Contribution of ALOS PALSAR fully polarimetric data for tropical vegetation cartography
Cedric Lardeux(1), Pierre-Louis Frison(1), Guillaume Roussel(1), Calvin Dikongo Ndjomba(1) and Jean-Paul Rudant(1)
(1) Universitee de Marne La Vallee, IFI 5 bd Descartes, 77454 Marne la Vallee Cedex 2, France
The goal of this study is to investigate the contribution of the fully polarimetric data for land use estimation over different study sites located in tropical region: in the Equatorial Africa, in French Polynesia, and in Brazil. The frames of these studies concern with the biodiversity for the management of different natural parks in Equatorial Africa (Gabon and the Democratic Republic of the Congo), the land use planning and the monitoring of invasive vegetation species in French Polynesia, and a better understanding of podzolisation and defertilization processes through the cartography of the podzols in the Jau basin, near the Rio Negro, in Brazil.
The classification algorithm retained is the Support Vector Machine (SVM) due to its ability to take into account multiple and heterogeneous parameters. In particular, different polarimetric indicators, derived from fully polarimetric data, such as the H/A/alpha parameters derived from the Cloude and Pottier decomposition, the ones based on the Pauli formalism, or the degrees of coherence between linear or circular polarizations can be accounted for. Preliminary results show that significant more accurate classification can be obtained with the SVM than with the classical Wishart classifier for example.
The results are compared to those that would be obtained with different partial polarimetric modes like Dual Polarimetric, Alternate polarization or single polarization configurations allowed by the ALOS/PALSAR or ENVISAT/ASAR sensors for example, as well as with the so called pi/4 mode. The contribution of L band to C band data is also analyzed.
Several field campaigns allowed to gather precious ground truth data used for the validation of the results.