An Object-Oriented Classification of Muir Woods using the Synergy of LiDAR and Multispectral Data

Abstract

An Object Based Image Analysis (OBIA) was employed to classify four tree species in a temperate rainforest utilizing the high-resolution WorldView-2 (WV2) sensor (8 bands + panchromatic) and airborne LiDAR (minimally 2 points per square meter). Classification involved first performing a parametric Maximum Likelihood (ML), Spectral Angle Mapper (SAM), and OBIA classification to the study area. Secondly, for each classified image, a LiDAR-derived Canopy Height Model (CHM) was incorporated thereafter. Kappa and z statistics were calculated and compared for each classification. It was originally hypothesized that an OBIA will provide the best accuracy, and incorporation of a CHM would further increase classification accuracy for all outputs. A series of statistical tests indicated a lack of strength in utilizing the CHM, except when specifying the Coastal Redwood class at 50m. Kappa results are 59% for the OBIA, 46% for ML, and 24% for SAM. CHM increased kappa accuracy by an average of 4.5%.