See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/237372271

A Guide to LIDAR Data Acquisition and Processing for the Forests of the Pacific
Northwest

Article · July 2008

CITATIONS READS

45 2,922

2 authors, including:

Demetrios Gatziolis
US Forest Service
63 PUBLICATIONS   690 CITATIONS   

SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Forest Service Carbon Monitoring System View project

All content following this page was uploaded by Demetrios Gatziolis on 15 January 2015.

The user has requested enhancement of the downloaded file.

United States
Department of
Agriculture
A Guide to LIDAR Data
Forest Service Acquisition and Processing
Pacific Northwest
Research Station for the Forests of the Pacific
General Technical Report
PNW-GTR-768 Northwest
July 2008
Demetrios Gatziolis and Hans-Erik Andersen
The Forest Service of the U.S. Department of Agriculture is dedicated to the principle of
multiple use management of the Nation’s forest resources for sustained yields of wood,
water, forage, wildlife, and recreation. Through forestry research, cooperation with the
States and private forest owners, and management of the National Forests and National
Grasslands, it strives—as directed by Congress—to provide increasingly greater service
to a growing Nation.
The U.S. Department of Agriculture (USDA) prohibits discrimination in all its programs and
activities on the basis of race, color, national origin, age, disability, and where applicable,
sex, marital status, familial status, parental status, religion, sexual orientation, genetic
information, political beliefs, reprisal, or because all or part of an individual’s income
is derived from any public assistance program. (Not all prohibited bases apply to all
programs.) Persons with disabilities who require alternative means for communication of
program information (Braille, large print, audiotape, etc.) should contact USDA’s TARGET
Center at (202) 720-2600 (voice and TDD). To file a complaint of discrimination, write
USDA, Director, Office of Civil Rights, 1400 Independence Avenue, SW, Washington, DC
20250-9410 or call (800) 795-3272 (voice) or (202) 720-6382 (TDD). USDA is an equal
opportunity provider and employer.

Authors
Demetrios Gatziolis is a research forester, Forestry Sciences Laboratory, 620 SW
Main, Suite 400, Portland, OR 97205; Hans-Erik Andersen is a research forester,
Forestry Sciences laboratory, 3301 C St., Suite 200, Anchorage, AK 99503.
Abstract
Gatziolis, Demetrios; Andersen, Hans-Erik. 2008. A guide to LIDAR data
acquisition and processing for the forests of the Pacific Northwest. Gen. Tech.
Rep. PNW-GTR-768. Portland, OR: U.S. Department of Agriculture, Forest
Service, Pacific Northwest Research Station. 32 p.

Light detection and ranging (LIDAR) is an emerging remote-sensing technology

with promising potential to assist in mapping, monitoring, and assessment of for-
est resources. Continuous technological advancement and substantial reductions
in data acquisition cost have enabled acquisition of laser data over entire states
and regions. These developments have triggered an explosion of interest in LIDAR
technology. Despite a growing body of peer-reviewed literature documenting the
merits of LIDAR for forest assessment, management, and planning, there seems to
be little information describing in detail the acquisition, quality assessment, and
processing of laser data for forestry applications. This report addresses this infor-
mation deficit by providing a foundational knowledge base containing answers to
the most frequently asked questions.
Keywords: LIDAR, Pacific Northwest, FIA, forest inventory, laser, absolute
and relative accuracy, precision, registration, stand penetration, DEM, canopy
surface, resolution, data storage, data quality assessment, topography, scanning.
Contents
1 Introduction
1 LIDAR Systems
2 System Specifications
7 Data Attributes
9 Data Storage
11 Data Acquisition Considerations
16 Data Quality Control
17 Return Coordinates
21 Spatial Completeness
25 Consistency of Tabular Return Attributes
29 PNW-FIA Software for Quality Assessment of LIDAR Data
31 References
 A Guide to LIDAR Data Acquisition and Processing for the Forests of the Pacific Northwest

Introduction
Light detection and ranging (LIDAR), also known as airborne laser scanning
(ALS), is an emerging remote sensing technology with promising potential to
assisting mapping, monitoring, and assessment of forest resources. Compared to LIDAR offers nearly
traditional analog or digital passive optical remote sensing, LIDAR offers tangible perfect registration of
advantages, including nearly perfect registration of spatially distributed data spatially distributed
and the ability to penetrate the vertical profile of a forest canopy and quantify its data and the ability to
structure. LIDAR has been used in many parts of the world to successfully assess penetrate the vertical
height and size of individual trees or, at the stand level, to estimate canopy closure, profile of a forest
volume, and biomass of forest stands; to assess wildlife habitat; and to quantify canopy.
stand susceptibility to fire (Andersen et al. 2005, Hinsley et al. 2006, Means et al.
2000, Naesset 2002, Persson et al. 2002, Popescu and Zhao 2007). Continuous
technological advancement and competition among vendors in the United States
have resulted in substantial reductions in data acquisition cost and have enabled
acquisition of spatially complete laser data over entire states and regions. The U.S.
Geological Survey has recently announced a plan to coordinate the acquisition of
LIDAR data at a national scale (Stoker et al. 2007). Laser scanning data are regu-
larly acquired over several national forests in Western States. These developments
have triggered an explosion of interest in LIDAR technology. Despite a growing
body of peer-reviewed literature documenting the merits of LIDAR for forest
assessment, management, and planning, there seems to be a void in information
describing issues related to the acquisition and processing of laser data. In the past
year alone, the authors have received numerous requests for guidance on the techni-
cal specifications of planned data acquisitions, on instructions on how to perform
data quality assessment, and on whether scanning data can be used to meet specific
objectives. This article addresses this information deficit by providing a founda-
tional knowledge base containing answers to the most frequently asked questions. A LIDAR system
comprises the
LIDAR Systems laser device, an

A LIDAR system operating from an airborne platform comprises a set of instru- inertial navigational

ments: the laser device; an inertial navigational measurement unit (IMU), which measurement unit, a

continuously records the aircraft’s attitude vectors (orientation); a high-precision high-precision airborne

airborne global positioning system (GPS) unit, which records the three-dimensional global positioning

position of the aircraft; and a computer interface that manages communication system, and a

among devices and data storage. The system also requires that a GPS base station computer interface.

installed at a known location on the ground and in the vicinity (within 50 km) of the
aircraft, operate simultaneously in order to differentially correct, and thus improve
the precision of, the airborne GPS data.

GENERAL TECHNICAL REPORT PNW-GTR-768

The laser device emits pulses (or beams) of light to determine the range to a distant
target. The distance to the target is determined by precisely measuring the time
delay between the emission of the pulse and the detection of the reflected (backscat-
tered) signal. In topographic mapping and forestry applications, the wavelength
of the pulses is in the near-infrared part of the spectrum, typically between 1040
and 1065 nm. There are two types of LIDAR acquisition differentiated by how
backscattered laser energy is quantified and recorded by the system’s receiver. With
waveform LIDAR, the energy reflected back to the sensor is recorded as a (nearly)
continuous signal. With discrete-return, small-footprint LIDAR, reflected energy
is quantized at amplitude intervals and is recorded at precisely referenced points
in time and space. Popular alternatives to the term “point” include “return” and
“echo.” The energy amplitude pertaining to each return is known as intensity. This
article addresses only small-footprint, discrete-return LIDAR.

System Specifications
LIDAR systems have been evolving for more than a decade, and will likely
continue to evolve even faster in the years to come. Hence, when planning data
acquisition, it is essential to obtain specifications of currently available systems.
Such specifications will determine both data acquisition costs and, quite likely, the
feasibility of projects the acquired data are expected to support. Baltsavias (1999a)
provided a good (if somewhat out-of-date) overview of the basic engineering and
geometric concepts underlying airborne laser scanning, and Baltsavias (1999b)
illustrated the variability in specifications among commercial systems. The major
operational specifications of a LIDAR system are outlined below:
• Scanning frequency is the number of pulses or beams emitted by the laser
instrument in 1 second. Older instruments emitted a few thousand pulses
per second. Modern systems support frequencies of up to 167 kHz (167,000
pulses per second). Sometimes they can be operated at lower-than-maxi-
mum frequencies, typically 100 kHz or 71 kHz, but seldom at low frequen-
cies, say, 10 kHz. The scanning frequency is directly related to the density
of discrete returns obtained. Thus a system operating at 150 kHz onboard
an aircraft flying at constant speed at a standard height above a target will
generate a much higher number of returns than when operating at 71 kHz.
Equivalently, a high-frequency system can generate desired return densities
by operating on an aircraft that flies higher and faster than an aircraft car-
rying a lower frequency system, thereby reducing flying time and acquisi-
tion costs.


 A Guide to LIDAR Data Acquisition and Processing for the Forests of the Pacific Northwest

• Scanning pattern is the spatial arrangement of pulse returns that would

be expected from a flat surface and depends on the mechanism used to
direct pulses across the flight line. Of the four scanning patterns supported
by instruments used in acquiring laser data for forestry applications, the
seesaw pattern (fig. 1a) and its stabilized equivalent (fig. 1b) are the most
common. In these two patterns, the pulse is directed across the scanning
swath by an oscillating mirror, and returns are continuously generated
in both directions of the scan. Although this configuration is designed
to preserve the spacing between returns, in practice, pulse density is not
uniform and returns tend to “bunch up” at the end of the swath because of
mirror deceleration. The nonuniform spacing of returns can be partially

Figure 1—Nadir view of theoretical scanning patterns of LIDAR instruments.


GENERAL TECHNICAL REPORT PNW-GTR-768

mitigated, but not eliminated, with the use of galvanometers. In the parallel
line pattern (fig. 1c), a rotating polygonal mirror directs pulses along
parallel lines across the swath, and data are generated in one direction of
the scan only. The elliptical pattern (fig. 1d) is generated via a rotating
mirror that revolves about an axis perpendicular to the rotation plane.
• Beam divergence. Unlike a true laser system, the trajectories of photons
in a beam emitted from a LIDAR instrument deviate slightly from the
beam propagation line (axis) and form a narrow cone rather than the thin
cylinder typical of true laser systems. The term “beam divergence” refers
to the increase in beam diameter that occurs as the distance between the
laser instrument and a plane that intersects the beam axis increases. Typical
beam divergence settings range from 0.1 to 1.0 millirad. At 0.3 millirad,
the diameter of the beam at a distance of 1000 m from the instrument is
approximately 30 cm (fig. 2). Because the total amount of pulse energy
remains constant regardless of the beam divergence, at a larger beam
divergence, the pulse energy is spread over a larger area, leading to a lower
signal-to-noise ratio.
• Scanning angle is the angle the beam axis is directed away from the
“focal” plane of the LIDAR instrument (fig. 3) It should not be confused

Figure 2—Illustration of LIDAR beam divergence. Horizontal and vertical distances are drawn in
different scales.


 A Guide to LIDAR Data Acquisition and Processing for the Forests of the Pacific Northwest

with the angle formed between the beam axis vector and a vertical plane
(nadir view), because the latter angle is affected by the attitude of the air-
craft. The maximum angle supported by most systems does not exceed 15
degrees. The angle is recorded as positive toward the starboard and nega-
tive toward the port side of the aircraft. The combination of scanning angle
and aboveground flight height determines the scanning swath (fig. 3).
• Footprint diameter is the diameter of a beam intercepted by a plane posi-
tioned perpendicularly to the beam axis at a distance from the instrument
equal to the nominal flight height (fig. 2). It is thus a function of both beam
divergence and the above-target flight height. The distribution of pulse
energy is not uniform over the extent of the footprint. It decreases radially
from the center and can be approximated by a two-dimensional Gaussian
distribution.

Figure 3—Illustration of scanning attributes of LIDAR data acquisition. Aircraft flying parallel to the ground and seesaw scanning
pattern are assumed.


GENERAL TECHNICAL REPORT PNW-GTR-768

• Pulse length is the duration of the pulse, in nanoseconds (ns). Along with
discretization settings (below), it determines the range resolution of the
pulse in multiple return systems, or the minimum distance between con-
secutive returns from a pulse.
• Number of returns (per beam/pulse) is the maximum number of individ-
ual returns that can be extracted from a single beam. Certain systems can
identify either the first or the first and last returns. Most modern systems
can identify multiple returns (e.g., up to five) from a single beam.
• Footprint spacing is the nominal distance between the centers of consecu-
tive beams along and between the scanning lines (fig. 3), which, along with
the beam divergence, determines the spatial resolution of LIDAR data. The
footprint spacing is a function of scanning frequency, the aboveground
flight height, and the velocity of the aircraft.
• Discretization settings are specifications integral to the processing of the
backscattered energy of a pulse to identify individual returns (fig. 4). They
are system-specific and proprietary, and sometimes are referred to as digi-
tization settings. They control the minimum energy amplitude necessary
to produce a return and, along with the pulse length, determine the mini-
mal distance between consecutive returns (discretization tolerance) from

Figure 4—Illustration of the discretiza-

tion process used to identify individual
returns by processing the backscattered
energy of a laser pulse.


 A Guide to LIDAR Data Acquisition and Processing for the Forests of the Pacific Northwest

the same pulse. Modern instruments can process the energy-backscatter

pertaining to a single beam and identify up to six returns, but the majority
support only up to four. The optimal settings for forestry applications likely Small-footprint LIDAR
depend on acquisition objectives and vegetation structure. data comprise a set of
return coordinates in
Data Attributes three dimensions with

Small-footprint LIDAR data comprise a set of return coordinates in three dimen- each return usually

sions with each return usually carrying attribute values that relate either to that carrying attribute

return or to the pulse from which the return was generated. values that relate either

• Pulse density is a direct function of the footprint spacing (described above) to that return or to the
2
over a hypothetical flat plane: pulse density = 1/(footprint spacing ). This is pulse from which the

the most consistent measure of the spatial resolution of a LIDAR data set. return was generated.

• Return density is the most common term used in describing a data set,
and is often confused with pulse density. It is the mean number of returns
in the data set present (in two dimensions) in a unit square area, typically
1 m2. With the exception of single-return systems, return density is con-
trolled by the specifications and operation mode of a LIDAR system and by
the target scanned. Assuming that all other specifications remain the same,
the return density generated by a four-return-per-square-meter-capable sys-
tem over a forest stand will be much higher than the density generated over
a nearby pasture (fig. 5), because in the latter case, virtually all the energy
returned falls within a single quantum (distance class). Because of this
scene-dependent variability, users should specify a minimum pulse density
for a given acquisition, instead of return density.
• Return intensity or simply intensity, is an attribute that describes the
strength of the beam backscattering pertaining to the return in question. It
depends on the reflectance properties of the target, and hence it can poten-
tially be used in target discrimination. Its utility for object classification
is often reduced because of its dependence on bidirectional reflectance
distribution function effects, the distance (range) to the laser instrument,
the total number of returns identified in the parent beam, the rank of the
return (first, second, etc.) in the parent beam, and the receiver’s gain factor.
The latter term describes the scaling of the receiver’s sensitivity designed
to prevent hardware damage in the event that it receives an extraordinarily
high amount of backscattered energy as can occur with high reflectivity
targets. Such reduction in sensor sensitivity is practically instantaneous. The
reverse scaling, an increase in sensitivity in the presence of continuously
weak energy backscattering, usually takes several seconds. The presence


GENERAL TECHNICAL REPORT PNW-GTR-768

of an isolated, single high-reflectivity target scanned in one flight line can

thus lead to substantial discrepancy in the mean intensity of returns on
the overlapping part of two adjacent flight lines. Additional, object-indepen-
dent variability in intensity values is introduced by suspected fluctuations in
the energy emitted by the laser instrument. Personal communication with
scientists involved in LIDAR research have revealed that these fluctuations
can sometimes amount to 30 percent of the mean pulse energy and that they
are likely more pronounced for high-frequency systems. Although LIDAR
instruments currently do not record gain factors and energy output levels,
persistent user requests to enable their logging could facilitate intensity nor-
malization in the future and thus improve intensity-based classification of
objects. 1 Return intensity is recorded in 8 bits (values 1 to 255), 12 bits (1 to
1023), and less often as a fraction in the 0 to 1 range or in 16 bits (1 to 65535).

Figure 5—(a) False color (near infrared, red, green) digital aerial photograph and (b) correspond-
ing gray-scale raster of LIDAR returns per square meter, with lighter tones depicting higher return
count.

1
Hyppä, J. Plenary session, 2007 International Society for Photogrammetry and Remote
Sensing Workshop, Espoo, Finland.


 A Guide to LIDAR Data Acquisition and Processing for the Forests of the Pacific Northwest

• Return number refers to the rank of a return among those generated from
one beam. It is meaningful only for systems that support multiple returns
per beam. The return number should not be confused with the number of
returns, a beam attribute.
• Attributes that a return inherits from its parent beam include the scan
angle, usually recorded in degrees; the end-of-scan-line, a binary (true/
false) attribute indicating whether the parent beam marked the edge of a
scanning line; and those sometimes assigned at the data postprocessing
phase such as indices to flight lines or classification schemes, and GPS
time, an indication of the precise time that a pulse was emitted. Provided
sufficient precision is used for storing GPS time, this attribute can be used
as a unique identifier for a pulse.

Additional information is usually organized in the form of metadata, and often

contains spatial geographic information system (GIS) layers with the spatial extent
of the data acquisition, flight lines, the date and time range, the model and charac-
teristics of the LIDAR instrument, etc.

Data Storage
The LIDAR data files are very large and can quickly fill up computer hard drives.
The need for efficient access to and storage of scan data, coupled with the absence
of a universal format standard, has led developers of LIDAR software to implement
their own, proprietary storage format, which, with few exceptions, pay little atten-
tion to enabling import/export options. Only recently a file format (LAS) endorsed
by the American Society for Photogrammetry and Remote Sensing (ASPRS) has
been gaining popularity and support. As revealed in personal communications with
several LIDAR data vendors across the United States in the last 2 years, the lack of
significant progress in format standardization has prompted data delivery requests
in ASCII (text) format in more than two-thirds of all acquisitions. Data delivered
in most of those acquisitions consisted of X, Y, and Z coordinates and intensity
only. This preference for the text format is rooted in the fact that, unlike any binary
alternative, the contents of text files are easily accessible via a text editor. Assuming
delimited format (text, space, tab, etc.) and that each file line carries data for one
return, the data can be easily imported into popular databases and subsequently
queried, merged, grouped into subsets, and rearranged as needed.
However, ASCII text is a poor format choice from the standpoint of data storage
efficiency. To illustrate this issue, consider a LIDAR data file comprising a modest
1 million returns with coordinates of two-digit (centimeter) precision (universal
transverse mercator projection) and 8- bit intensity being the only return attribute.


GENERAL TECHNICAL REPORT PNW-GTR-768

The size of this file will be approximately 32,134,000 bytes in text format and only
14,000,024 bytes in binary format (24 bytes are used to describe a transformation
of scale in return coordinates from a two-decimal real number to a long integer),
a gain in storage efficiency by a factor of 2.3. If the same file were to include all
data attributes mentioned in the previous section, its size in text format would be
approximately 64,094,000 bytes, and in LAS binary format it would be 28,000,227,
a storage efficiency gain of also 2.3. In either file configuration (intensity only vs.
all attributes), the time required for reading from or writing to the file in text format
Efficiency in accessing
would be, depending on the hardware configuration of the computer, nearly an
files is important in
order of magnitude longer than for binary format. Efficiency in accessing files is
research efforts and
important in research efforts and in applications that require files to be read mul-
in applications that
tiple times.
require files to be read
A less-evident implication of the file format is realized when considering how
multiple times.
LIDAR data are organized in individual files. A LIDAR data file would typically
contain returns either from a rectangular portion of the acquisition area, sometimes
referred to as “bin” or “tile,” or from individual flight lines (fig. 6). Compared
to files representing smaller bins, files corresponding to larger ones will have a
smaller percentage of returns near the borders of the bin, and thus introduce fewer

Figure 6—Illustration of the spatial extent of data

in LIDAR files. Shaded, single-digit-numbered
stripes correspond to individual LIDAR files with
data from one flight line. Odd- and even-numbered
stripes are flown in opposite directions. Darker
shading shows stripe overlap. Dash-outlined,
two-digit-numbered rectangles correspond to files
containing returns from multiple flight lines.

10
 A Guide to LIDAR Data Acquisition and Processing for the Forests of the Pacific Northwest

discontinuities or artifacts in data derivatives and metrics calculated along the bin
borders. Assuming an interest in minimizing border effects, maximum bin, and
therefore file, sizes should be targeted. Table 1 shows the limits in file size and
corresponding bin area imposed by a 32-bit computer operating system for various
data storage formats and return density configurations. All data from an acquisition
with a mere 100 million returns can be stored in just one binary file. If, instead, text Data acquisition
format is preferred, the data would have to be split into two or more files. Note that planning should be
switching from a 32- to a 64-bit operating system would eliminate this issue, as the based on a careful
file size supported by 64-bit operating systems is practically unlimited. evaluation of the
project objectives while
Data Acquisition Considerations considering potential
Data acquisition planning should be based on a careful evaluation of the project limitations imposed
objectives while considering potential limitations imposed by budget constraints, by budget constraints,
availability of LIDAR instruments with specific capabilities, terrain, and vegetation availability of LIDAR
structure and phenology. Often, acquisition planning is challenging, as it involves instruments with
many decisions among equally appealing or contrasting tradeoffs. The discussion specific capabilities,
below provides a synthesis of LIDAR data analysis objectives and their relation to terrain, and vegetation
system specifications and acquisition parameters. structure and
phenology.
Table 1—Attributes of a single LIDAR data file

Density (returns /m 2)
Operating File Return Number of
system format attributes returns 1 4 8
- - - - Bin area (ha) - - - -
32-bit Text X, Y, Z, intensity 133,658,000 13,366 3,341 1,671
Binary X, Y, Z, intensity 306,783,000 30,678 7,670 3,835
Text All 67,010,000 6,701 1,675 838
Binary All 153,392,000 15,339 3,835 1,917
64-bit Any Any Practically unlimited
Note: Text format assumes universal transverse mercator coordinates with 2-digit precision and 8-bit intensity.

Quantification of forest structure and assessment of tree height and volume via
LIDAR data is typically performed either at the individual tree or plot/stand level.
There is general agreement among researchers that the identification of individual
trees requires a minimal return density of approximately four returns per square
meter. This density often implicitly assumes systems that support multiple returns
per pulse. High-scanning-frequency systems can achieve this density when using
aircraft that fly high and fast to reduce acquisition costs. However, two data sets
with equal return density acquired over the same area by instruments operating
at different scanning frequencies can have very different return distributions in

11
GENERAL TECHNICAL REPORT PNW-GTR-768

three dimensions. This is in part because the energy carried by a single pulse in
high-scanning-frequency systems, and therefore its ability to penetrate vegetation,
is much lower than the energy of a pulse in a slower system. The high-scanning-
frequency system should be expected to generate proportionally more returns from
the upper part of the canopy. Conversely, the low-scanning-frequency system will
likely have a higher proportion of returns from the understory or the ground. In
forest stands with tall and very dense vegetation, common in the Pacific Northwest
(PNW), it is likely that a lower frequency system could generate more ground
returns than a faster system, even where the overall density of the faster system
is much greater than the density generated by the slower system. Although these
assumptions have not been tested formally, they are indirectly supported by the
fact that the proportion of ground-to-total returns generated by low-frequency laser
systems mounted on low-flying aircraft or helicopters over high-density tropical
forests (Clark et al. 2004) is much higher than the one achieved over comparably
dense PNW forests scanned by high-frequency systems flying approximately 1000
meters above terrain (Gatziolis 2007). Hence, where a precise and accurate descrip-
tion of the ground surface under dense vegetation is important, the option of using a
lower scanning frequency setting should be seriously considered.
Although the three-dimensional distribution of returns is also affected by
the discretization process, the absence of specific information on the settings of
alternative systems usually precludes a meaningful evaluation of comparative
advantages offered by each system. Fine sensitivity in pulse discretization, that is to
set the minimum that the amplitude of the backscattered pulse energy would need
to exceed for a return to be identified to a low value (fig. 4), will tend to produce
returns closer to the top of the canopy and support a more detailed description of
vegetation surfaces, including leader stems typical of many conifer species. Fine
sensitivity, though, is associated with lower positional precision of returns from
lower vegetation strata. Fine distance tolerances between consecutive returns from
a given beam would tend to produce more returns from the upper layers of tall,
dense, and healthy vegetation at the expense of fewer returns from the ground.
Coarse distance tolerances would prevent ground returns where the magnitude of
the tolerance exceeds the mean vegetation height.
As stated in the “Data Attributes” section, the local return density would differ
among vegetation types and structures. To avoid misunderstandings, data acquisi-
tion requests should specifically mention the minimum pulse density (not return
density) that is acceptable over a particular forest or land type. Data vendors with
experience in local acquisitions will likely be able to assess the flight height above

12
 A Guide to LIDAR Data Acquisition and Processing for the Forests of the Pacific Northwest

ground and aircraft speed for which their laser instrument can meet the requested
pulse density over a forest type of interest.
To determine the preferred beam divergence setting (wide vs. narrow), one
should consider how this setting affects the interaction of the beam with vegetation.
In wide divergence, the canopy volume illuminated (or sampled) by the beam is
larger than in narrow divergence. The ratio of canopy volume “sampled” by each
setting is actually the square of the divergence ratio (table 2). Because wide diver-
gence affords a more comprehensive coverage of the canopy, sampling plot-
or stand-based metrics computed from wide divergence LIDAR data would likely
be more robust and exhibit lower variance than those computed using narrow
divergence data. Because, as stated previously, in wide divergence the cross-
section pulse energy is spread over a larger area, the reduced photon density leads
to a lower signal-to-noise ratio in backscattered energy. This reduces the three-
dimensional precision of returns, and often causes backscattering from leader stems
in coniferous species or from the ground in dense stands to be of an amplitude too
weak to be identified as returns during the pulse discretization process. Recent
research in the Pacific Northwest has demonstrated that the accuracy of LIDAR-
based individual conifer tree height measurements obtained at a narrow beam
divergence (0.33 m) setting are significantly more accurate than those obtained at a
wide divergence setting (0.8 m) (Andersen et al. 2006).

Table 2—Vegetation volume illuminated by two LIDAR beam divergence settingsa

Beam cross section area at distance (m)

from LIDAR instrument
Vegetation
970 985 1000 volume
- - - - - - - Square meters - - - - - - - Cubic meters
Wide divergence (0.7 millirad) 0.362 0.373 0.385 11.204
Narrow divergence (0.3 millirad) 0.067 0.069 0.071 2.058
Ratio 5.444b
a
Flight height assumed to be 1000 m above vegetation, and vegetation is 30 m tall.
b
Result (5.444) = (0.7/0.3)2.

Before discussing how scanning angles affect the fidelity of LIDAR data, it is
necessary to describe a phenomenon known as “path reflectance” or “multipath-
ing,” a term borrowed from optical remote sensing and GPS technologies. Energy
propagating through a medium is subject to attenuation, or reduction in both den-
sity and amplitude owing to scattering, absorption, and reflectance. The attenuation
of near-infrared pulses, although minimal in the atmosphere, can be substantial in
porous and heterogeneous media, such as the forest. Of interest here, is the change

13
GENERAL TECHNICAL REPORT PNW-GTR-768

in propagation (or path) direction that sometimes occurs when the beam hits the
surface of objects with considerable mass, such as tree branches, trunks, rocks, or
the ground. How often a change of direction occurs is believed to correlate to the
angle of incidence and, to a lesser extent, to the distance the pulse travels through
the canopy. As mentioned earlier, the LIDAR instrument records the pulse direction
vectors and the time difference between pulse emission and return. The identifica-
tion of individual returns performed by processing the backscattered energy is
based on the premise that the pulse has traveled along a straight line, an assumption
that is true for the majority of the pulses. For pulses with one or more changes in
direction, the actual path remains unknown. Returns from these pulses are recorded
as originating further away from the LIDAR instrument than their true, albeit
unknown, locations (fig. 7).
As the direction of the beam deviates progressively more from nadir, the beam’s
angle of incidence upon the ground increases. This causes the ground to behave less
as a diffuse and more like a directional reflector, thereby facilitating path reflec-
tance. The distance the pulse travels through the canopy increases too. In acquisi-
tions on a slope, both of these measures can increase even further. For example,

Figure 7—Illustration of LIDAR path reflectance. Black

line represents the beam propagation anticipated by the
LIDAR instrument. Red line represents the actual beam
propagation. Letters A, B, and C indicate the locus of
anticipated return, first, and second beam inflection points,
respectively.

14
 A Guide to LIDAR Data Acquisition and Processing for the Forests of the Pacific Northwest

the incidence angle to the ground of a beam emitted with a scanning angle of 12
degrees down a 100-percent slope will be 57 degrees. The distance traveled by the
beam through 30-m-tall vegetation in such terrain would be 38.95 m, an increase of
nearly 30 percent compared to the distance the beam would travel through the same
canopy at zero degrees scanning angle. Therefore, the canopy penetration rates for
laser pulses in a forest area are directly related to the off-nadir scanning angles of
the laser pulses.
Experience has determined that LIDAR data artifacts attributed to multipath-
ing, such as returns located well below the ground, proliferate when scanning
angles exceed 12 to 14 degrees over dense forest stands. For that reason, some data
vendors tend to remove from delivered data sets returns from beams at large scan-
ning angles. Others are reluctant to eliminate those returns, especially where the
Decisions on
resulting return density nears the minimum specified. Maximum absolute scanning
acquisition timing
angles of 12 degrees for flat or moderate-slope areas and 10 degrees for areas with
should be based on the
steeper slopes should produce scan data with minimal scan-angle-related artifacts.
seasonal progression
Decisions on acquisition timing should be based on the seasonal progression
and phenological
and phenological stage of vegetation while considering local weather patterns and
stage of vegetation
terrain. In the absence of disturbances, seasonal progression determines the devel-
while considering local
opment stage and density of foliage and, hence, canopy penetrability by LIDAR
weather patterns and
pulses for deciduous cover types. In practice, flight timing has little effect on
terrain.
coniferous canopies unless there is a deciduous understory. Where extraction of a
high-fidelity digital elevation model (DEM) is one of the products the laser acquisi-
tion is expected to provide, it is advisable to consider a leaf-off acquisition even if
deciduous species are only present as shrubs or brush. Note that often the extraction
of a DEM, if not of importance by itself, is an intermediate LIDAR data analysis
step necessary for the assessment of many stand or tree parameters, including
canopy/crown cover, tree height, and canopy base height.
Seasonal progression can be essential in research efforts designed to achieve
near-simultaneous collection of laser and field data. Sometimes it would be
necessary to “grow” the field data forward or backward in time to ensure a close
temporal match with the laser data. Seasonality also affects the values of indices or
output of established models that assess vegetation parameters by assuming, often
implicitly, a certain vertical vegetation structure that is seasonally dependent. Many
of the models used to estimate stand volume, basal area, and mean height, and even
indices of fuel accumulation and stand susceptibility to fire, rely on the proportion
of total returns present in selected quantiles or percentiles of vegetation height.
Foliage development as the growing season progresses would tend to increase the
percentage of returns at higher height percentiles, thereby altering model or index

15
GENERAL TECHNICAL REPORT PNW-GTR-768

values. When uncertainty exists about whether assumptions related to foliage

maturity are embedded in models expected to be used with the data, acquisitions
Unlike most forms of toward the end of the growing season would be preferred.
optical remote sensing, Unlike most forms of optical remote sensing, acquisition of LIDAR data is
acquisition of LIDAR practically independent of solar illumination and could, in fact, be done at night
data is practically when lower windspeeds interfere less with aircraft handling. Wet, rainy days should
independing of solar not be considered because infrared light does not penetrate water vapor. Persistent,
illumination and visibility-limiting fog or frequent rainfall may not permit acquisitions during
could, in fact, be done certain, predetermined periods. Flying the aircraft at constant aboveground height
at night when lower so as to maintain uniform return density, a task easily (and safely) accomplished
windspeeds interfere over flat areas, can be challenging over undulating or mountainous terrain, espe-
less with aircraft cially during windy or foggy days. These considerations indicate that, for example,
handling. LIDAR data acquisitions during leaf-off conditions in the coastal PNW region
might be delayed if the acquisition window happens to coincide with a period
of prolonged low-ceiling fog or continuous rain, conditions typical of the region
during the winter months. Similarly, the summer window for obtaining scan data
at high-elevation, snow-free conditions can shift and be of different length between
years. Note also that as many LIDAR data vendors share aircraft and instruments, it
is more convenient for them to plan a mission within a larger time window. There-
fore, planning large time windows for acquisition will likely provide flexibility with
logistical and weather-related limitations, but it may also hinder efforts to coordi-
nate the flights with field data collection. Perhaps the best approach to optimizing
the timing for laser data acquisition is to have, where possible, the data vendor be
awarded the acquisition contract well ahead of the anticipated flight day(s). This
allows ample time for the vendor and user to communicate concerning the number
of flight days, season progression, and vegetation conditions (tree leaf-out, senes-
Quality control for cence, etc.), and ultimately optimize the acquisition timing.
a LIDAR data set
involves an evaluation Data Quality Control
of return coordinate
Quality control for a LIDAR data set from a user’s perspective involves an evalu-
accuracy and
ation of return coordinate accuracy and precision, compliance with acquisition
precision, compliance
specifications, and data spatial consistency and completeness. Ideally, a report with
with acquisition
quantitative estimates of these data quality measures would be part of every laser
specifications,
data delivery. Unfortunately, such reports are rare, and when they are produced,
and data spatial
the information included is frequently selective or incomplete, in part because, as
consistency and
discussed below, the procedures involved in evaluating data quality are costly and
completeness.
time consuming, especially for data over mountainous forested areas.

16
 A Guide to LIDAR Data Acquisition and Processing for the Forests of the Pacific Northwest

Return Coordinates
Although the positional accuracy and precision of LIDAR data far exceeds those
afforded by traditional remotely sensed imagery, laser return coordinates do
contain random and systematic errors. Both error types originate in one or more
of the laser system’s components. Random errors relate to noise in the computed
location of the aircraft, in the recording of aircraft attitude and scanning angles, or
in the recording of time between pulse emission and backscatter reception, which
ultimately determines the distance (range) to the target. The magnitude of random
errors can be calculated during system calibration. Because LIDAR systems get
“out-of-tune” fairly quickly, periodic calibration is necessary. Assuming that the
magnitude and distribution of errors from each laser system component is known,
error propagation techniques can be used to estimate the nominal coordinate error
and determine if it is acceptable for an application. It should be noted that the error
estimates obtained via propagation techniques refer to returns from hard surfaces
on flat terrain, conditions uncommon in forests of the Pacific Northwest. Hence,
even in an optimally and recently calibrated system, random errors in derived
return coordinates over forested landscapes can be far from negligible. Random
errors are known to affect the absolute accuracy of return coordinates, but, perhaps
against commonly held expectations, they also affect their relative accuracy. The
latter term describes the integrity in the spatial arrangement of neighboring returns.
As an example, the same amount of noise in the measurement of the aircraft’s roll,
one of the attitude vectors, will cause errors along the edges of the scanning swath Biases in GPS, aircraft
to be larger than those in the middle, thereby degrading the relative accuracy of the attitude, scanning
return data. angle, and time
Biases in GPS, aircraft attitude, scanning angle, and time measurements cause measurements cause
systematic errors. If erroneously measured, the three-dimensional offset between systematic errors.
the onboard GPS unit and the pulse emission point, would cause errors that are
independent of the above-terrain flight height, but that do depend on the flying
direction. Errors in measurement of the scan angle can cause return-coordinate
errors that increase with flying height and flight direction. Where the above-terrain
flight height changes continuously and in the presence of slopes, the magnitude and
type of error embedded in return coordinates is spatially variable.
As is evident from the brief discussion above, the amount of complexity in
the assessment of the error budget in a laser data set is substantial. Experience
suggests that the use of error propagation techniques and system calibration data
alone tend to underestimate coordinate error in forested landscapes. More accurate
assessments of the error budget require field surveying of the ground and of objects
that are clearly discernible in visualizations of the return cloud, or in derivatives

17
GENERAL TECHNICAL REPORT PNW-GTR-768

of the laser data. In mountainous areas or forests of heterogeneous structure, more

surveying could be necessary. Whether the benefits from an accurate assessment
of coordinate error warrant the expense to obtain it should be determined by the
project objectives. The decision should consider that often the fidelity of data
products actually depends on the relative accuracy of return coordinates, even in
the presence of sizable, absolute errors.

Relative accuracy of LIDAR return coordinates—

The assessment of relative accuracy is possible only between data obtained in
overlapping swaths. It has the advantage that it can be performed exclusively by
data postprocessing in the office. Generally, the higher the swath overlap (over a
minimum of 20 percent), the more precise the assessment would be. There are two
approaches in computing relative accuracy, with different degrees of complex-
ity. The first approach relies on computing raster, or triangular irregular network
(TIN), surfaces of elevation or intensity for each (or part) of the two flight lines and
then evaluating their spatial correspondence (Gruen and Akca 2005, Maas 2002,
Okatani and Deguchi 2002). Often the evaluation is based on calculating the raster
difference. A mean value other than zero or a skewed distribution for values of the
difference raster is indicative of bias in the measurement of range. The presence
of clusters of cells with values consistently positive or negative and of shape that
resembles dominant landscape objects or structures indicate bias in the recording
of aircraft location. Figure 8 illustrates the latter case showing the highest return
digital surface models computed for the area common to two overlapping scanning
swaths (8c and 8e) and their difference surface (8d). The 0.013-m mean value of the
difference raster is reduced to 0.004 m when considering only returns from hard,
impermeable surfaces (rooftops, paved road). This level of range error between
scanning swaths is consistent with the magnitude anticipated by error propagation
for range measurements. Unlike the high precision in range measurements, the
difference raster in figure 8d shows substantial bias in two-dimensional (horizontal)
space. The bias is evident in the vicinity of human-made structures. The differ-
ence raster has negative values in the southwest corner of a building (blue arrow),
positive values on the opposite corner (red arrow), and nearly zero in between. The
amount and direction of the horizontal discrepancy between the two surface rasters
can be assessed approximately with measurements performed on a computer screen
or, alternatively, by using scripts that compute the offset at which the normalized
correlation of the surfaces is maximized. A more precise assessment can be accom-
plished by digitizing the boundaries of structures and then computing the rotation,
translation, and scale adjustments required for aligning the boundary vector pairs in
space. Likely the most precise assessment of positional accuracy among all methods

18
 A Guide to LIDAR Data Acquisition and Processing for the Forests of the Pacific Northwest

Figure 8—(a) Perspective view of return cloud colored by elevation, (b) false-color digital orthophotograph, (c, e) 1-m highest return
rasters computed from the point clouds of two overlapping flight lines (darker tones denote higher elevation), and (d) difference raster
computed by subtracting raster in (e) from raster in (c).

that rely on intermediate raster data derivatives can be achieved by comparing the
two surfaces in the frequency domain (e.g., via Fourier transforms).
Raster-based evaluations of positional error in laser data have a shortcoming
that is rooted in the form of the return data. Unlike typical remote sensing prod-
ucts, return clouds are abstract and cannot be easily compared to discrete objects.
Consider, for instance, the edge of one of the roofs visible in figure 8a or 8b and
the footprint of a pulse that hits that edge. Assuming negligible random error in
the data, we would expect a return right on the edge of a roof if exactly half of the
pulse’s footprint intersected the roof plane, or, in other words, if the center of the
footprint would coincide with the edge line. This seldom is the case. Instead, there
would be many returns with elevation value equal to that at the edge of the roof,
some of them positioned (in two dimensions) over the roof plane and the others hov-
ering in midair next to the structure. Higher return densities, narrower beam diver-
gence settings (smaller footprints), longer linear features of objects, and a higher

19
GENERAL TECHNICAL REPORT PNW-GTR-768

frequency of objects of a given size promotes more precise feature extraction for
an object of a given size. Because the positional uncertainty of objects delineated
on elevation raster surfaces is proportional to the area of a raster cell, determining
Regardless of the exact the optimal resolution for the raster is of importance. Raster resolutions resulting in
methodology used, the an average of four to six returns per cell are regarded as appropriate for evaluating,
assessment of relative and ultimately improving, the relative accuracy of a laser data set.
accuracy in LIDAR Regardless of the exact methodology used, the assessment of relative accuracy
data is susceptible in LIDAR data is susceptible to artifacts or errors introduced during the creation of
to artifacts or errors the derived products, such as raster surfaces or object (i.e., individual tree) segmen-
introduced during tations. Often the algorithms used to produce these derivatives are complex and
the creation of the highly sensitive to the input parameters. These postprocessing errors sometimes
derived products, such could generate the impression that the data are of poor relative accuracy when
as raster surfaces or in reality they are not. To avoid such pitfalls, sufficient experience and adequate
object segmentations. understanding of the processes involved in generation and quality assessment
of derived products is essential. Fortunately, the magnitude of the errors can be
deduced with precision that increases proportionally to the number of objects
identified and the area each occupies. However, prematurely increasing the area
over which a relative accuracy evaluation is performed could lead to erroneous or
biased results because, sooner or later, either the system’s acquisition parameters
will cease being stationary or the spatial distribution of objects will change.
The second approach used for assessing the relative accuracy of a laser data
set requires computation of a surface for only one of the two overlapping scan-
ning swaths. The other remains in its original return cloud form. Unlike the first
approach, the highest-return surface is now represented as a triangulated irregular
network. Spatial correspondence is evaluated by examining how well the network
triangles fit to the return cloud (Lee et al. 2005). The advantage of this approach
is that (a) it does not require decisions on the appropriate resolution of computed
surface rasters; (b) it is found to work well with continuous canopies, sparse forest,
or in the presence of human-made objects; and (c) it can be used over small areas.
The drawback to the point-based methods is that there is some speculation on
how to determine which returns should participate in the evaluation of fit for each
network triangle. As this approach is an active research project, refinements in its
application should be anticipated.

Absolute accuracy of return coordinates—

The evaluation of relative positional accuracy is usually followed by an evaluation
of absolute accuracy using information provided by a high-precision field survey.
In most acquisitions familiar to the authors in the PNW, the ground survey, if
performed at all, is a single transect along a road free from overhanging vegetation.

20
 A Guide to LIDAR Data Acquisition and Processing for the Forests of the Pacific Northwest

Where the area was inaccessible to vehicles, the surveyed transect was outside the
acquisition boundary. The absolute accuracy report for those acquisitions was based
on a comparison of transect coordinates with those of co-located (in two dimen-
sions) or nearly co-located returns. Given that horizontal discrepancies between
two point domains or coordinate systems cannot always be detected or measured
along a transect, linear surveys can support evaluations of only absolute elevation
accuracy. Discrepancies owing to a shift of the return cloud parallel to a plane
that approximates the local ground surface will be missed. A three-dimensional
evaluation would require intersecting transects or, alternatively, placing targets
with characteristic shape and reflectance that clearly distinguish them from their
surroundings within the acquisition area. Both of these options are costly and often
practically difficult to implement. Note that absolute accuracy reports based on a
single survey may not be representative of the acquisition area. Precise, kinematic-
GPS-based, surveys of bare ground performed on multiple intersecting transects in
a study area in coastal Oregon showed that although the relative accuracy between
overlapping scanning swaths was often nearly constant throughout the acquisition
area, the absolute accuracy varied substantially. In a handful of locations, it was The relative accuracy
almost an order of magnitude lower (worse) than the one mentioned in the report of LIDAR data is more
submitted with the laser data. The lower accuracy was attributed to aircraft location important than the
bias introduced when the aircraft was engaged in rapid ascending or descending. absolute accuracy
The relative accuracy of LIDAR data is more important than the absolute accu- in most forest
racy in most forest applications, including estimation of density, structure, basal applications.
area, volume, etc. Given the cost of quality control, investments in evaluation of
and measures to promote the relative accuracy will likely be more beneficial to the
data user than those targeting the absolute accuracy. This general rule does not nec-
essarily apply to data acquisitions for topographic mapping, research purposes, or
studies investigating forest growth and change with time. Nevertheless, the reader
should be aware that laser data with even suboptimal return coordinate accuracy
registration will be far superior, in terms of registration and internal consistency, to
any other form of remotely sensed data.

Spatial Completeness
The implications from lack of continuity or scanning uniformity in a LIDAR
data acquisition differ. They can range from simply being sources of artifacts and
local variability in data derivatives to precluding data analysis. Even sporadic
discontinuities in the laser data could, for example, prevent a successful delineation
of the drainage network or the computation of landform and vegetation structure
metrics. Figure 9 shows a case from a laser acquisition, where, in the interest of

21
GENERAL TECHNICAL REPORT PNW-GTR-768

minimizing costs, the sidelap of adjacent scanning swaths was set to only 10 per-
cent. Thanks to changes in aircraft attitude, a noticeable portion of the acquisition
area was not scanned.
The absence of scanning uniformity across the acquisition area is demonstrated
by variability in pulse density or return density over the same or similar objects and
vegetation. It occurs where the distance between adjacent flight lines or the flight
height above the ground does not remain constant (fig. 10), or, as shown above (fig.
9), in the presence of pronounced instability in the aircraft attitude vectors. Local
fluctuations in return density theoretically can be prevented over flat or undulat-
ing terrain if the percentage of sidelap between adjacent scanning swaths can be
represented by the formula N × 100/(N+1), where N is an integer greater than 0. The
value of the formula for N = 1 is 50 percent, the theoretical minimum sidelap that
produces scanning uniformity. For N = 3, the amount of sidelap would be 75 per-
cent. Larger N values produce unnecessarily high sidelap amounts. If the amount
of scanning swath sidelap specified does not comply with the formula, scanning
uniformity cannot be achieved. For instance, sidelap of 60 percent corresponds to
N = 1.5, which is not an integer. In that case, half of the acquisition area will be

Figure 9—(a) Gray-scale panchromatic orthophotograph, and (b) raster of laser pulse density. Color lines denote the spatial extent of
laser scanning swaths.

22
 A Guide to LIDAR Data Acquisition and Processing for the Forests of the Pacific Northwest

represented in two scanning swaths and the other half in three scanning swaths.
At sidelap of 30 percent, 57 percent of the acquisition area will be represented in
a single scanning swath and the remaining 43 percent in two scanning swaths. In
practice, constant return density even over homogeneous vegetation is difficult to
achieve, and some variability should be anticipated and tolerated.
Interesting insights on how aircraft attitude affects the laser scanning can be
obtained by examining figure 10. It can be seen in the figure that a few flight lines
almost coincide, whereas the others are regularly spaced. This is because persistent
atmospheric turbulence during data acquisition that translated into nearly constant
changes in aircraft pitch and roll had caused the scanner to miss many areas,
despite the intended 50-percent swath sidelap. To the credit of the data vendor, the
lack of spatial completeness was identified and certain lines were flown a second
time. Although the additional flight lines covered all the areas initially missed, the
minimum pulse density (2 pulses/m2) requested in the acquisition specifications
was not met everywhere. Areas scanned in swaths with excessive meandering,
denoted by “I” in figure 10b, were more susceptible to not meeting the pulse density
standard. Areas represented in many swaths had pulse density several times higher
than the one requested.
Variability in aircraft pitch also affects the local pulse density, which when
mapped, exhibits a striping effect oriented perpendicularly to the flight line. When
the aircraft flies horizontally (zero pitch), the forward propagation of successive

Figure 10—(a) False-color, high-resolution orthophotograph showing direction-colored flight lines of laser data acquisition mission, (b)
raster of laser pulse density, and (c) raster of laser return density. Black lines on the rasters denote the spatial extent of individual scan-
ning swaths. Areas identified with roman numerals demonstrate the effects of substantial instability in aircraft roll (I) and pitch (II).

23
GENERAL TECHNICAL REPORT PNW-GTR-768

scanning lines and the distance between them is constant, although some minimal
variability could occur on steep ridges and valleys. A sudden reduction in pitch
(aircraft dives) slows the forward propagation of the scanning lines and, if large
enough, it can even reverse it momentarily. Similarly, a swift increase in pitch (air-
craft ascent) accelerates the forward propagation of scanning lines and increases the
distance between them. The result is sporadic, higher-than-average pulse density
areas (figure 10c, II) usually followed, along the flight line, by below-average den-
sity areas. The combined effect of roll and pitch changes on the local return density
for a single flight line is shown in figure 11. The area depicted is covered with
continuous forest over rolling terrain and the overall laser return density is 4.8/m 2,
higher than the 4 returns/m2 anticipated. In this example, concentration of pulses in
portions of the area owing to aircraft attitude dropped the return density well below
the specification standard for approximately 60 percent of the area shown in the
figure, with the reduction in density being more pronounced along the edges of the
scanning swath. In the absence of pertinent documentation, the effects of locally
variable pulse density of laser data derivatives can only be speculated. These effects
are likely limited to areas with low pulse density and could perhaps be associated
with an increase in the variance of biophysical parameter estimates computed at the
individual tree level.

Figure 11—Perspective view of (a) a false-color, high-resolution orthophotograph of a forested area with a flight line and corresponding
laser scanning swath superimposed, and (b) co-located raster of return density derived from the swath’s laser data.

24
 A Guide to LIDAR Data Acquisition and Processing for the Forests of the Pacific Northwest

Another dimension in spatial completeness refers to the portion of returns that are
removed by postprocessing prior to the data delivery. The majority of returns that
are eliminated correspond to pulses with scanning angle exceeding the maximum
specified, or those believed to be affected by multipathing and thus positioned
below the true ground surface. Although the elimination of returns in the latter case
is theoretically legitimate, the procedure used to identify them can contain flaws,
at least in certain circumstances. It was found, for example, that proprietary algo-
rithms used to classify points in the return cloud as above, on, or below the ground
do not work particularly well in dense, coniferous forests in the PNW coastal region
(Gatziolis 2007). Similar problems have been identified in the classification of
returns acquired at leaf-on conditions over dense, multistory, deciduous stands. In
those conditions, many returns eliminated as being below the ground because of
multipathing, when compared with precise and accurate representations of terrain
obtained by alternate means, were found to be actually above the ground. This topic
is an active research field and new, better performing algorithms are introduced
frequently. The legitimacy of returns flagged by the algorithm employed by a
data vendor as belowground and eliminated from the delivered data set cannot be
assessed by an alternative, perhaps superior algorithm, at a later time. Data users
with interest in exploring alternative algorithms and data analysis techniques are
advised to request from the vendors that all data initially flagged for removal from Intensity is the most
the delivered data set be organized in a second set to accompany the first. commonly requested
and used return
Consistency of Tabular Return Attributes
attribute.
Intensity is the most commonly requested and used return attribute. Often it is the
only attribute delivered with the return coordinates. It is typically expressed as an
integer, but sometimes is archived as a floating point number in the 0 to 1 range.
The latter preference can introduce data consistency problems if the precision of the
float numbers is not adequate or the scaling of the backscattered energy in the 0 to 1
range is data-tile or flight-line specific. Figure 12 demonstrates one such case show-
ing the return intensity histograms from the overlapping portion of two adjacent
flight lines, with the intensity values scaled linearly in the 0 to 1 range. Evidently,
the shape of the histograms is the same, but the second histogram appears com-
pressed. This is because the scaling of the intensity values was based on the maxi-
mum per flight line, with the maximum being much higher in one flight line than
in the other. In the example of figure 12, the intensity values for the two flight lines
can be adjusted via histogram matching. If, however, a delivered laser data set with
inconsistently (flight-line-specific) scaled intensity is organized in tiles and without
information on the membership of returns or pulses to individual flight lines, then

25
GENERAL TECHNICAL REPORT PNW-GTR-768

Figure 12—Distribution of intensity for returns in two overlapping scanning swaths.

Figure 13—Histogram of 8-bit return intensity from 1 km 2 of forested land with homogeneous
structure showing a few intensity values with abnormally high frequency.

26
 A Guide to LIDAR Data Acquisition and Processing for the Forests of the Pacific Northwest

there is little the analyst can do to restore the intensity data to a consistent scale and
improve their utility.
A second issue with intensity data consistency, albeit of minimal implications
on their utility, relates to the circumstantial reduction in the precision of intensity
value recorded. It has been observed in nearly every data set examined by the
authors. It does not seem to be associated with data acquisition conditions, the
phenological stage of the vegetation, or the laser instrument used, and occurs only
at higher intensity levels. This issue can perhaps be better understood by examining
figure 13, which shows a histogram of return intensity values for 1 km2 of structur-
ally homogeneous, uneven-aged coniferous forest. Certain intensity values in the
histogram, all pertaining to rounded numbers and of consistent increment between
them (100, 120, …, 180), have frequency much higher than their neighboring values.
About half of the returns in the histogram bins with higher-than-anticipated fre-
quency were preceded chronologically by returns with very low intensity. One pos-
sible explanation for this phenomenon is an inability of the laser sensor to precisely
quantify one or very few backscatter events among many others of consistently
low energy. Another explanation is suboptimally performed discretization of the
backscattered energy.
Inadequate precision has more serious implications in the recording of GPS
time, a quantity that indicates the moment a pulse was emitted. The GPS time of a
pulse or return can be used to determine the distance in three dimensions between
the return and the antenna of the laser instrument, to establish the chronological
succession of pulses and returns, or determine return membership in individual
pulses. In older, lower frequency systems (< 10 kHz) four digits of precision in
the recording of GPS time were adequate. For modern, high-frequency systems
(> 100 kHz), at least six digits of precision are needed. Sometimes, an upgrade to
a higher frequency system is not met by a proper modification in the precision of
recorded GPS time. The effect of this omission is shown in table 3, which contains
data from a laser acquisition performed by using a 71-kHz instrument. Recording
GPS time with adequate precision (5 digits) allows us to deduce that the 17 returns
belong to 13 different pulses (1001 to 1013), with four pulses represented by two
returns each. Using inadequate precision (4 digits) would have indicated that the 17
returns belong to only two pulses (1001 and 1002) with each pulse generating 8 and
9 returns, respectively, a deduction that is absurd given that the system’s configura-
tion supported a maximum of four returns per pulse. Note that determining the
precision of GPS time is easy when the data are stored in ASCII format. When
examining binary data, it should be ensured that the data viewer or conversion-to-
text utility employed supports the precision level anticipated. Otherwise lack of
precision could be a software-introduced artifact.
27
GENERAL TECHNICAL REPORT PNW-GTR-768

Table 3—Laser return data showing effect of global positioning system (GPS) time precision
in identifying return membership to unique pulses via data postprocessing
5-digit precision 4-digit precision
Intensity Derived Derived
X Y Z (8-bit) GPS time pulse ID GPS time pulse ID
804.60 359.55 219.12 76 37614.24936 1001 37614.2494 1001
804.13 359.99 222.21 18 37614.24938 1002 37614.2494 1001
803.90 359.58 219.23 56 37614.24938 1002 37614.2494 1001
803.23 359.66 219.72 52 37614.24939 1003 37614.2494 1001
802.55 359.71 219.99 61 37614.24940 1004 37614.2494 1001
801.88 359.79 220.43 57 37614.24942 1005 37614.2494 1001
801.14 359.74 219.94 64 37614.24943 1006 37614.2494 1001
800.42 359.73 219.83 75 37614.24944 1007 37614.2494 1001
800.02 360.26 223.60 12 37614.24946 1008 37614.2495 1002
799.69 359.71 219.54 43 37614.24946 1008 37614.2495 1002
799.26 360.19 222.94 4 37614.24947 1009 37614.2495 1002
798.96 359.69 219.28 49 37614.24947 1009 37614.2495 1002
798.48 360.08 222.06 7 37614.24949 1010 37614.2495 1002
798.25 359.70 219.30 45 37614.24949 1010 37614.2495 1002
797.55 359.72 219.32 61 37614.24950 1111 37614.2495 1002
796.84 359.74 219.37 63 37614.24951 1112 37614.2495 1002
796.13 359.75 219.31 56 37614.24953 1113 37614.2495 1002
Note: Data were from 71-kHz instrument and stored in ASCII format.

28
 A Guide to LIDAR Data Acquisition and Processing for the Forests of the Pacific Northwest

PNW-FIA Software for Quality Assessment of LIDAR

Data
Most commercial software packages with LIDAR data analysis capabilities are The PNW FIA Program
costly and offer few or no options for data quality assessment. To address this has developed a
limitation, PNW Forest Inventory and Analysis (FIA) Program has developed a collection of script
collection of script executables compiled from code written in C programming executables compiled
language that enable numerous data quality assessment and analysis operations. from code written
Although in the current development stage the scripts can be executed only from a in C programming
command line, they will be, prior to their public release, also organized in a graphi- language that enable
cal user interface (GUI). numerous data quality
During script development, emphasis was given to optimizations that expedite assessment and
execution, efficient use of computer hardware resources (computer memory), and analysis operations.
support for both 32- and 64-bit operating systems and Windows or Unix/Linux
platforms. The optimizations have helped realize script execution speeds that are
3 to 10 times faster than those of equivalent, commercially available, software.
Many of the scripts make implicit, transparent to the user, calls to R-routines, which
are provided along with the executables. R (R Development Core Team 2007) is
a widely utilized, well-documented, open-source, and freely available software
package that is particularly popular among researchers in a variety of disciplines.
Porting C script output directly to R institutes flexibility in data quality assessment
efforts and subsequent analyses, simplifies otherwise complex analysis tasks, and
enables informative graphical representation of script output. In Windows plat-
forms, a number of the scripts have equivalent versions in the Fusion software suite,
which is available from the Forest Service Remote Sensing Applications Center
Web site (http://www.fs.fed.us/eng/rsac/fusion/). Fusion also contains an impressive
and informative laser data visualization interface.
The scripts require that the input laser data are in the LAS format described
in the “Data Storage” section. To ensure portability, routines converting popular
ASCII formats to LAS are provided. Typical script output includes contingency
matrices (pivot tables), histograms and distribution moments of all return attri-
butes present in an LAS file, and custom-resolution rasters of return attributes in
GridASCII format or as georeferenced portable network graphic (PNG) files. Tasks
specific to quality assessment of laser data range from examining LAS fields for
content validity, to evaluating the consistency of distance intervals between returns
originating from the same pulse, to identifying and mapping the spatial distribution
of pulses with missing (filtered) returns. A set of scripts enables subsetting an LAS
file by using one or more attribute value ranges, or by two- and three-dimensional
spatial constraints explicated via a bounding box, a circle, a binary image or raster

29
GENERAL TECHNICAL REPORT PNW-GTR-768

functioning as a mask, or via a shapefile. Where the computational efficiency is not

a concern, spatial and tabular constraints can alternatively be combined simultane-
ously into a complex, structured query language (SQL) query. Table 4 provides a
brief description of the scripts and their functionality. A detailed version containing
command line syntax, examples, data requirement, and the algorithm(s) employed
in each script will be available in an upcoming publication.

Table 4­—Functionality description of the Pacific Northwest Forest Inventory and Analysis LIDAR utilities
Type Script name Description
Portability ASC2LAS Converts laser data from text to binary LAS format.
LAS2ASC Converts laser data from binary LAS to text format.
Rasterization LAS2Raster Generates rasters of distribution moments for each return/pulse attribute in
GridASCII or georeferenced portable network graphic (PNG) format.
Statistics LAS_Stats Generates histograms and ASCII tabular statistics for each return attribute
present in an LAS file.
Aggregation LAS_Merge Combines multiple LAS files into one.
Subsetting Subset_by_attribute Extracts returns from an LAS file within a range of values for a single return
attribute and saves them in a new LAS file.
Subset_w_raster Eliminates returns using a raster mask in GridASCII format.
Subset_w_BMP Eliminates returns using a bitmap mask.
Subset_w_Shapefile Distributes returns to separate LAS file according to attributes of a polygon
shapefile.
Subset_w_SQL Extracts returns from an LAS file that meet complex tabular and spatial
criteria provided as a structured query language (SQL) query.
Quality ID_complete_pulse Splits an LAS into two new files, one containing data from pulses with
assessment missing returns, and the other containing the remaining (nonfiltered) returns.
MinDist_bw_returns Computes the minimum distance in three dimensions of returns that originate
from a single pulse and have consecutive return numbers.
Update_Header Queries the contents of an LAS file and updates the information present in the
file header if necessary.

30
 A Guide to LIDAR Data Acquisition and Processing for the Forests of the Pacific Northwest

References
Andersen, H.-E.; McGaughey, R.J.; Reutebuch, S.E. 2005. Estimating canopy
fuel parameters using airborne LIDAR data. Remote Sensing of Environment.
94: 441-449.

Andersen, H.-E.; Reutebuch, S.E.; McGaughey, R.J. 2006. A rigorous

assessment of tree height measurements obtained using airborne LIDAR and
conventional field methods. Canadian Journal of Remote Sensing. 32(5): 355-366.

Baltsavias, E.P. 1999a. Airborne laser scanning: basic relations and formulas.
ISPRS Journal of Photogrammetry and Remote Sensing. 54: 199-214.

Baltsavias, E.P. 1999b. Airborne laser scanning: existing system and firms and
other resources. ISPRS Journal of Photogrammetry and Remote Sensing. 54:
164-198.

Clark, M.L.; Clark, D.B.; Roberts, D.A. 2004. Small-footprint LIDAR estimation
of sub-canopy elevation and tree height in a tropical rain forest landscape.
Remote Sensing of Environment. 91: 68-89.

Gatziolis, D. 2007. LIDAR-derived site index in the the U.S. Pacific Northwest–
challenges and opportunities. Espoo, Finland: International Archives of
Photogrammetry Remote Sensing and Spatial Information Sciences. 36 (Part
3/W52): 136-143.

Gruen, A.; Akca, D. 2005. Least squares 3D surface and curve matching. ISPRS
Journal of Photogrammetry and Remote Sensing. 59: 151-174.

Hinsley, S.A.; Hill, R.A.; Bellamy, P.E.; Baltzer, H. 2006. The application of
LIDAR in woodland bird ecology: climate, canopy structure and habitat quality.
Photogrammetric Engineering and Remote Sensing. 72: 1399-1406.

Lee, J.; Yu, K.; Kim, Y.; Habib, A. 2005. Segmentation and extraction of linear
features for detecting discrepancies between LIDAR strips. IEEE International
Geoscience and Remote Sensing Symposium. Los Alamitos, CA: Institute of
Electrical and Electronic Engineers. 7: 4954-4957.

Maas, H.-G. 2002. Methods for measuring height and planimetry discrepancies in
airborne laserscanner data. Photogrammetric Engineering and Remote Sensing.
68(9): 933-940.

Means, J.; Acker, S.; Fitt, B.; Renslow, M.; Emerson, L.; Hendrix, C.
2000. Predicting forest stand characteristics with airborne scanning LIDAR.
Photogrammetric Engineering and Remote Sensing. 66: 1367-1371.

31
GENERAL TECHNICAL REPORT PNW-GTR-768

Naesset, E. 2002. Predicting forest stand characteristics with airborne laser using
a practical two-stage procedure and field data. Remote Sensing of Environment.
80: 88-99.

Okatani, I.S.; Deguchi, K. 2002. A method for fine registration of multiple view
range images considering the measurement error properties. Computer Vision
and Image Understanding. 87: 66–77.

Persson, A.; Holmgren, J.; Sodermann, U. 2002. Detecting and measuring

individual trees using an airborne laser scanner. Photogrammetric Engineering
and Remote Sensing. 68: 925-932.

Popescu, S.; Zhao, K. 2007. A voxel-based LIDAR method for estimating crown
base height for deciduous and pine trees. Remote Sensing of Environment.
doi:10.1016/j.rse.2007.06.011.

R Development Core Team. 2007. R: a language and environment for statistical

computing. Vienna, Austria: R Foundation for Statistical Computing. http://
www.R-project.org. (May 29, 2008).

Stoker, J.; Parrish, J.; Gisclair, D.; Harding, D.; Haugerud, R.; Flood, M.;
Andersen, H.E.; Schuckman, K.; Maune, D.; Rooney, P.; Waters, K.; Habib,
A.; Wiggins, E.; Ellingson, B.; Jones, B.; Nechero, S.; Nayegandhi, A.;
Saultz, T.; Lee, G. 2007. Report of the first national LIDAR Initiative meeting.
Open-File Report OF2007-1189. U.S. Department of the Interior, U.S. Geological
Survey. http://www.lidar.cr.usgs.gov/presentations/NLImeetingReport-final.pdf.
(June 2, 2008).

32
This Page Left Blank Intentionally
This Page Left Blank Intentionally
 Pacific Northwest Research Station
Web site http://www.fs.fed.us/pnw/
Telephone (503) 808-2592
Publication requests (503) 808-2138
FAX (503) 808-2130
E-mail [email protected]
Mailing address Publications Distribution
Pacific Northwest Research Station
P.O. Box 3890
Portland, OR 97208-3890
 U.S. Department of Agriculture
Pacific Northwest Research Station
333 SW First Avenue
P.O. Box 3890
Portland, OR 97208-3890
Official Business
Penalty for Private Use, $300

View publication stats