Forest Ecology and Management 293 (2013) 149–160

Contents lists available at SciVerse ScienceDirect

Forest Ecology and Management

journal homepage: www.elsevier.com/locate/foreco

Forest biomass patterns across northeast China are strongly shaped by forest height
Xiangping Wang a,b,⇑, Shuai Ouyang a, Osbert Jianxin Sun a, Jingyun Fang b
a
The Key Laboratory of Silviculture and Conservation of the Ministry of Education, College of Forestry, Beijing Forestry University, Beijing 100083, China
b
Department of Ecology, College of Environmental Sciences, and Key Laboratory for Earth Surface Processes of the Ministry of Education, Peking University, Beijing 100871, China

a r t i c l e i n f o a b s t r a c t

Article history: The emergency of satellite-borne light detecting and ranging (LiDAR) technology in recent years have
Received 3 August 2012 provided a promising way to monitor worldwide patterns of forest biomass and carbon sources/sinks.
Received in revised form 28 December 2012 However, few studies have examined the roles of various abiotic and biotic factors in modulating the rela-
Accepted 2 January 2013
tionship between forest biomass and height at a large scale. This is important given the growing depen-
dence on LiDAR derived forest height as a predictor of forest biomass. In this analysis, we used 529 plots
across northeast China to examine this question, and to explore the method to estimate forest biomass
Keywords:
from height. Our results showed that, while forest height and average tree height showed close relation-
Forest biomass
Forest (tree) height
ships with stand biomass or mean biomass per stem (R2 between 0.57 and 0.78), stand biomass could not
Belowground biomass be reliably predicted with two methods based on average tree height. In contrast, when the effects of cli-
Canopy-tree life form mate and forest groups were included in the models, forest height could predict biomass patterns with a
Northeast China R2 between 0.74 (belowground) and 0.91 (total biomass), which was comparable to the widely accepted
biomass expansion factor method (R2 between 0.72 and 0.98). We also showed that the ratio of both
aboveground and belowground biomass to forest height (B/H ratio) was roughly similar at a large scale,
suggesting that forest biomass patterns are strongly shaped by forest height. However, B/H ratio showed
significant difference between deciduous and evergreen forests across northeast China. The life form of
canopy trees was the major factor modulating the relationships between stand biomass and forest height,
while climate, forest type and forest origin played a secondary role. Our results strongly support the use
of LiDAR to monitor the large-sale patterns of both above and belowground forest biomass. Our analysis
also found that the lack of forest height information in previous literatures has caused most of the bio-
mass data could not be utilized to estimate biomass patterns from height, and we advocate future anal-
yses to report forest height together with field-observed biomass.
Ó 2013 Elsevier B.V. All rights reserved.

1. Introduction
Forest ecosystems accounted for a majority of global terrestrial
carbon pool (over 80% of aboveground carbon, Dixon et al., 1994),
and might become a major carbon source as a result of rapid land
use change and climatic warming (Houghton, 2005; IPCC, 2007).
Consequently, estimating the large scale biomass patterns accu-
rately is crucial for monitoring forest carbon source and sink
dynamics worldwide (e.g. Fang et al., 2001; Schimel et al., 2001).
It is widely accepted that estimating forest biomass from stock
volume (the biomass expansion factor method) is the most accu-
rate method for large scale biomass estimation (e.g. Fang et al.,
2001; Brown, 2002). However, this method also has some limita-
tions: (1) it involves intensive investigations of large amount of
widely-distributed field plots, which is time and efforts consuming.
⇑ Corresponding author at: College of Forestry, Beijing Forestry University, 35
East Qinghua Road, Haidian, Beijing 100083, China. Tel.: +86 10 6233 6985; fax: +86
10 6233 6400.
E-mail address: wangxiangping@bjfu.edu.cn (X. Wang).
(2) Many countries or regions have not conducted forest inventory
yet (Houghton, 2005; Massada et al., 2006) and thus this method
cannot be applied. (3) The time and spatial resolutions of forest
inventory data were generally coarse. For instance, forest inventory
data were reported on the province (or county) basis every 5 years
in China (Fang et al., 2001). However, forest biomass is strongly af-
fected by disturbances (e.g. Wang et al., 2008; Simard et al., 2011).
To capture the effects of disturbance and tree growth on spatial
biomass patterns (for a more accurate monitoring of carbon sink/
source), Houghton (2005) suggested that forest biomass are better
to be monitored at a resolution of 25–250 m. Consequently, a more
time and efforts saving method is needed to map forest biomass at
finer time and spatial resolutions, especially when some important
forest regions in the world (e.g. tropic and temperate forests) are
undergoing rapid deforestations (Kauppi et al., 2006; IPCC, 2007).
The progresses of light detecting and ranging (LiDAR) tech-
niques in the last two decades have provided a promising approach
for this purpose. Studies with airborne LiDAR have shown that var-
ious forest structure parameters (e.g. forest height, stem density,
and average tree height), as well as stand biomass and volume,

0378-1127/$ - see front matter Ó 2013 Elsevier B.V. All rights reserved.
http://dx.doi.org/10.1016/j.foreco.2013.01.001
150 X. Wang et al. / Forest Ecology and Management 293 (2013) 149–160
could be estimated with LiDAR satisfactorily (e.g. Dubayah and
Drake, 2000; Lefsky et al., 2002; Drake et al., 2003; van Leeuwen
and Nieuwenhuis, 2010). Recently, satellite-borne LiDAR data are
also available, and have promoted the mapping of forest height
from regional (Lefsky et al., 2005) to global scales (Lefsky, 2010;
Simard et al., 2011; Los et al., 2012). These forest height maps pro-
vided an ideal opportunity for estimating large-scale forest bio-
mass patterns based on the relationship between biomass and
forest height (e.g. Dubayah and Drake, 2000; Lefsky et al., 2005;
Fang et al., 2006). However, because most previous LiDAR studies
were based on airborne LiDAR at relative small scales, the large
scale biomass–height relationships have seldom been examined
in details.
We believe that such an examination is critical for monitoring
the large-scale dynamics of forest biomass. If biomass–height rela-
tionships showed great differences both across environmental gra-
dients and among forest types, then it is not only necessary to
construct biomass–height relationships for each forest type by
each region (e.g. Drake et al., 2003), but also required to monitor
the distribution of each forest type (e.g. with remote sensing data).
On the other hand, it might be that not all the factors affecting bio-
mass–height relationship are really important for large-scale bio-
mass estimations. For instance, if biomass–height relationship
differed only between some coarsely classified forest groups (e.g.
conifer vs. broadleaf, or deciduous vs. evergreen forests) at a large
scale, and did not differ significantly among forest types within
these coarse-defined forest groups (e.g. Drake et al., 2002; Chen,
2010), then the efforts in classifying forest types from remote sens-
ing images could be greatly reduced (which should be conducted
regularly to monitor forest distribution). As a result, identify the
major factors modulating biomass–height relationship is especially
important for monitoring forest biomass with LiDAR technologies
at the large to global scales.
In this study, we used 529 forest plots (either measured by us or
complied from literatures) across northeast China to explore the
large-scale relationship between forest biomass and height. There
are many data on field-observed biomass and tree height in litera-
tures (e.g. Cannell, 1982; Fang et al., 2006; Wang et al., 2008). If
these data can be utilized to develop models for estimating bio-
mass patterns from height, then a great amount of efforts in the
field investigation of forest biomass can be avoided. Consequently,
we also tested which kind of literature data could be utilized for
this purpose. Specifically, we examined four questions as follows.
(1) What are the major factors regulating the large-scale bio-
mass–height relationship (see above)?
(2) Can belowground stand biomass be well predicted from for-
est height? Belowground biomass is a major source of uncer-
tainties in large-scale biomass estimation, and has long been
a great challenge (Cairns et al., 1997; Brown, 2002). Previous
studies have shown that forest biomass belowground are
closely related with aboveground biomass (e.g. Mokany
et al., 2005; Wang et al., 2008), while aboveground biomass
is closely related to forest height (e.g. Dubayah and Drake,
2000; Lefsky et al., 2002, 2005; Fang et al., 2006). Thus we
hypothesized that biomass belowground may also be closely
related to forest height. If estimating belowground biomass
from forest height is possible, then the satellite-borne LiDAR
will provide an unprecedented opportunity for monitoring
global forest carbon belowground.
(3) Can forest biomass be well estimated from average tree
height? Forest height and maximum tree height are the
most easy parameters that could be retrieved from LiDAR
data (Drake et al., 2003; Lefsky, 2010; van Leeuwen and Nie-
uwenhuis, 2010; Simard et al., 2011). However, most litera-
tures on filed-observed stand biomass reported only average
tree height (for both canopy trees and trees under canopy).
This situation will greatly reduce the data available for con-
structing large scale biomass–height models. Recent studies
have shown that estimating average height for both canopy
and under-canopy trees from LiDAR data is possible (Malt-
amo et al., 2004; Lee and Lucas, 2007). Consequently, we
also tested the possibility to estimate stand biomass from
average tree height. If this is possible, then the abundant
biomass data in the literatures could also be utilized by
LiDAR studies.
(4) Why stand biomass can be estimated from forest height?
While previous studies have focused on predicting forest
biomass from LiDAR retrieved forest height (Lefsky et al.,
2005; van Leeuwen and Nieuwenhuis, 2010), the biological
bases for estimating biomass from height remains less atten-
tions. Fang et al. (2006) showed that the ratio of above-
ground biomass to forest height (i.e. biomass per cubic
meter of forest space) is similar in East Asia, Europe, USA
and Canada, and proposed that the differences in biomass
per area among continents was largely caused by difference
in forest height. Here we tested whether this hypothesis
could be applied to explain the geographic forest biomass
patterns within a huge region in East Asia, using forest plots
well distributed across northeast China.
2. Materials and methods
2.1. Study area and data
The northeast part of China is defined here using the same def-
inition as our previous study (Wang et al., 2008), i.e. including Hei-
longjiang, Jilin and Liaoning provinces, eastern Inner Mongolia
Autonomous Region, and northern Hebei province, covering an
area of ca. 1,600,000 km2. The topography, climate and forest veg-
etations of the study area have been described in Wang et al.
(2008) and thus not detailed here.
We constructed a database for forest biomass, which consist of
641 plots across northeast China. In Wang et al. (2008) there were
515 plots, and 126 plots were appended into the database since
then. Our database documented the following information for each
plot: (1) stand biomass (total, above and belowground); (2) geo-
graphic ordinations (latitude, longitude and altitude); (3) climate
variables, including mean annual and growing season (months
with mean temperature P5 °C) temperature and precipitation,
etc.; (4) forest structure parameters, including diameter at breast
height (DBH), forest height (or maximum tree height) and average
tree height, stem density and stand volume, etc.; (5) forest type,
dominant tree species, and forest origin (primary/secondary/
planted forest). Most of the plots (475) were compiled from litera-
tures and thus some of the stand structure or biomass variables
might be not available. Methods for data compilation, estimation
of climate variables have been reported in Wang et al. (2008)
and thus not described here. Another 166 plots across all the major
mountain ranges (Changbai Mountains, Zhangguangcai Mountains,
Xiaoxing’an Mountains, Daxing’an Mountains and Yanshan–Tai-
hang Mountains) and forest types in northeast China were sampled
by us using the method described in Wang et al. (2008) (in that
analysis 85 plots were sampled by ourselves). In a few Chinese lit-
eratures, the average tree height of a plot is calculated as the tree
height corresponding to the geometric mean of DBH (estimated
with DBH–height relationship of the plot) (e.g. Meng, 2006). How-
ever, for the literatures where our biomass data came from, none of
them have stated that the average heights were calculated in that
way. Because the arithmetic mean of tree height is a far more com-
monly used statistic in ecological studies (e.g. Cannell, 1982; Fang
et al., 2006), we assumed that all these literatures documented
 X. Wang et al. / Forest Ecology and Management 293 (2013) 149–160 151

arithmetic means (if a different definition for average height was

used, the literature should have clarified it). As a result, we also cal-
culated the arithmetic means of tree height for the 166 plots sam-
pled by us.
Our database documented 230, 230 and 228 data pairs for the
relationships between forest height and total, above and below-
ground stand biomass, respectively. At the same time, the database
also included 312–338 records for the relationships between aver-
age tree height and mean biomass per stem (total, above and
belowground), 330–373 records for the relationships between
average tree height and stand biomass, and 429–448 records for
relationship between stock volume and stand biomass. These re-
cords (totally 529 plots) were directly used in the data analyses
in this study. For the basic information of the plots, see Appendix A.
These plots included all the major forest types in NE China
(Zhou, 1997; Wang et al., 2006b; 2008): (1) coniferous and broad-
leaf mixed (CBM) forest; (2) mixed deciduous broadleaf forest
(dominated by Quercus and Tilia spp., and mixed with Fraxinus
mandschurica, Juglans mandshurica and Acer ssp., etc.); (3) Larix for-
est (Larix gmelini and L. olgensis); (4) Picea and Abies forest (Picea
jezoensis, Abies nephrolepis and Picea koraiensis); (5) Pinus tabulae-
formis forest; (6) Populus and Betula forest (Populus davidiana, Bet-
ula platyphylla and B. ermanii). We also grouped these forest types
into two canopy types based on the life form of dominant species:
(1) deciduous forests, including mixed deciduous broadleaf forest,
Larix forest, and Populus and Betula forest; (2) evergreen forests,
including Picea and Abies forest, P. tabulaeformis forest, and CBM
forest. Note that though CBM forests are mixed with various decid-
uous broadleaf species (Quercus and Tilia spp., etc.), evergreen
coniferous species (e.g. P. koraiensis) are the dominant species in
the top canopy layer (Zhou, 1997; Wang et al., 2006a). As a result,
CBM forests showed similar biomass–height relationship as ever-
green needle-leaf forests (see Section 3), instead of deciduous for-
ests. Consequently, in this analysis CBM forests were grouped into
evergreen forests for simplicity in data analysis.
In Wang et al. (2008), we have shown that our data covered
most of the geographic and climatic gradients in the study area
(see also Appendix A here), and were well distributed across the
forest region of northeast China. In addition, for each forest type
the geographic ranges of our dataset were similar with its distribu-
tion ranges in the study area (Fig. 1 and Table 1 in Wang et al.
(2008)). As a result, these plots were good representations for for-
ests across northeast China and were well suited for exploring the
methods to estimate forest biomass at a large scale.

2.2. Statistic analyses

We used the power function to fit the relationships between

biomass and height, which is a widely accepted function used to
describe allometric relationships (e.g. West et al., 1999; Niklas
and Enquist, 2001). The exponent function was also tried but ex-
cluded from final analyses because the power function performed
better in describing the relationships
log ðBÞ ¼ a þ b log ðHÞ
log ðMBÞ ¼ a þ b log ðHaÞ
log ðBÞ ¼ a þ b log ðHaÞ
where B is sand biomass, MB is mean biomass per stem, H is forest
height, Ha is average tree height, and a and b are coefficients.
The relationship between stand biomass and volume was fitted
with the biomass expansion factor (BEF) method (Fang et al., 2001;
Brown, 2002):
B¼aþb V
Fig. 1. Relationships between stand biomass and forest height (a), between mean
stem biomass and average tree height (b), and between stand biomass and average
height (c) for forest plots across northeast China. Total, above and below: total,
aboveground and belowground biomass, respectively. For R2s and sample sizes of
the relationships, see Table 1.
ð1Þ
where B is stand biomass, V the stock volume, a and b are coeffi-
cients. BEF (=a + b/V) is defined as the ratio of stand biomass to
ð2Þ
stock volume (Fang et al., 2001).
To test these methods to estimate forest biomass from height or
ð3Þ
volume, we randomly extracted ca. 15% of plots in each forest type
from the 227 plots that have documented all of the following infor-
mation: total, above and below ground biomass, forest height, aver-
age tree height, stem density and stand volume. The resulting 37
plots were used for model validation while other plots in the data-
set were used to construct models for estimating forest biomass.
ð4Þ We first used the simple relationships between biomass and
height to compare the ability of three methods to predict forest
152 X. Wang et al. / Forest Ecology and Management 293 (2013) 149–160

Table 1
Three types of biomass–height relationships across northeast China and their abilities to predict stand biomass. (1) log (B) = a + b  log (H), the relationship between stand
biomass (B) and forest height (H). (2) log (MB) = a + b  log (Ha), the relationship between mean stem biomass (MB) and average tree height (Ha). (3) log (B) = a + b  log (Ha), the
relationship between stand biomass (B) and average tree height (Ha). For each model, we reported the coefficients (a and b), R2 and number of plots (n). Another 37 independent
plots were used to test these models. For each test, the R2, slope and intercept between predicted and observed stand biomass were reported. CI, the 95% confidential interval of
the slope or intercept. All regressions were significant at P < 0.05 except for the regression with a R2 of 0.07 (P = 0.12).

Biomass Model Predicted vs. observed stand biomass

Component a b R2 n Slope (CI) Intercept (CI) R2
log (B) = a + b  log (H)
Total 1.69 1.10 0.73 193 0.72 (0.61–0.84) 31.26 (11.38–51.14) 0.78
Aboveground 1.48 1.09 0.73 193 0.72 (0.60–0.85) 25.37 (8.70–42.05) 0.75
Belowground 0.11 1.10 0.67 191 0.59 (0.49–0.72) 9.4 (4.93–13.87) 0.68
log (MB) = a + b  log (Ha)
Total 7.47 2.10 0.78 301 0.72 (0.53–0.98) 13.43 (27.16–54.01) 0.17
Aboveground 7.53 2.03 0.74 298 0.71 (0.52–0.96) 12.32 (18.89–43.52) 0.17
Belowground 9.06 2.11 0.70 275 0.60 (0.44–0.84) 5.75 (2.70–14.20) 0.07
log (B) = a + b  log (Ha)
Total 1.42 1.36 0.66 336 0.71 (0.58–0.88) 8.89 (17.3–35.09) 0.62
Aboveground 1.50 1.23 0.61 316 0.65 (0.52–0.80) 14.95 (4.22–34.13) 0.60
Belowground 0.05 1.27 0.57 293 0.55 (0.44–0.69) 5.94 (1.03–10.84) 0.57

biomass: (1) estimating stand biomass with forest height using Eq.
(1); (2) estimating mean biomass per stem with Eq. (2) from aver-
age tree height, and then stand biomass was calculated as mean
biomass  stem density; (3) estimating stand biomass directly
from average tree height using Eq. (3).
To analyze the potential factors affecting biomass–height rela-
tionship, we used general linear model (GLM) to explain log (bio-
mass) with log (height) and several factors together: (1) climate
(growing season temperature and precipitation, GST and GSP
respectively), (2) canopy-tree life form (evergreen vs. deciduous
forest), (3) forest type (the six major forest types in northeast china
described above), and (4) forest origin (primary/secondary/planted
forest). The interaction terms between log (height) and other
explanatory variables were also included in GLMs to examine
whether the regression slopes between log (biomass) and log
(height) were affected by climate and forest groups (canopy-tree
life form, forest type and origin).
Sequential (type I) sums of squares (Schmid et al., 2002) were
used for analysis of variances (ANOVAs). The advantage of this
method is that we can examine the effect of a variable when the
effects of other variables have been accounted (Schmid et al.,
2002; Wang et al., 2009). Climatic variables (GST and GSP) were
entered into GLMs the first, so that the effects of differences among
forest groups were evaluated under a similar climatic condition.
Canopy-tree life form had to be entered into GLMs before forest
type, because forest type had included the explanatory power of
canopy-tree life form. Forest origin could not significantly explain
variations in biomass data when entering GLMs before canopy-tree
life form and forest type, and thus was entered into the model after
forest type (forest origin only showed significant effects within for-
est types).
After the major factors affecting biomass–height relationship
were identified, we went further to obtain models to predict stand
biomass from height with the least variables. We used the model
selection approach based on Akaike Information Criterion (AIC),
to drop unnecessary variables (Johnson and Omland, 2004).
To validate the resulting models for estimating forest biomass,
we calculated predicted biomass for the 37 remaining plots with
each model. Predicted values were fitted against observed biomass
to obtain the regression R2, the 95% confidential intervals (CIs) for
regression slope and interception. If the R2 was high enough, and
the 95% CI for slope and interception enclosed 1 and 0, respectively,
then the model was judged as acceptable (e.g. Wang and Fang,
2012).
3. Results
3.1. Three relationships between biomass and height
Biomass was closely related to height in forests across northeast
China (Fig. 1). This was true for all the three types of relationships
examined in this study. Forest height explained 73%, 73% and 67%
of variations in total, aboveground and belowground stand bio-
mass, respectively (Table 1), and average tree height also explained
78%, 74% and 70% of variations in mean stem biomass, respectively.
The relationship between stand biomass and average tree height,
though less strong compared with the other two relationships, also
explained 57–66% of variations in stand biomass.
However, the three biomass–height relationships showed great
differences in the ability to predict stand biomass (Table 1). The
stand biomass vs. forest height models (Eq. (1)) predicted the bio-
mass of the validation plots with an R2 of 0.78, 0.75 and 0.68,
respectively for the total, aboveground and belowground compo-
nents. The slopes between predicted and observed biomass (0.72,
0.72 and 0.59, respectively) were not satisfied. However, this prob-
lem can be well resolved by including other factors influencing bio-
mass–height relationship into the models (see below). Thus this
method was analyzed in details in the following sections.
On the other hand, the mean stem biomass vs. average height
models (Eq. (2)) predicted stand biomass with an R2 of only 0.17,
0.17 and 0.07, respectively. Consequently this relationship was ex-
cluded from a potential method for estimating stand biomass. For
the stand biomass vs. average height models (Eq. (3)), the R2s be-
tween predicted and observed biomass (0.66, 0.61 and 0.57,
respectively) was lower than the forest height method (Eq. (1)),
while the regression slopes were similar (0.72, 0.71 and 0.60,
respectively). However, the problem of unsatisfied slopes could
not be well fixed by including other variables into Eq. (3) (see
below).
3.2. Factors regulating the relationship between forest biomass and
height
The ratio of stand biomass to forest height (B/H ratio hereafter)
was roughly similar for our plots across northeast China (Fig. 2a).
This is not only true for total stand biomass, but also for biomass
above and below ground. This suggested that the biomass per unit
of forest space was similar for closed-canopy forests (Fang et al.,
 X. Wang et al. / Forest Ecology and Management 293 (2013) 149–160 153

Fig. 2. Ratio of stand biomass to forest height (B/H ratio) for plots across northeast China. (a) B/H ratio plotted against forest height. (b–d): comparison of B/H ratio among six
forest types, respectively for total (b), aboveground (c) and belowground biomass (d). Forest types that shared a same letter did not showed significant difference at P < 0.05.
CBM, coniferous–broadleaf mixed forest, MDB, mixed deciduous broadleaf forest, La, Larix forest, PA, Picea and Abies forest, Pt, Pinus tabulaeformis forest, PB, Populus and Betula
forest. DB, La and PB are deciduous forest while others were classified as evergreen forest (see Section 2). B/H ratio data were highly skewed and thus were log-transformed
prior to analyses.

2006) and thus supported the hypothesis that forest biomass pat-
terns were roughly shaped by forest height at a large scale.
Fig. 2 revealed that, B/H ratios generally did not differed signif-
icantly among deciduous forests, including mixed deciduous
broadleaf forest, Larix forest and Populus and Betula forest. At the
same time, the ratios also did not differed among evergreen nee-
dle-leaf forests, including Picea and Abies forest, P. tabulaeformis
forest and coniferous–broadleaf mixed forest. However, B/H ratios
showed significant differences between deciduous and evergreen
forests, and suggested that canopy-tree life form (evergreen vs.
deciduous forest) was an important factor affecting the relation-
ships between stand biomass and forest height.
The GLMs analyses showed that (Table 2), climate (GST and
GSP), canopy-tree life form (CLF) and forest origin (FO) had signifi-
cant power in explaining forest biomass in addition to forest height.
However, forest type could not significantly explain the variations
within deciduous (or evergreen) forest for total and aboveground
biomass (P < 0.05 level). Canopy-tree life form explained 6.3%,
5.8% and 8.5% (total, above and below ground, respectively) of vari-
ations in forest biomass, which was by far higher than that of cli-
mate, forest type and forest origin (at most 2.6%). In all the
interaction terms between forest height and other factors, the inter-
action between forest height and canopy-tree life form was the only
term that showed significant effect. This suggested that the slopes
of biomass–height relationships differed significantly between
evergreen and deciduous forests (Fig. 3). Taking together, the GLMs
confirmed the results of Fig. 2, and further suggested that canopy-
tree life form is the major factor regulating biomass–height rela-
tionships, while climate, forest type and forest origin played a weak
role.
3.3. Models for estimation stand biomass
When the full models in Table 2 were submitted for model
selection based on AIC, the final model included forest height, cli-
mate, forest type and forest origin, and some interaction terms
(Appendix B, abbreviated as GLM1 hereafter). These models pre-
dicted the total, above and belowground biomass for the validation
plots with an R2 of 0.91, 0.89 and 0.78, respectively (Table 3).
Importantly, all the slopes between predicted and observed values
were not significantly different from 1 (see the 95% confidential
intervals), while the intercepts were also not significantly different
from 0. This suggested that the empirical data were predicted
(Fig. 4).
Canopy-tree life form was excluded from GLM1 (Appendix B) by
the model selection algorithm. However, this was because the
explanatory power of forest type has included that of canopy-tree
life form, and clearly did not mean that canopy-tree life form was
not important (see Table 2). To obtain the most parsimonious mod-
els, we excluded forest type from the full model and re-run the
154 X. Wang et al. / Forest Ecology and Management 293 (2013) 149–160

Table 2
Summary of general linear models for the effects of forest height (H), climate (growing season temperature and precipitation, GST and GSP respectively), canopy-tree life form
(CLF), forest type (FT), and forest origin (FO) on forest biomass (log-transformed) across northeast China. df, degree of freedom; MS, mean square; %SS, percentage of sum of
squares explained.

df MS P %SS df MS P %SS df MS P %SS

Total biomass Aboveground biomass Belowground biomass
log (H) 1 54.52 0.000 73.30 1 54.08 0.000 73.11 1 54.43 0.000 67.36
GST 1 0.32 0.027 0.43 1 0.27 0.045 0.37 1 0.45 0.018 0.56
GSP 1 1.11 0.000 1.50 1 1.57 0.000 2.13 1 0.71 0.003 0.88
CLF 1 4.66 0.000 6.26 1 4.32 0.000 5.84 1 6.83 0.000 8.45
FT 4 0.14 0.068 0.77 4 0.14 0.076 0.78 4 0.29 0.006 1.44
FO 2 0.53 0.000 1.43 2 0.33 0.008 0.90 2 1.03 0.000 2.55
log (H):GST 1 0.03 0.476 0.04 1 0.07 0.324 0.09 1 0.02 0.630 0.02
log (H):GSP 1 0.11 0.193 0.15 1 0.04 0.437 0.05 1 0.29 0.056 0.36
log (H):CLF 1 0.78 0.001 1.04 1 0.62 0.003 0.84 1 1.10 0.000 1.36
log (H):FT 4 0.02 0.831 0.13 4 0.03 0.788 0.15 4 0.05 0.617 0.26
log (H):FO 2 0.02 0.789 0.04 2 0.04 0.533 0.11 2 0.05 0.533 0.12
Residuals 173 0.06 14.91 173 0.07 15.63 171 0.08 16.62

model selection. The final models included only forest height, cli-
mate and canopy-tree life form for total and aboveground biomass,
but forest origin was retained in the model for belowground bio-
mass (Table 4, named as GLM2 hereafter). These models predicted
total, above and belowground biomass with an R2 of 0.86, 0.84 and
0.74, respectively (Table 3), and all the slopes and intercepts be-
tween predicted and observed values were also not significantly
different from 1 and 0, respectively. Thus the forest biomass were
still well predicted by GLM2 (Fig. 4), despite the fact that GLM2 in-
volved much less model parameters compared with GLM1 (Table 4
and Appendix B).
For comparison with these two models based on forest height,
we developed the relationship between stand biomass and volume
(Eq. (4)) for each forest type (Table 5). These equations predicted
total, above and belowground biomass with an R2 of 0.98, 0.98
and 0.72, respectively (Table 3). And the slopes and intercepts be-
tween predicted and observed values also had 95% confidential
intervals that included (or nearly included) 1 and 0, respectively
(Table 3 and Fig. 4).
To show that forest biomass could not be estimated from aver-
age tree height, we also constructed the GLM models for the two
methods based on average height (Eqs. (2) and (3)). Similarly, the
variables retained in the final models were also selected based
on AIC (Appendix C). The models based on the relationship be-
tween stand biomass and average tree height (GLM3 hereafter)
was not able to predict forest biomass accurately (Table 3), despite
the fact that GLM3 explained 77–82% of variations in stand bio-
mass (Appendix C). The R2s between predicted and observed values
ranged between 0.74 and 0.79 (which was not so bad), while the
regression slopes were only 0.66–0.80. Because the 95% confiden-
tial intervals for the slopes were clearly lower than 1, which sug-
gested a significant underestimation of stand biomass, this
method was not acceptable.
As for the models based on the relationship between mean stem
biomass and average tree height (GLM4, Appendix C), none of the
R2s (0.29–0.47) and slopes (0.57–0.67) between predicted and ob-
served values showed that this method was acceptable as a meth-
od for biomass estimation (Table 3).

4. Discussion

4.1. Stand biomass cannot be estimated from average tree height

Fig. 3. Relationships between stand biomass and forest height for deciduous (D) In this analysis, we examined three types of biomass–height
and evergreen (E) forests in northeast China. Deciduous forest included deciduous relationships for forest plots across northeast China (Fig. 1). We
broadleaf forest and deciduous needle-leaf forest, while evergreen forest included showed that forest biomass could not be predicted with the two
evergreen needle-leaf forest and coniferous–broadleaf mixed forest. methods based on average tree height (Table 3). It is well known
 X. Wang et al. / Forest Ecology and Management 293 (2013) 149–160
Table 3
Testing five methods for estimating forest biomass across northeast China, using 37 plots remained for model validation. GLM1 and GLM2, predicting biomass based on the
relationships between stand biomass and forest height (B–H) using the GLMs in Appendix B and Table 4, respectively. BEF, estimating biomass from stand volume using the
equations in Table 5. GLM3, using the GLMs based on the relationships between stand biomass and average tree height (B–Ha) (Appendix C). GLM4, using the GLMs based on the
relationships between mean stem biomass and average height (MB–Ha) (Appendix C). For each test, the R2, slope and intercept between predicted and observed values were
reported. The slopes and intercepts were boldfaced, if the 95% confidential intervals (CIs) of slope included 1 and the intercept CI also included 0.
Biomass Predicted vs. observed stand biomass
Component Slope (CI) Intercept (CI)
GLM1 (B–H, Appendix B)
Total 1.01 (0.91–1.12) 2.41 (15.56–20.39)
Aboveground 0.96 (0.86–1.08) 6.82 (7.79–21.43)
Belowground 0.98 (0.83–1.15) 1.50 (4.67–7.67)
BEF (Table 5)
Total 0.97 (0.92–1.02) 7.51 (1.23–16.25)
Aboveground 0.93 (0.88–0.99) 8.53 (1.70–15.36)
Belowground 0.99 (0.83–1.18) 2.26 (4.72–9.23)
GLM4 (MB–Ha, Appendix C)
Total 0.67 (0.52–0.86) 34.74 (5.16–64.32)
Aboveground 0.63 (0.50–0.81) 30.14 (8.42–51.86)
Belowground 0.57 (0.43–0.76) 10.19 (3.5–16.89)
that the individual tree biomass and height data of forest plots are
commonly highly skewed, and thus arithmetic mean was not an
appropriate statistic for both of them (e.g. Chen et al., 1986; Zhu,
2005; Meng, 2006). Consequently, it is not surprising that stand
biomass could not be well estimated from average tree height (Ta-
ble 3), even when the effects of climate, forest type and forest ori-
gins were all included into the models (Appendix C).
Unfortunately, most of the previous literatures on field-ob-
served forest biomass reported average tree height instead of for-
est height (or maximum tree height). For instance, our database
included 641 plots for biomass observations, in which only 231
plots have documented forest height. Further, most (1 6 6) of these
231 plots were sampled by us and another 52 plots came from Ma
(1988). That is, the other 423 biomass records that were compiled
from literatures documented forest height for only 13 plots. In
comparison, these literatures documented averaged tree height
for 150 plots and stand volume for 251 plots. The lack of forest
height information in biomass literatures is actually a widespread
phenomenon. For another example, in the famous dataset of Can-
nell (1982), which compiled field-observed biomass data for plots
across the world, most plots documented average height rather
than forest height. However, our results showed that average tree
height actually had no practical use for the estimation of forest bio-
mass (Table 3). Consequently, though there are abundant data on
forest biomass in the literatures, most of them cannot be utilized
to construct models for estimating spatial biomass patterns with
forest height. As a result, we strongly suggest future studies on
field-observed biomass to report forest height (which is easy to
measure) in order to avoid unnecessary repeating investigations
of stand biomass (which are time and efforts consuming).
4.2. Why can stand biomass be estimated from forest height?
Our results showed that stand biomass could be well predicted
with models based on forest height (Table 3 and Fig. 4), consistent
with previous studies which found that forest biomass could be
estimated with LiDAR retrieved forest height satisfactorily (e.g.
Dubayah and Drake, 2000; Hurtt et al., 2004; Lefsky et al., 2005;
van Leeuwen and Nieuwenhuis, 2010). These findings raised an
interesting question: why the biomass of various forest communi-
ties (which had complicate and different structures) was so
strongly shaped by a simple variable (note that forest height alone
explained 67–73% of variations in stand biomass, see Table 1)?
155
Predicted vs. observed stand biomass
R2 Slope (CI) Intercept (CI) R2
GLM2 (B–H, Table 4)
0.91 0.98 (0.86–1.12) 2.94 (18.96–24.84) 0.86
0.89 0.94 (0.81–1.07) 7.38 (9.97–24.74) 0.84
0.78 1.02 (0.86–1.21) 0.17 (7.15–6.82) 0.74
GLM3 (B–Ha, Appendix C)
0.98 0.80 (0.68–0.93) 17.79 (3.96–39.55) 0.79
0.98 0.77 (0.65–0.92) 16.19 (1.96–34.34) 0.74
0.72 0.66 (0.55–0.77) 7.29 (3.02–11.56) 0.76
0.45
0.47
0.29
Fang et al. (2006) proposed a hypothesis that the aboveground
biomass per forest space (i.e. the ratio of aboveground biomass to
forest height) was similar for closed-canopy forests, and thus, at a
large scale the aboveground biomass per area could be roughly ex-
plained by differences in forest height. Our results extended this
hypothesis in that:
(1) We showed that the ratio of belowground biomass to forest
height is also roughly similar across northeast China
(Fig. 2a), which is not surprising because belowground bio-
mass showed close allometric relationship with above-
ground biomass (e.g. Cairns et al., 1997; Mokany et al.,
2005; Wang et al., 2008).
(2) While climate, forest type and forest origins showed weak
influences on the relationship between stand biomass and
forest height (Table 2), canopy-tree life form is an important
factor modulating this relationship (Figs. 2 and 3). Conse-
quently, it is more appropriate for the hypothesis to state
that ‘‘the ratio of stand biomass to forest height is roughly
similar within the deciduous or evergreen forests’’. Because
our dataset did not included evergreen broadleaf forests
(which are distributed in central and south China), this
hypothesis needs further tests before drawing a conclusion.
However, Drake et al. (2003) found that canopy deciduous-
ness is the major contributor affecting the relationship
between aboveground biomass and LiDAR derived metrics
of height in tropical forests. These similar findings in both
temperate (this analysis) and tropical (Drake et al., 2003)
forests also support our hypothesis. Previous studies that
estimate forest parameters with LiDAR have noticed the
importance of discriminating coniferous and broadleaf for-
ests, however, in many analyses the coniferous forests were
evergreen ones while the broadleaf forests were deciduous
ones (see van Leeuwen and Nieuwenhuis, 2010). Our results
showed that deciduous broadleaf and deciduous coniferous
forests did not differ in B/H ratios (Fig. 2), thus it may be that
the differences between coniferous and broadleaf forests
found by previous studies actually reflect the effect of can-
opy-tree life form (see also Chen, 2010).
(3) In Fang et al. (2006), the data outside China came from Can-
nell (1982). Because of the limited forest height data in Can-
nell’s dataset, in Fang et al. (2006) we had to analyze the
relationship between stand biomass and average tree height,
and thus got a ratio of aboveground biomass to average tree
156 X. Wang et al. / Forest Ecology and Management 293 (2013) 149–160
Fig. 4. Predicted stand biomass plotted to observed values for three of the methods
examined in this study (the diagonal is the 1:1 line). Both the GLM1 and GLM2
methods estimated stand biomass from forest height, but the former used the
general linear models (GLMs) based on forest types in Appendix B, while the GLM2
method used the GLMs based on canopy-tree life form in Table 4. For comparison
with these two methods, we also estimated biomass from stand volume (BEF) using
the equations in Table 5. The forest plots used here were remained for model
validations (see Section 2) and were independent of the procedure of model
coefficients estimation. For R2s, slopes and intercepts between predicted and
observed stand biomass, see Table 3.
height of ca. 1.0 kg m3 (0.93–1.05). Using the dataset in this
study, we obtained a ratio of aboveground biomass to forest
height of 0.63 kg m3 (SD = 0.16) for deciduous forests, and
0.86 kg m3 (SD = 0.27) for evergreen forests. Considering
that forest biomass cannot be predicted from average tree
height (Table 3), we suggest that the ratios reported here
may be more appropriate. We also advocate future studies
to examine whether these ratios are similar in other regions
of the world. If this is true, then this hypothesis may provide
a sound biological basis for estimating forest biomass pat-
terns from satellite LiDAR data.
However, these results clearly do not mean that forest height
alone can estimate forest biomass accurately. First, currently there
are still many uncertainties in the forest height estimates from Li-
DAR, especially in regions with steep and complicate topography
or at the large scales (e.g. Lefsky, 2010; Simard et al., 2011; Los
et al., 2012). Second, the relationships between forest biomass
and height were significantly affected by abiotic ad biotic factors
(Table 2). Consistent with biomass estimation studies using LiDAR
retrieved height (e.g. Drake et al., 2003; van Leeuwen and Nie-
uwenhuis, 2010), our results also showed that considering these
effects is necessary to increase the accuracy of biomass estimation
(comparing Tables 1 and 3).
4.3. The models for estimating forest biomass from forest height
In this analysis, we compared several methods for estimating
forest biomass (Table 3). Our results confirm previous findings that
the BEF method (predicting biomass from stock volume) (e.g. Fang
et al., 2001; Brown, 2002) is the most accurate method for the esti-
mation of plot-level biomass (Table 3 and Fig. 4). However, the R2
between predicted and observed biomass for GLM1 and GLM2
(based on forest height) was only slightly lower than the BEF meth-
od (0.74–0.91 vs. 0.72–0.98). Considering that the great efforts in
measuring stock volume for large number of field plots (required
by the BEF method) could be avoid by LiDAR technology in large-
scale biomass estimations, the slightly lower predictive accuracy
of the forest height models (e.g. GLM2) is acceptable. Further, it
should be noted that there are also many uncertainties in the re-
gional- or country-level forest inventory data on stock volume.
For instance, many uncertainties may be introduced during the
procedure of scaling up stock volume from plot to regional scales
(see e.g. The forestry administration of China, 2003). Thus it may
be that, at a large scale, forest biomass estimated based on the
BEF method is actually not better than the method based on forest
height (if the LiDAR derived forest heights are accurate enough).
Our results showed that forest biomass patterns could not be
accurately predicted with forest height alone (Table 1), and the dif-
ferences among forest groups should be considered (Table 3). This
means that monitoring forest distributions is necessary for moni-
toring biomass patterns with LiDAR. While monitoring the detailed
distributions for each forest type with remote sensing images can
be a great task at the large scale, a simple classification of ever-
green vs. deciduous forests is much easier. Comparing GLM1 and
GLM2, the predictive accuracy was only slightly lower for GLM2
(Table 3), which was natural because GLM2 included remarkably
less model parameters (Table 4 and Appendix B). However, GLM2
would involve much less efforts during the monitoring of forest
distributions. Consequently, we suggest that a more parsimonious
model (e.g. GLM2) may be more useful in the practical works of
forest biomass monitoring. Similar as our results, a recent analysis
on retrieving forest height from LiDAR also suggest that only the
differences between woodlands (deciduous) vs. conifer (evergreen)
forests need to be considered, while the differences among com-
munity types within woodlands or conifer forests could be omitted
(Chen, 2010). In another study in tropical forests, Drake et al.
(2002) also showed that a single relationship between above-
ground biomass and LiDAR data could well predict the plot bio-
mass for primary, secondary and agroforestry forests, suggesting
that the differences among forest origins could be omitted. This
is consistent with our results that forest origin was excluded from
the GLM2 for aboveground biomass (Table 4). Taking together,
these evidences support our hypothesis that biomass–height
 X. Wang et al. / Forest Ecology and Management 293 (2013) 149–160 157

Table 4
General linear models for predicting stand biomass (log-transformed) with forest height (H), growing season temperature and precipitation (GST and GSP, respectively), canopy-
tree life form (CLF), and forest origin (FO). CLF included two categories: deciduous forest and evergreen needle-leaf forest (E). FO included three categories: primary (Prim),
secondary (Sec) and planted forest. Unnecessary explanatory terms were dropped based on AIC. SE, standard error.

Estimate SE P Estimate SE P Estimate SE P

Total biomass Aboveground biomass Belowground biomass
(Intercept) 2.46 0.64 0.000 (Intercept) 1.04 0.28 0.000 (Intercept) 0.83 1.25 0.508
log (H) 0.64 0.22 0.004 log (H) 1.05 0.09 0.000 log (H) 1.19 0.45 0.009
GST 0.02 0.01 0.016 GST 0.01 0.01 0.063 GST 0.20 0.07 0.004
GSP 0.00 0.00 0.094 GSP 0.00 0.00 0.015 GSP 0.00 0.00 0.017
CLF_E 0.35 0.28 0.207 CLF_E 0.14 0.28 0.615 CLF_E 0.87 0.36 0.018
log (H):GSP 0.00 0.00 0.059 log (H):CLF_E 0.16 0.10 0.095 FO_Prim 0.15 0.10 0.112
log (H):CLF_E 0.24 0.10 0.015 FO_Sec 0.02 0.07 0.791
log (H):GST 0.06 0.02 0.021
log (H):GSP 0.00 0.00 0.021
log (H):CLF_E 0.44 0.12 0.000
R2 = 0.82, adjusted R2 = 0.82 R2 = 0.82, adjusted R2 = 0.81 R2 = 0.81, adjusted R2 = 0.80

Table 5
Relationship between stand biomass and stock volume for each forest type in northeast China (B = a + bV, where B is biomass, V is stand volume, and a and b are coefficients).
Abbreviations: n, sample size, CBM, coniferous–broadleaf mixed forest, MDB, mixed deciduous broadleaf forest, La, Larix forest, PA, Picea and Abies forest, Pt, Pinus tabulaeformis
forest, PB, Populus and Betula forest.

a b R2 n a b R2 n a b R2 n
Total biomass Aboveground Belowground
CBM 13.16 0.58 0.92 79 18.70 0.41 0.82 74 6.65 0.17 0.87 73
MDB 35.55 0.62 0.82 76 21.89 0.53 0.80 77 13.07 0.09 0.72 74
La 26.81 0.55 0.93 73 16.10 0.45 0.91 75 9.66 0.10 0.58 71
PA 28.89 0.53 0.94 38 27.76 0.40 0.94 37 0.74 0.14 0.82 35
Pt 8.81 0.78 0.98 62 6.90 0.63 0.97 64 0.42 0.17 0.95 61
PB 28.54 0.64 0.84 82 20.66 0.50 0.83 84 7.56 0.14 0.52 78

relationship may be similar for various forest types within decidu-
ous (or evergreen) forests, and thus different deciduous (or ever-
green) forests can be grouped to estimate aboveground biomass
from LiDAR data.
In contrast to biomass aboveground, great uncertainty still ex-
ists in estimating both the magnitude and spatial distribution of
belowground biomass (Brown, 2002), and some studies suggested
that the worldwide belowground biomass have been markedly
underestimated by traditional root: shoot biomass ratios (R/S ratio)
(e.g. Cairns et al., 1997; Mokany et al., 2005). In a previous study,
we showed that the belowground forest biomass across northeast
China could be predicted from aboveground biomass with an R2 of
0.71 through improved R/S ratios (Wang et al., 2008). Interestingly,
the GLM2 reported here actually showed better predictive accu-
racy (R2 = 0.74) than the improved R/S ratio method in that analysis
(and also avoid the great efforts to investigate aboveground bio-
mass!). Consequently, our results strongly support the LiDAR tech-
nology as a promising approach to monitor global forest biomass
belowground. Different from the total and aboveground biomass
models, forest origin was retained in the GLM2 for belowground
biomass (Table 4). This is consistent with previous findings that
natural and planted forests showed significant differences in the
above vs. below ground allocation of biomass (Mokany et al.,
2005; Wang et al., 2008). Thus the distribution map of planted for-
ests may still be needed for an accurate mapping of belowground
patterns.
Acknowledgments
This work was supported by the Special Research Program for
Public-welfare of the State Forestry Administration of China
(#200804001), the Research Fund of the Beijing Municipal Com-
mission of Education for development of Key Laboratory for Silvi-
culture and Conservation, the Research Funds for Doctoral
Education in Universities (#20090014120002) of China, the Project
for subject development and graduate student education of Beijing
(for Ecology in BJFU, 2009), and CFERN&GENE Award Funds on Eco-
logical Paper.

Appendix A

Basic information for plots used in this study. Abbreviations: Da, average diameter at breast height; Ha, mean tree height; Hm, maxi-
mum tree height; CBM, coniferous–broadleaf mixed forest; MDB, mixed deciduous broadleaf forest; La, Larix forest; PA, Picea and Abies for-
est; Pt, Pinus tabulaeformis forest; PB, Populus and Betula forest.

Forest type Geographic location Stand characteristics

Latitude Longitude Altitude Age Density (stem/ Da Ha Hm Volume (m3/
(°N) (°E) (m) (Year) ha) (cm) (m) (m) ha)
CBM Mean 43.4 125.9 622 61 1925 16.1 12.5 22.8 255.2
Max 50.7 133.5 2550 238 7052 37.4 27.0 34.8 714.9
Min 39.9 115.0 200 10 200 2.6 2.9 11.1 8.1

(continued on next page)

158 X. Wang et al. / Forest Ecology and Management 293 (2013) 149–160

Appendix A (continued)
Forest type Geographic location Stand characteristics
Latitude Longitude Altitude Age Density (stem/ Da Ha Hm Volume (m3/
(°N) (°E) (m) (Year) ha) (cm) (m) (m) ha)

MDB Mean 42.3 122.3 587 52 1532 13.2 9.6 16.1 125.2
Max 51.7 134.0 1652 157 8326 22.2 21.0 32.0 385.0
Min 39.9 115.4 177 9 226 3.3 2.7 8.0 22.1

La Mean 48.5 123.5 853 75 2101 16.6 14.1 22.8 193.1

Max 52.7 131.8 2650 195 15,782 37.8 26.3 34.3 648.3
Min 39.9 115.0 300 5 213 4.7 5.4 13.0 53.4

PA Mean 45.2 126.7 995 56 1779 13.9 9.2 24.7 269.3

Max 52.6 131.8 2250 170 4225 23.6 15.9 34.1 609.8
Min 39.9 115.0 280 10 631 0.8 0.8 18.0 0.2

Pt Mean 41.6 120.0 644 32 2015 13.1 9.4 11.4 121.6

Max 42.7 129.5 1800 95 5137 26.9 18.8 21.3 455.8
Min 39.8 111.0 190 18 517 5.6 3.3 4.1 18.2

PB Mean 44.6 124.2 816 36 1495 13.7 11.7 19.5 161.2

Max 52.5 134.0 2350 89 7320 20.0 19.5 32.6 333.9
Min 39.8 111.0 145 8 184 6.3 5.4 9.4 38.9

Overall Mean 44.2 123.6 738 50 1797 14.4 11.2 18.0 181.1
Max 52.7 134.0 2650 238 15,782 37.8 27.0 34.8 714.9
Min 39.8 111.0 145 5 184 0.8 0.8 4.1 0.2

Appendix B

General linear models for predicting stand biomass (log transformed) with forest height (H), growing season temperature and precip-
itation (GST and GSP, respectively), forest type (FT), and forest origin (FO) in northeast China (i.e. GLM1 in main text and Table 3). FT in-
cluded six forest types: coniferous–broadleaf mixed forest, mixed deciduous broadleaf forest (MDB), Larix forest (La), Picea and Abies forest
(PA), Pinus tabulaeformis forest (Pt), and Populus and Betula forest (PB). FO included three categories: primary (Prim), secondary (Sec) and
planted forest. Unnecessary explanatory terms were dropped based on AIC. SE, standard error.

Estimate SE P Estimate SE P Estimate SE P

Total biomass Aboveground Biomass Belowground Biomass
(Intercept) 1.44 0.96 0.134 1.51 0.97 0.119 0.37 1.06 0.731
log (H) 0.94 0.35 0.008 0.82 0.35 0.020 1.09 0.39 0.006
GST 0.03 0.01 0.035 0.03 0.02 0.028 0.03 0.02 0.130
GSP 0.00 0.00 0.116 0.00 0.00 0.179 0.00 0.00 0.095
FT_MDB 1.15 0.70 0.103 0.50 0.73 0.493 2.23 0.79 0.005
FT_La 0.68 0.72 0.342 0.39 0.73 0.597 0.57 0.81 0.482
FT_PA 0.82 1.06 0.440 0.64 1.09 0.557 0.13 1.19 0.913
FT_Pt 0.40 0.61 0.509 -0.13 0.64 0.844 1.08 0.69 0.121
FT_PB 1.14 0.73 0.122 0.70 0.76 0.360 1.47 0.83 0.077
FO_Prim 0.19 0.11 0.070 0.18 0.11 0.103 0.23 0.12 0.051
FO_Sec 0.16 0.07 0.028 0.15 0.07 0.037 0.18 0.08 0.027
log (H):GSP 0.00 0.00 0.066 0.00 0.00 0.093 0.00 0.00 0.076
log (H):FT_MDB 0.49 0.24 0.043 0.28 0.25 0.254 0.84 0.27 0.002
log (H):FT_La 0.29 0.24 0.217 0.19 0.24 0.420 0.28 0.26 0.287
log (H):FT_PA 0.28 0.34 0.413 0.23 0.35 0.510 0.04 0.38 0.905
log (H):FT_Pt 0.01 0.22 0.959 0.16 0.22 0.471 0.26 0.24 0.294
log (H):FT_PB 0.47 0.25 0.059 0.32 0.26 0.215 0.65 0.28 0.022

R2 = 0.85, adjusted R2 = 0.84 R2 = 0.84, adjusted R2 = 0.83 R2 = 0.83, adjusted R2 = 0.82

 X. Wang et al. / Forest Ecology and Management 293 (2013) 149–160 159

Appendix C

General linear models for explaining log-transformed stand biomass (GLM3) or mean biomass per stem (GLM4) with average tree height
(Ha), climate, forest type and forest origin. All other abbreviations were same as Appendix B.

GLM3 Estimate SE P Estimate SE P Estimate SE P

Total biomass Aboveground biomass Belowground biomass
(Intercept) 3.02 0.63 0.000 3.00 0.65 0.000 0.02 0.68 0.975
log (Ha) 0.59 0.27 0.029 0.39 0.28 0.166 0.99 0.29 0.001
GST 0.02 0.02 0.129 0.03 0.02 0.098
GSP 0.00 0.00 0.007 0.00 0.00 0.000 0.00 0.00 0.034
FT_MDB 0.96 0.44 0.032 0.99 0.48 0.039 0.14 0.51 0.787
FT_La 0.14 0.48 0.767 0.31 0.49 0.524 0.42 0.51 0.410
FT_PA 1.77 0.33 0.000 0.96 0.41 0.021 1.09 0.43 0.011
FT_Pt 0.34 0.36 0.345 0.16 0.42 0.709 0.65 0.43 0.133
FT_PB 0.37 0.52 0.486 0.35 0.61 0.567 0.30 0.65 0.645
FO_Prim 1.43 0.50 0.005 2.28 0.55 0.000 2.00 0.57 0.001
FO_Sec 1.43 0.38 0.000 1.75 0.40 0.000 2.41 0.42 0.000
log (Ha):GSP 0.00 0.00 0.002 0.00 0.00 0.000 0.00 0.00 0.026
log (Ha):FT_MDB 0.38 0.19 0.051 0.37 0.20 0.069 0.10 0.22 0.647
log (Ha):FT_La 0.10 0.19 0.603 0.01 0.19 0.959 0.28 0.20 0.175
log (Ha):FT_PA 0.83 0.15 0.000 0.50 0.18 0.006 0.61 0.18 0.001
log (Ha):FT_Pt 0.19 0.16 0.243 0.10 0.18 0.566 0.16 0.19 0.404
log (Ha):FT_PB 0.01 0.22 0.967 0.03 0.25 0.916 0.01 0.27 0.975
log (Ha):FO_Prim 0.38 0.19 0.048 0.74 0.21 0.000 0.52 0.22 0.017
log (Ha):FO_Sec 0.47 0.16 0.004 0.62 0.17 0.000292 0.82 0.18 0.000

R2 = 0.82, adjusted R2 = 0.81 R2 = 0.77, adjusted R2 = 0.75 R2 = 0.79, adjusted R2 = 0.78

GLM4 Mean total biomass Mean aboveground biomass Mean belowground biomass
(Intercept) 8.48 0.37 0.000 8.45 0.95 0.000 11.317 0.416 0.000
log (Ha) 2.19 0.07 0.000 2.09 0.41 0.000 2.429 0.097 0.000
GST 0.03 0.02 0.123 0.04 0.02 0.050 0.077 0.021 0.000
GSP 0.00 0.00 0.106 0.000 0.000 0.128
FT_MDB 0.22 0.11 0.054 0.92 0.74 0.214 0.384 0.115 0.001
FT_La 0.35 0.10 0.001 0.14 0.65 0.830 0.242 0.108 0.025
FT_PA 0.11 0.11 0.346 0.89 0.57 0.115 0.330 0.124 0.008
FT_Pt 0.25 0.10 0.012 0.77 0.57 0.179 0.466 0.102 0.000
FT_PB 0.13 0.12 0.271 1.03 0.88 0.246 0.130 0.124 0.295
FO_Prim 1.28 0.60 0.034 2.10 0.77 0.006 2.147 0.623 0.001
FO_Sec 2.09 0.35 0.000 2.09 0.66 0.002 3.127 0.377 0.000
log (Ha):GSP 0.00 0.00 0.153
log (Ha):FT_MDB 0.32 0.31 0.312
log (Ha):FT_La 0.08 0.26 0.751
log (Ha):FT_PA 0.28 0.24 0.257
log (Ha):FT_Pt 0.24 0.25 0.327
log (Ha):FT_PB 0.48 0.37 0.196
log (Ha):FO_Prim 0.18 0.22 0.419 0.55 0.29 0.059 0.416 0.229 0.071
log (Ha):FO_Sec 0.72 0.15 0.000 0.74 0.28 0.009 1.058 0.161 0.000

R2 = 0.85, adjusted R2 = 0.85 R2 = 0.82, adjusted R2 = 0.81 R2 = 0.85, adjusted R2 = 0.84

References Chen, Q., 2010. Retrieving vegetation height of forests and woodlands over
mountainous areas in the Pacific Coast region using satellite laser altimetry.
Remote Sensing of Environment 114, 1610–1627.
Brown, S., 2002. Measuring carbon in forests: current status and future challenges.
Dixon, R.K., Brown, S., Houghton, R.A., Solomon, A.M., Trexler, M.C., Wisniewski, J.,
Environmental Pollution 116, 363–372.
1994. Cabon pools and flux of global forest ecosystems. Science 263, 185–188.
Cairns, M.A., Brown, S., Helmer, E.H., Baumgardner, G.A., 1997. Root biomass
Drake, J.B., Dubayah, R.O., Clark, D.B., Knox, R.G., Blair, J.B., Hofton, M.A., Chazdon,
allocation in the world’s upland forests. Oecologia 111, 1–11.
R.L., Weishampel, J.F., Prince, S., 2002. Estimation of tropical forest structural
Cannell, M.G.R., 1982. World Forest Biomass and Primary Production Data.
characteristics using large-footprint lidar. Remote Sensing of Environment 79,
Academic Press, London.
305–319.
Chen, L.Z., Chen, Q.L., Bao, X.C., Ren, J.K., Miao, Y.G., Hu, Y.H., 1986. Study on Chinese
Drake, J.B., Knox, R.G., Dubayah, R.O., Clark, D.B., Condit, R., Blair, J.B., Hofton, M.,
arborvitae (Platycladus orientalis) forest and its biomass in Beijing. Acta
2003. Above-ground biomass estimation in closed canopy Neotropical forests
Phytoecologica et Geobotanica Sinica 10, 17–25.
160 X. Wang et al. / Forest Ecology and Management 293 (2013) 149–160
using lidar remote sensing: factors affecting the generality of relationships.
Global Ecology and Biogeography 12, 147–159.
Dubayah, R., Drake, J.B., 2000. Lidar remote sensing for forestry applications. Journal
of Forestry 98, 44–46.
Fang, J.Y., Brown, S., Tang, Y.H., Naruurs, G.-J., Wang, X.P., Shen, H.H., 2006.
Overestimated biomass carbon pools of the northern mid- and high latitude
forests. Climatic Change 74, 355–368.
Fang, J.Y., Chen, A.P., Peng, C.H., Zhao, S.Q., Ci, L.J., 2001. Changes in forest biomass
carbon storage in China between 1949 and 1998. Science 292, 2320–2322.
Houghton, R.A., 2005. Aboveground Forest Biomass and the Global Carbon Balance.
Global Change Biology 11, 945–958.
Hurtt, G.C., Dubayah, R., Drake, J., Moorcroft, P.R., Pacala, S.W., Blair, J.B., Fearon,
M.G., 2004. Beyond potential vegetation: combining Lidar data and a height-
structured model for carbon studies. Ecological Applications 14, 873–883.
IPCC, 2007. Climate Change 2007: Impacts, Adaptation, and Vulnerability.
Cambridge University Press, New York.
Johnson, J.B., Omland, K.S., 2004. Model selection in ecology and evolution. Trends
in Ecology and Evolution 19, 101–108.
Kauppi, P.E., Ausubel, J.H., Fang, J., Mather, A.S., Sedjo, R.A., Waggoner, P.E., 2006.
Returning forests analyzed with the forest identity. PNAS 103, 17574–17579.
Lee, A.C., Lucas, R.M., 2007. A LiDAR-derived canopy density model for tree stem and
crown mapping in Australian forests. Remote Sensing of Environment 111, 493–
518.
Lefsky, M.A., 2010. A global forest canopy height map from the moderate resolution
imaging spectroradiometer and the geoscience laser altimeter system.
Geophysical Research Letters 37, L15401.
Lefsky, M.A., Cohen, W.B., Harding, D.J., Parker, G.G., Acker, S.A., Gower, S.T., 2002.
Lidar remote sensing of aboveground biomass in three biomes. Global Ecology
and Biogeography 11, 393–399.
Lefsky, M.A., Harding, D.J., Keller, M., Cohen, W.B., Carabajal, C.C., Espirito-Santo,
F.D.B., Hunter, M.O., de Oliveira Jr., R., 2005. Estimates of forest canopy height
and aboveground biomass using ICESat. Geophysical Research Letters 32,
L22S02.
Los, S., Rosette, J., Kljun, N., North, P., Chasmer, L., Suarez, J., Hopkinson, C., Hill, R.,
van Gorsel, E., Mahoney, C., 2012. Vegetation height products between 60° S
and 60° N from ICESat GLAS data. Geoscientific Model Development 5, 413–432.
Ma, Q.Y., 1988. Biomass and Primary Productivity for Pinus tabulaeformis Forest.
Beijing Forestry University, Beijing.
Maltamo, M., Eerikäinen, K., Pitkänen, J., Hyyppä, J., Vehmas, M., 2004. Estimation of
timber volume and stem density based on scanning laser altimetry and
expected tree size distribution functions. Remote Sensing of Environment 90,
319–330.
Massada, A.B., Carrnel, Y., Tzur, G.E., Grunzweig, J.M., Yakir, D., 2006. Assessment of
temporal changes in aboveground forest tree biomass using aerial photographs
and allometric equations. Canadian Journal of Forest Research 36, 2585–2594.
Meng, X. (Ed.), 2006. Forest Mensuration. Forestry Press of China, Beijing.
Mokany, K., Raison, R.J., Prokushkin, A.S., 2005. Critical analysis of root: shoot ratios
in terrestrial biomes. Global Change Biology 11, 1–13.
Niklas, K.J., Enquist, B.J., 2001. Invariant scaling relationships for interspecific plant
biomass production rates and body size. PNAS 98, 2922–2927.
Schimel, D.S., House, J.I., Hibbard, K.A., Bousquet, P., Ciais, P., Peylin, P., Braswell,
B.H., Apps, M.J., Baker, D., Bondeau, A., Canadell, J., Churkina, G., Cramer, W.,
Denning, A.S., Field, C.B., Friedlingstein, P., Goodale, C., Heimann, M., Houghton,
R.A., Melillo, J.M., Moore III, B., Murdiyarso, D., Noble, I., Pacala, S.W., Prentice,
I.C., Raupach, M.R., Rayner, P.J., Scholes, R.J., Steffen, W.L., Wirth, C., 2001. Recent
patterns and mechanisms of carbon exchange by terrestrial ecosystems. Nature
414, 169–172.
Schmid, B., Hector, A., Huston, M.A., Inchausti, P., Nijs, I., Leadley, P.W., Tilman, D.,
2002. The design and analysis of biodiversity experiments. In: Loreau, M.,
Naeem, S., Inchausti, P. (Eds.), Biodiversity and Ecosystem Functioning:
Synthesis and Perspectives. Oxford University Press, Oxford, New York, pp.
66–75.
Simard, M., Pinto, N., Fisher, J.B., Baccini, A., 2011. Mapping forest canopy height
globally with spaceborne lidar. Journal of Geophysical Research 116, G04021.
The Forestry Administration of China, 2003. The Protocol for Forest Inventory in
China, Beijing (unpublished).
van Leeuwen, M., Nieuwenhuis, M., 2010. Retrieval of forest structural parameters
using LiDAR remote sensing. European Journal of Forest Research 129, 749–770.
Wang, X.P., Fang, J.Y., 2012. Constraining null models with environmental
gradients: a new method for evaluating the effects of environmental factors
and geometric constraints on geographic diversity patterns. Ecography. http://
dx.doi.org/10.1111/j.1600-0587.2012.07384.x.
Wang, X.P., Fang, J.Y., Sanders, N.J., White, P.S., Tang, Z.Y., 2009. Relative importance
of climate vs. local factors in shaping the regional patterns of forest plant
richness across Northeast China. Ecography 32, 133–142.
Wang, X.P., Fang, J.Y., Tang, Z.Y., Zhu, B., 2006a. Climatic control of primary forest
structure and DBH–height allometry in Northeast China. Forest Ecology and
Management 234, 264–274.
Wang, X.P., Fang, J.Y., Zhu, B., 2008. Forest biomass and root–shoot allocation in
northeast China. Forest Ecology and Management 255, 4007–4020.
Wang, X.P., Tang, Z.Y., Fang, J.Y., 2006b. Climatic control on forests and tree species
distribution in the forest region of Northeast China. Journal of Integrative Plant
Biology 48, 778–789.
West, G.B., Brown, J.H., Enquist, B.J., 1999. A general model for the structure and
allometry of plant vascular systems. Nature 400, 664–667.
Zhou, Y.L. (Ed.), 1997. Geography of the Vegetation in Northeast China. Science
Press, Beijing.
Zhu, B., 2005. Carbon Stocks of Main Forest Ecosystems in Northeast China. Peking
University, Beijing, Master degree thesis.