Study of a Simple Volume Scattering Model on Burned Forest Using Polarimetric PALSAR-2 Data

Several studies have taken advantage of polarimetric synthetic aperture radar (PolSAR) to monitor forest disturbance caused by wildfire given its higher sensitivity to forest structure compared to single polarization SAR. This letter explores the capability of a simple volume scattering model (SVSM) to characterize burned forested area caused by wildfire. The SVSM considers a shape factor and geometric randomness to model a nth cosine probability density function (PDF) assumption of the rotation angle with respect to the line of sight. The shape factor describes the shape of elements that constitute the forest canopy, while the geometric randomness represents the variance of the PDF. Two quad polarization L-band PALSAR-2 data acquired over Fort McMurray, AB, Canada, in 2015 and 2016 before and after a severe wildfire are used for this exploration. The ability of the shape factor is evaluated first for the coniferous and broadleaf tree classification, which achieves an overall accuracy as high as 77.41% and kappa of 0.55. The simple linear regression between the burn classes and geometric randomness change shows that the geometric randomness change has a high potential for the light, modest, and severe burn classes estimation.