Assessing the accuracy for area-based tree species classification using Sentinel-1 C-band SAR data

Abstract: Forest type (FTY) and tree species classification (SPP) over the Remn-ingstorp test site were performed using ground-based field observations and remote sensing data sources. The field inventory for the forest estate and for the surrounding natural reserve of Eahagen was carried out in 2016. The re-mote sensing data used were C-band Synthetic Aperture Radar (SAR) data from Sentinel-1. Dual polarization backscatter values were extracted for the period October 2017 - February 2019 and the area-based method was applied. The metrics obtained, i.e. monthly mean backscatter, were used to perform classification by machine learning models’ random forest (RF) and linear dis-criminant analysis (LDA). The models were evaluated with the leave-one-out cross-validation method and the classification outcomes were compared with reference values in terms of confusion matrixes. The best performing model was LDA with an overall accuracy of 88% for FTY and 61% for SPP, whereas RF achieved values of 84% for FTY and 56% for SPP. It was concluded that C-band SAR data can be used for FTY and SPP classification, but further investigation is needed to determine which factors affect the backscatter in order to obtain more accurate classifications.