Synergy of Airborne LiDAR Data and VHR Satellite Optical Imagery for Individual Crown and Tree Species Identification
Accurate data on individual tree crowns and their species within stands is still limited affecting many remote sensing studies on allometric equations,timber volume,above ground biomass and carbon exchange.This study evaluated the synergistic use of fine resolution multispectral imagery (WorldView-2,2 m) and high density LiDAR data (160 points/m-2) for individual crown segmentation and species identification and classification of two conifer species (Scots Pine Pinus sylvestris L.and Mountain pine Pinus uncinata Mill.Ex Mirb),in a mountainous area of the southern French Alps.The integration of WorldView-2 multispectral imagery and LiDAR data was considered during image segmentation and subsequent species identification and classification on a premise of complementarity.Three individual segmentation and species identification schemes were examined namely; segmentation and species identification based on LiDAR layers,spectral layers and a combination of the two datasets.A region growing segmentation approach was used.For each scheme,individual treetops were identified using a fixed-window local maxima approach and were used as seed pixels to grow individual tree crowns.The individual crown segments were subsequently used to derive one spectral and three physical characteristics for species identification.Tree height,crown diameter and the coefficient of variation of LiDAR intensity were the physical parameters derived from LiDAR data whereas the maximum satellite albedo reflectance was the spectral attribute derived from the optical satellite data.Logistic Regression and Classification and Regression Trees (CART) modelling approaches were used to identify each tree to either Scots or Mountain pine.Quantitative segmentation quality assessment showed that the LiDAR derived segments were superior (Segmentation goodness = 86.4%) to the optical segments.However,given the distortions in the multispectral image,integration of the datasets for individual crown segmentation was not possible.Classification accuracy results showed that the integration of spectral and LiDAR data improved the species identification compared to using either data sources independently.The highest classification accuracy (Kappa = 54%) was acquired when using both spectral and LiDAR derived metrics and a CART approach.This study concluded that the integration of LiDAR and the multispectral imagery provided increased interpretation capabilities and more reliable results.
C.B.Kukunda Y.A.Hussin H.van Gils
Department of Natural Resources, Faculty of Geo-information Science and Earth Observation, University of Twente, Hengelosstraat 99, 7500 AA, Enschede, The Netherlands
国际会议
北京
英文
210-217
2013-10-09（万方平台首次上网日期，不代表论文的发表时间）