of using multivariate techniques such as partial least squares regression rather than univariate methods such as vegetation indices for providing enhanced estimates of heterogeneous grass canopy characteristics. The newly introduced subset selection algorithm based on average absolute error (AAE) indicated that a carefully selected spectral subset contains adequate information for a successful model inversion. The results of the study demonstrated that, through the inversion of a radiative transfer model, grass canopy characteristics such as LAI and canopy chlorophyll content can be estimated with accuracies comparable to those of statistical approaches. Given thatthe accuracies obtained through the inversion of a radiative transfer model were comparable to those of statistical approaches, and considering the lack of robustness and transferability of statistical models for varying environmental conditions (Asner et al, 2003 ; Gobron et al., 1997), the radiative transfer models may be considered proper alternatives. In summary, the study contributes to the field of information extraction from hyperspectral measurements and enhances our understanding of vegetation biophysical and biochemical characteristics estimation. Several achievements have been registered in exploiting spectral information for the retrieval of vegetation biophysical and biochemical parameters using statistical and physical approaches. These involve the derivation of new vegetation indices and the successful implementation of a radiative transfer model inversion (with extensive validation), which comprised the development of a new method to subset the spectral data based on average absolute error.

Mots clés : vegetation / stand characteristics / remote sensing / spectral analysis / physical models / mathematical models / monitoring


Version intégrale

Page publiée le 16 janvier 2015, mise à jour le 18 décembre 2016