Systematic variations in multi-spectral lidar representations of canopy height profiles and gap probability

Abstract:

Airborne and terrestrial lidar are increasingly used in forest attribute modeling for carbon, ecosystem and resource monitoring. The near infra-red wavelength at 1064nm has been utilised most in airborne applications due to, for example, diode manufacture costs, surface reflectance and eye safety. Foliage reflects well at 1064nm and most of the literature on airborne lidar forest structure is based on data from this wavelength. However, lidar systems also operate at wavelengths further from the visible spectrum (e.g. 1550nm) for eye safety reasons. This corresponds to a water absorption band and can be sensitive to attenuation if surfaces contain moisture. Alternatively, some systems operate in the visible range (e.g. 532nm) for specialised applications requiring simultaneous mapping of terrestrial and bathymetric surfaces. All these wavelengths provide analogous 3D canopy structure reconstructions and thus offer the potential to be combined for spatial comparisons or temporal monitoring. However, a systematic comparison of wavelength-dependent foliage profile and gap probability (index of transmittance) is needed. Here we report on two multispectral lidar missions carried out in 2013 and 2015 over conifer, deciduous and mixed stands in Ontario, Canada. The first used separate lidar sensors acquiring comparable data at three wavelengths, while the second used a single sensor with 3 integrated laser systems. In both cases, wavelenegths sampled were 532nm, 1064nm and 1550nm. The experiment revealed significant differences in proportions of returns at ground level, the vertical foliage distribution and gap probability across wavelengths. Canopy attenuation was greatest at 532nm due to photosynthetic plant tissue absorption. Relative to 1064nm, foliage was systematically undersampled at the 10% to 60% height percentiles at both 1550nm and 532nm (this was confirmed with coincident terrestrial lidar data). When using all returns to calculate gap probability, all wavelengths were within 6% but when using first returns only, gap probability was overestimated by 67% at 532nm and 11% by 1550nm. These results demonstrate that each wavelength contains distinct information about canopy attributes and models must account for variations in wavelength if applied to data for monitoring purposes.