HR: 11:35h
AN: B42B-05 [Abstracts]
TI: Automatic Tree Crown Delineation Using Discrete Return Lidar and its Application in ICEsat Vegetation Product Validation
AU: * Pang, Y
EM: caf.pang@gmail.com
AF: Center for Ecological Applications of Lidar, College of Natural Resources, Colorado State
University, 1472 Campus Delivery, Fort Collins, CO 80523, United States
AU: Lefsky, M
EM: lefsky@cnr.colostate.edu
AF: Center for Ecological Applications of Lidar, College of Natural Resources, Colorado State
University, 1472 Campus Delivery, Fort Collins, CO 80523, United States
AU: Miller, M E
EM: memiller@warnercnr.colostate.edu
AF: Center for Ecological Applications of Lidar, College of Natural Resources, Colorado State
University, 1472 Campus Delivery, Fort Collins, CO 80523, United States
AU: Sherrill, K
EM: sherrill@warnercnr.colostate.edu
AF: Center for Ecological Applications of Lidar, College of Natural Resources, Colorado State
University, 1472 Campus Delivery, Fort Collins, CO 80523, United States
AU: Andersen, H
EM: handersen@fs.fed.us
AF: Forest Inventory and Analysis, USDA Forest Service PNW Research Station, Anchorage
Forestry Sciences Laboratory, 3301 C Street, Suite 200, Anchorage, AK 99503, United States
AB:
The Geoscience Laser Altimeter System (GLAS) has acquired over 250 million individual lidar observations over
forest regions globally; an unprecedented dataset of vegetation heights. The vertical extent of waveforms collected
by GLAS increases as a function of terrain slope and footprint size (the area on the ground that is illuminated by
the laser), but we have demonstrated the ability to retrieve accurate vegetation heights from them. It is infeasible
to characterize the elevation of terrain and the crown geometry of trees within the 50-70m diameter GLAS footprint
at the level of precision required for our current analyses. In addition, it would be prohibitively expensive to collect
the amount of data required to develop a consistent field relationship for forests globally. Airborne discrete return
lidar (DRL) has the potential to characterize the vegetation and terrain surfaces within the footprint with a level of
detail and precision not available from fieldwork. Furthermore, high density DRL datasets are becoming generally
available for forested sites worldwide.
Numerous studies have used DRL point cloud data to estimate individual tree parameters successfully. In the
GLAS waveform, tree crowns are represented in proportion to their crown area, and therefore estimated forest
height is weighted towards those trees with the largest crowns. This assumption is validated by a three
dimensional lidar waveform model. To estimate average heights of trees (rather than the average heights of DRL
points), we developed a crown delineation algorithm to automatically identify dominant and co-dominant tree
crowns from DRL data. We then calculated a crown size weighted average height for 511 plots located at five
sites along the west coast of North America from California to Alaska. Parameterizing the existing height
estimation equation with the new plot dataset results in very consistent relationships between the DRL forest
height and forest height estimated from GLAS. The R2 is 0.64 and RMSE is 6.5m for 511 plots.
Keywords:
discrete-return lidar, crown delineation, ICEsat, GLAS, forest height, validation
DE: 0428 Carbon cycling (4806)
DE: 0439 Ecosystems, structure and dynamics (4815)
DE: 0480 Remote sensing
SC: Biogeosciences [B]
MN: 2007 Fall Meeting