Please use this identifier to cite or link to this item:
|Title:||Mapping of Vegetation Biophysical Variables using Radiative Transfer Models - Coping with the Ill-posed Inverse Problem|
|Citation:||Proceedings of the Remote Sensing and Photogrammetry Society Annual Conference, 2008 p. 1-4|
|Publisher:||University of Exeter|
|Type:||Articles in periodicals and books|
|Abstract:||Optical remote sensing plays a fundament role in attempts to map repeatedly, in various spatial resolutions, and in a cost efficient way, the status of the Earth system. For the retrieval of vegetation biophysical variables such as leaf area index, physically based radiative transfer models (RTM) have been developed. The RTMs link the variables of interest with the measured spectral and/or directional signature of the canopy. To map the biophysical variables, the RTMs have to be inverted. For the inversion, different strategies have been developed (e.g., look-up-table, neural nets), each having specific advantages and disadvantages. Common to all inversion strategies, however, is the ill-posed nature of the inverse problem, meaning that the inverse solution is not always unique leading to large uncertainties in the estimated fields. The uncertainties arise from several facts: (i) the dimensionality of the observed spectral/directional variables is generally lower than the number of unknowns within the RTM, (ii) several variables within the RTM can counterbalance each other, and (iii) both, spectral measurements and RTMs are not free of error. To cope with the ill-posed inverse problem, various strategies have been proposed such as the use of prior info within the cost function and the use of information provided by the continuous temporal evolution of key vegetation structural variables. The present contribution aims to further develop an approach that takes the signature of adjacent pixels into account [RSE (2004), 93: 53-67]. The idea builds on the geographical principle that near-by locations are probably more alike than locations at a larger distance. This principle holds for (semi-)natural environments (e.g., forests, shrub- and grasslands) as well as highly managed targets (e.g., agricultural fields). The proposed object-based inversion allows to effectively constrain the inverse problem. The pros and cons of the developed approach will be presented and discussed with respect to alternative solutions. Implementation issues will be described. Using mainly synthetic data, retrieval accuracies of main biophysical variables will be compared to those obtained without regularisation.|
|JRC Directorate:||Space, Security and Migration|
Files in This Item:
There are no files associated with this item.
Items in repository are protected by copyright, with all rights reserved, unless otherwise indicated.