Abstract
In this paper, we present an innovative workflow for retrieving biophysical traits of vegetation by means of inverting the radiative transfer model PROSAIL from UAV borne hyperspectral data. The approach makes use of spectral images acquired with a Fabry-Perot interferometer in the visible and near infrared spectral range. Even of the reduced spatial coverage of UAV acquisition, the high spatial resolution of such images makes model inversion computationally highly demanding. To overcome this, we made use of a machine learning method. Firstly, we generated look up tables by means of model forward runs based on variating model parameters according to prior knowledge. Random forests were then trained and applied to radiometrically calibrated UAV borne images. This allowed to retrieve model parameters for the area of interest (AOI). The approach showed to be computationally efficient and usable for ecosystems with high spatial variability.