Cover type classification and biomass estimation by spectral analysis

摘要

This paper was written within the context of scaling up forest attributes, especially cover type and biomass, from ground plot inventory data to near infrared aerial photos. The material used is represented by ground plots georeferenced on aerial photos. Areas of 150 by 150 m are decomposed into proportions of spectrally distinct land cover elements: shadow and sunlit. Pielou's non-randomness index is used as a surrogate of tree spatial distribution. Cover type is predicted with 71% of accuracy by blue, red, and green bands and by Pielou's index. Meanwhile, only the blue and green bands are significant explanatory variables of biomass with a poor accuracy. More intensive use of photo texture information should improve the biomass prediction.