Evaluation of simulated bands in airborne optical sensors for tree species identification

Airborne multispectral remote sensing devices have been used in automatic identification of tree species, and the spatial and spectral properties of the sensors affect the remote sensing measurement results. Previous work based on a simulation model with ground-level measured reflectance data of pine (Pinus sylvestris L.), spruce (Picea abies (L.) H. Karst.), and birch (Betula pubescens Ehrh. and Betula pendula Roth) tree species and idealized Leica ADS80 sensitivities suggested that the addition of a fifth sensitivity band in the red edge wavelength region to the existing Leica ADS80 system significantly improves the classification performance. In this paper, we extend this analysis using a simulated model with accurate spectral sensitivity information and airborne AisaEAGLE II hyperspectral data for these three tree species. We simulated multispectral responses using spectral sensitivity characteristics of the Leica ADS40, the Vexcel UltraCam-D, the Intergraph-Z/I Digital mapping camera and the Leica ADS40 system with an added band in the 691–785 nm region. We evaluated the tree species classification performance of these simulated responses using Discriminant Analysis and Support Vector Machine classifiers. The classification experiment result showed that the simulated responses of the 5-band multispectral system yielded the most robust classification performance with approximately 98% accuracy. This result was similar to the accuracy obtained with the hyperspectral data. Although differences were observed in the sensitivity functions of the 4-band systems, there were no large differences observed in the classification performances between them. With the simulated 5-band system, there was an increase of 5–13% points in classification accuracy when compared to the accuracies of the 4-band systems. The results obtained via proposed 5-band system support results from previous studies suggesting that there is a need for a sensitivity band in the red edge wavelength region for applications in tree species classification.