Estimation of canopy height using lidar and radar interferometry: an assessment of combination methods and sensitivity to instrument, terrain and canopy height profile

Abstract:

The combined use of Lidar and radar interferometry to estimate canopy height can be classified into 3 categories: cross-validation, simple combination and fusion methods. In this presentation, we investigate the potential of each category for local and regional scale applications, and assess their sensitivity to instrument configuration, terrain topography and variations in the vertical forest canopy profiles.

In addition to field data, we use data from TanDEM-X, UAVSAR (Uninhabited Aerial Vehicle Synthetic Aperture Radar), LVIS (Laser Vegetation Imaging Sensor) and a commercial discrete lidar. TanDEM-X is a pair of X-band spaceborne radars flying in formation to provide a global digital surface model and can also be used to perform polarimetric synthetic aperture radar (polinSAR) inversion of canopy height. The UAVSAR is an airborne fully polarimetric radar enabling repeat-pass interferometry and has been used for polinsar. While LVIS records the full waveform within a 20m footprint, the discrete lidar collects a cloud of points. The lidar data can be used to validate the polinSAR results (validation), to obtain ground elevation (simple combination with radar surface models) or within the polinSAR inversion model through a common model framework.

The data was collected over the Laurentides Wildlife Reserve, a managed territory covering 7861km2 which is located between Québec city and Saguenay. The variety of management practices offers the possibility for long term and comparative studies of natural forest dynamics as well as the impact of human, fires and insect disturbances. The large elevational gradient of the region (~1000m) allows study of variations in structure and type of forests.

Depending on the method used, several factors may degrade the accuracy of canopy height estimates from the combined use of lidar and radar interferometry. Here we will consider misregistration of datasets, differences in spatial resolution and viewing geometry, geometric decorrelation and the vertical wavenumber. Finally we investigate the sensitivity of estimate to forest vertical profile and terrain topography.