Contribution of polarimetric SAR attributes for modeling of the tropical forest biomass affected by fire

This study presents the capability of full polarimetric PALSAR ALOS attributes to improve the aboveground biomass (AGB) modeling of forests affected by fires in the Brazilian Amazon (State of Roraima). To perform this study, we carried out multivariate regression, using coherent and incoherent SAR attributes and AGB values estimated through biophysical parameters obtained during the forest inventory. The PALSAR data were calibrated both radiometrically and geometrically to extract the polarimetric informations. In summary the results were: (a) considering the polarimetric signatures, we observed a higher scattering contribution of VV polarization in lightly burned forests, whereas there was a dominance of HH polarization scattering in the areas with the interannual recurrence of biomass burning; (b) the most significant variables for this tropical forest biomass modeling were: the anisotropy (A) derived from the eigenvectors and eigenvalues of Cloude & Pottier target decomposition; the orientation angle (ψ2) of second component of Touzi decomposition; the double-bounce scattering component (Pd) of Freeman & Durden decomposition; and also, the Volumetric Scattering Index (VSI) of Pope model. In order to validate this model using PALSAR images, a cross-validation method (leave-one-out) was performed, which indicated a prediction average error of 23% to estimate the stock density of forests affected by fire actions, where the AGB levels were reduced up to 60%. This study confirm that L-band PALSAR data can be applied to quantify and monitoring the carbon stocks in the tropical forest affected by fire, with an adequate accuracy, similar to that presented by traditional forest inventories.