Comparing integrated LiDAR and multispectral data with field measurements in hardwood stands

Height, species, and crown size were attributed to polygons generated from eight crown radii of trees in a bottomland hardwood forest in East Mississippi. These attributed polygons were used in tandem with eCognition (version 2.1) generated objects and attributes to identify matched field and automated metrics for testing at the tree level. Small-footprint, multi-return LiDAR surfaces and two multispectral image sources were used to create these image objects.; LiDAR analysis did not accurately measure tree heights, although field and LiDAR measurements depicted good correlation (R2 = 0.6231) and high model significance. Fused LiDAR and multispectral data performed moderately well in classifying individual tree crowns to their appropriate species (overall accuracy = 0.5420, KHAT = 0.4543) and merchantability (overall accuracy = 0.6260, KHAT = 0.5063) classes. Automated individual crown delineation and density measures were significantly different from their field counterparts (alpha = 0.05) with a poor overall matching of 32.11%.