An airborne Lidar canopy segmentation approach for estimating above-ground biomass in coastal eucalypt forests

摘要

There is growing interest in airborne lidar for forest carbon accounting and precisionforestry purposes. Airborne lidar systems offer a cost-effective, versatile, operationallyflexible and robust sampling tool for forest managers. The objective of this study was todevelop and test lidar canopy surface enhancement and segmentation processes forestimating dominant above-ground biomass (DAB) in a harvested eucalypt forest on theCentral Coast of New South Wales (Australia).The Crown Infill, Trim and Smooth (CITS) process, incorporating a series of filters,algorithms, and selective multi-stage smoothing, was used to enhance lidar canopysurfaces prior to segmentation. Canopy segmentation was achieved using a verticalcrown template approach termed the Spatially and Morphologically Isolated Crest(SMIC) process. SMIC delineates dominant tree crowns by detecting elevated crowncrests within a 3D lidar canopy surface. Consolidated crown units constitute the basicsampling, analysis and reporting units for wall-to-wall forest inventory. Theperformance, sensitivity and limitations of these procedures were evaluated using acombination of simulated forest models and actual lidar forest data.Automated crown polygons were used as a sampling template to extract dominant treeheight values which were converted to DAB estimates via height-to-biomassrelationships derived from field survey and on-site destructive sampling. Results werecompared with field based tree height and biomass estimates.Compared against a manually derived crown map from a 2ha field plot, canopysegmentation results revealed a producer's accuracy of 76% and overall accuracy of67%. Results indicated a trend toward greater crown splitting (fragmentation) as treesincrease in age, height, stem diameter and crown size. Extracted dominant tree heightvalues were highly correlated with ground survey height estimates (r2 0.95 for precisionsurvey and r2 0.69 for standard survey). There was also no significant differencebetween SMIC and manual crown height estimates.SMIC units overestimated ground-based DAB by 5%; this increased to 36% with theinclusion of segmentation errors. However, SMIC estimation of total plot above-groundbiomass (AGB) was within 9% of the ground-based estimate. Results are encouragingconsidering the mixed-species, multi-aged composition of the forest, and the combinedeffects of SMIC segmentation and lidar height errors.