Multi-Frequency Sar Utilization in Mountainous Regions in Bulgaria for Forest Boundaries Determination With Impact of Seasonality

This study focuses on an accuracy assessment of the forest boundary layer in a mountainous forest located in the north-west part of Bulgaria. The objective is an accuracy assessment of the forest boundaries determined via three different SAR bands – X, C, and L, whereas which method gives the highest accuracy with respect to seasonality. Polarimetric, interferometric, and time series SAR methods were utilized with POL-SAR images from ALOS PALSAR, bistatic InSAR data from TanDEM-X, and a four-year time series from Sentinel-1. Some SAR images were available in different seasons, allowing estimation of seasonality variations. Validation was performed with 60 points randomly generated. Overall Accuracy (OA) and Cohen’s Kappa were calculated to represent the accuracy of the masks. Strong influence of the seasonality was observed because of the higher ground contribution during the forest leaf-off. Frequency also plays a major role in accuracy. Supervised and unsupervised classifications were elaborated to determine forest classes. A Wishart unsupervised classifier was used for POL-SAR data. Assessment showed the highest OA above 90% of the C-band time series forest mask. The L-band mask also showed high accuracy with OA above 89%. The least accurate forest mask with an overall accuracy of 68% belongs to the X-band bistatic InSAR measurement in the winter. In conclusion, the SAR time series from Sentinel-1 showed the highest feasibility in the forest boundaries determination. The polarimetric SAR is better utilized for the purpose of the study. Accuracy is higher for the summer periods when scattering phase centers are higher from the ground, and volume scattering is better delineated from the ground contribution.