Fusion of airborne hyperspectral and LiDAR data for tree species classification in the temperate forest of northeast China

Abstract

Tree species classification using hyperspectral data has a problem over the cases of different tree species with similar spectral signals, especially in the natural boreal forests which have abundant tree species. In this paper, we assessed the data fusion capacity of airborne hyperspectral and LiDAR for mapping dominant tree species in Liangshui National Nature Reserve, Heilongjiang Province, China. Training samples were extracted by Mixture Tuned Matched Filtering (MTMF) method for non-forest area, and combining with Canopy Height Model (CHM) from the first LiDAR return and first spectral derivative for forest area, respectively. The classification results from SVM classifier, were compared between hyperspectral data only and the fused data. It showed an encourage result that the fused hyperspectral and LiDAR data had great potentials for tree species classification, which increased both overall accuracy (86.88%) and Kappa coefficient (0.836) than hyperspectral data only (80.67%, 0.783). In particular, the CHM was effective for the discrimination of tree species with similar spectra but different heights.