Role of tree size in moist tropical forest carbon cycling and water
deficit responses
Victoria Meakem1, Alan J. Tepley1, Erika B. Gonzalez-Akre1, Valentine Herrmann1, Helene C. Muller-Landau2,
S. Joseph Wright2, Stephen P. Hubbell2,3, Richard Condit2 and Kristina J. Anderson-Teixeira1,2
1Conservation Ecology Center, Smithsonian Conservation Biology Institute, Front Royal, VA 22630, USA; 2Center for Tropical Forest Science, Smithsonian Tropical Research Institute, Balboa
Ancon, Panama, Republic of Panama; 3Department of Ecology and Evolutionary Biology, University of California, Los Angeles, Los Angeles, CA 90095, USA
Author for correspondence:
Kristina J. Anderson-Teixeira
Tel: +1 540 635 6546
Email: teixeirak@si.edu
Received: 3 March 2017
Accepted: 27 April 2017
New Phytologist (2017)
doi: 10.1111/nph.14633
Key words: biomass, drought adaptation, El
Ni~no, tree mortality, tree size, tropical moist
forest, woody productivity.
Summary

 Drought disproportionately affects larger trees in tropical forests, but implications for forest
composition and carbon (C) cycling in relation to dry season intensity remain poorly under-
stood.

 In order to characterize how C cycling is shaped by tree size and drought adaptations and
how these patterns relate to spatial and temporal variation in water deficit, we analyze data
from three forest dynamics plots spanning a moisture gradient in Panama that have experi-
enced El Ni~no droughts.

 At all sites, aboveground C cycle contributions peaked below 50-cm stem diameter, with
stems ≥ 50 cm accounting for on average 59% of live aboveground biomass, 45% of woody
productivity and 49% of woody mortality. The dominance of drought-avoidance strategies
increased interactively with stem diameter and dry season intensity. Although size-related C
cycle contributions did not vary systematically across the moisture gradient under nondrought
conditions, woody mortality of larger trees was disproportionately elevated under El Ni~no
drought stress.

 Thus, large (> 50 cm) stems, which strongly mediate but do not necessarily dominate C
cycling, have drought adaptations that compensate for their more challenging hydraulic envi-
ronment, particularly in drier climates. However, these adaptations do not fully buffer the
effects of severe drought, and increased large tree mortality dominates ecosystem-level
drought responses.
Introduction
Tropical forests play critical roles in the global carbon and cli-
mate cycles. They contain an estimated 34% of terrestrial carbon
(C; US DOE, 2012), account for 34% of global gross primary
productivity (Beer et al., 2010), and influence climate on local to
global scales through their high rates of evapotranspiration (Sny-
der et al., 2004; Lawrence & Vandecar, 2015). Across the tropics,
anticipated changes in the spatial and temporal availability of
water (IPCC, 2013) are expected to alter the ecophysiology and
composition of forests, with consequent feedbacks to the climate
system. Given the importance of tropical forests to the climate
system, these feedbacks may be quite significant, yet our under-
standing of tropical forest responses to water deficit remains lim-
ited (e.g. Sitch et al., 2008; Huntingford et al., 2013; Powell
et al., 2013; Anderegg et al., 2015; Corlett, 2016). To precisely
describe or predict forest responses to climate variation or change,
it is necessary to characterize both the exact relationship between
tree size and C cycling, and the differential responses of trees of
different sizes to hydraulic stress.
Despite a common conception that larger trees dominate trop-
ical forest C cycling (e.g. Fauset et al., 2015; Bastin et al., 2015),
there is a surprising paucity of studies quantifying in detail how
contributions to key components of forest C budgets – specifi-
cally, live aboveground biomass (Cag,live), woody productivity
(ANPPstem) and woody mortality (M) – vary as a function of tree
size. The proportional contribution of large trees to total biomass
is variable, with contributions of trees ≥ 70 cm diameter at breast
height (DBH) ranging up to 45% in neotropical forests (Schietti
et al., 2016, and reference therein). With regards to biomass
turnover (i.e. ANPPstem and M), some theoretical work suggests
energy or C-flux equivalence across size classes (Enquist et al.,
2009), but these predictions do not account for differences in
light availability to the different size classes (Muller-Landau et al.,
2006a) or deviations from power-law scaling of size–abundance
relationships (Coomes et al., 2003; Muller-Landau et al., 2006b).
The exact form of the relationship between tree size and ecosys-
tem-level C contributions in tropical forests has yet to be
described, and we do not know how this relationship is shaped
by climate.

 2017 Smithsonian Institute
New Phytologist 
 2017 New Phytologist Trust
New Phytologist (2017) 1
www.newphytologist.com
Research
Trees of different sizes respond differently to variation in
hydraulic conditions; in tropical forests worldwide, drought tends
to have a greater impact on the growth and mortality of large
than small trees (Phillips et al., 2010; Bennett et al., 2015). These
differences can be quite pronounced; in some cases, drought has
actually increased the growth rate of small trees while decreasing
the growth rate of large trees (Bennett et al., 2015). The greater
drought sensitivity of large tropical trees is likely driven by a com-
bination of the greater hydraulic challenge of lifting water to
greater height against the effects of gravity and path length-
associated resistance (Ryan et al., 2006; Zhang et al., 2009;
McDowell et al., 2011; McDowell & Allen, 2015) and greater
abiotic stress associated with an exposed canopy position (Roberts
et al., 1990; Nepstad et al., 2007; Bennett et al., 2015). It remains
unclear, however, how the more challenging hydraulic situation
of larger trees affects forest composition and carbon cycling under
nondrought conditions.
Trees exhibit a variety of adaptations to minimize moisture
loss during periods of dry season or droughts. Deciduousness is a
drought avoidance strategy, whereby leaf loss during times of cli-
matic stress reduces transpiration and the associated risk of
embolism (Wolfe et al., 2016). Trees that are transpiring during
periods of hydraulic stress require adaptations to avoid hydraulic
failure, such as deep roots or high wood density, which is often
positively correlated with resistance to embolism formation
(Hacke et al., 2001). However, higher wood density comes at the
expense of growth rate and height gain, which is an important
factor in competitive light-limited environments such as tropical
forests (Swenson & Enquist, 2007). Although drought avoidance
and drought tolerance represent two largely independent strate-
gies, there is significant diversity in species’ trait combinations
(Markesteijn & Poorter, 2009). As a result, the most useful single
metric of water deficit tolerance for species within a given size
class may be geographical distribution across moisture gradients
(e.g. Condit et al., 2013). Given the more challenging hydraulic
environment faced by large trees, we expect that at sites subject to
predictable periods of water limitation (i.e. dry seasons), or where
mild to moderate drought occurs frequently, canopy species
should exhibit stronger drought adaptations compared to under-
story trees. We further expect that, when drought adaptations are
insufficient to compensate for the harsher microclimate faced by
larger trees, the relative contributions of larger trees to ecosystem-
level productivity and biomass should be reduced under more
arid conditions.
The greater sensitivity of large trees to drought stress should
have important implications for ecosystem-level C cycling and
forest feedbacks to climate change, yet this remains poorly under-
stood. Only a couple of studies have quantified the role of tree
size in ecosystem-level C cycle drought responses of tropical
forests (Bennett et al., 2015), both showing large reductions of
live-tree biomass as a result of the more pronounced drought
response of larger trees (Nepstad et al., 2007; da Costa et al.,
2010). Most ecosystem models do not yet incorporate size-related
variation in hydraulic traits, but those that do are better able to
reproduce observed forest ecosystem responses to drought (Row-
land et al., 2015; Christoffersen et al., 2016). Improved
quantification of C cycle contributions, drought adaptations and
drought responses as a function of tree size, will be key to
improving our understanding of tropical forest ecosystem
responses to spatial and temporal variation in moisture stress.
A series of well-studied moist tropical forest plots spanning a
gradient of dry season moisture availability across the isthmus of
Panama and subject to droughts during El Ni~no events provide
the opportunity to better understand the C cycle contributions,
drought adaptations and drought responses of trees of different
sizes (Table 1; e.g. Leigh et al., 1990; Condit et al., 1995, 2000,
2013; Condit, 1998a). In 1981, a 50-ha long-term monitoring
plot, the first of the Center for Tropical Forest Science-Forest
Global Earth Observatory (Anderson-Teixeira et al., 2015), was
established on Barro Colorado Island (Condit, 1998a; Hubbell
et al., 1999). Shortly thereafter, the major El Ni~no event of 1982–
83 caused severe dry season drought, resulting in high mortality,
particularly among large trees, followed by a rapid rebound in
terms of leaf area and forest structure (Leigh et al., 1990; Condit
et al., 1995, 1999). Additional sites were established at the drier
(Cocoli) and wetter (San Lorenzo) ends of the moisture gradient
in 1994 and 1996, respectively (Condit et al., 2000, 2004, 2013).
These plots were resurveyed before and after another major El
Ni~no event in 1997–98, which significantly increased mortality
only at the driest site, where again larger trees suffered more
(Condit et al., 2004). It remains to be quantified how whole-
ecosystem C cycling is shaped by trees of different sizes and
drought adaptations across this gradient and how these patterns
relate to spatial and temporal variation in water availability.
Here, we analyze data from these three sites (Table 1) to test
three hypotheses regarding how Cag,live, ANPPstem and M are
shaped by trees of different sizes and drought adaptations and
how these patterns relate to spatial and temporal variation in
moisture stress: (1) contributions to Cag,live, ANPPstem and M
increase with tree size under nondrought conditions, with large
trees contributing proportionately less at the drier end of the
moisture gradient; (2) community composition is such that
drought adaptations (deciduousness, high wood density and
water deficit tolerance, a metric based on species distribution in
response to moisture amounts) are accentuated in the larger size
classes, particularly at the drier end of the gradient, and therefore
drought adapted species contribute disproportionately to Cag,live,
ANPPstem and M; and (3) focusing on the two instances where
the effects of El Ni~no drought stress were evident through
increases in tree mortality (1982–83 El Ni~no at Barro Colorado
Island and 1997–98 El Ni~no at Cocoli; Condit et al., 1995,
1999), we expect negative impacts to the ecosystem C balance
(e.g. elevated mortality and decreased ANPPstem) to increase with
tree size.
Materials and Methods
Study sites and data
Tree censuses were conducted at three sites spanning a gradient
of dry season moisture availability (‘moisture’ for brevity) in
Panama: Cocoli, Barro Colorado Island and San Lorenzo, also
New Phytologist (2017) 
 2017 Smithsonian Institute
New Phytologist
 2017 New Phytologist Trustwww.newphytologist.com
Research
New
Phytologist2
known as Sherman (Table 1; Condit, 1998a,b; Hubbell et al.,
1999, 2010; Condit et al., 2000). All sites are tropical moist low-
land forests with differing proportions of evergreen and decidu-
ous species (Table 1; Condit et al., 2000). Cocoli is on the drier
Pacific side of the gradient, and is a secondary forest c. 100 yr old
(Condit, 1998b). The 50-ha Barro Colorado Island plot is
located on a 1500-ha island in Gatun Lake and is primarily old-
growth forest that has been undisturbed by humans for over
500 yr (Condit, 1998b). The site with the highest moisture, San
Lorenzo, is a mature forest near the Atlantic coast that has been
subject to some logging or clearing activity during the last 150 yr
(Condit, 1998b). We excluded from the analyses a 1-ha patch of
young, secondary forest within San Lorenzo (Condit et al., 2004)
and a 1.9-ha patch at Barro Colorado Island (Harms et al.,
2001), because these patches of secondary forest differ from the
rest of the plot in species composition and are expected to differ
in growth and mortality patterns. Plots were censused following a
standardized protocol in which all stems ≥ 1 cm diameter at
breast height (DBH) were mapped, tagged, identified to species,
and measured in DBH (1.3 m) or, for all censuses except the first
Barro Colorado Island census, above any buttresses or other stem
irregularities (Manokaran et al., 1990; Condit, 1998a). Data cur-
rent as of February 29, 2016 were downloaded from the Center
for Tropical Forest Science database (http://ctfs.si.edu/ctfsrep).
Three metrics related to drought adaptation were available for
the majority of species at these three plots: deciduousness, wood
density and a moisture association index (MAI). Deciduous
species were defined based on surveys and expert knowledge as
those species capable of deciduousness at any site or in any size
class (Supporting Information Methods S1; Table S1; Condit
et al., 2000). Analyses were conducted both including and
excluding brevideciduous species, which are species that experi-
ence a very brief loss of leaves. Wood density values (g cm3)
were obtained from central Panama (for methods see Wright
et al., 2010) or from the Center for Tropical Forest Science wood
density dataset (http://ctfs.si.edu/Public/Datasets/CTFSWood
Density/). If there was no value at the species level, we used the
genus-level mean (used for 16.6% of total species) or the family-
level mean (7.1% of total species). At the site level, this corre-
sponded to genus-level values for 7.0%, 8.3% and 15.9% of
species and family-level values for 2.3%, 3.8% and 9.5% of
species at Cocoli, Barro Colorado Island and San Lorenzo,
respectively. These species tended to be rare (in total representing
just 2.2% of individuals). If an individual was unidentified or
had no available wood density value at any taxonomic level, the
mean of all other species at the site was used. Wood density val-
ues were lacking for only one rare tree (Besleria robusta, n = 4). A
species-level MAI was assigned to 80% of species based on a
study using environmental predictors to model tree distributions
in Panama (Condit et al., 2013). The model used eight climatic
and soil factors, one of which was dry season moisture deficit, in
a hierarchical Guassian logistic regression to predict species
occurrence, and we defined MAI as the species-specific moisture
response parameter returned by the model.
Precipitation and temperature data were obtained from local
weather stations for each site. Measurements for Barro Colorado
Island were taken at ‘El Claro’, a station established in a clearing
c. 2 km from the plot in 1972. Data for San Lorenzo were col-
lected from a crane built in 1997 adjacent to the plot, and data
gaps were filled using records from Gatun West, a Panama Canal
Authority (Autoridad del Canal de Panama – ACP) station
located 5.3 km southeast of San Lorenzo. Cocoli records were
Table 1 Basic information on the three sites spanning the Panama moisture gradient
Cocoli Barro Colorado Island San Lorenzo
Plot information
Latitude, longitude 8.9877,79.6166 9.1543,79.8461 9.2815,79.974
Size (ha) 4 50 (48.1*) 6 (4.96*)
Census years 1994, 1997, 1998 1981, 1985, 1990, 1995, 2000, 2005, 2010 1996, 1997, 1998, 2009
Focal non-El Ni~no census period 1994†–1997 1990–1995† 1996†–1997
Plot descriptions Condit et al. (2000, 2004) Condit (1998a), Hubbell et al. (1999) Condit et al. (2000, 2004)
Climate (1995–2010 mean)
Mean annual temperature (°C) 26.0 27.4 25.6
Mean annual precipitation (mm) 1808 2167 3197
Mean maximum dry season moisture deficit (mm)‡ 575 514 492
Vegetation
Stem density (stems ≥ 1 cm DBH; ha1) 2470 5155 3441
Species richness (stems ≥ 1 cm DBH; full plot) 173 323 268
% of canopy species deciduous§ 42.1 32.2 23.8
Unless otherwise noted, vegetation properties are as calculated in this study for censuses between 1994 and 1996.
*Plot size after relatively young 1–2-ha patches of secondary forest were excluded from these analyses.
†Indicates focal census for stem density and biomass calculations.
‡Values from Condit et al. (2013). Briefly, dry season moisture deficit is calculated as the sum of daily deficit values (Dd; mm per month), where Dd is the
difference between potential evapotranspiration (PET) and precipitation, and the maximum cumulative deficit is averaged across years. To avoid breaking
the dry season into separate calendar years, D was calculated for September to July of the following year. Precipitation and PET are interpolated based on
local weather stations.
§Canopy defined as diameter at breast height (DBH) ≥ 30 cm. These values are similar to those reported by Condit et al. (2000), but have been updated
according to the list of deciduous species used here (Supporting Information Table S1).

 2017 Smithsonian Institute
New Phytologist 
 2017 New Phytologist Trust
New Phytologist (2017)
www.newphytologist.com
New
Phytologist Research 3
taken from the Parque Natural Metropolitano canopy crane
established in 1995 at the northwestern edge of Panama City,
and missing values were filled by averaging data from two nearby
ACP stations – Albrook Airbase and Balboa Heights, located
4–5 km away. Rainfall measurements from ACP stations were
corrected to match plot stations for both Cocoli and San Lorenzo
using a linear regression developed by Steve Paton (http://bioge
odb.stri.si.edu/physical_monitoring/). At Barro Colorado Island
only, soil moisture was sampled from depths of 0–10 cm at 10
sites around the Lutz catchment, and solar radiation was mea-
sured by a pyranometer on top of the Lutz tower, close to the
meteorological station ‘El Claro.’
Analyses
Our analyses for stem density and total site biomass focused on
the census nearest to 1995 (before 1997–98 El Ni~no), whereas
those for growth and mortality focused on census periods that
were the closest to overlapping and with no major El Ni~no events
(Cocoli: 1994–1997, Barro Colorado Island: 1990–1995, San
Lorenzo: 1996–1997; Table 1).
All analyses were conducted at the stem (ramet) level; i.e. those
stems that arose from the same root system or collar were exam-
ined individually as opposed to jointly at the tree (genet) level.
To account for the fact that individual stems of multi-stemmed
individuals were not assigned unique identifiers during censuses,
we developed an algorithm to determine the probable alignment
of stem IDs from one census to the next (Methods S2). For stems
measured at a height other than the standard 1.3 m height of
measurement – including 0.98–3.78%, 0.51–1.69% and 1.26–
1.36% of stems at Cocoli, Barro Colorado Island and San
Lorenzo, respectively – we applied a taper correction to give an
equivalent DBH at 1.3 m (Cushman et al., 2014). All tree ferns
(Cyatheaceae) and strangler figs (Ficus bullenei, F. colubrinae,
F. costaricana, F. citrifolia, F. pertusa and F. popenoei) were
excluded from these analyses because their growth is not well
characterized by trunk diameter. Additional corrections to the
data and exclusions of outliers are detailed in Methods S2.
Carbon cycling variables Variables describing ecosystem-level
carbon (C) cycling were calculated as follows. Biomass was esti-
mated based on allometries developed by Chave et al. (2014). The
equation selected was designed for use when tree height measure-
ments are unavailable and instead accounts for diameter–height
allometries using an environmental stress parameter (E), which is
a function of climatic water deficit, temperature seasonality, and
precipitation seasonality. Values for E were extracted from http://
chave.ups-tlse.fr/pantropical_allometry.htm and were found to be
0.0748 for Cocoli, 0.0518 for Barro Colorado Island and
0.0565 for San Lorenzo. Biomass was converted to C using the
approximation that biomass is 47% C (IPCC, 2006). For com-
parison, we also estimated biomass using the same equation for all
sites; that is, instead of incorporating different E values, all three
sites used the allometry designed for ‘moist’ forests (Chave et al.,
2005). We found that this allometry resulted in higher carbon
estimates for all sites, but did not result in substantive differences
among analyses (Table S2; Fig. S1), and we therefore only
reported the results obtained from the Chave et al. (2014) allo-
metric equation. For palm trees (Arecaceae), biomass was esti-
mated using the family-level equation based on diameter
developed by Goodman et al. (2013), and was adjusted for log
transformation bias using a correction factor (Chave et al., 2005).
Aboveground net primary production of stem biomass C
(ANPPstem; Mg C ha
1 yr1) was calculated as the sum of annual
biomass C growth for stems that were alive at the beginning and
end of a census period, plus the biomass C in stems that recruited
into the census, all divided by the census interval (following vari-
able definition in Anderson-Teixeira et al., 2016). Woody mor-
tality (M; Mg C ha1 yr1) was calculated as the sum of the
aboveground biomass C of all stems that died divided by the cen-
sus interval, with biomass C estimates based on DBH measure-
ments from the most recent census before death. Net biomass C
change was calculated as ANPPstem – M. Initial Cag,live was defined
as the sum of live biomass C for stems at the initial census for
each census period. For the first census period on Barro Colorado
Island, a correction was applied to estimate ANPPstem, M, and
net biomass change because all measurements were made at
1.3 m, including around buttresses or other stem abnormalities,
for this one census (Table S3; Methods S2). These values were
determined for all plants ≥ 1 cm DBH and for the following
diameter classes: 1–10, 10–50, and ≥ 50 cm (Table S4).
Size-related variation In order to analyze how the variables of
interest varied with stem size, stems were grouped into 23
approximately log-even bins based on their initial diameter. Stem
density (n ha1), Cag,live, ANPPstem andM per cm DBH were cal-
culated for each size class, as was the mean initial diameter (D0;
i.e. diameter measured at the first census of each interval). For all
variables, we fitted the following function to loge-transformed
data:
y ¼ aD0b expcD0 Eqn 1
Here, a, b and c are fitted parameters, where c = 0 gives a power
function and negative or positive values of c give hump- or
U-shaped fits, respectively. Ninety-five percent confidence inter-
vals were calculated by randomly sampling 109 10 m subplots
for each variable using 1000 bootstrap replicates. We also quanti-
fied size-related variation in several of the underlying variables:
individual biomass C, diameter growth, individual biomass C
growth and stem mortality (Methods S3; Fig. S2).
The variation of interspecific functional trait distributions
among stem size classes also was analyzed. Mean deciduousness,
wood density and MAI were calculated for each of eight approxi-
mately log-even size classes and for the community as a whole. A
linear-log function was then fitted using linear least squares
regression to loge-transformed values of DBH. Weighted means
for deciduousness, wood density and MAI were calculated as
∑ Ti(Ci/Ctot), where Ti is the functional trait value attributed to
each individual stem based on its species identity and Ci and Ctot
are Cag,live, ANPPstem or M of the individual or entire commu-
nity, respectively.
New Phytologist (2017) 
 2017 Smithsonian Institute
New Phytologist
 2017 New Phytologist Trustwww.newphytologist.com
Research
New
Phytologist4
Results
C cycling and tree size across the moisture gradient
Contrary to expectations, we reject the first component of
Hypothesis 1, that Cag,live, ANPPstem and M increase with tree
size under nondrought conditions, and instead find that these
variables peaked below 50 cm DBH (Fig. 1b–d). When consid-
ering ecosystem-level attributes as a function of DBH on a lin-
ear scale (i.e. per cm increase in DBH; Fig. 1), stem density
declined sharply with DBH with an accelerating decline at all
sites; that is, the fit parameters b and c (Eqn 1) were consis-
tently negative (Fig. 1a; all P < 0.001; Table S5). Because indi-
vidual biomass C increases steeply with DBH (Fig. S2a;
Table S5), Cag,live increased with DBH across the lower end of
the size range (Fig. 1b; all b > 0.67; all P < 0.001), peaking
between mid and 50 cm DBH, and declining in the largest size
classes (all c < 0; all P < 0.001). Diameter growth rate and indi-
vidual biomass C growth rates both increased monotonically
with DBH (Fig. S2b,c; Table S5). ANPPstem displayed hump-
shaped relationships with DBH at Barro Colorado Island and
San Lorenzo (Fig. 1c; both b > 0.14, all c <0.02, both
P < 0.04), but there was no significant trend at Cocoli
(P = 0.44). Stem mortality rate decreased with DBH (Fig. S2d;
Table S5), whereas M increased across most of the size range at
all sites (Fig. 1d, all b > 0.27; all P ≤ 0.03), before declining at
DBH > 50 cm at Barro Colorado Island and San Lorenzo (both
c <0.01, both P ≤ 0.001).
Expressed in terms of three broad size classes (1–10, 10–50
and > 50 cm DBH), ecosystem-level C cycling (Cag,live,
ANPPstem, M) was dominated by large and mid-sized stems
(Fig. 1; Table S4). Specifically, across sites and censuses, stems
≥ 50 cm had the lowest stem densities (mean of 32 stems ha1;
range: 30–38), but still contributed the most to C cycling, repre-
senting on average 45% of ANPPstem (range: 29–64%), 49% of
M (range: 19–63%) and 59% of Cag,live (range: 49–68%;
Table S4). Stems 10–50 cm DBH had an average stem density of
400 stems ha1 (range: 244–493) and contributed almost as
much C as large stems, comprising on average 44% of ANPPstem
(range: 29–55%), 45% of M (range: 33–72%) and 37% of Cag,
live (range: 30–47; Table S4). By contrast, the smallest stems
(< 10 cm DBH) had a mean stem density of 4020 stems ha1
(range: 2282–5284), and contributed least to C cycling, repre-
senting only an average 11% of ANPPstem (range: 6–16%), 6%
ofM (range: 3–9%) and 4% of Cag,live (range: 2–5%; Table S4).
Across the gradient, there were no consistent trends in ecosys-
tem-level C cycling or its partitioning across size classes. Specifi-
cally, at the ecosystem level, Cag,live tended to increase with
decreasing dry season intensity, but with overlapping 95% CIs
(Tables 2, S4) – a trend that is potentially confounded by the fact
that the driest site (Cocoli) is a secondary forest. This trend was
less pronounced when allometries from Chave et al. (2005) were
used, with San Lorenzo and Barro Colorado Island displaying
similar values (Table S2). M also tended to increase with decreas-
ing dry-season intensity, whereas ANPPstem showed no consistent
(a) (b)
(c) (d)
Fig. 1 Size-related variation in (a) stem
density, (b) aboveground live biomass carbon
(C) (Cag,live), (c) aboveground woody
productivity (ANPPstem) and (d) woody
mortality (M) at all three sites during the
focal non-El Ni~no census periods (Table 1).
Totals for each size bin are divided by the
width of the size bin (cm) such that
relationships depict how these variables
change with diameter on a linear scale.
Dashed lines indicate a nonsignificant trend.
Fit parameters and statistics are given in
Supporting Information Table S5. Vertical
lines depict 95% confidence intervals based
on bootstrapping over subplots.

 2017 Smithsonian Institute
New Phytologist 
 2017 New Phytologist Trust
New Phytologist (2017)
www.newphytologist.com
New
Phytologist Research 5
trends across the gradient (Table 2). Net biomass C change was
significantly positive in Cocoli, the secondary forest, and was not
significantly different from zero at the other two sites. For all of
these variables, there was little evidence of systematic differences
in the relative contributions of trees of different sizes across the
moisture gradient (Figs 1, S2; Tables S4, S5); thus, we reject the
second component of Hypothesis 1, that large trees contribute
proportionately less to C cycling at the drier end of the moisture
gradient.
Tree size and drought adaptations
The dominance of deciduous species increased with DBH at all
sites, particularly at the drier sites (Fig. 2a,b; Table S6), in
concordance with Hypothesis 2. Specifically, the fractional abun-
dance of deciduous species increased significantly with DBH at
all sites, regardless of whether brevideciduous species were classi-
fied as evergreen or deciduous (all P ≤ 0.01). As expected, the
steepness of the slope of this relationship increased with climatic
water deficit; that is, whereas the three sites had similarly low
fractions of deciduous species in the small size classes, the decidu-
ous fraction of larger trees increased from the wettest to the driest
site (Fig. 2a). This resulted in weighted mean deciduousness
being greater for Cag,live, ANPPstem and M than for stem density,
indicating that relative to their abundance, deciduous species
contributed disproportionately to biomass C and changes therein
(Fig. 2b). Although total fractions of deciduous species were simi-
lar across sites (largely due to the high abundance of evergreen
Table 2 Ecosystem-level carbon (C) variables including live biomass C (Cag,live), woody productivity (ANPPstem), woody mortality (M) and net biomass C
change for all three sites during non-El Ni~no census periods
Cag,live (95% CIs)
(Mg C ha1)
ANPPstem (95% CIs)
(Mg C ha1 yr1)
M (95% CIs)
(Mg C ha1 yr1)
Net biomass C change
(95% CIs) (Mg C ha1 yr1)
Cocoli 1994*–97 120 (105,132) 3.07 (2.63, 3.52) 1.18 (0.61, 1.99) 1.89 (0.90, 2.72)
Barro Colorado Island 1990–95* 136 (129,143) 2.76 (2.62, 2.91) 2.43 (2.04, 2.85) 0.32 (0.14, 0.70)
Non-El Ni~no mean 136 3.20 2.63 0.57
San Lorenzo 1996*–97 144 (130,157) 3.44 (2.98, 3.95) 2.83 (1.89, 3.95) 0.61 (0.59, 1.75)
Non-El Ni~no mean 146 2.78 2.94 0.16
Shown are records for our focal census periods (Table 1) and the mean for all non-El Ni~no census periods.
*Indicates year for which Cag,live is reported.
(a) (b)
(c) (d)
(e) (f)
Fig. 2 Deciduousness (a, b), wood density
(c, d) and moisture association index (MAI; e, f)
averaged by size class (a, c, e) and as means
weighted according to stems’ contributions
to total stem density, live biomass carbon (C)
(Cag,live), woody productivity (ANPPstem) and
woody mortality (M) (b, d, f). For
deciduousness, results are presented
counting brevideciduous species as
evergreen (solid colors) or deciduous (pale
colors), with 0 indicating nondeciduous and 1
indicating deciduous. For MAI, negative
values represent species associated with drier
climates and positive values correspond to
species associated with wetter climates
(Condit et al., 2013). Dashed lines indicate
nonsignificant trends. Results apply to the
focal censuses identified in Table 1.
New Phytologist (2017) 
 2017 Smithsonian Institute
New Phytologist
 2017 New Phytologist Trustwww.newphytologist.com
Research
New
Phytologist6
stems in the small size classes; Table 1; Fig. 2a,b), the contribu-
tions of deciduous species to Cag,live and ANPPstem increased with
climatic water deficit, indicating that the disproportionate influ-
ence of deciduous species on C cycling was greatest at the driest
site (Fig. 2b).
As with deciduousness, differences in wood density were
accentuated in the larger size classes and at drier sites (Fig. 2c);
however, contrary to Hypothesis 2, the larger trees at drier sites
had lower wood density. Specifically, mean wood density
decreased with DBH at the two drier sites (both P < 0.004),
but not at San Lorenzo (P = 0.71; Fig. 2c; Table S6). Species
with higher wood density were more abundant at Cocoli than
the wetter sites across most of the size spectrum, but this pat-
tern reversed in the largest size classes (Fig. 2c). At Cocoli and
Barro Colorado Island, species with low wood density con-
tributed disproportionately to C cycling, as illustrated by lower
weighted mean wood density for Cag,live, ANPPstem and M than
for stem density (Fig. 2d). Furthermore, although community-
wide mean wood density decreased with increasing moisture
across the gradient, weighted mean wood density increased
with moisture for Cag,live, ANPPstem and M; that is, the relative
importance of low wood density species in the larger size classes
increased with climatic water deficit (Fig. 2d). The decrease in
wood density with stem size was driven primarily by deciduous
canopy species with low wood density, particularly at Cocoli,
where exclusion of deciduous species made this trend disappear
(P = 0.99).
As expected, and consistent with how MAI is defined, the
mean MAI value varied across the moisture gradient, with the
greater abundance of xerophytic species at Cocoli and of meso-
phytic species at San Lorenzo (Fig. 2e). Interestingly, differences
were more pronounced in the larger size classes; mean MAI
decreased with DBH at Cocoli (P = 0.009), did not vary signifi-
cantly with DBH at Barro Colorado Island (P = 0.45) and
increased with DBH at San Lorenzo (P = 0.03; Fig. 2e;
Table S6). As with deciduousness and wood density, the MAIs
associated with larger trees tended to be disproportionately influ-
ential to C cycling, such that weighted mean MAI for Cag,live, M,
ANPPstem varied more markedly across the moisture gradient
than did mean MAI based on abundance.
Responses to El Ni~no events
The El Ni~no events disproportionately affected the largest trees at
both Barro Colorado Island and Cocoli, as predicted in Hypothe-
sis 3. At Barro Colorado Island, the 1982–83 El Ni~no was char-
acterized by anomalously warm temperatures lasting from May
1982 to June 1983 (peak mean monthly temperature of 29.30°C
in April 1983 was the highest on record from 1980 to 2010), low
November–April precipitation (289 mm in 1982–83 compared
to a 1980–2010 mean of 1041 mm) and the lowest soil moisture
on record from 1980 to 2010 (27.1% water by wet weight com-
pared to a 1980–2010 mean of 38.9%). The drought stress
resulted in high tree mortality (Condit et al., 1995), such that M
(4.86Mg C ha1 yr1) was almost double the mean value for
census periods lacking a major El Ni~no event
(2.63Mg C ha1 yr1; Table S4). Increases in M above the non-
El Ni~no mean were particularly pronounced for larger stems,
with stems ≥ 50 cm DBH responsible for 63% of total M (3.06
Mg C ha1 yr1; Fig. 3e; Tables 3, S4). ANPPstem was also ele-
vated during the 1981–1985 census period, effectively compen-
sating for the high M (net biomass C change = 0.70
Mg C ha1 yr1; Fig. 3a,c; Tables 3, S4). The largest stems
(≥ 50 cm) were the only size class to have a negative net biomass
C change value (0.24Mg C ha1 yr1; Tables 3, S4). It should
be noted that although we sought to correct for the changes in
measurement height protocol between the 1981 and 1985 Barro
Colorado Island censuses (Methods S2), values for this census
period are less accurate than the others reported here.
At Cocoli, the 1997–98 El Ni~no resulted in elevated Novem-
ber–April temperature (26.9°C, compared to mean of 26.2°C for
1995–2010 non-El Ni~no years) and decreased November–April
rainfall (347 mm; 1995–2010 non-El Ni~no mean: 550 mm).
ANPPstem decreased relative to the preceding non-El Ni~no period
whereas M increased, resulting in a net decline in Cag,live (Fig. 3;
Table 3). Specifically, ANPPstem declined 26% (from 3.07 to
2.28Mg C ha1 yr1; Table S4). This was driven by the larger
size classes (Fig. 3d): although ANPPstem of stems ≥ 50 cm DBH
declined, ANPPstem of stems < 10 cm increased (Table 3). Mean-
while, M increased 170% (from 1.18 to 3.18Mg C ha1 yr1),
with the largest contribution (63%) coming from stems ≥ 50 cm
DBH (Table 2; Fig. 3f). Total Cag,live declined overall, increasing
for smaller stems while decreasing for larger stems (Fig. 3b;
Tables 3, S4).
Discussion
Across three large forest plots in Panama, ecosystem-level carbon
(C) cycling was dominated by mid- to large-sized trees, with con-
tributions per unit diameter at breast height (DBH) typically
peaking in the 10–50 cm DBH range and trees ≥ 50 cm DBH
representing an average of 59% of live aboveground biomass
(Cag,live) and contributing somewhat less to changes therein (45%
woody productivity, ANPPstem, and 49% woody mortality, M;
Fig. 1, Table S4), despite their low stem density (0.8%, on aver-
age; Table S4). We found little difference in the relative C cycle
contributions of large vs small trees across the moisture gradient
(Fig. 1), indicating that the observed differences in community
composition were sufficient to compensate for any differential
biophysical challenges faced by larger trees in drier climates.
Indeed, larger stems showed evidence of stronger drought adapta-
tions, having higher fractions of deciduous species and more pro-
nounced sorting across a geographical moisture gradient (Fig. 2;
Condit et al., 2000, 2013). Despite these adaptations, in associa-
tion with El Ni~no drought stress at Barro Colorado Island in
1982–83 and at Cocoli in 1997–98, larger trees suffered greater
increases in mortality, dominating ecosystem-level C cycle
responses (Fig. 3e,f; Table 3). Thus, it is generally the mid-sized
to large trees – those with at least some chance of being in an
exposed canopy position (Muller-Landau et al., 2006a) – that dis-
play the most pronounced drought adaptations, suffer most
under drought (see also Condit et al., 1995, 2004; Bennett et al.,

 2017 Smithsonian Institute
New Phytologist 
 2017 New Phytologist Trust
New Phytologist (2017)
www.newphytologist.com
New
Phytologist Research 7
2015), and most strongly mediate forest C cycle responses to
hydraulic stress.
One novel finding of this study is that – contrary to Hypothe-
sis 1 – aboveground biomass and C cycling are not dominated by
the largest trees, but tend rather to peak at intermediate stem
diameters (Figs 1b-d, S3). Specifically, when C cycle contribu-
tions were expressed as a linear function of DBH – consistent
with previous literature on size scaling in forests (e.g.
Muller-Landau et al., 2006b; West et al., 2009; Lutz et al., 2012)
– Cag,live peaked at 27–50 cm DBH, whereas maximum contribu-
tions to ANPPstem and M occurred at < 50 cm DBH. It is impor-
tant to note that interpretations of C cycle contributions as a
function of DBH are influenced by the way that size bins are
defined. When DBH was expressed on a logarithmic scale – i.e.
size bin width increasing with DBH – C cycle contributions
increased continuously with DBH across most of the size spec-
trum (Fig. S3). Nevertheless, under either approach, it was not
the largest trees that contributed most to biomass and C cycling;
rather, their rarity made their contributions less than those of
intermediate-sized stems. By contrast, although stems < 10 cm
DBH contributed relatively little to live aboveground biomass at
these sites (≤ 5.2%), their contributions to ANPPstem andM were
more significant (ranging up to 15.9% and 8.7%, respectively). It
is particularly striking that at Barro Colorado Island and Cocoli,
ANPPstem per cm DBH was similar across all size classes below
c. 50 cm DBH and that contributions to ANPPstem declined
above this threshold (Fig. 1c). The relatively high contributions
of small stems to ANPPstem and M reflect high biomass turnover
rates, driven by relatively high stem mortality (Fig. S2d) and
mass-specific growth rates (i.e. individual net biomass C change/
Cag,live; Fig. S2a,b). Although the contributions of small stems
should not be ignored, it is the mid- to large-sized trees that dom-
inate aboveground C cycling (Fig. 1; Table S4) and should be
most important in driving tropical forest C cycle responses to
climatic variation in space and time.
Fig. 3 Comparison of (a, b) net biomass
carbon (C) change, (c, d) woody productivity
(ANPPstem), and (e, f) woody mortality (M)
by size class in El Ni~no and non-El Ni~no years
at Barro Colorado Island (1982–83 El Ni~no)
and Cocoli (1997–98 El Ni~no). Non-El Ni~no
means include all census periods that did not
include a major El Ni~no event (see Table 1).
Stems were divided into eight log-even bins
and the total value for each size bin was
divided by the width of the size bin (cm) to
depict how each variable changes with
diameter on a linear scale. Vertical lines
depict 95% confidence intervals based on
bootstrapping over subplots.
Table 3 El Ni~no-driven changes in woody productivity (ANPPstem), woody
mortality (M) and net biomass carbon (C) change by size class, with
change expressed relative to non-El Ni~no census period means (Table 1
and Supporting Information Table S4).
Event Size class (cm) ANPPstem M
Net biomass
C change
Barro Colorado Island 1–10 +0.11 +0.03 +0.09
1982–83 El Ni~no 10–50 +0.84 +0.54 +0.30
≥ 50 +1.40 +1.65 0.26
All (≥ 1) +2.35 +2.22 +0.13
Cocoli 1–10 +0.06 +0.03 +0.02
1997–98 El Ni~no 10–50 0.09 +0.44 0.53
≥ 50 0.76 +1.53 2.28
All (≥ 1) 0.79 +2.00 2.79
All variables have units of Mg C ha1 yr1.
New Phytologist (2017) 
 2017 Smithsonian Institute
New Phytologist
 2017 New Phytologist Trustwww.newphytologist.com
Research
New
Phytologist8
Contrary to our first hypothesis, we found little evidence of
systematic, directional differences in the C cycle contributions of
trees of different sizes across the moisture gradient (Fig. 1; Tables
S4, S5). Based on the observed greater drought sensitivity of
larger trees (Bennett et al., 2015), we may have expected relatively
smaller C cycle contributions of larger trees under drier condi-
tions (Hypothesis 1). However, across this modest moisture gra-
dient, the drought adaptations associated with larger trees in
drier climates were sufficient to compensate for any stronger
hydraulic stress experienced by these trees because of their canopy
position. Of course, extending into far drier climates, large trees
– and their carbon cycle contributions – completely disappear.
Species’ drought adaptations varied with stem size (Fig. 2).
The wettest site and understories at all sites were dominated by
evergreen species with relatively high wood density, whereas the
dominance of deciduous species increased, and mean wood den-
sity decreased, with increasing tree size and dry season intensity
(Fig. 2a). These patterns are largely consistent with our second
hypothesis and with the principle that larger, taller trees face
more challenging hydraulic constraints than do their understory
counterparts. Canopy trees are exposed to higher solar radiation
and leaf-to-air vapor pressure deficit, which may make it difficult
to simultaneously maintain hydraulic safety, regulate leaf tem-
perature, and maintain a positive C balance during dry condi-
tions (Roberts et al., 1990; Nepstad et al., 2007; Bennett et al.,
2015). Dry season deciduousness is one drought adaptation
strategy that allows trees to avoid these stressors (Markesteijn &
Poorter, 2009), and our results (Fig. 2) – along with previous
findings (Frankie et al., 1974; Wright, 1991; Condit et al.,
2000) – make it apparent that this strategy is increasingly
favored under drier conditions and for larger trees, supporting
Hypothesis 2. In fact, the magnitude of the observed increase in
deciduousness with stem size (Fig. 2a) is likely underestimated in
this study because individual species are often deciduous as big
trees but not as juveniles (Condit et al., 2000). Size trends in
deciduousness (Fig. 2a) and C cycle contributions (Fig. 1) com-
bined such that deciduous species contributed disproportionately
to forest C cycling relative to their abundance in the community
(Fig. 2b), particularly at the two drier sites. Thus, capturing the
observed size trend in deciduousness will be essential to accu-
rately modeling C cycling and its seasonality in semi-deciduous
tropical forests.
Counterintuitive to the principle that larger trees require
stronger drought adaptations (Hypothesis 2) is the fact that wood
density declined with stem size at the two drier sites (Fig. 2c). All
else being equal, we would expect that trees facing higher water
deficits would have higher wood density; however, Panamanian
tree species display a wide variety of hydraulic strategies. Low
wood density species can be strongly drought-adapted if decidu-
ous, and there was no association between wood density and
moisture association index (MAI) among the species included in
this analysis (R2= 0.002; P = 0.40). Thus, we interpret the lower
average wood density of large individuals (Fig. 2c) as being driven
primarily by the facts that low wood density trees can achieve the
same strength at lower costs by investing in thicker trunks (Lar-
javaara & Muller-Landau, 2010) and that, all else being equal,
lower wood density species have faster diameter growth and
therefore can reach large diameter faster than high wood-density
species. The latter may explain the pronounced dominance of
low wood density, mostly deciduous species in the largest size
classes at Cocoli, which is a secondary forest (Fig. 2); however,
this trend was not observed at San Lorenzo despite its history of
selective logging, which has been shown to favor low wood den-
sity stands (Carre~no-Rocabado et al., 2012). At the two drier
sites, declines in wood density with stem size (Fig. 2c) combined
with size trends in C cycle contributions (Fig. 1) such that the
weighted average wood density of stems contributing to C cycling
was consistently lower than community-wide wood density
means (Fig. 2d). For forests such as these, models or analyses
assuming that community mean wood densities apply across size
classes may overestimate biomass, ANPPstem andM.
A suite of hydraulic traits, including but by no means limited
to deciduousness and wood density, shape species’ overall water
deficit tolerance and distribution across geographic gradients, as
reflected in our MAI (Condit et al., 2013). This metric is not
suitable for direct comparison of drought tolerance across size
classes because understory and canopy species are subject to dif-
ferent microclimates. However, consistent with Hypothesis 2,
larger trees display stronger geographical sorting across the mois-
ture gradient (Fig. 2e). This suggests that water stress plays a
stronger role in shaping their geographical distributions than
those of understory species, which experience a more buffered
microclimate. Moreover, these results indicate that species associ-
ating more strongly with one end of the geographical moisture
gradient contribute more to C cycling (Fig. 2f), primarily because
of their larger size.
Although the relative C cycle contributions of trees of different
sizes did not vary across the moisture gradient, their responses to
the El Ni~no drought events differed. For the El Ni~no droughts at
both Barro Colorado Island (1982–83) and Cocoli (1997–98),
the larger trees were more strongly impacted in terms of mortality
(Fig. 3; Table S4; Condit et al., 1995, 1999, 2004; Bennett et al.,
2015). Consistent with our third hypothesis, the implication for
the C balance was, in both cases, a large increase (c.
2 Mg C ha1 yr1) in woody mortality, driven by the dispropor-
tionate importance of larger trees (Fig. 3; Table 3). Growth
responses differed between these two events. At Cocoli, consistent
with Hypothesis 3, the 1997–98 El Ni~no reduced growth in the
larger size classes and increased growth in the smaller size classes,
resulting in net declines in ANPPstem and Cag,live (Fig. 3;
Table 3). By contrast, high woody mortality associated with the
1982–83 El Ni~no at Barro Colorado Island appears to have been
compensated for by elevated ANPPstem during the same census
period (Figs 3; Table 3); this was perhaps driven by competitive
release or by positive El Ni~no growth responses of some species,
likely due to the alleviation of light limitation by reduced cloud
cover (Graham et al., 2003). In all cases, despite the fact that they
did not dominate aboveground C cycling during non-El Ni~no
conditions, it was the response of the larger trees that drove
ecosystem-level responses to the El Ni~no events (Table 3).
Here, we elucidated how spatial and temporal variation in
water deficit interact with tree size to shape C cycling in

 2017 Smithsonian Institute
New Phytologist 
 2017 New Phytologist Trust
New Phytologist (2017)
www.newphytologist.com
New
Phytologist Research 9
Panamanian tropical forests, findings that can yield insight into
the likely climate change responses of these and other tropical
forests. Panamanian forests are adapted to regular dry seasons
and moderate droughts, and – in cases where we have data – have
shown high resilience to the major El Ni~no events of 1982–83
and 1997–98 in terms of forest structure and C cycling (Fig. 3c;
Leigh et al., 1990; Condit et al., 2004). Thus, moderate climate-
change associated droughts are unlikely to dramatically alter
forest structure and function. Patterns across the moisture gradi-
ent suggest that a gradual drying trend – as may be expected if
temperature increases are not accompanied by significant
increases in precipitation – would likely result in shifts in com-
munity composition, with increasing prevalence of drought-
adapted (e.g. deciduous) species, particularly in the larger size
classes (Fig. 2). However, to the extent that species compositional
changes keep pace with climate change, major changes in C
cycling – or size trends therein – may be unlikely across the range
of climatic water deficit examined here (Fig. 1). By contrast, a
rapid increase in the frequency or intensity of severe El Ni~no
droughts could have substantial impacts on forest size structure
and C cycling. Because severe El Ni~no events disproportionately
impact the larger – and commonly older – trees, they stand to
have both substantive impacts on the C cycle and relatively long-
lasting impacts on forest structure. Thus, if climate change
increases severe droughts beyond what these forests have experi-
enced historically, there is potential for eventual deterioration of
forest resilience. Better understanding of the factors that confer
resilience and vulnerability of mid- to large-sized trees to drought
will therefore be particularly important for predicting tropical
forest responses to climate change.
Acknowledgements
We gratefully acknowledge the many people who contributed to
the forest census data used here. Thanks to Rolando Perez and
Steve Paton for information on deciduousness and climate,
respectively. The analyses presented here were funded by a Smith-
sonian Competitive Grants Program for Science to K.J.A-T.
The forest dynamics research was made possible by National
Science Foundation grants to S.P.H.: DEB-0640386, DEB-
0425651, DEB-0346488, DEB-0129874, DEB-00753102,
DEB-9909347, DEB-9615226, DEB-9615226, DEB-9405933,
DEB-9221033, DEB-9100058, DEB-8906869, DEB-8605042,
DEB-8206992, DEB-7922197, support from the Center for
Tropical Forest Science, the Smithsonian Tropical Research
Institute, the John D. and Catherine T. MacArthur Foundation,
the Mellon Foundation, the Small World Institute Fund, and
numerous private individuals. Some climate data were obtained
from the Meteorology and Hydrology Branch, Panama Canal
Authority, Republic of Panama.
Author contributions
V.M. and K.J.A-T. designed the research with input from coau-
thors; R.C., S.P.H. and S.J.W. collected forest census and trait
data; V.M. carried out data analysis and interpretation with
assistance from K.J.A-T., A.J.T., E.B.G-A., V.H., H.C.M-L.,
S.J.W. and R.C.; and V.M. and K.J.A-T. wrote the manuscript,
which was reviewed by all coauthors.
References
Anderegg WRL, Schwalm C, Biondi F, Camarero JJ, Koch G, Litvak M, Ogle
K, Shaw JD, Shevliakova E, Williams AP et al. 2015. Pervasive drought
legacies in forest ecosystems and their implications for carbon cycle models.
Science 349: 528–532.
Anderson-Teixeira KJ, Davies SJ, Bennett AC, Gonzalez-Akre EB, Muller-
Landau HC, Joseph Wright S, Abu Salim K, Almeyda Zambrano AM, Alonso
A, Baltzer JL et al. 2015. CTFS-ForestGEO: a worldwide network monitoring
forests in an era of global change. Global Change Biology 21: 528–549.
Anderson-Teixeira KJ, Wang MMH, McGarvey JC, LeBauer DS. 2016. Carbon
dynamics of mature and regrowth tropical forests derived from a pantropical
database (TropForC-db). Global Change Biology 22: 1690–1709.
Bastin J-F, Barbier N, Rejou-Mechain M, Fayolle A, Gourlet-Fleury S, Maniatis
D, de Haulleville T, Baya F, Beeckman H, Beina D et al. 2015. Seeing
Central African forests through their largest trees. Scientific Reports 5: 13 156.
Beer C, Reichstein M, Tomelleri E, Ciais P, Jung M, Carvalhais N, R€odenbeck
C, Arain MA, Baldocchi D, Bonan GB et al. 2010. Terrestrial gross carbon
dioxide uptake: global distribution and covariation with climate. Science 329:
834–838.
Bennett AC, McDowell NG, Allen CD, Anderson-Teixeira KJ. 2015. Larger
trees suffer most during drought in forests worldwide. Nature Plants 1: 15139.
Carre~no-Rocabado G, Pe~na-Claros M, Bongers F, Alarcon A, Licona J-C,
Poorter L. 2012. Effects of disturbance intensity on species and functional
diversity in a tropical forest. Journal of Ecology 100: 1453–1463.
Chave J, Andalo C, Brown S, Cairns MA, Chambers JQ, Eamus D, F€olster H,
Fromard F, Higuchi N, Kira T et al. 2005. Tree allometry and improved
estimation of carbon stocks and balance in tropical forests. Oecologia 145: 87–
99.
Chave J, Rejou-Mechain M, Burquez A, Chidumayo E, Colgan MS, Delitti
WBC, Duque A, Eid T, Fearnside PM, Goodman RC et al. 2014. Improved
allometric models to estimate the aboveground biomass of tropical trees. Global
Change Biology 20: 3177–3190.
Christoffersen BO, Gloor M, Fauset S, Fyllas NM, Galbraith DR, Baker TR,
Kruijt B, Rowland L, Fisher RA, Binks OJ et al. 2016. Linking hydraulic traits
to tropical forest function in a size-structured and trait-driven model (TFS v. 1-
Hydro). Geoscientific Model Development 9: 4227–4255.
Condit RS. 1998a. Tropical forest census plots – methods and results from Barro
Colorado Island, Panama and a comparison with other plots. Berlin, Germany
and Georgetown, TX, USA: Springer and R.G. Landes.
Condit R. 1998b. Ecological implications of changes in drought patterns: shifts
in forest composition in Panama. Climatic Change 39: 413–427.
Condit R, Aguilar S, Hernandez A, Perez R, Lao S, Angehr G, Hubbell SP,
Foster RB. 2004. Tropical forest dynamics across a rainfall gradient and the
impact of an El Nino dry season. Journal of Tropical Ecology 20: 51–72.
Condit R, Ashton PS, Manokaran N, LaFrankie JV, Hubbell SP, Foster RB.
1999. Dynamics of the forest communities at Pasoh and Barro Colorado:
comparing two 50-ha plots. Philosophical Transactions of the Royal Society B
354: 1739–1748.
Condit R, Engelbrecht BMJ, Pino D, Perez R, Turner BL. 2013. Species
distributions in response to individual soil nutrients and seasonal drought
across a community of tropical trees. Proceedings of the National Academy of
Sciences, USA 110: 5064–5068.
Condit R, Hubbell SP, Foster RB. 1995.Mortality rates of 205 neotropical tree
and shrub species and the impact of a severe drought. Ecological Monographs
65: 419–439.
Condit R, Watts K, Bohlman SA, Perez R, Foster RB, Hubbell SP. 2000.
Quantifying the deciduousness of tropical forest canopies under varying
climates. Journal of Vegetation Science 11: 649–658.
Coomes D, Duncan R, Allen R, Truscott J. 2003. Disturbances prevent stem
size-density distributions in natural forests from following scaling relationships.
Ecology Letters 6: 980–989.
New Phytologist (2017) 
 2017 Smithsonian Institute
New Phytologist
 2017 New Phytologist Trustwww.newphytologist.com
Research
New
Phytologist10
Corlett RT. 2016. The impacts of droughts in tropical forests. Trends in Plant
Science 21: 584–593.
da Costa ACL, Galbraith D, Almeida S, Portela BTT, da Costa M, de Athaydes
Silva J Jr, Braga AP, de Gonc
alves PHL, de Oliveira AA, Fisher R et al. 2010.
Effect of 7 yr of experimental drought on vegetation dynamics and biomass
storage of an eastern Amazonian rainforest. New Phytologist 187: 579–591.
Cushman KC, Muller-Landau HC, Condit RS, Hubbell SP. 2014. Improving
estimates of biomass change in buttressed trees using tree taper models.Methods
in Ecology and Evolution 5: 573–582.
Enquist BJ, West GB, Brown JH. 2009. Extensions and evaluations of a general
quantitative theory of forest structure and dynamics. Proceedings of the National
Academy of Sciences, USA 106: 7046–7051.
Fauset S, Johnson MO, Gloor M, Baker TR, Monteagudo MA, Brienen RJW,
Feldpausch TR, Lopez-Gonzalez G, Malhi Y, ter Steege H et al. 2015.
Hyperdominance in Amazonian forest carbon cycling. Nature Communications
6: 6857.
Frankie GW, Baker HG, Opler PA. 1974. Comparative phenological studies of
trees in tropical wet and dry forests in the lowlands of Costa Rica. Journal of
Ecology 62: 881–919.
Goodman RC, Phillips OL, del Castillo Torres D, Freitas L, Cortese ST,
Monteagudo A, Baker TR. 2013. Amazon palm biomass and allometry. Forest
Ecology and Management 310: 994–1004.
Graham EA, Mulkey SS, Kitajima K, Phillips NG, Wright SJ. 2003. Cloud cover
limits net CO2 uptake and growth of a rainforest tree during tropical rainy
seasons. Proceedings of the National Academy of Sciences, USA 100: 572–576.
Hacke UG, Sperry JS, Pockman WT, Davis SD, McCulloh KA. 2001. Trends in
wood density and structure are linked to prevention of xylem implosion by
negative pressure. Oecologia 126: 457–461.
Harms KE, Condit R, Hubbell SP, Foster RB. 2001.Habitat associations of
trees and shrubs in a 50-ha neotropical forest plot. Journal of Ecology 89: 947–
959.
Hubbell SP, Condit RS, Foster RB. 2010. Barro Colorado forest census plot
data. [WWW document] URL https://ctfs.arnarb.harvard.edu/webatlas/datase
ts/bci [accessed 29 February 2016].
Hubbell SP, Foster RB, O’Brien ST, Harms KE, Condit R, Wechsler B, Wright
SJ, de Lao SL. 1999. Light-gap disturbances, recruitment limitation, and tree
diversity in a neotropical forest. Science 283: 554–557.
Huntingford C, Zelazowski P, Galbraith D, Mercado LM, Sitch S, Fisher R,
Lomas M, Walker AP, Jones CD, Booth BBB et al. 2013. Simulated resilience
of tropical rainforests to CO2-induced climate change. Nature Geoscience 6:
268–273.
IPCC. 2006. Chapter 4. Forest land. In: Eggleston S, Buendia L, Miwa K, Ngara
T, Tanabe K, eds. 2006 IPCC guidelines for national greenhouse gas inventories,
vol. 4, agriculture, forestry, and other land uses. Hayama, Japan: Institute for
Global Environmental Strategies.
IPCC. 2013. Climate Change 2013: The Physical Science Basis. In: Stocker TF,
Quin D, Plattner G-K, Tignor M, Allen SK, Boschung J, Nauels A, Xia Y, Bex
V, Midgley PM, eds. Contribution of Working Group I to the Fifth Assessment
Report of the Intergovernmental Panel on Climate Change. Cambridge, UK and
New York, NY, USA: Cambridge University Press.
Larjavaara M, Muller-Landau HC. 2010. Rethinking the value of high wood
density. Functional Ecology 24: 701–705.
Lawrence D, Vandecar K. 2015. Effects of tropical deforestation on climate and
agriculture. Nature Climate Change 5: 27–36.
Leigh EG, Windsor DM, Rand AS, Foster RB. 1990. The impact of the ‘El
Ni~no’ drought of 1982–83 on a Panamanian semideciduous forest. In: Glynn
PW, ed. Global ecological consequences of the 1982–83 El Nino-Southern
Oscillation. Amsterdam, the Netherlands: Elsevier, 473–486.
Lutz JA, Larson AJ, Swanson ME, Freund JA. 2012. Ecological importance of
large-diameter trees in a temperate mixed-conifer forest. PLoS ONE 7: e36131.
Manokaran N, LaFrankie JV, Kochummen K, Quah ES, Klahn J, Ashton PS,
Hubbell SP, Chan HT. 1990.Methodology for the fifty hectare research plot
at Pasoh Forest Reserve. Research Pamphlet, Forest Research Institute of Malaysia
104: 1–69.
Markesteijn L, Poorter L. 2009. Seedling root morphology and biomass
allocation of 62 tropical tree species in relation to drought- and shade-
tolerance. Journal of Ecology 97: 311–325.
McDowell NG, Allen CD. 2015. Darcy’s law predicts widespread forest
mortality under climate warming. Nature Climate Change 5: 669–672.
McDowell NG, Bond BJ, Hill L, Ryan MG, Whitehead D. 2011. Relationships
between tree height and carbon isotope discrimination. In: Meinzer FC,
Lachenbruch B, Dawson TE, eds. Size and age related changes in tree structure
and function. Dordrecht, the Netherlands: Springer, 255–286.
Muller-Landau HC, Condit RS, Chave J, Thomas SC, Bohlman SA,
Bunyavejchewin S, Davies S, Foster R, Gunatilleke S, Gunatilleke N et al.
2006a. Testing metabolic ecology theory for allometric scaling of tree size,
growth and mortality in tropical forests. Ecology Letters 9: 575–588.
Muller-Landau HC, Condit RS, Harms KE, Marks CO, Thomas SC,
Bunyavejchewin S, Chuyong G, Co L, Davies S, Foster R et al. 2006b.
Comparing tropical forest tree size distributions with the predictions of
metabolic ecology and equilibrium models. Ecology Letters 9: 589–602.
Nepstad DC, Tohver IM, Ray D, Moutinho P, Cardinot G. 2007.Mortality of
large trees and lianas following experimental drought in an amazon forest.
Ecology 88: 2259–2269.
Phillips OL, van der Heijden G, Lewis SL, Lopez-Gonzalez G, Arag~ao LEOC,
Lloyd J, Malhi Y, Monteagudo A, Almeida S, Davila EA et al. 2010.
Drought–mortality relationships for tropical forests. New Phytologist 187: 631–
646.
Powell TL, Galbraith DR, Christoffersen BO, Harper A, Imbuzeiro HMA,
Rowland L, Almeida S, Brando PM, da Costa ACL, Costa MH et al.
2013. Confronting model predictions of carbon fluxes with measurements
of Amazon forests subjected to experimental drought. New Phytologist 200:
350–365.
Roberts J, Cabral OMR, Aguiar LFD. 1990. Stomatal and boundary-layer
conductances in an Amazonian terra firme rain forest. Journal of Applied Ecology
27: 336–353.
Rowland L, da Costa ACL, Galbraith DR, Oliveira RS, Binks OJ, Oliveira AA,
Pullen AM, Doughty CE, Metcalfe DB, Vasconcelos SS et al. 2015. Death
from drought in tropical forests is triggered by hydraulics not carbon starvation.
Nature 528: 119–122.
Ryan MG, Phillips N, Bond BJ. 2006. The hydraulic limitation hypothesis
revisited. Plant, Cell & Environment 29: 367–381.
Schietti J, Martins D, Emilio T, Souza PF, Levis C, Baccaro FB, da Veiga Pinto
JLP, Moulatlet GM, Stark SC, Sarmento K et al. 2016. Forest structure along
a 600 km transect of natural disturbances and seasonality gradients in central-
southern Amazonia. Journal of Ecology 104: 1335–1346.
Sitch S, Huntingford C, Gedney N, Levy PE, Lomas M, Piao SL, Betts R, Ciais
P, Cox P, Friedlingstein P et al. 2008. Evaluation of the terrestrial carbon
cycle, future plant geography and climate-carbon cycle feedbacks using five
Dynamic Global Vegetation Models (DGVMs). Global Change Biology 14:
2015–2039.
Snyder PK, Delire C, Foley JA. 2004. Evaluating the influence of different
vegetation biomes on the global climate. Climate Dynamics 23: 279–302.
Swenson NG, Enquist BJ. 2007. Ecological and evolutionary determinants of a
key plant functional trait: wood density and its community-wide variation
across latitude and elevation. American Journal of Botany 94: 451–459.
US DOE. 2012. Research priorities for tropical ecosystems under climate change
workshop report, DOE/SC-0153. US Department of Energy Office of Science.
West GB, Enquist BJ, Brown JH. 2009. A general quantitative theory of forest
structure and dynamics. Proceedings of the National Academy of Science, USA
106: 7040–7045.
Wolfe BT, Sperry JS, Kursar TA. 2016. Does leaf shedding protect stems from
cavitation during seasonal droughts? A test of the hydraulic fuse hypothesis.
New Phytologist 212: 1007–1018.
Wright SJ. 1991. Seasonal drought and the phenology of understory shrubs in a
tropical moist forest. Ecology 72: 1643–1657.
Wright SJ, Kitajima K, Kraft NJB, Reich PB, Wright IJ, Bunker DE, Condit R,
Dalling JW, Davies SJ, Dıaz S et al. 2010. Functional traits and the growth–
mortality trade-off in tropical trees. Ecology 91: 3664–3674.
Zhang Y-J, Meinzer FC, Hao G-Y, Scholz FG, Bucci SJ, Takahashi FSC,
Villalobos-Vega R, Giraldo JP, Cao K-F, Hoffmann WA et al. 2009. Size-
dependent mortality in a Neotropical savanna tree: the role of height-related
adjustments in hydraulic architecture and carbon allocation. Plant, Cell &
Environment 32: 1456–1466.

 2017 Smithsonian Institute
New Phytologist 
 2017 New Phytologist Trust
New Phytologist (2017)
www.newphytologist.com
New
Phytologist Research 11
Supporting Information
Additional Supporting Information may be found online in the
Supporting Information tab for this article:
Fig. S1 Size scaling of stem density, aboveground live biomass C,
aboveground woody productivity and woody mortality using
biomass allometries from Chave et al. (2005).
Fig. S2 Size scaling of mean individual Cag,live, individual diame-
ter growth, individual biomass C growth and stem mortality rate.
Fig. S3 Size scaling of stem density, aboveground live biomass C,
aboveground woody productivity and woody mortality on a non-
linear scale.
Table S1 List of species classified as deciduous in this study along
with source of deciduous observation
Table S2 Ecosystem-level C variables for all three sites during
non-El Ni~no census periods using biomass allometries from
Chave et al. (2005)
Table S3 Adjusted woody mortality values for the 1981–1985
Barro Colorado Island census period
Table S4Demographic rates and C cycle variables by size class
for each site and census period
Table S5 Fitted parameters corresponding to Fig. 1
Table S6 Fitted parameters corresponding to Fig. 2
Methods S1Methods for classifying deciduous species.
Methods S2Corrections to forest census data and exclusion of
outliers.
Methods S3Demographic variables.
Please note: Wiley Blackwell are not responsible for the content
or functionality of any Supporting Information supplied by the
authors. Any queries (other than missing material) should be
directed to the New Phytologist Central Office.
New Phytologist is an electronic (online-only) journal owned by the New Phytologist Trust, a not-for-profit organization dedicated
to the promotion of plant science, facilitating projects from symposia to free access for our Tansley reviews. 
Regular papers, Letters, Research reviews, Rapid reports and both Modelling/Theory and Methods papers are encouraged. 
We are committed to rapid processing, from online submission through to publication ‘as ready’ via Early View – our average time
to decision is <26 days. There are no page or colour charges and a PDF version will be provided for each article. 
The journal is available online at Wiley Online Library. Visit www.newphytologist.com to search the articles and register for table
of contents email alerts.
If you have any questions, do get in touch with Central Office (np-centraloffice@lancaster.ac.uk) or, if it is more convenient,
our USA Office (np-usaoffice@lancaster.ac.uk)
For submission instructions, subscription and all the latest information visit www.newphytologist.com
New Phytologist (2017) 
 2017 Smithsonian Institute
New Phytologist
 2017 New Phytologist Trustwww.newphytologist.com
Research
New
Phytologist12