Tree species classification using airborne laser and optical scanner data for forest management in a test site in Germany

摘要

This study is part of a R&D project in the German Federal State of Northrhine-Westfalia with the overall goal to develop an operational system for information extraction for forest management on stand level. With the improved availability of new airborne laser scanners (ALS) and optical line scanners (AOS) there is an increased number of options to develop assessment methods and the corresponding software to process this ever increasing quantity of data. Several aerial missions equipped with ALS (TopoSys Falcon II, Riegl LMS-Q560) and AOS (TopoSys line scanner, DLR HRSC-A) were flown in two forest districts in Northrhine-Westfalia in the years 2004-2007. Reliability of tree species classification is one of the key issues for the successful use of remote sensing when it is about deriving stand attributes for forest management planning. The objective of our research was to find features in ALS and AOS data useful to distinguish tree species, where in the study stands there were 9 different tree species (Picea abies, Fagus sylvatica, Quercus petrea, Acer pseudoplatanus, Fraxinus excelsior, Pseudotsuga menzlesii, Larix decidua, Larix kaempferi, Sorbus aucuparia). In a first step individual trees were identified and separated using a segmentation approach. Laser points and spectral values of digital orthophotos were extracted for every segment. A 3D shape parameter was calculated as additional variable, fitting a generalized ellipsoid of revolution to the lidar points of a crown segment. In a second step we used exploratory analyses of variance and logistic regression models to identify patterns in the data that support prediction of tree species and/or taxonomic groups. Best subsets were identified for specific combinations of ALS and AOS data variables and classification accuracies.