The use of LIDAR for automated forest measurement Hans-Erik Andersen Precision Forestry Cooperative

advertisement
The use of LIDAR for automated
forest measurement
Hans-Erik Andersen
Precision Forestry Cooperative
University of Washington
College of Forest Resources
LIDAR (LIght Detection And
Ranging)
• Active airborne sensor
emits several thousand
laser pulses per second
• Many sensors can record
multiple returns from a
single laser pulse
• Recorded pulse returns are
post-processed and
delivered as X-Y-Z
coordinates
LIDAR for Forest Measurement
• Ability to penetrate forest canopy
• Some laser pulses reach forest floor, while intermediate
returns reflect from canopy and sub-canopy vegetation
• Allows for detailed modeling of terrain surface
• Forest measurement
– Most work to date has emphasized stand-level analysis
– Little work has taken advantage of the high spatial resolution of
LIDAR data (> 1 return per m2)
– This high resolution allows for patterns in the data to be
recognized and measurement carried out at the scale of an
individual tree
Canopy Surface Model generated from filtered LIDAR returns
One-acre area within Capitol State Forest, WA
Research Objective:
Automated forest measurement using
airborne LIDAR data
• Quantitative analysis of geometric structure (shape and size content)
within canopy surface model using mathematical morphology, a settheoretic approach to image understanding
•Morphological analysis drives individual tree crown recognition and
measurement algorithms
1/5 acre inventory plot within Capitol State Forest, WA
Raw LIDAR returns within 1/5 acre plot in
Capitol State Forest, WA (3382 returns)
Canopy Surface Model (green) and Terrain Surface Model (purple)
generated from LIDAR for 1/5 acre plot in Capitol State Forest, WA
3-D Morphological
Individual Tree Crown recognition algorithm
Profile of forested area
3-D Morphological
Individual Tree Crown recognition algorithm
Stylized profile of LIDAR returns
over forested area
3-D Morphological
Individual Tree Crown recognition algorithm
Generation of LIDAR canopy surface
model
3-D Morphological
Individual Tree Crown recognition algorithm
LIDAR Canopy surface model
3-D Morphological
Individual Tree Crown recognition algorithm
Morphological Opening Operation:
• Disk of specified radius is pressed up hard against surface
• Morphological opening is the elevation of the disk as it
slides under the entire surface
3-D Morphological
Individual Tree Crown recognition algorithm
• TOP-HAT TRANSFORM extracts high points in image
• Scale invariant – as opposed to thresholding operation, will
extract tops of large and small trees
• Used to mask out LIDAR returns near tree top
• Maximum LIDAR return within this mask gives a very good
estimate of the position of the individual tree top
Original surface
-
Opening of surface
=
Top-hat transform
Canopy Surface Model (green) and Terrain Surface Model (purple)
generated from LIDAR for 1/5 acre plot in Capitol State Forest, WA
Areas masked by morphological opening operation on
LIDAR Canopy Surface Model, 1/5th acre plot
These areas correspond to possible location of individual trees
Output of 3-D Morphological Individual Tree Crown recognition
algorithm applied to LIDAR-derived Canopy Surface Model for 1/5
acre plot in Capitol State Forest, WA
Comparison of automated LIDAR-based individual tree measurement (blue) vs.
manual (photogrammetric) individual tree measurement (yellow)
80 % of LIDAR-measurements within 3 ft. of photo-measured tree position!
Current and Future Research Objectives:
Automated forest measurement using
airborne LIDAR data
• To date have used only helicopter-based LIDAR data; need high-quality ground
DTM from fixed wing data set to assess differences in LIDAR density and acquisition
methods.
• Use more detailed DTM to increase accuracy of individual tree height measurement
• Model 3-D spatial distribution of lidar returns within an individual tree crown
• Incorporate prior knowledge relating to spatial distribution and interaction of trees
into this model
• Infer the most probable location and dimensions of individual trees given the
distribution of LIDAR returns
Download