Evaluating the potential of airborne hyperspectral LiDAR for assessing forest insects and diseases with 3D Radiative Transfer Modeling

Abstract

Accurately monitoring forest insects and diseases disturbances is important for managing forests and implementing effective forest quality improvement measures. However, monitoring disturbances in the lower and middle parts of the forest canopy with traditional remote sensing technologies is difficult. A new sensor called hyperspectral LiDAR (HSL) makes the monitoring of these forest disturbances possible, however, its applicability to canopy scale has not been fully studied due to the current hardware limitation. This paper assessed the potential of airborne hyperspectral LiDAR (AHSL) for monitoring forest insects and diseases stress using 3D radiative transfer modeling and in-situ measurements. A virtual 3D forest scene with explicitly described structures was first reconstructed from terrestrial laser scanning and field measurement data, upon which a number of different insects and diseases disturbance scenarios with different damage locations and stress levels were defined. AHSL point cloud and the corresponding hyperspectral image (HI) were then simulated with large-scale and remote sensing image simulation (LESS) model for each combination of the different damage locations and stress levels. LiDAR point cloud from different layers of the simulated AHSL point cloud were then extracted and rasterized into images with 3-m spatial resolution, which, along with the hyperspectral images, were used to test the insect and disease monitoring ability by using a random forest model. Results show that AHSL has significant higher overall accuracies (OA: 65.95% ∼ 89.45%) than HI (OA: 33.99% ∼ 57.02%) for predicting insects and diseases stress levels. Compared to AHSL, HI is affected by a variety of factors such as soil and shadows, resulting in poorer monitoring capability for different damage locations, with the highest classification accuracy in the case of entire canopy damage (OA: 57.02%). For AHSL, it has good classification accuracy for all damage locations, with the lowest accuracy in the case of lower canopy damage (OA: 65.95%). This study demonstrates that AHSL is a promising and reliable tool to monitor forest disturbances, especially for structural and spectral changes in the lower and middle parts of canopy, which may have great potential for early detecting forest insects and diseases.