-
extracted text (extracttext:extracted_text)
-
Using LiDAR Data to Estimate Effective Leaf Area Index,
Determine Biometrics and Visualize Canopy Structure
in a Central Oregon Forest with Complex Terrain
by
Evan Anthony Hayduk
A Thesis
Submitted in partial fulfillment
of the requirements for the degree
Master of Environmental Studies
The Evergreen State College
December 2012
©2012 by Evan Anthony Hayduk. All rights reserved.
This Thesis for the Master of Environmental Studies Degree
by
Evan Anthony Hayduk
has been approved for
The Evergreen State College
by
________________________
Judith Bayard Cushing
Member of the Faculty (Ecology Informatics and Computer Science)
December 20, 2012
________________________
Date
ABSTRACT
Using LiDAR Data to Estimate Effective Leaf Area Index,
Determine Biometrics and Visualize Canopy Structure
in a Central Oregon Forest with Complex Terrain
Evan Anthony Hayduk
Leaf Area Index (LAI), the total one-sided area of leaf tissue per unit ground
surface area, is an important parameter in many ecological models. LAI is
important for determining interception loss, and can be used potentially as a
surrogate for other ecosystem parameters when studying ecosystem processes and
services. Estimation of LAI at the watershed scale is difficult since traditional,
direct destructive methods are cumbersome and possible only on small spatial
scales. Furthermore, estimation of LAI in steep terrain has proven challenging for
indirect methods using tools that utilize lasers to estimate light penetration
through canopies. In this study, digital hemispherical photographs were used to
ground-truth a Light Detecting and Ranging (LiDAR) method of estimating
effective LAI at both the plot and watershed scales using canopy volume from
LiDAR point cloud data. Effective LAI differs from true LAI in that it includes
non-leaf material, such as branches, in the calculation. The LiDAR model seems
to underestimate effective LAI when compared to ground based methods (R2=
0.3346, p<.001) for 19 of the 133 vegetation plots in Watershed 1 of H.J.
Andrews Experimental Forest.
LiDAR data were also used to calculate biometrics (height, crown diameter, and
stem location) of individual trees and to visualize forest structure. When
compared to vegetation surveys completed for all permanent vegetation plots,
82% of live trees were identified using LiDAR data. The results of this work can
be used for modeling throughfall, canopy storage and interception loss for the
watershed, either scaling from branch, to plot, to watershed leaf area or using
allometric equations with the identified individual trees. The visualizations
presented could assist researchers by allowing them to see gaps in the canopy and
assess variability on a subplot scale. Future research includes assessing what
factors affect the accuracy of tree identification and how software programs can
be improved for more accurate tree identification and LAI estimation in complex
terrain.
TABLE OF CONTENTS
Chapter 1- Introduction: Leaf Area Index, LiDAR and Motivation
Introduction……………………………………………………………….2
Leaf Area Index…………………………………………………...5
Direct Methods of LAI Measurement……………………………..5
Indirect Methods of LAI Measurement……………………….......7
What is LiDAR? ……………………………..………………….10
Study Site – The H. J. Andrews Experimental Forest………..….13
Geology…………………………………………………..13
Soils ...…………………………………………………...14
Vegetation and Management History…………….……...15
Harvest History…..………………………………………15
Interdisciplinary Nature of this Research………..………………16
Thesis Organization Overview……………………..……………17
Chapter 2- Estimating LAI from Digital Hemispherical Photographs and
LiDAR Data
Introduction and Background……………………………………………20
Remotely Sensed LAI Measurements……………………………20
LiDAR and LAI Estimation from the Literature……...…………23
Selection of a LAI-LiDAR Model…………………………….…25
Methods…………………………………………………………..………26
Hemispherical Photograph Acquisition………………………….26
Hemispherical Photograph Analysis…………………………..…27
LiDAR Data Acquisition………………………...………………31
LiDAR Data Analysis……………………………………………32
FUSION Executables Used………………………………………33
Volume Calculation for LAI Model………………..……………36
Results……………………………………………………………………37
LAI Results from Digital Hemispherical Photographs using SLIM
Software………………………………………………….37
Simple Linear Regression of LAI Values…………..……………39
LiDAR LAI Model Based on Surface Volume of CHM…...……41
Discussion………………………………………..………………………45
Underestimation of LAI from Hemispherical Photograph
Analysis…………………………………………………………..47
Slope Correction for Hemispherical Photograph Analysis………49
LiDAR LAI Estimates…………………………………………...51
Alternative LiDAR Systems……………………………………..53
Conclusions………………………………………………………………55
Chapter 3- LiDAR used to Describe Stand Characteristics and Identify
Individual Tree Height and Location
Introduction and Background……………………………………………59
Biomass and Carbon Stocks……………………………………...59
Using LiDAR Data to Describe Forest Structure and Cover….…61
iv
Stem Mapping……………………………………………………67
Methods…………………………………………………..………………70
TreeVaW Individual Tree Identification and Stem Mapping……70
H.J. Andrews Vegetation Survey Data……………………..……71
DBH-Tree Height Relationship and Comparison………..………71
Results……………………………………………………………………72
Total Trees Identified in Plots……………………………………72
Vegetation Survey and TreeVaW Comparison by Plot………….73
Discussion……………………………………………………………..…84
New Methods in Tree Extraction from LiDAR Data………….…88
Conclusions………………………………………………………………91
Chapter 4- LiDAR Data Visualization and Overview of Visualization in
Natural Resource Management
Introduction and Background……………………………………………94
Methods…………………………………………………………………101
Visualization of LiDAR Data for Plots and Transects………….102
Visualization of TreeVaW Output Trees……………………….103
Results…………………………………………………………………..103
Watershed 1 Visualizations……………………………………..103
Visualizations of LiDAR Point Cloud and TreeVaW Identified
Trees…………………………………………………….107
Discussion………………………………………………………………114
Visualization in Ecosystem Monitoring and Processes………...116
Landscape Visualization………………………………………..118
Conclusions……………………………………………………………..120
Chapter 5- Summary of Conclusions and Future Research
Objectives Revisited……………………………………………………123
Future Research Directions……………………………………………..125
Literature Cited……………….………………………………………………128
Appendix A: LiDAR Calculated LAI, Surface Volume and Cover*Height Value
for All Permanent Vegetation Plots…………………………………….142
Appendix B: TreeVaW Tree Identification Results for all 133 Permanent
Vegetation Plots………………..…………………………………..…...146
v
LIST OF FIGURES
Figure 1. Study site map showing WS1 at H.J. Andrews Experimental Forest.
Blue circles indicate all permanent vegetation plots in the watershed with
red circles representing plots in which DHPs were taken......................... 12
Figure 2. Trigonometric calculation for scope setting in SLIM software for DHP
photo analysis............................................................................................ 30
Figure 3. Q-Q plots for all LAI results: Plot LAI (right), Wide LAI (bottom left)
and Cardinal LAI (top left). Axes represent theoretical and sample
quantiles. Only Plot LAI is normally distributed. ..................................... 38
Figure 4. Regression analysis for ln(Plot LAI) compared to ln(Wide_LAI). The
relationship is not significant (alpha= .05). .............................................. 39
Figure 5. Regression analysis for ln(Plot LAI) compared to ln(Cardinal LAI). The
relationship is somewhat strong and statistically significant (R2 = 0.3114,
p = 0.013). ................................................................................................. 40
Figure 6. Simple linear regression model of Plot LAI from hemispherical
photographs to LiDAR estimated LAI for each plot. ............................... 43
Figure 7. Simple linear regression between LiDAR calculated LAI and
Cover*Height values for plots in which DHP were taken. ....................... 44
Figure 8. Results of simple linear regression for all plots between LiDAR
calculated LAI and Cover*Height value................................................... 45
Figure 9. Previous estimates of LAI in the FEEL network in Watershed 1 at H.J.
Andrews. ................................................................................................... 47
Figure 10. Transect 1, plot 8. ................................................................................ 74
Figure 11. Transect 1, plot 9. ................................................................................ 74
Figure 12. Transect 1, plot 10. .............................................................................. 75
Figure 13. Transect 2, plot 5. ................................................................................ 76
Figure 14. Transect 2, plot 6. ................................................................................ 76
Figure 15. Transect 2, plot 7. ................................................................................ 77
Figure 16. Transect 2, plot 8. ................................................................................ 77
Figure 17. Transect 2, plot 9. ................................................................................ 78
Figure 18. Transect 2, plot 11. .............................................................................. 78
Figure 19. Transect 2, plot 12. .............................................................................. 79
Figure 20. Transect 2, plot 13. .............................................................................. 79
Figure 21. Transect 6, plot 8. ................................................................................ 80
Figure 22. Transect 6, plot 9. ................................................................................ 80
Figure 23. Transect 6, plot 10. .............................................................................. 81
Figure 24. Transect 6, plot 11. .............................................................................. 81
Figure 25. Transect 6, plot 12. .............................................................................. 82
Figure 26. Transect 6, plot 13. .............................................................................. 82
vi
Figure 27. Transect 6, plot 14. .............................................................................. 83
Figure 28. Transect 6, plot 15. .............................................................................. 83
Figure 29. Transect 6, plot 16. .............................................................................. 84
Figure 30. Transect 6, plot 17. .............................................................................. 84
Figure 31. A comparison of a) the newly developed HSCOI metric, and b) a
CHM. The high point in the CHM is transformed to a low point in the
HSCOI because it would be an area where penetration of LiDAR pulses
would be very low (Lee and Lucas 2007) . ............................................... 89
Figure 32. A diagram representing new types of visual analysis of LiDAR data
could occur because of research by Kao et al. (2005). ............................. 98
Figure 33. FUSION visualization of WS1 Digital Elevation Model (DEM) and
LiDAR data of 19 vegetation plots where digital hemispherical
photographs were taken. ......................................................................... 104
Figure 34. Overhead view visualization of WS1 DEM and LiDAR data of
vegetation plots. ...................................................................................... 105
Figure 35. Overhead view visualization of WS1 DEM and LiDAR data of
vegetation plots with digital orthophotograph of vegetation in the
Watershed overlaid. ................................................................................ 106
Figure 36. Visualization of LiDAR data and DEM of vegetation plots in transect 2
(above) and transect 6 (below). ............................................................... 107
Figure 37. Transect 1, plot 9. .............................................................................. 108
Figure 38. Transect 1, plot 10. ............................................................................ 108
Figure 39. Transect 2, plot 5. .............................................................................. 109
Figure 40. Transect 2, plot 6. .............................................................................. 109
Figure 41. Transect 2, plot 7. .............................................................................. 109
Figure 42. Transect 2, plot 8. .............................................................................. 110
Figure 43. Transect 2, plot 9. .............................................................................. 110
Figure 44. Transect 2, plot 12. ............................................................................ 110
Figure 45. Transect 2, plot 13. ............................................................................ 111
Figure 46. Transect 6, plot 8. .............................................................................. 111
Figure 47. Transect 6, plot 9. .............................................................................. 111
Figure 48. Transect 6, plot 10. ............................................................................ 112
Figure 49. Transect 6, plot 11. ............................................................................ 112
Figure 50. Transect 6, plot 12. ............................................................................ 112
Figure 51. Transect 6, plot 13. ............................................................................ 113
Figure 52. Transect 6, plot 14. ............................................................................ 113
Figure 53. Transect 6, plot 15. ............................................................................ 113
Figure 54. Transect 6, plot 16. ............................................................................ 114
Figure 55. Transect 6, plot 17. ............................................................................ 114
vii
LIST OF TABLES
Table 1. Resulting resolution and accuracy of LiDAR data. ................................ 32
Table 2. LAI and gap fraction results from SLIM software. Plot LAI was
calculated with a limited scope of view to capture LAI within each
vegetation plot. Wide LAI used the default 60 degree scope for calculation
and Cardinal LAI is the average of the four cardinal direction LAI values
with a 60 degree scope. ............................................................................. 37
Table 3. Plot LAI, Volume of Canopy Height Model, and LiDAR calucalted LAI
for each plot studied. ................................................................................. 42
Table 4. DBH-Height asymtotic equation and regression coefficients used for
each species (Garman et al. 1995). ........................................................... 72
Table 5. Observed trees, trees predicted by TreeVaW and percentage of trees
identified by TreeVaW in plots where digital hemispherical photographs
were taken. ................................................................................................ 73
viii
ACKNOWLEDGEMENTS
I would especially like to thank my reader, Judy Cushing, for all the guidance and
expertise she provided throughout my time in the MES program. A special thanks
to the last minute heroics of Fox Peterson, for helping in the final editing of this
thesis. Also, thank you to Barbara Bond and Scott Allen for the framing of this
research. Also, I would like to greatly thank everyone who assisted me in the
completion of this thesis, including:
Jerilyn Walley, Nalini Nadkarni, Mark Schulze, Lee Zeman, Anne McIntosh,
Adam Kennedy, Kirsten Winters, Jeff Richardson, Robert McGaughey, Soren
Popescu, Jason Cornell, Lorrain Garr, Carri Leroy, Martha Henderson, Gerardo
Chin-Leo, Ralph Murphy, my entire MES cohort and all the other MES’ers.
This thesis research is a part of the VISTAS project. The VISTAS project is
supported by the U.S. National Science Foundation/ BIO/DBI 1062572. Any
opinions, findings, and conclusions or recommendations expressed in this
material are those of the author and do not necessarily reflect the views of the
National Science Foundation. Data sets were provided by the H.J. Andrews
Experimental Forest research program, funded by the National Science
Foundation's Long-Term Ecological Research Program (DEB 08-23380), US
Forest Service Pacific Northwest Research Station, and Oregon State University.
A final and thank you to my wonderfully patient and beautiful partner Jen Motley
and my family.
ix
Chapter 1- Introduction: Leaf Area Index, LiDAR and Motivation
The major motivation for this thesis, which focuses on estimating Leaf Area
Index and related visualizations for Watershed 1 (WS1) in the H.J. Andrews
Experimental Forest, is rooted in the Visualization of Terrestrial and Aquatic
Systems (VISTAS) project. VISTAS is a collaboration among computer and
social scientists, ecologists, and students at The Evergreen State College, Oregon
State University, and the University of Minnesota. The project was born of
necessity: to solve problems apparent in the current age of massive data stores and
grand challenge environmental science. Adding to the challenge of addressing
critical environmental science questions in the presence of massive amounts of
data are difficulties inherent when dealing with multiple spatial and temporal
scales, complex physical terrain, and highly distributed and heterogeneous data
(VISTAS Summary). The VISTAS project posits that visualizing natural
phenomena can help scientists develop intuition and hypotheses at multiple spatial
scales, thus improving their ability to formulate new insights about ecosystem
services, patterns and processes in complex systems and to communicate these
insights to the wider community (VISTAS Summary). The objectives of the
VISTAS project are:
1) Conduct EcoInformatics research to enable the required visual analytics
and implement a proof of concept software tool
2) Co-develop VISTAS software with environmental scientists who will use
it in studies spanning spatial and temporal scales
3) Apply social science methods to study the co-development and usability of
VISTAS and its visual analytics.
2
Barbara Bond, co-principle investigator (coPI) of the VISTAS project and
former PI of the Andrews Long Term Ecological Research (LTER) site, and her
students have studied ecohydrological function in WS1 for decades. Her former
Master’s student, Scott Allen, now a PhD candidate at Louisiana State University,
recently completed work in WS1. Allen and others’ work examines the influence
of spatial patterns of canopy storage, interception, and throughfall on isotopic
composition of water in the watershed. A portion of his work is related to prior
research including that by Keim et al. (2005) which used simulated rainfall on
harvest branches to determine the storage capacity of branches with a known leaf
area. This branch scale leaf area can then be scaled up to the tree level to estimate
the capability of individual trees to intercept and store precipitation (Waring et al.,
1977; Gholz et al., 1979; Laurenroth et al., 1993; Bond et al., 2002). Scaling
Keim’s work to the watershed level requires accurate measurement of Leaf Area
Index for the watershed, the third objective of this thesis project (stated below).
The main objectives of this thesis are as follows:
1) Determine accurate LAI estimates for a subset of permanent vegetation
plots in WS1 using digital hemispherical photography (DHP)
2) Use estimates of LAI obtained from DHP (1) to build a LiDAR based
model of LAI for all 133 permanent vegetation plots
3) Calibrate the LiDAR based model (2) for LAI to create LAI maps for the
entire watershed
4) Test the ability of software programs to extract and identify individual
trees from LiDAR data in all 133 permanent vegetation plots in WS1
3
5) Create novel visualizations of individual trees in the vegetation plots using
LiDAR data
LAI is not only important for measuring interception loss, but as a potentially
easy way to measure ecosystem parameters that can be used as surrogates for
calculating ecosystem processes and services. For example, relationships between
LAI and biomass may facilitate the calculation of carbon (C) stores in forests, and
relationships between LAI and stomatal aperture may facilitate the calculation of
gas exchange and C fluxes from forests. Additionally, color-identification
software may be used in conjunction with LAI to produce estimates of forest
albedo, which can be used to study global heat flux, and the potential for forests
to contribute to or protect us from, varying wavelengths of radiation. LAI may
also be directly related to foliar biomass, which is often useful in biogeochemical
analyses over time as a proxy for plant nutrient allocation, indicative of soil
nitrogen (N) concentrations, and potentially related to rates of root development
and turnover. Most importantly, LAI may be used in all of the above ways to aid
in calculation of global net primary production (NPP) and net ecosystem
exchange (NEE), which dictate the planet’s ability to mitigate changes in
atmospheric C concentrations. Accurate NPP and NEE calculations could be
particularly useful from a management standpoint as they allow the assignment of
a relative economic value to forests for their service as a carbon sink, may aid in
the quantification of ecosystem services, help with management decisions, and be
a useful measure for policy and decision makers to compare forest capacities
worldwide.
4
Leaf Area Index (LAI)
Leaf area index (LAI) was first described by Watson (1947) as the “total onesided area of leaf tissue per unit ground surface area." Watson labeled LAI as a
dimensionless unit that characterizes the canopy of a forest stand. LAI determines
canopy water interception and radiation extinction, thereby influencing water and
carbon gas exchange, within- and below- canopy microclimates, and aiding in the
understanding of biogeochemical cycling in ecosystems (Breda, 2003). Accurate
measurements of LAI can contribute to a better understanding of water resource
dynamics and photosynthetic productivity on an individual, stand, or watershed
scale (Breda, 2003).
Since Watson (1947) first defined LAI, various techniques have been used
to measure LAI in ecosystems from simple cropland to dense, complex temperate
rain forests, and have also been summarized in review articles such as Breda
(2003), Jonckheere et al. (2004) and Weiss et al. (2004). Breda (2003) breaks
down in situ LAI measurement methods into two different categories: direct and
indirect measurements.
Direct Methods of LAI Measurement
Direct methods of determining LAI include harvesting, allometric methods and
litter collection. Harvesting is a destructive method of measurement that requires
harvesting and measuring all vegetation within a given area. Leaf samples are
collected from a site that has a known ground-surface area and then dried and
weighed. LAI is then computed by multiplying leaf dry mass (g m-2) by the
specific leaf area (m2 g-1 SLA). Since m2 and g are present in both the numerator
5
(leaf area) and denominator (ground area) these measures cancel out, leaving LAI
as a unitless value. This method is used extensively with crop species, but the
destructive nature and tedious approach is not applicable to large areas, especially
forest stands with large trees.
A less destructive method of sampling is the application of allometric
relationships between directly (destructively) measured sapwood and leaf area of
individual trees. The theory supporting this technique is based on the hypothesis
that leaf area is directly proportionate to conducting tissue (Grier and Waring,
1974; Makela, 1986; Waring et al., 1977). Consequently, allometric relationships
are sensitive to site and species, and in extreme cases, to the year of measurement.
Breda (2003) also recommends replacing sapwood area as the base for
measurement due to the difficulty in measuring conductive area. Using easily
measurable variables, such as diameter at breast height (DBH), is a more efficient
way to measure LAI using allometric relationships, and can be done without
directly removing a core sample from trees to quantify sapwood.
The final direct method described by Breda (2003) consists of collecting
leaves as litter in traps distributed below the canopy during leaf fall. This
technique is used widely in deciduous stands that shed leaves during the fall
season. Litter is collected in a set number of traps within a known collection area,
and harvested often to avoid loss of litter and decomposition. The collected litter
is dried and weighed, then scaled to trap size in m2 to compute the dry mass of
litter as g m-2. Leaf dry mass is then converted to leaf area by multiplying the
collected biomass by the Specific Leaf Area (SLA). SLA is found by measuring
6
the projected leaf area of leaves and dividing that value by the leaf area dry mass
of the leaves. SLA relates biomass to cover and is indicative of photosynthetic
processing and C allocation (Evans and Poorter, 2001). To obtain LAI, the total
accumulated leaf area spanning collection times is calculated by adding together
the leaf area calculations for each trap and from each collection and scaling to the
plot or landscape scale.
Indirect Methods of LAI Measurement
Jonckheere et al. (2004) describe the benefits of indirectly measuring LAI
whereby leaf area is inferred from one or more other variables. Indirect methods
are faster and amendable to automation and thus allow for larger spatial samples.
Additionally, these techniques do not require destructive sampling and can be
applied in areas where destructive sampling is not possible. Indirect methods fall
into one of two categories: indirect contact methods and indirect non-contact
methods.
The first indirect contact method described by Jonckheere et al. (2004) is
the inclined point quadrat technique used by Wilson (1960, 1963). According to
Jonckheere (2004) this method “consists of piercing a vegetation canopy with a
long thin needle (point quadrat) under known elevation (i.e., the angle between
the needle and the horizontal plane when vertically projected) and azimuth angles
(i.e., the bearing of the needle from North when horizontally projected) and
counting the number of hits or contacts of the point quadrat with ‘green’ canopy
elements.” These data are then input into a simple equation based on a radiation
penetration model:
7
LAI= 1.1 x N(32.5)
where N (32.5) is the number of contacts with an elevation angle of 32.5 degrees.
The more times the needle is dropped into a vegetative canopy, the more reliable
are the estimates of LAI. When the needle is dropped repeatedly with different
angles of insertion, the formula used becomes:
Ni = LKi
where L is LAI, Ni is the number of contacts of the needle dropped with elevation
i and Ki the extinction coefficient with elevation i. This method, according to
Jonckheere et al. (2004) is attractive because it does not require an assumption of
random leaf distribution, but its major drawback is the considerable field work
involved, often requiring over 1000 insertions to obtain a reliable estimate of LAI.
This method is also not applicable in canopies taller than 1.5 m, and thus it has
been limited to determination of LAI on cropland.
Jonckheere et al. (2004) also consider allometric techniques for the
measurement of LAI, as mentioned above, which they categorize as an indirect
contact method, and which rely on destructive sampling to measure sapwood, or
non-destructive measurement of basal area or diameter at breast height (DBH).
The destructive sampling of sapwood would fall into the direct methods category,
mentioned above. Measurement of other variables such as DBH and basal area are
8
considered indirect non-contact due to the non-destructive nature of the
measurements. Other indirect non-contact measurements of LAI are based on
measuring light transmission through canopies and are more commonly
implemented.
Indirect non-contact methods apply the Beer-Lambert law, which accounts
for the total radiation intercepted by canopy layers and depends on incident
irradiance, canopy structure, and optical properties. Indirect non-contact methods
require forest floor based measurement of total, direct, and/or diffuse radiation
transmittance to the ground. In the last few decades, a wide range of instruments
have been developed to measure LAI in real time within plant canopies. These
instruments use either gap fraction analysis, or gap size distribution analysis, to
determine LAI. Devices such as the Digital Plant Canopy Imager CI 100 MVI,
measure gap fraction by incorporating canopy image analysis techniques. Other
devices, such as Accupar, Demon, Licor LAI-2000 Plant Canopy Analyzer, use
gap fraction and calculate LAI by comparing differential light measurements
above and below the canopy (sensu Cutini, 1998). However, the maximum
measurable LAI is lower for devices that measure gap fraction, because LAI
reaches an asymptotic saturation level at a value of 5, which causes gap fraction
saturation as LAI approaches and exceeds 5-6. Other devices use gap size
distribution to measure LAI, including the Tracing Radiation and Architecture of
Canopies (TRAC) instrument.
Hemispherical photography also uses gap size distribution to estimate LAI
in forest canopies. Jonckheere et al. (2004) focus extensively on the use of
9
hemispherical photographs in their review of LAI measurement methods, and
state that a hemispherical photograph provides a permanent record and is
therefore a valuable information source for position, size, density and distribution
of canopy gaps. Hemispherical photography requires the use of a fish-eye lens to
photograph the canopy from the forest floor with the camera oriented towards
zenith. This technique also captures the species-, site- and age-related differences
in the architecture of canopies based on light attenuation and the difference in the
photograph of light (sky) and dark (canopy).
TRAC and hemispherical photography, which use gap size distribution to
measure LAI, do not distinguish photosynthetically active leaf tissue from other
plant elements such as branches, stems and flowers. For this reason, alternative
terms for LAI have been proposed, with Chen and Black (1992) settling on the
most widely used term: “effective LAI”.
What is LiDAR?
A description of Light Detecting and Ranging (LiDAR) systems and LiDAR data
is found in the FUSION software manual (McGaughey, 2012): LiDAR systems
use laser light to measure distances between the source of the LiDAR and the
object(s) surveyed. Aerial laser scanning is an aircraft based LiDAR system
which provides accurate, detailed 3D measurement of ground, vegetation, and
buildings. In open areas, ground contours can be measured within 6 inches of
actual elevation. In steep, forested areas accuracy is typically .3m to .6m,
depending on density of canopy cover and the spacing of the laser pulses. The
speed and accuracy of LiDAR systems allow for highly detailed mapping of large
10
areas previously possibly only with time-consuming and expensive ground
surveys (McGaughey, 2012), and are increasingly being used to infer forest
biometrics.
Aerial LiDAR systems are mounted on a single- or twin-engine plane or
helicopter for data collection over a large area. An aerial LiDAR system consists
of four pieces of equipment: 1) a laser emitter/receiver scanning unit on the
aircraft, 2) global positioning system (GPS) units on the aircraft and on the
ground, 3) an inertial measurement unit (IMU) attached to the scanner to measure
roll, pitch, and yam of the aircraft, and 4) a computer system to control the entire
system and store the data. Several types of LiDAR systems exist, with the most
commonly used version for forestry being discrete-return, small-footprint
systems. The term small-footprint refers to the size of the laser beam diameter at
ground level, typically from .02m to 1m wide. Up to 200,000 pulses of light per
second are emitted by the laser scanner, and the time it takes for the pulse to
return to the receiver is measured. The times are used to compute the distance to
the object on the ground, with the GPS and IMU units used determining the
precise location and attitude of the laser scanner as the pulses are emitted. All this
information is used to calculate exact coordinates for each point. Depending on
the unit, an oscillating mirror or rotating prism on the laser scanner is used to
sweep light pulses across a wide swath of the landscape. Large areas are surveyed
using a series of parallel flight paths. The only weather conditions required are
clear skies and flights can be performed day or night since the system emits its
own light (Lefsky et al., 2002; Lefsky et al., 2005).
11
The laser altimetry calculated by the aerial system yields direct 3D
measurements of the ground surface, vegetation, roads, and buildings. The
millions of data points create a 3D point cloud. After the flight, additional
calculations are performed to create the final data points, and results can be
produced in weeks to months. The initial acre of the LiDAR flight is expensive,
accounting for the costs of the aircraft, equipment and personnel. However, when
large areas are measured, the costs can drop to as low as $1 to $2 per acre.
Figure 1. Study site map showing WS1 at H.J. Andrews Experimental Forest. Blue circles indicate all
permanent vegetation plots in the watershed with red circles representing plots in which DHPs were
taken.
12
Study Site – H.J. Andrews Experimental Forest
The site where this thesis research was conducted is Watershed 1 on the H. J.
Andrews Long Term Ecological Research site near Blue River Oregon, USA, in
the Cascade Mountains (Figure 1). Study site information for Watershed 1 (WS1)
was obtained from the H.J. Andrews website (andrewsforest.oregonstate.edu/).
WS1 is a small watershed of 96 ha located on a first order stream draining to
Lookout Creek along the McKenzie River in the Western Cascades range of
Oregon. The bounding coordinates (in decimal degrees) for WS1 are North:
44.20851700, South: 44.19901700, East: -122.23581300, and West 122.25683100. WS1 is a well-known low-elevation watershed nestled within the
greater H.J. Andrews Experimental forest and exhibits "complex terrain"
characteristic of the region. The minimum elevation on WS1 is 450 meters,
maximum elevation is 1027 meters, the mean slope measured using ground based
clinometry is 59.35% and the watershed outlet faces an aspect of 286 degrees.
Mean January temperature is 35 F (1.6 C), mean July temperature is 69 degrees F
(20.6 C), although heating and cooling patterns are asymmetric, with cold air
pooling occurring on approximately 80% of summer nights (Rothacher, 1965;
Pypker et al., 2007).
Geology: Swanson and James (1975) describe the geology of WS1 and the H.J.
Andrews (HJA). The HJA is underlain by bedrock of volcanic origin of mixed
mineralogy. Three geologic formations have been mapped for the HJA and
correspond roughly with elevation. Little Butte Formation bedrock
(approximately 760 m elevation), dated as Oligocene to lower Miocene, consists
13
of massive tuffs and breccias derived from mudflows and pyroclastic flows.
Sardine Formation bedrock (760 m to 1200 m), dated as middle to late Miocene,
consists of two units: a lower unit containing welded and non-welded ash flows,
and an upper unit containing basalt and andesite lava flows. Andesitic and basaltic
lava flows (>1200 m), or the "Pliocascade" Formation, were deposited during the
Pliocene and overlie Sardine Formation material. Watershed 1 spans the Little
Butte-Sardine contact area. A large caprock, visible in aerial photography,
particularly after harvest, and in LiDAR bare earth images, demarcates a distinct
separation in mineralogy observed by Peterson (2012). Watershed 1 has both
basaltic and andesitic mineralogy: in the extreme northeast corner of WS1, the
upper elevations are underlain by deposits of Sardine andesitic flow rock. Most of
the bedrock of this type in WS1 is slow to weather, displaying rugged
escarpments and outcroppings (Swanson and James, 1975). However, recent
research by Pett-Ridge (Peterson, personal communication) suggests that
weathering rates on parts of WS1 may be relatively rapid; these analyses are
based on silicate material losses in dissolved water and stream outflow.
Soils: Soils have not been formally mapped for WS1, although Rothacher (1965)
characterized several potential soil groups. Additionally, several soil pits have
been dug on the watershed (1957, mid-1960's, 1982, 2002, 2010) which indicate
high heterogeneity in soil physical and chemical characteristics (Peterson and
Lajtha, 2012). Land movements of varying duration and intensity are common on
the HJA; glacial, fluvial and mass wasting processes are the main factors affecting
soil development and spatial distribution, especially in geologic contact zones and
14
steeper areas. In WS1, soils vary from shallow and stony to moderately deep with
well-developed profile features. These soils may affect the establishment and
survival of vegetative communities.
Vegetation and Management History: Rothacher et al. (1967) describe the prelogging condition of WS1. Douglas-fir (Psuedotsuga menziesii) was the dominant
species, ranging in age from 100 to 500 years. Western Hemlock (Tsuga
heterophylla) was intermixed and generally younger; some western red-cedar
(Thuja plicata) was also present, mostly in drainage areas. Pacific yew (Taxus
brevifolia) was present in the understory. Hardwood species were common in
stem density, but had relatively low biomass compared to conifers. However,
hardwoods were first to establish on many plots, and created environments on
which conifers would later thrive. Current hardwoods, in terms of relative
abundance (number) and in decreasing order are big-leaf maple (Acer
macrophyllum), Pacific dogwood (Cornus nuttallii), golden chinkapin
(Chrysolepis chrysophylla), and red alder (Alnus rubra). Some of these, such as
Alnus rubra, are found mixed within the coniferous plots, while others, such as
Chrysolepis chrysophylla, clearly dominate dry plots. Six understory plant
communities were present: 1) hazel-salal (10% of watershed area), 2)
rhododendron-salal (10%), 3) vine maple-salal (10%), 4) vine maple-Oregon
grape (25%), 5) gold-thread (25%) and 6) sword-fern (20%).
Harvest History: Watershed 1 was originally part of a paired watershed study with
a control watershed (WS2) and a patch clear cut watershed (WS3) of similar size
and topographic characteristic. WS1 was 100% clear-cut over four year period
15
from fall 1962 to summer 1966. Initially, it was suspected that high lead logging
could be used, but the method was switched to skyline following the harvest of
one unit. Even the skyline logging techniques of the day, however, proved to be
difficult for the initially hired group, Ballinger Logging, and Swiss contractors
(Wyssen Logging) were called in to construct a method for harvesting larger
trees. During this time between logging events, regrowth of understory species,
specifically Ceanothus spp occurred, which may have impacted soil conditions.
After the entire watershed was harvested, it was burned in 1967 in a burn that was
"hot and satisfactory," and all debris littering the stream was removed. Since
1952, the watershed stream outflow has been monitored and regrowth inventoried
approximately on a six year basis. These inventories and stream measurements
are the fundamental components of the current biomass and nutrient studies
conducted on WS1 to assess the impacts of harvest on watershed productivity and
nutrient budgets (Peterson 2012).
Interdisciplinary Nature of this Research
This thesis follows the VISTAS project with an interdisciplinary approach to
completing stated objectives. The work’s major goals are those related to ecology
and pose fundamental questions about how forest ecosystems function. However,
the methods used to answer these questions are based in the computer sciences,
including computer programming and data visualization, and the researcher wears
both hats to carry out the work. It would be difficult for a computer scientist to
complete field work to acquire hemispherical photographs that are later analyzed
for LAI calculations. Similarly, it would be difficult for most ecologists to write
16
code to create an executable program, or even to use the complex software needed
for such work. Furthermore, the results of this study can be used to accomplish
goals in several disciplines. The estimates of LAI for the watershed can be used
by ecologists to determine how precipitation interacts with the canopy, either
being stored or falling as throughfall, and as the first check to the water balance of
the system. The estimates of total trees in the plots and watershed can be used for
biogeochemical cycling models to determine nutrient transport and carbon
sequestration within the watershed. The methods of visualization and tree counts
could be used by natural resource managers for decision making. And, finally,
the suggestions offered for improvement of the LiDAR analysis and visualization
software can be used by computer scientists to improve these tools, and to build
even better tools in the future.
Thesis Organization Overview
This thesis is organized into four chapters as follows. In Chapter 2, the use of
remote sensing to estimate LAI is explored. Then, we describe how we used
digital hemispherical photography (DHP) of 19 plots to estimate LAI for all 133
plots in WS1, and how we subsequently used LAI estimates from those plots to
‘ground-truth’ a LiDAR based model to determine LAI for the remaining plots
and the watershed as whole. In Chapter 3, we present an overview of how LIDAR
has been used to estimate canopy and stand characteristics. We then describe how
we used the TreeVaW software to extract and identify individual trees in all 133
permanent vegetation plots in WS1. TreeVaW results for the 19 plots we
surveyed are compared to comprehensive vegetation surveys completed at about
17
the same time as the LiDAR data were collected; and TreeVaW and survey data
of individual tree height are compared. Chapter 4 provides background
information about forest data visualization and then presents visualizations of
LiDAR data from WS1 created with the FUSION software, as well as
visualization of trees identified in 19 plots by TreeVaW. Chapter 5 concludes this
thesis, and proposes future directions for related research.
18
Chapter 2- Estimating LAI from Digital Hemispherical Photographs and
LiDAR Data
19
INTRODUCTION AND BACKGROUND
Remotely Sensed LAI Measurements
Breda (2003) chose to consider only ground based measurements of LAI, but
noted that remotely sensed vegetation indices, either from satellite or aerial highresolution imagery, have novel potential for estimating LAI at larger scales. His
review mentions that remotely sensed indices at the time (2003) required a siteand stand-specific calibration with ground-based measurements of LAI since
remote sensing in itself does not yield accurate LAI measurement for complex
canopies, especially those with high LAI values. Jonckheere et al. (2004) also
describe air- and space-borne methods for LAI determination at the forest and
landscape level, but note that the description of those techniques were beyond the
scope of their review.
Zheng and Moskal (2009) offer a complete review of estimating LAI
using remote sensing techniques including theoretical background, methods used
and sensors utilized. This paper covers the ground based methods discussed
above, but focuses on the use of remote sensing technologies and includes a
discussion of sources of error and scaling issues. Aerial passive sensors are
superior to images obtained from satellites because of their much finer spatial
resolution. Unfortunately greater resolution produces in shadows from tree
canopies that obscure each other and adds bias to estimation of LAI and makes
simulating a radiation regime difficult without using a geometric optical model.
In a forest canopy, different angular distributions of foliage elements
result in solar radiation interaction with foliage at four different scales: a) within
20
groups of trees, b) within individual tree crowns, c) within branches, and d) within
shoots (Zheng and Moskal, 2009). To correctly estimate LAI, a geometric-optical
model must use the calculated shape of canopy crowns and spatial distribution of
canopy elements. The proportion of shadows cast as a function of view direction,
relative to the hot spot direction, are then calculated, and spectral characteristics
are obtained based on geometrical shape and arrangement. Finally, the spectral
reflectance of individual trees or whole canopies can be calculated with the
geometric optical model (Zheng and Moskal, 2009).
Satellite retrieval of LAI measurements is based on the unique spectral
response characteristic of green leaves as opposed to other materials such as bark.
The absorption of solar radiation of green leaves, with high absorption of visible
light and red light make vegetation indices such as Simple Ratio (SR),
Normalized Difference Vegetation Index (NDVI), Enhanced Vegetation Index
(EVI) and the Reduced Simple Ratio (RSR) possible (Zheng and Moskal 2009).
Each of these indices has certain advantages over the others, depending on the
forest or ecosystem type being evaluated.
Other passive sensors discussed by Zheng and Moskal (2009) are Landsat
series sensors and hyperspectral remote sensing data sets. The Landsat series
sensors, including thematic mapper and enhanced thematic mapper are used most
often because of the balance of spectral, spatial and temporal resolution possible
with the sensors. These sensors have been developed to estimate and map LAI at
landscape and global levels, based on linear regression relationships between
21
vegetation indices and LAI, as well as on the linear and non-linear estimation
model.
The final type of remotely sensed LAI methods reviewed by Zheng and
Moskal (2009) are those that use active sensors that emit a certain wavelength
signal and capture the echoes reflected by target objects without receiving the
reflected solar radiation by land surfaces. Radio Detection and Range (RADAR)
and Light Detection and Ranging (LiDAR) are the two most commonly used
active remote sensing systems. LiDAR and other active remote sensing have a
distinct advantage over optical passive remote sensing in their ability to capture
detail of three-dimensional structure of the forest. The passive systems can only
provide two-dimensional information. LiDAR systems can be terrestrial, airborne,
or satellite-based, and either discrete or full waveform. Discrete LiDAR systems
provide single or multiple returns for each laser pulse, while the full waveform
LiDAR provides the waveform for one pulse. Terrestrial LiDAR is a discrete
system, yet only one return is recorded for each laser pulse, while most other
discrete systems have three or more echoes bounce back for each laser pulse.
Discrete systems measure the distance between the emitting sensor and objects by
recording the time of flight of the laser. Each laser pulse returns two different
types of information: spatial information in x,y,z coordinates and a corresponding
intensity value. Aerial, discrete LiDAR systems have been used extensively to
characterize and explore forest canopies.
22
LiDAR and LAI Estimation from the Literature
One of the more recent methods to determine LAI of forest stands uses LiDAR
data. Morsdorf et al. (2006) evaluated the potential of aerial discrete return
LiDAR systems to derive fractional cover and LAI for a stand in Switzerland. The
objective of their work was to establish a predictor variable of LAI that resembles
the way LAI is estimated using indirect methods in the field. Morsdorf et al.
(2006) started with an equation from previous literature (Weiss et al. 2004) in
which LAI at a certain heights is a function of leaf density. The authors added
contact frequency as a predictor and set the projection function to 0.5, assuming a
spherical foliage distribution in a study area dominated by conifer species. The
LAI proxy from LiDAR data used by Morsdorf et al. follows:
where EFE, ELE, and ESE stand for first echo, last echo, and single echo, which
were recorded by the LiDAR system used by the authors. The authors used the
results of the LAI model to produce maps of LAI for the study area.
In similar research, Riano et al. (2004) also estimated LAI and canopy
properties, as well as covered ground using hemispherical photography in three
oak (Quercus myrenaica) and eight pine (Pinus sylvestris) forest plots in the
Sierra de Guadarrama mountains of central Spain. The purpose of their research
was to assess the capacity of LiDAR data to estimate LAI and covered ground at
different spatial scales. To accomplish this, LiDAR data were processed for
23
different radii, from 0.5 to 2.5 meters (0.5 meter increments) and from 2.5 to 20
meters (2.5 meter increments). The LiDAR predictive variables were: 50, 75 and
90 percentile of heights, average height, maximum height and percentage of
canopy hits (returns above 3 meters). The best LAI prediction radius for the oakdominated forests was from 7.5-10 meters, while the prediction radius for pine
forests was 10-12.5 meters. Overall, estimations of LAI for the oak forests were
more accurate than those for the pine forests. The results of the regressions of
hemispherical LAI and the LiDAR parameters were then used by the authors to
map LAI and covered ground for both the oak and pine forest stands.
Solberg et al (2006) examined the use of LiDAR derived gap fraction
measurements taken before and after an insect outbreak in a Scots pine stand in
Norway. The authors compared LAI obtained from a LICOR 2000 device and
digital hemispherical photographs with gap fraction values obtained with discrete
return LiDAR data. By ground-truthing the discrete return LiDAR data, Solberg
et al. (2006) validated the hypothesis that airborne laser scanning could be used to
map defoliation at high spatial resolution over large areas. The authors concluded
that this application of LiDAR data could be used for: 1) ad-hoc mapping of acute
forest damage, 2) routine monitoring of crown density in stands, and 3) producing
large-scale ground-truth data sets for satellite surveys.
Lim et al. (2003) focused LiDAR and LAI research on sugar maple (Acer
saccharum Marsh.) and yellow birch (Betula alleghaniensis Britton) stands in the
Turkey Lakes watershed of Sault Ste. Marie, Ontario, Canada. LiDAR derived
metrics for the research included maximum laser height, mean laser height, and
24
mean laser height calculated from LiDAR returns based on a threshold applied to
the intensity of return values. Maximum laser height was found to be the best
estimator of LAI for the study area. The authors concluded that ‘laser height
metrics’ are a viable option for estimation of plot heights, stem density,
aboveground biomass and volume, and other canopy related measures, including
LAI.
Selection of a LAI-LiDAR Model
Research by Richardson et al. (2009), reviewed four models from the literature for
estimating LAI using aerial LiDAR in the Washington Park Arboretum in Seattle,
WA. The four LAI models reviewed used LiDAR metrics based on the mean
elevation of returns (Lim et al. 2009), fraction of canopy returns to total returns
(Riano et al. 2004), the ratio of returns above 2m to returns below 2m (Solberg et
al. 2006), and a canopy volume metric (Lefsky et al. 1999). Similar to work done
by Coops et al. (2009), Richardson et al. (2009) adapted the methods of Lefsky et
al. (1999) to use discrete return LiDAR rather than the full waveform LiDAR
used by Lefsky et al. (1999). These four models were considered for the
determination of LAI from LiDAR data for this project. The model created by
Lefsky et al. (1999) was chosen for two reasons: 1) the research took place in a
similar forest type similar (namely, watershed 1(WS1) at H.J. Andrews), and 2)
the range of viable LAI produced by the model fell within the expected range of
LAI for WS1 (Nadkarni, personal communication). The other models were
calibrated for different forest types and produced lower LAI values: 0.5-4.0 for
Lim et al. (2003), 0-3 for Riano et al. (2004) and 0-1.6 for Solberg et al. (2006).
25
Richardson et al. (2009) noted that these models performed poorly in “saturated”
conditions where LAI is high.
METHODS
Hemispherical Photograph Acquisition
For the research reported in this thesis, Digital Hemispherical Photographs
(DHPs) were taken over two days in May 2012 in Watershed 1 at the H.J.
Andrews Experimental Forest. The camera used was a Canon PowerShot S3IS
with an Opteka for Canon 58 mm super wide fisheye lens (0.20X). DHPs were
taken early in the morning and late in the evening when the sun was below the
horizon or otherwise blocked by the surrounding terrain, since uniform overcast
skies were not present. Uniform overcast skies are needed for accurate analysis to
limit sunflecks and direct sunlight affecting the photograph. The camera setting
mode button was adjusted to F1 function, the setting used for hemispherical
photographs with a resulting circular image. The settings were also set to P, or
programmed automatic, so that the camera automatically adjusted the shutter
speed to the amount of light available. The camera was also set to ‘auto-bracket’
mode, which takes three photographs at three slightly different exposures:
underexposed, normal and overexposed. The DHP showing the greatest contrast
between foliage and sky was then used for analysis.
The camera was mounted to a tripod and leveled with a bubble level
roughly one meter above the ground level. The top of the camera was oriented to
the north with a compass, so that the top of the resulting DHP was due north. The
10-second self-timer function was utilized to limit shaking of the camera while
26
depressing the picture button. Five sets (3 auto-bracketed photos for each) DHPs
were taken for each 9m radius plot: one in the center of the plot, and four in each
cardinal direction (N, S, E, and W) roughly 4.5 meters from the center of the plot.
Immediately after taking a set of DHPs, the photos were transferred to a
computer and coded according to how other data in the watershed were collected.
Information included in the code was plot [P], unit [1], watershed [1], transect
[1,2,6], plot number [01-26], direction [C-center, N-North, E- east, S- south, Wwest], and camera exposure [N-normal, U-underexposed, O-overexposed]. For
example, a normal exposure DHP taken in the center of Plot 8 of transect 1 would
be coded as P1108CN.
DHPs were taken in 21 of 133 permanent vegetation plots. Before
analysis, two plots (P11108 and P11211) were removed from the data due to low
light conditions (P11108) and high light conditions (P11211) at the time the plots
were photographed. These plots were included in the stem mapping and
comparison of stem counts derived from LiDAR data with TreeVaW and actual
tree surveys in the plots (See chapter 3).
Hemispherical Photograph Analysis
DHP analysis was completed with the SLIM (Spot Light Interception Model)
software package
(www.ualberta.ca/~pcomeau/Light_Modeling/Lite_and_slim_intro.html)
developed by Phil Comeau. The program is part of the SLIM/LITE software
package. SLIM is designed to estimate LAI, gap fraction, and fractional
transmittance from DHPs.
27
Analysis of the DHPs is subjective (Comeau, SLIM). SLIM analyzes
DHPs by classifying each pixel as either a sky or canopy pixel. This converts the
true color digital photograph into a black and white, ‘thresholded’ image. Since
uniform sky and light conditions were not present when the DHPs were taken,
each DHP was analyzed individually for the light and sky conditions at the time
each photo was taken by hand-selecting areas of the DHP that were sky or
canopy. Five left clicks on the mouse were used to denote sky and five right clicks
denoted canopy. The resulting image was visually analyzed to make sure that fine
details, such as a single branch, were visible in the thresholded image. If fine
detail was not visible, the brightness of the image was adjusted using a slide bar to
ensure all canopy features were recognized. SLIM converted the thresholded
image to a Below Canopy Readings (BCR) image. This grayscale image divided
the thresholded image into 480 segments. The BCR was then used by the program
to estimate Leaf Area Index and Canopy Gap Fraction.
SLIM estimates LAI in three different ways: Poisson, Binomial and Linear
Average. The Poisson model assumes a random canopy distribution and random
location of leaves and needles in the plane of projection (Jonckheere et al. 2004).
The canopy layers are considered thin enough to decrease the probability of
having more than one contact between incoming rays of light and vegetation
within one layer (Jonckheere et al. 2004). In most natural systems, this is not the
case. Binomial models for estimating LAI are described by Nilson (1971) and
reviewed by Jonckheere et al (2004). The negative binomial model is more
appropriate for canopies with clumped or more regularly distributed leaves. The
28
final SLIM method for estimating LAI, Linear Average, was first proposed by
Lang and Xiang (1986), and combines local linear averaging with larger-scale
logarithmic-linear averaging of transmittance data. The Linear Average method in
SLIM requires an estimation of leaf angle, data which was not available for this
research. The Binomial method requires an estimation of clumping, which was
available from previous work done in WS1 (Kennedy, personal communication).
Due to its availability of clumping and presence of non-random leaf distribution,
the Binomial method was used to calculate LAI values in this research.
For this thesis research, three methods were used to estimate and compare
LAI for the vegetation plots and surrounding area. The SLIM program uses scope
as an input parameter, which determines the angle of view for analysis of canopy
attributes. The default setting for the program is a scope of 60 degrees, resulting
in a 120 degree view for analysis. The 60 degrees refers to the angle from zenith
(directly above) that is analyzed. The three methods for determining LAI were: 1)
LAI for only the plot (LAI_plot), 2) LAI for the plot and immediate area (scope
60 degrees), and 3) a wider view of LAI around the plot.
The first method, LAI_plot required trigonometric calculations and
ArcMap (ESRI, 2012) analysis to determine the correct scope that would only
calculate the LAI for the 9 meter radius plot. First, a raster layer provided by the
HJA for vegetation height was used. This layer was created by researchers at HJA
and is similar to a Digital Elevation Model (DEM). In ArcMAP (ESRI, 2012), this
layer was loaded, as well as the shape file for all 133 vegetation plots. The plots
where the DHPs were taken were selected and a new layer of only these 19 plots
29
was created. From this layer, the create buffer tool was used to create polygons of
the 9 m radius DHP plots. The polygons represented the borders of the plots. The
zone by mask tool was then used to clip the vegetation raster to the vegetation
plots, creating a new layer that only held raster data for the vegetation plots. The
zonal statistics tool was used to determine the mean value of the vegetation raster
for each plot. The clipped plots then represented the average canopy height inside
each plot. Average canopy height was used to calculate the correct scope of view
for analysis in SLIM. Because average canopy height and plot radius transposed
to canopy were known, the angle of scope could be calculated with basic
trigonometry (see Figure 2).
Figure 2. Trigonometric calculation for scope setting in SLIM software for DHP photo analysis.
The second and third methods required less computation. The second
method used the same center DHP as the previous method, but used a default 60
degree setting for LAI analysis. The third method used DHP and computed
average values of all four cardinal directions ~4 meters from the center of each
plot. For these DHPs, a default scope of 60 degrees was also used. The average
30
LAI values determined by SLIM were calculated as the ‘wider view’ of LAI
around each plot. The plot LAI was compared to both the wider LAI and the
average cardinal direction LAI values using a simple linear regression.
LiDAR Data Acquisition
LiDAR data were collected by Watershed Sciences, Inc., (WS) at the HJA and the
Willamette National Forest on August 10th and 11th, 2008. The total collection
area, including a 100 meter buffer, was 19,493 acres. The LiDAR survey used a
Leica ALS50 Phase II laser system with a sensor scan angle of ±14 degrees from
nadir (the perpendicular vector to the ground directly below the aircraft) and a
pulse rate of ≥ eight points per square meter of terrestrial surfaces. All areas were
surveyed with an opposing flight line side-lap of ≥50% (≥100% overlap) to
reduce laser shadowing and increase surface laser painting. The system allows for
up to four range measurements (returns) per pulse, and all discernible laser returns
were processed for the output data set. An onboard differential GPS unit
measured the aircraft position twice per second (2 Hz) to accurately solve for
laser point position (geographic coordinates, x, y, z). Aircraft attitude was
measure 200 times per second (200 Hz) as pitch, roll, and yaw (heading) with an
onboard inertial measurement unit. Aircraft/sensor position and attitude were
indexed by GPS time to allow for post-processing correction and calibration.
Table 1shows the resulting resolution and accuracy with specification and
achieved values (Watershed Sciences Report, 2008).
31
Targeted
Achieved
Resolution:
≥ 8 points/m2
9.14 points/m2
Vertical Accuracy (1 sd)
< 13 cm
2.1 cm
Table 1. Resulting resolution and accuracy of LiDAR data.
LiDAR Data Analysis
LiDAR data form a point cloud that represents areas in which the light pulse from
the plane is reflected back to the receiver after it intercepts canopy or ground. The
visualization software tool FUSION (McGaughey, 2012
http://forsys.cfr.washington.edu/JFSP06/lidar_&_ifsar_tools.htm), was developed
as specialized remote sensing software to process, analyze and display extremely
large LiDAR data sets. FUSION creates 3-dimentional terrain and canopy surface
models and fuses LiDAR data with 2-dimentional imagery such as
orthophotographs, topographic maps, and satellite images. FUSION also includes
algorithms that allow users to manually measure individual tree attributes or
automatic capabilities to characterize individual trees.
The analysis and visualization software consists of two programs,
FUSION and LiDAR data viewer (LDV), and several other task-specific
command line programs (FUSION manual, McGaughey 2012). The primary user
interface for FUSION is a graphical display window and a control window. The
display window shows all project data in a 2D display, similar to geographic
information systems. Input for FUSION consists of several data types, including
shapefiles, images, digital terrain models, canopy surface models, and LiDAR
return data. The LDV program creates and displays 3D visualizations for
examination and measurement of spatially explicit data subsets. The command
32
line programs provide specific analysis and data processing capabilities to make
FUSION suitable for processing large LiDAR acquisitions. Command line
programs utilities (or executables) used for this thesis project include
CanopyModel, ClipData, CloudMetrics, and PolyClipData.
FUSION Executables Used
In addition to the two main programs of FUSION and LDV, the FUSION package
includes numerous task-specific executables, or command line utility and
processing programs. As a part of the FUSION program download, a
comprehensive manual provided guidance and instructions on proper use of the
executables.
The first executable used was ClipData, which creates a sub-sample of the
LiDAR data for use in other analysis. The options for these selections are
rectangular or round, and can be created around any point, such as a plot center or
GPS point. For this project, selections of the LiDAR data were clipped for each
plot. The syntax for the command line program for ClipData follows:
ClipData [switches] InputSpecifier SampleFile [MinX MinY MaxX MaxY]
where ClipData refers to the program to be used, and
[switches] identifies options that can be utilized
InputSpecifier identifies the raw LiDAR data files to be clipped
SampleFile is the output file created
33
[MinX MinY MaxX MaxY] are the spatial coordinates for the sample area
to be clipped.
Switches used were “Shape: 1” to denote a circle selection, and “dtm:file” and
“height” to identify the bare earth model and to normalize the elevation data from
the raw LiDAR data.
The second executable used was CloudMetrics, which computes various
statistical parameters describing a sub-set of LiDAR data. CloudMetrics was
executed on each of the plots using the output of the ClipData executable
mentioned above. The output of CloudMetrics is a .csv file. The syntax for
CloudMetrics follows:
CloudMetrics [switches] InputDataSpecifier OutputFileName
where CloudMetrics refers to the program to be used, and
[switches] to specific options that can be utilized
InputDataSpecifier to the raw LiDAR data files
OutputFileName to the newly created output .csv file
Switches used included “minht:#” and “above:#” which compute additional
metrics above a given height break, in this case 3 meters (Fusion Manual,
McGaughey 2012).
The next executable utilized was CanopyModel to create a canopy surface
model from the LiDAR point cloud for the entirety of WS1, as well as for each
individual plot where DHPs were taken. CanopyModel assigns the elevation of
34
the highest return in each grid cell to the grid cell center, and smooths the surface
using a median or a mean value for the grid cells. When a bare earth model is
used with a switch, a canopy height model (CHM) is created. The output of
CanopyModel is in a PLANS format DTM file that uses floating point elevation
values and contains coordinate projection information for easy input into GIS
systems. The syntax for CanopyModel follows:
CanopyModel [switches] surfacefile cellsize xyunits zunits coordsys zone
horizdatum vertdatum datafile1 datafile2
where CanopyModel refers to the program used, and
[switches] to specific options that can be utilized
surface file to the name of the output file
cellsize to the desired grid cell size (0.5 meters)
xy units and y zunits to units for LiDAR data (M for meters)
coordsys to the coordinate system for the canopy model (1 for UTM)
zone to the coordinate system zone for the canopy model (10)
horizdatum to the horizontal datum used (2 for NAD83)
vertdatum to the vertical datum used (2 for NAVD88) and datafile1
datafile2 to the raw LiDAR data files used to create the canopy model.
The switch “ground:file” was used to specify the corresponding bare earth model
used to normalize the LiDAR data and create a canopy height model. The canopy
height model created a canopy volume metric that was used in the LAI LiDAR
35
model; the canopy volume metric was also input into TreeVaW for identification
and stem mapping of individual trees.
The final executable used was PolyClipData. This program is similar to the
Clipdata executable, but uses a shapefile to clip data rather than a set square or
circle. For this executable, a shapefile of WS1 was used to clip the LiDAR data to
the extents of the watershed. The syntax for PolyClipData follows:
PolyClipData [switches] PolyFile OutputFile Datafile
where PolyClipData refers to the program used, and
[switches] to specific options that can be utilized
PolyFile to the name of the ESRI shapefile containing polygons
OutputFile to the name of the output file
Datafile to the name of the LiDAR data file or a list of data files with a .txt
extension
Volume Calculation for LAI model
A CHM model created with FUSION was used to estimate canopy volume for the
LAI_LiDAR model. The CHM was loaded in FUSION, and the export terrain
model tool was used to export the CHM as an ASCII grid file. In ArcMap, the
ASCII to raster conversion tool was used to create a raster from the CHM. The
new CHM raster output was loaded into ArcMap, along with the bare earth model
mentioned above. The volume surface calculate tool in ArcMap was used to
measure the volume of canopy present above the bare earth model for each plot.
These values were stored in a .csv file for regression analysis using the statistical
package R (The R project).
36
After the surface volume was calculated for each plot in which DHPs were
taken, a simple linear regression was performed between those values and the LAI
plot values from DHP analysis. The regression coefficients, slope and intercept,
were then used for the LAI from LiDAR model.
RESULTS
LAI Results from Digital Hemispherical Photographs using SLIM Software
PLOT
P11109
P11110
P11205
P11206
P11207
P11208
P11209
P11212
P11213
P11608
P11609
P11610
P11611
P11612
P11613
P11614
P11615
P11616
P11617
Plot LAI Plot Gap Fraction Wide LAI Wide Gap Fraction Cardinal LAI Cardinal Gap Fraction
8.24
13
6.42
20.3
6.5
18.525
2.91
58.7
6.68
21.1
6.63
19.675
6.18
30.9
8.41
8.8
7.875
10.525
9.08
9.6
7.06
13.7
7.5625
11.125
7.42
20.2
7.78
11.8
7.555
12.5
9.32
9.3
7.58
9.7
7.7275
10.8
10.39
7.7
7.48
11
7.7425
13.65
5.95
32.7
6.29
22.8
6.41
20.85
6.97
14.5
6.03
17.7
6.92
14.275
8.67
7.5
7.17
10.1
7.4475
9.775
7.15
22.8
7.57
9.7
7.725
8.925
7.66
19
7.31
11.6
7.35
11.025
7.5
13.8
7.19
10.4
7.2425
10.325
9.85
7.3
7.21
11.9
7.595
9.775
8.34
10.8
7.51
9.7
7.6925
9
9.22
7.2
7.85
8.1
7.9
8.7
8.01
20.9
7.96
11.3
7.8825
9.8
9.54
6.1
7.33
10.5
7.555
8.975
8.55
11.5
7.44
11.8
7.5225
9.55
Table 2. LAI and gap fraction results from SLIM software. Plot LAI was calculated with a limited scope of
view to capture LAI within each vegetation plot. Wide LAI used the default 60 degree scope for
calculation and Cardinal LAI is the average of the four cardinal direction LAI values with a 60 degree
scope.
Table 2 displays the results of analysis by SLIM software. Plot LAI is
computed for a scope of view that captured only canopy directly above individual
plots. LAI for the plots ranged from 2.91 to 10.39. Wide LAI was computed using
a scope of 60 degrees (from zenith, 120 degrees of view). Wide LAI showed both
less variation among values and overall lower values in comparison to Plot LAI
(means). The lowest value for the Wide LAI was 6.03, while the highest was 7.96.
37
Cardinal LAI is the average of the four DHPs taken in the four cardinal directions
from center for each plot (also with a 60 degree scope, 120 degrees of view for
analysis). The Cardinal LAI values are relatively consistent among plots, ranging
from a 6.41 to 7.8825.
Before analysis of LAI values, a Q-Q plot of each set of LAI values was
performed to test for normal distribution. Figure 3 displays the Q-Q plots for each
set of LAI values. According to these plots, only the Plot LAI measurements are
normally distributed. Consequently, a natural log transformation was performed
on the data for regression and correlation analysis in order to reduce variability in
the tail ends of the distribution.
Figure 3. Q-Q plots for all LAI results: Plot LAI (right), Wide LAI (bottom left) and Cardinal LAI (top
left). Axes represent theoretical and sample quantiles. Only Plot LAI is normally distributed.
38
Simple Linear Regression of LAI Values
After natural log transformations were performed on all three sets of LAI values,
a regression analysis was performed to assess how the Wide LAI and Cardinal
LAI values compare to the Plot LAI values (Figures 4 and 5).
Figure 4. Regression analysis for ln(Plot LAI) compared to ln(Wide_LAI). The relationship is not
significant (alpha= .05).
39
Figure 5. Regression analysis for ln(Plot LAI) compared to ln(Cardinal LAI)
(R2 = 0.3114, p = 0.013).
The relationship between Plot LAI and Wide LAI is neither strong (R2= 0.08) nor
significant (p= 0.217). The analysis reveals a statistically significant relationship
between Plot LAI and Cardinal LAI (R2 = 0.31, p = 0.013). Outlier plots heavily
influence the relationships between both plot LAI and wider LAI and plot LAI
and cardinal LAI. We suspect that although a strong correlation may exist, it may
be driven primarily by the presence of extremely high and low values of cardinal
LAI. If the correlations do have physical meaning, however, correlations between
cardinal LAI and plot LAI as opposed to plot LAI and wider LAI is indicative of
anisotropy in LAI patterns on the watershed, and suggestive that some form of
40
topographic directionality, potentially driven by mechanisms such as downslope
water movement or solar angle, may exist in LAI.
The Plot LAI values were the only normally distributed values. They also
represent the highest values for the LAI measurements. These values were closest
to what was expected for LAI range for an even aged, mostly Douglas fir stand
(Nadkarni, personal communication). For these reasons, only the LAI plot values
were used for subsequent analysis.
LiDAR LAI Model Based on Surface Volume of CHM
In order to assess the relationship between DHP Plot LAI and LiDAR, the Plot
LAI values were used in a simple linear regression model against the surface
volume metric from the LiDAR data. Table 3 shows the volume of the Canopy
Height Model (CHM) for each plot, as well as the corresponding Plot LAI value
for each plot.
The coefficients (slope and intercept) of the best fit line from the simple
linear regression performed between Plot LAI and the volume of each plot was
used to estimate the value of LAI from LiDAR. The equation, from Richardson et
al. (2009) was derived by work originally completed by Lefsky et al. (1999), but
adapted for use with discrete return aerial LiDAR data.
41
PLOT
P11109
P11110
P11205
P11206
P11207
P11208
P11209
P11212
P11213
P11608
P11609
P11610
P11611
P11612
P11613
P11614
P11615
P11616
P11617
Plot LAI Volume LiDAR LAI
8.24 4313.057 8.323283
2.91 3457.183 6.120263
6.18 3481.565 6.183021
9.08 3984.748 7.478215
7.42 3992.862
7.4991
9.32 3793.143 6.985023
10.39 4104.921 7.787541
5.95 3682.819 6.70105
6.97 4412.353 8.578869
8.67 4208.022 8.052922
7.15 4046.029 7.635951
7.66
4153.8 7.913353
7.5 4704.49 9.330831
9.85 4419.239 8.596594
8.34 4625.235 9.126827
9.22 4697.408 9.312602
8.01 4299.383 8.288085
9.54 4674.676 9.254088
8.55 4109.038 7.798137
Table 3. Plot LAI, Volume of Canopy Height Model, and LiDAR calucalted LAI for each plot studied.
The slope of the best fit line (0.0026) is the value for α and the intercept (2.7785) is the value for β. The following equation from Richardon et al. (2009)
was used:
LiDAR LAI= α+β(Volume)
This equation was applied to each plot and the results are presented in Table 3.
The LiDAR calculated LAI was then compared to the Plot LAI values obtained
from the DHP by running a simple linear regression (Figure 6). The regression
revealed a somewhat strong (R2= 0.3314) and statistically significant (p=0.0099)
relationship.
42
Figure 6. Simple linear regression model of Plot LAI from hemispherical photographs to LiDAR
estimated LAI for each plot.
The LiDAR LAI model was applied to the rest of the permanent
vegetation plots in WS1. Plots P11108, P11202, P11417, P11418, P11427 and
P11519 were excluded due to missing values. The surface volume values, LiDAR
calculated LAI and Cover*Height value for all plots are displayed in Appendix A.
Cover*Height was originally created as a proxy for biomass in prior research in
the watershed. Peterson and Lajtha (2012) exponentially regressed a
Cover*Height metric derived from the 1m LiDAR onto allometrically calculated
values of biomass using the equations in the Pacific Northwest Biomass
Component Equation Library (Halpern and Means, 2011). These equations were
43
validated against similar sets presented by Lutz (2005). Two simple linear
regressions were completed between the LiDAR LAI calculations and
Cover*Height values. First, for plots in which DHP were taken (Figure 7) and
second for all plots in the watershed (Figure 8). The relationship for the DHP
plots was very strong (R2= 0.75) and statistically significant (p< 1.578e-8). This
relationship for all plots was strong (R2= 0.5467) and statistically significant (p<
2.2e-16).
Figure 7. Simple linear regression between LiDAR calculated LAI and Cover*Height values for plots
in which DHP were taken.
44
Figure 8. Results of simple linear regression for all plots between LiDAR calculated LAI and
Cover*Height value.
DISCUSSION
In this research, a model of LAI from LiDAR data was successfully run to
estimate LAI for 133 plots in a small watershed at the H.J. Andrews Experimental
Forest in Central Oregon. The LiDAR based estimates, and the hemispherical
photograph LAI estimates, with which the model was calibrated, are both slightly
less than what would be expected for LAI in the area. The LAI for an even aged,
dense, roughly 40 year old mostly Douglas fir stand would be expected to be in
the range of 9-12 (Vose et al. 1994; Waring, 1998). Both estimates, from DHP
and LiDAR produced estimates ranging from 7-10. The slightly lower estimates
in DHP and LiDAR may be due to sub-optimality of field and sky conditions on
the days of measurement.
45
Previously, a LICOR-2000 device was used to estimate LAI in the
watershed and, among other errors, also resulted in underestimation of expected
values. An informal document produced by previous researchers working in the
watershed detailed issues retrieving LAI for the 133 permanent vegetation plots
and the whole watershed (Peterson, 2010). The document states that protocol for
the use of LI-COR in mountainous terrain (i.e., WS1) is complicated because the
hill slope may shade the hemispherical instrument lens. As mentioned in the site
description, slopes on WS1 average nearly 60% when measured in the field, such
that they intercept the view of the hemispheric lens. Special covers exist to direct
the view to open space away from the hillslope, but experimental error in this case
is high. Because of this, previous attempts at estimating LAI in the Forest
Ecohydrology and tELemetry Transect (FEEL), which corresponds with transect
1 of WS1 have been highly variable (Figure 9). Peterson (2010) also attributes
this error to both variation in tree foliage over time and the experimental error
mentioned above. On the FEEL transect, tree species present include Psuedotsuga
menziesii, Tsuga heterophylla, Arbetus menziesii, Alnus rubra, Acer circinatum,
and Acer macrophyllum. During the early summer season of the most recent
measurement, the abundance of Acer spp. understory contributed severely to the
perceived canopy cover, whereas in measurements made prior and in other
season, hardwood species did not have such prolific coverage. Allometric
equations have also been used in the watershed with similar results showing
variability in foliar contribution depending on the selection of species, stand age,
and stand vigor used. Kennedy (personal communication) suggested that a scalar
46
correction factor of 1.89 be used to express the density of coniferous leaf area that
would not be accounted for in measurements of hardwoods in the watershed.
However, recent calculations of LAI using sapwood values did not use this
correction factor (Peterson, 2010). As a result of these prior results, I chose for
this thesis project digital hemispherical photography as the best non-LiDAR
method of estimation of LAI for the watershed. However, the results of my
research suggest that this method is prone to errors similar to previous methods.
Figure 9. Previous estimates of LAI in the FEEL network in Watershed 1 at H.J. Andrews.
Underestimation of LAI from Hemispherical Photograph Analysis
A review of the literature reveals that underestimation and other issues
surrounding estimation of LAI from hemispherical photographs is common.
Martens (1993) used four different instruments to estimate LAI in conifer and
47
hardwood forest stands. These instruments fell into two categories: line and
hemispheric sensors, with the latter including hemispherical photography.
Different methods in obtaining LAI from hemispherical photographs either
underestimated (Campbell’s method) or overestimated LAI (Beer-Lambert
method) when compared to direct measurements taken in the same area. These
methods are similar to how SLIM estimates LAI in three different ways. Most
estimation methods for LAI require the assumption that canopy elements are
randomly dispersed, which is highly unlikely in any natural forest stand. This
conclusion is similar to that drawn by Dufrene and Breda (1995) in their research
using one semi-direct and three indirect measurements of LAI in a deciduous
forest. All three indirect methods underestimated LAI due in part to local
clumping of architectural canopy components, and in particular, the spatial
dispositions of branchlets and leaves not being independent. This results in a nonrandom distribution of these two canopy elements.
Clark and Murphy (2011) thoroughly outline the issues surrounding
hemispherical photography in the estimation of gap fraction and LAI. The
limitations start with equipment used, including camera spatial, radiometric and
spectral resolution and software. Accurate estimates are also dependent on
weather conditions, time of day, crown closure, ground slope, and many other
factors. Sunny conditions (present at the time of capture of DHPs for this thesis),
time of day (early morning and late evening) and low crown closure can all lead
to increased direct sunlight, which can saturate photographs and negatively
impact the accuracy of gap fraction and effective LAI estimates. The camera and
48
lens used to acquire photographs can also affect LAI estimates. Color blurring in
digital pictures can result in measurement errors in canopy gaps, edge detection,
and lead to “blooming”, especially near zenith (straight up) in sunny conditions.
Clark and Murphy point out that leaf clumping can affect gap fraction and LAI
measurements; with the threshold level of light between “gap” and “plant” being a
subjective measure that varies with individual users and which is subject to bias.
Slope Correction for Hemispherical Photograph Analysis
Hemispherical photographs are taken to capture the canopy directly above the plot
or area that is being studied. When this area is flat, capturing the area directly
above the camera is straightforward. However, when the area is sloped, as is
much of WS1, capturing the area directly above is more difficult. Some
researchers have taken this into account, and slope correction for hemispherical
photograph analysis is an option to correct errors due to slope. Walter and
Torquebiau (2000) addressed this matter, noting that although much of the
world’s forests grow on sloped terrain, the issue of ground slope has not been
addressed in indirect measurements of LAI.
Schleppi et al. (2007) considered how LAI estimates are obtained from
hemispherical photographs. LAI is indirectly estimated by measuring the light
transmission through canopies, and the angle at which the transmission is
measured enters the calculation at two distinct points: 1) as the angle of incidence
determining the travel distance of a light ray through the whole canopy, and 2)
relative to the zenith for the statistical distribution of the angle at which the single
foliage elements are seen. Schleppi et al. (2007) note that any angle on flat ground
49
is sufficient to describe the zenith angle. However, on sloped ground the angle of
incidence is not identical to the zenith angle, and each direction in the canopy
must be classified according to both angles.
Schleppit et al (2007), referencing Nilson (1971) note that the
transmission of light through and ideal canopy is described as a function of the
zenith angle θ:
G(θ)L = K(θ) = -lnT(θ)cos(θ), θ < π/2
where L is Leaf Area Index, G(θ) is the mean projection ratio of leaves in the
zenith angle, which is a function of the statistical distribution of the leaf
inclination angles, T(θ) is the light transmission at angle zenith θ, and K(θ) is the
‘contact number’, which represents the average number of contacts that a light
probe would make by passing through the canopy at the zenith angle, relative to
the thickness of the canopy. This equation is only valid on flat ground. On sloped
ground, with an angle of (v), and the sensor held horizontally, light travels a
shorter way through the canopy in the downhill compared to the uphill direction.
If an observer looked downhill the canopy would appear lighter, and darker when
looking uphill. Therefore, the zenith angle (θ) can be replaced in the equation on
the right side by the angle of incidence (τ):
G(θ)L = K(θ) = -lnT(θ, τ)cos(τ), θ < π/2, τ < π/2
50
This correction would necessarily have to occur within the computer software
algorithms used to estimate LAI from hemispherical photographs.
LiDAR LAI Estimates
Since the LiDAR estimates of LAI were calibrated based on the DHP estimates of
LAI, they are slightly less than expected as well. However, the high r-squared
value (R2= 0.57) when the estimates for all plots were compared to the
Cover*Height measurements for each plot is promising in that they compare to a
widely used metric within the watershed. One way to reassess LAI for the plots
would be to use a so-called LiDAR only metric, or a model that only uses metrics
derived from the LiDAR data itself, and does not rely on possibly unreliable
hemispherical photograph estimates of LAI.
Many of these LiDAR only models exist in the literature. Jensen et al.
(2008) noted that most previous attempts to estimate LAI from remote sensing
have relied on empirical relationships between field-measured observations and
various spectral vegetation indices (SVIs) derived from optical imagery or the
inversion of canopy radiative transfer models. Jensen et al. (2008) used LiDAR
data along with SPOT5- derived SVIs to estimate LAI, but found that LiDAR data
alone was adequate to do so. The researchers calculated many LiDAR-derived
model covariates, including canopy height metrics, canopy cover metrics, and
height distribution metrics. These covariates were compared to known LAI values
for plots within a multiple regression framework. In estimated LAI for two stands
in Idaho, nine different models were created and tested, with r-squared values
ranging from 0.6971 to as high as 0.8612 (Jensen et al. 2008).
51
Overall, the LiDAR derived covariates explained the largest proportion of
variation in LAI, with most models incorporating LiDAR covariates associated
with upper story metrics, and all models containing the covariate MAX_HEIGHT.
The authors point out the logic for including LiDAR: increases in canopy height
should correlate to increases in LAI. However, the covariate MAX_HEIGHT
alone did not significantly correlate with the known LAI quantities. When
considering the vertical foliage distribution covariates, the calculated differences
in percentile heights played an important role. Similarly to the MAX_HEIGHT
covariate, the covariate L95_C25, or the 95th percentile value minus the 25th
percentile value, is present in each LAI model.
Morrison et al. (2011) also set out to compute LAI using LiDAR with
minimum field data for use in remote areas like the Canadian boreal forest. Their
research reviewed many LiDAR metrics from articles mentioned above (Lim et
al. 2003, Riano et al. 2004, Solberg et al. 2006, and Lefsky et al. 1999), but
concluded that each model was developed for a specific forest type and required
calibration. Instead, the authors modified an intensity based gap fraction model
developed by Hopkinson and Chasmer (2007). This model classified LiDAR
returns into four echo classes (first, single, intermediate, last) and generated grids
of intensity by summing returns within a raster cell, and takes into account twoway power transmission loss by intermediate and last return hits using a square
root function. The subset of first and single hits at or below 1.3 meters represents
ground returns and an equation combining all these values is used to estimate gap
52
fraction. The gap fraction (P) estimates are then used to estimate effective LAI
(LAIe) based on the Beer-Lambert Law:
LAIe = -ln(P)/k
where k is the extinction coefficient, which is a function of leaf angle distribution,
radiation type and direction, and canopy structure and clumping. A standardized,
mid-value of 0.5 was used by the researchers to represent a spherical (random)
projection coefficient for leaves of any shape. Using the standard 0.5 extinction
coefficient, LAIe for conifer species was underestimated, and LAIe was
overestimated for broad-leaved aspen species. The authors then optimized the
extinction coefficient for each species studied using DHP LAI estimates. This
resulted in better LAIe estimates from the LiDAR data.
Alternative LiDAR Systems
The LiDAR system used for collection of data for this thesis was a discrete return,
small footprint system, with similar systems used extensively in the research field
to estimate LAI and other forest attributes. However, two other systems hold
promise for estimation of LAI: full waveform aerial LiDAR and ground-based
terrestrial LiDAR systems. Full waveform systems have already been described in
this thesis, and were used by Lefsky et al. (1999) for their estimation of LAI.
Adams et al (2012) utilized full waveform LiDAR in their research, noticing that
with all uses of LiDAR, performance was hindered by an inability to distinguish
the source of the LiDAR returns as foliage, stems, understory and the ground,
53
other than the relative position of the return. The goal of their research was to
determine whether drawing distinctions between the type of material that a return
was hitting would improve analysis with LiDAR data, and if full waveform
metrics could provide information on foliage density and improve forest health
and growth measurements.
Adams et al. (2012) also cover the major differences between waveform
systems and discrete return systems and why using full waveform systems might
better suit the needs of researchers. The major drawback of discrete return
systems is a ‘blind-spot’ that occurs following each detected return, during which
no other returns can be detected. Adams et al. (2012) quantified waveform shape
with various curve-fitting methods, including peak height, half-height width and
an exponential decay function attributed to each return. However, due to the
complexity of the surfaces encountered and the multitude of angles, textures and
paths each new waveform metric showed more potential variation within a
surface type that it did between surface types. However, ground peaks on average
showed waveforms with higher peaks, shorter widths and faster decays. Foliage
returns in turn averaged lower peaks, wider pulses and slower decays. This
classification could lead to more accurate DEM production from full waveform
LiDAR, but the clear distinction between surface types still seems impractical for
the time being.
Terrestrial LiDAR systems represent an alternative approach for collecting
LiDAR data, where collection occurs from below the canopy rather than from an
aircraft or satellite. Seidl et al. (2012) examined the use of ground-based laser
54
scanning in the analysis of mature forest structure and compared their findings to
hemispherical photography methods. In all, 35 groups of trees were analyzed by
the researchers to generate 3D point clouds of the tree axes and leaves. The
images were used to generate hemispheric views of the canopy. These images
were compared to actual hemispherical photographs taken in the same area. The
authors found that their method was problematic for identifying small canopy
gaps, and wind-induced movements at the time of collection further complicated
the issue. However, improvements in the systems should speed up the operation
of the system and/or produce a smaller beam, both of which would help alleviate
the issue. The authors see a future application of this work being the creation of
canopy models of growth and photosynthetic carbon gain in mature trees based on
the 3D canopy structure data collected, which their study showed was well
represented by terrestrial laser scanning.
CONCLUSIONS
Estimation of Leaf Area Index in Watershed 1 at H.J. Andrews Experimental
Forest using prior methods had proven difficult. The use of hemispherical
photographs to estimate LAI in this thesis seems prone to the same difficulties.
The steepness of the terrain and the high density of the vegetation make the
estimation of LAI very uncertain. However, using LiDAR data to estimate LAI
seems promising for the watershed. The strong correlation between the LAI
LiDAR estimates from this work to the Cover*Height metric used in the
watershed provide evidence that the results are validated against a metric known
and often used by researchers in the area.
55
The lack of correlation of the plot level LAI to the wider scope LAI and
cardinal direction LAI is curious. This may be caused by either the hill slope
affecting the two wider measures of LAI, or the different ways in which light is
being measured in the photographs within the software itself. Exploring how light
is reacting differently on the sloped terrain compared to a flat surface may provide
insight and better results for the wider LAI measurements. Developing new
algorithms within software programs for specialized use on sloped terrain could
perhaps solve to this issue.
A possible next step in the estimation of LAI for the watershed is to
develop and test a LiDAR model that does not rely on ground-truthed LAI
estimates. In theory, a model of LAI from LiDAR data that used hemispherical
photographs, or other methods, as a ground-truth will only be as good as those
original LAI estimates. A LiDAR only model may be the answer to estimating
LAI in a steep, densely vegetated watershed, since it has been proven that LiDAR
can adequately measure these areas.
Finally, exploration into alternative LiDAR systems may be helpful in the
estimation of LAI in the watershed. An examination of how full waveform or
terrestrial systems could improve estimates is certainly possible. The discrete
return, aerial system used for this thesis seems capable of providing information
on the upper canopy, but seems to lack the capability to accurately measure and
assess lower canopy and understory features. An interesting approach to solve this
issue could be linking the aerial LiDAR data with terrestrial LiDAR data. This
56
would provide a perspective from above and below the canopy, and may provide
a more complete picture and insight into the complete structure of the forest.
57
Chapter 3- LiDAR used to Describe Stand Characteristics and Identify
Individual Tree Height and Location
58
INTRODUCTION AND BACKGROUND
LiDAR data have been used extensively in recent years and have the potential to
generate high resolution digital terrain surfaces accurately. The resulting surface,
precise within 15 cm, represents complex natural and semi-natural environments
at a range of scales (Large and Heritage 2009). One of the first commercial uses
of LiDAR data in the United States was the identification of encroaching
vegetation on power line corridors (FUSION manual). Federal agencies in the
United States, such as the Federal Emergency Management Administration
(FEMA) and the U.S. Geological Survey (USGS), have used LiDAR with county
and state agencies to map flood plains and earthquake hazard zones (FUSION
manual). The following represents a fraction of LiDAR applications in the
forestry sector: a) the estimation of biomass and carbon stocks, b) description and
quantification of forest structure and cover, and c) identification of individual
trees and stem mapping.
Biomass and Carbon Stocks
Omasa et al. (2007) assessed the capability of LiDAR to measure carbon (C)
stocks in forests. Using LiDAR to quantify forest C stocks leads to a more
complete understanding of terrestrial C cycling, which is important to quantify in
light of recent climate change research. The ability of a forest to store and
sequester C is often valued as ecosystem services, so an apt understanding of their
capacity to do so is essential for both ecological and economical decision making.
LiDAR is a novel tool for quantifying forest biomass because it allows for remote
sensing of highly specific biomass components. For example, Garcia et al. (2010)
59
also evaluated LiDAR use to estimate total aboveground, branch and foliar
biomass in an unmanaged forest in Spain using models based on LiDAR extracted
height, LiDAR point cloud intensity, or height and intensity data combined. The
researchers determined that normalizing LiDAR intensity data to a standard range
removed the range dependence of the intensity signal. The intensity-based models
proved the most effective and provided more accurate predictions of the
breakdown of biomass into branch and foliar fractions. They also found that using
species-specific models of the dominant species in the area improved estimates
for biomass. Overall, the research demonstrated that LiDAR intensity data could
be used to segment above ground to branch and foliar biomass from total biomass
determination. Other variables, derived from LiDAR data and similar to those
created using CloudMetrics and FUSION software (see Chapter 2), were included
as explanatory variables in these biomass models.
Zhao et al (2009) conducted further work with LiDAR data and forest
biomass, but aimed to move beyond scale-dependent the models that first need to
be fitted and applied at the same scale or pixel size. The research goal was to
create methods for scale invariant estimation of forest biomass using LiDAR data,
and resulted in two models: a linear functional model that used LiDAR-derived
canopy height distributions and a functionally equivalent nonlinear model that
used canopy height quantile functions as parameters. These models used a LiDAR
tree delineation approach to create a fine-resolution biomass map that captured
individual tree component biomass in Eastern Texas, and the authors’ work
validated the use of canopy height distributions and canopy height quantiles as
60
LiDAR metrics for estimating biomass, as well as for mapping biomass at a range
of spatial scales. Furthermore, the results of this work are viable for estimating
other forest characteristics including belowground biomass, timber volume, crown
fuel weight and Leaf Area Index.
Using LiDAR Data to Describe Forest Structure and Cover
Aerial LiDAR data have been used extensively for describing, measuring and
quantifying canopy cover and structure of forests all over the world. Magnussen
and Boudewyn (1998) used canopy-based quantile estimators to derive stand
height from LiDAR data. Knowing that the proportion of laser pulses returned
from or above a given reference height is directly proportional to the fraction of
leaf area above it, the authors hypothesized that an unbiased estimate of this
relationship could be obtained using the quantile of LiDAR derived canopy
heights matching the fraction of leaf area above a desired height. Their work
found a strong relationship between field and LiDAR estimates of stand height,
and statistical tests supported their hypothesis. Their work demonstrated that
estimating stand height from LiDAR data based on maximum canopy height
value in each cell of a fixed grid has been and is likely to continue to be
successful.
Smith et al. (2009) used discrete return LiDAR data to compare estimates
of forest canopy cover from LiDAR and spectral methods and ground based
measurements. This research used imagery from the Advanced Spaceborne
Thermal Emission and Reflection Radiometer (ASTER), and explored sources of
error if this technology were used on a large scale. The researchers found that
61
78% of the variability in field-based canopy cover metrics could be described by
the derived LiDAR metric for canopy cover. They surmised that the other 22% is
likely due to challenges of using LiDAR to sense understory vegetation and shrub
dominated plots.
Hopkinson and Chasmer (2009) compared four models of fractional cover
to hemispherical photograph fractional cover measurements across five distinct
ecozones, eight forest species and multiple LiDAR survey configurations. Their
models used four different LiDAR metrics: 1) a canopy-to-total first returns ratio,
2) a canopy-to-total returns ratio, 3) an intensity return ratio, and 4) a Beer’s Law
modified intensity return ratio. Although they found that the intensity based forest
cover model had the highest R2 value, the forest cover method using Beer’s law
was more useful since its best fit line passes though the origin and has a slope
near unity. The models used showed promise across all the ecozones, but short
canopies (less than 2m) and open canopy forest plots posed the greatest challenge
to the models when predicting forest cover.
In research by Coops et al. (2007), the authors noted that the variation in
vertical and horizontal forest structure is difficult to quantify with labor-intensive
field methods or with passive optical remote sensing techniques that are limited in
their ability to distinguish structural changes below the top of the canopy.
Working in primarily Douglas-fir (Pseudotsuga menziesii) and western hemlock
(Tsuga heterophylla) stands on Vancouver Island, Canada, the researchers chose
stands that represented a wide range of stand development ages. This research
built on previous work that used full waveform data to examine open and filled
62
volumes within the canopy itself (Lefsky et al. 1999). The resulting model
produced a three dimensional canopy structure that was termed “canopy volume
profiles” or 1 meter tall cells or voxels that were classified either as empty or
filled depending on whether a LiDAR return was present within the 1m voxel.
The voxels in the upper most 65% of the canopy were considered “euphotic”, and
those below that threshold deemed “oligophotic”. The euphotic zone refers to the
area that intercepts the majority of light within the canopy. Oligophotic refers to
the area beneath the euphotic zone, which receives less light compared to the
euphotic zone.
Coops et al. (2007) adapted the full waveform SLICER LiDAR methods
used by Lefsky et al. (1999) for their discrete return LiDAR data. The returns
present into 5x5 meter plot subsets were ‘binned’, and the number of returns
within each subplot was counted at 1m height intervals. Bin areas were then
classified as oligophotic or euphotic depending on their placement within the
canopy, thus creating canopy volume profiles for the plots, that were found to
correlate with ground measured stand attributes including crown volume, stem
density and basal area. The overall canopy surface structure was shown to
suitably characterize the total amount of open gap area using the methods of
binning LiDAR into 1m voxels. The limited number of observations precluded
developing regression models to estimate forest parameters, but the authors
concluded that the relationship between the canopy volume variables and
structural attributes suggest that models could be developed for this forest type
regardless of stand age.
63
Concurrent with the work of Lefsky et al. (1999) was a research paper by
Means et al. (1999). Using the same SLICER system, the authors compared
ground measurements for height, basal area, total biomass, and leaf biomass to
those obtained from the full waveform LiDAR system. The SLICER derived
measurements correlated well with ground-measured attributes, with R2 values of
0.95 for height, 0.96 for basal area, 0.96 for total biomass and 0.84 for leaf
biomass. The relationships found were strong up to a height of 52m, a basal area
of 132 m2/ha, and a total biomass of 1200 Mg/ha.
Much research has compared LiDAR estimates of forest attributes to field
measured data, but Smith et al (2009) also considered spectral estimates when
comparing field and LiDAR estimates of forest canopy cover. This work, situated
in northern Idaho over an area of 25,000 hectares of mixed conifer forest,
compared cover measurements from reflective spectral satellite data, and LiDAR
and field collected measurements with variables measured using spherical
densiometers. . The research had two goals: 1) to evaluate the overall accuracy of
spectral and LiDAR derived cover metrics, and 2) to determine whether LiDAR
data could quantify and reveal the sources of error observed in the spectral-based
canopy cover metrics. The LiDAR metrics outperformed the spectral metrics
when compared to field gathered data. However, all metrics were sensitive to the
presence of herbaceous vegetation, shrubs, seedlings, saplings, and other
subcanopy vegetation. This work is particularly relevant to this analysis because
the stand type and site analyzed is similar to the one addressed at H.J. Andrews.
64
Lovell et al (2003) highlighted the importance of obtaining canopy
structure from LiDAR data since that information is not available from other
remote sensing methods but is essential for ecological assessments in forest
inventory. Canopy architecture is particularly relevant to predictions of moisture
and gas exchange that describe the overall functionality and productivity of the
forest ecosystem. Their work used both aerial laser scanning, as well as ground
based ranging systems in measuring important forestry parameters compared to
standard field inventory, hemispherical photographs, and optical point-quadrat
sampling. Simple models were developed, including determining predominant
height of stand by aerial LiDAR, and Leaf Area Index from the ground based
scanning system. The results of this work justify the further development of
instrumentation and analysis to combine results from multiple systems to describe
forest attributes such as height, cover, LAI, and foliage profile.
Many models using LiDAR data to describe forest structure are based on
where the LiDAR points are distributed, namely by height. However, Hopkinson
and Chasmer (2007) used LiDAR intensity values in modeling canopy gap
fraction by using a modified Beer-Lambert approach. This may be more
physiologically appropriate since a Beer-Lambert calculation reflects the capacity
of the canopy to absorb radiation and deflect moisture, whereas a height
distribution is a descriptive metric with less direct physiological significance. The
authors related the ratio of ground return power/total return power to the canopy
gap fraction derived from digital hemispherical photographs. They found that the
LiDAR intensity based power distribution ratio provided a higher correlation to
65
the DHP gap fraction than the more often used ground to total return ratio.
Furthermore, they modified the intensity power distribution ratio to account for
secondary two-way pulse transmission losses within the canopy. This step created
a model that requires no calibration and provides an accurate estimate of overhead
canopy cover.
Hyde et al (2005) pointed out that LiDAR has been used extensively for
measuring forest attributes in many ecosystems, including tropical, boreal, and
mid-latitude forests. However, the authors noted that few studies have taken place
in montane forests, and examined the ability of large footprint LiDAR systems to
retrieve forest structural attributes in highly variable terrain and different canopy
conditions in the Sierra Nevada mountains in California. The authors examined
the effects of slope, elevation, aspect, canopy cover, crown shape, and spatial
arrangement of canopy forming trees on the accuracy of the LiDAR estimates of
height, cover and biomass and found good agreement between field and LiDAR
measurements of height, cover, and biomass at the footprint level, and canopy
height and biomass at the stand level. The differences encountered between field
and LiDAR measurements was attributed to the spatial configuration of canopy
elements, and were less affected by topography, crown shape, or canopy cover.
LiDAR data have also been utilized to measure, quantify and map the
structure of understory characteristics. Korpela (2008) used LiDAR data to
identify and map understory lichens in a barren pine grove in Southern Finland,
using two different LiDAR data sets and examining the backscatter properties and
intensity of LiDAR returns to differentiate bare ground from ground covered by
66
reindeer lichen. The remote sensing capabilities of the LiDAR made large scale
mapping of lichen possible, whereas conventional field methods would have
proved time-consuming and thus impractical.
Stem Mapping
LiDAR data have been used extensively to describe forest structure and cover,
and to estimate biomass and C stocks, and more recently to identify individual
trees and take ‘inventory’ of forest stands at various spatial scales. These
techniques replace older, time consuming field measurements that identify
individual trees one at a time during forest surveys (Hawk, 1970). Identifying
individual trees in a stand with LiDAR data is important for this thesis as it
suggests an alternative to mapping LAI at the watershed scale. If all trees can be
identified in a stand, then allometric equations of DBH to height (Curtis 1967;
Waring et al., 1977; Garman et al. 1995) and stem/crown dimension and
biomass/needle area ratios (Bartelink, 1996) might be used to correctly estimate
the amount of leaf area for individual trees. The estimates could then be used to
scale up tree-based estimates to LAI at the stand, watershed or even landscape
scale.
Strunk et al (2008) used aerial LiDAR data to estimate basal area for a
complex forest on the Fort Lewis Military Installation in southeast Puget Sound
(now Joint Base Lewis-McChord). The authors used the program FUSION
(McGaughey, 2012) and similar software cited in Chapter 2 of this thesis
(ClipData, CloudMetrics, GridMetrics) to produce LiDAR metrics that describe
canopy size and vertical distribution. These metrics calculated basal area using
67
regression analysis. While the research did not explicitly identify individual trees
from the stands studied, the estimation of basal area of a stand as well as stand
density, predominant height, crown cover, foliage projected cover, and foliage
branch projected cover are structural variables that are used to characterize forests
in inventory, mapping programs and conservation (Specht and Specht, 1999).
Popescu and Wynne (2004) first set out to develop processing and analysis
techniques to facilitate the use of discrete return LiDAR data to estimate plotlevel tree height and measure individual trees from three-dimensional LiDAR
surfaces. Popescu has been a leader in this field, developing TreeVaW, the tree
extraction software used in this research. His work was conducted in both
deciduous and coniferous stands, and the methods of using LiDAR derived tree
metrics with regression models and cross validation of tree heights were more
successful in the conifer stands than in the deciduous. The process included
filtering the point cloud data with a circular window for the conifer stands, and a
square window for the deciduous stands, and calibrating the search window based
on forest type led to better results overall.
Similar to methods used by Popescu and Wynne (2004), Tiede et al.
(2005) used local maximum filtering within a GIS environment to extract and
delineate single trees from the LiDAR data point cloud. After using local maxima
to identify individual trees from the point cloud, the authors developed a regionspecific growing algorithm to delineate tree crowns from their data. This method
used the original laser points rather than a derived raster data set. The authors
achieved their stated goal for this research, developing and demonstrating a
68
complete GIS-based method from LiDAR data pre-processing, algorithm
development, analysis and visualization, but noted that a complete count and
validation of findings required field-verification, especially in complex multitiered deciduous and young stands.
Chen et al (2006) also derived local maxima in a canopy height model to
identify individual tree tops using variable sized windows. However, in this
research the window size was determined by the lower-limit of the prediction
intervals of the regression curve between crown size and tree height. This was
done because of ‘commission errors’, or non-treetop local maxima that were
incorrectly classified as treetops. The authors manually measured tree crown size
and tree height from the CHM used for analysis. This method was preferable to
measuring in the field since: 1) it was easy to identify individual trees from the
LiDAR data set due to the high pulse density present, and 2) sampling within the
CHM greatly reduced workload and was not limited by factors such as
accessibility in the field. The authors then used a watershed segmentation method
by inverting the CHM so local maximums became local minimums, and a canopy
minimum model (CMM) was produced. The ‘flooding’ of the CMM took place
and the algorithm developed built ‘dams’ along the divide line between
neighboring catchment basins (trees). The ‘dams’ were called watershed lines and
were used to partition individual trees from the flooded CMM.
Korpela et al (2007) used large-scale aerial imagery, a Digital Terrain
Model (DTM), and LiDAR data to develop a single-tree remote sensing (STRS)
system to identify 3D treetop positioning, height estimation, species recognition,
69
crown width estimation, and model-based estimation of individual stem diameter.
Each step was a part of a semiautomatic system using the tree data sources listed
above. LiDAR-based crown width estimates were completed using crown
modeling, where parametric crowns were iteratively fitted with LiDAR data.
Image-based 3D treetop position and crown width estimation was down with
multi-scale template matching, and species recognition was done manually by
visual photo-interpretation. Overall, this method underestimated stem diameters,
and calibration of the system again would require time-consuming field
observations and measurements.
METHODS
TreeVaW Individual Tree Identification and Stem Mapping
For this thesis, a canopy height model (CHM) was created using FUSION and
LiDAR data for WS1 (see Chapter 2). This CHM was clipped to the extents of
each plot in the watershed using the extract by mask tool in ArcMap (ESRI).
Because TreeVaW reads an older ENVI file format, with a .dat extension, the
ArcMap copy raster function was utilized so that these files could be imported
into TreeVaW, . The copy raster tool also creates an .hdr, or header file, that
TreeVaW requires. Files were converted for each plot, after which the .dat
extension was removed, which is the final step before TreeVaW would read the
data. TreeVaW also requires unique parameters to calibrate its algorithms to the
forest type present. Keith Olsen, a researcher at Oregon State University
completing similar work at H.J. Andrews using TreeVaW provided these values
(Keith Olsen, personal communication). Each plot file was input into TreeVaW
70
one at a time, and the software output a .csv file with XY location, crown
diameter, and tree height for each tree identified. The following equation relating
crown diameter (y) to height (x) was used for TreeVaW input:
y = 0.0000310796916048496 * x3 - 0.00267405906767456 * x2 +
0.195530509685481 * x + 1.61296048520958
H.J. Andrews Vegetation Survey Data
To compare the output of individual trees produced by TreeVaW, I used a data set
from the H.J. Andrews website detailing comprehensive vegetation surveys
completed in WS1 (http://andrewsforest.oregonstate.edu/). The most recent data
were used (2007) since it would closest temporally to the date when the LiDAR
data were collected (2008). The HJA data set required pre-processing to allow for
comparison to the TreeVaW output: 1) Because the data set contained plots for
both WS1 and WS3, all WS3 data were removed, 2) Next, all trees that were
noted as “mortalities” were removed, so to analyze only live trees, 3) Finally,
trees that were too small to measure diameter at breast height (DBH) were
removed, since those trees would also be too small for the TreeVaW program to
recognize.
DBH-Tree Height Relationship and Comparison
After the HJA data were prepared it was then compared to the TreeVaW output
for each plot. However, there was no direct way to ‘match up’ the trees present in
each data set because TreeVaW identified individual trees by XY location, crown
71
diameter and tree height, while the HJA data set identified trees only by DBH and
quadrant location (NW, NE, SW, SE). In order to identify which trees TreeVaW
was finding, asymptotic equations to convert DBH to height were used for each
individual species (Research Contribution 10, Garman et al. 1995). The equations
from that report provide predictive regional estimates of height-diameter trends
for 24 tree species over a wide range of diameters, and used the same equation for
each species, with regression coefficients for each derived from trees measured in
8,727 fixed and variable radius plots representing managed and natural stands.
The equation used along with coefficients for each species converted in this thesis
is found in Table 4.
Height = 1.37 + (b0[1-exp(b1*DBH)]^b2
Species
CODON
Douglas Fir (Pseudotsuga menziesii ) PSME
Big Leaf Maple (Acer macrophyllum ) ACMA
Western Hemlock (Tsuga heterophylla ) TSHE
Red Alder (Alnus rubra )
ALRU
Chinkapin (Castanopsis chrysophylla ) CACH
B0
61.6358
30.4131
57.4756
35.55
40.6648
B1
-0.01469
-0.03425
-0.01677
-0.02832
-0.01778
B2
0.92706
0.6821
1.02854
0.79602
0.87363
Table 4. DBH-Height asymptotic equation and regression coefficients used for each species (Garman et
al. 1995).
RESULTS
Total Trees Identified in Plots
Overall, in all the plots of WS1 TreeVaW identified 2,810 trees of the 3,407 trees
observed (82.48%) in the vegetation surveys (see Appendix B for complete list of
trees by plot). Trees identified and trees observed for all the plots in which DHPs
were taken are presented in Table 5. When TreeVaW was used to identify all trees
72
in the watershed, 74,299 trees were identified. However, since all trees in the
watershed have not been counted, there is no way to assess the accuracy of this
measure.
Plot
P11108
P11109
P11110
P11205
P11206
P11207
P11208
P11209
P11211
P11212
P11213
P11608
P11609
P11610
P11611
P11612
P11613
P11614
P11615
P11616
P11617
Observed Trees Predicted Trees % of Trees Id'ed
28
12
42.86%
23
17
73.91%
23
18
78.26%
17
17
100.00%
17
15
88.24%
15
18
120.00%
12
14
116.67%
25
11
44.00%
38
15
39.47%
46
22
47.83%
28
17
60.71%
24
17
70.83%
15
12
80.00%
18
14
77.78%
19
14
73.68%
21
17
80.95%
30
15
50.00%
19
15
78.95%
25
20
80.00%
21
14
66.67%
13
14
107.69%
Table 5. Observed trees, trees predicted by TreeVaW and percentage of trees identified by TreeVaW
in plots where digital hemispherical photographs were taken.
Vegetation Survey and TreeVaW Comparison by Plot
In plot P11108, TreeVaW identified 12 of 28 live trees (42.86%). The trees
identified were the tallest 12 trees present in the plot, according to the vegetation
survey. Also, the height of the trees calculated by TreeVaw was consistently
higher than the actual trees in the plot.
73
Figure 10. Transect 1, plot 8.
In plot P11109 TreeVaW identified 17 of 23 live trees present (73.91%).
Similarly to P11108, the 17 trees identified by TreeVaW were the tallest 17 trees
present in the plot. Also, the heights of the trees identified by TreeVaW were
(again) consistently taller than those in the actual vegetation survey.
Figure 11. Transect 1, plot 9.
In plot P11110, TreeVaW identified 18 of 23 live trees present (78.26%). Unlike
P11108 and P11109, however, TreeVaW identified both tall and short trees, but
not trees in the middle of the height range. The heights of the taller trees identified
74
by TreeVaw were overestimated and the heights for the shorter trees
underestimated.
Figure 12. Transect 1, plot 10.
In plot P11205, TreeVaW identified 17 of 17 live trees (100%). The heights of all
the trees identified by TreeVaW were very similar to the actual tree heights from
the vegetation survey, although heights of the three smallest trees were
underestimated. This may be due to the fact that these trees were ACMA (Acer
macrophyllum, or bigleaf maple), while the rest of the trees \ were PSME
(Psuedotsuga menziesii, or Douglas Fir). Additionally, the DBH data for the
ACMA (and Castanopsis chrysophylla or CACH) trees on WS1 is difficult to
incorporate into single bole models such as TreeVaw because it represents the
average DBH of a clump of stems, so that the stem representing the main
structural bole (likely the tallest bole) will be de-emphasized by the presence of
thinner copice stems of secondary boles.
75
Figure 13. Transect 2, plot 5.
In plot P11206, TreeVaW identified 15 of 17 live trees present (88.24%). The
heights of the trees identified by TreeVaW were slightly overestimated for the 11
taller trees and slightly underestimated for the 4 shortest trees.
Figure 14. Transect 2, plot 6.
In plot P11207, TreeVaW identified 18 trees although only 15 live trees were
present (120%). Heights for the taller trees identified by TreeVaW were
overestimated while the heights of the shorter trees were quite close to the
measured trees heights. The identification of three extra live trees not present may
be due to identification of snags (dead trees) by TreeVaW that were removed
from the vegetation survey.
76
P11207
35
Tree Height (m)
30
25
20
15
Measured
10
TreeVaW
5
0
0
5
10
15
20
Number of Trees
Figure 15. Transect 2, plot 7.
In plot P11208, TreeVaW identified 14 trees while only 12 live trees were present
(116.67%). The two extra trees identified by TreeVaW were smaller than any live
tree present in the vegetation survey. These extra trees may have been small snags
that were removed from the vegetation survey. The heights of all trees identified
by TreeVaw in this plot were overestimated, consistently about 3m taller than the
trees present according to the vegetation survey.
P11208
30
Tree Height (m)
25
20
15
Measured
10
TreeVaW
5
0
0
5
10
15
Number of Trees
Figure 16. Transect 2, plot 8.
In plot P11209, TreeVaW identified 11 of 25 live trees present (44%). The trees
identified by TreeVaW for this plot were the tallest trees present according to the
vegetation survey. Also, the heights of the trees identified by TreeVaW were
consistently overestimated by about 1-2 meters.
77
Tree Height (m)
P11209
40
35
30
25
20
15
10
5
0
Measured
TreeVaW
0
5
10
15
20
25
30
Number of Trees
Figure 17. Transect 2, plot 9.
In plot P11211, TreeVaW identified 15 of 38 live trees present (39.47%). The
trees identified by TreeVaW were the tallest trees present according to the
vegetation survey, and the heights of the trees identified were consistently 7-8
meters taller than the trees present. Plot 211 is a particularly unique plot on WS1
because a mortality event occurring on the plot above it, Plot 212, has led to the
bending and breakage of many trees on Plot 211. We would expect to see this
over-estimation of height on the model versus the field because the model does
not take into account the poor tree morphology on this plot due to local
disturbance.
Figure 18. Transect 2, plot 11.
In plot P11212, TreeVaW identified 22 of 46 live trees present (47.83%).
TreeVaW identified both the tallest and the shortest trees present in this plot. For
78
the shorter trees, TreeVaW consistently underestimated the height of the trees and
consistently overestimated the height of the tallest trees. Again, the local mortality
event on this plot may play a role. While large and small trees may have survived
the event due to structural resilience or flexibility, respectively, mid-sized trees
may have been damaged to the extent that they are unrecognizable in LiDAR
imagery.
Figure 19. Transect 2, plot 12.
In plot P11213, TreeVaW identified 17 of 28 live trees present (60.71%). All of
the trees identified by TreeVaW except one were the tallest trees in the plot.
TreeVaw overestimated the height of all trees identified.
Figure 20. Transect 2, plot 13.
79
In plot P11608, TreeVaW identified 17 of 24 live trees present (70.83%).
TreeVaW identified the 17 tallest trees in this plot with exceptional precision,
only overestimating one tree by roughly three meters.
Figure 21. Transect 6, plot 8.
In plot P11609, TreeVaW identified 12 of 15 live trees present (80%). TreeVaW
identified the 11 tallest trees as well as the shortest tree in this plot. It
overestimated the height for three trees by 2-8 meters and underestimated the
height of the tallest tree present in the plot.
Figure 22. Transect 6, plot 9.
In plot P11610, TreeVaW identified 14 of 18 live trees present (77.78%).
TreeVaW identified the 12 tallest trees as well as the two shortest trees in this
80
plot. The five tallest trees and two shortest trees were identified accurately for
height, but TreeVaW slightly overestimated the height for the remaining 7 trees.
P11610
35
Tree Height (m)
30
25
20
15
Measured
10
TreeVaW
5
0
0
5
10
15
20
Number of Trees
Figure 23. Transect 6, plot 10.
In plot P11611, TreeVaW identified 14 of 19 live trees present (73.68%).
TreeVaW again estimated the 14 tallest trees in this plot, fairly accurately with all
estimated heights falling within two meters of the actual measurements.
Figure 24. Transect 6, plot 11.
In plot P11612, TreeVaW identified 17 of 21 live trees present (80.95%). The
four trees that TreeVaW failed to identify were in the lower third of tree heights.
The heights for the three smallest trees identified were accurate, and TreeVaW
slightly overestimated the majority of the other trees identified.
81
Figure 25. Transect 6, plot 12.
In plot P11613, TreeVaW identified 15 of 30 live trees present (50%). The
shortest tree and the 14 tallest trees were identified by TreeVaW within this plot.
TreeVaW slightly underestimated the height of the shortest tree and slightly
overestimated (1-2 meters) the remaining tallest trees identified.
P11613
35
Tree Height (m)
30
25
20
15
Measured
10
TreeVaW
5
0
0
10
20
30
40
Number of Trees
Figure 26. Transect 6, plot 13.
In plot P11614, TreeVaW identified 15 of 19 live trees present (78.95%). The
four trees not identified were the four smallest trees within the plot. Overall,
TreeVaW slightly overestimated height for all trees identified, with higher
overestimations for the tallest trees (2-3 meters each).
82
Figure 27. Transect 6, plot 14.
In plot P11615, TreeVaW identified 20 of 25 live trees present (80%). Trees
identified were among the tallest and shortest trees. For shorter trees identified,
over- and under-estimation of height occurred while all of the heights for taller
trees identified were overestimated by TreeVaW.
Figure 28. Transect 6, plot 15.
In plot P11616, TreeVaW identified 14 of 21 live trees (66.67%). The tallest trees
in this plot were the trees that TreeVaW identified, and the heights of each were
overestimated by 1-5 meters.
83
Figure 29. Transect 6, plot 16.
In plot P11617, TreeVaW identified 14 live trees with only 13 live trees present
(107.69%). The heights of all but the shortest and tallest trees were
overestimated.
Figure 30. Transect 6, plot 17.
DISCUSSION
The TreeVaW software identified most of the tallest trees in most of the plots
surveyed. This was evident in the graphs for each plot in which DHPs were taken
84
for LAI analysis. These results are expected, given that the method used to
identify trees uses a CHM. The tallest trees in stands form the peaks and valleys
of the CHM, and thus are easier to segment and identify (Richardson and Moskal
2011). In a dense stand such as WS1, the tallest trees would mask or cover the
smaller trees in the understory, and any existing software or method to identify
trees would likely miss them. Richardson and Moskal (2011) found that the
precision of their method was higher for the two taller height classes than the two
shorter height classes. These findings, also supported by previous studies,
strongly indicate that LiDAR data are consistently accurate in identifying the
tallest, or dominant, trees in a stand. Additionally, on WS1, due to complex
terrain, dominance is in part a function of slope. On steeper slopes, which are
found deeper "within" the watershed (for example, P11108 and P11211),
dominant trees are codominant with respect to the plots "above" them on the
slope. Along the ridgeline (for example, all plots on transect 6, dominance is more
representative of actual tree heights.
Andersen et al. (2001) encountered similar results when using LiDAR data
to create a canopy surface model that measured individual trees. The
morphologically based tree measurement algorithm performed better where tree
crowns were larger and more widely dispersed. In denser areas, the canopy
surface algorithm output errors of omission and commission. Morphologically,
there are also distinct differences in tree DBH to tree height relationships on WS1,
especially with respect to aspect. For PSME of the same age, individual trees on
the north-facing slope are of smaller DBH and closer spacing than those on the
85
south-facing slope. We suggest that the ecological mechanism driving these
morphological differences is soil moisture availability; although trees on the south
facing slope will likely have greater radiation, soil moisture on these exposed
slopes is generally low and weathering is more extensive than on north-facing
slopes. The initial establishment of the plantation on these slopes was poor, and
young PSME benefited from growing further apart and slowly under the
protection of drought-tolerant hardwoods. On the north-facing slope, initial
establishment of the plantation was highly successful, and PSME with adequate
soil water and nutrients allocated much biomass to stem height to compete for
limited radiation. This more rapid height growth of PSME on the north-facing
slope lead to trees of the same age being at different stages of stand development,
with north facing PSME undergoing canopy closure sometime between 1995 and
2001. This led to local mortality events and overall productivity decline. Although
model algorithms are able to identify general patterns in PSME height and foliar
component based on DBH, the specifics of stand physiology on WS1 due to
complex terrain may not be fully represented, and this may account for a number
of discrepancies we see in modeled versus measured height.
Hirata et al (2009) tested individual tree identification in stands of variable
thinning by using a digital canopy model (DCM) and assessed the pulse
penetration of the LiDAR itself. The researchers inverted the DCM and performed
watershed segmentation in the basins and drainage divides to identify individual
trees. The total number of trees identified (607 of 748 or 81.1%) is remarkably
similar to the tree identification results in this thesis. More trees were accurately
86
identified in the areas where heavy or moderate thinning occurred compared to an
area that was not thinned. A potential explanation for this is that plots with a
comparatively sparse distribution of trees allowed for LiDAR penetration of the
understory. Furthermore, the ratio of unidentified trees to all standing trees was
higher for trees in the smallest height classes compared to the taller height classes
of trees. Also similar to some plots in this thesis (P11108, P11110 and P11212),
Hirata et al. (2009) found that LiDAR underestimated the heights of trees shorter
than 15 meters, and overestimated heights of trees taller than 15 meters.
In their research, Edson and Wing (2011) used three different tree
extraction methods to identify individual trees in mixed conifer stands. They used
a watershed segmentation method of inverting a canopy height model, manual
extraction using the FUSION software, and automatic extraction using TreeVaW
(as in this thesis). These three methods had varying results, but the FUSION
method was quickly abandoned because how long it took to find individual trees
and issues in identifying understory trees in dense stands. The authors also noted
that where and how many LiDAR pulses strike and reflect off trees impacts the
identification and measurement of individual trees. Especially with conifer
species, the odds of a pulse striking the highest, single apex point of a tree are
low. The odds decrease further when the apex of the tree is below the upper
canopy. Similar to the research previously discussed, Edson and Wing found it
relatively easy to identify upper trees in dense plots, but smaller trees were
shrouded by larger ones, and identification of smaller trees was difficult at best.
They found identifying young conifers in a clear cut also difficult: even with a
87
pulse rate of 8-10 pulses per square meter, the sparseness of young conifer foliage
caused only one or two LiDAR pulses to strike individual trees, making
identification difficult to impossible.
New Methods in Tree Extraction from LiDAR Data
Seeking to improve upon known methods of individual tree identification, Lee
and Lucas (2007) noted that most previous work focused primarily on the use of
Canopy Height Models, and this approach has proven only mildly successful for
mapping and attributing stems in complex, multilayered forests. They therefore
developed a novel complementary approach using a Height-Scaled Crown
Openness Index (HSCOI), which provides a quantitative measure of the
penetration of LiDAR pulses into the canopy. To quantify the penetration of
LiDAR pulses into the canopy, the data were transformed into a 3D voxel matrix
of 1 cubic meter squares. Within the 3D matrix, canopy voxels containing returns
were attributed with the tallest recorded LiDAR height value within the voxel
space. The HSCOI was then constructed using the weighted summation of a
proxy variable of the inverse canopy density, or 1/the number of voxels
containing returns per 1 square meter vertical column. The HSCOI metric thus
translates the LiDAR point observations into a measure of relative penetration of
the LiDAR pulses by scaling them from the top of the canopy so that 0%
indicated no penetration and 100% full penetration of the pulse to the ground.
Figure 31 shows the difference between the created HSCOI and a traditional
CHM.
88
Figure 31. A comparison of a) the newly developed HSCOI metric, and b) a CHM. The high point in
the CHM is transformed to a low point in the HSCOI because it would be an area where penetration of
LiDAR pulses would be very low (Lee and Lucas 2007) .
Lee and Lucas (2007) applied the new metric (HSCOI) to mixed species
forests in Queensland, Australia, which facilitated the mapping of the forest areas,
delineation of tree crowns and clusters, and estimation of canopy cover.
Computed tree densities compared well with field measurements at the stand
level, and the most consistent results were from stem densities of less than 700
stems/hectare. This metric was combined with the CHM shown above to estimate
dominant stem height, crown cover, and foliage and branch projective cover to
sufficient levels of inventory for the stands. However, this method resulted in less
accurate measurements when applied to a different forest type with increased
average height and canopy closure.
Li et al (2012) also set out to develop a new approach that built on the use
of CHMs in order to segment individual trees, using a similar LiDAR system as
this thesis, and begun by separating ground returns from aboveground returns.
The data were ‘normalized’ by subtracting the vegetation point cloud from the
ground point DEM. As a result, the elevation value at any point represented its
89
height from the ground. The new method relies on the relative spacing between
trees, which tends to increase from the base of trees to the top of trees. The
authors note that, although there is overlapping of trees in dense forests, spacing
between trees exists at higher parts of the canopy. Coniferous trees have, in
simplicity, a conical shape, such that the crown radius at the base of the crown
(where it would be measured from a ground based measurement) will be greater
than that at the top of the crown. Thus, the point density of LiDAR returns, or any
metric modeling canopy architecture, will be sparser at heights further from the
canopy base. This method works from the top of a tree, and ‘grows’ the tree by
including nearby points and excluding points from other trees based on their
relative spacing, and becomes more of a challenge farther down the tree as
spacing decreases and trees overlap. However, classifying points sequentially,
from highest to lowest, overcomes this challenge. By developing and defining
appropriate spacing thresholds, most points can be assigned to their corresponding
trees. Trees are segmented using three variables: 1) either these fixed or adaptive
thresholds, 2) a minimum spacing rule, and 3) a horizontal profile of tree shape.
Reducing undersegmentation is accomplished by using a small threshold, and
oversegmentation is reduced using the shape and distribution of the points. This
top to bottom approach is iterated until all points have been classified into their
corresponding sets, with each set corresponding to an individual tree.
The authors tested their algorithm in a conifer stand in the Sierra Nevada
range of California and found that it increased the accuracy of individual tree
detection compared to other methods. 86% of trees were identified, and 94 % of
90
the segmented trees were correct. The authors state that this method holds
promise for use in similar, mixed and complex coniferous forests, but its
effectiveness in other forest types, namely deciduous forests, requires further
evaluation. This method is similar to other methods that use a CHM in that it
identifies local maxima as the top of a tree, differing in that the new algorithm
identifies the global maximum (highest point), segments that tree, and then
removes the data points associated with that tree from the point cloud. Then, the
process starts again, finding the ‘next’ global maximum and segmenting that tree,
iterating until all points have been classified. Using untransformed LiDAR data
avoids interpolation errors that emerge when a point cloud is transformed into a
CHM.
CONCLUSIONS
LiDAR data have been used extensively to measure canopy characteristics and
attributes, and this thesis has continued that work by using LiDAR data to identify
individual tree location, height, and crown width within the H. J. Andrews
Experimental Forest. The software used, TreeVaW, was originally developed for
a much different forest type, and species and site specific parameters are
available, problems still persist. These problems are most evident for crown width
estimates, which were not used for analysis since they seemed unreasonably
small. The stand type for which TreeVaW was developed consists of much less
dense, more spread out trees, and the program seems to have issues delineating
points for individual trees for the crown width estimates in denser stands. Also,
the denseness of the canopy seems to limit the ability of TreeVaW to identify
shorter trees in the understory. This is not entirely surprising since the software
91
relies on a canopy height model (CHM) for extraction of individual trees, and the
taller trees are the most visible in the CHM.
TreeVaW performed very well in some plots, but in many plots it
underperformed by failing to identify all or most trees). The program seemed to
perform better in plots lower density, and well as in plots with mostly taller trees.
TreeVaW tended to slightly overestimate heights of taller trees, and when it did
identify shorter trees it often underestimated the heights of those trees. Reliance
on the CHM for tree extraction may be why these errors occur and the use of a
Height-Scaled Crown Openness Index (HSCOI) presented by Lee and Lucas
(2007) might solve these issues. This said, the overall percentage of trees
identified by TreeVaW in all 133 vegetation plots (~82%) compares well to other
published results.
92
Chapter 4- LiDAR Data Visualization and Overview of Visualization in
Natural Resource Management
93
INTRODUCTION AND BACKGROUND
The use of data-driven visualizations in forestry management has become
increasingly common in the last two decades. Bergen et al (1998) assessed the use
of data-driven simulation, dimensional accuracy and realism in a landscape
visualization tool as a part of their research. Even then, computer-based
simulations of landscapes were recognized as an effective tool for assessing the
potential impact of land-use decisions and s were commonly used by decision
makers to assess the visual impact of forest operations, such as road building and
harvesting, and as an aid in designing mitigation strategies. Early tools for
decision makers relied on using computers to manipulate two-dimensional
scanned photographic or videotape images for visual assessment and mitigation
design after harvest. This computerized image manipulation carried the
advantages of speed, efficiency and flexibility, but with two major drawbacks.
First, image manipulation as a method of landscape simulation was primary an
‘artistic technique’, and therefore disconnected from project data. The second
major drawback was that image manipulation methods lacked dimensional
accuracy, and translating design information from a manipulated 2D image to a
3D landscape was difficult, if not impossible.
Bergen et al. (1998) also offered a review of the early 3D landscape
visualization tools, and noted that the development of these methods closely
paralleled developments in computer technology. Early computer tools included
the Perspective Plot software and PREVIEW, both of which were based on digital
terrain model (DTM) representations of a landscape and displayed crude line
drawings of terrain and vegetation. More complex, realistic and accurate
94
simulations followed with the development of the SmartForest and Vantage Point
software, which at the time of this article represented the state-of-the-art. Both
draw individual trees to form forest stands over a DTM. Vantage Point generated
and displayed color images of forest landscapes up to 8000 ha, and was used in
the evaluation of visual quality and the visual impact of forest operations. Both
systems used measured data, which added validity and credibility to the resulting
image. The tools solved the two drawbacks mentioned previously, by relying on
actual data and solving the dimensional accuracy issue by allowing the user to
interact more directly with 3D landscape data. Users could display and manipulate
design information on a perspective image of the landscape.
Bergen et al. (1999) concluded by stating that several obstacles remained
before landscape simulation tools could find wider acceptance and use. The 3D
visualization tools of the time, while offering flexibility in representing and
altering data sets, lacked realism. Images that appeared more realistic were less
flexible and often not based on measured data. The authors posed two remaining
questions: 1) From a theoretical perspective, how much realism and accuracy is
required to make value judgments? 2) Is it practical and economical to gather
enough data to create more realistic simulations? The authors claimed that more
complex computer systems could create more detailed images faster, approaching
real-time manipulation of photo-realistic images representing 3D data sets, but it
was unknown whether theory and data gathering abilities would keep pace with
the technology.
95
McGaughey (1998) also reviewed techniques for visualizing the effects of
forestry operations. His research compared how geometric modeling, where
mathematical models of individual components are built and then assembled to
create a model of a forest stand or landscape, compares with video imaging. The
research also explores a hybrid of these two methods as well as the method of
video draping, or overlaying an image over a DTM. These four methods were
compared with respect to their data requirements, level of realism in final scene,
operational complexity and data integrity. The ultimate consideration in choosing
a method was found to be reliant on the size of the project area, overall goal of the
visualizations, amount of detail that must be shown, and amount of available data
describing the area. Finally, the author reviewed the available software at the
time, noting that commercial systems can be expensive, often need a specialized
operator to produce results, and require that typical forestry data be converted into
a suitable format. However, public domain visualization and image-editing
capabilities suitable to forestry visualization were available at the time for little or
no cost.
Since the time when SmartForest and Vantage Point were considered
state-of-the-art, visualization tools have come a long way, and one reason for this
is the rise in use of remote sensing capabilities such as LiDAR. Kao et al. (2005)
point out in their research that the development of remote sensing capabilities has
ushered in an era where large quantities of multidimensional and multivariate data
are routinely analyzed. Most work to date, they noted, relied on statistical
summaries (a.k.a. data aggregations) to characterize the distribution of data with a
96
small set of descriptors. These methods reduce the dimensionality of the data set,
often making it impossible to access the primary data from the visualization, but
making visualization straightforward. However, this approach fails when the
distributions are nonparametric, and when they are multimodal. Kao et al. (2005)
proposed instead allowing for exploration, query and comparison of LiDAR data
distributions, in order to increase opportunities to query multi-valued data in new
ways that better help scientists and other stakeholders understand the spatial
distributions of geophysical and ecological phenomena. This, in theory, could be
done both at single locations and across the spatial domain. The authors utilized
so-called spatially distributed probability density functions (pdfs), created from
multiple return LiDAR data, and noted that the major contribution of their work
was a paradigm shift that allowed ecologists to think of and analyze data in terms
of full distributions, not just summary statistics. The major contributions of this
paper, as listed by the authors, were: 1) provide automated and interactive ways to
analyze forest canopy distributions from LiDAR data, 2) make it easier for
scientists to analyze distributions derived from LiDAR data, and 3) allow
scientists to query distribution data for special features and then identify areas of
the spatial field with similar distributions and discover potentially interesting
distributions and their locations. Figure 32 shows a diagram of the visual analysis
of distribution data made possible by this work.
97
Figure 32. A diagram representing new types of visual analysis of LiDAR data could occur because of
research by Kao et al. (2005).
Fujisaki et al. (2007) offered an update to forest visualization systems in
their research paper describing stand assessment through LiDAR based forest
visualization using immersive virtual environment technology. The systems
mentioned include INFORMS, FMIS, Landscape Management System, and the
Stand Visualization System (SVS), all of which aimed for higher perceptional
effectiveness in forest visualization. These systems use advanced graphics
techniques and display technologies to develop fully interactive 3D visualizations,
referred to as immersive virtual environments (IVE). IVEs are used in many fields
including vehicle simulations, entertainment, architectural design, medicine and
surgery, and education, and the potential use of the technology for forest
visualization was first noted by McGaughey and Carson (2003).
98
In their research, Fujisaki et al. (2007) set up an experiment to compare
IVE methods to field recorded videos of immature and mature loblolly pine
(Pinus taeda) stands in Mississippi. LiDAR data was used to create a virtual
forest, which was then projected with an interactive room-sized stereoscopic
display, so that participants could ‘walk’ through the forest. Other participants
viewed only the field video recordings of the stands. Study participants’ estimates
of stand characteristics were then compared, and significant differences were
found in the two groups’ estimates of height class and rotation stage, even though
estimates of stocking, tree size class, stand structure, and hardwood competition
were similar. These results led the researchers to conclude that IVE technologies
and visualizations in general could be potentially useful in natural resources
management, and that further study of the economic aspects and interface
development were required to further develop such technologies into an
operational system.
Stoltman et al. (2004) saw visualization of forests as a pathway for public
participation in the forest planning process. Working within the state of
Wisconsin, USA, these researchers noticed that visualization technology at the
time was used solely by researchers and consultants, and not by natural resource
managers. The authors reviewed a 3D forest visualization system, developed for
use by the Wisconsin Department of Natural Resources, which incorporated a
library of photographs of trees, snags, and logging debris to realistically depict
forest management activities. The system was linked to a GIS so that available
forest survey data could be incorporated, and was built to be as user friendly as
99
possible, for use by managers without extensive computer knowledge. The
authors found the system to be a success, noting that computer visualization of
forest management facilitated continued public involvement in forest
management. The key, the authors claimed, is getting the system into the hands of
managers. Finally, the authors state that the development of forest visualization
parallels the development of GIS, moving from being in the hands of only a few
computer-literate individuals to being widely used by the broad scientific and
natural resource management community.
Similarly, Kopytko et al. (2008) attempted to connect resource managers
to visualizations and in doing so connect the managers with ecologists attempting
to answer fundamental scientific questions regarding the natural world. This
research points out the fundamental differences between how ecologists conduct
their research and how managers make decisions. The translation of ecological
values into management procedures is difficult, but there exists a push from the
public to incorporate ecological values into decision making, in addition to
traditional revenue maximization. The results of this research allowed for better
characterization of canopy crowns, with an informatics tool to provide structure
summaries that better enable people to look at the data and classify or cluster trees
according to structural similarity. This said, the authors noted that new
information technology was still needed to accomplish the ecology research goals,
including 1) interpolation and extrapolation of missing data on geographically
complex topographies, 2) better models and tools to develop complex situations,
100
and 3) better pattern recognition and visualization so data can be compared and
contrasted.
METHODS
This section describes the methods used in this thesis research to achieve data
visualization. Before visualization with FUSION could begin, the software
requires that a FUSION project be built from various sources of data. First the
raw LiDAR data was loaded. The original LiDAR data acquisition completed by
Watershed Science, Inc., segmented the LiDAR data into 500 mB bins, which
separated ground returns into ‘ground bins’, and canopy or non-ground returns
into high_bins (Spies et al., 2011). This segmentation was done because otherwise
the size of the data file would have been exceptionally large; it also separated
ground returns from canopy returns which simplified analysis. Thus, the full HJA
data set did not have to be loaded, but instead only the six bins that covered the
spatial extent of WS1. The ground and high bins are related to one another by
number, and the following ground and high bins were used: bin_017, bin_020,
bin_21, bin_025, bin_026 and bin_031. Using the ground bins, a bare earth model
was created in FUSION using the ASCII raster terrain model tool, which converts
ASCII points from the bins into a .dtm file that FUSION can read and analyze.
The ground bins were combined into one .dtm file, labeled All_WS1.dtm.
After the raw data and bare earth model were loaded, an image of WS1
was acquired. This image was downloaded from the USGS seamless image
viewer/Oregon Imagery Explorer (http://nationalmap.gov/viewer.html) This site
has orthophotographs for the entire state of Oregon, and the online interface
101
allows a user to zoom into an area, select the extent of image desired, and
download the image as a 4 band National Agriculture Imagery Program (NAIP)
image. An orthophotograph is an aerial image that is geometrically corrected, or
orthorectified, so that the scale is uniform and distances can be accurately
measured on the image itself. After the orthoimage was loaded into FUSION, the
vegetation plot point shapefile (created during hemispherical photograph analysis
(Chapter 2) was loaded into FUSION as a point of interest (POI) file.
Visualization of LiDAR Data for Plots and Transects
Once all the required data files were loaded into FUSION, creating visualizations
using LDV was straightforward. In the FUSION interface a user can stoke or
select any size rectangle or circle and LiDAR Data Viewer (LDV) automatically
loads with the selected data visualized. The sample options button in the FUSION
user interface was utilized to create original visualizations of the LiDAR point
cloud for each plot and wider views of entire transects in which DHPs were taken.
Since the plot centers were loaded into FUSION as a POI file, snap sample points
to nearest POI point was selected from the options menu. Also, since a ground
model had been loaded, the subtract ground elevation from each return was
selected to normalize the point cloud, and give elevation of points above the
ground rather than above sea level. The sample shape selected was a fixed circle
with a diameter of 18 meters (to align with the 9 meter radius plots). The set
shape, along with the snapping to the nearest POI ensured that not only a user
click near a plot aligned with the center of the plot, but also that the outer edge of
102
the selection was matched to the plot itself. The visualizations of each plot were
colored by height, normalized by ground model (Figures 37-55, left).
Visualization of TreeVaW Output Trees
Special help for this section of the thesis came from Lee Zeman, a collaborator on
the VISTAS project who created images of TreeVaW’s output using the visual
programming language Processing (http://processing.org). The position and
heights of the trees were read from TreeVaW’s output files. DBH was
approximated from height using the allometric equations given in Garman (1995);
crown radius was approximated from DBH using the two-term allometric
equation given in Gill (2000). All trees were assumed to be Psuedotsuga
menziesii.
RESULTS
Watershed 1 Visualizations
Figures 33-36 are visualizations of LiDAR data created using FUSION software.
103
Figure 33. FUSION visualization of WS1 Digital Elevation Model (DEM) and LiDAR data of 19 vegetation
plots where digital hemispherical photographs were taken.
104
Figure 34. Overhead view visualization of WS1 DEM and LiDAR data of vegetation plots.
105
Figure 35. Overhead view visualization of WS1 DEM and LiDAR data of vegetation plots with digital
orthophotograph of vegetation in the Watershed overlaid.
106
Figure 36. Visualization of LiDAR data and DEM of vegetation plots in transect 2 (above) and transect 6
(below).
Visualizations of LiDAR Point Cloud and TreeVaW Identified Trees
Figures 37-55 are visualizations created for each plot where digital hemispherical
photographs were taken for LAI measurements (see Chapter 2). The image on the
left of each figure is the visualization produced with FUSION, and the image on
107
the right is the visualization of trees identified using TreeVaW, created using
Processing from data output by TreeVaW.
Figure 37. Transect 1, plot 9.
Figure 38. Transect 1, plot 10.
108
Figure 39. Transect 2, plot 5.
Figure 40. Transect 2, plot 6.
Figure 41. Transect 2, plot 7.
109
Figure 42. Transect 2, plot 8.
Figure 43. Transect 2, plot 9.
Figure 44. Transect 2, plot 12.
110
Figure 45. Transect 2, plot 13.
Figure 46. Transect 6, plot 8.
Figure 47. Transect 6, plot 9.
111
Figure 48. Transect 6, plot 10.
Figure 49. Transect 6, plot 11.
Figure 50. Transect 6, plot 12.
112
Figure 51. Transect 6, plot 13.
Figure 52. Transect 6, plot 14.
Figure 53. Transect 6, plot 15.
113
Figure 54. Transect 6, plot 16.
Figure 55. Transect 6, plot 17.
DISCUSSION
The FUSION visualizations of LiDAR data using are presented above. These
visualizations show the lay of the land in WS1, highlighting the steepness of the
terrain and the denseness of the forest canopy within the vegetation stands. These
images show why estimation of LAI has been a difficulty in the past (see Chapter
2), and why on complex terrain with variable stand composition, visualization of
forest structure has continued to be an issue even within this thesis.
114
The visualizations of 3D stem maps from the TreeVaW output are also
presented. These images show the trees identified by TreeVaW in 19 vegetative
plots in the WS1. However, since no x-y location information was available for
the vegetation surveys, similar visualizations of the surveyed trees could not be
completed. Although stem maps exist for some of the plots in WS1, they are not
readily available and have not been verified recently enough to compare to
LiDAR derived metrics and TreeVaW visualizations (Peterson, personal
communication; Wooley, personal communication). If recent stem maps from the
watershed’s vegetation plots could be verified a visual comparison could be made.
One purpose visualizing the TreeVaW output is to visually examine the trees that
TreeVaW identified. With those visualizations, it became clear that where a thick
over story of taller trees is present, as in P11109 (Figure 37) and P11616 (Figure
54). TreeVaw is able to identify more trees and predict their heights more
accurately than where stand composition is complex, and understory trees are
present in P11211 and P11212 (Figure 44). In the field, these two plots (P11109
and P11616) fall on areas where harvest regeneration attempts were highly
successful. The north-facing aspect of WS1 was replanted with 1-1 seedlings (1
year total age, 1 year development at the nursery) only once, but soil conditions
(deep, moist) fostered early coniferous development. Transect 1, at a low
elevation, also has sufficient belowground moisture to foster a dense stand for
much of the year. Transect 6, located along the ridgeline, is very near the second
original landing from the harvest event. Areas around the landing were replanted
specially, and they have very flat slopes and deep soils. Tree growth on this plot is
115
particularly well organized, with 1-1 and 2-1 (2 years total age, 1 year
development at the nursery) seedlings planted in row arrangement that is still
visible today. Strong canopy coverage from these successful plantings, combined
with purposeful row spacing, leads to the development of a healthy overstory and
precludes the establishment of understory species, with the exception of hardy
shrubs. Thus, we note that on these plots, TreeVaw did not identify many, if any,
smaller trees in the understory. However, where there were gaps present in the
taller trees, such as P11110 (Figure 38) and P11208 (Figure 42) TreeVaW was
better able to identify the smaller trees found underneath these gaps. On P11110,
a windthrow event, likely corresponding with the floods of 1994-1995 and the
odd weather patterns that season, caused great instantaneous mortality leading to
the establishment of gaps. These gaps were readily seeded by Tsuga heterophylla
blown in from neighboring stands. On P11208, which is located on the south
facing slope, the presence of local topography, specifically a basaltic caprock
feature, does not allow for the contiguous establishment of deep rooted conifers,
thus smaller trees such as Prunus emarginata, Cornus spp., and Castanopsis
chrysophylla have established, needing less belowground resources for growth
and structure and benefiting from the increased radiation on the south-facing
slope (Peterson, 2012).
Visualization in Ecosystem Monitoring and Processes
The research presented in the introduction of this chapter explains how
visualization of forests and trees have been used in forestry management; separate
from this is visualization of forest stands for the purposes of understanding
116
ecosystem processes and function. Few articles in the literature have focused on
visualization of ecological phenomena for purposes of understanding these basic
ecological phenomena and functions. Below, I selected a short list of these
studies.
Krisp (2004) examined the use of 3D visualization to measure ecological
barriers to movement for species in Finland. To build a model for visualizing the
effects of barriers, the authors assumed that different land covers and land uses
have a dissimilar impact on wildlife movement, and they assigned a qualitative
value to each landscape component considered, on a species-species basis. In
general, the values assigned were low for natural elements and high for artificial
ones. Then, by linking the qualitative value to the landscape components in a map
the movement limitations for species could be assessed, first using a 2D map, and
then a 3D map that had the ability to change the perspective or view. This work of
modeling barrier effects in 3D can help visually identify ecological corridors,
networks or bottlenecks. This research represents early work in environmental
phenomena visualization.
Omasa et al. (2007) explored the use of LiDAR to produce 3D imaging
not only for understanding canopy structure, similar to Chapter 3 of this thesis,
but also for detecting and understanding plant responses to varying phenomena.
The authors reviewed the development of LiDAR systems and their application
from the leaf to canopy level remote sensing. Plant properties explored with 3D
LiDAR imaging included canopy height, canopy structure, carbon stock, and
species distribution, and plant responses (change in growth and shape) were also
117
assessed. Canopy height, an important variable needed to estimate the 3D
properties of trees, was historically underestimated in LiDAR metrics because of
low likelihood of pulses with a density of <1 pulse m-2 hitting the highest point
on any given tree. More recently, high pulse densities have been used more often
to solve this underestimation problem. Ground-based LiDAR provides a more
precise 3D image of individual trees, with resolutions ranging from .05 -10 cm.
Tree heights can be measured from these ground based systems, but if tree tops
are blocked by other trees, estimating tree height can be difficult. A way around
this problem is to take ground based LiDAR measurements in multiple locations
around a tree of interest, and images from the different locations can be coregistered and merged.
Landscape Visualization
While visualization of forest processes has been scarce, visualizations of
landscape phenomena have been more common. Paar (2005) examined the role of
landscape visualization software for landscape and environmental planning in
Germany. The author questioned hundreds of private consultancies and public
authorities about the current and future use of 3D landscape visualizations in
environmental decision making. Overall, the respondents had great expectations
about where 3D visualizations would take landscape planning, with 91% of
respondents believing that 3D visualizations would bring additional benefits to
landscape planning. Many of the issues confronted by the respondents were based
on software problems. Respondents who use visualization software mentioned
technical problems, including insufficient representation of plants and habitats,
118
and slow rendering. The group representing ‘non-users’ of software mentioned
the high cost of existing software as a barrier to use. Respondents from both
groups mentioned training and usability of available software as a difficulty. The
results of Paar’s surveys show that the application of visualizations facilitates
improved communication between experts and laypersons, and the author
concludes that new technological developments within computer graphics will
continue to aid landscape planners and environmental managers with use of 3D
landscape visualization tools, but that innovative, new 3D landscape visualization
tools are needed to continue this trend.
Wang et al. (2006) described 3D visualizations of forest landscapes as
“quantitative ecological information-based techniques that can be used to
visualize forest structure, dynamics, landscape transformations and regional
plans.” The authors posited that the use of existing public data sets to create
visualizations can replace time consuming and expensive stand-level surveys. The
authors utilized the Forest Inventory and Analysis d (http://www.fia.fs.fed.us) and
several other existing vegetation databases to visualize a section of the
Chequamegon National Forest in Wisconsin. The authors created visualizations
from the individual stand to the landscape level, using two software packages:
Tree Professional 5 and Visual Nature Studio 2.01, and concluded that public data
sets are suitable and useful for visualizing the dynamics of forests and landscapes.
This research shows that large pools of data are available, but they may be un- or
under-utilized by researchers and decision makers.
119
Sherren et al (2011) examined lessons learned from visualizations of the
landscape and in habitat assessment in the current trend of tree decline in
Australia. The authors presented photo-realistic visualizations of landscapes to
stakeholders representing the results of modeling of different future scenarios of
tree decline. The visualizations represent likely outcomes of the ‘status quo’
compared to alternative tree planting remediation efforts in an attempt to show the
consequences of each on landscape aesthetics and biodiversity. Through the
visualizations the authors found that the current trend of tree decline is contrary to
stakeholders’ values concerning the region’s social and ecological well-being.
The authors showed that the visualizations could be used to create
interdisciplinary collaboration and engage stakeholder involvement in forest
management.
CONCLUSIONS
This chapter outlined the importance of visualizing forest and landscapes in the
natural resource management and ecosystem research. Visualization of forest
landscapes and structure for resource management has become common in recent
years, but visualization of forest processes and function is just beginning. The
data intensive nature of the work requires novel visualization tools to understand
many and massive current data sets. Visualization also stands as a key
communication tool in keeping decision makers, stakeholders and the general
public connected and informed in decision making. The visualizations presented
in this chapter suggest how LiDAR data might be used to visualize forest
structure and represent actual trees. The use of remote sensing technologies such
120
as LiDAR greatly reduces the time, resources, and personnel needed to create
visualizations for use in both natural resource management and ecosystem
research.
121
Chapter 5- Conclusions and Future Research
122
This chapter summarizes conclusions already mentioned in Chapters 2-4, and
looks forward to research directions for this field. The objectives of the thesis
frame the following conclusions:
Thesis Objective 1: Determine accurate LAI estimates for a subset of
permanent vegetation plots in WS1 using digital hemispherical photography
(DHP)
Hemispherical photographs taken in 19 vegetation plots in WS1 were analyzed
three different ways using SLIM software: 1) limiting the ‘scope’, or angle of
view, to estimate for the plot only, 2) using a wider scope to estimate the
immediate area around the plot, and 3) using the average LAI values for DHPs in
four cardinal directions around the center of the plot. The three resulting estimates
were not statistically correlated, which may be caused by either 1) the hillslope
affecting the two wider measures of LAI, or 2) how light was measured in the
DHPs for the software analysis.
Issues encountered in these LAI estimates are similar to previous work in
WS1. The steepness of the terrain and the high vegetation density make accurate
estimation of LAI very difficult.
Thesis Objective 2: Use estimates of LAI obtained from DHP to build a
LiDAR based model of LAI for all 133 permanent vegetation plots
Using the estimates of LAI for the plots obtained from DHP, we developed a
LiDAR based model to estimate LAI using the LiDAR metric surface volume,
and used this to estimate LAI for all 133 vegetation plots in the watershed. The
resulting LAI values also revealed a strong relationship when compared to
123
Cover*Height for each plot (R2 = 0.5467, p < 2.2e-16). The use of LiDAR to
estimate LAI for the watershed thus seems promising. The development of a
LiDAR only model to estimate LAI is a possible solution to estimating LAI in
steep, densely vegetated watershed, since LiDAR seems to adequately measure
complex terrain.
Thesis Objective 3: Calibrate the LiDAR based model for LAI to create LAI
maps for the entire watershed.
The LiDAR model for LAI could not be calibrated to create LAI maps for the
watershed as a whole. A reason for this is the LiDAR derived volume metric used
is not directly related to LAI, but rather to leaf area, which is mechanistically
different because LAI incorporates ground area and is a dimensionless value
(Richardson, personal communication). Because of this, TreeVaW software was
used to estimate the total number of trees in the watershed, which could then be
used with allometric relationships as an alternative for estimating LAI for the
entire watershed. Future research will need to validate the number of trees in the
watershed either through field measurements or another remotely sensed metric.
Thesis Objective 4: Test the ability of software programs to extract and
identify individual trees from LiDAR data in all 133 permanent vegetation
plots in WS1
The TreeVaW software program was used to extract and identify individual trees
within all 133 vegetation plots in the watershed. Overall, TreeVaW identified
over 82% of trees when compared to vegetation surveys completed in the plots.
TreeVaW performed better in plots with lower tree density and those with taller
124
trees. TreeVaW was also used to identify all trees in the watershed, and identified
nearly 77,300 trees. Since no comprehensive tree counts have been taken in the
watershed the estimate of total trees could not be validated. However, the
vegetation plots represent the entire range of slope, aspect, and elevation present
in the watershed, it is reasonable to assume that the estimate of total trees would
be comparable to the 82% of trees identified in the vegetation plots.
Thesis Objective 5: Create novel visualizations of LiDAR data and individual
trees in the vegetation plots.
Visualizations were created from the LiDAR point cloud data using the FUSION
and LiDAR Data Viewer (LDV) software. The output of individual trees from
TreeVaW was also used to create 3D stem maps (stand visualizations) using the
programming language Processing (http://processing.org). The latter
visualizations clearly displayed gaps in the canopy of the plots where TreeVaW
was better able to identify shorter trees, and also showed where the canopy was
dense and limited TreeVaW’s ability to identify understory trees.
Future Research Directions
This thesis research has answered some questions about LAI, LiDAR data and
tree extraction, but has also made apparent potential directions for future research.
First, a comprehensive look at why ground based measurements of LAI based on
hemispherical photographs differ from LiDAR based methods is in order. The
importance of LAI in ecosystems studies has been demonstrated by other prior
research, but difficulties remain in obtaining accurate estimates LAI in densely
vegetated, steep terrain. The use of LiDAR is promising for estimating LAI.
125
Incorporating slope correction into digital hemispherical photograph analysis and
developing a site-specific model for correcting LiDAR estimates could make LAI
estimates more accurate.
Secondly, a comprehensive examination of why TreeVaW performed very
well in some vegetation plots but poorly in others is needed. The initial analysis
showed that TreeVaW performed better in plots with lower tree density and taller
trees. However, comparing the aspect, slope, or elevation of the plots may provide
insight into what factors affected TreeVaW’s identification of individual trees.
Since the forest type in Texas, where TreeVaW was developed, differs
considerably from the Pacific Northwest Cascades, changes to algorithms and the
software itself may provide more accurate identification of trees. This could
especially improve the estimates of crown width which were unusable for this
research.
Thirdly, the visualizations of individual trees presented in this thesis could
assist researchers in canopy gap research. These visualizations give the researcher
the ability to “see” the size and shape of the gaps in the canopy, which could then
be compared to remotely sensed metrics and topographic and climatic
measurement. This gives the researcher the chance to assess variability on a
subplot scale, without having to break the data down to the subplot scale.
Finally, further empirical data needs to be gathered to validate the
estimations of LAI, number of trees, and tree heights at the individual tree scale.
Current field measurements of LAI are inconsistent, and repeated sampling of the
same geographic locations under a variety of seasons and meteorological
126
conditions could help determine which measures are most accurate under which
conditions. The number of trees in the entire watershed also needs to be calculated
to assess TreeVaW’s accuracy on the watershed scale, perhaps from derived
metrics other than the LiDAR data (local “peaks” in height at a sub-crown
diameter scale). Finally, validation of tree heights from the ground needs to be
conducted specifically on WS1; although some validations exist for the LiDAR
data on the H.J. Andrews at large, these measurements were taken from the tallest
trees (old growth) and may not be applicable to those on WS1.
The ability to accurately map individual trees and measure LAI on a
watershed scale is increasingly important in both forest ecosystem research and
management. This thesis combines ground and remotely sensed methods to
measure LAI and utilizes current state-of-the-art software to identify individual
and visualize individual trees. This research represents an interdisciplinary
approach to solving complex environmental questions, and contributes novel
analysis to the field. Moving forward, the combination of visualization, remote
sensing analysis and software programming can move us closer to understanding
environmental phenomena and processes from an ecological and management
perspective.
127
LITERATURE CITED
Adams, T., P. Beets, and C. Parrish. 2012. Extracting more data from lidar in
forested areas by analyzing waveform shape. Remote Sensing 4: 682-702
Andersen, H.E., S. Reutebuch and G.F. Schreuder. Automated individual tree
measurement through morphological analysis of LiDAR-based canopy surface
model. Proceedings of the First International Forestry Cooperative Symposium.
Seattle, WA. 2001.
Barilotti, A., S. Turco, and G. Alberti. 2006. LAI determination in forestry
ecosystems by lidar data analysis. 3D Remote Sensing in Forestry international
workshop. Online (www.wau.boku.ac.at/fileadmin/_/.../3drsforestry/.../8b.3barilotti.pdf). Accessed 24 November 2012.
Barilotti, A., and S. Turco. 2006. A 3-D GIS for the sustainable management of
forest resources. IUFRO Landscape Ecology Conference, Locorotondo, Bari,
Italy. Online
(http://geomatica.uniud.it/pubblicazioni/pdf/2006/IUFRO%2006%20Barilotti.pdf)
Accessed 24 November 2012.
Barilotti, A., F. Sepic, E. Abramo, and F. Crosilla. 2007. Improving the
morphological analysis for tree extraction: a dynamic approach to lidar data.
ISPRS Workshop on Laser Scanning 2007 and SilviLaser 2007, Finland. Online
(www.isprs.org/proceedings/XXXVI/3-W52/.../Barilotti_2007.pdf). Accessed 24
November 2012.
Barilotti, A., and F. Sepic. 2010. Assessment of forestry parameters at single-tree
level by using methods of lidar data analysis and processing. Ambiência
Guarapuava 6: 81-92.
Bartelink, H.H. 1996. Allometric relationships on biomass and needle area of
douglas-fir. Forest Ecology and Management 86: 193-203.
Bergen, S.D., R.J. McGaughey and J.L. Fridley. 1998. Data-driven simulation,
dimensional accuracy and realism in a landscape visualization tool. Landscape
and Urban Planning 40: 283-293.
Bond, B.J., J.A. Jones, G. Moore, N. Phillips, D. Post and J.J. McDonnell. 2002.
The zone of vegetation influence on baseflow revealed by diel patterns of
streamflow and vegetation water use in a headwater basin. Hydrological
Processes 16: 1671-1677.
128
Breda, N.J.J. 2003. Ground-based measurements of leaf area index: a review of
methods, instruments, and current controversies. Journal of Experimental Botany
54(392): 2403-2417.
Carlson, T.N., and D.A. Ripley. 1997. On the relation between NDVI, fractional
vegetation cover, and leaf area index. Remote Sensing of Environment 62: 241252.
Chen, J.M. and T.A. Black. 1992. Defining Leaf-Area Index for non-flat leaves.
Plant Cell Environment 15: 421-429.
Chen, J.M., and J. Cihlar. 1995. Plant canopy gap-size analysis theory for
improving optical measurement of leaf area index. Applied Optics 34(27): 62116222.
Chen, X., L. Vierling, E. Rowell, and T. DeFelice. 2004. Using lidar and effective
LAI data to evaluate IKONOS and landsat 7 ETM+ vegetation cover estimates in
a ponderosa pine forest. Remote Sensing of Environment 91: 14-26.
Chen. Q., D. Baldocchi, P. Gong and M. Kelly. 2006. Isolating individual trees in
a savanna woodland using small footprint lidar data. Photogrammetric
engineering and Remote Sensing. 72(8): 923-932.
Chertov, O., A. Komarov, G. Andrienko, N. Andrienko, and P. Gatalsky. 2002.
Integrating forest simulation models and spatial-temporal interactive visualization
for decision making at landscape level. Ecological Modelling 148: 47-65.
Clark, J. and G. Murphy. 2011. Estimating forest biomass components with
hemispherical photography for Douglas-fir stands in northwest Oregon. Canadian
Journal of Forest Research. 41: 1060-1074.
Cohen, W.B., T.A. Spies, R.J. Alig, D.R. Oetter, T.K. Maiersperger, and M.
Fiorella. 2002. Characterizing 23 years (1972-95) of stand replacement
disturbance in western Oregon forests with landsat imagery. Ecosystems 5(2):
122-137.
Comeau, Phil. LITE and SLIM Programs for Estimating Light Levels Beneath
Tree Canopies.
http://www.ualberta.ca/~pcomeau/Light_Modeling/Lite_and_slim_intro.html
Accessed April 2012.
Coops, N.C., T. Hilker, M.A. Wulder, B. St-Onge, G. Newnham, A. Siggins, and
J.A. Trofymow. 2007. Estimating canopy structure of douglas-fir forest stands
from discrete return lidar. Trees 21: 295-310.
129
Curtis, R.O. 1967. Height-diameter and height-diameter-age equations for secondgrowth douglas-fir. Forest Science 13(4): 365-375.
Cushing, J. VISTAS Project Summary. National Science Foundation (NSF)
proposal number 1062572. http://blogs.evergreen.edu/vistas
Cutini, A., G. Matteucci, and G. Mugozza. 1998. Estimation of leaf area index
with the Licor LAI 2000 in deciduous forests. Forest Ecology and Management.
105 (1-3) 55-65.
Deblonde, G., M. Penner, and A. Royer. 1994. Measuring leaf area index with the
li-cor LAI-2000 in pine stands. Ecology 75(5): 1507-1511.
Dubayah, R., R. Knox, M. Hofton, J.B. Blair, J. Drake. Land surface
characterization using lidar remote sensing. Pages 25-55 in Hill, M.J., and R.J.
Aspinall. 2000. Spatial Information for Land Use Management. Gordan and
Breach Science Publishers, The Netherlands.
Dufrene, E. and N. Breda. 1995. Estimation of deciduous forest leaf area index
using direct and indirect methods. Oecologia 104: 156-162.
Edson, C., and M.G. Wing. 2011. Airborne light detection and ranging (lidar) for
individual tree stem location, height, and biomass measurements. Remote Sensing
3: 2494-2528.
Erdody, T.L., and L.M. Moskal. 2008. Lidar derived canopy fuel metrics for
wildlife modeling. Factsheet #4 Remote Sensing and Geospatial Application
Laboratory, University of Washington, Seattle, WA. Online
(http://dept.washington.edu/rsgal/). Accessed 24 November 2012.
Espana, M.L., F. Baret, and M. Weiss. 2008. Slope correction for LAI estimation
from gap fraction measurements. Agricultural and Forest Meteorology 148: 15531562.
ESRI (Environmental Systems Resource Institute). 2012. ArcMap 10. ESRI,
Redlands, California.
Evans, J.R. and H. Poorter. 2001. Photosynthetic Acclimation of plants to growth
irradiance. Specific Leaf Area and N-fixation. Plant, Cell and Environment 24:
755-767.
Evans, J.S., and A.T. Hudak. 2007. A multiscale curvature algorithm for
classifying discrete return lidar in forested environments. IEEE Transactions on
Geoscience and Remote Sensing 45(4): 1029-1038.
130
Falkowksi, M.J., A.M.S. Smith, P.E. Gessler, A.T. Hudak, L.A. Vierling, and J.S.
Evans. 2008. The influence of conifer forest canopy cover on the accuracy of two
individual tree measurement algorithms using lidar data. Canadian Journal of
Remote Sensing 34(2): S1-S13.
Feng, L., P. de Reffye, P. Dreyfus, and D. Auclair. 2012. Connecting an
architectural plant model to a forest stand dynamics model—application to
Austrian black pine stand visualization. Annals of Forest Science 69: 245-255.
Frazer, G.W., J.A. Trofymow, and K.P. Lertzman. 2000. Canopy openness and
leaf area in chronosequence of coastal temperate rainforests. Canadian Journal of
Forest Research 30(2): 239-256.
Fry, B. and Reas, C., Processing Library for Visual Arts and Design,
http://www.processing.org. Accessed November 2012.
Fujisaki, I., D.L. Evans, R.J. Moorhead, D.W. Irby, M.J. Mohammadi-Aragh,
S.D. Roberts, and P.D. Gerard. (2008) Stand assessment through LiDAR-based
forest visualization using immersive virtual environment technology. Forest
Science 54(1): 1-7.
Garcia, M., D. Riano, E. Chuvieco, and F.M. Danson. 2010. Estimating biomass
carbon stocks for a Mediterranean forest in central Spain using lidar height and
intensity data. Remote Sensing of Environment 114: 816-830.
Garman, S.L., S.A. Acker, J.L. Ohmann, and T.A. Spies. 1995. Asymptotic
height-diameter equations for twenty-four tree species in western Oregon.
Research contribution 10. College of Forestry Forest Research Laboratory,
Oregon State University, Corvallis, OR. Online
(ir.library.oregonstate.edu/xmlui/handle/1957/7655). Accessed 24 November
2012.
Gholz, H.L., C.C. Grier, A.G. Campbell, and A.T. Brown. 1979. Equations for
estimating biomass and leaf area of plants in the pacific northwest. Research
paper 41, Forest Research Lab, School of Forestry, Oregon State University,
Corvallis, OR.
Gonsamo, A., and P. Pellikka. 2008. Methodology comparison for slope
correction in canopy leaf area index estimation using hemispherical photography.
Forest Ecology and Management 256: 749-759.
Gonsamo, A., J.N. Walter, and P. Pellikka. 2010. Sampling gap fraction and size
for estimating leaf area and clumping indices from hemispherical photographs.
Canadian Journal of Forest Research 40: 1588-1603.
131
Goodwin, N.R., N.C. Coops, and D.S. Culvenor. 2007. Development of a
simulation model to predict lidar interception in forested environments. Remote
Sensing of Environment 111: 481-492.
Grier, C.C. and R.H. Waring. 1974. Conifer foliage mass related to sapwood area.
Forest Science. 20: 205-206.
Halpern, C. and J. Means. 2011. Pacific Northwest Plant Biomass Component
Equation Library. Long-Term Ecological Research. Forest Science Data Bank,
Corvallis, OR. [Database]
Hawk, G.M. 1970. Vegetation and stem mapping of watershed 10 H.J. Andrews
experimental forest. Internal Report 97 US Internal Biological Program,
Coniferous forest biome. Online
(andrewsforest.oregonstate.edu/pubs/pdf/pub1862.pdf). Accessed 24 November
2012.
He, Q., C. Cao, E. Chen, G. Sun, F. Ling, Y. Pang, H. Zhang, W. Ni, M. Xu, Z.
Li, and X. Li. 2012. Forest stand biomass estimation using ALOS PALSAR data
based on lidar-derived prior knowledge in the Qilian mountain, western China.
International Journal of Remote Sensing 33(3): 710-729.
Heritage, G. and A. Large, eds. Laser Scanning for the Environmental Sciences.
Wiley-Blackwell, 2009.
Hirata, Y., N. Furuya, M. Suzuki and H. Yamamoto. 2009. Airborne laser
scanning in forest management: Individual tree identification and laser pulse
penetration in a stand with different levels of thinning. Forest Ecology and
Management 258: 752-760.
H.J. Andrews Experimental Forest LTER. http://andrewsforest.oregonstate.edu/
2011. Accessed June 2012.
Hopkinson, C., and L.E. Chasmer. 2007. Modelling canopy gap fraction from
lidar intensity. ISPRS Workshop on Laser Scanning 2007 and Silvilaser 2007,
Finland. Online (http://www.isprs.org/proceedings/XXXVI/3W52/final_papers/Hopkinson_2007.pdf). Accessed 24 November 2012.
Hopkinson, C., and L. Chasmer. 2009. Testinf lidar models of fractional cover
across multiple forest ecozones. Remote Sensing of Environment 113: 275-288.
Hosoi, F., and K. Omasa. 2009. Estimating vertical leaf area density profiles of
tree canopies using three-dimensional portable lidar imaging. Online
(http://park.itc.u-tokyo.ac.jp/joho/Omasa/462.pdf). Accessed 24 November 2012.
132
Houldcroft, C.J., C.L. Campbell, I.J. Davenport, R.J. Gurney, and N. Holden.
2005. Measurement of canopy geometry characterisitics using lidar laser
altimetry: a feasibility study. IEEE Transactions on Geoscience and Remote
Sensing 43(10): 2270-2282.
Hyde, P., R. Dubayah, B. Peterson, J.B. Blair, M. Hofton, C. Hunsaker, R. Knox,
and W. Walker. 2005. Mapping forest structure for wildlife habitat analysis using
waveform lidar: validation of montane ecosystems. Remote Sensing of
Environment 96: 427-437.
Hyde, P., R. Dubayah, W. Walker, J.B. Blair, M. Hofton, and C. Hunsacker.
2006. Mapping forest structure for wildlife habitat analysis using multi-sensor
(lidar, sar/insar, etm+, quickbird) synergy. Remote Sensing of Environment 102:
63-73.
Jensen, J.L.R., K.S. Humes, L.A. Vierling, and A.T. Hudak. 2008. Discrete return
lidar-based prediction of leaf area index in two conifer forests. Remote Sensing of
Environment 112: 3947-3957.
Jensen, J.L.R, K.S. Humes, A.T. Hudak, L.A. Vierling, and E. Delmelle. 2011.
Evaluation of the MODIS LAI product using independent lidar-derived LAI: a
case study in mixed conifer forest. Remote Sensing of Environment 115: 36253639.
Jonckheere, I., S. Fleck, K. Nackaerts, B. Muys, P. Coppin, M. Weiss, and F.
Baret. 2004. Review of methods for in situ leaf area index determination part I.
theories, sensors, and hemispherical photography. Agricultural and Forest
Meteorology 121: 19-35.
Junttila, V., T. Kauranne, and V. Leppänen. 2010. Estimation of forest stand
parameters from airborne laser scanning using calibrated plot databases. Forest
Science 56(3): 257-270.
Kao, D.L., M.G. Kramer, A.L. Love, J.L. Dungan and A.T. Pain. (2005)
Visualizing distributions from multi-return LiDAR data to understand forest
structure. The Cartographic Journal 42(1): 35-47.
Kopytko, N., J.B. Cushing, L. Zeman, N. Stevenson-Molnar, F. Martin, and E.S.
Keeley. "Making ecology research results useful for resource management: a case
study in visual analytics." In Proceedings of the 10th Annual International
Conference on Digital Government Research: Social Networks: Making
Connections between Citizens, Data and Government, pp. 206-215. Digital
Government Society of North America, 2009.
133
Korhonen, L., I. Korpela, J. Heiskanen, and M. Maltamo. 2011. Airborne
discrete-return lidar data in the estimation of vertical canopy cover, angular
canopy closure and leaf area index. Remote Sensing of Environment 115: 10651080.
Korpela, I. B. Dahlin, H. Schäfer, E. Bruun, F. Haapaniemi, J. Honkasalo, S.
Ilvesniemi, V. Kuutti, M. Linkosalmi, J. Mustonen, M. Salo, O. Suomi, and H.
Virtanen. 2007. Single tree forest inventory using lidar and aerial images for 3D
treetop positioning, species recognition, height and crown width estimation.
IAPRES Volume XXXVI, Part 3/W 52.
Korpela, I.S. 2008. Mapping of understory lichens with airborne discrete-return
lidar data. Remote Sensing of Environment 112: 3891-3897.
Krisp, J.M. 2004. Three-dimensional visualization of ecological barriers. Applied
Geography. 24: 23-34.
Lang, A.R.G. and Y. Xiang. 1986. Estimation of leaf area index from
transmission of direct sunlight in discontinuous canopies. Agricultural and Forest
Meteorology 37: 229-243.
Lauenroth, W.K., D.L. Urban, D.P. Coffin, W.J. Parton, H.H. Shugart, T.B.
Kirchner and T.M. Smith. 1993. Modeling vegetation structure- ecosystem
process interactions across sites and ecosystems. Ecological Modeling 67: 49-80.
Leblanc, S.G., J.M. Chen, R. Fernandes, D.W. Deering, and A. Conley. 2005.
Methodology comparison for canopy structure parameters extraction from digital
hemispherical photography in boreal forests. Agricultural and Forest Meteorology
129: 187-207.
Lee, A.C., and R.M. Lucas. 2007. A lidar-derived canopy density model for tree
stem and crown mapping in Australian forests. Remote Sensing of Environment
111: 493-518.
Lefsky, M.A., W.B. Cohen, S.A. Acker, G.G. Parker, T.A. Spies, and D. Harding.
1999. Lidar remote sensing of the canopy structure and biophysical properties of
douglas-fir western hemlock forests. Remote Sensing of Environment 70: 339361.
Lefsky, M.A., W.B. Cohen, D.J. Harding, G.G. Parker, S.A. Acker, and S.T.
Gower. 2001. Lidar remote sensing of aboveground biomass in three biomes.
International Archives of Photogrammetry and Remote Sensing XXXIV-3/W4:
22-24.
134
Lefsky, M.A., W.B. Cohen, G.G. Parker, and D.J. Harding. 2002. Lidar remote
sensing for ecosystem studies. BioScience 52(1): 19-30.
Lefsky, M.A. 2005. Geographic variability in lidar predictions of forest stand
structure in the Pacific Northwest. Remote Sensing of Environment 95(4): 532548.
Li, W., Q. Guo, M.K. Jakubowski, and M. Kelly. 2012. A new method for
segmenting individual trees from lidar point cloud. Photogrammetric Engineering
& Remote Sensing 78(1): 75-84.
Lim, K., P. Treitz, K. Baldwin, I. Morrison and J. Green. 2003. Lidar remote
sensing of biophysical properties of tolerant northern hardwood forests. Canadian
Journal of Remote Sensing 29(5): 658-678.
Lovell, J.L., D.L.B Jupp, D.S. Culvenor, and N.C. Coops. 2003. Using airborne
and ground-based ranging lidar to measure canopy structure in Australian forests.
Canadian Journal of Remote Sensing 29(5): 607-622.
Lutz, J.A. 2005. The contribution of mortality to early coniferous forest
development. Seattle, WA: University of Washington. 95 p. M.S. thesis.
Macfarlane, C., M. Hoffman, D. Eamus, N. Kerp, S. Higginson, R. McMurtrie,
and M. Adams. 2007. Estimation of leaf area index in eucalypt forest using digital
photography. Agricultural and Forest Meteorology 143: 176-188.
Makela, A. 1986. Implications of the pipe model theory on dry matter partitioning
and height growth in trees. Journal of Theoretical Biology 123: 103-120.
Magnussen, S., and P. Boudewyn. 1998. Derivations of stand heights from
airborne laser scanner data with canopy-based quantile estimators. Canadian
Journal of Forest Research 28: 1016-1031.
Marcell, W., M. Eriksson, and S. Popescu. 2009. Systematic sampling of scanning
lidar swaths. Online
(ftp://www.iscc.org/people/vanaardt/Silvilaser/.../S5.../Marcell_64.pdf). Accessed
24 November 2012.
McGaughey, R. 1998. Techniques for visualizing the appearance of forestry
operations. Journal of Forestry 96(6): 9-14.
McGaughey, R.J., AND W.W. Carson. 2003. Fusing LIDAR data, photographs,
and other data using 2D and 3D visualization techniques. P. 16–24 in Proc. of
Terrain Data: Applications and Visualization-Making the Connection,
135
Charleston, SC. American Society for Photogrammetry and Remote Sensing,
Bethesda, MD.
McGaughey, R.J. Fusion/LDV: Software for LiDAR Data Analysis and
Visualization; USDA Forest Service, Pacific Northwest Research Station:
Portland, OR, USA, 2012. Available online:
http://forsys.cfr.washington.edu/JFSP06/lidar_&_ifsar_tools.htm Accessed:
January 2012.
Martens, S.N., S.L. Ustin, and R.A. Rouseau. 1993. Estimation of tree canopy leaf
area index by gap fraction analysis. Forest Ecology and Management 61: 91-108.
Means, J.E., S.A. Acker, D.J. Harding, J.B. Blair, M.A. Lefsky, W.B. Cohen,
M.E. Harmon, and W.A. McKee. 1999. Use of large-footprint scanning airborne
lidar to estimate forest stand characteristics in the Western Cascades of Oregon.
Remote Sensing of Environment 67: 298-308.
Montes, F., P. Pita, A. Rubio, I. Canellas. 2007. Leaf area index estimation in
mountain even-aged Pinus silvestris L. stands from hemispherical photographs.
Agricultural and Forest Meteorology 145: 215-228.
Morrison, H., C. Hopkinson, L. Chasmer, and N. Kljun. 2011. Generating an
automated approach to optimize effective leaf area index by Canadian boreal
forest species using airborne lidar. Silvilaser 11th international conference on
lidar applications for assessing forest ecosystems Hobart, Australia. Online
(http://www.iufro.org/download/file/8239/5065/40205-silvilaser2011_pdf).
Accessed 24 November 2012.
Morrison, H., C. Hopkinson, and M.A. Wulder. 2011. Optimal lidar gridding
parameterization for effective leaf area estimation in the boreal forest Yukon
Territory, Canada. SilviLaser 2011, Hobart, Australia. Online
(www.cabdirect.org/abstracts/20123180564.html). Accessed 24 November 2012.
Morsdorf, F., B. Kötz, E. Meier, K.I. Itten, and B. Allgöwer. 2006. Estimation of
LAI and fractional cover from small footprint airborne laser scanning data based
on gap fraction. Remote Sensing of Environment 104: 50-61.
Moskal, L.M., T. Erdody, A. Kato, J. Richardson, G. Zheng, and D. Briggs. 2009.
Lidar applications in precision forestry. Silvilaser College Station, TX.
Negron Juarez, R.I., H. Ribeiro da Rocha, A.M. Silva e Figueira, M.L. Goulden,
and S.D. Miller. 2009. An improved estimate of leaf area index based on the
histogram analysis of hemispherical photographs. Agricultural and Forest
Meteorology 149: 920-928.
136
Norman, D. 2012. The detection of forest structures in the Monongahela National
Forest using lidar. Thesis. Marshall University. 53 pages.
Omasa, K., F. Hosoi, and A. Konishi. 2007. 3D lidar imaging for detecting and
understanding plant response and canopy structure. Journal of Experimental
Botany 58(4): 881-898.
Paar, P. 2006. Landscape visualizations: Applications and requirements of 3D
visualization software for environmental planning. Computers, Environment and
Urban Systems 30: 815-839.
Peduzzi, A., R.H. Wynne, T.R. Fox, R.F. Nelson, and V.A. Thomas. 2012.
Estimating leaf area index in intensively managed pine plantations using airborne
laser scanner data. Forest Ecology and Management 270: 54-65.
Pekin, B., and C. Macfarlane. 2009. Measurement of crown cover and leaf area
index using digital cover photography and its application to remote sensing.
Remote Sensing 1: 1298-1320.
Peterson, F. 2012. Post-harvest establishment influences ANPP, Soil C, and DOC
export in complex terrain. Doctoral Dissertation, Oregon State University,
Corvallis, OR.
Peterson, F. and K. Lajtha. 2012. Soil C and DOC dynamics in complex terrain.
Submitted to JGR-Biogeosciences, December 2012.
Peterson, K.S. 2010. Calculating leaf area index on watershed one and
relationships with other indices. Unpublished internal document. Oregon State
University, Corvallis, OR.
Popescu. S.C. and R. Wynne. 2004. Seeing the trees in the forest: Using LiDAR
and multispectral data fusion with local filtering and variable window size for
estimating tree height. Photogrammetric Engineering and Remote Sensing 70(5):
589-604.
Popescu, S.C., and K. Zhao. 2008. A voxel-based lidar method for estimating
crown base height for deciduous and pine trees. Remote Sensing of Environment
112: 767-781.
Popescu, S.C., K. Zhao, A. Neuenschwander, and C. Lin. 2011. Satellite lidar vs.
small footprint airborne lidar: comparing the accuracy of aboveground biomass
estimates and forest structure metrics at footprint level. Remote Sensing of
Environment 115: 2786-2797.
137
Popescu, S.C. TreeVaW: Tree Variable Window; Spatial Sciences Laboratory,
Texas A&M University: College Station, TX, USA. Available online:
http://ssl.tamu.edu/personnel/s_popescu/TreeVaW/download.htm Accessed 7th
July 7 2012
Poulos, H.M., A.E. Camp, R.G. Gatewood, and L. Loomis. 2007. A hierarchical
approach for scaling forest inventory and fuels data from local to landscape scales
in the Davis mountains, Texas, USA. Forest Ecology and Management 244: 1-15.
Pypker, T.G., M.H. Unsworth, A.C. Mix, W. Rugh, T. Ocheltree, K. Alstad and
B.J. Bond. 2007. Using nocturnal cold air drainage flow to monitor ecosystem
processes in complex terrain. Ecological Applications 17: 702-714.
R Development Core Team (2012). R: A language and environment for statistical
computing, reference index version 2.15.12. R Foundation for Statistical
Computing, Vienna, Austria. ISBN 3-900051-07-0, URL http://www.Rproject.org. Accessed July 2012.
Riano, D., J. Sanchez-Pena, M. Patricio, F. Valladares, J. Greenberg and S.L.
Ustin. 2006. A simple model to calculate leaf area index from lidar data.
American Geophysical Union, Fall Meeting.
http://adsabs.harvard.edu/abs/2006AGUFM.B41A0160R Accesed: 24th
November 2012.
Riano, D., F. Valladares, S. Condes, and E. Chuvieco. 2004. Estimation of leaf
area index and covered ground from airborne laser scanner (lidar) in two
contrasting forests. Agricultural and Forest Meteorology short communication.
Richardson, J.J. 2008. Estimating leaf area index from multiple return, small
footprint aerial lidar at the Washington park arboretum. Thesis. College of Forest
Resources, University of Washington, Seattle, WA. M.S. Thesis.
Richardson, J.J., L.M. Moskal, and S. Kim. 2008. Estimating urban forest leaf
area index (LAI) from aerial lidar. Factsheet #5. Remote Sensing and Geospatial
Application Laboratory, University of Washington, Seattle, WA. Online
(http://dept.washington.edu/rsgal/). Accessed 24 November 2012.
Richardson, J.J., L.M. Moskal, and S. Kim. 2009. Modeling approaches to
estimate effective leaf area index from aerial discrete-return lidar. Agricultural
and Forest Meteorology 149: 1152-1160.
Richardson, A.D., D.B. Dail, and D.Y. Hollinger. 2011. Leaf area index
uncertainty estimates for model-data fusion applications. Agricultural and Forest
Meteorology 151: 1287-1292.
138
Richardson, J.J., and L.M. Moskal. 2011. Strengths and limitations of assessing
forest density and spatial configuration with aerial lidar. Remote Sensing of
Environment 115: 2640-2651.
Roth, B.E., K. Clint Slatton, and M.J. Cohen. 2007. On the potential for highresolution lidar to improve rainfall interception estimates in forest ecosystems.
Frontiers of Ecology and the Environment 5(8): 421-428.
Rothacher, J., C.T. Dyrness, and R.L. Fredriksen. 1967. Hydrologic and related
characteristics of three small watersheds in the Oregon Cascades. Portland, OR:
U.S. Department of Agriculture, Forest Service, Pacific Northwest Forest and
Range Experiment Station. 54 p.
Ryu, Y., T. Nilson, H. Kobayashi, O. Sonnentage, B.E. Law, and D.D. Baldocchi.
2010. On the correct estimation of effective leaf area index: does it reveal
information on clumping effects? Agricultural and Forest Meteorology 150: 463472.
Schleppi, P., M. Conedera, I. Sedivy, and A. Thimonier. 2007. Correcting nonlinearity and slope effects in the estimation of the leaf area index of forests from
hemispherical photographs. Agricultural and Forest Meteorology 144: 236-242.
Sea, W.B., P. Choler, J. Beringer, R.A. Weinmann, L.B. Hutley, and R. Leuning.
2011. Documenting improvement in leaf area index estimates from MODIS using
hemispherical photos from Australian savannas. Agricultural and Forest
Meteorology 151: 1453-1461.
Seidel, D., S. Fleck, and C. Leuschner. 2012. Analyzing forest canopies with
ground-based laser scanning: A comparison with hemispherical photography.
Agricultural and Forest Meteorology 154-155: 1-8.
Seidl, R., T.A. Spies, W. Rammer, E.A. Steel, R.J. Pabst, and K. Olsen. 2012.
Multi-scale drivers of spatial variation in old-growth forest carbon density
disentangled with lidar and an individual-based landscape model. Ecosystems. 15
pages.
Sherren, K., J. Fischer, H. Clayton, A. Hauldren, and S. Dovers. Lessons from
visualizing the landscape and habitat implications of tree decline—and its
remediation through tree planting—in Australia’s grazing landscapes. Landscape
and Urban Planning 103: 248-258.
Smith, A.M.S., M.J. Falkowski, A.T. Hudak, J.S. Evans, A.P. Robinson, and
C.M. Steele. 2009. A cross-comparison of field, spectral, and lidar estimates of
forest canopy cover. Canadian Journal of Remote Sensing 35(5): 447-459.
139
Solberg, S., E. Naesset, K. Holt Hanssen, and E. Christiansen. 2006. Mapping
defoliation during a severe insect attack on Scots pine using airborne laser
scanning. Remote Sensing of Environment 102: 364-376.
Specht, R.L and Specht, A. Australian plant communities: dynamics of structure,
growth and biodiversity. 1999.
Spies, T. 2011. LiDAR data (August 2008) for the Andrews Experimental Forest
and Willamette National Forest study areas. H.J. Andrews Experimental Forest.
Forest Science Data Bank, Corvallis, OR. [Database]
Stojanova, D., P. Panov, V. Gjorgjioski, A. Kobler, and S. Džeroski. 2010.
Estimating vegetation height and canopy cover from remotely sensed data with
machine learning. Ecological Informatics 5(4): 256-266.
Stoltman, A. M., V.C. Radeloff and D.J. Mladenoff. (2004) Forest Visualization
for management and planning in Wisconsin. Journal of Forestry 102(4) 7-13.
Strunk, J.L., S.E. Reutebuch, and J.R. Foster. 2008. Lidar inventory and
monitoring of a complex forest. ASPR Annual Conference, Portland, OR. Online
(www.asprs.org/a/publications/proceedings/portland08/0076.pdf). Accessed 24
November 2012.
Suarez, J.C., C. Ontiveros, S. Smith and S. Snape. 2005. Use of airborne LiDAR
and aerial photography in the estimation of individual tree heights in forestry.
Computers and Geosciences 31: 253-262.
Tiede, D., G. Hocheimer and T. Blaschke. 2005. A full GIS-based workflow for
tree identification and tree crown delineation using laser scanning. In CMRT05,
Vol. XXXVI, Part 3/W24. Vienna, Austria.
Turner, D.P., W.B. Cohen, R.E. Kennedy, K.S. Fassnacht, and J.M. Briggs. 1999.
Relationships between leaf area index and landsat tm spectral vegetation indices
across three temperate zone sites. Remote Sensing of Environment 70: 52-68.
United States Department of Agriculture (USDA), Forest Service. Forest
Inventory and Analysis National Program. http://www.fia.fs.fed.us/ December
2012.
United States Geological Survey (USGS). The National Map Viewer.
http://nationalmap.gov/viewer.html. December 2012 Accessed August 2012.
Van Gardingen, P.R., G.E. Jackson, S. Hernandez-Daumas, G. Russell, and L.
Sharp. 1999. Leaf area index estimates obtained for clumped canopies using
hemispherical photography. Agricultural and Forest Meteorology 94: 243-257.
140
Vose, J.M., P.M. Dougherty, J.N. Long, F.W. Smith, H.L. Gholz and P.J. Curran.
1994. Factors influencing the amount and distribution of leaf area of pine stands.
Ecological Bulletin (Copenhagen) 43: 102-114.
Walter, J.N., and E.F. Torquebiau. 2000. The computation of forest leaf area
index on slope using fish eye sensors. Life Sciences 323: 801-813.
Wang, W., B. Song, J. Chen, D. Zheng and T.R. Crow. 2006. Visualizing forest
landscapes using public data sources. Landscape and Urban Planning 75: 111124.
Waring, R.H., H.L. Gholz, C.C. Grier and M.L. Plummer. 1977. Evaluating stem
conducting tissue as an estimator of leaf area in four woody angiosperms.
Canadian Journal of Botany 55:1474-1477.
Waring, R.H. 1998. Lessons learned while extending physiological principles
from growth chambers to satellite studies. Tree Physiology. 18: 491-497.
Watershed Sciences. Lidar Remote Sensing Data Collection: HJ Andrews &
Willamette National Forest; Watershed Sciences Inc.: Corvallis, OR, USA
November 2008; 25 p.
Weiss, M., F. Baret, G.J. Smith, I. Jonckheere, and P. Coppin. 2004. Review of
methods for in situ leaf area index (LAI) determination part II. Estimation of LAI,
errors and sampling. Agricultural and Forest Meteorology 121: 37-53.
Wilson, J.W. 1960. Inclined point quadrats. New Phytol. 59: 1-8.
Wilson, J.W. 1963. Estimation of foliage denseness and foliage angle by inclined
point quadrats. Australian Journal of Botany 11: 95-105.
Zhao, K., S. Popescu, and R. Nelson. 2009. Lidar remote sensing of forest
biomass: a scale-invariant estimation approach using airborne lasers. Remote
Sensing of Environment 113: 182-196.
Zhao, K., and S. Popescu. 2009. Lidar-based mapping of leaf area index and its
use for validating GLOBCARBON satellite LAI product in a temperate forest of
the southern USA. Remote Sensing of Environment 113: 1628-1645.
Zheng, G., and L.M. Moskal. 2009. Retrieving leaf area index (LAI) using remote
sensing: theories, methods, and sensors. Sensors 9: 2719-2745.
141
APPENDIX A: LiDAR Calculated LAI, Surface Volume and Cover*Height
Value for All Permanent Vegetation Plots.
Plot
P11101
P11102
P11103
P11104
P11105
P11107
P11109
P11110
P11111
P11112
P11201
P11203
P11205
P11206
P11207
P11208
P11209
P11210
P11211
P11212
P11213
P11214
P11215
P11216
P11217
P11218
P11301
P11302
P11303
P11304
P11305
P11306
P11307
P11308
P11309
P11310
P11311
LiDAR
LAI
4.56
7.21
9.36
6.47
9.66
5.87
8.32
6.12
9.59
15.97
0.64
1.77
6.18
7.48
7.50
6.99
7.79
7.85
10.17
6.70
8.58
7.61
6.84
7.85
6.43
8.51
8.34
8.56
6.27
5.39
5.11
0.77
4.06
0.78
2.48
1.58
5.54
Volume
2851.99
3881.91
4716.75
3592.73
4832.69
3361.66
4313.06
3457.18
4804.17
7283.14
1327.74
1765.31
3481.56
3984.75
3992.86
3793.14
4104.92
4130.72
5029.63
3682.82
4412.35
4036.64
3737.21
4130.42
3579.44
4384.02
4317.79
4406.24
3513.99
3173.55
3066.43
1377.15
2656.61
1384.11
2042.33
1695.09
3231.37
Cover*Height
1304.86
1810.46
2108.04
1688.23
2129.69
1499.25
1970.07
1552.28
2191.74
3269.34
537.41
1634.25
1851.11
1709.26
1723.01
1841.66
1857.40
2321.76
1482.66
1928.78
1883.05
1684.53
1876.32
1558.30
1976.15
1930.42
2036.53
1626.86
1466.32
1449.60
538.83
1142.32
581.40
733.49
698.31
1465.79
1546.48
142
P11312
P11313
P11314
P11315
P11316
P11317
P11318
P11319
P11320
P11321
P11322
P11323
P11324
P11401
P11402
P11403
P11404
P11405
P11406
P11407
P11408
P11409
P11410
P11411
P11412
P11413
P11414
P11415
P11416
P11419
P11420
P11421
P11422
P11423
P11424
P11425
P11426
P11501
P11502
P11503
P11504
5.42
9.49
9.68
10.32
10.07
9.33
7.63
8.13
7.51
7.22
6.91
6.71
9.59
7.79
6.90
5.75
7.26
9.25
8.71
6.39
5.60
4.74
8.04
2.60
5.31
8.26
8.18
6.03
8.32
3.26
4.10
1.96
6.56
7.80
2.02
3.55
3.10
7.73
5.20
5.74
8.47
3186.17
4766.61
4840.79
5089.58
4990.19
4703.48
4044.01
4237.20
3996.70
3883.21
3764.93
3687.65
4804.72
4107.68
3759.48
3312.53
3901.58
4674.50
4464.90
3561.29
3255.33
2921.07
4203.01
2090.99
3140.85
4286.63
4257.79
3422.51
4312.89
2347.27
2672.14
1840.46
3628.07
4108.20
1864.88
2458.25
2283.98
4081.92
3100.67
3308.71
4368.29
2147.66
2150.88
2302.55
2250.21
2096.08
1867.13
1925.33
1758.08
1773.62
1712.32
1683.36
2254.25
1925.11
1777.88
1529.96
1835.90
2125.42
2039.21
1576.06
1550.84
1406.38
1916.22
932.02
1478.81
1913.67
1944.52
1471.17
1840.87
328.52
1210.87
836.08
1686.62
1923.78
879.63
1185.66
977.30
1814.15
1913.26
1496.04
1510.65
1993.34
143
P11505
P11506
P11507
P11508
P11509
P11510
P11511
P11512
P11513
P11514
P11515
P11516
P11517
P11518
P11520
P11521
P11522
P11523
P11524
P11525
P11526
P11527
P11601
P11602
P11603
P11604
P11605
P11606
P11607
P11608
P11609
P11610
P11611
P11612
P11613
P11614
P11615
P11616
P11617
P11618
P11619
3.64
5.98
9.12
8.47
7.30
6.11
6.43
8.44
7.83
8.39
7.82
2.00
1.95
4.59
5.45
0.28
2.56
1.97
7.14
5.29
5.44
6.16
6.21
5.48
5.75
5.74
6.92
5.81
8.04
8.05
7.64
7.91
9.33
8.60
9.13
9.31
8.29
9.25
7.80
8.90
6.77
2492.18
3403.64
4623.31
4369.43
3913.96
3451.69
3576.48
4359.61
4122.42
4337.04
4118.74
1858.20
1836.35
2861.19
3196.52
1189.14
2073.50
1844.66
3853.28
3133.00
3192.00
3471.17
3493.93
3209.01
3313.64
3309.35
3767.20
3337.78
4203.94
4208.02
4046.03
4153.80
4704.49
4419.24
4625.23
4697.41
4299.38
4674.68
4109.04
4538.78
3709.65
1026.51
1602.89
1846.26
1971.16
1814.08
1615.55
1558.77
1985.29
1843.08
1967.35
1889.00
835.13
875.49
1377.29
1430.11
402.75
898.35
768.83
1799.12
1467.19
1376.24
1542.90
1572.30
1543.77
1522.65
1576.71
1740.94
1526.60
1907.45
1907.70
1864.89
1915.59
2150.82
1994.95
2067.91
2135.11
1872.71
2048.82
1936.86
2073.92
1752.57
144
P11620
P11621
P11622
P11623
P11624
P11625
P11626
10.10
8.77
10.47
7.85
6.43
5.76
4.70
5002.62
4486.22
5145.32
4127.45
3576.04
3317.49
2907.34
2233.84
2025.14
2291.59
1865.50
1655.44
1557.57
1341.70
145
APPENDIX B: TreeVaW Tree Identification Results for all 133 Permanent
Vegetation Plots
Observed
Trees
Predicted
Trees
P11108
P11109
P11110
P11205
P11206
P11207
P11208
P11209
P11211
P11212
P11213
P11608
P11609
P11610
P11611
P11612
P11613
P11614
P11615
P11616
P11617
28
23
23
17
17
15
12
25
38
46
28
24
15
18
19
21
30
19
25
21
13
12
17
18
17
15
18
14
11
15
22
17
17
12
14
14
17
15
15
20
14
14
P11102
P11103
P11104
P11105
P11107
P11111
P11112
P11201
P11202
P11203
P11210
P11214
P11215
35
43
16
14
23
46
80
18
5
11
18
54
45
15
13
19
14
13
11
8
48
30
29
21
15
15
PLOT
% Trees Id'ed
Plots in which DHPs were
taken
42.86%
73.91%
78.26%
100.00%
88.24%
120.00%
116.67%
44.00%
39.47%
47.83%
60.71%
70.83%
80.00%
77.78%
73.68%
80.95%
50.00%
78.95%
80.00%
66.67%
107.69%
All other WS1 plots
42.86%
30.23%
118.75%
100.00%
56.52%
23.91%
10.00%
266.67%
600.00%
263.64%
116.67%
27.78%
33.33%
146
P11216
P11217
P11218
P11301
P11302
P11303
P11304
P11305
P11306
P11307
P11308
P11309
P11310
P11311
P11312
P11313
P11314
P11315
P11316
P11317
P11318
P11319
P11320
P11321
P11322
P11323
P11324
P11401
P11402
P11403
P11404
P11405
P11406
P11407
P11408
P11409
P11410
P11411
P11412
P11413
P11414
27
62
52
36
18
44
29
34
4
17
6
18
11
7
5
30
44
22
40
46
36
20
24
35
36
64
44
27
16
8
14
21
13
23
9
20
16
15
49
47
23
14
14
12
16
16
21
20
21
31
24
54
35
39
33
28
14
14
14
16
18
17
15
17
23
19
12
10
15
19
15
15
12
10
14
42
37
15
46
21
12
11
51.85%
22.58%
23.08%
44.44%
88.89%
47.73%
68.97%
61.76%
775.00%
141.18%
900.00%
194.44%
354.55%
471.43%
560.00%
46.67%
31.82%
63.64%
40.00%
39.13%
47.22%
75.00%
70.83%
65.71%
52.78%
18.75%
22.73%
55.56%
118.75%
187.50%
107.14%
57.14%
76.92%
60.87%
466.67%
185.00%
93.75%
306.67%
42.86%
25.53%
47.83%
147
P11415
P11416
P11417
P11418
P11419
P11420
P11421
P11422
P11423
P11424
P11425
P11426
P11427
P11501
P11502
P11503
P11504
P11505
P11506
P11507
P11508
P11509
P11510
P11511
P11512
P11513
P11514
P11515
P11516
P11517
P11518
P11519
P11520
P11521
P11522
P11523
P11524
P11525
P11526
P11527
P11601
16
15
11
14
20
28
22
16
34
28
24
13
43
30
22
37
36
12
11
18
23
18
32
8
15
24
17
23
11
3
28
4
9
8
22
21
32
16
22
70
24
14
11
68
29
29
30
46
16
14
55
31
30
26
14
20
19
14
26
18
9
12
13
19
34
16
13
16
14
34
53
25
71
29
64
39
40
14
18
23
16
21
87.50%
73.33%
618.18%
207.14%
145.00%
107.14%
209.09%
100.00%
41.18%
196.43%
129.17%
230.77%
60.47%
46.67%
90.91%
51.35%
38.89%
216.67%
163.64%
50.00%
52.17%
72.22%
59.38%
425.00%
106.67%
54.17%
94.12%
60.87%
309.09%
1766.67%
89.29%
1775.00%
322.22%
800.00%
177.27%
190.48%
43.75%
112.50%
104.55%
22.86%
87.50%
148
P11602
P11603
P11604
P11605
P11606
P11607
P11618
P11619
P11620
P11621
P11622
P11623
P11624
P11625
P11626
TOTAL
13
28
29
34
19
22
16
17
26
35
30
27
51
84
99
26
24
14
15
22
22
14
24
10
16
14
17
19
15
25
3407
200.00%
85.71%
48.28%
44.12%
115.79%
100.00%
87.50%
141.18%
38.46%
45.71%
46.67%
62.96%
37.25%
17.86%
25.25%
2810
82.48%
149