allometric growth and allocation in forests: a perspective from fluxnet
TRANSCRIPT
Ecological Applications, 21(5), 2011, pp. 1546–1556� 2011 by the Ecological Society of America
Allometric growth and allocation in forests:a perspective from FLUXNET
ADAM WOLF,1 CHRISTOPHER B. FIELD, AND JOSEPH A. BERRY
Department of Global Ecology, Carnegie Institution for Science, Stanford, California 94305 USA
Abstract. To develop a scheme for partitioning the products of photosynthesis towarddifferent biomass components in land-surface models, a database on component mass and netprimary productivity (NPP), collected from FLUXNET sites, was examined to determineallometric patterns of allocation. We found that NPP per individual of foliage (Gfol), stem andbranches (Gstem), coarse roots (Gcroot) and fine roots (Gfroot) in individual trees is largelyexplained (r2¼ 67–91%) by the magnitude of total NPP per individual (G). Gfol scales with Gisometrically, meaning it is a fixed fraction of G (;25%). Root–shoot trade-offs were manifestas a slow decline in Gfroot, as a fraction of G, from 50% to 25% as stands increased inbiomass, with Gstem and Gcroot increasing as a consequence. These results indicate that afunctional trade-off between aboveground and belowground allocation is essentially capturedby variations in G, which itself is largely governed by stand biomass and only secondarily bysite-specific resource availability. We argue that forests are characterized by strongcompetition for light, observed as a race for individual trees to ascend by increasingpartitioning toward wood, rather than by growing more leaves, and that this competitionstrongly constrains the allocational plasticity that trees may be capable of. The residualvariation in partitioning was not related to climatic or edaphic factors, nor did plots withnutrient or water additions show a pattern of partitioning distinct from that predicted by Galone. These findings leverage short-term process studies of the terrestrial carbon cycle toimprove decade-scale predictions of biomass accumulation in forests. An algorithm forcalculating partitioning in land-surface models is presented.
Key words: allocation of biomass; allometry; data assimilation; FLUXNET; forest stand biomass; land-surface model; net primary productivity, NPP; partitioning of photosynthesis products; tree biomasscompounds; tree plasticity.
INTRODUCTION
The central challenge in modeling the land surface of
the Earth is translating the theoretical and empirical
understanding gained from small-scale process studies
into a model that describes the state and change of
physical quantities at much larger scales. The scaling
problem for terrestrial carbon-cycle science is that
photosynthesis, growth, and mortality are directly
observable only at small spatial scales, such as individual
leaves or trees, and at short temporal scales, especially
for process studies, but the scale of vegetation’s
interaction with climate-change impact is global and
long term. Direct measurement of photosynthesis takes
place on individual leaves or branches in cuvettes, but
grid-cell-scale modeling of photosynthesis requires some
scaling approach to translate leaf-level observations of
fluxes to the canopy scale, such as the widely used ‘‘big
leaf’’ approach to accommodate light gradients within
the canopy (Sellers et al. 1997). The accumulation of
biomass in forest stands is also a challenge to model
because secondary succession can extend well past a
century (Wirth et al. 2009) but process studies of
growth, allocation and turnover typically span a year
or less (Luyssaert et al. 2008), and even forest-inventory
programs rarely have observations extending past a few
decades. Current thought on the terrestrial carbon cycle
(Schulze 2005) suggests that the processes that are most
prominent in the short term, such as photosynthesis or
leaf growth and senescence, are less important determi-
nants to the long-term carbon budget than are processes
that are slow or infrequent, such as decomposition of
coarse woody debris (Harmon et al. 1986) or gap-phase
dynamics (Moorcroft et al. 2001).
We argue in a companion paper (A. Wolf, P. Ciais, V.
Bellassen, N. Delbart, C. B. Field, and J. A. Berry,
unpublished manuscript) that the treatment of vegetation
growth in land-surface models (LSMs), namely the
selective allocation of photosynthate toward different
organs (Fig. 1), has really not adequately addressed the
scaling of biomass from that of individual trees into
stands. The simulation of biomass in LSMs is an
important diagnostic because it represents the realism
of long-term integrals of land–atmosphere carbon fluxes
Manuscript received 16 June 2010; revised 17 September2010; accepted 19 October 2010. Corresponding Editor: D. S.Schimel.
1 Present address: Department of Ecology and EvolutionaryBiology, Princeton University, Princeton, New Jersey, USA.E-mail: [email protected]
1546
that are central to uncertainty in future feedbacks in the
carbon cycle (Friedlingstein et al. 2006). A. Wolf, P.
Ciais, V. Bellassen, N. Delbart, C. B. Field, and J. A.
Berry (unpublished manuscript) found that simulated
biomass in most LSMs is at odds with the large body of
empirical and theoretical work on forest allometry
(West et al. 1999).
Allometry can also play a constructive role in linking
LSMs with remote sensing to constrain estimates of
biomass (Wolf et al. 2010). Many of the forest attributes
that show the strongest allometric interrelationships
with biomass—such as height, population density, and
crown radius—also impact lidar and radar retrievals
(Sun and Ranson 2000, Goodwin et al. 2007) and multi-
angle optical remote sensing (Chopping et al. 2008).
Improving links between modeled and remotely sensed
biomass estimates, such as those from forthcoming
DESDynI, Carbon-3D, and BIOMASS missions (Hese
et al. 2005, Dubayah et al. 2010), would allow improved
monitoring of the state of the biosphere in a rapidly
changing world. However, for this interaction to
proceed, we need stand-level models that can accom-
modate these individual-level data in a self-consistent
manner without bias.
Allometry is a quantitative approach that character-
izes trees in forest stands (e.g., their size, mass, number,
and growth rate), using scaling laws to describe the
properties that emerge as a consequence of the
biophysical constraints of being a tree competing for
survival in a forest. The important concept to gain from
this scaling argument is that size matters: the proportion
of leaf, trunk, and coarse- and fine-root masses in a tree
is not fixed, but changes as the body size of the tree
changes (Enquist and Niklas 2002), and as a forest
matures the number of trees decreases as the mass of the
trees in the stand increase (Enquist et al. 1998, West et
al. 2009). While real forests experience crowding that
results in competition for resources, mortality of
branches and leaves that underperform (‘‘self-pruning’’),
and death of whole trees that have negative carbon
balance (‘‘stand self-thinning’’), LSMs generally lump
these processes into allocation and turnover of abstract
pools that represent the aggregate biomass of the entire
stand. While no model behaved perfectly with the
respect to the observations, the models that did best,
including Orchidee (Bellassen et al. 2009) and ED
(Moorcroft et al. 2001), were those that made the
attempt to treat stands as collections of individuals
subject to allometric constraints.
While the study by A. Wolf, P. Ciais, V. Bellassen, N.
Delbart, C. B. Field, and J. A. Berry (unpublished
manuscript) identified a bias in the biomass patterns
simulated by a variety of global models, it suggested no
means for fixing the problem. Accumulation of biomass
over time is an outcome of both the carbon devoted to
the growth and maintenance of those pools, and the loss
of carbon from those pools by respiration or mortality.
The rate of loss can be conceptualized using a turnover
rate following first-order kinetics. Discerning the frac-
tional allocation of photosynthate to different plant
parts is not easily accomplished from observations of
biomass alone, because the turnover of the pools (leaf,
wood, and fine roots) is different. We will use the term
‘‘partitioning’’ in this paper to refer to the fractional
distribution of a plant’s net primary productivity (NPP)
among different plant parts, where NPP includes both
biomass accumulation and subsequent mortality, but
excludes respiration for growth or maintenance. As
Friedlingstein et al. (1999) point out, woody and
herbaceous vegetation could in principle partition
biomass equally, but the longevity of lignified wood
results in the predominance of the stem pool in forest
biomass. In addition, biomass data alone are inadequate
for estimating partitioning because partitioning is
unlikely to be constant over the life of a tree, due to
the fractal nature of the branches needed to physically
FIG. 1. The role of net primary productivity (NPP) partitioning schemes for biomass component growth in land-surface models.‘‘R’’ stands for respiration, and ‘‘hetero’’ for heterotrophic. Partitioning schemes that determine fractional allocation include (1)fixed proportion, (2) root/shoot competition for resources; (3) seasonal phenology, and (4) size dependent (allometry); seeIntroduction for discussion of partitioning schemes.
July 2011 1547GROWTH ALLOMETRY FROM FLUXNET
support the leaves and fine roots (Mohler et al. 1978,
West et al. 1999).
Friedlingstein et al. (1999) identified several additional
impediments to applying the existing literature on
partitioning to a scheme appropriate for LSMs. The
process-level understanding of partitioning, to the extent
it is understood at all, is essentially at the level of the
individual (Levin et al. 1989, Cannell and Dewar 1994,
Bartelink 1998, Lacointe 2000, Poorter and Nagel 2000,
Ogle and Pacala 2009). Superimposed on this uncer-
tainty of individual partitioning are additional phenom-
ena that are evident at the stand level, such as self-
pruning, self-thinning, and change in species composi-
tion. These community-level phenomena enhance the
mortality of whole trees and tree organs (e.g., branches
and coarse roots) while changing the size distribution of
individuals in the remaining growing tree stock. Such
larger-scale processes do not have to be represented in
models intending to simulate short-term flux data, but
are essential to modeling the carbon budget of a forest
stand that extends for years or decades. Although an
individual-level representation of forest biomass was
discarded decades ago by LSMs to speed up computa-
tion (Running and Gower 1991), it is not necessary (or
reasonable) to discard the allometric constraints or
functional balance of the individual in the pursuit of a
model representing growth and biomass distribution at
the stand level (Purves and Pacala 2008). We base this
on a line of evidence that the scaling of individuals into
stands follows some general ‘‘rules’’: the competition of
individuals for space in a stand leads to predictable
patterns of self-thinning (Enquist et al. 1998), growth
(Niklas and Enquist 2002), and resultant biomass
(Enquist and Niklas 2002) that emerge as a consequence
of structural and vascular constraints imposed on
individuals (West et al. 1999). Because large databases
of stand-level forest inventory repeatedly corroborate
these individual-level scaling patterns, they should be
amenable to use in LSMs that also operate at the stand
level. However, we do not currently have a way of
incorporating these scaling rules into a LSM.
A good partitioning scheme should possess a few key
attributes. First, the partitioning should honor the scale
dependence of partitioning, i.e., that partitioning de-
pends on the characteristic tree size in a stand, and that
the balance of biomass in different organs must conform
to observed allometries. Second, a partitioning scheme
should also distinguish between above ground and
belowground wood, because aboveground wood is
widely measured by forest inventory and facilitates
validation against data and use in data assimilation.
Third, a partitioning scheme must represent fine roots,
because this pool is globally important in the carbon
cycle (Jackson et al. 1997, Matamala et al. 2003), but
largely neglected by allometric scaling theory (Enquist
and Niklas 2002, Niklas and Enquist 2002). Finally,
because most models represent some sort of functional
competition of allocation between leaves and fine roots
in response to limitations in light, water, or nutrients
(e.g., Tilman 1988, Running and Gower 1991), we must
address the relative importance of allometry (i.e., size
dependence of mechanical support) vs. resource limita-
tion in driving variation in partitioning.
The goals of the present study are to use a recently
compiled database on component net primary produc-
tivity (NPP) collected at FLUXNET sites (Luyssaert et
al. 2007), and to use these data to develop a partitioning
scheme that obeys the allometric scaling observed in
nature for use in LSMs. There are several motivations
for adding this study to the literature on allocation and
partitioning. This study is the largest we are aware of to
separately analyze fine roots as a component of
partitioning (cf. Litton et al. [2007] and Enquist and
Niklas [2002]). Fine roots are functionally distinct from
coarse roots because of their dramatically higher
turnover rate, and are generally modeled as a separate
biomass pool in LSMs, so understanding partitioning to
fine roots is essential to be relevant for these modeling
efforts. Also, this study considers the role of stand
thinning in apparent allocation by separately controlling
population density from individual allocation in scaling
from individuals to stands. Finally, we hope that
synthesizing the literature on growth allometry
(Enquist and Niklas 2002) with that of partitioning in
LSMs will give improved confidence that LSMs
realistically represent long-term integrals of forest–
atmosphere carbon flux, which could reduce this source
of uncertainty in predictions of future climate (Fung et
al. 2005, Friedlingstein et al. 2006).
MATERIALS AND METHODS
The FLUXNET program synthesizes data that have
been collected at a large number of sites (.400 sites)
where net ecosystem exchange (NEE) between the
terrestrial biosphere have been intensively measured
and pooled to permit synthesis activities (Baldocchi et
al. 2001, Baldocchi and Valentini 2004, Luyssaert et al.
2007). A subset of these sites were summarized in a
database presented by Luyssaert et al. (2007; hereafter
‘‘the Luyssaert database’’), which compiled annual
fluxes of component net primary productivity (NPP)
separated by foliage, branch, trunk, coarse roots, and
fine roots whenever available. The NPP of these
components includes growth that was subsequently lost
to mortality (e.g., litterfall), such that summing compo-
nent NPP and ecosystem respiration should in principle
equal the NEE measured by eddy covariance; the
closure between bottom-up and top-down estimates is
generally within 5%. Because the Luyssaert database is
compiled from a large number of studies, methods
employed to estimate component NPP vary between
sites, and not all components are available for all sites.
For the purposes of this paper, we included only sites
where all components of NPP (foliage, branch, trunk,
coarse root, fine root) are reported, and where stand
density N is also reported (n ¼ 95 sites). Total NPP per
ADAM WOLF ET AL.1548 Ecological ApplicationsVol. 21, No. 5
individual (G, in kg dry matter�tree�1�yr�1) was calcu-
lated as the sum of all component NPP (kg C�ha�1�yr�1)divided by stand density N (trees/ha): G ¼ (NPPfol þNPPtrunk þ NPPbranch þ NPPcroot þ NPPfroot)/N
with appropriate unit conversions from kilograms C to
kilograms dry matter (DM). Similarly, each component
NPP was calculated from the component NPP per
individual: Gfol ¼ NPPfol/N; Gstem ¼ (NPPtrunk þNPPbranch)/N; Gcroot ¼ NPPcroot/N; Gfroot ¼NPPfroot/N. Total biomass per individual (M, in kg
C) and component biomass per individual (Mfol,
Mstem, Mcroot, Mfroot) was calculated analogously.
Woody biomass is calculated an aggregate of stem and
coarse-root biomass such that Mwood ¼ Mstem þMcroot and Gwood ¼ Gstem þ Gcroot. The conversion
of stand-level values of NPP and biomass to an
individual basis using the stand density (N) follow
Enquist and Niklas (2002) and Niklas and Enquist
(2002) for their allometric analysis of the Cannell (1982)
database. While this approach elides the variation within
stands of tree size and size-related growth, it facilitates
comparison between stands differing in population
density on the basis of the biomass and growth of the
‘‘average’’ tree.
Among the 95 useable sites, 27 sites are dominated by
angiosperms (Fig. 2: open circles), and 68 sites are
dominated by gymnosperms (Fig. 2: solid circles). The
sites are generally located at temperate (n¼ 71 sites) and
boreal (n ¼ 20 sites) regions, with few in tropical (n ¼ 3
sites) and Mediterranean (n ¼ 1 site) regions. The sites
were given a variety of codes to characterize manage-
ment of the stands; most stands were managed forests (n
¼48 sites), many were natural forests (n¼20 sites), some
stands were categorized as recently disturbed (n ¼ 8
sites), a small number were given fertilizer or irrigation
(n¼ 9 sites), and for some no information was available
(n¼10 sites). For the sake of comparison, of the 63 plots
from Litton et al. (2007), only 14 plots are included here:
3 plots from Ryan et al., (1996), 4 plots fromMaier et al.
(2004), 2 plots from Law et al. (2001), and 5 plots from
Curtis et al. (2002). The remainder either lacked data on
fine roots or were unavailable for reference as Master’s
theses or book chapters. The Maier et al. (2004) and
Ryan et al. (1996) studies represent seven of the nine
fertilization/irrigation plots included in this study, with
the remaining two control and treatment plots reported
in Linder and Agren (1998) and Gower et al. (2001),
respectively. The fertilized and irrigated plots are
included with all other plots in the regression analysis,
but their departures from the mean slope will be
additionally discussed as the only direct measure of the
impact of nutrient and water additions on partitioning.
Direct measurements of soil nutrients are not otherwise
available in the Luyssaert database. Finally, in one study
(Gielen et al. 2005), three irrigated control plots
included in the regression calculation have three
corresponding CO2-enriched plots, which will be shown
for comparison.
A range of techniques were employed to measure fine
root NPP, including ‘‘higher quality’’ measurements of
biomass and in situ turnover using minirhizotrons or
root windows (n¼ 54 sites), ‘‘modest quality’’ techniques
such as biomass with an assumed turnover rate (n ¼ 18
sites) or root ingrowth techniques (n ¼ 3 sites), and
‘‘lower quality’’ techniques based on sequential coring (n
¼ 16 sites). Some sites used techniques for fine-root NPP
that were ambiguous from the study text (n ¼ 4 sites).
The majority of studies to estimate coarse-root NPP
were derived from allometric considerations (n ¼ 64
sites), with the remaining sites deriving coarse-root NPP
from biomass and in situ turnover observations (n ¼ 9
sites), ingrowth techniques (n¼4 sites), sequential coring
(n ¼ 17 sites), and one site with unknown methodology
(n ¼ 1 site). Because the coarse-root biomass and NPP
are largely derived from allometric relationships with
stem biomass and NPP, readers should be cautious in
interpretating Gcroot and Mcroot separately from
Gstem and Mstem. Readers are encouraged to refer to
Luyssaert et al. (2007) for more detail on the methods
used to compile the data hierarchically, as well as for all
citations to the original data comprising the database.
The world forest-production database compliled by
Cannell (Cannell 1982; hereafter ‘‘the Cannell data-
base’’) is included here as a benchmark database to
facilitate comparison with other work in this area. The
Cannell database reports stand biomass and NPP, with
additional details on stand age, diameter, height,
density, and basal area where available. Biomass and
NPP are reported separately for foliage, branches, bark,
stem, reproductive structures, and roots. NPP whenever
possible accounts for mortality such as litterfall. Note
that the Cannell database does not distinguish growth or
mass of fine roots from coarse roots, so Mroot and
Groot in the Cannell database reflects the mass and
growth rate, respectively, of the aggregate root pool.
Global relationships with climate were evaluated
using the CRU (Climate Research Unit) high-resolution
climatology from 1961–1990 used in the IPCC AR3
(Intergovernmental Panel on Climate Change, Third
Assessment Report [2001], available online)2 (also see
Mitchell and Jones 2005). These data include mean
monthly maximum, minimum, and average tempera-
ture, wet days, frost-free days, precipitation, water
vapor content, diurnal temperature range, and cloud
cover. These data were converted to representative
annual values expressing climate limitations, including
annual maximum vapor pressure deficit, minumum
annual relative humidity, annual frost-free days, annual
precipitation, and Thornthwaite’s aridity index (P –
PET)/PET, where P is precipitation and PET is
potential evapotranspiration calculated using the
Thornthwaite equation (Thornthwaite 1948). Soil tex-
ture was used as an integrated measure of soil resources
2 hhttp://www.ipcc-data.org/obs/cru_ts2_1.htmli
July 2011 1549GROWTH ALLOMETRY FROM FLUXNET
such as nutrients and aeration. Texture data were
gathered from the 0.0833 degree IGBP (International
Geosphere–Biosphere Programme) soil database
(Global Soil Data Task Force 2000).
Allometric scaling employs a power-law relationship
of the form Y ¼ aXb. Taking the logarithm to this
function yields the equation log(Y )¼ log(a)þ blog(X ),
whose parameters can be solved by linear regression.
The scaling parameters were estimated using Type II
(reduced major axis [RMA]) regression (Sokal and
Rohlf 1995: section 14.13). Most of the scaling relations
explored in this paper represent the scaling of NPP of a
component of a tree in relation to the whole, for
example Gfol } G. In this case, the solved parameters
represent the fractional allocation of net primary
productivity to that component, i.e., the partitioning.
Due to noise in the data, the parameters estimated in
this way do not sum identically to 1 (i.e., fractions of
total annual NPP), and were therefore renormalized.
RESULTS
The results presented below show first the scaling of
component NPP (net primary productivity) to total
NPP (G) using regression analysis, then an analysis of
the residuals against climatic and edaphic factors. We
then compare the allometry drawn from the Luyssaert
FIG. 2. Partitioning toward (A) foliage, (B) stem (i.e., trunk and branches) (C) fine roots, and (D) coarse roots. Both Type I(solid line: least squares, LS) and Type II (dashed line: reduced major axis, RMA) regressions are shown. G is total NPP (netprimary productivity) per individual (in kg dry matter [DM]�tree�1�yr�1).
ADAM WOLF ET AL.1550 Ecological ApplicationsVol. 21, No. 5
database (Luyssaert et al. 2007) with those from the
benchmark Cannell database (Cannell 1982). Finally we
integrate the partitioning results at the individual level
to the stand level using G : M (where M is total biomass
per individual [in kg C]) andM : N allometry (where N is
stand density [in trees/ha]). The results conclude with an
algorithm for calculating partitioning in land-surface
models (LSMs).
The partitioning of G among the different biomass
pools is plotted in Fig. 2 (abbreviations: fol, foliage;
croot, coarse roots; and froot, fine roots). The linear
regressions on log-transformed variables assume that the
component NPP follows a power law that falls to 0 as G
goes to 0, a functional relationship that is widely used in
the allometry literature, and is intended to capture
changes in the response variable over several orders of
magnitude. Gfol is isometric with G (bGfol ¼ 1.106 6
0.054, r2 ¼ 0.914, n ¼ 95 sites). Isometric scaling means
that the scaling exponent is not significantly different
from 1, and that Gfol and G vary in a fixed proportion,
as predicted by Niklas and Enquist (2002). By contrast,
allocation toward woody biomass pools (Gstem and
Gcroot) increases with G (bGstem ¼ 1.165 6 0.045, r2 ¼0.942, n¼95 sites; bGcroot¼1.310 6 0.045, r2¼0.953, n¼95 sites). Note that the scalings of Gstem and Gcroot are
not significantly different from one another, indicating
that the ratio of partitioning to aboveground and
belowground woody biomass does not change with G.
The fraction of Gwood (i.e., GstemþGcroot) devoted to
Gstem is estimated as 0.808 (Gstem¼0.8083Gwood0.987
(r2 ¼ 0.995, n ¼ 95 sites; 0.822 ¼ 10�0.085). As a
consequence of the increase in partitioning to Gwood,
the partitioning to Gfroot drops as G increases (bGfroot¼0.987 6 0.078, r2 ¼ 0.821, n ¼ 95 sites).
The regressions of each component G against G show
that the majority (70–93%) of all variation can be
explained by the magnitude of G itself. The residuals of
component G (Gfol, Gstem, Gcroot, Gfroot vs. G) had
no significant or systematic relationship with nutrient or
water additions within the studies that had these
treatments (two-sided t tests nonsignificant), although
G itself increased in all plots with resource additions as
would be expected (Fig. 3). The study that included a
CO2 enrichment treatment (Gielen et al. 2005) showed a
slight increase in fine-root allocation, but this study had
some of the largest residuals overall (Fig. 3B–D),
perhaps because these data were collected from a 1–2
year old plantation of Populus spp. grown with drip
irrigation on agricultural land, which was considerably
different from the other forests in terms of age and
management. The residuals were regressed against
climate variables from the CRU (Climate Research
Unit) data set (Mitchelll and Jones 2005), and against
the aridity index, and none were found to be signifi-
cantly related to the unexplained variation in compo-
nent G. Likewise, the residuals were not significantly
related to soil texture. The residuals of log(Gfol) and
log(Gfroot) were uncorrelated with one another (r2 ¼
0.03). Together these results suggest that there is not a
trade-off in partitioning toward foliage and fine roots,
and that this putative trade-off, at the stand level, is
furthermore not related to climatic resources such as
aridity, season length, or light availability, nor soil
resources linked to texture. Other attributes of stands
such as taxonomic affiliation (angiosperm vs. gymno-
sperm), site management (managed vs. unmanaged), and
stand age were not significantly related to the unex-
plained variation in component G. The strong relation-
ship of partitioning with G itself, and the independence
of partitioning from climate and taxon indicate that at
the resolution of this study, partitioning is a relatively
conservative, non-plastic attribute, and is consequently
generalizeable for trees of given NPP.
The allometric scaling of mass and NPP among
different biomass components is not significantly differ-
ent in the Luyssaert database from the Cannell database
(Appendix Table A1), suggesting that the essential
features of allometry in the Luyssaert database are
consistent with previous studies (Enquist and Niklas
2002, Niklas and Enquist 2002). Although labile
biomass components (Mfol and Mfroot) have scaling
exponents against woody biomass (Mstem and Mcroot)
that are significantly less than 1 (Appendix Table A1),
the fractional distribution of mass among different
biomass components is fairly constant with total tree
size M because all biomass components asymptote fairly
quickly (Fig. 4A and D; regressions shown in Appendix
Fig. A1). After an initial condition dominated by labile
components of biomass, the stands quickly transition to
a state where the overwhelming majority of biomass is
woody. Mstem and Mcroot scale in fixed proportion of
approximately 3:1 (Mcroot ¼ 0.246 3 Mstem1.014, r2 ¼0.99, n ¼ 37 sites), although it should be cautioned that
most of the coarse-root biomass data in the Luyssaert
database is allometrically derived from stem-biomass
data.
The rates of component G at the stand level were
estimated by estimating M as a function of G (M¼5.357
3G1.575, r2¼ 0.70, n¼ 40 sites) and N as a function of M
(N ¼ 18 793 3 M�0.552, r2 ¼ 0.67, n ¼ 42 sites), which
invert metabolic rate–body size allometry and self-
thinning allometry, respectively. Note that the correla-
tion between G and M, and M to N is not nearly as
strong as the correlations between component G (Gfol,
Gstem, Gcroot, Gfroot) and G, so stand biomass is a
weaker predictor of partitioning than G, and therefore
factors that impact G, such as favorable soils or climate,
have a stronger influence on partitioning than does tand
biomass per se. Nevertheless, the consequence of
upscaling to the stand level by including the strong
decline of population density with mass accumulation
(Fig. 4C and F) is to enhance the drop-off of
partitioning to fine roots for stands of modest biomass
(300–500 MG DM/ha), which creates stronger differ-
ences among stands for biomass devoted to fine roots.
July 2011 1551GROWTH ALLOMETRY FROM FLUXNET
By contrast, partitioning to foliage at the stand level is
nearly constant for all stand sizes.
These results suggest the hypothesis that an algorithm
for calculating partitioning in land-surface models
(LSMs) would take the following form:
1) Calculate NPP.
2) Calculate N (population density). If N is not
simulated in the model, diagnose N from M 3 N by
inverting the self-thinning relationship (equations in
previous paragraph).
3) Calculate G ¼NPP/N.
4) Calculate Gfol, Gstem, Gcroot, Gfroot from G
(equations in Fig. 2).
5) Renormalize Gfol, Gstem, Gcroot, Gfroot so that
they sum to G.
DISCUSSION
This paper synthesizes the partitioning data from a
large, detailed database of component net primary
productivity, NPP, with an emphasis on identifying the
scale dependence of component NPP and trade-offs in
NPP among various plant organs, particularly fine
roots, and explicitly accounting for stand self-thinning
in accounting for variation among stands. Particular
results presented above are in agreement with a variety
of previous studies, but which have not been synthesized
in a form useful for application to efforts to model the
FIG. 3. Residuals of component growth from the RMA regressions of Fig. 2, emphasizing studies with fertilizer, water, or CO2
additions: (A) foliage, (B) stem (i.e., trunk and branches), (C) fine roots, (D) coarse roots. Original units for G (individual NPP)are kg DM�tree�1�yr�1; RMA stands for reduced major axis; r is a correlation coefficient, n is sample size. FACE refers to studiesincluding free-air carbon enrichment (e.g., Gielen et al. 2005). Symbol color identifies the data sources: brown, Maier et al.(2004); blue, Ryan et al. (1996); red, Gielen et al. (2005); green, Lender and Agren (1998) and Gower et al. (2001). The small solidgray circles represent studies with no manipulation.
ADAM WOLF ET AL.1552 Ecological ApplicationsVol. 21, No. 5
global biosphere. For example, the component allome-
try from this study is not significantly different
(Appendix Table A1) from the component mass
allometry in Enquist and Niklas (2002) or the compo-
nent allometry of NPP in Niklas and Enquist (2002).
However, neither of those studies considered fine-root
mass or growth, which greatly limits their applicability
in carbon-cycle science because this pool is critically
important in the global carbon cycle (Jackson et al.
1997). The estimate of partitioning to Gfroot estimated
here is similar to the value of 1/3 estimated globally by
Jackson et al. (1997), but their value was not based on
direct measurement, but inferred from biomass and an
assumed turnover rate of once per year.
The results of our present study are likewise consistent
with the findings of Litton et al. (2007), who also
presented a synthesis of stand-level partitioning (Fig. 4C
and F), with the caveat that few of the studies they cite
include fine-root data, so they were compelled to group
fine roots and coarse roots as an aggregate ‘‘below-
ground’’ pool. Litton et al. (2007) found that stand-level
individual-tree foliage, Gfol, was fixed at ;25% for
stands of all sizes, which is supported by the data here.
Also, Litton et al. (2007) show a trade-off in partitioning
between Gstem and belowground components (GcrootþGfroot): stands with low NPP partitioned ;50% of their
NPP belowground and ;25% to Gstem, and with higher
stand-level NPP, this proportion flipped to 25%belowground and 50% Gstem. Our results show that
fast-turnover fine roots (‘‘froot’’) and woody coarse
roots (‘‘croot’’) are functionally distinct in this context,
and that the trade-off is apparently between Gwood and
Gfroot (Fig. 4C), and not Gstem and Groot as suggested
by Litton et al.’s (2007) results. This result is intriguing
because it suggests that in forests, the functional trade-
off for aboveground and belowground resources that is
so widely postulated (Tilman 1988), is manifest not as a
trade-off between root and shoot allocation per se (e.g.,
Levin et al. 1989), nor between foliage and fine roots
(e.g., Friedlingstein et al. 1999, Poorter and Nagel 2000),
but as a trade-off between the woody shoot and roots.
The Gfroot�Gwood trade-off appears to be a basic
feature of forest stand ontogeny, largely explained by
changes in the magnitude of G, which generally increases
as a forest accumulates biomass. Only secondarily does
this trade-off appear to be driven by changes in G that
FIG. 4. (A, D) Fractional distribution of mass and (B, C, E, F) fractional partitioning of growth. Upper plots are modeleddistributions of fractional component mass per individual, fractional component growth per individual, and fractional componentgrowth vs. stand biomass, and were developed from power-law relationships. Lower plots show original data with modeledfunctions superimposed.
July 2011 1553GROWTH ALLOMETRY FROM FLUXNET
are mediated by site resources independently from forest
size. To the extent that climate, site resources, and even
fertilization and irrigation change partitioning, we found
no evidence that these factors have effects beyond those
attributable to their impact on G itself (Fig. 3).
Clearly, plants in general are hugely plastic in their
partitioning towards roots, shoots, and foliage, resulting
in wide differences in root : shoot ratios as an adapta-
tion to local site conditions (Mokany et al. 2006), which
corroborates the basic validity of the functional trade-
off hypothesis (Tilman 1988). Nevertheless, our present
study found that the pattern of partitioning responds
consistently to the large variety of site conditions that
could enhance or reduce NPP, including fertilizer and
irrigation, which suggests that forests are fairly con-
strained in the variety of partitioning strategies that are
successfully employed. Why should this constrained set
of strategies basically consist of increasing partitioning
to stem and reducing partitioning to fine roots when
NPP is large? The net primary productivity of forests is
well above the mean NPP of terrestrial biomes
worldwide (Baldocchi and Valentini 2004, Luyssaert et
al. 2008). That is, forests are a biome that occurs in
places that are relatively free of growth constraints
climatically and edaphically. The findings in this study
indicate that in such a rich resource environment,
competition for light is likely driving trees to allocate
evermore photosynthate toward the woody stem in a
race to ascend. In this context, more foliage per se is
apparently not as important as lofting the foliage into
the sunlit canopy, because the allocation toward foliage
itself is a constant fraction of NPP. Therefore, while
stem biomass (Mstem) per se has no ‘‘functional’’ benefit
in terms of resource capture, it is critically important in
helping to realize the potential resource capture of the
foliage. Indeed, the biomass scaling of Mfol } Mstem3/4
(Enquist and Niklas 1998) shows that trees need to
create an ever larger amount of woody stem to support a
given amount of foliage: Mstem } Mfol4/3. This observa-
tion has been borne out in a number of studies using
different datasets (e.g., Wolf et al. 2010, Cheng and Niklas
2007), but importantly all of these studies are derived from
stand-level inventory data. The Mfol :Mstem scaling in
these studies is therefore only partly a consequence of the
vasculature and mechanical support needed to maintain a
given mass of leaves, and indeed the mass of branches is a
small fraction of total aboveground woody biomass
(Mwood). The stands in this study (and forests generally)
become taller over the course of stand development (H
[height] scales positively withM ), even though the mass of
leaves is constant at the stand level. It is perhaps significant
in this context that althoughWest et al. (1999) predict that
H should scale to the ¼ power of M, but instead grow
substantially taller in the Cannell database for a given
biomass:H } Mstem0.346 (95% CI: 0.321–0.373, r¼0.82, n
¼ 315 sites). We conclude from this that the allometric
scaling of biomass (Enquist andNiklas 1998) is only partly
a consequence of the vascular arguments made in that
paper, but largely a consequence of competition between
trees leading to major investments in stem biomass.
Together, the results presented above indicate that
incorporation of an allometrically sound partitioning
scheme in LSMs is straightforward, because the com-
plications introduced by differences in site potential and
other factors that may affect partitioning but that are
not represented in LSMs are essentially captured by
variations in NPP, which is ubiquitous in LSMs. This is
an incremental step in reconciling the biomass allometry
simulated in land-surface models with that seen in nature
(A. Wolf, P. Ciais, V. Bellassen, N. Delbart, C. B. Field,
and J. A. Berry, unpublished manuscript). However, this
still leaves open the issue of representing population
density in LSMs, because few models have any
representation of an individual. In addition, the recon-
ciliation of mass allometry between models is not made
possible simply by coming up with an allometric
partitioning scheme, because the turnover times of
leaves, wood, and fine roots is an additional free
parameter that may be in error in LSMs. The turnover
times (M/G from Appendix Fig. A1 divided by Fig. 2)
for fine roots and gymnosperm foliage in the Luyssaert
database is consistent with literature values, and has a
mean of ;3 years (Appendix Fig. A2). The woody
biomass, by contrast, shows a large increase in turnover
times of 15–45 years without reaching an asymptote.
This could be interpreted as a decrease in the lability of
woody tissue as branches and trunks become coarser in
size and decrease in rate mortality as the stand matures
(e.g., Ishii and McDowell 2002). Regardless, the absence
of a constant turnover time for woody tissue makes the
goal of simulating a forest with the proper biomass
allometry a greater challenge than simply calculating the
partitioning correctly, as implemented by the algorithm
presented in the Results. For the given partitioning
scheme to result in the patterns of biomass in Appendix
Fig. A2, a model would need to adjust the turnover
times in order to keep on the biomass accumulation
trajectory shown in the regressions. To this end, it might
be possible to implement a version of data assimilation
that adjusted the turnover times of biomass pools
dynamically, in order to align better with the inventory
data, as summarized by their allometric scaling equa-
tions. Finally, our conclusion that partitioning is
essentially predicted by total NPP places a premium on
understanding controls on NPP, about which there is
still considerable uncertainty among LSMs (Beer et al.
2010).
LITERATURE CITED
Baldocchi, D., et al. 2001. FLUXNET: a new tool to study thetemporal and spatial variability of ecosystem-scale carbondioxide, water vapor, and energy flux densities. Bulletin ofthe American Meteorological Society 82:2415–2434.
Baldocchi, D., and R. Valentini. 2004. Geographic andtemporal variation of carbon exchange by ecosystems andtheir sensitivity to environmental perturbations. Pages 295–316 in C. B. Field and M. R. Raupach, editors. The global
ADAM WOLF ET AL.1554 Ecological ApplicationsVol. 21, No. 5
carbon cycle: integrating humans, climate and the naturalworld. Island Press, Washington, D.C., USA.
Bartelink, H. H. 1998. A model of dry matter partitioning intrees. Tree Physiology 18: 91–101.
Beer, C., et al. 2010. Terrestrial Gross Carbon Dioxide Uptake:Global Distribution and Covariation with Climate. Science329(5993):834–838.
Bellassen, V., G. Le Maire, J. F. Dhote, P. Ciais, and N. Viovy.2009. Modelling forest management within a global vegeta-tion model—Part 1. Model structure and standard results.Ecological Modelling 221:2458–2474.
Cannell, M. G. R. 1982. World forest biomass and primaryproduction data. Academic Press, London, UK.
Cannell, M. G. R., and R. C. Dewar. 1994. Carbon allocationin trees—a review of concepts for modeling. Advances inEcological Research 25:59–104.
Cheng, D. L., and K. J. Niklas. 2007. Above- and below-ground biomass relationships across 1534 forested commu-nities. Annals of Botany 99:95–102.
Chopping, M., G. G. Moisen, L. Su, A. Laliberte, A. Rango,J. V. Mortonchik, and D. P. C. Peters. 2008. Large areamapping of southwestern forest crown cover, canopy height,and biomass using the NASA multiangle imaging spectro-radiometer. Remote Sensing of Environment 112:2051–2063.
Curtis, P. S., P. J. Hanson, P. Bolstad, C. Barford, J. C.Randolph, H. P. Schmid, and K. B. Wilson. 2002. Biometricand eddy-covariance based estimates of annual carbonstorage in five eastern North American deciduous forests.Agricultural and Forest Meteorology 113:3–19.
Dubayah, R. O., S. L. Sheldon, D. B. Clark, M. A. Hofton,J. B. Blair, G. C. Hurtt, and R. L. Chazdon. 2010. Estimationof tropical dorest height and biomass dynamics using lidarremote sensing at La Selva, Costa Rica. Journal ofGeophysical Research 115, G00E09.
Enquist, B. J., J. H. Brown, and G. B. West. 1998. Allometricscaling of plant energetics and population density. Nature395(6698):163–165.
Enquist, B. J., and K. J. Niklas. 2002. Global allocation rulesfor patterns of biomass partitioning in seed plants. Science295(5559):1517–1520.
Friedlingstein, P. G., et al., 2006. Climate–carbon cyclefeedback analysis: results from the C4MIP model intercom-parison. Journal of Climate 19:3337–3353.
Friedlingstein, P., G. Joel, C. B. Field, and I. Y. Fung. 1999.Toward an allocation scheme for global terrestrial carbonmodels. Global Change Biology 5:755–770.
Fung, I. Y., S. C. Doney, K. Lindsay, and J. John. 2005.Evolution of carbon sinks in a changing climate. Proceedingsof the National Academy of Sciences USA 102(32):11201–11206.
Gielen, B., et al., 2005. Net carbon storage in a poplarplantation (POPFACE) after three years of free-air CO2
enrichment. Tree Physiology 25:1399–1408.Global Soil Data Task Force. 2000. Global soil data products.CD-ROM. International Geosphere-Biosphere Programme,Data and Information System, Potsdam, Germany.
Goodwin, N. R., N. C. Coops, and D. S. Culvenor. 2007.Development of a simulation model to predict LiDARinterception in forested environments. Remote Sensing ofEnvironment 111:481–492.
Gower, S. T., O. Krankina, R. J. Olson, M. Apps, S. Linder,and C. Wang. 2001. Net primary production and carbonallocation patterns of boreal forest ecosystems. EcologicalApplications 11:1395–1411.
Harmon, M. E., J. F. Franklin, F. J. Swanson, P. Sollins, S. V.Gregory, J. O. Lattin, J. R. Anderson, G. W. Lienkaemper,K. Cromack, and K. W. Cummins. 1986. Ecology of coarsewoody debris in temperate ecosystems. Advances in Ecolog-ical Research 15:133–302.
Hese, S., W. Lucht, C. Schmullius, M. Barnsley, R. Dubayah,D. Knorr, K. Neumann, T. Riedel, and K. Schroter. 2005.
Global biomass mapping for an improved understanding ofthe CO2 balance—the Earth observation mission Carbon-3D.Remote Sensing of Environment 94:94–104.
Ishii, H., and N. McDowell. 2002. Age-related development ofcrown structure in coastal Douglas-fir trees. Forest Ecologyand Management 169:257–270.
Jackson, R. B., H. A. Mooney, and E. D. Schulze. 1997. Aglobal budget for fine root biomass, surface area, andnutrient contents. Proceedings of the National Academy ofSciences USA 94(14):7362–7366.
Lacointe, A. 2000. Carbon allocation among tree organs: Areview of basic processes and representation in functional-structural tree models. Annals of Forest Science 57(5–6):521–533.
Law, B. E., P. E. Thornton, J. Irvine, P. M. Anthoni, and S.Van Tuyl. 2001. Carbon storage and fluxes in ponderosa pineforests at different developmental stages. Global ChangeBiology 7:755–777.
Levin, S. A., H. A. Mooney, and C. Field. 1989. Thedependence of plant root:shoot ratios on internal nitrogenconcentration. Annals of Botany 64:71–75.
Linder, S., and G. I. Agren. 1998. NPP boreal forest sites. OakRidge National Laboratory Distributed Active ArchiveCenter, Oak Ridge, Tennessee, USA. hhttp://www.daac.ornl.gov.i
Litton, C. M., J. W. Raich, and M. G. Ryan. 2007. Carbonallocation in forest ecosystems. Global Change Biology13:2089–2109.
Luyssaert, S., et al., 2007. CO2 balance of boreal, temperate,and tropical forests derived from a global database. GlobalChange Biology 13:2509–2537.
Luyssaert, S., E. D. Schulze, A. Borner, A. Knohl, D.Hessenmoller, B. E. Law, P. Ciais, and J. Grace. 2008. Old-growth forests as global carbon sinks. Nature 455(7210):213–215.
Maier, C. A., T. J. Albaugh, H. L. Allen, and P. M. Dougherty.2004. Respiratory carbon use and carbon storage in mid-rotation loblolly pine (Pinus taeda L.) plantations: the effectof site resources on the stand carbon balance. Global ChangeBiology 10:1335–1350.
Matamala, R., M. A. Gonzalez-Meler, J. D. Jastrow, R. J.Norby, and W. H. Schlesinger. 2003. Impacts of fine rootturnover on forest NPP and soil C sequestration potential.Science 302(5649):1385–1387.
Mitchell, T. D., and P. D. Jones. 2005. An improved method ofconstructing a database of monthly climate observations andassociated high-resolution grids. International Journal ofClimatology 25:693–712.
Mohler, C. L., P. L. Marks, and D. G. Sprugel. 1978. Standstructure and allometry of trees during self-thinning of purestands. Journal of Ecology 66:599–614.
Mokany, K., J. R. Raison, and A. S. Prokushkin. 2006. Criticalanalysis of root : shoot ratios in terrestrial biomes. GlobalChange Biology 12:84–96.
Moorcroft, P. R., G. C. Hurtt, and S. W. Pacala. 2001. Amethod for scaling vegetation dynamics: the ecosystemdemography model (ED). Ecological Monographs 71:557–586.
Niklas, K. J., and B. J. Enquist. 2002. On the vegetativebiomass partitioning of seed plant leaves, stems, and roots.American Naturalist 159:482–497.
Ogle, K., and S. W. Pacala. 2009. A modeling framework forinferring tree growth and allocation from physiological,morphological and allometric traits. Tree Physiology 29:587–605.
Poorter, H., and O. Nagel. 2000. The role of biomass allocationin the growth response of plants to different levels of light,CO2, nutrients and water: a quantitative review. AustralianJournal of Plant Physiology 27:595–607.
Purves, D., and S. Pacala. 2008. Predictive models of forestdynamics. Science 320(5882):1452–1453.
July 2011 1555GROWTH ALLOMETRY FROM FLUXNET
Running, S. W., and S. T. Gower. 1991. Forest-BGC, a general-model of forest ecosystem processes for regional applications.II. Dynamic carbon allocation and nitrogen budgets. TreePhysiology 9(1-2):147–160.
Ryan, M. G., R. M. Hubbard, S. Pongracic, R. J. Raison, andR. E. McMurtrie. 1996. Foliage, fine-root, woody-tissue andstand respiration in Pinus radiata in relation to nutrientstatus. Tree Physiology 16:333–343.
Schulze, E. D. 2005. Biological control of the terrestrial carbonsink. Biogeosciences Discussions 2:1283–1329.
Sellers, P. J., et al. 1997. Modeling the exchanges of energy,water, and carbon between continents and the atmosphere.Science 275(5299): 502–509.
Sokal, R. R., and F. J. Rohlf. 1995. Biometry: the principlesand practice of statistics in biological research. Third edition.W.H. Freeman and Company, New York, New York, USA.
Sun, G. Q., and K. J. Ranson. 2000. Modeling lidar returnsfrom forest canopies. IEEE Transactions on Geoscience andRemote Sensing 38:2617–2626.
Thornthwaite, C. W. 1948. An approach toward a rationalclassification of climate. Geographical Review 38:55–94.
Tilman, D. 1988. Plant strategies and the dynamics andstructure of plant communities. Monographs in populationbiology 26. Princeton University Press, Princeton, NewJersey, USA.
West, G. B., J. H. Brown, and B. J. Enquist. 1999. A generalmodel for the structure and allometry of plant vascularsystems. Nature 400(6745):664–667.
West, G. B., B. J. Enquist, and J. H, Brown. 2009. A generalquantitative theory of forest structure and dynamics.Proceedings of National Academy of Science USA106(17):7040–7045.
Wirth, C., G. Gleixner, and M. Heimann. 2009. Old-growthforests: function, fate and value. Ecological Studies 27.Springer-Verlag, New York, New York, USA.
Wolf, A., J. A. Berry, and G. P. Asner. (2010). Allometricconstraints on sources of variability in multi-angle reflectancemeasurements. Remote Sensing of Environment 114:1205–1219.
APPENDIX
A table comparing regression coefficients for allometric relationships between tree biomass components in different databases, afigure showing regressions for scaling of component mass with total tree mass, and a figure showing turnover times for componentsof biomass from the Luyssaert database (Ecological Archives A021-072-A1).
ADAM WOLF ET AL.1556 Ecological ApplicationsVol. 21, No. 5