The article of Yi Lin & Martin Herold: Tree species classification based on explicit tree structure feature parameters derived from static terrestrial laser scanning data, has been pubished in Agricultural and Forest Meteorology, Volume 216, 15 January 2016, Pages 105–114.
Tree species information is essential for forest studies such as forest meteorology, botany and ecology, and across the relevant fields new techniques efficient for classifying tree species are desperately demanded. For this requirement, the state-of-the-art remote sensing technology of static terrestrial laser scanning (TLS) shows the potential of handling it, since TLS can represent tree structures in details. However, the literature review suggests that the endeavors of introducing TLS into tree species classification are still in shortage. Aimed at this technical gap, this study attempted TLS for distinguishing four typical boreal tree species, i.e. Norway spruce (Picea abies), Scots pine (Pinus sylvestris), European aspen (Populus tremula) and pedunculate oak (Quercus robur). After a theoretical comparison of the generic mechanisms of TLS- and airborne laser scanning (ALS)-based forest mapping, explicit tree structure (ETS) feature parameters, rather than conventional ALS-derived statistical-sense feature parameters, were proposed and derived from TLS point clouds. Then, based on a support-vector-machine (SVM) classifier, the specific classification was operated in a leave-one-out-for-cross-validation (LOOCV) mode. Tests indicated that the proposed TLS-based ETS feature parameters and the used classification algorithm can be validated for implementing the task of classifying the four tree species (the maximum total accuracy reaches 90.00% and the robust total accuracy reaches 77.5%). Overall, as a leading endeavor, this study has developed a procedural frame for TLS-based tree species classification.
Keywords: Tree species classification; Static terrestrial laser scanning (TLS); Explicit tree structure feature parameters; Support vector machine (SVM) classifier; Leave-one-out-for-cross-validation (LOOCV)