EARSeL eProceedings 12, 1/2013 25 CANOPY STRUCTURE EFFECT ON SAR IMAGE TEXTURE VERSUS FOREST BIOMASS RELATIONSHIPS Isabelle Champion 1 , Jean Pierre Da Costa 2 , Adrien Godineau 2 , Ludovic Villard 3 , Pascale Dubois-Fernandez 4 , and Tuy Le Toan 3 1. INRA, UR1263 EPHYSE, F-33140 Villenave d’Ornon, France; e-mail: champion(at)bordeaux.inra.fr 2. CNRS, UMR 5218 IMS, F-33405 Talence, France 3. Centre d'Etudes Spatiales de la Biosphère (CESBIO), CNRS-CNES Université Paul Sabatier, 18 Avenue Edouard Belin, 31401 Toulouse Cedex 4, France 4. Office National d’Etudes et Recherches Aérospatiales (ONERA), Centre de Salon de Provence - BA 701, 13661 Salon, France ABSTRACT Quantifying forest biomass is of crucial importance for estimating carbon fluxes on the regional and global scale in climate change studies. Significant relationships have already been established between radar mean intensity and forest biomass, but these relationships show a reduced sensitivity to biomass variations for mature stands (about 80 t/ha and more). On the contrary, recent studies have shown that image texture is significantly related to biomass even for mature stands for a temperate, monospecific, even-aged forest the biomass of which is 140 t/ha at its highest point. The present paper aims at extending these observations to tropical forests which represent a large terrestrial biomass pool with values higher than 450 t/ha. Radar images were acquired during the TropiSAR experiment in 2009, which took place over a tropical rain forest located in French Guiana at P band and cross-polarization with the use of SETHI ONERA airborne instrument. Three sets of treatments applied to 15 forest stands provided biomass values from 268 to 466 t/ha where permanent zones of 6.25 ha each were mapped and regularly measured. Homogeneous patches were selected inside each of the 15 experimental stands. Statistical features were then derived for each patch: a) from grey level statistics; b) from the statistics of pixel pairs on the basis of the gray level co-occurrence matrix. It is shown that linear relationships between texture features and forest biomass are heavily influenced by stand structure and the local topography and soil of the experimental stands. But, when stands are separated on two structural groups using texture descriptors, texture/biomass regressions reveal to be very significant. INTRODUCTION With growing awareness of the role of forests in the global carbon cycle, the need for methods of monitoring biomass has become more urgent. Tropical forests form a large part of the terrestrial carbon pool and the carbon sources generated by deforestation in the tropics. However, little information is available on the global scale about these regions which are difficult to get to. Remote sensing techniques are therefore of particular interest with respect to studies on global climate. In this context, Synthetic Aperture Radar (SAR) systems provide images independently of cloud cover contrary to optical systems. Moreover, SAR systems have demonstrated their potential to discriminate forest status, especially at low frequencies. In order to determine the biomass of a forest, significant relationships are therefore established between mean intensity of the backscattering coefficient (or σ o ) and biophysical variables (stand age, biomass and related variables). Carbon quantities are estimated by inferring wood biomass from forest biomass, and then converting it into carbon by using a value of approximately 0.5 ton of carbon for 1 ton of wood. However, for mature stands (about 80 t/ha and more) increasing biomass reduces the sensitivity of the σ o /biomass relationships (1). DOI: 10.12760/01-2013-1-03