Forest plantation species classification using Full-Pol-Time-Averaged SAR scattering powers

Abstract

Forest plantation species identification is essential for biodiversity conservation, sustainable management, and carbon sequestration. Fully Polarimetric Synthetic Aperture Radar (Full-Pol-SAR) remote sensing data aids as an alternative to time-consuming field inventory and cloud-covered optical data for forest plantation species identification. This study tried to identify forest plantation species using Fully Polarimetric Time-series Averaged SAR (Full-Pol-Time-Averaged SAR) datasets. Due to the availability of time series quad polarization data of TerraSAR-X from December 20, 2014 to March 07, 2015 and near-real-time field with us for the Haldwani forest area, we have utilized these data sets for the proposed work. The temporal effect of phenological changes can be characterized using the Full-Pol-SAR 6-component Scattering power Decomposition (6SD) and is utilized to identify different forest species. Misclassification problem exists in forests with single-date 6SD images, an improved forest species classification approach based on time-averaged 6SD and a supervised Random Forest (RF) classifier is proposed. This study reveals that Teak, Eucalyptus, Poplar, mixed, and non-forest could be distinctly identified and classified in time-averaged 6SD images. The overall classification accuracy of the proposed approach (∼83.72%) was, higher than the single-date Full-Pol-SAR results (mean overall accuracy with standard deviation 68.949 ± 2.096%).