Conference PaperPDF Available

Mapping Forest Biomass Carbon by Fusing Terrestrial and Airborne LiDAR Datasets


Abstract and Figures

The storage and flux of terrestrial carbon (C) is one of the most uncertain components of the global C budget and detailed quantification of forest C remains difficult to measure on a large scale. Remote sensing of forests with airborne LiDAR has proven to be an effective method of bridging the gap between data from plot-level mensuration and landscape-scale C storage estimates, but the standard method of assessing forest C is typically based on national or regional-scale allometric equations that are often not representative on the local-scale. The goals of this study were to determine if terrestrial LiDAR (TLS) can be used to improve local allometric equations and what are the implications of fusing these equations to airborne LiDAR data sources for biomass estimation. Our non-destructively determined allometric relationships had lower error than national-scale equations (R 2 = 0.98, RMSE = 125 kg). National allometric equations overestimated biomass in the species studied and total site-wide biomass significantly decreased by over 35 Mg/ha when our equation was used. The proposed pipeline has the potential to significantly reduce labor costs associated with destructive sampling and simplify the process of detailed C monitoring, while improving the quality and quantity of data retrieved.
Content may be subject to copyright.
Atticus Stovall 1
Author: Atticus E. L. Stovall, University of Virginia
Advisor: Herman H. Shugart, University of Virginia
1.0: Abstract
The storage and flux of terrestrial carbon (C)
is one of the most uncertain components of the
global C budget and detailed quantification of
forest C remains difficult to measure on a
large scale. Remote sensing of forests with
airborne LiDAR has proven to be an effective
method of bridging the gap between data from
plot-level mensuration and landscape-scale C
storage estimates, but the standard method of
assessing forest C is typically based on
national or regional-scale allometric equations
that are often not representative on the local-
scale. The goals of this study were to
determine if terrestrial LiDAR (TLS) can be
used to improve local allometric equations and
what are the implications of fusing these
equations to airborne LiDAR data sources for
biomass estimation. Our non-destructively
determined allometric relationships had lower
error than national-scale equations (R2= 0.98,
RMSE = 125 kg). National allometric
equations overestimated biomass in the
species studied and total site-wide biomass
significantly decreased by over 35 Mg/ha
when our equation was used. The proposed
pipeline has the potential to significantly
reduce labor costs associated with destructive
sampling and simplify the process of detailed
C monitoring, while improving the quality and
quantity of data retrieved.
2.0: Introduction
Current rising atmospheric CO2
concentrations are a major concern with
significant global ramifications and the
quantification of terrestrial carbon (C) storage
lies on the forefront of scientific
understanding of the global C cycle. Of the C
fluxes that are known to occur on Earth, the
terrestrial sink has the greatest amount of
uncertainty and human-induced change of
landscape-scale ecosystems has complicated
the C cycle, creating a necessity for
monitoring the dynamics of these global C
stocks. Existing mature and secondary
successional forests act as an extremely
effective means of active C sequestration,
potentially mitigating the predicted future
changes in climate.
Actions have been taken by the United
Nations Framework Convention on Climate
Change (UNFCCC) to improve monitoring of
forest cover for Reduced Deforestation and
Degradation (REDD). The three tier system
implemented in these assessments of C
storage are categorized specifically according
to the level of uncertainty and level of detail
within the estimates, with Tier 3 being the
most accurate (IPCC 2006). Forest C storage
is typically based on national- or regional-
scale allometric equations using traditional
methods of determining aboveground
biomass. These estimates vary significantly
depending on region and environmental
factors (Jenkins et al. 2003; Chojnacky, Heath,
and Jenkins 2014). Furthermore, current
methods for developing these relationships
require destruction of the sample and trees that
are located in protected areas often go
unmeasured due to their ecological and
cultural significance (Picard, Saint-Andre, and
Henry 2012). Improvement of these
measurements is necessary in order for
collaborative international carbon monitoring
programs to be successful in areas with tree
Atticus Stovall 2
species that have insufficiently documented
allometric relationships (Asner 2009).
Strengthening confidence in these estimates
will aid in development of landscape-scale
modeling projections of carbon storage.
Furthermore, the study of the structural
allocation of this carbon is needed in order to
fully understand the role of forests in the
carbon cycle (Shugart, Saatchi, and Hall
Remote sensing has become the primary
method of monitoring C storage and fluxes on
a global scale, offering increasingly higher
spatial resolutions as technology improves.
Satellites such as the newly launched Landsat
8 greatly improve the understanding of land-
cover changes; however, resolving C stocks in
areas with high vegetation density (leaf area
index (LAI) > 3) is rarely successful without
ground-based validation (Lefsky et al. 2002).
Airborne Light Detection and Ranging
(LiDAR) provides extremely high-resolution
data that is useful for quantifying biomass and
forest structure over large areas (Cook et al.
2013). Ground validation of these estimates is
essential in order for these data to be useful
for C quantification.
Terrestrial Laser Scanners (TLS) are
stationary ground-based LiDAR systems that
emit a high number of pulses. Multiple TLS
scans registered together can resolve
vegetation structure with a previously
unprecedented level of detail. These devices
are being used more frequently in the field of
precision forestry (Hopkinson et al. 2004) and
techniques have been developed that fuse ALS
and TLS to form a comprehensive view of the
forest (Popescu, Wynne, and Scrivani 2004).
Volumetric measurements of vegetation are
made with digital analysis, resulting in a fine-
scale plot level estimation of biomass with
minimal effort compared to labor intensive
field-based methods (Srinivasan et al. 2014).
The focus of this work is to determine if
terrestrial LiDAR can be used as an effective
tool to improve local allometric equations and
what might be the implications of fusing these
equations to airborne LiDAR data sources for
biomass and carbon estimation across the
3.0: Methodology
3.1: Data Sources
The acquisition of data for this project are
from a mixture of publicly available data sets
and field-based data collection. A stem map
from a 25.6 ha Smithsonian Institute Global
Earth Observatory (SIGEO) plot in the
Smithsonian Conservation Biology Institute
near Front Royal, VA was used, containing
data on all woody stems > 1 cm DBH (Bourg
et al. 2013). The allometric relationship used
in this work is within the Hardwood -
Figure 1: Map of site and plot distribution.
Atticus Stovall 3
Fagaceae, deciduous classification, as
described by Chojnacky et al. (2014).
3.2: Field Sampling Techniques
Sampling took place in the SIGEO plot in
Front Royal, Virginia, USA. 14 1/10 ha plots
in each sampling location were chosen at
random using ArcGIS mapping software.
Individual plots were located using a handheld
Garmin eTrex unit and plot centers were
found on the marked grid intersections of the
SIGEO plot. We used the Faro Focus3D120
terrestrial laser-scanner to collect the
following data: location relative to the scanner
in xyz space, intensity values determined by
the 905 nm laser wavelength reflectance, and
RGB values. Scan resolution and incidence of
error are on the order of ±2 mm. In order to
reduce occlusion from the presence of high-
density vegetation we scanned 5-6 times in
different locations on each plot.
Time elapsed for an individual TLS scan
averages about 5.5 minutes, depending on the
specific settings selected. Prior research
suggests that the ideal scanning parameters for
resolving individual tree features is ¼
resolution (10 kpts/360°), resulting in a 2.5
minute scan time with point spacing of 0.6 cm
at a 10 m distance (Pueschel et al. 2013). The
faster scan times reduce the error induced by
movement of objects in the scan area (i.e.
wind-blown trees and leaves). The center scan
takes place at the marked plot center and all
subsequent scan locations are geo-registered
from that point.
3.3: Post Processing
Individual scans were registered using the
automatic registration algorithms included in
Faro’s proprietary software package SCENE
and overlapping redundant points were
deleted/fused together to create a seamless 3D
point cloud that was used for analysis. A “dark
scan point” filter was used to reduce noise
caused by low intensity returns and a 2 cm
“density filter” reduces the presence of uneven
distribution of points. The fully registered
scan were then exported with intensity values
and RGB values for further analysis. The
ground level was estimated using the Quick
Terrain Modeler Automatic Ground Level
(AGL) tool and a Digital Elevation Model
(DEM) was exported. The ground-level points
were then removed from the point cloud to
reduce the statistical bias from the high
density of ground points in TLS data. Once
leveled, the data was isolated between 1.3 and
1.4 m above ground level and exported for use
in automatic stem map creation and diameter
(DBH) measurement.
3.4: Individual Tree Measurements
To automate and increase the accuracy of
DBH measurements we developed an R-script
that completes this task with the least squares
circle fitting algorithm by minimizing the sum
of the squared residuals between a 2 cm DBH-
height slice in the returns and the optimally
fitting circle (Coope 1993). Preliminary
testing indicates this method to be ideal for
noisier scenes, where vegetation may be
Figure 2: Study design and workflow.
TLS Data Collection
Data Fusion
Model Evaluation
Plot location
Volume Algorithm
Site Stem Map Biomass
Airborne LiDAR Biomass
Convex-hullVoxel building
TLS Allometric Relationships
DBH Height Biomass
Chojnacky et al. 2014 Biomass
Filter Registration
Clip Tree Segmentation
5 TLS Scans
Atticus Stovall 4
occluding a portion of the trunk section. To
further validate these three-dimensional
measurements, the DBH values from the stem
map on the SIGEO plot were used as a cross-
reference to determine the accuracy of this
TLS-based measurement.
3.6: Biomass Estimations from TLS Data
In order to provide a standard estimate of
aboveground biomass for a clear comparison
to our improved measurements we calculated
plot-level biomass from measured DBH based
on the newly updated United States allometric
equations (Chojnacky, Heath, and Jenkins
For this project, volume was estimated with
TLS data using a novel combination of a
convex-hull algorithm and the voxelization
process. Several studies have produced
successful algorithms for volume estimation in
trees ranging from simple stem diameter
calculations (Bienert et al. 2006) and
quantitative surface models (Hackenberg et al.
2014; Calders et al. 2015) to voxel-based
models (Fumiki Hosoi, Nakai, and Omasa
2013; Bienert et al. 2014). We developed an
optimized voxel-convex-hull-based algorithm
that combines several of the techniques
implemented in the above work. Voxelization
of the entire point cloud is the essential
preliminary step in volume quantification.
Once completed, the bulk of the algorithm
development was focused on effective voxel
infilling of the tree point clouds. The method
developed by Hosoi et al. in 2013 uses a voxel
grid to generate surface contours before
infilling takes place, but this can lead to
overestimates of the total wood volume if
filled voxels are used instead of interpolated
original points.
In order to attempt to overcome these stated
limitations of the previous methods described
our algorithm relies on a convex-hull method
of creating a solid exterior model of the tree.
The algorithm essentially traces the outermost
points of the trunk to create a seamless outline
of the trunk, from which a volume can be
determined. Appropriate segmentation of the
tree is essential for accurate volume
estimations from the TLS data. Since the
current version of the algorithm cannot
differentiate between branching points,
branches larger than the voxel size must be
separated out individually in order for the
functions that the algorithm relies on to be
valid. Furthermore, areas where branches
become bent downward must be segmented at
the vertex of the bend. If correct segmentation
is not completed the volume will invariably be
significantly overestimated. Once all branches
are segmented the algorithm can be initiated
across both the central bole of the tree, as well
as all branches and segments larger than the
specified voxel size. The process, at this point,
is entirely automated and completed on
individual 2 cm thick slices from the point
cloud, moving from Zmin ≤ Z ≤ Zmax,
Figure 3: Example trunk section before
and after processing the TLS data with
the volume infill algorithm.
Atticus Stovall 5
producing a completely solid volumetric
model of an individual tree. The final process
can also be completed with a height buffer of
specified thickness in order to infer
information from the areas of missing data on
the trunk, reducing the likelihood of
underestimation of volume. The final step
invokes a voxelizing algorithm to estimate
volume attributed to branches smaller than the
stated voxel size. In order to complete this,
each remaining voxel is assumed to represent
a 1 cm diameter by 2 cm long cylinder and
small-branch volume is then estimated. The
combined volume estimates were then
converted to aboveground biomass by
applying species-based wood density values.
This will provide two pieces of essential
[a] An individual species specific non-
destructively estimated biomass data point that
will be used to improve and create new local
allometric equations.
[b] A high resolution model of the vertical
distribution of biomass and carbon on a per-
plot level, information of which will be
essential in development of predictive models
of these values using the point cloud
3.7: Creation of TLS-based Allometric
The process of allometric equation
development is one that is time consuming
and expensive. The methods of non-
destructive biomass estimation described
above attempt to improve allometric
equations via inclusion of previously
prohibitively complex trees that were not
ideal candidates for destructive sampling. In
order to link the TLS-based biomass estimates
to both the whole forest stem map and
airborne LiDAR dataset we created our own
allometric equation from 24 digitally sampled
trees across the forest within the Hardwood -
Fagaceae, deciduous classification, as
described by Chojnacky et al. (2014). The
final equation uses diameter as a predictor for
aboveground biomass as a log-linear
3.8: Fusion of TLS and Airborne LiDAR
The final step in our analysis was the
assessment of how feasible our TLS-based
allometric equation would be when included
in a conventional airborne LiDAR processing
pipeline. Following standard procedure for
airborne LiDAR analysis, we used the
CloudMetrics algorithm included within the
Fusion LiDAR Data Toolkit to produce 94
different point cloud metrics for each of the 14
TLS sample plots. A subset of these metrics
was used to find the most logical combination
of variables that produce the strongest
relationship to the plot-level biomass
measurements through stepwise regression.
Between 1 and 4 metrics were used in order to
form a meaningful model for the relationship
Figure 4: DBH is reliably measured using a Least
Squares circle fit algorithm (see inset). Model fit
versus field based measured values gives a high R2
and low RMSE (R2 = 0.96, RMSE = 2.40 cm)
Atticus Stovall 6
between the LiDAR returns and aboveground
biomass. The final predictive model includes a
combination of height metrics, such as mean
canopy height and percentiles, percent cover
estimates, and structural metrics, such as
skewness and kurtosis of the point cloud
distribution at each plot location. The final
model was applied across the landscape and a
map of LiDAR-derived biomass created for
4.0: Results and Discussion
4.1: Individual Tree Measurements
TLS is an effective tool for automatically
recovering DBH measurements with minimal
effort. By using the Least Squares circle fit
algorithm DBH was reliably measured (R2 =
0.96, RMSE = 2.40 cm) with a slight
underestimation at larger diameters (Figure
4). The likely cause of this underestimation is
the non-circular (or non-cylindrical) nature of
large diameter trees. Many of the individuals
sampled in larger DBH size classes had
multiple trunks, resulting in a much more
oblong or ellipsoidal shape than circular.
Moreover, in our study site we have found it
common that larger trees generally exhibit
novel trunk shapes, likely as a result of
variations in growing conditions over the
lifetime of the individual. We anticipate that
improvements to this method of automatic
DBH recovery can be made with the
application of the above described volume
algorithm. Since the very nature of our
algorithm allows for a wide range of shapes
that can occur in trees, the retrieval of
diameter would be an ideal application. We
have experimented with trunk diameter
retrieval by calculating the area at a specific
vertical interval of the point cloud and back-
calculating an estimated equivalent circle, in
turn allowing for DBH measurement.
Interestingly, we can use this same process to
estimate diameter at a fine vertical resolution
moving up the tree, allowing for detailed
observations of stem taper, where occlusion of
the base points is not significant.
4.2: Creation of TLS-based Allometric
We were able to successfully create a TLS-
based local allometric equation from the 24
individual trees that fell within the Hardwood
- Fagaceae, deciduous classification, as
described by Chojnacky et al. (2014). With a
significantly smaller sample size (n = 24) our
equation for the same class of trees displayed
a much higher R2, describing 98% of the
variation in our data, as opposed to 84%, as
seen in the Chojnacky et al. equation (Figure
5a). The explanation for this observation can
primarily be attributed to the increased
amount of variation that will inherently be
found in larger sample sizes, even without the
consideration that the Chojnacky et al.
equations are of a national rather than a local
scale. RMSE, although not reported in the
Chojnacky et al. paper, is higher than the
RMSE found in a previous and frequently
applied set of allometric equations (Jenkins et
al. 2003) at 0.29 log(Biomass (kg)) versus
0.24 log(Biomass (kg)). Our equation may
have higher RMSE because of the small
sample size, making our dataset more
susceptible to outliers. This finding will be
investigated in future work and likely
improved with the addition of a larger number
of data points.
When comparing the two equations, the most
notable difference in the two equations is a
significant decrease in the biomass for large
trees as seen in Figure 5b and 5c. The
implications of such a large change in biomass
are of great interest. Currently, the most
Atticus Stovall 7
commonly applied allometric equations are of
a national rather than a local scale. There is
evidence that suggests application of local as
opposed to national scale biomass equations
can lead to significant differences, especially
in the context of LiDAR applications (Zhao,
Guo, and Kelly 2012). In our case, since this
study focused primarily on a single species
group, we anticipated only a slight change in
the site level biomass observed. However,
Figure 5: a) Final log-linear biomass equation produced from the three-dimensionally
modeled TLS dataset. b) Chojnacky et al. 2014 Hardwood - Fagaceae, deciduous equation
(red) compared to the TLS-derived allometric equation (RMSE = 125.33 kg). c) Comparison
of Chojnacky et al. 2014 Hardwood - Fagaceae, deciduous equation to the TLS-derived
allometric equation across all modeled trees (grey dotted-line is the 1:1 line). d) Site-wide
extrapolation of biomass density using the two equations, resulting in over 35 Mg/ha decrease
(error bars indicate 95% confidence).
a) b)
c) d)
Atticus Stovall 8
when applied across the entirety of the 25.6 ha
area we observed a 35 Mg/ha decrease in
biomass when applying the local equation to
the stem map. The large difference in biomass
density can be attributed to the prevalence of
the species group observed. Within our study
forest the dominant species (as defined by
largest basal area per plot) are oak, hickory,
and other hardwood species, so, we would
infer that small differences in the allometric
equation used will result in large differences
in biomass density across the plot.
Furthermore, this species group is especially
biomass dense, in that the wood density is
quite high compared to other dominant species
that are also large in size (i.e. yellow tulip
4.3: Fusion of TLS and Airborne LiDAR
Although this study included a relatively small
number of sample plots, the creation of
predictive multiple linear regression model for
biomass was possible, describing around 71%
if the biomass density variation across the
landscape. The sampling plan for this study
allowed us to include 5.47% (1.4 ha) of the
total plot area, which is larger than many
studies that utilize airborne LiDAR for
biomass predictive models. This relatively
high percentage of sampled area is helpful in
Figure 5: a) Comparison between plot level biomass estimates (per 1/10 ha) from the two
predictive multiple linear regression models (inset is a table of the four predictive point cloud
statistics used to create the final model). b) National scale biomass prediction using the
multiple linear regression model. c) TLS-based biomass prediction using the multiple linear
regression model.
Variable Definition
H_Skew Skewness of point cloud returns
30_P 30th percentile of returns
CCE_1m Canopy cover estimate using 1 m cutoff
CCE_Mean Canopy cover estimate using mean height cutoff
Atticus Stovall 9
creating a final predictive model, however, it
should be noted that the LiDAR dataset used
was not of the appropriate resolution to predict
biomass with a high level of confidence (point
spacing average was 1.7 m), which leads to
higher uncertainty in our final estimates. The
major findings from this portion of our study
are as follows: 1) Biomass was predicted
using the TLS-derived equations as well as the
national scale equations, with little influence
on the percentage of variation described in the
final model, and 2) biomass decreased across
all LiDAR plots with the implementation of
our TLS-based equation. If we take into
account the higher confidence due to the
higher amount of variation described in the
allometric equation development (Section
4.2), the TLS-based local allometric equation
would be the ideal selection when applying a
biomass relationship, although the R2 values
using the LiDAR data are comparable.
5.0: Conclusions and Future Work
This work has shown a potential application
and pipeline for analysis of terrestrial LiDAR
data in the context of biomass estimation in
forests. The algorithm developed for this work
performs well when applied within a complex
temperate deciduous forest during the leaf-off
season. Automatic DBH measurement was
shown to be a realistic alternative to manual
measurement with relatively low error and a
slight bias at large diameters, however we
anticipate improvements to this methodology
with the application of the described
algorithm. Our local allometric equation
derived from the TLS point cloud was low in
RMSE and described 98% of the data using a
log-linear relationship. The application of our
biomass relationship to the site-wide data
resulted in over 35 Mg/ha difference in
biomass density. When applied to the airborne
LiDAR dataset we found that our equation
performed as well as the national scale
equation, but resulted in significant site-wide
decreases in biomass density.
We are in the process of improving the
volume algorithm by comparing destructively
harvested tree samples to the TLS data. The
major current limitation is the presence of
noise in the data that will invariably lead to
over estimation in tree-level biomass. Some
exploratory methods will be used to reduce the
impact of noise when using the convex-hull
portion of our algorithm. Acquisition of high
density airborne LiDAR data and application
of the pipeline will be necessary in the future
to determine the efficacy of this process.
6.0: Acknowledgments
Thank you to my advisor Dr. Hank Shugart
for planting the seed that became this project,
Dr. Robert Washington-Allen for giving me a
foundation of TLS methods for which to build
this work, Dr. Howie Epstein, and Dr. Todd
Scanlon. Many thanks to all of those who
helped with field work including Adrienne
and Moses Stovall, Adrianna Foster, and Bella
7.0: References
Asner, Gregory P. 2009. “Tropical Forest
Carbon Assessment: Integrating
Satellite and Airborne Mapping
Approaches.” Environmental Research
Letters 4 (3): 034009.
Bourg, Norman A., William J. McShea,
Jonathan R. Thompson, Jennifer C.
McGarvey, and Xiaoli Shen. 2013.
“Initial Census, Woody Seedling, Seed
Rain, and Stand Structure Data for the
SCBI SIGEO Large Forest Dynamics
Plot: Ecological Archives E094-195.”
Ecology 94 (9): 211112.
Chojnacky, D. C., L. S. Heath, and J. C.
Jenkins. 2014. “Updated Generalized
Atticus Stovall 10
Biomass Equations for North
American Tree Species.” Forestry 87
(1): 12951.
Cook, Bruce D., Ross F. Nelson, Elizabeth M.
Middleton, Douglas C. Morton, Joel T.
McCorkel, Jeffrey G. Masek, Kenneth
J. Ranson, Vuong Ly, Paul M.
Montesano, and others. 2013. “NASA
Goddard’s LiDAR, Hyperspectral and
Thermal (G-LiHT) Airborne Imager.”
Remote Sensing 5 (8): 404566.
Jenkins, Jennifer C., David C. Chojnacky,
Linda S. Heath, and Richard A.
Birdsey. 2003. “National-Scale
Biomass Estimators for United States
Tree Species.” Forest Science 49 (1):
Lefsky, Michael A., Warren B. Cohen, David
J. Harding, Geoffrey G. Parker, Steven
A. Acker, and S. Thomas Gower.
2002. “Lidar Remote Sensing of
above-Ground Biomass in Three
Biomes.” Global Ecology and
Biogeography 11 (5): 39399.
Picard, Nicolas, Laurent Saint-Andre, and
Matieu Henry. 2012. Manual for
Building Tree Volume and Biomass
Allometric Equations: From Field
Measurement to Prediction. Food and
Agricultural Organization of the
United Nations and Centre de
Coopération Internationale en
Recherche Agronomique pour le
Popescu, Sorin C., Randolph H. Wynne, and
John A. Scrivani. 2004. “Fusion of
Small-Footprint Lidar and
Multispectral Data to Estimate Plot-
Level Volume and Biomass in
Deciduous and Pine Forests in
Virginia, USA.” Forest Science 50 (4):
Pueschel, Pyare, Glenn Newnham, Gilles
Rock, Thomas Udelhoven, Willy
Werner, and Joachim Hill. 2013. “The
Influence of Scan Mode and Circle
Fitting on Tree Stem Detection, Stem
Diameter and Volume Extraction from
Terrestrial Laser Scans.” ISPRS
Journal of Photogrammetry and
Remote Sensing 77 (March): 4456.
Shugart, H. H., S. Saatchi, and F. G. Hall.
2010. “Importance of Structure and Its
Measurement in Quantifying Function
of Forest Ecosystems.” Journal of
Geophysical Research 115
Srinivasan, Shruthi, Sorin C. Popescu, Marian
Eriksson, Ryan D. Sheridan, and Nian-
Wei Ku. 2014. “Multi-Temporal
Terrestrial Laser Scanning for
Modeling Tree Biomass Change.”
Forest Ecology and Management 318
(April): 30417.
Yang, Xiaoyuan, Alan H. Strahler, Crystal B.
Schaaf, David L.B. Jupp, Tian Yao,
Feng Zhao, Zhuosen Wang, et al.
2013. “Three-Dimensional Forest
Reconstruction and Structural
Parameter Retrievals Using a
Terrestrial Full-Waveform Lidar
Instrument (Echidna®).” Remote
Sensing of Environment 135 (August):
3651. doi:10.1016/j.rse.2013.03.020.
Zhao, Feng, Qinghua Guo, and Maggi Kelly.
2012. “Allometric Equation Choice
Impacts Lidar-Based Forest Biomass
Estimates: A Case Study from the
Sierra National Forest, CA.”
Agricultural and Forest Meteorology
165 (November): 6472.
ResearchGate has not been able to resolve any citations for this publication.
Full-text available
The combination of LiDAR and optical remotely sensed data provides unique information about ecosystem structure and function. Here, we describe the development, validation and application of a new airborne system that integrates commercial off the shelf LiDAR hyperspectral and thermal components in a compact, lightweight and portable system. Goddard's LiDAR, Hyperspectral and Thermal (G-LiHT) airborne imager is a unique system that permits simultaneous measurements of vegetation structure, foliar spectra and surface temperatures at very high spatial resolution (approximate to 1 m) on a wide range of airborne platforms. The complementary nature of LiDAR, optical and thermal data provide an analytical framework for the development of new algorithms to map plant species composition, plant functional types, biodiversity, biomass and carbon stocks, and plant growth. In addition, G-LiHT data enhance our ability to validate data from existing satellite missions and support NASA Earth Science research. G-LiHT's data processing and distribution system is designed to give scientists open access to both low- and high-level data products (, which will stimulate the community development of synergistic data fusion algorithms. G-LiHT has been used to collect more than 6,500 km(2) of data for NASA-sponsored studies across a broad range of ecoregions in the USA and Mexico. In this paper, we document G-LiHT design considerations, physical specifications, instrument performance and calibration and acquisition parameters. In addition, we describe the data processing system and higher-level data products that are freely distributed under NASA's Data and Information policy.
Full-text available
We present data from the first five years (2008–2012) of the establishment of the 25.6-ha Smithsonian Conservation Biology Institute (SCBI) Large Forest Dynamics Plot, comprising the initial woody stem census, woody seedling plot surveys, seed rain, and dendrochronological data. The plot is in mature secondary mixed deciduous forest 5 km south of Front Royal, Virginia, USA. The initial plot census enumerated 38 932 free-standing living stems and 29 991 living individuals ≥1 cm dbh comprising 62 species, 38 genera, and 26 families, along with an additional 1248 dead/missing standing stems, for a total of 40 180 stems. Dominant canopy trees include tulip poplar (Liriodendron tulipifera), hickories (Carya spp.), oaks (Quercus spp.), white ash (Fraxinus americana), and black gum (Nyssa sylvatica). Prominent understory components include spicebush (Lindera benzoin), pawpaw (Asimina triloba), American hornbeam (Carpinus caroliniana), witchhazel (Hamamelis virginiana), and eastern redbud (Cercis canadensis). Few species predominate numerically on the plot; seven species have >1000 individuals (71.3% of the total). Mean stand density was 1179 living individuals/ha, while mean basal area was 34.1 m2/ha. Of the total plot area, 4 ha have had white-tailed deer (Odocoileus virginianus) exclusion since 1990. Woody seedling surveys from 2010–2012 in 354 1-m2 plots measured 19 415 seedlings of 47 species, from new germinants up to 1 cm dbh. Community-wide seed rain data from 200 0.5-m2 litterfall traps yielded a total of 9197 records from 37 species. Long-term seed data collected from 1986–2011 for Quercus and Carya within the exclosure and two replicate sites are also presented, documenting considerable annual variation in mast production. Dendrochronological data from 492 tree cores suggested the major canopy trees established circa 1900, but scattered trees of several species existed earlier. Large-scale forest dynamics plots employing standardized methodology have a long, rich history in the tropics. Similar plots in the temperate zone have been largely lacking, however. The SCBI plot represents one of the first of its kind in the Smithsonian Global Earth Observatory's recently established network of such plots, complementing its well-known network of tropical forest plots and enabling comparative studies on forest ecology and climate change at the global scale.
Full-text available
The structure of forests, the three-dimensional arrangement of individual trees, has a profound effect on how ecosystems function and cycle carbon, water, and nutrients. The increased need to understand local to global dynamics of ecosystems, a prerequisite to understand the coupling of the biosphere to other components of Earth systems, has created a demand for extensive ecosystem structure data. Repeated satellite observations of vegetation patterns in two dimensions have made significant contributions to our understanding of the state and dynamics of the global biosphere. Recent advances in remote sensing technology allow us to view the biosphere in three dimensions and provide us with refined measurements of horizontal, as well as vertical, structure of forests. This paper provides an introductory review of the importance of the three-dimensional characterization of terrestrial ecosystem structure of forests and woodlands and its potential measurement from space. We discuss the relevance of these measurements for reducing the uncertainties of terrestrial carbon cycle and the response of ecosystems to future climate. By relating the 3-D structure to forest biomass, carbon content, disturbance characteristics, and habitat diversity, we examine the requirements for future satellite sensors in terms of precision and spatial and temporal resolutions. In particular, we focus this review on measurements from lidar and radar sensors that provide vertical and horizontal characterization of vegetation and are currently recommended for next generation of NASA's Earth observing and European Earth Explorer systems.
Full-text available
Large-scale carbon mapping is needed to support the UNFCCC program to reduce deforestation and forest degradation (REDD). Managers of forested land can potentially increase their carbon credits via detailed monitoring of forest cover, loss and gain (hectares), and periodic estimates of changes in forest carbon density (tons ha−1). Satellites provide an opportunity to monitor changes in forest carbon caused by deforestation and degradation, but only after initial carbon densities have been assessed. New airborne approaches, especially light detection and ranging (LiDAR), provide a means to estimate forest carbon density over large areas, which greatly assists in the development of practical baselines. Here I present an integrated satellite–airborne mapping approach that supports high-resolution carbon stock assessment and monitoring in tropical forest regions. The approach yields a spatially resolved, regional state-of-the-forest carbon baseline, followed by high-resolution monitoring of forest cover and disturbance to estimate carbon emissions. Rapid advances and decreasing costs in the satellite and airborne mapping sectors are already making high-resolution carbon stock and emissions assessments viable anywhere in the world.
Historically, tree biomass at large scales has been estimated by applying dimensional analysis techniques and field measurements such as diameter at breast height (dbh) in allometric regression equations. Equations often have been developed using differing methods and applied only to certain species or isolated areas. We previously had compiled and combined (in meta-analysis) available diameter-based allometric regression equations for estimating total aboveground and component dry-weight biomass for US trees. This had resulted in a set of 10 consistent, national-scale aboveground biomass regression equations for US species, as well as equations for predicting biomass of tree components as proportions of total aboveground biomass. In this update of our published equation database and refinement of our model, we developed equations based on allometric scaling theory, using taxonomic groupings and wood specific gravity as surrogates for scaling parameters that we could not estimate. The new approach resulted in 35 theoretically based generalized equations (13 conifer, 18 hardwood, 4 woodland), compared with the previous empirically grouped 10. For trees from USDA Forest Inventory and Analysis Program (FIA) plots, with forest types grouped into conifers and hardwoods, previous and updated equations produced nearly identical estimates that predicted ∼20 per cent higher biomass than FIA estimates. Differences were observed between previous and updated equation estimates when comparisons were made using individual FIA forest types. © 2013 Institute of Chartered Foresters, All rights reserved. For Permissions, please e-mail: [email protected] /* */
Estimation of the amount of carbon stored in forests is a key challenge for understanding the global carbon cycle, one which remote sensing is expected to help address. However, estimation of carbon storage in moderate to high biomass forests is difficult for conventional optical and radar sensors. Lidar ( li ght d etection and r anging) instruments measure the vertical structure of forests and thus hold great promise for remotely sensing the quantity and spatial organization of forest biomass. In this study, we compare the relationships between lidar-measured canopy structure and coincident field measurements of above-ground biomass at sites in the temperate deciduous, temperate coniferous, and boreal coniferous biomes. A single regression for all three sites is compared with equations derived for each site individually. The single equation explains 84% of variance in above-ground biomass ( P < 0.0001) and shows no statistically significant bias in its predictions for any individual site.
Terrestrial laser scanning (TLS) has been used to estimate a number of biophysical and structural vegetation parameters. Of these stem diameter is a primary input to traditional forest inventory. While many experimental studies have confirmed the potential for TLS to successfully extract stem diameter, the estimation accuracies differ strongly for these studies – due to differences in experimental design, data processing and test plot characteristics. In order to provide consistency and maximize estimation accuracy, a systematic study into the impact of these variables is required. To contribute to such an approach, 12 scans were acquired with a FARO photon 120 at two test plots (Beech, Douglas fir) to assess the effects of scan mode and circle fitting on the extraction of stem diameter and volume. An automated tree stem detection algorithm based on the range images of single scans was developed and applied to the data. Extraction of stem diameter was achieved by slicing the point cloud and fitting circles to the slices using three different algorithms (Lemen, Pratt and Taubin), resulting in diameter profiles for each detected tree. Diameter at breast height (DBH) was determined using both the single value for the diameter fitted at the nominal breast height and by a linear fit of the stem diameter vertical profile. The latter is intended to reduce the influence of outliers and errors in the ground level determination. TLS-extracted DBH was compared to tape-measured DBH. Results show that tree stems with an unobstructed view to the scanner can be successfully extracted automatically from range images of the TLS data with detection rates of 94% for Beech and 96% for Douglas fir. If occlusion of trees is accounted for stem detection rates decrease to 85% (Beech) and 84% (Douglas fir). As far as the DBH estimation is concerned, both DBH extraction methods yield estimates which agree with reference measurements, however, the linear fit based approach proved to be more robust for the single scan DBH extraction (RMSE range 1.39–1.74 cm compared to 1.47–2.43 cm). With regard to the different circle fit algorithms applied, the algorithm by Lemen showed the best overall performance (RMSE range 1.39–1.65 cm compared to 1.49–2.43 cm). The Lemen algorithm was also found to be more robust in case of noisy data. Compared to the single scans, the DBH extraction from the merged scan data proved to be superior with significant lower RMSE’s (0.66–1.21 cm). The influence of scan mode and circle fitting is reflected in the stem volume estimates, too. Stem volumes extracted from the single scans exhibit a large variability with deviations from the reference volumes ranging from −34% to 44%. By contrast volumes extracted from the merged scans only vary weakly (−2% to 6%) and show a marginal influence of circle fitting.
Allometric equation choice impacts lidar-based forest biomass estimates: A case study from the Sierra National Forest, CA a b s t r a c t Regression modeling of biomass estimates from an airborne, multiple return lidar system using regional biomass allometric equations differs significantly from those using national scale Jenkins allometric equa-tions with respect to the amount of variation explained, variables selected and variables of importance. Our discrete return lidar data were collected in September 2007 at 121 plots in a conifer dominated forest site in the Sierra Nevada mountains that include a full range of forest density. We regressed field plot-level estimates of biomass derived from field data and two different allometric equations with a range of lidar metrics. We compared regression performance across eight models: (1) point clouds alone, (2) point clouds with an empirical relationship between DBH and height (i.e., volume), (3) individual tree-level metrics, and (4) all data combined, and across two allometric equations – (A) Forest Inventory Analysis (FIA), and (B) Jenkins. In lower biomass plots, the reference above ground biomass (AGB) esti-mates from regional allometric equations and Jenkins equations were closely related; in plots with large biomass they were different. This finding suggests that published equations from large biomass plots are either not readily available or less represented in national scale allometric equation compiling. Models using reference AGBs calculated from regional allometric equations performed much better than those using reference AGBs calculated from Jenkins allometric equations. In these cases adjusted R 2 improve-ment ranged from 0.07 to 0.11. The regression model that used regional allometric equations with lidar metrics and individual tree data provided the best overall R 2 (0.79) with lowest RMSE suggesting that in most conditions regional biomass equations should be preferred over national equations. The inclu-sion of volumetric metrics shows that lidar variables are more sensitive to the reference AGBs calculated from regional allometric equations, and care should be taken when substituting regional equations using national scale compiled allometric equations in regional biomass studies. In addition, consistent with previous studies, the mean height of individual trees identified was chosen by both models with both reference AGBs calculated from regional allometric equations and those calculated from Jenkins equa-tions, supporting the need to identify individual trees for biomass prediction. Based on these results, we conclude that the selection of allometric equations can influence the capacity of lidar data to estimate biomass significantly, and a careful selection of allometric equations is required for regional lidar biomass studies.