Tree species classification in a temperate mixed forest using a combination of imaging spectroscopy and airborne laser scanning

Abstract

Knowledge of the spatial distribution of tree species is important for efficiently managing and monitoring forested ecosystems, especially in mixed forests of the temperate zone. In this study, we fused imaging spectroscopy (IS) data with leaf-on and off small-footprint airborne laser scanning (ALS) data, for tree species identification in a dense temperate forest in Switzerland. In addition to the spectral reflectance of the sunlit part of the tree crowns, structural features computed based on the height, intensity and point distribution of ALS data in both the vertical and horizontal dimensions are used as features. Features were extracted using a pixel-based (1 m × 1 m) and an individual tree crown approach. In addition, applying a floating forward feature selection approach revealed that the ALS-derived features provided relevant structural information for species identification, while IS-derived features added complementary biochemical information. Comparing the accuracies of three different combinations of ALS and IS data, shows the highest classification accuracy (kappa = 90.3%) was obtained by fusing a selected set of features at individual tree crowns (ITC), while the best kappa accuracies resulting from IS or ALS data alone were 74.7% and 75.1%, respectively. Inclusion of the ITC information improved the classification results for all datasets, however, this improvement is significantly higher for ALS derived datasets (+31%). Our results show that accurate ITC information drastically improves classification accuracy of tree species in dense forests and that multi-seasonal ALS structural attributes play a major part in species discrimination.