This is the presentation from my Master's Thesis defense at UGA in Spring 2013. The results were subsequently published in the journal Photogrammetric Engineering and Remote Sensing.

1. Christopher W. Strother
Master’s Thesis
Center for Geospatial Research
Department of Geography
University of Georgia
Spring 2013
DETECTION AND ANALYSIS OF EXTRAORDINARY
TREE HEIGHTS IN THE GREAT SMOKY
MOUNTAINS NATIONAL PARK USING REGIONAL
SCALE LIDAR DATA

2. OUTLINE
• Introduction
• Thesis Objectives
• Detection of Tall Trees
• OLS Analysis
• Conclusions
• References

3. INTRODUCTION

4. • Manuscript-style thesis with 2 articles for
publication -
• LiDAR Detection of the Ten Tallest Trees in the
Tennessee Portion of the Great Smoky
Mountains National Park
• Ordinary Least Squares Analysis of a LiDAR-
Derived Tree Height Database

5. • The Importance of Tall Trees
• Location of old growth communities
• Conditions favorable for growth potential
• Ecological importance as biomass and carbon sinks
• Habitat for other species of plants and animals
• Evidence that tall trees on the decline due to the
warming associated with climate change (Laurance,
2012)

6. • American Recovery and Reinvestment Act of 2009
• U.S. Geological Survey (USGS)
• Center for Remote Sensing and Mapping Science at
UGA
• Institute for Environmental and Spatial Analysis at GSC
• Photo Science, Inc.
• Eastern Native Tree Society
• National Park Service

8. • Great Smoky Mountains National Park
• The Park was established in 1934 to mitigate erosion and
fire damage caused by logging (Houk, 2000).
• The GRSM has approximately 209,000 ha of forest cover –
most expansive virgin forest land on East Coast (NPS,
1981).
• The Park receives up to 10 million visitors a year and has
been designated an International Biosphere Reserve and a
U.N. World Heritage Site (Welch et al., 2002).

9. • LiDAR (Light Detection And Ranging)
• LiDAR is a cost effective active remote sensing system used
to derive elevation information (Bossler et al., 2002; Maune,
2001).
• LiDAR systems use a near-infrared laser (1064 nm) and
accurate and precise clock to measure distance between the
sensor and the feature using the equation:
• 𝑅 =
1
2 𝑡𝑐
Where R = distance; t = time; c = speed of light
(Jensen, 2007)

10. http://forsys.cfr.washington.edu/JFSP06/lidar_technology.htm
• LiDAR Principles - acquisition

11. • LiDAR Principles – the point cloud

12. • LiDAR Principles – multiple returns

13. • LiDAR Principles - classification
• Classified according to ASPRS standards (LAS 1.2)
• 4 categories:
• 1 = Nonground
• 2 = Ground
• 7 = Noise
• 12 = Overlap

14. • LiDAR Principles – digital elevation model
(DEM)

15. • LiDAR Use In Forestry
• Airborne LIDAR has been used extensively in the last twenty
years to obtain accurate measurements of forested areas
(Nilsson, 1996; Maune, 2001; Andersen et al., 2006; Jensen,
2007).
• Maximum tree height is an indicator of ecological and
environmental quantities in tree communities regarding
biomass and resource use (Kempes et al., 2011).
• Errors inherent in LIDAR data include post spacing issues
(Fig. 2), which create misrepresentation of crown structure
(Zimble et al., 2003).

16. Zimble et al., 2003

17. • Ground Based Tree Height Measurement Techniques
• Accurate direct measurements of trees in the field are
difficult (Andersen et al., 2006).
• The USFS indicates that the best measurements are made
using a laser rangefinder with a built-in clinometer like the
Impulse100 (USFS, 2005).
h = hd (tan ρ + tan Θ)

18. THESIS OBJECTIVES

19. • Primary Goal – To investigate LiDAR as a remote
sensing tool for assessing vegetation structure and
providing resource managers with detailed information
on canopy height.

20. • Detection of maximum tree heights in the GRSM by creating a
methodology for processing a large dataset (724 tiles – each
representing 225 ha in area and around 200 – 300 Mb file size)
of recently acquired (2011) LiDAR data to identify potential trees
of extraordinary height and to assess the environmental
conditions at the top ten sites (Chapter 2).
• Assessment of LiDAR-derived tree height databases to predict
tree heights in a highly variable forested environment using
multivariate regression (Chapter 3).

21. Strother, C.W., M. Madden, T. Jordan, and A. Presotto. To be submitted to
Photogrammatic Engineering & Remote Sensing.
CHAPTER 2 - LIDAR DETECTION OF THE TEN
TALLEST TREES IN THE TENNESSEE PORTION OF
THE GREAT SMOKY MOUNTAINS NATIONAL PARK

22. INTRODUCTION

23. • June 2011 – Correspondence between Michael Davie
of the Eastern Native Tree Society (ENTS) and Dr.
Marguerite Madden began
• August 2011 – Intrepid and youthful new graduate
student became interested in the search

24. • Breckheimer (2011) work led to the discovery of a tulip
tree 58.0 meters tall in NC portion of the GRSM
• Tallest tree in TN portion of the GRSM listed as a tulip
tree 52.7 meters tall

26. DATA
724 tiles of LIDAR data, CIR imagery, and DEMs

27. METHODOLOGY

28. • Convert .las point cloud data to multipoint shapefiles for
ArcGIS processing

29. • Create Digital
Surface Models (DSMs)

30. • Create normalized
DSMs (nDSMs)

31. • Classify nDSM rasters for values of >51.8 m (170 ft)
and mosaic

33. • Convert raster values to points and query for height
values 52 – 59 m in the park

34. • Manual removal of noise, man-made objects, and
points outside of park boundary

35. • List the top ten height clusters with coordinates for field
verification

36. RESULTS

38. Site Lidar
Height (m)
Field
Height (m)
% Error vs.
Field
Elevation
(m)
Degree
Slope
Aspect Overstory Tree Type
1
59.0 Unknown Unknown 376.3 35.1 SW PIs-T Unknown
2
59.0 Unknown Unknown 358.6 51.9 N OmH/T Unknown
3
55.9 72.8 -30.2 494.2 10.3 E CHxA-T Pine
4
57.0 Unknown Unknown 477.9 39.4 NW PIs-T Unknown
5
57.0 56.6 0.7 394.0 54.7 NW PI Pine
6
57.0 61.5 -7.9 367.4 80.5 NW PIs White oak
7
55.0 58.4 -6.2 785.3 28.6 NE CHx Tulip poplar
8
56.0 56.9 -1.6 765.0 26.8 E CHx Tulip poplar
9
56.0 51.6 7.9 761.1 19.0 NE CHx Tulip poplar
10
55.0 56.4 -2.5 761.4 31.7 N CHx Tulip poplar

39. CONCLUSIONS AND RECOMMENDATIONS

40. • All ten sites are taller than the current height record
holder in the Tennessee portion of the GRSM
• Field measurement in rugged terrain is difficult
• More rigorous examination of the environmental and
ecological conditions at these sites is needed

41. Strother, C.W., M. Madden, T. Jordan, and S. Holloway. To be submitted to
The Professional Geographer.
CHAPTER 3 – ORDINARY LEAST SQUARES
ANALYSIS OF A LIDAR-DERIVED TREE HEIGHT
DATABASE

42. INTRODUCTION

43. • LiDAR data format provides large numbers of possible
observations for statistical analysis
• Ordinary Least Squares (OLS) analysis is a linear,
unbiased estimator that is useful in multivariate
regression
• With a wealth of canopy height observations, it should
be possible to model optimal conditions for growth that
can be used to predict recoverable carbon stock after
destructive events

44. False Gap

46. METHODOLOGY

47. • LiDAR Analyst was used to extract 22,187 tree points
from the LiDAR point cloud

48. • DEM of the study area was used to create slope and
aspect rasters

49. • Overstory vegetation, soil, and stream layers were
added

51. • All layers joined in ArcGIS to create database of 22,187
trees with elevation, slope, aspect, soil type, vegetation
community, distance to stream, and tree height
attributes
• Results imported to STATA IC 10 statistical analysis
software for modeling

52. RESULTS AND DISCUSSION

53. 0
1020304050
600 800 1000 1200
Elevation
TreeHeight Fitted values

54. 0
1020304050
0 20 40 60 80
Slope
TreeHeight Fitted values

55. 0
1020304050
0 100 200 300 400
NEAR_DIST
TreeHeight Fitted values

56. 05
10152025
Ditney Soco Spivey

57. 05
10152025
E N NE NW S SE SW W

58. 0
102030
LowElevationMixedPine-XericOak
MontaneAlluvialForest
SACoveHardwoods
SAEarlySuccessionalHardwoods
SAMixedHardwoodswithoutOaks
SAMixedHardwoods,Acidic
SANorthernHardwoods
SubmesictoMesicOak/Hardwoods

59. treeheight = b0 + b1*elevation + b2*near_dist +b3*slope
+ b4*s1 + b5*s2 + b6*a1 + b7*a2 + b8*a4 + b9*a5 + b10*a6
+ b11*a7 + b12*a8 + b13*n1 + b14*n2 + b15*n3 + b16*n4
+ b17*n5 + b18*n7 + b19*n8
R² = 0.2390
MODEL 1

60. treeheight = b0 + b1*elevation + b2*near_dist + b4*s1 + b5*s2 +
b6*a1 + b7*a2 + b8*a4 + b9*a5 + b10*a6
+ b11*a7 + b12*a8
R² = 0.2057
MODEL 2

61. CONCLUSIONS AND RECOMMENDATIONS

62. • LiDAR data format = large number of observations (n)
• Environmental factors such as elevation, distance to water,
aspect, and soil type significantly affect tree heights in this
highly variable environment
• Model only accounted for 20% of variability – more work is
needed to identify other variables that may contribute
• Complex natural environments are difficult to model
effectively

63. THESIS CONCLUSIONS

64. • LiDAR data collected in forested environments provide
an “embarrassment of riches” for researchers
• Consideration of data processing workflows and
computational limitations should be addressed
• New LiDAR technologies such as terrestrial and flash
LiDAR should be examined and fused with current
airborne collections to provide even more rigorous
datasets

65. THANK YOU!