JOURNAL OF GEOPHYSICAL RESEARCH, VOL. 116, G04021, doi:10.

1029/2011JG001708, 2011

Mapping forest canopy height globally with spaceborne lidar

Marc Simard,1 Naiara Pinto,2 Joshua B. Fisher,1 and Alessandro Baccini3
Received 17 March 2011; revised 19 August 2011; accepted 27 August 2011; published 19 November 2011.
[1] Data from spaceborne light detection and ranging (lidar) opens the possibility to map
forest vertical structure globally. We present a wall‐to‐wall, global map of canopy
height at 1‐km spatial resolution, using 2005 data from the Geoscience Laser Altimeter
System (GLAS) aboard ICESat (Ice, Cloud, and land Elevation Satellite). A challenge in
the use of GLAS data for global vegetation studies is the sparse coverage of lidar shots
(mean = 121 data points/degree2 for the L3C campaign). However, GLAS‐derived
canopy height (RH100) values were highly correlated with other, more spatially dense,
ancillary variables available globally, which allowed us to model global RH100 from
forest type, tree cover, elevation, and climatology maps. The difference between the model
predicted RH100 and footprint level lidar‐derived RH100 values showed that error
increased in closed broadleaved forests such as the Amazon, underscoring the challenges
in mapping tall (>40 m) canopies. The resulting map was validated with field
measurements from 66 FLUXNET sites. The modeled RH100 versus in situ canopy height
error (RMSE = 6.1 m, R2 = 0.5; or, RMSE = 4.4 m, R2 = 0.7 without 7 outliers) is
conservative as it also includes measurement uncertainty and sub pixel variability within
the 1‐km pixels. Our results were compared against a recently published canopy height
map. We found our values to be in general taller and more strongly correlated with
FLUXNET data. Our map reveals a global latitudinal gradient in canopy height, increasing
towards the equator, as well as coarse forest disturbance patterns.
Citation: Simard, M., N. Pinto, J. B. Fisher, and A. Baccini (2011), Mapping forest canopy height globally with spaceborne
lidar, J. Geophys. Res., 116, G04021, doi:10.1029/2011JG001708.

1. Introduction requires exploiting the relationship between footprint level

lidar‐derived canopy height estimates and spatially continu-
[2] Forest vertical structure remains poorly characterized
ous ancillary variables, such as data from the Moderate
despite being a predictor of aboveground live biomass Resolution Imaging Spectroradiometer (MODIS).
[Lefsky et al., 2002; Drake et al., 2002; Anderson et al.,
[4] There are numerous approaches to associate the sparse
2006], primary productivity [Thomas et al., 2008], and
lidar footprints with the spatially continuous ancillary vari-
biodiversity [Goetz et al., 2007]. Here, we model forest
ables. One option is to segment or classify the study site to
vertical structure using data from the Geoscience Laser
obtain patches that share a meaningful ecological parameter
Altimeter System (GLAS) aboard ICESat (Ice, Cloud, and
(e.g. age, species composition), such that lidar measurements
land Elevation Satellite). The altimeter transmitted a 1024 nm
can be scaled up to the patch level [i.e., Lefsky et al., 2005a;
light pulse and recorded the reflected signal (waveform)
Boudreau et al., 2008]. A recent global canopy height map
[Zwally et al., 2002]. We use these data to construct a global
[Lefsky, 2010] was produced by segmenting MODIS reflec-
wall‐to‐wall map of forest canopy height.
tance data to delineate forest patches. We argue that those
[3] ICESat/GLAS acquired data globally between 2003 results can be difficult to interpret at coarse resolution given
and 2009. However, lidar shots provide an incomplete cov-
that (i) MODIS images were segmented based on spectral/
erage of the Earth. GLAS footprints were approximately 65 m
textural heterogeneity thresholds that do not readily translate
in diameter, spaced by 170 m along track and several tens of into forest stand properties; (ii) model error can be attributed
kilometers across tracks, a distance that increased in the tro-
to both sub‐pixel (500‐m MODIS) and sub‐patch (1–900
pics. To compound this data dearth, obstruction by clouds
MODIS pixels) variability. As an alternative, we propose to
often hindered coverage. Thus, producing a wall‐to‐wall map
associate selected GLAS shots with 1‐km pixels as we
believe that in the absence of high‐resolution forest distur-
1
Jet Propulsion Laboratory, California Institute of Technology, bance/age maps to serve as segments, the pixel approach
Pasadena, California, USA. deserves evaluation.
2
Department of Geography, University of Maryland, College Park,
Maryland, USA.
[5] We produced a global wall‐to‐wall canopy height map
3
Woods Hole Research Center, Falmouth, Massachusetts, USA. by combining GLAS RH100 estimates and global ancillary
variables. This paper also extends previous work [Lefsky,
Copyright 2011 by the American Geophysical Union. 2010] by validating results against field measurements and
0148‐0227/11/2011JG001708

G04021 1 of 12
G04021 SIMARD ET AL.: FOREST HEIGHT WITH SPACEBORNE LIDAR G04021

considering the impact of sub‐pixel variability on model noise standard deviation. The location of the Gaussian dis-
accuracy. In mapping canopy height globally, we are inter- tributions is constrained to be between the signal beginning
ested in characterizing fine‐scale variability (attributable to and end. Generally, the ground can be determined as the last
disturbance) against a backdrop of generally more coarsely Gaussian peak, which works best in flat areas and open
changing edaphic and climatic gradients. With that goal in canopies. Within closed canopies, locating the ground is
mind, we selected globally available climate, elevation, and sometimes difficult (e.g. when the last peak has low ampli-
vegetation cover layers. This includes the MOD44B percent tude relative to another neighboring peak) [Boudreau et al.,
tree cover product [Hansen et al., 2003] from MODIS, ele- 2008]. It has been shown that using a regression through
vation from the Shuttle Radar Topography Mission, SRTM waveform extent and a terrain index derived from an ancil-
[Farr et al., 2007], as well as climatology maps from the lary DEM can alleviate ground detection issues and improve
Tropical Rainfall Measuring Mission, TRMM [Kummerow canopy height estimates [Lefsky et al., 2005b; Rosette et al.,
et al., 1998] and Worldclim database [Hijmans et al., 2005]. 2008]. However, the regressions may be site specific and
All ancillary variables were resampled to bring the resolution may introduce significant biases. On the other hand, Rosette
of the output wall‐to‐wall map to 1‐km. et al. [2008] obtained reasonable results using the location of
[6] In the context of building a systematic algorithm, the the last Gaussian peak as the ground level and, importantly,
first step was to develop an objective procedure to select found it had the lowest mean error (0.39 m). Since the
waveforms and correct slope‐induced distortions in RH100 regression tree methodology essentially represents overall
[Lefsky et al., 2005b] and calibrate canopy height estimates; trends, it is important that potential bias in the GLAS esti-
RH100 estimates were calibrated with field measurements. mate of canopy height be minimized.
Second, we used a regression tree approach to model lidar [10] After the systematic selection process described in
points from the available ancillary variables. Finally, model section 2.2, the RH100 values were rounded to the nearest
predictions were independently validated against field esti- meter and used as input in the regression tree to produce a
mates from 66 FLUXNET sites distributed globally and wall‐to‐wall canopy height map (see section 2.3). If more
covering a broad range of forest types. than one GLAS shot intersected a 1‐km pixel, all points
[7] Our error analysis accounted for the disparity in scales were used. Instead of locally combining multiple GLAS
between lidar footprints and the resolution of the ancillary shots, model averaging is performed as the last step of the
variables. In our model, lidar shots effectively represented regression tree approach (see section 2.3).
samples of the underlying forest structure, and did not always
intersect the tallest tree within a 1‐km pixel. Conversely, 2.2. Waveform Selection
ancillary variables might not always reflect the environmental [11] We selected the data acquired with laser L3C
conditions at the lidar footprint, but rather comprise an between 2005‐05‐20 and 2005‐06‐23. This campaign was
average over 1‐km2. Due to the scarcity of fine‐resolution chosen due to its temporal overlap with the 2005 MODIS
global vegetation maps, we have examined the impact of sub Percent Tree Cover product (MOD44B).
pixel variability on model error indirectly by looking at two [12] The overall goal of the waveform selection procedure
surrogates of forest heterogeneity: degree of disturbance and was to isolate data points from forested sites while reducing
forest type. First, we hypothesize that our estimates at pro- the impact of slopes and cloud contamination on canopy
tected sites (as defined by the UN World Database on Pro- height estimates. We selected GLAS shots that fell within a
tected Areas, www.wdpa.org) have less error due to less forest class as defined by the Globcover map [Hagolle et al.,
structural variability associated with anthropogenic distur- 2005]. Because the 65 m GLAS footprint samples only a
bance. Second, we hypothesize that across forest types, fraction of the 1‐km2 land cover pixel, not all waveforms are
model error increases with variance in canopy height. from forest canopies. Those cases were problematic given
[8] The resulting wall‐to‐wall map can be downloaded our objective to model the tallest canopies as opposed to
from the web (http://lidarradar.jpl.nasa.gov) and reveals gaps or forest edges. To ensure that shots were reflected
regional canopy height gradients as well as coarse distur- from a forest canopy, we selected waveforms characterized
bance patterns. Our results were compared against another by more than one Gaussian peak, assuming that waveforms
canopy height product [Lefsky, 2010], and differences were with a single peak are due to ground reflection only.
examined in light of the choice of calibration algorithm, [13] GLA14 waveforms were also filtered using engi-
ancillary variables, and modeling procedures. neering and signal parameters to account for cloud cover
and terrain slope. The GLA14 product contains a cloud
2. Methods detection flag that we found to be too stringent. Instead, we
computed the Signal‐to‐Noise Ratio (SNR) to detect
2.1. Canopy Height Estimation waveform hindered by clouds as well as waveforms with
[9] Our analyses were based on the GLA14 land product other unidentified measurement and system issues. To fur-
version 31. Each GLA14 waveform is a fit of the original ther remove waveforms with signal dominated by clouds,
GLAS waveform, modeled by a maximum of 6 Gaussian we selected waveforms that were co‐located close to the
distributions [Brenner et al., 2003]. The GLAS‐derived ground as defined by SRTM (i.e. within 80 m to account for
estimate of canopy height is the waveform metric RH100, forest height and SRTM elevation errors).
defined as the distance between signal beginning and the [14] Terrain slope is the main factor contributing to canopy
location of the lidar ground peak [Harding and Carabajal, height estimation error [Lefsky et al., 2005b; Duncanson
2005; Sun et al., 2008; Boudreau et al., 2008]. In the et al., 2010]. The impact of slope is to broaden the lidar
GLA14 product version 31, the signal beginning is defined waveform, thereby introducing a bias in canopy height
as the location at which the signal is 3.5 times above the estimates. For a large footprint lidar such as GLAS, pulse

2 of 12
G04021 SIMARD ET AL.: FOREST HEIGHT WITH SPACEBORNE LIDAR G04021

Table 1. Ancillary Variables Used in the Regression Tree Approach to Model Canopy Height (RH100) Estimates From GLAS
Variable Dataset Year Resolution Reference
Annual Mean Precipitation (mm) Worldclim 1950–2000 0.00833 Deg Hijmans et al. [2005]
TRMM 2001–2008 0.25 Deg Kummerow et al. [1998]
Precipitation Seasonality Worldclim 1950–2000 0.00833 Deg Hijmans et al. [2005]
100 * SD (mm)/mean (mm) TRMM 2001–2008 0.25 Deg Kummerow et al. [1998]
Annual Mean Worldclim 1950–2000 1 km Hijmans et al. [2005]
Temperature °C * 10
Temperature Seasonality Worldclim 1950–2000 1 km Hijmans et al. [2005]
100 * SD (°C)
Elevation (m) SRTM + GTOPO 2000 1 km U.S. Geological Survey [2006]
Tree Cover (%) MOD44B 2005 500 m Hansen et al. [2003]
Protection Status (7 classes) UN World Database on 2010 Vector format www.wdpa.org
Protected Areas

broadening can be significant. Assuming homogeneous The list of splitting rules is stored in a “tree” object. RF has
canopy height within the footprint, the maximum broad- been shown to outperform other machine learning algo-
ening of the waveform extent can be calculated as the rithms due to its “bagging” strategy [Prasad et al., 2006].
distance between the lowest ground point and the top of Essentially, before growing each tree, RF selects a user‐
the highest tree within the footprint. This implies a potential defined number of input points and subsamples the predictor
bias = Df *tan(slope) [Chen, 2010], where Df is the footprint variables. This procedure is expected to decrease the cor-
diameter. As an example, laser L3 with a ∼65 m footprint relation among individual “trees,” which improves model
may exhibit an 11.5 m canopy height estimation bias over a accuracy [Breiman, 2001]. After the model is run, a single
10 degree slope. prediction is obtained by averaging the predictions from all
[15] To minimize canopy height estimation bias due to “trees.”
terrain slope, we produced a slope map extending between [17] We included 7 ancillary variables in our regression tree
60S and 60N, from the 90 m SRTM elevation data to esti- model, primarily corresponding to climate and vegetation
mate the potential bias. Building on this simplified model, we characteristics (Table 1). Some datasets were combined to
only preserved waveforms located in slopes below 5 degrees achieve global coverage, as follows. The 3B43_V6 Accu-
and for which the necessary bias correction was less than 25% mulated Precipitation product from the Tropical Rainfall
of the measured RH100. For instance, a waveform located on Measurement Mission (TRMM) covers latitudes between
a 5° hill was selected if the original RH100 was greater than −50 and 50. For large latitudes, the precipitation estimates
23 m. We found this step significantly improved predictions were obtained from the Worldclim product (Table 1).
in comparison with an earlier product version [Simard et al., Similarly, the SRTM elevation product is restricted to
2008]. Although, slope compensation should always reduce latitudes −60 and 60, and the coverage was extended using
the measured canopy height (RH100), the random DEM the GTOPO product. We found pixels above 70N that were
errors cause under‐and over‐estimation of slopes (∼±5°). identified as forest sites by the Globcover product, but yet
This generates a random compensation error (∼6 m) that is not covered by the MOD44 map. For these sites, we defined
reduced through averaging within the regression tree model. the percent tree cover as the average percent tree cover for
On the other hand, a heterogeneous canopy within the GLAS each Globcover class, calculated for latitudes 68–70N. All
footprint may be over‐compensated by our simplistic cor- maps were interpolated to 1‐km using majority rule (for
rection model that assumes homogeneous stands at the scale categorical variables) or bilinear interpolation (for continu-
of a GLAS footprint. ous variables).
[18] The RF package used here was implemented in C++
2.3. Producing a Global Wall‐to‐Wall Forest Canopy with a wrapper in R language (R Development Core Team,
Height Map http://www.R-project.org, 2011), and includes both regression
and classification functions. Here, we employ a regression
[16] We employed the regression tree method Random
tree approach to handle a continuous, ordinal response vari-
Forest (RF) [Breiman, 2001] to model RH100 values based
able (RH100). We generated one “RF forest” per Globcover
on global ancillary variables, and to estimate canopy height
class [Hagolle et al., 2005]. We considered 12 classes here,
values for areas not covered by GLAS waveforms. RF has
been formalized in the context of non‐parametric statistics including forest/cropland mosaics but excluding shrublands.
Each “RF forest” contained 500 “RF trees,” and individual
and machine learning algorithms. The method has been
“RF trees” were grown from a subset of 25000 randomly
successfully used in initiatives to map vegetation flood state
[Whitcomb et al., 2009], biomass [Baccini et al., 2008; selected points and 4 predictor variables. The sampling was
performed with replacement, that is, there is some overlap in
Powell et al., 2010] and species distribution [Prasad et al.,
the input data among “RF trees.” The results were mosaicked
2006]. One main advantage is that multiple predictor vari-
to produce the wall‐to‐wall map.
ables can be incorporated without making assumptions
[19] An initial modeling attempt included slope maps
about their statistical distribution or covariance structure. RF
derived from SRTM, but that introduced artifacts in the
iteratively splits the response variable (RH100) into two
results so slope was removed from the list of ancillary vari-
groups. At each split the model chooses the predictor vari-
ables. Inclusion of Normalized Difference Vegetation Index
able that minimizes the within‐group variance in RH100.

3 of 12
G04021 SIMARD ET AL.: FOREST HEIGHT WITH SPACEBORNE LIDAR G04021

[22] The Uganda dataset (above), as well as canopy

height data from the FLUXNET La Thuille database
[Baldocchi, 2008; Baldocchi et al., 2001] were used to
validate the 1‐km wall‐to‐wall map and assess model
error/uncertainty. The FLUXNET database was selected
for three primary reasons: (i) most sites are relatively
homogeneous and flat to 1‐km2, which minimizes error
due to sub pixel variability; (ii) the sites have global
coverage, include most major biome types and climates,
and represent one of the largest available datasets of eco-
system measurements; and, (iii) data quality is generally
high due to the presence of a physical above‐canopy tower
at each site, as well as repeated measurements for these
well‐instrumented sites that require accurate canopy height
for their core flux measurements. The FLUXNET database
contains 475 sites, of which 120 sites contained tree height
data relevant to the ICESat time period, 86 sites had in situ
Figure 1. Signal‐to‐Noise Ratio (SNR) versus RH100 esti- data and were covered by our map. Finally, 66 FLUXNET
mates from GLAS points. sites with vegetation known to be greater than 5 m were
covered by our map. This represents 9 vegetation classes,
primarily dominated by closed (>40%) broadleaved
(NDVI) maps from MODIS, did not improve model accu- deciduous forest (16 sites) and closed (>40%) needleleaved
racy. Furthermore, the model averaging employed by Ran- evergreen forest (15 sites). Some error is introduced
dom Forest creates a trade‐off between overall model because the longitude/latitude for each site was given for
accuracy and ability to predict extreme values. We updated the tower location, but the canopy (e.g., footprint) that the
the model to address this issue with two modifications: (1) by tower instruments measure is often adjacent to the tower,
taking the maximum prediction across all 500 trees, instead of rather than symmetrically surrounding the tower. The accu-
calculating the average prediction; and (2) by developing a racy and method of the in situ canopy height measurements
stratified sampling strategy to ensure that all height classes was generally not reported and non‐uniform, though from
are equally represented in the training data. These two strat- experience working at some of these sites and taking such
egies led to a significant decrease in model accuracy. We measurements the in situ uncertainty typically increases lin-
decided for using the model in its original inception, which early with height from negligible to up to 10% for the tallest
facilitates reproducibility. (>30 m) trees. Method of measurement ranges from: ground‐
based or airborne lidar, laser rangefinder/clinometer, and
2.4. Field Calibration/Validation climbing the tower or tree and measuring distance to ground
[20] We present two forms of field validation—one for the with a tape measure (or directly holding up a tape measure to
individual GLAS estimates of canopy height within the 65 m shorter vegetation). At seven sites, land cover heterogeneity
footprints, and the other for the resultant modeled 1‐km surrounding the site was significant and led to severe pixel
pixels from the wall‐to‐wall map. For clarity, we will refer to contamination. We present results for both with and without
the former as the calibration procedure since it is to ensure those seven “outliers.”
the footprint‐level RH100 measurements correspond to the
top canopy height. On the other hand, comparison of field
data with the resulting canopy height map will be considered
as the validation procedure.
[21] Calibration of the RH100 derived from GLAS
waveforms was based on a set of 98 co‐located field mea-
surements collected in tropical ecosystems in Uganda,
Africa, during September 2008. Each field sample covered
an area of 1600 m2 centered on a GLAS shot. The height of
the three tallest trees within the plot was measured using a
clinometer. To minimize the effect of slope [Lefsky et al.,
2005b; Yang et al., 2011] in the field data, we considered
only GLAS shots on terrain with a slope smaller than 10%
(slope was measured with a clinometer). To minimize the
effect of roughness we limited the area where tree height
was collected and measured tree height only of the 3 tallest
trees within a radius of 20 meters. By doing so, the mea-
surements are derived from the footprint region with highest
gain [Lee et al., 2011]. The height of the tallest tree and the Figure 2. Difference between GLAS ground estimates and
average height of the three tallest trees were compared to the SRTM elevation (m). Very large values (>1000) in general
co‐located footprint level estimates of RH100. had low Signal‐to‐Noise Ratio. These artifacts are likely
due to cloud contamination.

4 of 12
G04021 SIMARD ET AL.: FOREST HEIGHT WITH SPACEBORNE LIDAR G04021

Figure 3. Density of GLAS shots (per degree) used as input in the tree regression procedure.

2.5. Error Analysis which is also based on GLAS data. Although both maps aim
[23] The main objective of the error analysis was to assess at characterizing canopy height globally, there are important
the influence of terrain and sub pixel variability on model differences in terms of canopy height metric and modeling
accuracy. That is, how consistently accurate are predictions in strategy. We calculate and discuss the differences between
disturbed vs. pristine sites, in tall vs. short forest types, and in the two maps in light of the different calibration procedures,
flat vs. steep terrain? For this task, we compared model pre-
dictions against the RH100 data points used to derive the
regression tree. The root mean square error (RMSE) was
calculated for each forest type as:

vffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
uN
uP
u ð^yi  yi Þ2
ti¼1
RMSE ðmÞ ¼
N

Where yi is the canopy height observed from GLAS (RH100)

and ^yi is the modeled canopy height in the wall‐to‐wall map.
[24] In addition, we calculated the percent error for each
GLAS point as:

sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
ð^yi  yi Þ2
Percent Error ¼
yi

A t‐test was used to examine the impact of protection status

(protected or not protected) on percent error. We hypothe-
size that protected areas have lower error due to lower
variability in canopy height within 1 km pixels. Figure 4. Relationship between percent tree cover esti-
mated from MODIS Vegetation Continuous Fields and
2.6. Comparison With the Existing Global Canopy GLAS‐derived canopy height (RH100). The dashed vertical
Height Product lines indicate the range in GLAS RH100, the horizontal line
[25] We compare our wall‐to‐wall estimates of canopy within the boxes corresponds to the median canopy height
height map with the recent product from Lefsky [2010], and the boxes contain 50% of the points.

5 of 12
G04021 SIMARD ET AL.: FOREST HEIGHT WITH SPACEBORNE LIDAR G04021

considering the difference in terrain elevation estimates

between GLAS and SRTM data. Figure 2 shows the
elevation difference between these two datasets. For dif-
ferences greater than a few tens of meters the GLAS
returns likely originated from clouds, explaining the low
SNR values encountered. Finally, only waveforms in
terrain slopes less than 5° and requiring a slope correction
less than 25% of the uncorrected RH100 were selected to
minimize the impact of terrain slope on the regression.
[27] The waveform selection procedure generated a clean
set of 2.5 million data points from the initial 39 million land
shots for campaign L3C (6.4%). The resulting clean GLAS
dataset is globally distributed with a denser sampling in flat,
forested areas such as the temperate and boreal forest, the
Figure 5. Mean GLAS‐derived canopy height (RH100) as Amazon and the Congo Basin (Figure 3).
a function of latitude.
3.2. Global Canopy Height Gradients
ancillary variables, and modeling approaches. We also [28] Canopy height estimates from the filtered GLAS
compare the individual pixels from Lefsky [2010] to the shots revealed two main patterns. RH100 values increases
FLUXNET validation dataset. with MODIS tree cover estimates (Figure 4). In general,
canopy height decreases with latitude and elevation, except
for a peak around 40S (Figure 5). Tall forest stands (in
3. Results Victoria, Tasmania, and New Zealand) had a disproportional
contribution to this latitudinal band, whereas in the North
3.1. Waveform Filtering and Slope Correction canopy height values were averaged across a broader range
[26] We first selected waveforms with multiple peaks of forest types.
that belonged to a Globcover forest class. We found the
GLA14 cloud detection flag removed lidar waveforms 3.3. Wall‐to‐Wall Canopy Height Map
that could still be used for canopy height estimation. On [29] The wall‐to‐wall canopy height map (Figure 6) can
the other hand, the choice of a minimum acceptable SNR be downloaded from http://lidarradar.jpl.nasa.gov. In the
value was easily identified. The plot in Figure 1 shows Amazon basin, the wall‐to‐wall map shows differences in
that low SNR is a good indicator of the measurement canopy height as a function of distance from rivers, as well
quality since the large amount of very large canopy height as edge effects associated with road construction and along
(>50 m) for SNR below 50 is not realistic. Our choice of the Amazonian arc of deforestation (Figure 6, inset). The
a minimum acceptable SNR value can be strengthened by distribution of canopy height values was in general slightly

Figure 6. Wall‐to‐wall map produced by modeling GLAS points with a regression tree approach. The
inset shows a disturbance gradient in the Amazon (color scale was adjusted to increase contrast).

6 of 12
G04021 SIMARD ET AL.: FOREST HEIGHT WITH SPACEBORNE LIDAR G04021

Table 2. RH100 Mean and Standard Deviation (m) for Input

GLAS Points and the Wall‐to‐Wall Canopy Height Map
Mean (m) Standard Deviation (m)
Map values 16.9 8.0
GLAS values 18.1 11.5

narrower in the wall‐to‐wall map as compared to the orig-

inal GLAS points (Table 2), but also showed a right‐skewed
distribution (Figure 7).

3.4. Ground Validation and Error Analysis

[30] Because the 98 points collected in Africa were coin-
cident with the GLAS data, we used them to calibrate the
estimates of canopy height from GLAS waveforms (i.e. local
estimates of RH100) as well as validate the model estimates
(i.e. the final spatially continuous map). The result is shown
in Figure 8 with (a) local RH100 calibration error of 4.1 m
(R2 = 0.84) and (b) wall‐to‐wall model RMSE of 6.6 m (R2 =
0.64). In the latter case, the wall‐to‐wall map tended to
underestimate canopy height of tall forest stands (Figure 8b).
[31] For the FLUXNET validation, the RMSE, coefficient
of determination (R2), and slope of the linear regression
were calculated to assess the deviation between the in situ
measurements and the wall‐to‐wall map. Figure 9 shows
our model has an RMSE of 6.1 m. By removing 7 outliers
with error larger than 10 m (triangles), the RMSE reduces
to 4.4 m with a significant r2 of 0.69.
[32] Model accuracy was also examined by looking at the
difference between the wall‐to‐wall map and the footprint
level GLAS estimates of canopy height. Figure 10 shows the
geographical distribution of model RMSE with respect to
the footprint level GLAS RH100 estimates within a 1° cells.
Mean percent model error was lower in protected sites
(26%, as compared to 32% in non‐protected sites), and this
difference was statistically significant at 0.95 confidence
level (t = 48.98, P < 0.001). Furthermore, RMSE differed
among forest cover classes. Results indicate that model Figure 8. Validation of GLAS and regression tree predicted
accuracy is lower in tall, closed broadleaved forests (a class top canopy height using the African dataset. (a) Field mea-
sured top canopy within GLAS footprint versus GLAS esti-
mate of the canopy height. The estimates of canopy top
height from GLAS shots correspond closely to the field esti-
mates within the footprint (RMSE = 4.1 m and R2 = 0.84).
(b) The regression tree model predicted height versus aver-
age field measured canopy height. The model tends to under-
estimate taller canopies (RMSE = 6.6 m and R2 = 0.64).

that includes the Amazon) and in mosaic land cover classes

(Table 3).

3.5. Comparison With Previous Global Map

of Canopy Height
[33] We compared our canopy height map (Figure 6) with
the Lefsky [2010] map shown with the same color coding
and geographical projection (Geographical, WGS84) in
Figure 11. There are large differences in vegetation cover-
age that can be attributed to our inclusion of land cover
Figure 7. Comparison of the distribution of canopy height classes with low forest cover densities such as mosaic crops,
values from GLAS points and the wall‐to‐wall map. open forest, and saline flooded forests. In addition, open

7 of 12
G04021 SIMARD ET AL.: FOREST HEIGHT WITH SPACEBORNE LIDAR G04021

boreal zone, which is dominated by open needleleaf decid-

uous or evergreen forest our map has a canopy average 7 m
(±8.3 m) taller. However, this land cover class is not fully
covered in the Lefsky [2010] map, which greatly contributes
to this mean difference.
[35] In areas where the Lefsky [2010] map shows taller
canopies (Figure 12), we noticed a general spatial pattern
that corresponds to regions of severe topography. However,
we found no correlation between slope steepness and the
amount of difference between the map estimates. Instead, to
verify this observation, we plotted the cumulative distribu-
tion of the area where the Lefsky [2010] map is taller than
ours as a function of terrain slope (Figure 13). Terrain slope
was computed using SRTM elevation dataset. The plot
shows that around 40% of these areas occur in terrain where
slopes are steeper than 5°. We can only conclude that the
Lefsky [2010] map generally has taller forests in the pres-
ence of topography.
[36] The difference in height between the Lefsky [2010]
estimates and FLUXNET measurements should not be
Figure 9. Comparison of derived heights and field mea- impacted by topography since sites are generally flat. How-
sured height from the FLUXNET dataset. ever, we can expect to see a bias due to the different metrics
that will not affect the r2 in the FLUXNET validation. We
found poor correspondence in the Lefsky [2010] map with
needleleaf deciduous or evergreen forest and closed to the FLUXNET validation data (r2 = 0.01, RMSE = 9.6 m;
open mixed broadleaf and needleleaf forest, are only par- with 17 “outliers” removed: r2 = 0.26, RMSE = 6.1 m)
tially covered in the Lefsky [2010] map, but fully covered (Figure 14).
in our map.
[34] We calculated the difference between the two pro-
ducts, and found our canopy height estimates to be in 4. Discussion
general taller though not everywhere (Figure 12). This is [37] The global distribution of RH100 estimates from
partially explained since we modeled the top canopy height processed GLAS waveforms revealed a latitudinal gradi-
instead of Lorey’s height, which is a tree‐size weighted ent in which canopy height peaks at low latitudes and in
mean height. For closed to open broadleaf evergreen or Southern temperate rainforests (Figure 5). The peak
semi‐deciduous forest, which dominates the tropical belt, around latitude 40S coincides with forests in SE Australia
our estimates are on average 12 m (±8.9 m) taller. In the and Tasmania that harbor the species Eucalyptus regnans,

Figure 10. Model RMSE with respect to RH100 estimated from GLAS shots within 1° cells.

8 of 12
G04021 SIMARD ET AL.: FOREST HEIGHT WITH SPACEBORNE LIDAR G04021

Table 3. Mean and Standard Deviation by Vegetation Type With Model RMSE Computed From Individual GLAS
Footprint
Class Namea Mean RH100 (m) SD RH100 (m) RMSE (m)
Mosaic Cropland/Vegetation 14.0 9.9 6.0
Mosaic Vegetation/Cropland 13.1 9.6 6.0
Closed to Open Broadleaved or Semi‐Deciduous Forest 29.9 11.7 7.8
Closed Broadleaved Deciduous Forest 19.0 8.9 6.4
Open Broadleaved Deciduous Forest/Woodland 13.0 7.0 4.4
Closed Needleleaved Evergreen Forest 20.3 9.8 5.7
Open Needleleaved Deciduous or Evergreen Forest 17.2 8.4 5.8
Closed to Open Mixed Forest 19.8 8.8 5.4
Mosaic Forest or Shrubland/Grassland 12.8 8.7 5.2
Mosaic Grassland/Forest or Shrubland 12.4 8.3 4.5
Closed to Open Broadleaved Forest Regularly 26.1 10.8 5.2
Flooded – Fresh or Brackish Water
Closed Broadleaved Forest or Shrubland Permanently 16.0 9.1 3.8
Flooded – Saline or Brackish Water
a
Abridged from Hagolle et al. [2005].

one of the world’s tallest flowering plants [Keith et al., [39] Canopy height values were aggregated when GLAS
2009]. shots fell in the same forest patch in the Lefsky [2010] map.
[38] A few issues complicate the comparison of our map In our case, predicted canopy heights were averaged during
and the existing canopy height map [Lefsky, 2010]. Our the tree regression procedure. Thus both products represent
product (Figure 6) maps the top of the canopy as defined by aggregated canopy height measures, while differing with
the waveform percentile metric RH100. Lefsky [2010] has respect to the fundamental unit and the stage at which the
mapped Lorey’s Height (Figure 11), which is empirically aggregation is performed. In addition, the land cover prod-
derived from waveform shape parameters and is expected to uct used to delineate forest types improved our results. We
be lower than RH100. At the lidar footprint level, Lorey’s used more inclusive forest categories from [Hagolle et al.,
Height shows a robust relationship with aboveground bio- 2005] that included mosaics, thus our map has a broader
mass (Mg/ha) [Lefsky, 2010]. On the other hand RH100 is coverage. On the other hand, the use of an ecoregion map
a more direct measurement, which can also be related to [Olson et al., 2001] in the Lefsky [2010] product likely
biomass and has been used in ecosystem dynamics model- allowed the identification of extreme canopy height values
ing [Thomas et al., 2008]. Beyond differences in canopy associated with particular associations (for example, Red
height metrics, product utility will also depend on the Woods in the Northwest USA). The observed difference in
sampling and prediction errors that incur during the gener- regions with severe topography (Figure 13) indicates that
ation of the wall‐to‐wall maps. the impact of topography on large footprint lidar estimates

Figure 11. Global forest height by Lefsky [2010] using matching color table and coordinate system (i.e.
geographical, WGS84).

9 of 12
G04021 SIMARD ET AL.: FOREST HEIGHT WITH SPACEBORNE LIDAR G04021

Figure 12. Canopy height difference between forest height maps from this study and Lefsky [2010].

of height remains an open issue. In our case, the errors pared to the Lefsky [2010] map. This is expected given that
introduced by terrain slope were mostly removed by the Lefsky [2010] map models Lorey’s height and under-
excluding GLAS shots that required significant slope cor- scores the fact that the two products have different ecolog-
rection and through the model’s averaging process. How- ical meanings and potential applications (Figure 9), although
ever, the GLAS shot selection process means our model more field data are necessary to perform a better comparison
assumes canopy height for a given forest type and cover, of the two products. We argue that our use of a constant
whether on steep or flat terrain, is the same if it has the same resolution (1‐km) facilitates efforts to quantify the impact of
climatic conditions, elevation, and percent tree cover. Any sub pixel variability on model error. Indeed, a recent paper
residual error due to terrain slope is included in the vali- mapping forest carbon stocks (Mg/ha) over the tropics
dation with the FLUXNET data. [Saatchi et al., 2011] exploits the footprint‐level Lorey’s
[40] Our product shows a better correspondence with Height estimates from Lefsky [2010] but models biomass
unweighted canopy height from FLUXNET sites, as com- over a 1‐km grid. Lefsky [2010] argues for the use of patches,

Figure 13. Cumulative distribution of points where esti-

mates from the Lefsky [2010] map are larger than the ones
reported in this study as a function of terrain slope. Areas Figure 14. Comparison of Lefsky [2010] canopy height
where estimates of canopy height from this study are taller map versus FLUXNET dataset (same field measurements
than Lefsky’s are not shown here. used in Figure 9).

10 of 12
G04021 SIMARD ET AL.: FOREST HEIGHT WITH SPACEBORNE LIDAR G04021

not pixels, as the fundamental unit used to model lidar and/or forest age maps. During our work, canopy height
points. We agree there is not a single optimum resolution to data from FLUXNET sites emerged as a critical dataset for
sample forests given variation in canopy geometry, eleva- validation of forest structure maps, enabling comparison
tion, and successional stage. But there is also a constellation across a wide range of forest biomes. We recommend that
of clustering algorithms that can be used to delineate forest FLUXNET canopy height measurements (e.g. top, mean and
patches. Ultimately, the question is whether these can identify Lorey’s canopy height) be standardized and expanded to all
meaningful ecological units such as even‐aged stands [e.g., forest sites. In addition, this dataset should be centralized
Lefsky et al., 2005a]. Alternatively if units are small enough, and distributed publically over the internet to facilitate the
their area is comparable to the lidar footprint [e.g., Kimes validation and comparison of future maps by international
et al., 2006] such that searching for homogeneous areas may investigators. An 890 MB GeoTIFF file of the global 1‐km
not be necessary given dense lidar coverage. canopy height map for public download can be found on the
[41] Model error when comparing predictions against website http://lidarradar.jpl.nasa.gov/. Users are encouraged
relatively homogeneous forest stands near FLUXNET tow- to contact the first author and provide feedback about the
ers and field plots in Africa was ∼6.1 m (or 4.4 m without product.
outliers) and 6.6 m respectively. While these results limit
predictions at the local scale, the map produced here con-
tains one of the best descriptions of forest vertical structure at [43] Acknowledgments. The research was carried out at the Jet Pro-
pulsion Laboratory, California Institute of Technology, under a contract
regional and global scales currently available. For example, with the National Aeronautics and Space Administration. Funded through
regional patterns were observed in the Amazonian arc of the MEaSUREs program (project WBS 547714.04.14.01.13). N. Pinto
deforestation, where smaller trees were predicted along for- was funded by the NASA Postdoctoral Program (NPP). The authors would
like to thank fruitful discussions with Claudia Carabajal, Mike Kobrick,
est edges. There are still limits to our ability to capture tall Ralph Dubayah as well as the anonymous reviewers for their constructive
(>40 m) canopy height values at 1‐km resolution (Figures 7 comments. M. Simard is funded by the NASA MEaSUREs program. Can-
and 8b). Model accuracy was mainly influenced by (i) sat- opy height data were provided to J. Fisher by the PI’s part of FLUXNET.
uration of the relationship between canopy height and the
ancillary variables and (ii) sub pixel variability. The relative References
contribution of these factors likely influenced the differences
Anderson, J., M. E. Martin, M. L. Smith, R. O. Dubayah, M. A. Hofton,
in model accuracy across forest types (Table 3). In mosaic P. Hyde, B. E. Peterson, J. B. Blair, and R. G. Knox (2006), The use of
classes, lidar points are more likely to miss tall trees and the waveform lidar to measure northern temperate mixed conifer and decid-
mismatch between lidar‐derived estimates and the 1‐km uous forest structure in New Hampshire, Remote Sens. Environ., 105,
248–261, doi:10.1016/j.rse.2006.07.001.
ancillary variables is more pronounced. Tall mature tropical Baccini, A., N. Laporte, S. J. Goetz, M. Sun, and H. Dong (2008), A first
forests, on the other hand, present a challenge due to the map of tropical Africa’s above‐ground biomass derived from satellite
saturation of tree cover with canopy height (Figure 4) and the imagery, Environ. Res. Lett., 3, 045011, doi:10.1088/1748-9326/3/4/
broader range of tree heights (Table 3 and Figure 8b), 045011.
Baldocchi, D. (2008), “Breathing” of the terrestrial biosphere: Lessons
meaning lidar samples might not always intersect emergent learned from a global network of carbon dioxide flux measurement sys-
trees. The geographical distribution of model RMSE as tems, Aust. J. Bot., 56, 1–26, doi:10.1071/BT07151.
compared to the RH100 derived from GLAS (Figure 10), Baldocchi, D., et al. (2001), FLUXNET: A new tool to study the temporal
and spatial variability of ecosystem‐scale carbon dioxide, water vapor,
which is larger in forested areas, also indicates the spatial and energy flux densities, Bull. Am. Meteorol. Soc., 82, 2415–2434,
resolution of the input layers may not be sufficient to capture doi:10.1175/1520-0477(2001)082<2415:FANTTS>2.3.CO;2.
canopy heterogeneities. As expected, model accuracy was Boudreau, J., R. F. Nelson, H. A. Margolis, A. Beaudoin, L. Guindon, and
D. Kimes (2008), An analysis of regional forest biomass using airborne
higher in protected forests, underscoring the challenges in and spaceborne lidar in Quebec, Remote Sens. Environ., 112, 3876–3890,
modeling forest structure in the presence of anthropogenic doi:10.1016/j.rse.2008.06.003.
disturbance. Future initiatives to model vertical canopy Breiman, L. (2001), Random forests, Mach. Learn., 45, 5–32, doi:10.1023/
structure globally will have to address these sampling design A:1010933404324.
Brenner, A. C., et al. (2003), Derivation of range and range distributions from
issues. One question is whether fine‐resolution continental laser pulse waveform analysis for surface elevations, roughness, slope,
mosaics (Landsat, ALOS/PALSAR) can be used to model and vegetation heights, algorithm theoretical basis document, version 3.0,
fine‐scale disturbances and improve these results. pp. 33–36, NASA Goddard Space Flight Cent., Greenbelt, Md.
Chen, Q. (2010), Retrieving canopy height of forests and woodlands over
mountainous areas in the Pacific coast region using satellite laser altimetry,
Remote Sens. Environ., 114, 1610–1627, doi:10.1016/j.rse.2010.02.016.
5. Conclusions Drake, J. B., R. O. Dubayah, R. G. Knox, D. Clark, and J. B. Blair
[42] We presented a systematic approach to select, pro- (2002), Sensitivity of large‐footprint lidar to canopy structure and bio-
mass in a neotropical rainforest, Remote Sens. Environ., 81, 378–392,
cess, and model lidar‐derived canopy height estimates doi:10.1016/S0034-4257(02)00013-5.
from GLAS to produce global wall‐to‐wall canopy height Duncanson, L. I., K. O. Niemann, and M. A. Wulder (2010), Estimating
maps. The prediction map shows a reasonable correspon- forest canopy height and terrain relief from GLAS waveform metrics,
Remote Sens. Environ., 114, 138–154, doi:10.1016/j.rse.2009.08.018.
dence with 66 FLUXNET sites (RMSE = 6.1 m, R2 = 0.49; Farr, T. G., et al. (2007), The Shuttle Radar Topography Mission, Rev.
RMSE = 4.4 m and R2 = 0.69 without outliers). A compar- Geophys., 45, RG2004, doi:10.1029/2005RG000183.
ison between the model and the input lidar points shows that Goetz, S. J., D. Steinberg, R. O. Dubayah, and J. B. Blair (2007), Laser
remote sensing of canopy habitat heterogeneity as a predictor of bird spe-
error differs across forest types and increases as a function of cies in an eastern temperate forest, USA, Remote Sens. Environ., 108,
canopy height. The results shown here have implications for 254–263, doi:10.1016/j.rse.2006.11.016.
the design of active‐passive sensor fusion algorithms for Hagolle, O., A. Lobo, P. Maisongrande, F. Cabot, B. Duchemin, and
mapping forest vertical structure. We expect that maps with A. De Pereyra (2005), Quality assessment and improvement of temporally
composited products of remotely sensed imagery by combination of
error less than 6 m can be obtained when lidar‐derived canopy VEGETATION 1 and 2 images, Remote Sens. Environ., 94, 172–186,
height estimates are integrated with fine‐resolution maps, doi:10.1016/j.rse.2004.09.008.

11 of 12
G04021 SIMARD ET AL.: FOREST HEIGHT WITH SPACEBORNE LIDAR G04021

Hansen, M., R. S. DeFries, J. R. G. Townshend, M. Carroll, C. Dimiceli, mass dynamics with Landsat time‐series and field inventory data: A com-
and R. A. Sohlberg (2003), Global percent tree cover at a spatial resolu- parison of empirical modeling approaches, Remote Sens. Environ., 114,
tion of 500 meters: First results of the MODIS vegetation continuous 1053–1068, doi:10.1016/j.rse.2009.12.018.
fields algorithm, Earth Interact., 7, 1–15, doi:10.1175/1087-3562 Prasad, A. M., L. R. Iverson, and A. Liaw (2006), Newer classification and
(2003)007<0001:GPTCAA>2.0.CO;2. regression tree techniques: Bagging and random forests for ecological
Harding, D. J., and C. C. Carabajal (2005), ICESat waveform measure- prediction, Ecosystems, 9, 181–199, doi:10.1007/s10021-005-0054-1.
ments of within‐footprint topographic relief and vegetation vertical struc- Rosette, J. A. B., P. R. J. North, and J. C. Suárez (2008), Vegetation height
ture, Geophys. Res. Lett., 32, L21S10, doi:10.1029/2005GL023471. estimates for a mixed temperate forest using satellite laser altimetry, Int.
Hijmans, R. J., S. E. Cameron, J. L. Parra, P. G. Jones, and A. Jarvis J. Remote Sens., 29, 1475–1493, doi:10.1080/01431160701736380.
(2005), Very high resolution interpolated climate surfaces for global land Saatchi, S., et al. (2011), Benchmark map of forest carbon stocks in tropical
areas, Int. J. Climatol., 25, 1965–1978, doi:10.1002/joc.1276. regions across three continents, Proc. Natl. Acad. Sci. U. S. A., 108,
Keith, H., D. B. Mackey, and D. B. Lindenmayer (2009), Re‐evaluation of 9899–9904, doi:10.1073/pnas.1019576108.
forest biomass carbon stocks and lessons from the world’s most carbon‐ Simard, M., N. Pinto, and S. Riddick (2008), Characterizing vegetation
dense forests, Proc. Natl. Acad. Sci. U. S. A., 106, 11,635–11,640. 3D structure globally using spaceborne lidar and radar, Eos Trans.
Kimes, D. S., K. J. Ranson, G. Sun, and J. B. Blair (2006), Predicting lidar AGU, 89(53), Fall Meet. Suppl., Abstract B41C‐0402.
measured forest vertical structure from multi‐angle spectral data, Remote Sun, G., K. J. Ranson, D. S. Kimes, J. B. Blair, and K. Kovacs (2008),
Sens. Environ., 100, 503–511, doi:10.1016/j.rse.2005.11.004. Forest vertical structure from GLAS: An evaluation using LVIS and
Kummerow, C., W. Barnes, T. Kozu, J. Shiue, and J. Simpson (1998), The SRTM data, Remote Sens. Environ., 112, 107–117, doi:10.1016/j.
Tropical Rainfall Measuring Mission (TRMM) sensor package, J. Atmos. rse.2006.09.036.
Oceanic Technol., 15, 809–817, doi:10.1175/1520-0426(1998) Thomas, R. Q., G. C. Hurtt, R. Dubayah, and M. H. Schilz (2008), Using
015<0809:TTRMMT>2.0.CO;2. lidar data and height‐structured ecosystem model to estimate forest car-
Lee, S., W. Ni‐Meister, W. Yang, and Q. Chen (2011), Physically based bon stocks and fluxes over mountainous terrain, Can. J. Rem. Sens., 34,
vertical vegetation structure retrieval from ICESat data: Validation using S351–S363, doi:10.5589/m08-036.
LVIS in White Mountain National Forest, New Hampshire, USA, U.S. Geological Survey (2006), Shuttle Radar Topography Mission, 30 Arc
Remote Sens. Environ., 115, 2776–2785, doi:10.1016/j.rse.2010.08.026. Second product, version 2.0, http://www.glcf.umd.edu/data/srtm/, Global
Lefsky, M. A. (2010), A global forest canopy height map from the Land Cover Facil., Univ. of Md., College Park, Md.
Moderate Resolution Imaging Spectroradiometer and the Geoscience Whitcomb, J., M. Moghaddam, K. McDonald, J. Kellndorfer, and E. Podest
Laser Altimeter System, Geophys. Res. Lett., 37, L15401, doi:10.1029/ (2009), Mapping vegetated wetlands of Alaska using L‐band radar satel-
2010GL043622. lite imagery, Can. J. Remote Sens., 35, 54–72, doi:10.5589/m08-080.
Lefsky, M. A., W. B. Cohen, D. J. Harding, G. G. Parker, S. A. Acker, and Yang, W., W. Ni‐Meister, and S. Lee (2011), Assessment of the impacts of
T. Gower (2002), Lidar remote sensing of above‐ground biomass in three surface topography, off‐nadir pointing and vegetation structure on vege-
biomes, Glob. Ecol. Biogeogr., 11, 393–399, doi:10.1046/j.1466-822x. tation lidar waveforms using an extended geometric optical and radiative
2002.00303.x. transfer model, Remote Sens. Environ., 115, 2810–2822, doi:10.1016/j.
Lefsky, M. A., D. P. Turner, M. Guzy, and W. B. Cohen (2005a), Combin- rse.2010.02.021.
ing lidar estimates of aboveground biomass and Landsat estimates of Zwally, H. J., et al. (2002), ICESat’s laser measurements of polar ice, atmo-
stand age for spatially extensive validation of modeled forest productiv- sphere, ocean, and land, J. Geodyn., 34, 405–445, doi:10.1016/S0264-
ity, Remote Sens. Environ., 95, 549–558, doi:10.1016/j.rse.2004.12.022. 3707(02)00042-X.
Lefsky, M. A., D. J. Harding, M. Keller, W. B. Cohen, C. C. Carabajal,
F. B. Del Born Espirito‐Santo, M. O. Hunter, and R. Oliveira (2005b), A. Baccini, Woods Hole Research Center, 149 Woods Hole Rd.,
Estimates of forest canopy height and aboveground biomass using
Falmouth, MA 02540, USA.
ICESat, Geophys. Res. Lett., 32, L22S02, doi:10.1029/2005GL023971. J. B. Fisher and M. Simard, Jet Propulsion Laboratory, California
Olson, D., et al. (2001), Terrestrial ecoregions of the world: A new map of Institute of Technology, 4800 Oak Grove Dr., Pasadena, CA 91109,
life on earth, BioScience, 51, 933–938, doi:10.1641/0006-3568(2001)051 USA. ([email protected])
[0933:TEOTWA]2.0.CO;2. N. Pinto, Department of Geography, University of Maryland, 2181
Powell, S., W. B. Cohen, S. P. Healey, R. E. Kennedy, G. G. Moisen, K. B.
LeFrak Hall, College Park, MD 20742, USA.
Pierce, and J. L. Ohmann (2010), Quantification of live aboveground bio-

12 of 12