Abstract:

Remote sensing technology has been proved useful in mapping grassland vegetation properties. Spectral features of vegetation cover can be recorded by optical sensors on board of different platforms. With increasing popularity of applying unmanned aerial vehicle (UAV) to mapping plant cover, the study aims to investigate the possible applications and potential issues related to mapping leaf area index (LAI) through integration of remote sensing imagery collected by multiple sensors.

This paper applied the collected spectral data through field-based (FLD) and UAV-borne spectroradiometer to map LAI in a Sino–German experiment pasture located in the Xilingol grassland, Inner Mongolia, China. Spectroradiometers on FLD and UAV platforms were taken to measure spectral reflectance related to the targeted vegetation properties. Based on eight vegetation indices (VIs) computed from the collected hyperspectral data, regression models were used to inverse LAI. The spectral responses between FLD and UAV platforms were compared, and the regression models relating LAI with VIs from FLD and UAV were established. The modeled LAIs by UAV and FLD platforms were analyzed in order to evaluate the feasibility of potential integration of spectra data for mapping vegetation from the two platforms.

Results indicated that the spectral reflectance between FLD and UAV showed critical gaps in the green and near-infrared regions of the spectrum over densely vegetated areas, while the gaps were small over sparsely vegetated areas. The VI values from FLD spectra were greater than their UAV-based counterparts. Out of all the VIs, broadband generalized soil-adjusted vegetation index (GESAVI) and narrow-band nNDVI2 were found to achieve the best results in terms of the accuracy of the inversed LAIs for both FLD and UAV platforms. We conclude that GESAVI and nNDVI2 are the two promising VIs for both platforms and thus preferred for LAI inversion to carry spectra integration of the two platforms. We suggest that accuracy on the LAI inversion could be improved by applying more advanced functions (e.g. non-linear) considering the observed bias for the difference between the UAV- and FLD-inversed LAIs, especially when LAI was low.

Key words: grassland, leaf area index, unmanned aerial vehicle, vegetation index, remote sensing