Use of LIDAR for Forest Inventory and

Forest Management Application

Birgit Peterson1, Ralph Dubayah2, Peter Hyde3, Michelle characteristics have been identified that help quantify these
Hofton4, J. Bryan Blair5, and JoAnn Fites-Kaufman6 fuel loads: canopy bulk density (CBD) and canopy base height
(CBH). These have been adopted for fire behavior modeling
Abstract.—A significant impediment to forest (Sando and Wick 1972, Scott and Reinhardt 2001). CBD is
managers has been the difficulty in obtaining large- defined as the mass of available canopy fuel per unit canopy
area forest structure and fuel characteristics at useful volume and CBH is the lowest height in the canopy where there
resolutions and accuracies. This paper demonstrates is sufficient fuel to propagate fire vertically into the canopy
how LIDAR data were used to predict canopy bulk (Scott and Reinhardt 2001).
density (CBD) and canopy base height (CBH) for an
area in the Sierra National Forest. The LIDAR data This article provides a brief, simple description of the
were used to generate maps of canopy fuels for input different types of LIDAR systems and how they work and
into a fire behavior model (FARSITE). The results summarizes previous research utilizing LIDAR for landsurface
indicate that LIDAR metrics are significant predictors characterization. It also examines the use of large-footprint,
of both CBD (r2 = 0.71) and CBH (r2 = 0.59). In sum- waveform-digitizing LIDAR data to predict and create maps of
mary, LIDAR is no longer an experimental technique CBD and CBH as well as the use of LIDAR-derived products
and has become accepted as a source of accurate and to run a fire behavior model. LIDAR metrics are compared
dependable data that are suitable for forest inventory to field-based estimates of CBD and CBH and, based on the
and assessment. regression models resulting from these comparisons, maps of
CBD and CBH are generated that are then tested as inputs into
a fire behavior model.
Introduction

In this article we present an overview of the use of LIDAR LIDAR

for forest inventory and canopy structure mapping and
explore the efficacy of a large-footprint, waveform-digitizing LIDAR (frequently used synonymously with the term laser
LIDAR for the estimation of canopy fuels for utilization in altimetry) provides a direct and elegant means to measure the
fire behavior simulation models. Because of its ability to structure of vegetation canopies (Dubayah and Drake 2000).
measure the vertical structure of forest canopies, LIDAR is LIDAR is an active remote sensing technique in which a
uniquely suited among remote sensing instruments to observe pulse of light is sent to the Earth’s surface from an airborne or
canopy structure characteristics, including those relevant spaceborne laser. The pulse reflects off of canopy materials such
to fuels characterization, and may help address the relative as leaves and branches. The returned energy is collected back
lack of spatially explicit fuels data. Two canopy structure at the instrument by a telescope. The time taken for the pulse

1
U.S. Department of Agriculture (USDA), Forest Service, Rocky Mountain Research Station; currently at U.S. Geological Survey Center for Earth Resources
Observation and Science, Sioux Falls, SD 57198. E-mail: [email protected].
2
Department of Geography, 1150 LeFrak Hall, University of Maryland, College Park, MD 20742. E-mail: [email protected].
3
Department of Geography, 1150 LeFrak Hall, University of Maryland, College Park, MD 20742. E-mail: [email protected].
4
Department of Geography, 1150 LeFrak Hall, University of Maryland, College Park, MD 20742. E-mail: [email protected].
5
National Aeronautics and Space Administration, Goddard Space Flight Center, Building 32, Greenbelt, MD 20771. E-mail: [email protected].
6
USDA Forest Service, Tahoe National Forest, 631 Coyote Street, Nevada City, CA 95959. E-mail: [email protected].

2005 Proceedings of the Seventh Annual Forest Inventory and Analysis Symposium 193
to travel from the instrument, reflect off of the surface, and be Subcanopy topography, canopy height, basal area, canopy
collected at the telescope is recorded. From this ranging infor- cover, and biomass have all been successfully derived from
mation various structure metrics can be calculated, inferred, large-footprint LIDAR waveform data in a variety of forest
or modeled (Dubayah and Drake 2000). A variety of LIDAR types (Drake et al. 2002a, Dubayah and Drake 2000, Hofton
systems have been used to measure vegetation characteristics. et al. 2002, Hyde et al. 2005, Lefsky et al. 1999, Means et al.
Most of these are small-footprint, high pulse rate, first- or 1999, Peterson 2000). For example, results from Hofton et al.
last-return-only airborne systems that fly at low altitudes. (2002) show that large-footprint LIDAR measured subcanopy
Other, experimental LIDAR systems are large footprint and full topography in a dense, wet tropical rainforest with an accuracy
waveform digitizing and provide greater vertical detail about better than that of the best operational digital elevation models
the vegetation canopy. Dubayah et al. (2000) and Lefsky et (such as U.S. Geological Survey 30-m DEM products). Means
al. (2002) provide thorough overviews of use of LIDAR for et al. (1999) used large-footprint LIDAR to recover mean
landsurface characterization and forest studies. stand height (r2 = 0.95) for conifer stands of various ages in
the Western Cascades of Oregon. Drake et al. (2002a) found
Canopy height, basal area, timber volume and biomass have that metrics from a large-footprint LIDAR system were able
all been successfully derived from LIDAR data (Drake et al. to model plot-level biomass (r2 = 0.93) for a wet tropical
2002a, Drake et al. 2002b, Hyde et al. 2005, Lefsky et al. 1999, rainforest. Dubayah et al. (2000), Dubayah and Drake (2000),
Maclean and Krabill 1986, Magnussen and Boudewyn 1998, and Lefsky (2002) provide a thorough overview of forest
Means et al. 1999, Naesset 1997, Nelson et al. 1984, Nelson et structure derived using large-footprint LIDAR. In sum, LIDAR
al. 1988, Nelson 1997, Nilsson 1996, Peterson 2000). Many of is a proven method for deriving many characteristics relevant
these studies rely on small-footprint systems. Small-footprint to forest management. LIDAR data have also been used to
LIDARs have the advantage of providing very detailed mea- measure canopy structure relevant to fire behavior modeling
surements of the canopy top topography. Most small-footprint
(5-cm to 1-m diameters) systems are low flying and have a high
Figure 1.—Illustrations showing sample waveforms for
sampling frequency (1,000 to 10,000 Hz). Although small-foot-
different cover types in the Sierra Nevada. (a) Waveform return
print systems typically do not digitize the return waveforms, from bare ground—no canopy return. (b) Waveform return
the high frequency sampling produces a dense coverage of the for a short, dense forest stand. The canopy return blends in
overflown area. This can provide a very detailed view of the with the ground return. (c) Waveform return for a tall, dense
forest stand. The waveform shows layering in the canopy and
vegetation canopy topography; however, the internal structure
the ground return is clearly defined. (d) Waveform return for a
of the canopy is difficult to reconstruct because data from the tall, sparse forest stand. The waveform shows a distinct upper
canopy interior are sparse (Dubayah et al. 2000). canopy layer and a layer of low-lying vegetation that mixes in
with the ground return. The stand diagrams were created with
the Stand Visualization System based on field measurements.
Recently, LIDARs have been developed that are optimized
for the measurement of vegetation (Blair et al. 1994, Blair et
(a) (b)
al. 1999). These systems have larger footprints (5- to 25-m
diameters) and are fully waveform digitizing, meaning that the
complete reflected laser pulse return is collected by the system. Bare ground

LIDAR remote sensing using waveform digitization records Waveform amplitude Waveform amplitude

the vertical distribution of surface areas between the canopy

(c) (d)
top and the ground. For any particular height in the canopy, the
waveform denotes the amount of energy (i.e., the amplitude of
the waveform) returned for that layer (Dubayah et al. 2000).
Waveform amplitude Waveform amplitude
The amplitude is related to the volume and density of canopy
material located at that height (fig. 1).
194 2005 Proceedings of the Seventh Annual Forest Inventory and Analysis Symposium
(Andersen et al. 2005, Morsdorf et al. 2004, Riaño et al. 2003, Service and privately owned lands. The topography varies
Riaño et al. 2004, Seielstad and Queen 2003) and this specific considerably with some areas characterized by very steep slopes
application is explored in the remainder of this paper. and an elevation range between approximately 850 and 2,700 m.

The LIDAR data used in this study were collected by the Laser
Study Site and Data Collection Vegetation Imaging Sensor (LVIS) (Blair et al. 1999). LVIS is
a large-footprint LIDAR system optimized to measure canopy
The study area is located in the Sierra National Forest in the structure characteristics. LVIS mapped a 25- by 6-km area of
Sierra Nevada mountains of California near Fresno (fig. 2) and the Sierra National Forest in October of 1999 in a series of
covers a wide range of vegetation types (e.g., fir, pine, mixed flight tracks (fig. 2). Flying onboard a NASA C-130 at 8 km
conifer, mixed hardwood/conifer, meadow), canopy cover, and above ground level and operating at 320 Hz, LVIS produced
elevation. Common species of the region include red fir (Abies thousands of 25-m diameter footprints at the surface.
magnifica A. Murr.), white fir (Abies concolor (God. & Glend.)
Hildebr.), ponderosa pine (Pinus ponderosa Dougl. ex Laws.), Field data were collected in the summers of 2000–02 in the
Jeffrey pine (Pinus jeffreyi Grev. & Balf.), and incense cedar Sierra National Forest. Circular plots were centered on LIDAR
(Libocedrus decurrens Torr.), among others. Canopy cover footprints and measured 15 m in radius. The 15-m radius was
can range from completely open in meadows or ridge tops to chosen to ensure complete overlap with the LVIS footprint and
very dense, especially in fir stands. The study area extends to account for trees located beyond the 12.5-m radius of the
over nearly 18,000 ha of U.S. Department of Agriculture Forest footprint with crowns overhanging the footprint. Within these
plots all trees over 10-cm diameter at breast height (d.b.h.)
Figure 2.— Schematic showing the location of the study
(diameter at breast height) were sampled. Measurements
site, plot distribution, and footprint-centered plot design. (a)
Locator map of the study area in the Sierra Nevada, northeast included tree height, height to partial crown, partial crown
of Fresno. (b) The study area was delimited by swaths of LVIS wedge angle, height to full crown, four crown radius
data covering the region. The combined area of the swaths measurements, and distance and azimuth relative to the plot
is approximately 25 by 6 km. (c) The individual plots were
center. Tree crown shape and species were also recorded.
colocated with the LVIS footprints. Each circular plot (15-m
radius) is centered on an LVIS footprint with its own waveform.

(b) Derivation of CBD and CBH

(a)

The data from the 135 plots were used to calculate field-based
CBD according to an inventory-based method. The original
methodology was presented in Sando and Wick (1972) and
relied on conventional field-sampled data (e.g., height, d.b.h.,
stem count density) to derive quantitative observations of
canopy fuels. This method was subsequently modified for
(c)
inclusion in Fire and Fuels Extension to the Forest Vegetation
Simulator (Beukema et al. 1997). As described by Scott and
Reinhardt (2001) a vertical profile of bulk density is derived
by first calculating the foliage and fine branch biomass for
each tree in the plot, then dividing that fuel equally into 1-foot
(0.3048-m) horizontal layers from the base of the tree’s crown
through to the maximum tree height and finally summing the
LVIS = Laser Vegetation Imaging Sensor.

2005 Proceedings of the Seventh Annual Forest Inventory and Analysis Symposium 195
fuel loads contributed by each tree in the plot for all 1-foot seg- case approximately 30-cm deep) by the total energy in the
ments. CBD is estimated by finding the maximum of a 4.5-m waveform. The normalization process accounts for flight-to-
deep running average for the horizontal layers of CBD. CBH is flight as well as footprint-to-footprint variations in energy in the
typically defined as the height in the profile at which the CBD waveform, caused, for example, by flying at day versus night or
reaches a predetermined threshold value. In this study, CBH by the incident angle of the laser beam. Normalization allows
is defined as the height in the profile at which the bulk density for easier comparison of waveform-derived metrics.
equals or exceeds 0.011 kg/m3 (Scott and Reinhardt 2001).
Third, the waveform metrics listed above were calculated for
CBD and CBH were derived from LIDAR data for waveforms each of the normalized waveforms. HT was determined by
that were coincident with the study’s field plots. This process subtracting the range to the ground (defined as the midpoint
involved several steps. First, LIDAR metrics were identified of the last peak) from that of the first detectable canopy return
as potential predictors based on previous work deriving other above noise. HT2 is the squared value of HT. CE and GE are
biophysical characteristics from waveform data such as canopy derived by separating the waveform into a canopy portion and
cover, basal area, and biomass. The LIDAR metrics selected a ground portion and then summing the bin values for those
were canopy height (HT), canopy height squared (HT2), canopy portions of the waveform. L is the height of the bottom of the
energy (CE), canopy energy/ground energy ratio (CE/GE), canopy portion of the waveform. D is the vertical extent of the
lowest canopy return (L), canopy depth (D), peak amplitude canopy portion of the waveform. MAX is the peak amplitude
(MAX), and the height of median cumulative canopy energy value in the canopy portion of the waveform. HMCE is the
(HMCE) (fig. 3). height at which the cumulative energy in the canopy portion
of the waveform reaches the 50th percentile. Several additional
Second, individual waveforms were normalized by dividing metrics were derived to predict CBH from the cumulative
the energy present in each waveform bin (representing the canopy energy profile. The additional LIDAR-derived CBH
energy returned for each vertical resolution unit, in this metrics include the 0.5th-, 1st-, 5th-, and 10th-percentile heights of
the cumulative canopy energy.
Figure 3.—Schematic of an individual LIDAR waveform
showing LIDAR metrics. A pulse of laser energy reflects off A transformation was also applied to the LVIS waveforms.
canopy (e.g., leaves and branches) and the ground beneath,
Some previous studies (Lefsky et al. 1999, Means et al. 1999)
resulting in a waveform. The amplitudes of individual peaks
in the waveform are a function of the number of reflecting have maintained that LIDAR waveform data need to be adjusted
surfaces at that height. The different LIDAR metrics used in to correct for shading of lower foliage and branches by higher
this study are superimposed on the waveform. foliage and branches. This adjustment consists of applying an
exponential transform to the waveform (modified MacArthur-
Horn [1969] method) and is described in detail in Lefsky et al.
(1999). The transform has the effect of increasing the amplitude
of the waveform return from the lower part of the canopy.

Once the LIDAR metrics were calculated, they were used as

explanatory variables in multiple linear regression analyses
to determine which set of metrics best predicted CBD and
CBH. Separate regression equations were derived for different
vegetation types. The vegetation type categories used in this
study were red fir, white fir, ponderosa pine, miscellaneous
Waveform amplitude
pine (comprised of Jeffrey pine, sugar pine, and lodgepole

196 2005 Proceedings of the Seventh Annual Forest Inventory and Analysis Symposium
pine), Sierra mixed conifer, mixed hardwood conifer/mixed from the field data could be overestimating the canopy loads
hardwood, and meadow/bare ground. Because the number of of the codominant and subdominant trees. The trees in denser
plots in two of the vegetation classes (white fir and ponderosa stands have crowns that are often irregular in shape, meaning
pine) was small, some explanatory variables were dropped that actual fuel load for these trees is likely much lower than
out of the regression equations for these classes. Stepwise predicted when a regular shape is assumed in an algorithm. In
regression techniques were used to determine which variables addition, there is considerable variation in crown shape among
should be dropped because they had relatively low explanatory species. White fir, for example, tends to be rather cone shaped
power. The same suite of LIDAR metrics were recalculated while sugar or ponderosa pine crowns are more parabolic.
from the waveforms once the modified MacArthur-Horn Furthermore, the field-based estimates of CBD only consider
transformation was applied. The metrics derived from the the fraction of fuels made up of fine (e.g., foliar) material rather
transformed waveforms were then used as variables in the same than the total biomass in the plot, which is recorded by the
series of regression analyses for the different vegetation types LIDAR waveform.
as described above.
We believe that at least part of the error in the CBH derivation
The LIDAR-predicted and field-derived CBD compared rather can be attributed to the fact that trees less than 10 cm d.b.h.
well. The r2 value of 0.71 (p < 0.0001, root square error (RSE) were not sampled in the field. For certain plots (especially
= 0.036) is based on the correlation between the collective mixed conifer) this excludes a significant number of smaller
observed and predicted estimates of CBD. The regression stems and could lead to an erroneously high derivation of CBH
analyses were then repeated using the transformed LIDAR from the field data. The omission of smaller trees could cause
data. This dropped the r2 value to 0.67. The comparison the amount of material assigned to the lower part of the density
between the LIDAR-based and field-based estimates of CBH is profile to be less than it should be.
also rather good. For CBH, the regression model was improved
when using the LIDAR metrics derived from the transformed Other factors such as slope and varying footprint size (due to
waveform. The r2 using the transformed data was 0.59 (p < changes in surface elevation) were explored to determine if they
0.0001, RSE = 0.573) as compared to an r2 of 0.48 using the might be a source of error for both the CBD and CBH LIDAR
metrics from the original waveform. Again, the reported r2 for derivations. No relationship between the residuals of the
the CBH derivation is based on the correlation between the regression and these factors could be discerned, however.
collective observed and predicted estimates.
Interestingly, the results of the CBH regression analyses
The differences between the various vegetation-type specific show that LVIS metrics that were derived from waveforms
regression models most likely reflect structural differences transformed using the modified MacArthur-Horn method were
among the various forest stands included in the study. For most better able to predict CBH (r2 = 0.59) than the untransformed
of the vegetation types the relationship between the LIDAR- metrics (r2 = 0.48). The transform increases the amplitude of
metrics and field-derived CBD is fairly strong (i.e., r2 > 0.6), the return in the lower portion of the waveform and therefore it
the exception being the mixed conifer class (r2 = 0.3811), has a greater impact on the metrics derived from that part of the
where, in the higher range of values, the predicted CBD was waveform. The overall effect of the transform was to lower the
lower than the observed CBD. The greatest error in predict- height of several metrics. This caused the correlation between
ing CBD occurred in stands characterized by a dense canopy predicted and observed CBH at the shorter end of the range
layer of mid and understory trees with a few dominant tree (0–2 m) to improve, thereby also improving the overall r2. The
crowns interspersed. The equations used to calculate CBD poorest results were again for the mixed conifer class.

2005 Proceedings of the Seventh Annual Forest Inventory and Analysis Symposium 197
FARSITE Simulations The wind and weather input data used for the two model runs
were representative of a dry, warm day and the simulated
The fire behavior model used in this study is the Fire Area duration was set to 40 hours.
Simulator (FARSITE, Finney 1998). FARSITE is a Geographic
Information System-based fire behavior model in common use Figure 4 shows the output (crown fire/no crown fire status) for
with agencies throughout the Unites States. In all, FARSITE the two model runs. The extent of the fire spread is very similar
has eight input layers (Finney 1998). The first five (elevation, for both of the model runs. Though occurring in similar loca-
slope, aspect, fuel model, and canopy cover) are all that are tions, the occurrence of crown fire as discrete clusters in the
needed to simulate surface fires. The last three (canopy height, LVIS output is very different from the larger, continuous areas
CBD, and CBH) are needed to model crown fire behavior. of crown fire shown in the CONV output grid. In the LVIS out-
put grid the crown fire clusters appear to be associated with the
Once the regression models for CBD and CBH were developed presence of higher CBD values and lower CBH values, which
they were used to derive CBD and CBH from all of the LVIS are assumed to promote the spread of fire to the canopy. Future
waveforms in the study area. First, the required LIDAR metrics research will explore not only the effect of increasing or de-
were calculated from the waveforms. Then the LVIS data creasing the canopy structure values on model outputs but also
were classified by land cover type and the vegetation-type the effect of increased spatial heterogeneity in the input layers.
specific regression models were applied. This created point
data of CBD and CBH for the entire study area. These point
Figure 4.—FARSITE crown-fire/no-crown-fire outputs for two
data were then gridded into 25-m raster layers using ArcInfo.
model runs using LVIS and CONV canopy structure inputs.
These grids are hereafter referred to as LVIS grids. An inverse
difference weighting (IDW) technique was used for gridding
and to compensate for gaps in the data caused by irregularities
in the flight lines. To complete the set of canopy structure data
needed to run FARSITE, an LVIS-derived canopy height grid
was also created. Hyde et al. (2005) validated the LVIS canopy
height measurement for the Sierra Nevada study site. For this Crown fire No crown fire
study, the height data were also gridded to 25 m using the IDW
CONV = conventional; LVIS = Laser Vegetation Imaging Sensor.
technique.

Once the LVIS grids were created they were first compared to Conclusions
canopy height, CBD, and CBH data layers that were generated
using conventional methods, referred to hereafter as CONV LIDAR systems of different types have had success in recover-
grids. The CONV grids were only available for a smaller part ing forest structure characteristics for a variety of vegetation
of the study area—at the far southeastern end of the flight lines. types in a comparatively simple and direct manner. In recent
Therefore, the LVIS grids were clipped to match the extent of years LIDAR has become recognized as a valuable remote
the CONV grids. There are obvious differences between the two sensing tool for forest inventory and structure mapping and is
sets of data. Of particular note is the increased spatial heteroge- gaining in use for informing forest management decisionmak-
neity contained in the LVIS grids relative to the CONV grids. ing. Because of its ability to measure canopy structure both
horizontally and vertically, LIDAR has potential for providing
FARSITE was then run twice: once using the LVIS canopy the type of forest structure required for fuels estimation and
structure grids and once using the CONV input grids. All other fire behavior modeling. The results of this paper demonstrate
spatial inputs were kept constant as was the point of ignition. that waveform data from a large-footprint system may provide

198 2005 Proceedings of the Seventh Annual Forest Inventory and Analysis Symposium
the spatially explicit forest structure needed for fire behavior Blair, J.B.; Coyle, D.B.; Bufton, J.; Harding, D. 1994.
modeling. We will continue to explore and improve on methods Optimization of an airborne laser altimeter for remote sensing
for deriving CBD and CBH from LIDAR. One option to be of vegetation and tree canopies. In: Proceedings, IEEE
considered is to incorporate various remote sensing data from international geoscience and remote sensing symposium.
other sensor types into a fusion-based approach for deriving the Los Alamitos, CA. 2: 939-941.
canopy structure variables. These results also have implications
for remote-sensing-based inventory at larger scales. ICESAT Blair, J.B.; Rabine, D.L.; Hofton, M.A. 1999. The Laser
and other near-future space-based LIDAR systems are or will Vegetation Imaging Sensor (LVIS): a medium-altitude,
likely be large footprint and waveform digitizing. Though these digitization-only, airborne laser altimeter for mapping
are not imaging systems, the global samples of three-dimen- vegetation and topography. ISPRS Journal of Photogrammetry
sional structure that they will provide can be incorporated into and Remote Sensing. 54: 115-122.
forest inventory.
Drake, J.B.; Dubayah, R.O.; Clark, D.; Knox, R.; Blair, J.B.;
Hofton, M.; Chazdon, R.L.; Weishampel, J.F.; Prince, S. 2002a.
Acknowledgments Estimation of tropical forest structural characteristics using large-
footprint LIDAR. Remote Sensing of Environment. 79: 305-319.
The authors thank Carolyn Hunsaker, Wayne Walker, Steve
Wilcox, Craig Dobson, Leland Pierce, Malcolm North, and Drake, J.B.; Dubayah, R.O.; Knox, R.G.; Clark, D.B.; Blair, J.B.
Brian Boroski for all of their effort in developing the sampling 2002b. Sensitivity of large-footprint LIDAR to canopy structure
protocol and organizing and conducting the field sampling and biomass in a neotropical rainforest. Remote Sensing of
effort. The authors also thank David Rabine for his work in Environment. 81: 378-392.
collecting and processing the LVIS data and Michelle West, J.
Meghan Salmon, Josh Rhoads, Ryan Wilson, Sharon Pronchik, Dubayah, R.; Drake, J. 2000. LIDAR remote sensing for
Aviva Pearlman, John Williams, Brian Emmett, and all the forestry. Journal of Forestry. 98: 44-46.
others who collected and recorded the field data. This work
was supported by a grant to Ralph Dubayah from the NASA Dubayah, R.; Knox, R.; Hofton, M.; Blair, B.; Drake, J. 2000.
Terrestrial Ecology Program. Land surface characterization using LIDAR remote sensing.
In: Hill, M.; Aspinall, R., eds. Spatial information for land use
management. Singapore: Gordon and Breach: 25-38.
Literature Cited
Finney, M.A. 1998. FARSITE: Fire Area Simulator—model
Andersen, H.E.; McGaughey, R.J.; Reutebuch, S.E. 2005. development and evaluation. Res. Pap. RMRS-RP-4. Ft. Collins,
Estimating forest canopy fuel parameters using LIDAR data. CO: U.S. Department of Agriculture, Forest Service, Rocky
Remote Sensing of Environment. 94: 441-449. Mountain Research Station. 47 p.

Beukema, S.J.; Greenough, D.C.; Robinson, C.E.; Kurtz, W.A.; Hofton, M.; Rocchio, L.; Blair, J.B.; Dubayah, R. 2002.
Reinhardt, E.D.; Crookston, N.L.; Brown, J.K.; Hardy, C.C.; Validation of vegetation canopy LIDAR sub-canopy
Sage, A.R. 1997. An introduction to the fire and fuels extension topography measurements for dense tropical forest. Journal of
to FVS. In: Teck, R.; Mouer, M.; Adams, J., eds. Proceedings, Geodynamics. 34: 491-502.
forest vegetation simulator conference. Gen. Tech. Rep. INT-
GTR-373. Ogden, UT: U.S. Department of Agriculture, Forest
Service, Intermountain Research Station: 191-195.

2005 Proceedings of the Seventh Annual Forest Inventory and Analysis Symposium 199
Hyde, P.; Dubayah, R.; Peterson, B.; Blair, J.B.; Hofton, M.; Nelson, R. 1997. Modeling forest canopy heights: the effects of
Hunsaker, C.; Knox, R.; Walker, W. 2005. Mapping forest canopy shape. Remote Sensing of Environment. 60: 327-334.
structure for wildlife analysis using waveform LIDAR:
validation of montane ecosystems. Remote Sensing of Nelson, R.; Krabill, W.; Maclean, G. 1984. Determining forest
Environment. 96: 427-437. canopy characteristics using airborne laser data. Remote
Sensing of Environment. 15: 201-212.
Lefsky, M.A.; Cohen, W.B.; Parker, G.G.; Harding, D.J. 2002.
LIDAR remote sensing for ecosystem studies. BioScience. 52: Nelson, R.; Swift, R.; Krabill, W. 1988. Using airborne lasers
19-30. to estimate forest canopy and stand characteristics. Journal of
Forestry. 86: 31-38.
Lefsky, M.A.; Harding, D.; Cohen, W.B.; Parker, G.; Shugart,
H.H. 1999. Surface LIDAR remote sensing of basal areas Nilsson, M. 1996. Estimation of tree heights and stand
and biomass in deciduous forests of eastern Maryland, USA. volume using an airborne LIDAR system. Remote Sensing of
Remote Sensing of Environment. 67: 83-98. Environment. 56: 1-7.

MacArthur, R.; Horn, H. 1969. Foliage profile by vertical Peterson, B. 2000. Recovery of forest canopy heights using
measurements. Ecology. 5: 802-804. large-footprint LIDAR. College Park, MD: University of
Maryland. 59 p. M.A. thesis.
Maclean, G.; Krabill, W.B. 1986. Gross-merchantable timber
volume estimation using an airborne LIDAR system. Canadian Riaño, D.; Chuvieco, E.; Condés, S.; González-Matesanz, J.;
Journal of Remote Sensing. 12: 7-18. Ustin, S. 2004. Generation of crown bulk density for Pinus
sylvertris L. from LIDAR. Remote Sensing of Environment.
Magnussen, S.; Boudewyn, P. 1998. Derivations of stand 92: 345-352.
heights from airborne laser scanner data with canopy-based
quantile estimators. Canadian Journal of Forest Research. 28: Riaño, D.; Meier, E.; Allgöwer, B.; Chuvieco, E.; Ustin, S.
1016-1031. 2003. Modeling airborne laser scanning data for the spatial
generation of critical forest parameters in fire behavior
Means, J.E.; Acker, S.A.; Harding, D.J.; Blair, J.B.; Lefsky, modeling. Remote Sensing of Environment. 86: 177-186.
M.A.; Cohen, W.B.; Harmon, M.E.; McKee, W.A. 1999. Use
of large-footprint scanning airborne LIDAR to estimate forest Sando, R.W.; Wick, C.H. 1972. A method of evaluating crown
stand characteristics in the western Cascades of Oregon. fuels in forest stands. Res. Pap. NC-84. St. Paul, MN: U.S.
Remote Sensing of Environment. 67: 298-308. Department of Agriculture, Forest Service, North Central
Forest Experiment Station. 10 p.
Morsdorf, F.; Meier, E.; Kötz, B.; Itten, K.I.; Dobbertin, M.;
Allgöwer, B. 2004. LIDAR-based geometric reconstruction Scott, J.H.; Reinhardt, E.D. 2001. Assessing crown fire potential
of boreal type forest stands at single tree level for forest and by linking models of surface and crown fire behavior. Res. Pap.
wildland fire management. Remote Sensing of Environment. RMRS-RP-29. Ft. Collins, CO: U.S. Department of Agriculture,
92: 353-362. Forest Service, Rocky Mountain Research Station. 59 p.

Naesset, E. 1997. Determination of mean tree height of forest Seielstad, C.A.; Queen, L.P. 2000. Using airborne laser
stands using airborne laser scanner data. Remote Sensing of altimetry to determine fuel models for estimating fire behavior.
Environment. 61: 246-253. Journal of Forestry. 101: 10-15.

200 2005 Proceedings of the Seventh Annual Forest Inventory and Analysis Symposium
Area-Independent Sampling for Basal Area

James W. Flewelling1 Sample Protocol and Estimators

Abstract.—An unbiased direct estimator of total The sampling protocol addressed here is that of a regular grid of
basal area for a stand (Flewelling and Iles 2004) sample points. The orientation of the grid is predetermined. A
is reviewed. Stand area need not be known. The starting point is randomly located within an area corresponding
estimator’s primary application is in conjunction with to a grid cell, and the sample points extend indefinitely to areas
a randomly positioned grid of sample points. The inside and outside the stand. Other protocols are addressed by
points may be centers for horizontal point samples or Flewelling and Iles (2004). Each sample point may be the center
fixed-area plots. The sample space extends beyond of a fixed-area plot or a horizontal point sample. No distinction
a stand’s boundary, though only trees within the is made between sample points that fall inside the stand, and
boundary are tallied. Measured distances from sample those that fall outside the stand. At each point, only the sample
trees to stand boundaries are not required. trees within the stand are considered.

For fixed-area plots, the estimator of total basal area is

Introduction
Ĝ =Ag × Σ gi (1)
Most methods of estimating basal area for a stand are area
dependent in that they are the product of an estimated basal where Ag is the area per grid point as established by the grid
area per hectare and a known or estimated area. A major spacing, gi is the basal area per hectare on the ith sample plot,
concern in applying these methods is the avoidance of edge and the summation is over all sample plots. For horizontal point
bias. Such bias can arise when the distance from a tree to the samples, the estimator is
stand’s edge affects its sampling probability and the estimator is
not able to fully account for the varying probabilities. Unbiased Ĝ = T × F × Ag (2)
estimators do exist, but are difficult or impossible to apply with
complex stand boundaries. Methods which adjust for edge bias where F is the basal area factor and T is the total tree count,
are reviewed by Schreuder et al. (1993). The “walkthrough” summed over all sample points. Modified versions of horizontal
solution (Ducey et al. 2004) offers an operationally simpler point sample may use several different basal area factors
alternative to the mirage method of Schmid-Haas (1969, 1982). depending on tree size, and may invoke fixed-area plots for
The foregoing methods confine sample points to being within certain ranges of tree sizes. The generalized estimator for these
the stand boundaries. Schmid-Haas (1982: 264) also suggested modified samples is
a substantively different approach to the edge bias problem:
“One possibility is obvious; sample plots whose centre lies (3)
outside the area under investigation are also included in the
sample, care being taken to ensure that the probability density where Fv is a variable basal area factor, the first summation is
for such plot centers is the same as for those within the (stand) over all sample points, and the second summation is over all the
area.” That concept is embodied in the toss-back method by trees at a particular sample point. For tree sizes being sampled
Iles (2001) and in the area-independent method reviewed here. with an angle gauge, Fv is the basal area factor of the gauge. For

1
Consulting biometrician, 9320 40th Avenue N.E., Seattle, WA 98115. E-mail: [email protected].

2005 Proceedings of the Seventh Annual Forest Inventory and Analysis Symposium 201
tree sizes being sampled with fixed area plots, Fv is the ratio of Literature Cited
the tree’s basal area to the plot area.
Ducey, M.G.; Gove, J.H.; Valentine, H.T. 2004. A walkthrough
solution to the boundary overlap problem. Forest Science. 50:
Discussion 427-435.

The most likely application of the area-independent estimator Flewelling, J.W.; Iles, K. 2004. Area-independent sampling for
is for stands where the area is unknown. An example is in the total basal area. Forest Science. 50: 512-517.
determination of the basal area of that portion of a stand which
excludes riparian corridors whose extent and area are unknown. Iles, K. 2001. Edge effect in forest sampling, part II. Inventory
and Cruising Newsletter. 53: 1‑3. http://www.proaxis.com/
The appeal of the area-independent estimator is not limited to ~johnbell/. (7 January 2006).
stands with unknown areas. This estimator and the toss-back
method both are unbiased for any stand geometry and are rela- Reams, G.A.; Smith, W.D.; Hansen, M.H.; Bechtold, W.A.;
tively easy to use. The exact delineation of the stand boundary Roesch, F.A.; Moisen, G.G. 2005. The forest inventory and
in the vicinity of the sample points is not required. Independent analysis sampling frame. In: Bechtold, W.A.; Patterson, P.L.,
random errors in the location of sample points would seem eds. The enhanced Forest Inventory and Analysis program—
not to introduce bias. This lack of sensitivity to location error national sampling design and estimation procedures. Gen. Tech.
is not shared by methods that limit sample points to a stand’s Rep. SRS-80. Asheville, NC: U.S. Department of Agriculture,
interior; this feature could be used to advantage by using hand- Forest Service, Southern Research Station: 11-26.
held Geographic Positioning System units to navigate to sample
points. An operational difficulty of the method is that some of Schmid-Haas, P. 1969. Stichproben am Walrand. Sample plots
the sample points outside of the stand may be inaccessible; for at forest stand margins. Mitteilungen der Schweizerischen
those sample points, the selection of sample trees will be much Anstalt für das forstliche Versuchswesen. 45: 234-303. English
more difficult than making a prism sweep. summary.

The Forest Inventory and Analysis (FIA) program is generally Schmid-Haas, P. 1982. Sampling at the forest edge. In:
not concerned with individual stands. Instead, forest attributes Ranneby, B., ed. Statistics in theory and practice: essays in
are sought within populations such as States or counties, and by honour of Bertil Matern. Umea, Sweden: Swedish University of
various condition classes such as forest cover type. The FIA’s Agricultural Science: 263-276.
grid of ground plots have a constant sampling density and could
be analyzed with the area-independent estimator to make unbi- Schreuder, H.T.; Gregoire, T.G.; Wood, G.B. 1993. Sampling
ased estimates of basal area by cover type. The FIA sampling methods for multi-resource inventories. New York: John Wiley.
program, however, is multiphase; the first phase measures or 446 p.
estimates forest area (Reams et al. 2005), and could potentially
subdivide the forested area into condition classes. Hence, area-
dependent estimators for basal area and other attributes are be-
ing used; these should be presumed to have lower variance than
would the area-independent estimators.

202 2005 Proceedings of the Seventh Annual Forest Inventory and Analysis Symposium