Ecology, 88(9), 2007, pp. 2259–2269
? 2007 by the Ecological Society of America
MORTALITY OF LARGE TREES AND LIANAS FOLLOWING
EXPERIMENTAL DROUGHT IN AN AMAZON FOREST
DANIEL C. NEPSTAD,1,2,3INGRID MARISA TOHVER,2DAVID RAY,1PAULO MOUTINHO,1,2AND GEORGINA CARDINOT2
1Woods Hole Research Center, 149 Woods Hole Road, Falmouth, Massachusetts 02540-1644 USA
2Instituto de Pesquisa Ambiental da Amazo ˆnia, Avenida Nazare´ 669, CEP-66035-170, Belem, Para ´, Brazil
Oscillation (ENSO) events influence large areas of tropical forest and may become more
frequent in the future. One of the most important forest responses to severe drought is tree
mortality, which alters forest structure, composition, carbon content, and flammability, and
which varies widely. This study tests the hypothesis that tree mortality increases abruptly
during drought episodes when plant-available soil water (PAW) declines below a critical
minimum threshold. It also examines the effect of tree size, plant life form (palm, liana, tree)
and potential canopy position (understory, midcanopy, overstory) on drought-induced plant
mortality. A severe, four-year drought episode was simulated by excluding 60% of incoming
throughfall during each wet season using plastic panels installed in the understory of a 1-ha
forest treatment plot, while a 1-ha control plot received normal rainfall. After 3.2 years, the
treatment resulted in a 38% increase in mortality rates across all stems .2 cm dbh. Mortality
rates increased 4.5-fold among large trees (.30 cm dbh) and twofold among medium trees
(10–30 cm dbh) in response to the treatment, whereas the smallest stems were less responsive.
Recruitment rates did not compensate for the elevated mortality of larger-diameter stems in
the treatment plot. Overall, lianas proved more susceptible to drought-induced mortality than
trees or palms, and potential overstory tree species were more vulnerable than midcanopy and
understory species. Large stems contributed to 90% of the pretreatment live aboveground
biomass in both plots. Large-tree mortality resulting from the treatment generated 3.4 times
more dead biomass than the control plot. The dramatic mortality response suggests
significant, adverse impacts on the global carbon cycle if climatic changes follow current
Severe drought episodes such as those associated with El Nin ˜ o Southern
El Nin˜o; ENSO; forest mortality; lianas; tropical forest.
Amazon rain forest; biomass; Brazil; carbon flux; climate change; dead biomass; drought;
Severe episodic droughts change the structure and
function of moist tropical forests in important ways.
Supra-annual drought events such as those associated
with El Nin ˜ o Southern Oscillation (ENSO) episodes are
sometimes accompanied by higher adult tree mortality
(Leighton and Wirawan 1986, Condit et al. 1995, 2004,
Kinnaird and O’Brien 1998, Nakagawa et al. 2000,
Williamson et al. 2000, van Nieuwstadt and Sheil 2005),
higher or lower seedling mortality (Engelbrecht et al.
2002, Condit et al. 2004, Edwards and Krockenberger
2006), mast fruiting of some tree species (Curran et al.
1999, Wright et al. 1999), reduced seed set in other tree
species (Brando et al. 2006, Wright and Calderon 2006),
and increased forest flammability (Woods 1989, Nepstad
et al. 1999, 2004, Ray et al. 2005, van Nieuwstadt and
Sheil 2005). Soil moisture depletion during severe
drought can stimulate stomatal closure, a decline in leaf
area (Nepstad et al. 1994), and a reduction in wood
production (Nepstad et al. 2002, 2004), perhaps sup-
pressing net primary productivity on regional scales
(Tian et al. 1998). Of these responses to severe drought
episodes, the mortality of adult trees may be the most
significant. Large trees determine the three-dimensional
structure of the forest, the degree of coupling between the
microclimate of the forest understory and leaf canopy,
and contain more carbon than any other moist tropical
forest pool (with the exception of soil carbon, much of
which cycles on millennial time scales [Trumbore et al.
1995]). High tree crowns that maintain substantial leaf
area duringseasonal droughts cast deep shade,helping to
keep air vapor pressure deficits low near the ground, and
preventing the fine fuel layer from drying sufficiently to
ignite when agricultural fires provide sources of ignition
(Ray et al. 2005). In this context, drought-induced tree
mortality can initiate a positive feedback loop in which
increased drying of the forest interior enhances the
likelihood of forest burning, and that burning kills more
trees, further diminishing the protective shade of the
high, dense leaf canopy (Nepstad et al. 1995, 2001,
Cochrane et al. 1999). The drought-induced death of
centuries-old trees may be registered in the structure and
carbon exchange of moist tropical forests for decades,
Manuscript received 19 June 2006; revised 26 January 2007;
accepted 2 February 2007. Corresponding Editor: F. C.
causing net outflux of carbon as large dead biomass
pools (dead wood and vegetation) decompose (Saleska et
al. 2003, Rice et al. 2004) or net uptake of carbon as
growing trees outpace respiratory losses. Our ability to
predict the future of moist tropical forests within
scenarios of more frequent and intense episodic droughts
(Trenberth and Hoar 1997, Foley et al. 2002, Cox et al.
2004, Meir et al. 2006) and the role of these forests in
contributing to the fluxes of carbon dioxide and other
greenhouse gases to the atmosphere is currently limited
by the paucity of our knowledge of how tropical trees
respond to severe drought.
Tropical forests in regions of little rainfall seasonality
have exhibited the highest stand-level tree mortality
following severe drought. Tree mortality recorded after
severe drought episodes ranges from a high of 26% in a
forest with aseasonal rainfall in East Kalimantan 21
months after the severe drought episode of the
1997/1998 ENSO event (trees .10 cm dbh [van
Nieuwstadt and Sheil 2005]), to no mortality response
in two of three seasonally dry forests studied in Panama
following the same event (1–10 cm and .10 cm dbh
[Condit et al. 2004]); dbh is diameter at breast height.
Within this broad range of mortality responses, several
intermediate tree mortality responses to ENSO episode
droughts have been reported. Tree mortality in one of
the Panama forests that exhibited no response in
1997/1998 increased 50% following the 1983 ENSO
episode, from 2% to 3% (Condit et al. 1995). In the
Central Amazon, which typically has a mild dry season,
tree mortality increased slightly from 1.1%/yr pre-ENSO
to 1.9%/yr following the 1997/1998 ENSO event (.10
cm dbh [Williamson et al. 2000]). In an aseasonal
Sarawak forest, tree mortality climbed from 0.9% pre-
ENSO to 6.4 and 4.3% during the 1997/1998 ENSO
episode, with little variation across topography (.10 cm
dbh [Nakagawa et al. 2000]). Tree mortality was 10% in
aseasonal Sumatra following the 1997/1998 ENSO
(Kinnaird and O’Brien 1998) and 14–24% following
the 1983 ENSO in Kalimantan (Leighton and Wirawan
1986). In a study of the response of nine forests on
Mount Kinabalu, Borneo (Sabah), tree mortality did
not respond to the 1997/1998 ENSO episode on upland
sites with nutrient-poor soils, but increased two- to
sixfold on the other sites (Aiba and Kitayama 2002).
It is more difficult to discern patterns in the
differential responses of plant types to severe drought.
In Panama and East Kalimantan, large trees suffered
higher mortality rates than juvenile saplings or shrubs
following the 1983 and 1997/1998 ENSO episodes,
respectively (Condit et al. 1995, van Nieuwstadt and
Sheil 2005), whereas in Sarawak small trees died more
than large trees (Nakagawa et al. 2000). In the Central
Amazon, post-ENSO mortality was even across tree size
and species (Williamson et al. 2000). Condit et al. (1995)
found that gap-colonizing tree species were more
vulnerable to drought-induced mortality than generalist
species, but only in smaller size classes. In some cases,
individual taxa or morphological features were associ-
ated with significantly higher drought-induced mortali-
ty, such as dipterocarps in Sarawak (Nakagawa et al.
2000) and trees with low wood density in East
Kalimantan (van Nieuwstadt and Sheil 2005).
One explanation for the considerable variation in
response of tree mortality to severe drought episodes
may be the effects of the drought on the amount of soil
water that is available to plants at each site (plant-
available soil water, PAW), and the ability of different
plant types to access this soil water. None of the studies
cited here present measurements of PAW, so it is
currently impossible to examine the role of PAW in
explaining this variation. With depletion of PAW in
surface soil layers, shallow-rooted plants may die more
rapidly than deeply rooting plants. As drought proceeds
and deeper PAW is exhausted, the mortality of large
trees (that tend to have deeper root systems) may exceed
that of small trees, since tall tree crowns have greater
exposure to sunlight and associated evaporative demand
than the partially shaded plants growing in the
understory. (Understory and midcanopy trees, in
contrast, receive partial shade from overstory species.)
Hence, tropical forest ecosystems may reach a critical
minimum threshold in their response to drought; this
threshold is defined by the rainfall history at which
PAW is depleted throughout the rooting zone. The
identification of such a threshold, if it exists, could
become an important task in the development of
vegetation models that predict the response of tropical
vegetation to future climate scenarios, and would
demand increasingly sophisticated knowledge of PAW
(Nepstad et al. 2004, Hutyra et al. 2005).
We established a large throughfall exclusion experi-
ment in an east-central Amazon forest and monitored
PAW to 11 m depth to simulate a severe four-year
drought episode, testing the hypothesis that PAW
depletion reaches a critical threshold beyond which
large trees begin to die. Like tall trees, lianas are also
exposed to high sunlight and evaporative demand, and
may also exhibit a threshold response. By experimentally
examining plant response to drought, we were able to
separate the effects of soil moisture depletion from the
effects of increasing solar radiation, factors that are
often confounded during natural drought episodes. We
addressed the following questions. (1) Is drought-
induced mortality of medium and large trees associated
with depletion of PAW beyond a critical threshold? (2)
Is the mortality response dependent upon life form
(trees, palms, and lianas)? (3) Is the mortality response
dependent upon the potential canopy position of the
individual (overstory, midcanopy, and understory)? And
(4), how much live biomass is killed by severe drought?
Site description and study design
The throughfall exclusion experiment is located in the
Tapajo ´ s National Forest in the east-central Brazilian
DANIEL C. NEPSTAD ET AL.2260Ecology, Vol. 88, No. 9
Amazon (285401200S, 5485701200W). The region receives
an average annual rainfall of ;2000 mm, varying
between 600 and 3000 mm/yr, and suffers severe drought
stress during ENSO years. Soils are deeply weathered
Oxisols (Haplustox), with no concretions or impeding
layers in the upper 12 m. The water table is ;100 m
below the soil surface. Beginning in late January 2000
through December 2004, ;60% of throughfall (equiva-
lent to approximately half the rainfall) was excluded
from the 1-ha treatment plot (dry plot, D) using a
system of ;5660 plastic panels located in the forest
understory during the wet seasons (see Nepstad et al.
2002 for a complete description of the study and site).
Treatment effects were assessed relative to measure-
ments collected from a similar 1-ha control plot (wet
plot, W) located nearby. Trenches (1.0–1.7 m deep) were
dug around the perimeters of both plots to inhibit root
access to water from the surrounding area for the D
plot, and to balance any damage to root systems in the
case of the W plot.
Plant-available water and leaf water potential
We measured soil water in both plots monthly using
time domain reflectometry (TDR) (Jipp et al. 1998). Six
deep soil shafts were equipped with 13 pairs of 30 cm
long TDR probes. One pair of probes per shaft was
installed vertically at the soil surface. The remaining
probes were installed horizontally into the walls of the
shaft at depths of 0.5, 1, and 2 m, and thereafter at 1-m
intervals to a depth of 11 m. Volumetric soil water
contents (VWC) obtained from the TDR measurements
were used to derive estimates of plant-available water
(PAW) within the soil profile over time.
PAW for the entire soil profile (intervals from 0 to 200
cm and 200–1100 cm deep) was determined by setting
the permanent wilting point (minimum PAW) equal to
the average minimum VWC recorded at each measure-
ment depth in the D plot. After establishing the
minimum values of VWC, PAW was determined as the
difference between the observed VWCs averaged across
sensors for each month and the global minimum values.
The final, derived PAW variable is expressed as a
percentage of the maximum potential for this soil
(%PAWmax¼ 100 3 [VWC at field capacity ? VWC at
minimum measured value]?1for each depth), and is
integrated over two depth intervals in the measured soil
profile (0–2 m and 2–11 m deep).
We measured predawn leaf water potentials (LWP;
leaf water potentials were always measured at this time)
of six common understory and midcanopy tree species
found in both plots, using a pressure chamber,
beginning in January 2000. Three mature individuals
per species were repeatedly measured in each plot by
sampling four leaves per individual biweekly. These data
were averaged across species in each plot at each date to
provide an estimate of plant water stress for each area.
We also calculated a water stress integral (WSI) for each
plot (Myers 1988). This procedure involved accumulat-
ing, over time, deviations in LWP relative to the
maximum observed value determined for each species,
providing a long-term index of plant water stress:
where LWPmaxis the least negative value recorded for
LWP over time, and n is the number of days in the
interval. The daily values for WSI determined in each area
were summed over the period of the study (t, in days). We
determined additional drought stress provoked by the
treatment as the difference in the WSI between the W and
D plots over time. When expressed as the difference
between plots over the long term, WSI is advantageously
linked directly to LWP, providing a physiologically based
estimate of the treatment’s cumulative impact on plant
water relations. The fact that LWP recovered during the
wet season in the D plot even while %PAWmaxcontinued
to decline provides evidence of this association.
We completed an initial survey of stems ?1 cm dbh in
March 2000, including identification to species and
measurement of diameter at breast height (dbh ¼ 1.3
m). Diameters of highly irregular or buttressed stems
were measured just above the irregularity. We remea-
sured all previously inventoried individuals ?2 cm in
August 2003, providing a census interval of 3.2 yr. Stems
1–2 cm dbh were remeasured in September 2003 in four
2-m2subplots within each 1-ha plot to provide an
estimate of recruitment into the ?2 cm dbh size class. In
addition, we monitored all trees .10 cm dbh monthly to
assess their condition (live vs. dead), and to measure their
radial increment using steel dendrometer bands. Plants
with their bases located within 5 m of the edge of the
trench were excluded from these analyses to minimize the
influence of root damage. Finally, we eliminated a 20-m2
area in the W plot from the analyses because the
pretreatment death and subsequent collapse of one large
tree damaged surrounding trees. Effective sampling areas
used in the study were 0.81 ha (D) and 0.77 ha (W).
Categories of stems
We categorized plants in the initial survey by size
classes, life forms, and potential canopy positions for
tree species. Four stem size classes were examined: 2.0–
5.0, 5.1–10.0, 10.1–30.0, and .30 cm dbh. Three life
forms were identified: trees, lianas, and palms. Trees
were classified according to overstory, midcanopy, and
understory, based on the species’ potential canopy
position at maturity (as described by Ribeiro et al.
1999). Tree species not found in the reference (n ¼ 17)
were omitted from this part of the analysis.
Mortality and recruitment rates
Mortality was assessed as death or disappearance of
previously tagged stems. Standing stems were considered
September 20072261MORTALITY OF AMAZON TREES
dead if their inner bark was dry and easily separated
from the sapwood. The following equation was used to
calculate annual mortality rates (m):
m ¼ ½1 ? ð1 ? SD=NÞ1=t?ð2Þ
where N is the initial number of stems, SDis the number
of dead stems at the end of the period, and t is the time
interval between surveys measured in years (Sheil and
Growth into larger diameter size classes was assessed
by annual recruitment rates (r) using
r ¼ ½ð1 þ X=NÞ1=t? 1?ð3Þ
where X is the number of stems growing into a larger
diameter class within the interval, and N and t are as
defined following Eq. 2 (Nakagawa et al. 2000).
Survivorship within the two largest diameter size
classes of trees (10–30 and .30 cm dbh) was determined
on a monthly basis. Percent survival at each sample date
is the ratio of the number of surviving stems to those
present in the initial inventory. Recruited stems were
omitted from this analysis.
The scientific name, number of individuals, and
number of deaths of each species in this study is
summarized in the Appendix.
We estimated the aboveground biomass of all
measured plants using the following allometric equa-
tions: (1) moist forest equation presented by Chambers
et al. (2001) for trees ?5 cm dbh; (2) primary forest
equation of Nascimento and Laurance (2002) for trees
2–5 cm dbh; (3) the equation of Gerwing and Farias
(2000) for woody lianas; and (4) the equation presented
by Hughes et al. (1999) for stemmed palms. Biomass
values were converted to carbon equivalents by multi-
plying dry mass values by 0.5.
We used a 2 3 2 contingency table analysis to assess
differences in mortality between plots. Initial stem
counts and the number of dead individuals at the end
of the census interval within each plot provided the cell
values. Hypothesis tests were carried out using one-
tailed probabilities based on Fisher’s exact test (Ha:
mortality in the treatment plot has increased due to the
drought). This procedure was followed for each of the 10
plant categories. Binomial probabilities were used to
estimate 95% CI of the annual mortality rates deter-
mined for each plant category.
The experiment excluded ;60% of throughfall from
the soil when the panels were in place; a total of 620–890
mm of throughfall was excluded on an annual basis. We
therefore induced an effective annual rainfall of 1810,
1120, and 1130 mm in 2000–2002 and 630 mm during
the first eight months of 2003 (Fig. 1). Since 2000 rainfall
was above average (2700 mm), effective rainfall during
year 1 of the experiment was similar to the average for
the site. In 2001 and 2002, effective rainfall was higher
than total rainfall during the 1998 ENSO episode, when
the site received only 630 mm of rainfall. In sum, the
rainfall regime simulated during the study period was
similar to rainfall during a moderate to severe ENSO
event, with the important difference that the low level of
effective rainfall was maintained for nearly three years.
In the upper 2 m of the soil profile PAW was reduced
to 41% of PAWmaxin the D plot, while the W plot
remained at 83% PAWmaxafter the second wet season
(January–July 2001) (Fig. 2b). Deep soil water under-
went a similar pattern of depletion during this period in
the D plot. Within 2–11 m of the soil profile, soil water
dropped to 47% and 87% PAWmaxin the D and W plots,
respectively (Fig. 2c). After the second season of
throughfall exclusion, drought stress also impacted
canopy water status, inducing a 0.66 MPa difference in
LWP between the two plots (Fig. 2a). During the fourth
exclusion period, soil water in the D plot declined to
14% and 3% PAWmaxwithin the 0–2 m and 2–11 m
profiles, respectively. The corresponding values of
%PAWmaxfor the W plot were uncommonly low, 58%
and 43% in August 2003, reflecting incomplete wet
season recharge of soil water following the unusually
strong dry season in 2002 (Fig. 2d). Following four
seasons of throughfall exclusion, the normal pattern of
soil water recharge and depletion, discernible in the W
plot, became indistinguishable in the D plot (Fig. 2).
Drought-induced changes in stem recruitment
and mortality rates
Stem distributions by size class were similar for both
plots in March 2000 when the initial inventory was
rainfall (rainfall minus water excluded by plastic panels;
measured in dry plot) during 3.75 years of the throughfall
Annual rainfall (measured in wet plot) and effective
DANIEL C. NEPSTAD ET AL.2262 Ecology, Vol. 88, No. 9
conducted (Fig. 3a). There were 242 tree, liana, and
palm species common to both plots, representing 73%
and 63% of the total number of species encountered in
the D (n¼339) and W (n¼392) plots, respectively. The
distribution of species across life forms was nearly
identical in both plots: trees represented ;78%, lianas
21%, and palms the remaining 1% of the total number of
species in the study plots. Considering only life forms,
trees composed 88% (W) and 86% (D), lianas 11% (W),
and 13% (D), and palms 1% (W) and 2% (D) of stems
.2 cm dbh. The distribution of trees by potential
canopy positions was likewise similar in both plots.
Small-diameter trees (2–5 cm dbh) were the most
abundant size class for each of the three canopy position
categories in both plots. About 50% of all potential
overstory and midcanopy species were represented as
experiment, showing (a) predawn leaf water potentials averaged across six species (mean 6 SE; n¼3 trees per species, n¼4 leaves
per tree) in both plots; (b) plant-available soil water as a percentage of the maximum value (%PAWmax) for 0–2 m; (c) %PAWmax
for 2–11 m in the soil profile; and (d) daily precipitation. Vertical hatching indicates periods when the throughfall exclusion system
was functioning during the wet season.
Selected components of the water balance within the wet (W) and dry (D) plots at the Tapajo ´ s throughfall exclusion
September 20072263 MORTALITY OF AMAZON TREES
juvenile saplings and 87% of all understory species were
represented as small trees falling into this size class.
Reduced growth rates (D. Nepstad, D. Ray, and P.
Brando, unpublished data) and increased mortality
changed recruitment patterns in the D plot. Despite
average recruitment rates of ;3–6%/yr across all size
classes in both plots, mortality in the larger size classes
surpassed recruitment in the dry plot (Fig. 3b, c).
Drought-induced mortality exceeded recruitment by
2.9%/yr and 6.7%/yr for stems 10–30 cm dbh and .30
cm dbh, respectively (Fig. 3c). By contrast, mortality
was offset by recruitment among smaller diameter
plants, thereby maintaining a positive structural change
in the 2–5 cm and 5–10 cm size classes. Thus, the net
effect of the drought treatment on forest structure was a
10% reduction in the number of larger trees (?10 cm
Partial throughfall exclusion amplified mortality rates
among all stems, but particularly for the larger trees
(Fig. 4a). Following 3.2 yr of measurements, the
community-wide mortality rate in the D plot
(3.77%/yr) significantly exceeded the W plot mortality
rate (2.72%/yr) by 38% (P , 0.001). Only the smallest
size class, 2–5 cm, did not undergo significantly greater
mortality rates in the D plot (P ¼ 0.13). In all other size
classes mortality rates were considerably greater in the
dry plot than in the wet plot. Mortality rates were 69%
higher in the dry plot than the wet plot in the 5–10 cm
size class, 98% greater in the 10–30 cm size class, and
445% higher in the .30 cm size class (Fig. 4a).
Independent of stem size, lianas demonstrated a greater
susceptibility to drought than either trees or palms (Fig.
4b). The mortality rate for lianas was ;70% greater in
the dry plot, 6.78%/yr, than in the wet plot, 3.78%/yr (P
¼ 0.004). Trees endured significantly higher mortality in
the D plot, 3.36%/yr, than in the W plot, 2.54%/yr, (P¼
0.009). Palms also underwent higher mortality rates of
similar magnitude to that of trees; however, the sample
(a) the distribution of stems by diameter size class in the wet
(W) and dry (D) plots at the time of the first inventory in March
2000; (b) stem recruitment rates by size class over the interval
between the initial inventory and August 2003 (ingrowth refers
to stems that advanced to a larger diameter class through stem
growth). Part (c) shows the percentage change in numbers of
stems in each size class based on the difference between
mortality and recruitment rates.
Aspects of forest structure at the study site, showing
vegetation in the wet (W) and dry (D) plots. Groupings include
(a) four diameter size classes, (b) three life forms, and (c) three
tree canopy positions. Asterisks above the paired bars indicate a
significantly higher mortality rate in the D than in the W plot
(one-tailed Fisher’s exact test: ** P , 0.01, *** P , 0.001).
Annual mortality rates for different groupings of
DANIEL C. NEPSTAD ET AL.2264 Ecology, Vol. 88, No. 9
size was much smaller for palms and the difference was
not significant (P ¼ 0.402).
Tree species that potentially grow to reach the upper
canopy at maturity were more vulnerable to drought
than those potentially reaching only the midcanopy or
understory (Fig. 4c). Mortality rates were more than
double in D than in W for the overstory species (mW¼
1.75%/yr, mD¼ 3.80%/yr; P , 0.0001), but remained
similar for the midcanopy species (mW¼3.35%/yr, mD¼
3.24%/yr; P ¼ 0.44), and only modestly elevated among
the understory species (mW¼ 2.52%/yr, mD¼ 3.18%/yr;
P¼0.087). Mortality rates and 95% confidence intervals
for all species in both plots are listed in the Appendix.
Medium- and large-tree survivorship
Changes in soil water and associated drought stress
resulting from the treatment are the best indicators of
large-tree mortality responses to drought. Elevations in
large-tree mortality were delayed until late in the third
dry season of the water exclusion treatment, when
%PAWmaxreached its lowest levels in both 0–2 and 2–11
m depth intervals (Figs. 2 and 5). When the large-tree
mortality response was triggered, trees exceeding 30 cm
dbh underwent the fastest decline in survivorship. From
the middle of the third to the fourth dry season of the
study period, the 10–30 cm size class declined from 94%
to 84% of its initial population in the D plot and from
95% to 93% in the W plot (Fig. 5a). In the same
sampling interval the .30 cm size class declined by 21%
of its initial population, from 100% to 79%, in the D plot
and remained at 95% survivorship in the W plot (Fig.
The difference between the two plots in survivorship
of medium and large trees appears to be associated with
a threshold difference in the plots’ water stress integral.
Prior to an accumulation of 100 MPa-days (the unit of
the WSI; see Methods: Plant-available water and leaf
water potential) of differential water stress between the
two plots in October 2002, there was no relationship
between water stress integral (WSI) and large-tree
survivorship (Fig. 5c). However in November 2002, the
plots crossed a difference threshold of 100 MPa-days
WSI and a positive association between WSI and large-
tree survivorship emerged. Subsequently, as the WSI
increased from 100 to 120 MPa-days, survivorship of the
original population of large trees in the W plot exceeded
the D plot by .10%. The escalating WSI in the D plot
during the fourth throughfall exclusion period in 2003
indicates that wet season recharge was insufficient to
curb drought stress.
The delayed mortality response in medium and large
trees was also linked to the cumulative changes in PAW
throughout the soil profile. Prior to the declines in large-
tree survivorship in the D plot, PAW fell below 30%
PAWmaxfor the entire soil profile (0–11 m) in October
2001 (Fig. 2). Subsequently, soil water was maintained
below 30% PAWmaxin the D plot for the rest of the
study period, except for a brief spike in November 2002
at 0–2 m depth due to heavy rains during the dry season.
In contrast, %PAWmaxin the W plot dropped below
30% only at the end of the 2002 dry season. Unlike the
protracted reduction in the D plot, this brief decline in
%PAWmax in the W plot was not accompanied by
increased tree mortality. The cumulative difference in
PAW between the two plots rose gradually for the 0–2 m
profile for the duration of the study period, but
increased markedly for the 2–11 m profile, particularly
after the 2001 dry season (Fig. 5d).
Biomass and dead biomass (necromass)
The aboveground standing biomass of stems ?2 cm
dbh was slightly higher in the D plot (310.6 Mg/ha) than
in the W plots (302.6 Mg/ha) in March 2000, when the
initial inventory was completed (Fig. 6a). At that time
stems .10 cm dbh accounted for ;90% of the
pretreatment standing live biomass in both plots.
Drought-induced mortality generated a large net in-
crease in the amount of dead biomass in the D plot.
Following four seasons of the treatment, the total
quantity of aboveground dead biomass was 46.5 Mg/ha
greater in the D plot (65.9 Mg/ha) than in the W plot
(19.4 Mg/ha). Furthermore, unlike the W plot, the
change in standing biomass in the D plot failed to offset
the dead biomass pools created (Fig. 6b). These changes
represent a net increase of ‘‘committed emissions to the
atmosphere’’ of 7.3 Mg?ha?1?yr?1over the study period
(Fig. 7). Transfer of live biomass to the dead biomass
pool represents ‘‘committed’’ carbon emission, insofar as
most of it will eventually be released to the atmosphere
through decay (Fearnside 1997). Larger diameter stems
(.10 cm dbh) accounted for 95% of the total dead
biomass generated, reflecting both their relatively higher
mortality rates and C content. Inputs to the dead
biomass pool were approximately two times higher for
the 5–10 and 10–30 cm dbh size classes and over five
times greater for the .30 cm dbh size class in the D plot
than in the W plot.
The seasonally dry forest of the Tapajo ´ s is an
extremely drought-tolerant ecosystem. Our partial
throughfall exclusion simulated effective rainfall of
1120, 1130, and 630 mm during the second through
fourth years of the study period (eight months in the
final year), well below evapotranspirational water loss of
;1400–1500 mm/yr measured at this site during the
same period (Saleska et al. 2003). But, surprisingly, the
mortality of large trees (.10 cm dbh) began only during
the final year of the experiment (Fig. 5). This drought
tolerance is conferred in part by deeply penetrating root
systems that absorb soil water to depths of at least 11 m
(Nepstad et al. 1994, 2004, Moreira et al. 1997, Jipp et
al. 1998, Bruno et al. 2006). It may also be facilitated by
hydraulic redistribution of soil water documented at this
site (Oliveira et al. 2005), in which the nighttime flow of
soil water upward through vertical roots and then away
September 2007 2265MORTALITY OF AMAZON TREES
from the stem into the soil via lateral roots facilitates
subsequent daytime water uptake from shallow soils.
Finally, drought tolerance may be associated with
significant uptake of water directly by leaves during
dry season rains and nighttime dew events, also
documented at this site (G. Cardinot, M. Moreira, L.
Sternberg, D. Nepstad, R. Oliveira, T. Dawson, and S.
Burgess, unpublished manuscript).
The study supports the hypothesis that the mortality
of large trees increases precipitously when soil water
reaches a critical minimum threshold. Mortality of
medium-sized trees (10–30 cm dbh) and large trees
(.30 cm dbh) increased dramatically as the water stress
integral difference reached 100 MPa-days (Fig. 5). The
very high levels of mortality during ENSO drought
observed in Borneo and Sumatra (Leighton and
Wirawan 1986, Kinnaird and O’Brien 1998, Nakagawa
et al. 2000, Aiba and Kitayama 2002, van Nieuwstadt
and Sheil 2005) may have been associated with the
depletion of soil water below a critical minimum
at the throughfall exclusion experiment; (c) relative differences between plots in terms of the cumulative water stress integral; and
(d) cumulative differences in plant-available soil water for two depth intervals in the soil profile. Vertical hatching indicates periods
when the throughfall exclusion system was functioning during the wet season.
Survivorship curves for trees (a) 10–30 cm dbh and (b) .30 cm dbh sampled monthly in the wet (W) and dry (D) plots
DANIEL C. NEPSTAD ET AL.2266 Ecology, Vol. 88, No. 9
threshold—a threshold that varies with rooting depths,
plant physiological tolerance to drought, impenetrable
horizons, soil texture, and rainfall. Relatively mild soil
water depletion may provoke greater tree mortality in
these aseasonal forests if they are composed of trees that
are physiologically drought sensitive or unable to obtain
water during extended periods of drought through deep
rooting or other mechanisms.
Both the synchrony of the tree mortality response for
the two tree size classes and the relatively low rates of
small-stem mortality observed for the size classes ,10
cm dbh (Fig. 4) argue against the hypothesis that there is
an initial drought response in which the mortality of
small trees is greatest because of their shallower rooting
depths (Fig. 4). However, the study results support the
hypothesis that lianas, like large trees, are susceptible to
drought-induced mortality. In addition to their exposure
to solar radiation, this drought response may be
associated with their vulnerability to xylem cavitation
(Schnitzer and Bongers 2002). This response is also
consistent with observations of deep root systems
(Restom and Nepstad 2004); lianas may avoid
drought-induced mortality or tissue damage by tapping
deep soil moisture, but are vulnerable to mortality when
deep soil moisture is depleted. The trend toward
increasing densities of lianas in recent years, indicated
by permanent plot studies (Phillips et al. 2005), could be
dampened under scenarios of increasingly severe
The degree to which the Tapajo ´ s is representative of
other tropical forests subjected to episodic severe
drought must be evaluated using both climatic and
edaphic criteria. The rainfall regime of this forest is
similar to approximately one-third of the Amazon forest
formation in both total annual amount and seasonal
distribution (Nepstad et al. 2004). PAWmax to 10 m
depth (1100 mm) is lower than all but 10% of the Basin
(Nepstad et al. 2004). The total annual rainfall that we
simulated during the second and third years of the
experiment (1120 and 1130 mm/yr) (Fig. 1) is lower than
all but 5% of the forests of the Amazon for any given
year during the period 1996–2004 (Nepstad et al. 2004,
updated with recent rainfall data), but is within the
range of values already measured at the Tapajo ´ s site.
The effective rainfall induced by this experiment is rare
in the Amazon today, but could become much more
common in the future through climate change (Oyama
and Nobre 2003, Cox et al. 2004).
Soil water content can be strongly influenced by
groundwater, which is deeper in the Tapajo ´ s forest (100
m) than in much of the Amazon (Costa et al. 2002).
Shallow water tables that can be reached by tree root
systems may increase soil water availability. (In the long
term, however, transpirational uptake of groundwater
will lower its level, unless the aquifer is being recharged
elsewhere.) The water table is dynamic in regions of
seasonal rainfall, varying by up to 8 m in some forests
(Nepstad et al. 1994), recharging the moisture of soils
dried through transpirational uptake. Little is known,
however, of the ability of root systems to survive the low
oxygen levels that develop when the water table rises,
and it is possible that forests on soils with shallow water
tables experience annual dieback of deep root systems as
they become submersed. Further research is needed to
determine the influence of depth to water table on the
drought tolerance of moist tropical forests.
The simulated drought transferred nearly 30 Mg/ha of
live aboveground biomass to the dead biomass pool
time of original inventory in June 2000. (b) Net change in
biomass for each size class by the time of the August 2003
(a) Standing biomass distribution by size class at the
carbon,’’ in both plots for each size class by the August 2003
inventory (see Results: Biomass and dead biomass).
Quantity of dead biomass, presented as ‘‘committed
September 20072267 MORTALITY OF AMAZON TREES
through the death of medium and large trees and may
have reduced growth of the live biomass pool by another
20 Mg/ha (Figs. 6 and 7). The effect of the increase in
dead biomass on forest carbon exchange with the
atmosphere will be determined by the balance between
dead biomass decomposition rates, the rate of further
additions to this dead biomass pool from posttreatment
mortality, and the posttreatment accumulation of
photosynthetically fixed carbon in new wood (5 Mg/ha
in the control plot). The long-term effect of a severe
drought such as the one simulated would depend upon
other ecological and economic processes underway in
the landscape. For example, the posttreatment forest has
lower leaf area index and higher light penetration than
the undisturbed forest, with abundant fuels from
collapsing trees and new light-dependent colonizing
trees and grasses. Where propagules of grasses and
forbs are available from neighboring cattle pastures and
crop fields, drought-induced tree mortality can open
moist tropical forests to invasion from these highly
flammable, light-demanding plants, increasing the like-
lihood that the forest will catch fire, and initiating a
positive feedback of further tree mortality and increas-
ing flammability (Nepstad et al. 2001).
Coupled climate and vegetation models designed to
simulate future interactions between terrestrial ecosys-
tems and climate currently capture some of these
processes through the gradual displacement of one plant
form by another as climatic changes shift competitive
advantages (e.g., Oyama and Nobre 2003, Cox et al.
2004). Episodic events, such as severe droughts, may
accelerate these simulated vegetation shifts through
mortality of centuries-old canopy trees. Current emis-
sions of carbon to the atmosphere from tropical
deforestation comprise approximately one-fifth of global
average anthropogenic emissions (Prentice et al. 2001),
and these emissions could increase through the com-
bined effect of severe episodic drought and fire.
M. Nascimento Aviz, O. Portela, S. Nascimento, and R.
Baena assisted with field data collection, P. Lefebvre assisted
with graphics and analysis, K. Schwalbe, G. Carvalho, and W.
Kingerlee assisted with editing, and P. Brando, W. Rocha, and
W. Riker helped process and analyze data. This research was
supported by grants from the National Science Foundation
(No. DEB-0213011), the National Aeronautics and Space
Administration (LBA-ECO), the U.S. Agency for International
Development, and the Brazilian Science Council (CNPq).
Aiba, S. I., and K. Kitayama. 2002. Effects of the 1997–98 El
Nino drought on rain forests of Mount Kinabalu, Borneo.
Journal of Tropical Ecology 18:215–230.
Brando, P., D. Ray, D. Nepstad, G. Cardinot, L. M. Curran,
and R. Oliveira. 2006. Effects of partial throughfall exclusion
on the phenology of Coussarea racemosa (Rubiaceae) in an
east-central Amazon rainforest. Oecologia 150:181–189.
Bruno, R. D., H. R. da Rocha, H. C. de Freitas, M. L.
Goulden, and S. D. Miller. 2006. Soil moisture dynamics in
an eastern Amazonian tropical forest. Hydrological Processes
Chambers, J. Q., J. D. Santos, R. J. Ribeiro, and N. Higuchi.
2001. Tree damage, allometric relationships, and above-
ground net primary production in central Amazon forest.
Forest Ecology and Management 152:73–84.
Cochrane, M. A., A. Alencar, M. D. Schulze, C. M. Souza, Jr.,
D. C. Nepstad, P. A. Lefebvre, and E. A. Davidson. 1999.
Positive feedbacks in the fire dynamic of closed canopy
tropical forests. Science 284:1832–1835.
Condit, R., S. Aguilar, A. Hernandez, R. Perez, S. Lao, G.
Angehr, S. P. Hubbell, and R. B. Foster. 2004. Tropical
forest dynamics across a rainfall gradient and the impact of
an El Nino dry season. Journal of Tropical Ecology 20:51–
Condit, R., S. P. Hubbell, and R. B. Foster. 1995. Mortality
rates of 205 neotropical tree and shrub species and the impact
of a severe drought. Ecological Monographs 65:419–439.
Costa, M. H., C. H. C. Oliveira, R. G. Andrade, T. R.
Bustamante, F. A. Silva, and M. T. Coe. 2002. A macroscale
hydrological data set of river flow routing parameters for the
Amazon Basin. Journal of Geophysical Research—Atmo-
spheres 107: Article no. 8039.
Cox, P. M., R. A. Betts, M. Collins, P. P. Harris, C.
Huntingford, and C. D. Jones. 2004. Amazonian forest
dieback under climate-carbon cycle projections for the 21st
century. Theoretical and Applied Climatology 78:137–156.
Curran, L. M., I. Caniago, G. D. Paoli, D. Astianti, M.
Kusneti, M. Leighton, C. E. Nirarita, and H. Haeruman.
1999. Impact of El Nin ˜ o and logging on canopy tree
recruitment in Borneo. Science 286:2184–2188.
Edwards, W., and A. Krockenberger. 2006. Seedling mortality
due to drought and fire associated with the 2002 El Nino
event in a tropical rain forest in north-east Queensland,
Australia. Biotropica 38:16–26.
Engelbrecht, B. M. J., S. J. Wright, and D. De Steven. 2002.
Survival and ecophysiology of tree seedlings during El Nino
drought in a tropical moist forest in Panama. Journal of
Tropical Ecology 18:569–579.
Fearnside, P. M. 1997. Greenhouse gases from deforestation in
Brazilian Amazonia: net committed emissions. Climatic
Foley, J. A., A. Botta, M. T. Coe, and M. H. Costa. 2002. El
Nino-Southern Oscillation and the climate, ecosystems and
rivers of Amazonia. Global Biogeochemical Cycles 16(4):1–
Gerwing, J. J., and D. L. Farias. 2000. Integrating liana
abundance and forest stature into an estimate of total
aboveground biomass for an eastern Amazonian forest.
Journal of Tropical Ecology 16:327–335.
Hughes, R. F., J. B. Kauffman, and V. J. Jaramillo. 1999.
Biomass, carbon, and nutrient dynamics of secondary forests
in a humid tropical region of Mexico. Ecology 80:1892–1907.
Hutyra, L. R., J. W. Munger, C. A. Nobre, S. R. Saleska, S. A.
Vieira, and S. C. Wofsy. 2005. Climatic variability and
vegetation vulnerability in Amazonia. Geophysical Research
Letters 32:L24712 [doi: 10.1029/2005GL024981].
Jipp, P., D. C. Nepstad, and K. Cassel. and C. J. R. d.
Carvalho. 1998. Deep soil moisture storage and transpiration
in forests and pastures of seasonally-dry Amazonia. Climatic
Kinnaird, M. F., and T. G. O’Brien. 1998. Ecological effects of
wildfire on lowland rainforest in Sumatra. Conservation
Leighton, M., and N. Wirawan. 1986. Catastrophic drought
and fire in Borneo tropical rain forest associated with the
1982–1983 El Nino Southern Oscillation Event. Pages 75–102
in G. T. Prance, editor. Tropical rain forests and the world
atmosphere. AAAS Selected Symposium 101. Westview
Press, Boulder, Colorado, USA.
Meir, P., P. M. Cox, and J. Grace. 2006. The influence of
terrestrial ecosystems on climate. Trends in Ecology and
DANIEL C. NEPSTAD ET AL.2268 Ecology, Vol. 88, No. 9
Moreira, M. Z., L. Sternberg, L. A. Martinelli, R. L. Victoria,
E. M. Barbosa, L. C. M. Bonates, and D. C. Nepstad. 1997.
Contribution of transpiration to forest ambient vapor based
on isotopic measurements. Global Change Biology 3:439–
Myers, B. J. 1988. Water stress integral—a link between short-
term stress and long-term growth. Tree Physiology 4:315–
Nakagawa, M., et al. 2000. Impact of severe drought associated
with the 1997–1998 El Nino in a tropical forest in Sarawak.
Journal of Tropical Ecology 16:355–367.
Nascimento, H. E. M., and W. F. Laurance. 2002. Total
aboveground biomass in central Amazonian rainforests: a
landscape-scale study. Forest Ecology and Management 168:
Nepstad, D. C., C. J. R. d. Carvalho, E. A. Davidson, P. Jipp,
P. A. Lefebvre, G. H. d. Negreiros, E. D. da Silva, T. A.
Stone, S. E. Trumbore, and S. Vieira. 1994. The role of deep
roots in the hydrological and carbon cycles of Amazonian
forests and pastures. Nature 372:666–669.
Nepstad, D. C., G. O. Carvalho, A. C. Barros, A. Alencar, J. P.
Capobianco, J. Bishop, P. Moutinho, P. A. Lefebvre, U. L.
Silva, and E. Prins. 2001. Road paving, fire regime feedbacks,
and the future of Amazon forests. Forest Ecology and
Nepstad, D. C., P. Jipp, P. R. d. S. Moutinho, G. H. d.
Negreiros, and S. Vieira. 1995. Forest recovery following
pasture abandonment in Amazonia: canopy seasonality, fire
resistance and ants. Pages 333–349 in D. Rapport, editor.
Evaluating and monitoring the health of large-scale ecosys-
tems. Springer-Verlag, New York, New York, USA.
Nepstad, D., P. Lefebvre, U. L. Silva, Jr., J. Tomasella, P.
Schlesinger, L. Solorzano, P. Moutinho, and D. Ray. 2004.
Amazon drought and its implications for forest flammability
and tree growth: a basin-wide analysis. Global Change
Nepstad, D. C., et al. 2002. The effects of rainfall exclusion on
canopy processes and biogeochemistry of an Amazon forest.
Journal of Geophysical Research 107:1–18.
Nepstad, D. C., A. Verı´ssimo, A. Alencar, C. A. Nobre, E.
Lima, P. A. Lefebvre, P. Schlesinger, C. Potter, P. R. d. S.
Moutinho, E. Mendoza, M. A. Cochrane, and V. Brooks.
1999. Large-scale impoverishment of Amazonian forests by
logging and fire. Nature 398:505–508.
Oliveira, R. S., T. E. Dawson, S. S. O. Burgess, and D. C.
Nepstad. 2005. Hydraulic redistribution in three Amazonian
trees. Oecologia 145:354–363.
Oyama, M. D., and C. A. Nobre. 2003. A new climate-
vegetation equilibrium state for tropical South America.
Geophysical Research Letters 30:Article no. 2199.
Phillips, O. L., R. Va ´ squez Martı´nez, A. Monteagudo
Mendoza, T. Baker, and P. Nu ´ n ˜ ez-Vargas. 2005. Large
lianas as hyperdynamic elements of the tropical forest
canopy. Ecology 86:1250–1258.
Prentice, I. C., G. D. Farquhar, M. J. R. Fasham, M. L.
Goulden, and M. Heimann. 2001. The carbon cycle and
atmospheric carbon dioxide. Pages 183–237 in J. T. Hough-
ton, editor. Climate change 2001: the scientific basis.
Contribution of Working Group I to the Third Assessment
Report of the Intergovernmental Panel on Climate Change.
Cambridge University Press, Cambridge, UK.
Ray, D., D. C. Nepstad, and P. R. d. S. Moutinho. 2005.
Micrometeorological and canopy controls of flammability in
mature and disturbed forests in an east-central Amazon
landscape. Ecological Applications 15:1664–1678.
Restom, T. G., and D. C. Nepstad. 2004. Seedling growth
dynamics of a deeply rooting liana in a secondary forest in
eastern Amazonia. Forest Ecology and Management 190:
Ribeiro, J. E., M. J. G. Hopkins, A. Vincentini, A. Sothers,
J. M. de Brito, and M. A. D. de Souza. 1999. Flora da
Reserva Ducke: Guia de identificac ¸ a ˜ o das plantas vasculares
de uma floresta de terra-firme na Amazo ˆ nia central. INPA,
Rice, A. H., E. H. Pyle, S. R. Saleska, L. Hutyra, M. Palace, M.
Keller, P. B. de Carmago, K. Portilho, D. F. Marques, and S.
Wofsy. 2004. Carbon balance and vegetation dynamics in an
old-growth Amazonian forest. Ecological Applications
Saleska, S. R., et al. 2003. Carbon in Amazon forests:
unexpected seasonal fluxes and disturbance-induced losses.
Schnitzer, S. A., and F. Bongers. 2002. The ecology of lianas
and their role in forests. Trends in Ecology and Evolution 17:
Sheil, D., and R. M. May. 1996. Mortality and recruitment rate
evaluations in heterogeneous tropical forests. Journal of
Tian, H., J. M. Melillo, D. W. Kicklighter, A. D. McGuire,
J. V. Helfrich III, B. Moore III, and C. J. Vo ¨ ro ¨ smarty. 1998.
Effect of interannual climate variability on carbon storage in
Amazonian ecosystems. Nature 396:664–667.
Trenberth, K. E., and T. J. Hoar. 1997. El Nin ˜ o and climate
change. Geophysical Research Letters 24:3057–3060.
Trumbore, S. E., E. A. Davidson, P. B. d. Camargo, D. C.
Nepstad, and L. A. Martinelli. 1995. Below-ground cycling of
carbon in forests and pastures of Eastern Amazonia. Global
Biogeochemical Cycles 9:515–528.
Van Nieuwstadt, M. G. L., and D. Sheil. 2005. Drought, fire
and tree survival in a Borneo rain forest, East Kalimantan,
Indonesia. Journal of Ecology 93:191–201.
Williamson, G. B., W. F. Laurance, A. A. Oliveira, P.
Delamonica, C. Gascon, T. E. Lovejoy, and L. Pohl. 2000.
Amazonia tree mortality during the 1997 El Nin ˜ o drought.
Conservation Biology 14:1538–1542.
Woods, P. 1989. Effects of logging, drought and fire on
structure and composition of tropical forests in Sabah,
Malaysia. Biotropica 21:290–298.
Wright, S. J., and O. Calderon. 2006. Seasonal, El Nino and
longer term changes in flower and seed production in a moist
tropical forest. Ecology Letters 9:35–44.
Wright, S. J., C. Carrasco, O. Calderon, and S. Paton. 1999.
The El Nino Southern Oscillation, variable fruit production,
and famine in a tropical forest. Ecology 80:1632–1647.
Annual percentage mortality rates and 95% confidence intervals for paired and unpaired species in wet and dry plots, in the
Tapajo ´ s National Forest, east-central Brazilian Amazon (Ecological Archives E088-136-A1).
September 20072269 MORTALITY OF AMAZON TREES