You are on page 1of 68

www.sciencemag.

org/cgi/content/full/323/5919/1344/DC1

Supporting Online Material for

Drought Sensitivity of the Amazon Rainforest

Oliver L. Phillips,* Luiz E. O. C. Aragão, Simon L. Lewis, Joshua B. Fisher, Jon Lloyd,
Gabriela López-González, Yadvinder Malhi, Abel Monteagudo, Julie Peacock, Carlos A. Quesada,
Geertje van der Heijden, Samuel Almeida, Iêda Amaral, Luzmila Arroyo, Gerardo Aymard,
Tim R. Baker, Olaf Bánki, Lilian Blanc, Damien Bonal, Paulo Brando, Jerome Chave,
Átila Cristina Alves de Oliveira, Nallaret Dávila Cardozo, Claudia I. Czimczik, Ted R. Feldpausch,
Maria Aparecida Freitas, Emanuel Gloor, Niro Higuchi, Eliana Jiménez, Gareth Lloyd,
Patrick Meir, Casimiro Mendoza, Alexandra Morel, David A. Neill, Daniel Nepstad, Sandra Patiño,
Maria Cristina Peñuela, Adriana Prieto, Fredy Ramírez, Michael Schwarz, Javier Silva,
Marcos Silveira, Anne Sota Thomas, Hans ter Steege, Juliana Stropp, Rodolfo Vásquez,
Przemyslaw Zelazowski, Esteban Alvarez Dávila, Sandy Andelman, Ana Andrade, Kuo-Jung Chao,
Terry Erwin, Anthony Di Fiore, Eurídice Honorio C., Helen Keeling, Tim J. Killeen,
William F. Laurance, Antonio Peña Cruz, Nigel C. A. Pitman, Percy Núñez Vargas,
Hirma Ramírez-Angulo, Agustín Rudas, Rafael Salamão, Natalino Silva, John Terborgh,
Armando Torres-Lezama

*To whom correspondence should be addressed. E-mail: o.phillips@leeds.ac.uk

Published 6 March 2009, Science 323, 1344 (2009)


DOI: 10.1126/science.1164033

This PDF file includes:

Materials and Methods


SOM Text
Figs. S1 to S8
Tables S1 to S7
References
Amazon drought sensitivity

Supporting Online Material


Author Contributions

Co-authors are listed alphabetically in four groups:

* contribution to developing the project and a critical important contribution to analysis;


* major contribution to developing the project or a leading contribution to analysis and to new
fieldwork;
* mostly field / lab participants, with important contribution to 2005, 2006, 2007 fieldwork
and/or significant data acquisition;
* mostly field / lab participants: contribution to unpublished earlier work for essential
historical context for 2005.

OP wrote the paper;


OP, JL, YM conceived the idea;
OP, SA, IA, DB, JC, NH, JL, YM, DNei, DNep, MCP, CQ, MS, RV, HS, EA, SA, TB, TE, K-
JC, TK, WL, CM, PM, NP, AR, HS, RS, NS, and JT designed the experiments and sampling
design (=sampling design of multiple plots per site, pan-Amazon remeasurement sampling
design, wrote proposal);
SL, SA, IA, LAra, LArr, GA, OB, LB, DB, PB, TB, CC, JC, ACA, NDC, JE, MF, TF, JF, NH,
EJ, GL, AMon, AMor, DN, DN, JP, OP SP, MCP, CQ, AP, FR, MS, JS, MS, MS, HS, RV,
PZ, AA, EA, TE, AF, KJC, EH, HK, TK, CM, HN, APC, NP, PNV, RS, NS, GvdH, JT, AT
collected the field data (=tree measurements, tree collection, tree ID, soil collection, met data
collection);
LAra, SL, LB, PB, TB, JC, JF, EG, NH, GL, AMor, JP, OP, CQ, MSch, GvdH, PZ
performed the laboratory analysis (processed multi-site met data into consistent format,
RAINFOR plot and functional trait database design, development, operation; non-database
computation of interval growth rates, biomass, mortality rates; error modelling, soil analyses,
AWC computation);
LAra, SL, JF, OP, EG contributed materials or analysis tools (developed evapotranspiration
modeling, climate error modelling, sampling correction techniques, TRMM data processing
and forest area analysis);
OP, SL, GvdH analysed the data (did the analyses presented in this paper).

1
Amazon drought sensitivity

SOM Methods
Further information on data and analyses are provided here.

1. Collating biometric data


2. Computing forest growth and dynamic parameters
3. Sampling effects
4. Assembling the meteorological datasets
5. Sampling and analysis of soils
6. Calculating moisture balance
7. Defining and estimating drought
8. Comparing wood density of dead trees
9. Estimating large-scale impacts

1. Collating biometric data


Located throughout Amazonia, the RAINFOR plots span the full range of variation in
substrate, rainfall, seasonality, drought-vulnerability, biomass, productivity, turnover, and
species composition (S1-8).

Sampling
The mean size of our 136 plots is 1.7 ha (median 1.0 ha), monitored for an average 11.5 years
(Table S1 for full details). Plots are distributed in 44 distinct “sites” (landscapes which we
define by maximum inter-plot distance of ~5 km), and are periodically remeasured typically
every three to five years (Fig. S1). Immediately following the 2005 drought we initiated an
intensive eight-nation remeasurement program, visiting 72 plots distributed in 27 sites across
Amazonia which were affected by the drought to differing extents (c.f. SOM Methods Section
4, 7 for drought characterization). Most plots were remeasured in 2006, after the drought had
ended and at a similar time of year to previous sentences to minimise possible hydrostatic
effects of wood shrink-swell. 65 plots (20 sites) have at least three censuses, allowing us to
define discrete ‘2005’ and ‘pre-2005’ intervals. Wherever possible we selected intervals of ~1
year to span the 2005 drought period and capped this to a maximum of five years, resulting in
55 plots (19 sites) with separate pre-2005 and 2005 intervals, for which the ‘2005 interval’
itself averaged 1.97 years.

2
Amazon drought sensitivity

Measurements and biomass estimation


For each census we used internationally standardised approaches to fieldwork and quality
control (S3,S6,S8,S9). Within fixed-area plots, diameters of all free-standing stems ≥100 mm
were measured at 1.3 m or above stem deformities. We identified 79.4%, 92.6% and 94.7% of
stems to species, genus and family, respectively. We applied on-site rechecking procedures to
screen for potential measurement errors, and after field-work we also assessed trees that
increased in diameter >40 mm a-1 or shrunk >5 mm to determine if they may have been
inaccurately measured in the field. For cases judged potentially inaccurate - for example a
slow-growing species growing abnormally fast - the diameter was either interpolated or
extrapolated using known measurements from the same stem from other censuses, or, when
only one accurate measurement was available, by applying the median growth rate for the
same diameter size-class (100-199, 200-399, and 400+ mm diameter). Median growth rates,
averaging 1-3 mm a-1, are always lower than mean growth rates so this is a conservative
procedure with respect to the hypothesis that growth was high; overall it affects 0.9% of
Amazon tree measurements. Tree diameter measurements were converted to tree carbon
stocks using the only published allometric equation exclusively based on Amazon data that
also includes terms for wood specific gravity (WSG) (S10). We assume biomass is 50%
carbon (S11). A new WSG database was compiled (S3,S12,S13), with 5761 independent
neotropical values representing 2504 neotropical species and 717 neotropical genera. Within
our dataset 52.2% of stems have a wood density value at species level, 89.4% at the genus
level, and 94.2% at the family level; across Amazonia most variation in wood density values is
expressed at the level of genus (S3). Species in both the WSG and plot databases were
standardised for orthography and synonymy (S14) to maximise matches between the
databases. For species without corresponding WSG values, we used mean generic- or family-
level values, as appropriate. For incompletely identified stems we used the next higher known
taxonomic category or the mean WSG of the plot.

For a few sites where we lacked tree-by-tree data, we estimated AGB dynamics based on tree
basal area (BA) values. Thus, following Higuchi’s data based on destructive harvesting of 319
trees adjacent to the BIONTE plots from central Amazonia, the relationship between tree basal
area (m2 ha-1) and fresh above-ground biomass of trees >10 cm diameter (AGFB, t ha-1) has the
linear form AGFB = 66.92 + 16.85(BA), with r2=0.85 (S15) . The relationship was tested and
found to be appropriate at another site in eastern Amazonia (S16). Within our dataset we
applied the basal area allometric equation to assess AGB dynamics at five of the 136 plots

3
Amazon drought sensitivity

(ANN-03, ELI-01, ELI-02, SCR-01, SCR-03) (S17-20), the latest of these was measured in
1991 and therefore do not directly affect the 2005 analyses. Among these five plots mean
measured AGB change is +0.56 Mg ha-1 a-1, less than the pan-Amazon pre-2005 mean rate
suggesting that inclusion of these plots may be conservative with respect to the hypothesis of
increasing AGB prior to 2005.

2. Computing forest growth and dynamic parameters


All biomass growth and mortality values were first census-interval corrected, to help prevent
the possibility that length of the census-interval per se would unduly affect the estimated rates
of gain and loss terms. In our analysis census interval length varies substantially both within-
and among plots. This presents a problem because estimates of mortality and growth obtained
from censuses of non-homogeneous populations decrease with increasing census interval,
because progressively longer intervals result in failure to observe a progressively greater
fraction of the mortality and growth and that non-observed fraction is a non-random subset of
all mortality and growth (S21,S22). Therefore rate estimates based on longer intervals can
substantially underestimate the true values.

Specifically, the gross annual mortality of biomass at census t1 includes the measured
mortality of biomass of trees alive at census t0, as well as two additional unobserved
components: the mortality of the additional biomass grown by the initial cohort of trees alive
at t0, and the mortality of trees which recruited in after t0. These latter terms are unmeasured
and need to be estimated by applying a correction that scales with the length of the census-
interval and the estimated rate of mortality. Similarly, the gross annual growth of biomass
includes the measured growth of surviving trees, measured recruitment (biomass of trees
reaching 100mm diameter the period), and the measured growth of recruited trees, as well as
two additional components: the growth of the initial cohort of trees which then died, and the
biomass of trees which recruited in after t0 but died before t1. As with the mortality estimates
these last two terms are unmeasured. The unmeasured growth and mortality terms will tend to
converge for a mature forest, and so theoretically the magnitude of the correction slope will be
approximately proportional to the product of the stand growth rate and the rate of fractional
loss of biomass through mortality. Therefore the correction slope may vary approximately as
the square of the growth rate (S5). We thus corrected raw growth and mortality values by
applying a correction that scales with the length of the census-interval and the estimated rate
of mortality. For sites with two or more censuses, raw values were corrected iteratively

4
Amazon drought sensitivity

following the approach in (S5) (Appendix: p 590), except that we substituted biomass gain
rates for basal area gain rates and only used pre-2005 data for computing correction factors for
individual plots, to prevent any unusual forest behaviour associated with the drought from
skewing the correction factor for the whole period. For sites without two intervals pre-2005
we again followed (S5) in fitting a pan-Amazon correction slope function to stand growth rate
data for all the plots where site-specific correction slopes could be calculated directly, and then
computed interval-specific correction factors for each remaining plot by applying the best fit
quadratic function for stand growth rate.

3. Sampling effects
Estimation of carbon gains and losses from census data might be biased if the sample size (in
time or in space) is not sufficiently large to capture disturbance events. This potentially
important point has been made by several commentators and has been summarized by the
pithy phrase "slow in, rapid out" (S23). A case has been made that networks like RAINFOR’s
will be affected by the influence of local disturbance events, but the practical significance of
this problem for interpreting the pan-Amazon finding of long-term biomass gain has not been
clear until now.

A recent paper (S24) attempted to specify the number of plots required to sample the annual
change in forest biomass, assuming that the size-frequency distribution of mortality events
follows a power-law. If the scaling parameter of the power low, α , is >2.0 then in most cases
there is little bias, but the bias increases for smaller values of α especially below 1.5 (S24). To
assess what this should be we fitted a power-law function to various disturbance size-
frequency data from tropical forests, including large-scale mortality events in Amazonia based
on satellite images (S25) and various plot datasets from Africa (S26), using the maximum
likelihood estimation (S27). All indicate the correct scaling parameter to be between 2.1 and
3.1, suggesting that disturbances in tropical forests are heavily dominated by small and
frequent events and that any bias in our multi-annual estimates of changes in carbon stocks
must be small (S24). The power-law was fitted so model outputs from S24 could be
compared, but alternative long-tail distributions may provide a better fit (S27). As power laws
provide the ‘fattest’ tails of any generalised extreme value distribution, fitting other
distributions would suggest that large disturbances are of even less consequence. We
therefore do not discuss the issue further here.

5
Amazon drought sensitivity

A different but related issue is whether a bias toward heavy sampling in one or a few localities
could be responsible for our results, or whether the patterns we find are in fact widespread.
Thus, are the pre-2005 net increases recorded only in plots in a few sites, or across Amazonia?
Are the 2005 declines restricted to one part of Amazonia or many? Fig. 3 suggests they are
indeed widespread but we explore these possibilities further in SOM Results by examining the
extent to which results are consistent when analysed at different sample unit scales.

Finally, it is important to acknowledge that our sample plots and sites vary in both total area
monitored and length of time monitored for, and that this uneven sampling among monitoring
locations could affect our estimates of biomass change. We therefore sought to weigh each
plot’s (and site’s) contribution by the monitoring effort (i.e., area and length of monitoring
period), and then create modified distributions of the census-interval corrected ΔAGB
distributions based on these weighted values. For the whole dataset and for the pre-2005 plot
dataset, optimum weightings were derived empirically by assessing patterns in the residuals of
sampling effort vs. biomass change for a series of different weightings. Weighting by the
square root of the number of years of monitoring, and weighting by the cube root of plot size
both independently best remove any pattern in the residuals, suggesting that these relationships
best describe the relationships between census length or plot size and sampling error
respectively (Fig. S2). The final weighting of the square root of the number of years of
monitoring plus cube root of plot size minus one (to avoid double-accounting) shows no
pattern in the residuals (Fig. S3). For the 2005 dataset we did not weight by area or by time.
Plots were all ~1 ha, and since we were testing for the possible impacts of a time-bounded
event shorter than the measurement intervals themselves, it would be inappropriate to allocate
greater weight to longer, more poorly-resolved intervals. For all plots monitored with at least
one interval prior to 2005 and at least one other through 2005, we weighted the 2005 – pre-
2005 difference between biomass gain rates by the square root of the pre-2005 interval. All
equivalent datasets for sites were weighted as described above for plots, except that we used
the square root of the total interval length and cube root of the total area, and for the 2005
intervals we weighted by area, as sites were comprised of varying numbers of plots and
therefore varying sample area. The weighting by time and area had subtle but small effects on
our estimates of change prior to, and during, the 2005 drought. These are shown numerically
in Table S3.

6
Amazon drought sensitivity

4. Assembling the meteorological datasets


The 2005 drought was driven mainly by elevated Atlantic sea surface temperatures (SSTs)
which affect dry season rainfall in southern Amazonia (S28), in contrast to Pacific SST-
induced droughts, which affect more the wet season rainfall. Thus in 2005 total annual
rainfall reductions were not generally severe; rather the dry season was strongly intensified
over a large area. The 2005 drought also tended to most affect the area with least inter-annual
variability (western and southern Amazon) (S29) (Fig. 3).

Daily radiation, relative humidity, maximum and average temperatures, and precipitation were
collected from multiple sources for the period 1996-2006. The time period was chosen in order
to capture the meteorological conditions during both the 1997-98 and 2005 drought years
across the Amazon and to match closely with available remotely sensed meteorological data
and biomass dynamics data.

Three sources of meteorological data were used for this analysis in order of priority: 1) data
collected adjacent to the plots; 2) data collected from the closest meteorological station(s)
within 50 km of the plot (accessed via Tutiempo, www.tutiempo.net, and Hidroweb,
http://hidroweb.ana.gov.br/); 3) precipitation data measured by the Tropical Rainfall
Measuring Mission (TRMM, 3B43 version 6) (S30). TRMM monthly mean precipitation
(mm/hour) was provided from 1998-2006 at 0.25° spatial resolution.

The methodology for data generation is detailed below.

4.1 Precipitation. For each site we sought an appropriate, distinct precipitation data set.
Meteorological data from a nearby station were accessed through either Tutiempo or
Hidroweb, accounting for ~67% of the daily rainfall values we used. Where neither Tutiempo
nor Hidroweb data were available we used TRMM data. For the MCWD analysis in the main
text we simply used the local TRMM monthly totals when rain-gauge data were not available,
to allow us to model the impact of uncertainty in rain-gauge data for defined months. TRMM
data represent the highest source of uncertainty due to the coarse resolution. Impacts of
TRMM-associated errors on moisture balance estimates are assessed below (section 6).

7
Amazon drought sensitivity

For the more complex moisture balance analysis in the SOM we needed daily rainfall
estimates, so to fill occasional missing periods an appropriate offset for monthly TRMM totals
was extracted for each field site and each meteorological station and compared with running
thirty day totals for each site. When nearby meteorological stations were lacking, a multiplier
was generated using the TRMM offset between the field site and closest met station by:

Multiplier = Absolute value [1 + (precipitation_TRMM_field_site –


precipitation_TRMM_met_station) / (precipitation_TRMM_met_station)]

In order to preserve the quasi-stochastic nature of the precipitation data the multiplier was
applied to generate daily data for the field site from the daily precipitation data of the
meteorological station. In cases where there were still gaps in precipitation data coinciding
with similarly spaced gaps in nearby stations, daily means were taken over the ten year span
and used to infill these gaps. This last method was applied to ~4% of daily values.

Quality check of all precipitation datasets was performed by comparison of monthly totals and
TRMM values for each site. A simple rule of thumb was applied – if there was a 100 mm or a
100% difference between the two, data were corrected. For the months where the precipitation
data fit this requirement a multiplier was generated (as above) to correct the discrepancy.

4.2 Relative humidity and temperature. With some exceptions, Tutiempo and field site
meteorological station datasets were complete, barring a few missing days. For infilling data
gaps of greater than 10 days we averaged the relative humidity or temperature readings for
each day of the year over the ten year period. For gaps of ten days or less, we infilled by
taking the average values from three days before and three days after the missing day(s).

4.3 Radiation. Daily net radiation (Rn) was used as an input for modelling evapotranspiration
(section 6 of the SOM). To generate site-specific time series of Rn (January 1996 to December
2006) we integrated two independent sources of data, the Clouds and the Earth’s Radiant
Energy System (CERES) mission (S31) and the Brazilian Institute for Space Research/Center
for Climate Forecast and Climatic Studies (INPE/CPTEC) GL 1.2 V2 total incoming radiation
data (S32).

8
Amazon drought sensitivity

Daily net radiation was first modelled from CERES mission data, by using the Terra dataset
from the Monthly Gridded Radiative Fluxes and Clouds (FSW) product. Data were obtained
from the NASA Langley Research Center Atmospheric Science Data Center. Daily CERES
net radiation data were derived from satellite overpasses with a sine wave model to represent
the diurnal variation of net radiation. Instantaneous radiation fluxes between 10 and 11am
local time were extracted at one degree spatial resolution for 1 March 2000 to 31 December
2005. 17% of values were spurious or missing from the product (ranging from 7% for BEE to
29% for ZAR), with gaps were spread evenly throughout the dataset, except for 13 days in
August 2000 and the whole of June 2001. The gaps were replaced with means of adjacent
correct measurements. The gap-filled dataset was then converted to daily average net radiation
(DANR) based on a sinusoidal model (S33). Rn values used in the final analysis represent 24
hourly averages. To validate the above method, we used our ground Rn measurements from
CAX (339 days between 1 January and 7 December 2005) and TAM (166 days between 12
March and 11 December 2005), and global solar radiation data collected by Harvard
University from Santarém (1352 days between 30 June 2000 and 12 March 2004). The r2
between those datasets and modelled Rn was 56%, 68%, and 37%, respectively.

The 5-years CERES time series was subsequently used to generate Rn values for the full 11-
years total incoming radiation (Rt) data from INPE/CPTEC. The INPE/CPTEC Rt data is
available at 0.4 degree spatial resolution, estimated from the Geostationary Satellite
information (GOES-8 and GOES-12) using the “Global Radiation” model GL 1.2 version 2
(29). This dataset encompasses five-day averaged Rt data from January 1996 to December
2006 (total of 803 surfaces) for the Amazon region bounded by 10.0oN, 75.2oW and 15.0oS,
44.8oW. To generate site-specific relationships between the CERES Rn and the INPE/CPTEC
Rt data, we first aggregated the daily CERES data into a 5-day averaged data, similar to the
INPE/CPTEC time series. Using the overlapping period between the two datasets we then
estimated Rn for the full 11 years from the 5 year statistical relationship between CERES net
radiation and INPE/CPTEC total incoming radiation. This procedure was based on the fact
that Rn is a direct linear function of Rt (S34). Correlation coefficients for all site-specific
relationships were highly significant (p<0.01).

We expect uncertainty in our final estimates due to methodological limitations and the
uncertainty embedded in each product. Nevertheless this should not invalidate our inter-annual
and inter-site analyses as the data were exposed to the same systematic errors for all years and

9
Amazon drought sensitivity

sites. The different spatial and temporal resolutions of the CERES and INPE/CPTEC datasets
are likely to be responsible for the scatter observed in the regressions. The sine wave model
applied to derive the CERES Rn data from the single satellite overpass may overestimate daily
net radiation as clouds in the Amazon region normally build from the late morning. Estimates
produced by the model GL 1.2 have a mean bias of ±10 Wm-2. However, for some localities
Rt may be overestimated by up to 50 Wm-2 due to the impossibility of representing aerosol
impact on radiation during the Amazonian fire season. This effect is likely to be reduced by
the integration with the CERES data, which explicitly consider aerosol in their calculation.

5. Soil sampling and analysis


Soil sampling for each RAINFOR plot was carried out using a standard protocol, and by
applying stratified random sampling to account for landscape features that could affect soil
variability. Thus, for homogeneous plots with flat topography five soil samples were taken to
2 m depth and a soil pit 1.5 x 1.5 m wide was dug to 2 m and subsequently sampled by auger
to a total depth of 4 m. In areas with greater spatial variability the number of samples was
increased accordingly, up to 12 cores per 1-ha plot. Pits were situated at locations
representative of the dominant soil and topographic position(s), just outside the plots.

Within each core, samples were collected using an undisturbed soil sampler (Eijkelkamp
Agrisearch Equipment BV, the Netherlands) and all soil retained for the following depths: 0-
0.05, 0.05-0.10, 0.10-0.20, 0.20-0.30, 0.30-0.50, 0.50-1.00, 1.00-1.50 and 1.50-2.00m. In soil
pits samples were collected from pit walls. Samples were air-dried as soon as practical with
roots, detritus, small rocks removed, and then sieved at 2mm in the laboratory. Soil
descriptions and an initial in situ soil classification (Reference Soil Group) were done by one
person (CQ) following standard protocols (S35) paying attention to: depth, depth to C horizon,
horizon distribution, color, rocks, presence/absence of concretions, ironstones or hardpans,
redox-features, root amount and distribution, drainage capacity, soil hardness and structure,
aggregate type and stability, organic matter content, faunal activity, particle size, easily
weatherable minerals, soil porosity, and topography. Three bulk density samples were
collected per depth, using container-rings of known volume (Eijkelkamp Agrisearch
Equipment), oven-dried at 105ºC to constant weight, and cooled to room temperature in sealed
desiccant before final weight determination. Particle size analysis was subsequently done by
the Boyoucos method cross-checked by the pipette method (S36). Following laboratory
analyses soils were classified using the recent World Reference Base for Soil Resources (S37).

10
Amazon drought sensitivity

Available Water Capacity (AWC) was only used for some soil moisture balance calculations
(section 6 below); estimates are available on request from the corresponding author. We used a
tropical soil pedotransfer function (S38) to derive AWC. For each depth soil texture was used
to estimate AWC, which was then integrated over the whole rooting profile. Because tropical
forest rooting depth is notoriously difficult to determine precisely, where the maximum
observed rooting depth and maximum sampling depth coincided we needed to make
assumptions about unobserved rooting. We adopted these rules:
* maximum rooting depth was assumed to be 4m, unless substantial physical impediments to
root growth were observed;
* arenosols and ferralsols were assumed to have roots to 4m;
* cambisols, fluvisols, plinthosols, and stagnosols were assumed have roots to 1m, or where
actually observed, whichever was the greatest;
* gleysols and saturated podzols have rooting only to where actually observed, as prolonged
anaerobosis makes deeper roots unlikely.

Where rooting was known or assumed to penetrate below where observed, AWC was
estimated for unobserved depths on basis of texture of the deepest observed soil. For plots
with more than one soil class, an area-weighted mean was an appropriate whole plot estimate.
We had sufficient plot-level particle–size information to directly derive AWC for 46 plots; for
the remaining 33 plots we estimated AWC according to these rules:
* if other regional plots had the same soil class we applied the mean of those plots (19 plots);
* if the soil was identified to a soil class without a locally-measured value, we applied the
mean measured AWC for Amazon soils of that class (7 plots)
* for remaining plots with uncertain soil class, some classes could nevertheless be excluded so
we applied the mean pan-Amazon value of the candidate soil class(es) (7 plots).

6. Calculating moisture balance


We used two different bucket-model approaches to derive water balance for our forests. The
first approach accounts for variation in monthly rainfall alone. The second accounts for
estimated variation in daily rainfall, storage capacity (soils), and daily loss rates
(evapotranspiration), but in so doing incorporates additional uncertainty. We therefore report
results using both approaches and various derived statistics thereof (see Methods 7 and 9,

11
Amazon drought sensitivity

below, and Results Figs. S5-8 and Table S5). For simplicity in the main text of the paper we
only report results from the first approach.

Our first approach is based simply on rainfall inputs and assumes a mean loss rate via
evapotranspiration (~3.33 mm day-1), based on empirical measurements and modelled
estimates showing a moist Amazon canopy transpires about 100 mm each month (S39-41).
This approach also has widespread precedent (S1, S29, S42). We estimated the maximum
cumulative water deficit (MCWD) in each year, cumulating such deficits on a monthly basis
(S17) in order to scale to the whole Amazon, as this parameter can be directly estimated from
TRMM data (c.f. SOM Methods 10, below) and this allows us to explore the impact of
TRMM-associated errors on our relative drought metric (c.f. final section of SOM Methods 6).

We also employed a moisture balance approach to estimate soil moisture content (SMC) based
on daily precipitation (PPT), estimated water holding capacity (AWC), and estimated
evapotranspiration (ET), with ET modified by soil moisture supply at low levels of SMC. The
bucket model had four constraints:
1) If PPT-ET+SMC0>AWC, then SMC=AWC
2) If PPT-ET+SMC0<AWC, then SMC=0
3) If PPT-ET+SMC0>0, then SMC=PPT-ET+SMC0
4) If SMC<0.4*AWC, then ETmodified = ETunmodified * SMC/AWC
The first constraint limits the maximum SMC to the size of the AWC. The second eliminates
negative water storage. The third adds water to the SMC in the previous time step (SMC0)
within the bounds of the first two constraints. The final constraint, involving a “ramp
function”, allows for a transition from being primarily radiation-limited to primarily water-
limited evaporation (S43,S44) as SWC/AWC falls below 0.4. Evapotranspiration is
independent of soil moisture until a threshold below which point it declines rapidly (S44,S45).
Lacking direct values for Amazonia, we have taken the 0.4 threshold value from our analysis
of data from West African tropical forest (S45), as suggested for other woody ecosystems
(e.g., S46).

Precipitation
Precipitation data were derived as described in section 4.1.

Soil water holding capacity

12
Amazon drought sensitivity

Soil water holding capacity was estimated for each site, as described in section 5.

Evapotranspiration
Evapotranspiration was modelled after S47. This model demonstrates robust validity at 16
global FLUXNET sites (r2=0.90, slope=0.96 for predicted vs. observed evapotranspiration),
and is relatively easy to parameterize. The model is validated at 21 pan-tropical eddy
covariance sites (r2=0.91, slope=0.92) (S41) so is appropriate for our pan-Amazon study. The
model requires data on relative humidity, air temperature and net radiation. We used known
Leaf Area Index to partition net radiation to the canopy versus soil. The model calculates five
constraints to the Priestley-Taylor (1972) equation (S43) - relative surface wetness, green
canopy fraction, plant temperature constraint, plant moisture constraint and soil moisture
constraint, and estimates the fraction of PAR absorbed or intercepted by green (fAPAR) or
total (fIPAR) vegetation cover from greenness vegetation indices (i.e., NDVI). These
estimates are used to calculate the green canopy fraction and a plant moisture constraint.
Because our forests are tropical evergreen, NDVI was consistently high (~0.80) and the green
canopy fraction is ~80%. The plant moisture constraint (fAPAR/ fAPARmax) was a non-
factor because there was no known dry-season decline in NDVI for the RAINFOR plots
(remotely-sensed data suggest a green-up during 2005 (S48)). Therefore ET was ultimately
radiation-, humidity- and temperature-based. Any effects of reduced canopy conductance
during extreme drought are implicitly reflected in the soil moisture constraint function.

Error assessment for moisture balance estimates


To assess the errors in our moisture deficit estimates, we focussed on our first approach to
modeling moisture balance in terms of MCWD. We computed the errors for rainfall estimated
by monthly TRMM by calibrating TRMM against monthly rain-gauge measures in Amazonia
(data from S25 and 10 additional locations near our plot clusters). To generate random samples
of precipitation and in turn MCWD records, we made random draws of the empirical
distribution of TRMM minus rain gauge data residuals, via the inverse of the cumulative
distribution function applied to uniformly distributed random numbers. We then propagated
the uncertainties through to assess their impact on estimated annual MCWD at each site, and
in turn the uncertainty in the difference between mean annual MCWD in the 2005 interval and
mean annual MCWD in the pre-2005 interval (“deltaMCWD”). We reconstructed the
deltaAGB vs. deltaMCWD relationship (Fig. 2), bootstrapping in R to account first for the
climate measurement errors and then for the uncertainty in the deltaAGB vs. deltaMCWD

13
Amazon drought sensitivity

relationship. Bootstrapping of deltaMCWD values was done 1000 times, with each dataset
then bootstrapped 100 times to account for uncertainty in the delta AGB and delta MCWD
relationship, resulting in 100,000 bootstrap samples on which the 95% confidence interval is
based.

7. Defining and estimating drought


Selecting an appropriate index of tropical forest droughting is challenging. Previous studies
have used, among others, the Palmer Drought Index (PDI), and the Standard Precipitation
Index (SPI). As recently pointed out (S49), the PDI is parameterized based on data from a few
stations in North America and is heavily influenced by 12-month cumulative rainfall deficits -
rather than the dry season anomalies which are expected to have greater impact on tropical
vegetation. The SPI, by contrast, can be used to represent precipitation deficit at the defined
time scale of interest (S50). Nevertheless, the census intervals of our forest dynamic dataset
vary from site to site, since it was not possible to measure forest dynamics simultaneously at
dozens of locations across Amazonia. Application of the SPI over a specified time-scale is
therefore incompatible with our aim of relating drought impacts to forest dynamics recorded
by site-specific monitoring intervals and therefore we sought more flexible indices suited to
our purpose.

Amazonia covers a wide climatic range from >4,000 mm in parts of the aseasonal northwest to
~1,400 mm annual rainfall on the southern fringes, with a similar but much more spatially
heterogeneous range in soil moisture capacity. We hypothesized that ecological impacts of
drought would be experienced by plants as a function of departure from their long-term mean
local environmental conditions, leading us to estimated droughting experienced during the
2005 interval relative to local, long-term mean values. Elsewhere (Fig. S6) we present results
based on absolute droughting metrics, which confirm that these are largely devoid of
predictive value.

Relative drought was estimated by deriving moisture availability on both daily and monthly
bases for the 2005 interval itself, and then for the entire pre-2005 interval. The 2005 interval
was usually shorter than the previous entire measurement period (mean 1.97 vs. 10.54 years),
so to avoid census-interval effects we expressed moisture deficits in each interval as annual
means before computing the difference in moisture availability between intervals. We
hypothesized that the impacts of moisture deficits would be determined by the maximum

14
Amazon drought sensitivity

depletion of moisture. Thus in this paper, unless noted otherwise, we define the drought
experienced by each plot as the mean maximum cumulative water deficit (MCWD) reached in
2005-interval year(s), relative to the mean MCWD reached in all pre-2005 interval year(s).
Because of the complexities in estimating moisture balance and plant-relevant drought indices
and the difficulty of accounting for all possible sources of error we also explored other metrics
of relative drought (see Fig. 2, and Figs. S5,S7,S8 and Table S5).

To avoid compromising the reliability of attribution of any change in biomass dynamics to


climate conditions in the period we did not attempt to derive drought parameters for sites for
which for which the interval spanning 2005 was >5 years (ACU, BOG, FRM, HCC, JAS,
LFB, PIB, TIP). For all sites with suitable data we computed the 2005 interval rates relative to
pre-2005 interval rates, and plotted them as a function of the 2005 interval metereology
relative to the pre-2005 interval meteorology. For some analyses (c.f. Table S6C,D and section
9 below) it was also necessary to define simply which individual plots and sites were
“droughted” in 2005, and which were not. We did this using our simple metric of relative
droughting intensity, comparing the mean MCWD reached in 2005-interval year(s) with that
reached in pre-2005 interval years. There were a number of marginal cases – for example for
12 plots the mean MCWD for the two intervals were within 10 mm of one another. We
therefore required that for a plot to be classed as “droughted” in 2005, both the intensity and
the length of the dry period in the 2005 interval must have surpassed the long-term local
mean.

8. Comparing wood density of dead trees


Wood specific gravity of dead trees was estimated as described above for live trees (c.f.
S.O.M. Methods 1). To estimate summary statistics of wood density of dead trees we
evaluated the mean wood density of trees that died, separately for the 2005 interval and for the
pre-2005 interval(s). To limit the impact of possible sampling errors and site-switching we
only included those plots which had >5 individual deaths in the 2005 census, and in all
previous censuses, and with a 2005 interval <5 years. We compared the mean wood density in
the 2005 interval with that of previous intervals for all plots, and separately for those classed
as “droughted” and “not-droughted” in 2005.

9. Estimating large-scale impacts

15
Amazon drought sensitivity

We use two different approaches to scale up from our field studies to the whole droughted
area. The first scales from our statistical distribution of plot-level (and site-level) data from
“droughted” locations, and requires the simplifying assumption that these are representative of
the wider droughted area. The second combines the demonstrated relationships between
biomass change and droughting with estimated rainfall. This requires making several
assumptions, notably that the effects of soil conditions on plant physiology are not confounded
with spatial patterns in climate and that any errors in the remotely-sensed data from TRMM
(Tropical Rainfall Measuring Mission) lack sufficient spatial or temporal structure to confound
the extrapolations. TRMM data are well-calibrated for Amazonia, being validated against
ground measurements (S29). Monthly rainfall values are well represented but may
overestimate rainfall lower than 50 mm month-1 and underestimate rainfall higher than 300
mm month-1. As our analysis is based on drought patterns (low intensity rainfall), and 2005
was characterised by intensification of the dry season, this limitation may tend to make our
scaling-up conservative with respect to the hypothesis of a large impact of the 2005 drought
(i.e., TRMM may slightly underestimate the intensity to which the 2005 drought departed
from long-term average conditions). TRMM data are only available in a readily-useable
format on a calendar-monthly basis.

We present results for both scaling-up approaches, and apply them to both the plot-based and
the site-based estimates of the drought sensitivity, to explore the how sensitive the result is to
varying the different assumptions involved in each analysis.

9.1 Scaling from the statistical distribution of biomass change data


All droughted plots for which we can compute AGB change in the 2005 interval and in the
pre-2005 interval were used. The whole period and pre-2005 values were weighted by
sampling effort (Figs. S2,S3), while the 2005 interval distribution was not weighted (cf. Table
S3). We then (1) scaled the mean, 2.5%, and 97.5% bootstrapped confidence intervals by the
droughted Amazon forest area estimated from TRMM calendar months <100mm relative to
pre-2005 (~3,304,092 km2), (2) multiplied by 1.97, since the mean 2005-interval spanned 1.97
years, (3) estimated the impact on unmeasured AGB (the estimated ratio of trees <10cm dbh
and lianas >1cm dbh to trees >10cm dbh in Amazonia is 0.099 (S15), (4) estimated the impact
on below-ground biomass (ratio of BGB:AGB = 0.37, (S51)), (5) summed for all biomass, and
(6) assumed all biomass to be 50% carbon following convention (S9). All analyses were
repeated at the site level.

16
Amazon drought sensitivity

9.2 Applying the biomass dynamics vs. climate relationship


For each plot monitored prior to and through 2005, we subtracted the rate of net change in
AGB measured in the pre-2005 interval from the rate of net change in AGB measured in the
2005 interval. These values were used in a weighted regression on the 2005 – pre-2005
difference in drought parameters, with each plot’s change in net biomass gain rates weighted
by its pre-2005 sampling effort (cf. Table S2). Note that as we are estimating plot-by-plot
differences in biomass change rates prior to and through the drought period, the weighting for
time and area needs to be applied after calculating the 2005-pre-2005 difference value, not
before. All analyses were repeated at the site level.

To scale to the wider region we used a climate parameter which can be directly estimated from
TRMM data to capture drought intensity (MCWD – maximum climatological water deficit,
i.e., the annual maximum cumulated deficit, with deficits accumulating for each month for
which rainfall was less than 100 mm (S17)). As discussed above, the threshold of 100mm
month-1, or 3.33mm day-1, was adopted because this equates to the mean evapotranspiration of
Amazonian canopies obtained by ground measurements in different locations and seasons in
Amazonia (S39-41). Hence, when monthly rainfall is less than 100 mm the forest will
normally enter into water deficit. We averaged MCWD values across the year(s) of each
interval in order to eliminate census interval effects when comparing variable length intervals,
since longer intervals are, by chance alone, more likely to have dry events. We then (1) used
TRMM data to estimate monthly precipitation variables for 0.25 cell pixel since 1998; (2)
predicted the change in AGB change rates for each 2005-droughted pixel, correcting each
pixel’s area by its 2006 proportion of forest based on the vegetation map of South America
from Global Land Cover 2000 Project (S52), updated for Brazil using the deforestation data up
to 2006 obtained from INPE’s Assessment of Deforestation in Brazilian Amazonia project
(PRODES) (http://www.inpe.br/). To carry out the first two steps, we extracted the values of
forest cover fraction and the annual values of MCWD from 1998 to 2006 for the 8,644
individual pixels at 0.25 spatial resolution (total area of 6,730,940 km2) within the
geographical boundaries of the Amazon region (S53). After the forest area correction the total
area decreased to 5,278,747 km2. Subsequently we (3) corrected for the small additional
fraction of Amazonia that had unreliable TRMM data during the period 1998-2005 (259,207
km2). The other computational procedures were applied as before, thus: (4) multiplying by
1.97, (5) estimating the impact on the unmeasured fraction of AGB, (6) estimating the impact

17
Amazon drought sensitivity

on below-ground biomass, (7) summing for all biomass, and (8) assuming all biomass to be
50% carbon. Estimated mean and bootstrapped confidence intervals for aggregate impacts are
based on the equivalent summed predicted impacts for each pixel. Some pixels were
droughted more severely than any of our plots (maximum estimated MCWD 2005 - previous
years = 262 mm, maximum measured in the field = 118 mm). We did not attempt to
extrapolate our model beyond the limits of measurement and so to these cells simply applied
the maximum AGB impact estimate within our measured range. Again, all analyses were
repeated at the level of the site sample unit.

18
Amazon drought sensitivity

Supporting Online Material, Results

1. Sampling effects
2. Sensitivity to drought modeling approach
3. Wood density of dead trees
4. Large-scale impacts

1. Sampling Effects
As described above (SOM Methods 3) we asked whether our pre-2005 results could be
explained by unusually heavy sampling intensity in a few landscapes that happened to be
gaining biomass, and conversely whether our 2005 results could be explained by unusually
heavy sampling intensity in a few landscapes that happened to lose biomass. For both plot and
site sample units we then quantified the possible contribution of sampling effects and
accounted for them in our results.

For practical reasons the pan-Amazon sample is evidently non-randomly distributed.

For pre-2005 data might this spatial bias might be driving the result – either by sampling
unusually heavily those regions that happened to be gaining biomass, and under-sampling
those that happened to lose biomass? This appears highly unlikely, since the long-term mean
net gain is essentially invariant with varying scale at which sample units are defined. Thus:
• of 123 pre-2005 plots 76% showed increases, mean per hectare net annual AGB
increase = 0.93 (0.62, 1.25) Mg (Table S3);
• of 41 “sites” (landscapes with ~5 km radii) 86% showed increases, mean per hectare
net annual AGB increase = 1.00 (0.61, 1.36) Mg (Table S3);
• of 11 “regions” (plot clusters with ~200 km radii spanning coherent geomorphological
and climatic units) all 11 increased, with mean per hectare net annual AGB increase in
AGB = 0.95 (0.51, 1.41) Mg.

Similarly, the shutdown in the biomass carbon sink around 2005 is not driven by a few
landscapes that happened to be anomalously disturbed: the annual net biomass change across
55 plots monitored in that period (-0.71 (-1.93, +0.30 Mg ha-1) is identical to that across 19
sites monitored in that period (-0.71 (-1.94, +0.35) Mg ha-1).

19
Amazon drought sensitivity

At the plot level we also attempted to take into account the sampling effort made in each
individual plot, since plots vary both in size and in length of pre-2005 monitoring, by
weighting the contribution of each plot according to the monitoring effort made. This slightly
weakened the magnitude of some results and consistently tightened the confidence intervals.
Thus (cf. Table S3):

• of 123 pre-2005 plots, mean annual AGB increase = 0.93 (0.62, 1.25) Mg ha-1 without
weighting, and 0.89 (0.65, 1.12) Mg ha-1 with weighting.
• of 41 pre-2005 sites, mean annual AGB increase = 1.00 (0.61, 1.36) Mg ha-1 without
weighting, and 0.94 (0.63, 1.27) Mg ha-1 with weighting.
• of 55 plots monitored through 2005, mean annual AGB increase = - 0.71 (-1.93, +0.30)
Mg ha-1 without weighting, and - 0.59 (- 1.66, +0.35) Mg ha-1 with weighting.
• of 19 sites monitored through 2005, mean annual AGB increase = - 0.71 (-1.94, +0.25)
Mg ha-1 without weighting, and - 0.74 (- 1.79, +0.28) Mg ha-1 with weighting.

The negligible impact of varying the spatial scale of analysis suggests that the Amazon plot
network provides a reasonably unbiased sample of forest behaviour. The small impact of
weighting by time and area suggests nevertheless that it is important not to treat all plots
equally – more precise understanding is gained with more monitoring effort.

2. Sensitivity to drought modeling approach


Modelling the hydrologic regime is error-prone, and it is difficult to account for these errors,
especially for such a vast and remote region as Amazonia. We therefore used a wide variety
of drought metrics, and repeated analyses of drought impacts on biomass dynamics at the site
level as well as the plot level.

For the monthly deltaMCWD drought modeling we took account of the TRMM-related errors
in rainfall assessment. Results of this simple drought modeling approach (Figs. 2, S8A) were
similar to those based on a more sophisticated approach that accounted for evapotranspiration
and soil water holding-capacity, and ran with a daily time-step (Figs. S5,S7B,C). This
suggests that the relationship with biomass dynamics is robust regardless of the detail of the

20
Amazon drought sensitivity

drought modeling approach, and that it is not unduly affected by poor representation of
evapotranspiration or soil water-holding capacity (also cf. Table S7).

3. Wood density of dead trees


Across all plots the mean wood density of dead trees in 2005 (0.58) was less than that of trees
that died in previous intervals (0.60) (Wilcoxon test one-tailed, n = 40, P=0.02). Among the
plots that were droughted in 2005, the mean wood density of dead trees in 2005 (0.57)
declined compared to previous intervals (0.60) (Wilcoxon test one-tailed, n = 25, P=0.02), but
among plots not droughted in 2005, the mean wood density of dead trees in 2005 (0.60) was
no different to that in previous intervals (0.61) (Wilcoxon test one-tailed, n = 15, P=0.84).
The drought was therefore associated with a detectable change in the kind of trees which died,
leading us to conclude that repeated droughts like that of 2005 would be capable of driving
shifts in tree biodiversity across Amazonia.

4. Large-scale impacts
The estimated impact of the drought on biomass carbon, as compared to the pre-drought
measurement period, varies between -1.21 (-2.01, -0.57) Pg C and -1.60 Pg C (-2.63, -0.83)
(Table S6C, E). The first value is based on simply scaling the per-plot impact by the total
droughted area; the second, slightly greater value is based on using remotely-sensed rainfall
data to scale from the relationship of biomass change data with relative drought intensity.
Site-level scaling-up yields similar and intermediate estimates, -1.57 (-2.40, -0.78) Pg C for
scaling the per-site impact by the total droughted area, and -1.47 (-2.61, -0.79) Pg C using
remotely-sensed rainfall data to scale from the biomass change / drought intensity relationship
(Table S6D, F).

The biomass dynamics/ climate approach makes use of more of the information on forest
response to drought than the simply scaling-up of the mean droughted plot and site impact, but
requires making additional assumptions which may introduce error and possibly bias. Neither
approach is flawless but the consistency of the results indicates a significant regional impact.
By comparison, these values greatly exceed the estimated range of annual C emissions due to
deforestation across the neotropics (0.3-0.8 Pg C) (55). They also outweigh estimates of the
long-term pre-2005 annual net carbon sink in undisturbed Amazon (0.40-0.42, (0.29, 0.57) Pg)
and neotropical forests (0.61-0.64 (0.44, 0.86) Pg), derived from measurements in the same
inventory plots and sites (Table S6A, B). The 2005 drought evidently not only shut down the

21
Amazon drought sensitivity

long-term sink but apparently reversed it over a large area. It has caused substantial carbon
emissions from Amazon forest, whether in 2005 itself and/or committed for following year(s).
Co-ordinated monitoring with a distributed network of plots is required to determine whether
this event represents a one-off hit for Amazon carbon stocks from which a full recovery is now
made, or the start of a longer-term climate-induced phase shift in which old-growth Amazonia
becomes carbon-neutral or a carbon-source.

References
S1. W. G. Sombroek, Amazon Soils, 303 pp., Cent. for Agric. Publ. and Doc.,
Wageningen, Netherlands (1966).
S2. W. G. Sombroek, Acta Amazonica 30, 81-100 (2000).
S3. T. R. Baker, O. L. Phillips, Y. Malhi, S. Almeida, L. Arroyo et al., Global Ch. Biol. 10,
545-562 (2004).
S4. Y. Malhi, J. Wright. Phil. Trans. Roy. Soc., Ser. B 359, 311-29 (2004).
S5. Y. Malhi, T. R. Baker, O. L. Phillips, S. Almeida, E. Alvarez et al., Global Ch. Biol.
10, 563-591 (2004).
S6. O. L. Phillips, T. R. Baker, L. Arroyo, N. Higuchi, T. Killeen et al., Philos. Trans. R.
Soc. London, Ser. B 359, 381-407 (2004).
S7. Y. Malhi, D. Wood, T. R. Baker, J. Wright, O. L. Phillips et al., Global Ch. Biol. 12,
1107-1138 (2006).
S8. H. ter Steege, N. C. A. Pitman, O. L. Phillips, J. Chave, D. Sabatier et al., Nature 443,
444-447 (2006).
S9. J. Chave, R. Condit, H.C. Muller-Landau, S. Thomas, P. S. Ashton et al., PLoS Biology
6, e45 (2008).
S10. J. Q. Chambers, J. dos Santos, R. J. Ribeiro, N. Higuchi, For. Ecol. Man. 152, 73–84
(2001).
S11. J. Chave, C. Andalo, S. Brown, N. A. Cairns, J. Q. Chambers et al., Oecologia 145, 87-
99 (2005).
S12. T. R. Baker, G. Lopez-Gonzalez, G., “Wood density”, in Functional Traits Database,
Version 4, G. Lopez-Gonzalez, O. L. Phillips, T. R. Baker, S. L. Lewis, J. Peacock et
al. Date of extraction July 2007.
S13. J. Chave, H. C. Muller-Landau, T. R. Baker, T. A. Easdale, H. Ter Steege et al., Ecol.
Appl. 16, 2356-2367 (2006).

22
Amazon drought sensitivity

S14. B. L. Boyle. TaxonScrubber Version 1.2. The SALVIAS Project,


http:/www.salvias.net. Date of extraction October 2005.
S15. O. L. Phillips, Y. Malhi, N. Higuchi, W. F. Laurance, P. Núñez V. et al. Science 282,
439-442 (1998).
S16. T. M. Araujo, N. Higuchi, J. A Carvalho Jr, Anais da Academia Brasileira de Ciencias
68, 35-41 (1996).
S17. J. Heuveldop, M. Neumann, Turrialba 33, 25-38 (1983).
S18. C. Uhl, K. Clark, N. Dezzeo, P. Maquirino Ecology 69, 751-763 (1988).
S19. R. Pélissier, B. Riéra, La Terre et la Vie 48, 21-33 (1993)
S20. J. Korning, H. Balslev, J. Veg. Sci. 4, 77-86 (1994).
S21. D. Sheil, R. May, J. Ecol. 84, 91–100 (1996).
S22. S. L. Lewis, O.L. Phillips, D. Sheil, B. Vinceti, T. Baker et al., J. Ecol. 92, 929-944
(2004).
S23. C. Körner, Science 300, 1242–1243 (2003).
S24. J. Fisher, G. Hurtt, J. Chambers, Q. Thomas, Ecology Letters 11, 554-563 (2008).
S25. B. W. Nelson, V. Kapos J. B. Adams, W. Oliveira, O. P. G. Braun et al. Ecology, 75,
853-858 (1994).
S26. S. L. Lewis, G. Lopez-Gonzalez, B. Sonké, K. Affum-Baffo, T. R. Baker et al. Nature
(in press).
S27. A. Clauset, C. R. Shalizi, M. E. J. Newman. arXive E-print arXiv:0706.1062 (2007).
S28. P. P. Harris, C. Huntingford, P. M. Cox, Philos. Trans. R. Soc. London, Ser. B 363,
1753-1759 (2008).
S29. L. E. O. C. Aragão, Y. Malhi, R. M. Roman-Cuesta, S. Saatchi, L. O. Anderson, Y. E.
Shimabukuro, Geophys. Res. Lett., 34, L07701, doi:10.1029/2006GL028946 (2007).
S30. North America Space Administration
http://disc.gsfc.nasa.gov/data/datapool/TRMM_DP/01_Data_Products/02_Gridded/07_
Monthly_Other_Data_Source_3B_43/) (accessed 2008).
S31. B. A. Wielicki, B. R. Barkstrom, E. F. Harrison, R. B. Lee III, G. L. Smith et al., Bull.
Amer. Meteor. Soc. 77, 853-868 (1996).
S32. J. C. Ceballos, M. J. Bottino, J. M. J. Geophys. Res. 109: D02211 (2004).
S33. G. Bisht, V. Venturini, S. Islam, L. Jiang, Rem. Sens. Envt. 97: 52-67 (2005).
S34. W. J. Shuttleworth, Quart. J. Roy. Meteor. Soc. 110, 1163-1169 (1984).

23
Amazon drought sensitivity

S35. “Guidelines for Soil Profile Description”, in Soil Resources Management and
Conservation Service (Land and Water Development Division, FAO, Rome, ed. 3,
1990).
S36. G. W. Gee, J. W. Bauder, in Methods of Soil Analysis Part 1, A. Klute, Ed. Agron.
Monogr. 9. (ASA and SSSA, Madison, WI , 1986), pp. 383–411.
S37. IUSS Working Group WRB. World reference base for soil resources 2006. A
framework for international classification, correlation and communication (World Soil
Resources Reports No. 103. FAO, Rome, 2006).
S38. J. Tomasella, M. G. Hodnett, Soil Sci., 163, 190–202 (1998).
S39. W. J. Shuttleworth, Phil. Trans. Roy. Soc., Ser. B, 324, 299– 334 (1989).
S40. Y. Malhi, L.E.O.C. Aragão, D. Galbraith, C. Huntingford, R. Fisher et al.,
Proceedings of the National Academy of Sciences, U.S.A. (in press).
S41. J. B. Fisher, Y. Malhi, A. C. de Araújo, D. Bonal, M. Gamo et al., Predicting the land-
atmosphere water flux in the tropics, Gl. Ch. Biol. (in press).
S42. P.M. Cox et al., Theoretical and Applied Climatology 78, 137-156 (2004).
S43. C. H. B. Priestley, R. J. Taylor, Monthly Weather Review 100, 81-92 (1972).
S44. M. G. Hodnett, L. Pimentel da Silva, H.R. da Rocha, R. Cruz Senna. J. Hydrol. 170,
233-254 (1995).
S45. B. A. Monteny, A. Casenave, Annales Geophysicae 7, 427-436 (1989).
S46. F. X. Dunin, A. R. Aston, Agric. Water Mangmt. 8, 305-323 (1984).
S47. J. B. Fisher, K. P. Tu, D. D. Baldocchi, Remote Sens. Environ. 112, 901-919 (2008).
S48. S. R. Saleska, K. Didan, A. R. Huete, H. R. da Rocha, Science 318, 612 (2007).
S49. W. Li, R. Fu. R. I. Negrón Juárez, K. Fernandes, Philos. Trans. R. Soc. London, Ser. B
363, 1767-1772 (2008).
S50. T. B. McKee, N. J. Doesken, J. Kleist, in Eight Conference on Applied Climatology
(Anaheim, CA, pp.179-194, 1993).
S51. O. L. Phillips, S. L. Lewis, T. R. Baker, K.-J. Chao, N. Higuchi, Philos. Trans. R. Soc.
London, Ser. B 363, 1819-1828 (2008).
S52. H. Eva, E. E. de Miranda, C. M. Di Bella, V. Gond, O. Huber et al. “A vegetation map
of South America” (EUR 20159 EN, European Commission, Luxembourg, 2002).
S53. H. Eva, O. Huber (eds.) “A proposal for defining the geographical boundaries of
Amazonia” (Rep. EUR 21808-EN, Off. Publ. Eur. Communities, Luxembourg, 2005).

24
Amazon drought sensitivity

S54. M. T. Nascimento, R. I. Barbosa, D. M. Villela, J. Proctor, Plant Ecology 192, 181-191


(2007).
S55. J. Heuveldop, M. Neumann, Turrialba 33, 25-38 (1983).

25
Amazon drought sensitivity

Supporting Online Material. Table S1. Plot attributes.

Location Effort Pre-2005 period 2005 interval


End Start Droughted and
Area Interval Start End Monitored Monitored
PlotCode Site Country Lat Long monitored
(ha) (y) date date pre-2005? 2005? 1
2005?1,2
ACU-01 1 Bolivia -15.25 -61.24 1.00 10.8 1996.4 2007.2 0 0 0
AGJ-01 2 Peru -11.89 -71.36 2.25 9.9 1993.8 2003.7 1 0 0
AGP-01 3 Colombia -3.72 -70.31 1.00 13.9 1992.2 2005.1 2005.1 2006.1 1 1 1
AGP-02 3 Colombia -3.72 -70.31 1.00 14.3 1991.9 2005.1 2005.1 2006.1 1 1 1
ALM-01 4 Peru -11.80 -71.47 2.00 10.0 1994.5 2004.5 1 0 0
ALP-1 5 Peru -3.95 -73.43 0.80 15.2 1990.9 2005.1 2005.1 2006.1 1 1 0
ALP-2 5 Peru -3.95 -73.44 0.88 15.2 1990.9 2005.1 2005.1 2006.1 1 1 0
ALP-30 5 Peru -3.95 -73.43 1.00 4.9 2001.3 2005.1 2005.1 2006.1 1 1 0
ANN-
6 Ecuador -0.53 -76.43 1.00 5.0 1986.0 1991.0 1 0 0
033
BDF-03 7 Brasil -2.41 -59.86 1.00 22.6 1981.1 2003.7 1 0 0
BDF-04 7 Brasil -2.41 -59.86 1.00 22.6 1981.1 2003.7 1 0 0
BDF-05 7 Brasil -2.41 -59.86 1.00 23.0 1981.2 2003.7 1 0 0
BDF-06 7 Brasil -2.41 -59.86 3.00 22.2 1981.5 2003.7 1 0 0
BDF-07 7 Brasil -2.38 -59.90 1.00 22.4 1981.6 2004.0 1 0 0
BDF-08 7 Brasil -2.38 -59.90 1.00 22.5 1981.6 2004.1 1 0 0
BDF-09 7 Brasil -2.39 -59.85 1.00 15.5 1987.0 2002.5 1 0 0
BDF-10 7 Brasil -2.38 -59.86 2.00 19.0 1983.5 2002.5 1 0 0
BDF-11 7 Brasil -2.38 -59.85 3.00 19.0 1983.5 2002.5 1 0 0
BDF-12 7 Brasil -2.39 -59.86 2.00 19.0 1983.5 2002.5 1 0 0
BDF-13 7 Brasil -2.39 -59.92 9.00 17.4 1985.9 2003.3 1 0 0
BDF-14 7 Brasil -2.40 -59.90 1.00 18.9 1984.2 2003.1 1 0 0
BEE-01 8 Bolivia -16.53 -64.58 1.00 4.6 2002.1 2004.8 2004.8 2006.6 1 1 0
BEE-05 8 Bolivia -16.53 -64.58 1.00 3.9 2002.7 2004.6 2004.6 2006.6 1 1 0
BNT-01 9 Brasil -2.63 -60.17 1.00 26.0 1980.5 2004.5 2004.5 2006.3 1 1 1
BNT-02 9 Brasil -2.63 -60.17 1.00 26.0 1980.5 2004.5 2004.5 2006.3 1 1 1
BNT-04 9 Brasil -2.63 -60.15 1.00 19.8 1986.5 2004.5 2004.5 2006.3 1 1 1
BNT-05 9 Brasil -2.63 -60.17 1.00 7.0 1986.5 1993.5 1 0 0
BNT-06 9 Brasil -2.63 -60.17 1.00 7.0 1986.5 1993.5 1 0 0
BNT-07 9 Brasil -2.63 -60.17 1.00 7.0 1986.5 1993.5 1 0 0
BOG-01 10 Ecuador -0.70 -76.48 1.00 11.6 1995.5 2007.1 1 0 0
BOG-02 10 Ecuador -0.70 -76.47 1.00 11.6 1995.5 2007.1 1 0 0
CAX-01 11 Brasil -1.74 -51.46 1.00 10.1 1994.5 2004.6 1 0 0
CAX-02 11 Brasil -1.74 -51.46 1.00 7.7 1995.5 2003.2 1 0 0
CAX-06 11 Brasil -1.72 -51.46 1.00 2.3 2004.6 2006.9 0 1 1
CAX-08 12 Brasil -1.86 -51.44 1.00 3.5 2003.4 2005.0 2005.0 2006.0 1 1 1
CHO-01 13 Bolivia -14.39 -61.15 1.00 4.9 1996.5 2001.5 1 0 0
CRP-01 14 Bolivia -14.54 -61.50 1.00 7.3 1994.2 2001.5 1 0 0
CRP-02 14 Bolivia -14.54 -61.50 1.00 7.2 1994.3 2001.5 1 0 0
CUZ-01 15 Peru -12.50 -68.97 1.00 17.2 1989.4 2003.8 2003.8 2006.6 1 1 1
CUZ-02 15 Peru -12.50 -68.97 1.00 17.2 1989.4 2003.7 2003.7 2006.6 1 1 1
CUZ-03 15 Peru -12.50 -68.96 1.00 17.2 1989.4 2003.7 2003.7 2006.6 1 1 1
CUZ-04 15 Peru -12.50 -68.96 1.00 17.2 1989.4 2003.8 2003.8 2006.6 1 1 1
ELD-
16 Venezuela 6.10 -61.40 1.00 32.2 1972.1 2004.3 1 0 0
1234
French
ELI-014 17 5.50 -53.00 0.78 10.1 1981.2 1991.3 1 0 0
Guiana
French
ELI-024 17 5.50 -53.00 1.00 10.0 1981.2 1991.2 1 0 0
Guiana
FRM-01 18 Guyana 5.22 -58.58 1.00 6.0 2000.5 2006.5 1 0 0
HCC-21 19 Bolivia -14.56 -60.75 1.00 10.2 1996.5 2006.7 1 0 0

26
Amazon drought sensitivity

Location Effort Pre-2005 period 2005 interval


End Start Droughted and
Area Interval Start End Monitored Monitored
PlotCode Site Country Lat Long monitored
(ha) (y) date date pre-2005? 2005? 1
2005?1,2
HCC-22 19 Bolivia -14.57 -60.75 1.00 10.2 1996.5 2006.7 1 0 0
JAC-01 9 Brasil -2.61 -60.21 5.00 6.0 1996.5 2002.5 1 0 0
JAC-02 9 Brasil -2.62 -60.20 5.00 6.0 1996.5 2002.5 1 0 0
JAS-02 20 Ecuador -1.08 -77.61 1.00 19.5 1987.6 2007.1 1 0 0
JAS-03 20 Ecuador -1.07 -77.61 1.00 18.2 1988.9 2007.1 1 0 0
JAS-05 20 Ecuador -1.06 -77.62 1.00 17.7 1989.4 2007.1 1 0 0
JAS-14 20 Ecuador -1.07 -77.61 1.00 16.6 1990.4 2007.1 1 0 0
JEN-11 21 Peru -4.88 -73.63 1.00 2.1 2005.2 2007.2 0 1 0
JEN-12 21 Peru -4.90 -73.63 1.00 2.1 2005.2 2007.3 0 1 0
JRI-01 22 Brasil -0.89 -52.19 1.00 11.0 1985.5 1996.5 1 0 0
LFB-01 23 Bolivia -14.56 -60.93 1.00 13.0 1993.6 2006.6 1 0 0
LFB-02 23 Bolivia -14.58 -60.83 1.00 13.0 1993.6 2006.6 1 0 0
LOR-01 24 Colombia -3.06 -69.99 1.00 2.3 2004.1 2006.4 0 1 1
LOR-U 24 Colombia -3.06 -69.99 1.00 13.9 1992.5 2004.1 2004.1 2006.4 1 1 1
LSL-01 13 Bolivia -14.41 -61.14 1.00 5.0 1996.5 2001.5 1 0 0
LSL-02 13 Bolivia -14.41 -61.14 1.00 5.0 1996.5 2001.5 1 0 0
MAR-01 25 Brasil -5.73 -49.05 2.00 7.6 1988.4 1996.0 1 0 0
MAR-02 25 Brasil -5.72 -49.03 2.00 8.2 1988.5 1996.6 1 0 0
MAR-03 25 Brasil -5.70 -49.00 2.00 7.5 1988.5 1996.0 1 0 0
MNU-01 26 Peru -11.87 -71.35 1.00 26.0 1974.7 2000.7 1 0 0
MNU-03 26 Peru -11.88 -71.35 2.00 10.0 1991.7 2001.7 1 0 0
MNU-04 26 Peru -11.88 -71.35 2.00 9.9 1991.7 2001.6 1 0 0
MNU-05 26 Peru -11.87 -71.35 2.00 10.0 1989.5 1999.5 1 0 0
MNU-06 26 Peru -11.87 -71.35 2.00 14.9 1989.8 2004.7 1 0 0
MNU-08 26 Peru -11.98 -71.18 2.00 9.9 1991.8 2001.7 1 0 0
MSH-01 5 Peru -3.78 -73.50 1.00 7.6 1983.0 1990.7 1 0 0
French
NOR-01 27 4.08 -52.67 10.00 9.6 1992.6 2002.2 1 0 0
Guiana
French
NOR-02 27 4.08 -52.67 11.00 7.5 1993.4 2000.9 1 0 0
Guiana
PAK-01 28 Peru -11.92 -71.25 1.00 4.0 1987.5 1991.5 1 0 0
French
PAR-01 28 5.25 -52.83 6.25 19.0 1984.7 2003.7 1 0 0
Guiana
French
PAR-06 29 5.25 -52.92 6.25 19.0 1984.7 2003.7 1 0 0
Guiana
French
PAR-11 29 5.25 -52.92 6.25 19.0 1984.7 2003.7 1 0 0
Guiana
French
PAR-13 29 5.25 -52.92 6.25 12.0 1991.7 2003.7 1 0 0
Guiana
French
PAR-14 29 5.25 -52.92 6.25 12.0 1991.7 2003.7 1 0 0
Guiana
French
PAR-15 29 5.25 -52.92 6.25 12.0 1991.7 2003.7 1 0 0
Guiana
French
PAR-20 29 5.28 -52.93 0.98 1.6 2004.5 2006.1 0 1 1
Guiana
French
PAR-21 29 5.28 -52.93 0.98 1.8 2004.4 2006.3 0 1 1
Guiana
French
PAR-22 29 5.28 -52.93 1.47 1.7 2004.5 2006.2 0 1 1
Guiana
French
PAR-23 29 5.28 -52.93 1.47 1.8 2004.4 2006.2 0 1 1
Guiana
PIB-05 30 Guyana 5.02 -58.62 1.00 13.0 1993.5 2006.5 1 0 0
PIB-06 30 Guyana 5.01 -58.62 1.00 13.0 1993.5 2006.5 1 0 0
PIB-12 30 Guyana 5.03 -58.60 1.00 13.0 1993.5 2006.5 1 0 0
RES-03 31 Brasil -10.82 -68.78 1.00 15.3 1991.4 2003.6 2003.6 2006.6 1 1 0
RES-04 31 Brasil -10.80 -68.77 1.00 15.3 1991.4 2003.6 2003.6 2006.6 1 1 0

27
Amazon drought sensitivity

Location Effort Pre-2005 period 2005 interval


End Start Droughted and
Area Interval Start End Monitored Monitored
PlotCode Site Country Lat Long monitored
(ha) (y) date date pre-2005? 2005? 1
2005?1,2
RES-05 32 Brasil -10.57 -68.31 1.00 15.4 1991.3 2003.6 2003.6 2006.7 1 1 1
RES-06 32 Brasil -10.56 -68.30 1.00 7.2 1999.5 2003.6 2003.6 2006.7 1 1 1
RIO-12 33 Venezuela 8.11 -61.69 0.50 32.2 1972.1 2004.3 1 0 0
ROM-
34 Brasil 3.42 -61.67 0.75 11.5 1991.8 2003.2 1 0 0
015
ROM-
34 Brasil 3.42 -61.67 0.75 11.5 1991.8 2003.2 1 0 0
025
ROM-
34 Brasil 3.42 -61.67 0.75 11.5 1991.8 2003.2 1 0 0
035
SCR-016 35 Venezuela 1.93 -67.02 1.00 10.3 1975.7 1986.0 1 0 0
SCR-037 35 Venezuela 1.75 -67.00 1.00 4.0 1975.5 1979.5 1 0 0
SCR-04 35 Venezuela 1.93 -67.04 1.00 30.7 1975.5 2004.2 2004.2 2006.2 1 1 0
SCR-05 35 Venezuela 1.93 -67.04 1.00 30.6 1975.6 2004.3 2004.3 2006.3 1 1 0
SCT-01 36 Bolivia -17.00 -64.77 1.00 4.7 2001.8 2003.6 2003.6 2006.5 1 1 0
SCT-06 36 Bolivia -17.00 -64.77 1.00 4.0 2002.7 2003.9 2003.9 2006.5 1 1 0
SUC-01 37 Peru -3.25 -72.91 1.00 14.0 1992.1 2005.1 2005.1 2006.1 1 1 0
SUC-02 37 Peru -3.25 -72.90 1.00 14.0 1992.1 2005.1 2005.1 2006.1 1 1 0
SUC-03 37 Peru -3.25 -72.92 1.00 5.0 2001.1 2005.1 2005.1 2006.1 1 1 1
SUC-04 37 Peru -3.25 -72.89 1.00 5.0 2001.2 2005.1 2005.1 2006.1 1 1 0
SUC-05 37 Peru -3.26 -72.90 1.00 4.0 2001.1 2005.1 1 0 0
TAM-01 38 Peru -12.84 -69.29 1.00 22.5 1983.8 2003.7 2003.7 2006.3 1 1 1
TAM-02 38 Peru -12.83 -69.29 1.00 26.0 1979.9 2003.7 2003.7 2006.3 1 1 1
TAM-04 38 Peru -12.84 -69.28 0.42 20.0 1983.8 2003.7 1 0 0
TAM-05 38 Peru -12.83 -69.27 1.00 22.0 1983.7 2003.7 2003.7 2006.0 1 1 1
TAM-06 38 Peru -12.84 -69.30 1.00 22.0 1983.7 2003.7 2003.7 2006.3 1 1 1
TAM-07 38 Peru -12.83 -69.26 1.00 22.0 1983.8 2003.7 2003.7 2006.3 1 1 1
TAM-08 38 Peru -12.83 -69.27 1.00 4.4 2001.5 2003.7 2003.7 2005.9 1 1 1
TAP-01 39 Brasil -3.31 -54.94 1.00 12.0 1983.5 1995.5 1 0 0
TAP-02 39 Brasil -3.31 -54.95 1.00 12.0 1983.5 1995.5 1 0 0
TAP-03 39 Brasil -3.31 -54.94 1.00 12.0 1983.5 1995.5 1 0 0
TAP-70 40 Brasil -2.89 -54.95 3.99 2.0 1999.5 2001.5 1 0 0
TAP-80 40 Brasil -2.90 -54.95 1.00 6.0 2000.0 2005.0 2005.0 2006.0 1 1 1
TAP-81 40 Brasil -2.90 -54.95 1.00 1.0 2000.0 2001.0 1 0 0
TEC-01 11 Brasil -1.71 -51.46 1.00 3.0 2002.9 2004.9 2004.9 2006.9 1 1 1
TEC-02 11 Brasil -1.74 -51.49 1.00 3.7 2003.2 2005.2 2005.2 2006.9 1 1 1
TEC-03 11 Brasil -1.73 -51.51 1.00 3.7 2003.2 2006.9 0 1 1
TEC-04 11 Brasil -1.75 -51.52 1.00 3.6 2003.3 2005.2 2005.2 2006.9 1 1 0
TEC-06 11 Brasil -1.73 -51.43 1.00 3.5 2003.3 2005.3 2005.3 2006.9 1 1 1
TEM-03 41 Brasil -2.43 -59.90 1.00 2.0 2003.9 2005.0 2005.0 2006.0 1 1 1
TEM-04 41 Brasil -2.44 -59.78 1.00 2.1 2003.9 2004.8 2004.8 2005.9 1 1 1
TEM-05 8 Brasil -2.61 -60.21 1.00 1.6 2004.0 2005.1 2005.1 2005.8 1 1 1
TEM-06 8 Brasil -2.60 -60.11 1.00 1.5 2004.2 2005.3 2005.3 2005.8 1 1 1
TIP-02 42 Ecuador -0.64 -76.15 1.00 9.0 1997.6 2007.2 1 0 0
TIP-03 42 Ecuador -0.64 -76.14 1.00 9.0 1998.5 2007.2 1 0 0
YAN-01 43 Peru -3.44 -72.85 1.00 22.0 1983.5 2005.1 1 0 0
YAN-02 43 Peru -3.43 -72.84 1.00 4.0 2001.1 2005.1 1 0 0
ZAR-01 44 Colombia -4.01 -69.91 1.00 2.0 2004.7 2006.7 0 1 1
ZAR-02 44 Colombia -4.00 -69.90 1.00 1.8 2004.9 2006.7 0 1 1
ZAR-03 44 Colombia -3.99 -69.90 1.00 1.8 2004.9 2006.7 0 1 1

1
with 2005 interval of less than 5 years
2
see SOM Methods for details on droughting definitions
3
basal area converted to biomass as described in Supporting Online Information, from S20.

28
Amazon drought sensitivity

4
from S19.
5
from S54.
6
basal area converted to biomass as described in Supporting Online Information, from S18.
7
basal area converted to biomass as described in Supporting Online Information, from S55.

29
Amazon drought sensitivity
SOM Table S2. Net biomass fluxes, before and after accounting for sampling effort, presented plot-by-plot and site-by-site.
Table shows change in AGB (Mg ha-1 a-1) for whole period, pre-2005, and 2005 datasets, and the difference in AGB change rates (Mg ha-1 a-1 a-1) in
2005 compared with pre-2005. All data were previously census-interval corrected.

A. Plots. Sampling effort adjustments were applied to the census-interval corrected data. Sampling effort weighting by area and time was only
performed for full period and pre-2005 data, as plot areas were equivalent (~1 ha) during 2005, and the hypothesis tested is that a drought shorter than
the measurement interval impacted biomass dynamics. See text for further details on dataset correction procedures.

Full Period Pre-2005 2005 all 2005 droughted 2005 - pre-2005, all 2005 - pre-2005, droughted
PlotCode Original Weighted Original Weighted Original Original Original Weighted Original Weighted
ACU-01 -1.417 -1.393
AGJ-01 0.386 0.399 0.386 0.408
AGP-01 0.954 1.064 0.974 1.070 0.729 0.729 -0.244 -0.469 -0.244 -0.469
AGP-02 -0.825 -0.931 0.217 0.242 -13.428 -13.428 -13.645 -11.306 -13.645 -11.306
ALM-01 0.933 0.954 0.933 0.976
ALP-1 0.040 0.046 0.455 0.514 -5.313 -5.768 -4.165
ALP-2 0.420 0.485 0.160 0.182 4.140 3.980 2.769
ALP-30 2.590 1.707 1.835 1.096 5.532 3.697 3.056
ANN-03 2.466 1.649 2.466 1.687
BDF-03 0.833 1.183 0.833 1.211
BDF-04 -2.375 -3.375 -2.375 -3.452
BDF-05 1.129 1.619 1.129 1.656
BDF-06 0.253 0.390 0.253 0.400
BDF-07 0.650 0.920 0.650 0.942
BDF-08 -0.036 -0.050 -0.036 -0.052
BDF-09 0.772 0.908 0.772 0.929
BDF-10 -0.044 -0.061 -0.044 -0.062
BDF-11 1.852 2.662 1.852 2.723
BDF-12 -0.561 -0.776 -0.561 -0.794
BDF-13 1.560 2.451 1.560 2.508
BDF-14 -0.060 -0.078 -0.060 -0.079
BEE-01 2.817 1.798 1.650 0.828 4.391 2.741 3.571
BEE-05 3.808 2.260 4.660 1.986 2.997 -1.663 1.119
BNT-01 1.238 1.888 1.469 2.225 -1.825 -1.825 -3.294 -4.055 -3.294 -4.055
BNT-02 2.066 3.150 1.978 2.996 3.261 3.261 1.283 0.274 1.283 0.274
BNT-04 1.116 1.483 1.213 1.574 0.148 0.148 -1.064 -1.426 -1.064 -1.426
BNT-05 2.681 2.121 2.681 2.170
BNT-06 -1.947 -1.540 -1.947 -1.576
BNT-07 1.404 1.111 1.404 1.136
30
Amazon drought sensitivity
BOG-01 2.196 2.240 3.991 3.149
BOG-02 0.698 0.712 1.727 1.363
CAX-01 1.598 1.518 1.598 1.553
CAX-02 -0.384 -0.318 -0.384 -0.326
CAX-06 0.111 0.050 0.154 0.154
CAX-08 1.245 0.697 1.899 0.744 -0.662 -0.662 -2.561 -1.294 -2.561 -1.294
CHO-01 -0.395 -0.261 -0.395 -0.268
CRP-01 0.175 0.141 0.175 0.144
CRP-02 1.804 1.444 1.804 1.477
CUZ-01 -0.032 -0.040 1.328 1.541 -6.832 -6.832 -8.160 -9.610 -8.160 -9.610
CUZ-02 0.824 1.023 0.944 1.094 0.251 0.251 -0.693 -0.797 -0.693 -0.797
CUZ-03 1.076 1.336 0.649 0.751 3.130 3.130 2.481 2.954 2.481 2.954
CUZ-04 0.772 0.957 1.004 1.163 -0.380 -0.380 -1.383 -1.611 -1.383 -1.611
ELD-1234 1.766 2.995 1.766 3.064
ELI-01 -0.265 -0.246 -0.265 -0.252
ELI-02 0.703 0.665 0.703 0.681
FRM-01 2.269 1.662 0.364 0.111
HCC-21 0.905 0.863 -0.272 -0.185
HCC-22 -0.067 -0.064 1.886 1.280
JAC-01 -0.695 -0.657 -0.695 -0.672
JAC-02 -0.813 -0.768 -0.813 -0.785
JAS-02 -1.923 -2.536 0.191 0.222
JAS-03 1.649 2.104 1.589 1.764
JAS-05 -0.611 -0.769 1.888 2.057
JAS-14 6.129 7.464 4.675 4.884
JEN-11 -2.573 -1.102 -2.573
JEN-12 3.670 1.579 3.670
JRI-01 -0.409 -0.405 -0.409 -0.415
LFB-01 2.760 2.978 1.992 1.700
LFB-02 2.195 2.367 1.731 1.477
LOR-01 -2.562 -1.094 -2.562 -2.562
LOR-U 1.244 1.376 0.757 0.790 3.813 3.813 3.056 3.247 3.056 3.247
LSL-01 1.987 1.329 1.987 1.360
LSL-02 6.387 4.271 6.387 4.370
MAR-01 0.041 0.037 0.041 0.038
MAR-02 0.201 0.188 0.201 0.192
MAR-03 -0.183 -0.164 -0.183 -0.168
MNU-01 0.379 0.579 0.379 0.592
MNU-03 0.742 0.759 0.742 0.776
MNU-04 1.104 1.125 1.104 1.151
31
Amazon drought sensitivity
MNU-05 1.147 1.173 1.147 1.200
MNU-06 -0.030 -0.037 -0.029 -0.037
MNU-08 0.686 0.700 0.686 0.716
MSH-01 1.176 0.972 1.176 0.994
NOR-01 2.290 2.911 2.290 2.979
NOR-02 0.400 0.474 0.400 0.485
PAK-01 -2.457 -1.470 -2.457 -1.504
PAR-01 0.223 0.346 0.223 0.354
PAR-06 0.704 1.095 0.704 1.120
PAR-11 1.181 1.836 1.181 1.878
PAR-13 1.852 2.385 1.852 2.441
PAR-14 1.835 2.363 1.835 2.417
PAR-15 2.564 3.301 2.564 3.378
PAR-20 -2.426 -0.917 -2.426 -2.426
PAR-21 -6.375 -2.573 -6.375 -6.375
PAR-22 -3.041 -1.319 -3.041 -3.041
PAR-23 -0.632 -0.279 -0.632 -0.632
PIB-05 1.929 2.080 0.540 0.437
PIB-06 1.270 1.370 -0.101 -0.082
PIB-12 1.838 1.982 0.799 0.647
RES-03 2.341 2.737 1.708 1.821 4.737 3.030 4.328
RES-04 0.577 0.675 1.048 1.117 -1.204 -2.252 -2.680
RES-05 1.235 1.448 1.878 2.011 -1.310 -1.310 -3.188 -3.607 -3.188 -3.607
RES-06 -0.153 -0.123 1.657 1.021 -2.472 -2.472 -4.129 -4.034 -4.129 -4.034
RIO-12 0.687 1.122 0.687 1.148
ROM-01 -0.847 -0.834 -0.847 -0.853
ROM-02 -1.878 -1.849 -1.878 -1.892
ROM-03 -1.546 -1.522 -1.546 -1.557
SCR-01 0.000 0.000 0.000 0.000
SCR-03 -0.090 -0.054 -0.090 -0.055
SCR-04 0.414 0.686 0.588 0.963 -1.954 -2.541 -2.966
SCR-05 -0.484 -0.801 -0.646 -1.057 1.371 2.017 2.477
SCT-01 1.481 0.959 1.665 0.682 1.371 -0.294 1.029
SCT-06 -4.330 -2.577 -4.210 -1.411 -4.351 -0.141 -3.807
SUC-01 0.234 0.262 0.190 0.209 3.661 3.471 2.513
SUC-02 -0.315 -0.352 -0.473 -0.521 1.678 2.152 1.768
SUC-03 1.111 0.745 1.949 1.192 -2.376 -2.376 -4.325 -3.190 -4.325 -3.190
SUC-04 1.518 1.014 0.881 0.538 4.030 3.150 2.414
SUC-05 1.768 1.062 1.768 1.086
TAM-01 -0.068 -0.097 -0.129 -0.175 0.376 0.376 0.505 0.607 0.505 0.607
32
Amazon drought sensitivity
TAM-02 0.474 0.723 0.404 0.603 1.097 1.097 0.693 0.653 0.693 0.653
TAM-04 1.324 1.669 1.324 1.708
TAM-05 0.993 1.393 0.950 1.343 1.829 1.829 0.878 0.680 0.878 0.680
TAM-06 2.095 2.939 2.166 3.060 0.741 0.741 -1.424 -2.212 -1.424 -2.212
TAM-07 -0.222 -0.311 0.529 0.723 -5.997 -5.997 -6.526 -7.550 -6.526 -7.550
TAM-08 0.883 0.553 3.106 1.416 -1.243 -1.243 -4.349 -2.753 -4.349 -2.753
TAP-01 3.025 3.134 3.025 3.206
TAP-02 3.549 3.677 3.549 3.761
TAP-03 2.723 2.821 2.723 2.887
TAP-70 3.700 2.213 3.700 2.264
TAP-80 1.251 0.916 2.560 1.751 -5.292 -5.292 -7.852 -6.156 -7.852 -6.156
TEC-01 -4.597 -5.904 -2.580 -1.983 -1.983 3.921 0.981 3.921 0.981
TEC-02 3.082 -2.369 5.070 2.227 0.626 0.626 -4.444 -1.615 -4.444 -1.615
TEC-03 -1.690 1.768 -1.690 -1.690
TEC-04 -0.651 -0.970 -0.834 -0.355 -0.437 0.397 -0.052
TEC-06 -2.604 -0.368 -3.078 -1.312 -2.077 -2.077 1.001 -0.710 1.001 -0.710
TEM-03 3.077 -1.465 3.406 1.068 2.747 2.747 -0.659 1.219 -0.659 1.219
TEM-04 2.765 1.305 3.747 1.117 2.238 2.238 -1.509 0.805 -1.509 0.805
TEM-05 -6.839 1.201 -1.415 -0.454 -15.501 -15.501 -14.085 -11.003 -14.085 -11.003
TEM-06 -1.151 -2.595 1.351 0.434 -6.656 -6.656 -8.007 -4.832 -8.007 -4.832
TIP-02 1.565 -0.427 0.617 0.398
TIP-03 1.195 1.404 1.780 1.038
YAN-01 1.136 1.072 1.136 1.630
YAN-02 3.202 1.594 3.202 1.962
ZAR-01 1.027 1.918 0.591 0.591
ZAR-02 2.203 0.431 0.692 0.692
ZAR-03 1.166 0.891 0.872 0.872

33
Amazon drought sensitivity
B. Sites. Sampling effort sampling adjustments were applied to the census-interval corrected data. Sampling effort weighting by area and time was
only performed for full period and pre-2005 data. Sampling effort weighted by area only for 2005 intervals. While both the total area monitored and
interval length did vary, the hypothesis being tested is that a drought shorter than the measurement interval impacted biomass dynamics so only
weighting by area is appropriate. See text for further details on dataset correction procedures.

Full Period Pre-2005 2005 all 2005 droughted 2005 - pre-2005, all 2005 - pre-2005, droughte
Site Original Weighted Original Weighted Original Weighted Original Weighted Original Weighted Original Weighted
1 -1.417 -0.797
2 0.386 0.228 0.386 0.240
3 0.065 0.062 0.595 0.576 -6.349 -5.686 -6.349 -5.686 -6.945 -6.262 -6.945 -6.262
4 0.933 0.546 0.933 0.574
5 1.056 1.290 0.906 1.120 1.453 1.486 0.547 0.367
6 2.466 0.944 2.466 0.992
7 0.552 1.645 0.552 1.729
8 3.279 1.840 2.913 1.352 3.677 3.855 0.764 2.503
9 -0.294 -0.604 1.097 2.316 -4.109 -4.993 -4.109 -4.993 -5.207 -8.036 -5.207 -7.309
10 1.447 1.259 2.859 2.007
11 -0.642 -0.783 -0.588 -0.623 -1.112 -1.255 -1.112 -1.255 -0.524 -0.632 -0.524 -0.632
12 1.245 0.399 1.899 0.435 -0.662 -0.471 -0.662 -0.471 -2.561 -0.906 -2.561 -0.906
13 2.660 1.965 2.660 2.066
14 0.990 0.687 0.990 0.723
15 0.660 1.004 0.981 1.441 -0.958 -1.080 -0.958 -1.080 -1.939 -2.522 -1.939 -2.522
16 1.766 1.714 1.766 1.802
17 0.219 0.176 0.219 0.185
18 2.269 0.951 0.364 0.065
19 0.419 0.342 0.807 0.493
20 0.729 1.020 2.201 3.153
21 0.549 0.215 0.549 0.504
22 -0.409 -0.232 -0.409 -0.244
23 2.478 2.274 1.862 1.409
24 0.750 0.542 0.748 0.458 0.625 0.560 0.625 0.560 -0.125 0.102 -0.125 0.102
25 0.020 0.019 0.020 0.020
26 0.571 0.997 0.571 1.048
27 1.461 1.474 1.461 1.549
28 -2.457 -0.841 -2.457 -0.884
29 0.930 2.140 1.237 2.669 -3.118 -3.768 -3.118 -3.768 -4.355 -6.438 -4.355 -6.438
30 1.679 1.922 0.413 0.373
31 1.459 1.446 1.378 1.286 1.767 2.303 0.389 1.017

34
Amazon drought sensitivity
Full Period Pre-2005 2005 all 2005 droughted 2005 - pre-2005, all 2005 - pre-2005, droughte
Site Original Weighted Original Weighted Original Weighted Original Weighted Original Weighted Original Weighted
32 0.792 0.679 1.767 1.367 -1.891 -1.693 -1.891 -1.693 -3.658 -3.061 -3.658 -3.061
33 0.687 0.918 0.687 0.966
34 -1.424 -1.504 -1.424 -1.581
35 -0.033 -0.053 -0.026 -0.042 -0.291 -0.310 -0.265 -0.268
36 -1.179 -0.646 -1.541 -0.509 -1.752 -2.203 -0.221 -1.694
37 0.301 0.371 0.417 0.515 1.701 2.090 1.283 1.574
38 0.750 1.636 0.913 1.950 -0.974 -1.257 -0.974 -1.257 -1.887 -3.207 -1.887 -3.207
39 1.436 1.061 2.031 1.196 -5.292 -3.763 -5.292 -3.763 -7.852 -4.958 -7.852 -4.958
40 3.099 3.418 3.099 3.593
41 2.921 1.155 3.568 1.124 2.492 2.232 2.492 2.232 -1.076 1.109 -1.076 1.109
42 1.380 1.064 1.149 0.641
43 1.454 1.334 1.454 1.402
44 1.445 0.732 0.718 0.736 0.718 0.736

35
Amazon drought sensitivity

SOM Table S3A. Summary of Net Biomass Fluxes, before and after accounting for sampling effects,
with bootstrapped confidence intervals. Status of locations that were ‘droughted’ and ‘non-droughted’
defined in terms of cumulative precipitation deficit. In plots classed as “droughted” in 2005 the intensity and
length of the dry period in the 2005 interval both surpassed the long-term mean (see S.O.M. Methods 7).

Table shows delta AGB (Mg ha-1 a-1) for whole period, pre-2005, and 2005 datasets, and change in delta AGB
(Mg ha-1 a-1 a-1) for 2005- pre-2005 difference datasets. All data were previously census-interval corrected.

A.1 Plots

Dataset Original Weighted by Sampling Effort *


Mean, median 2.5, 97.5% Mean, median 2.5, 97.5% confidence
confidence intervals intervals

Full Period, n=136 0.624, 0.798 0.291, 0.929 0.725, 0.811 0.479, 0.991

Pre-2005, n=123 0.933, 0.933 0.616, 1.250 0.890, 0.963 0.647, 1.115

2005 all plots, n=55 -0.713, 0.151 -1.929, 0.296 N/A N/A

2005 droughted, n=38 -1.617, -0.647 -3.164, -0.542 N/A N/A

2005 non-droughted, 1.292, 1.678 -0.458, 2.674 N/A N/A


n=17
2005 – pre-2005, all -1.527, -1.064 -2.878, -0.362 -1.497, -0.797 -3.009, -0.440
plots, n=43

2005 – pre-2005, -2.776, -1.467 -4.668, -1.311 -2.386, -1.518 -3.965, -1.117
droughted, n=28
2005 – pre-2005, non- 0.798, 2.017 -0.881, 1.992 0.758, 1.768 -0.779, 2.004
droughted, n=15
2005 – pre-2005 core -3.076, -1.467 -5.043, -1.669 -2.839, -1.482 -4.833, -1.477
droughted**, n=26

* Sampling effort adjustments were applied to the census-interval corrected data. Sampling effort weighting
by time and by area for pre-2005 interval. For 2005 intervals no weighting was performed as plots all ~1 ha,
and hypothesis being tested is that a drought shorter than the measurement interval impacted biomass
dynamics. See text for further details on dataset correction procedures.

** Core-droughted plots are defined as having had (1) greater mean monthly cumulative water deficit
(MCWD) in the 2005 interval, (2) greater mean length of water deficit in 2005 interval, and (3) <5% chance of
any TRMM-associated errors in MCWD leading us to have misclassified the location as 2005 MCWD-
droughted.

36
Amazon drought sensitivity

A.2 Sites

Dataset Original Weighted by Sampling Effort *


Mean, median 2.5, 97.5% Mean, median 2.5, 97.5% confidence
confidence intervals
intervals

Full Period, n=44 0.851, 0.771 0.504, 1.212 0.741, 0.900 0.452, 1.019

Pre-2005, n=41 0.998, 0.933 0.614, 1.360 0.942, 0.984 0.628, 1.267

2005 all sites, n=19 -0.712, -0.662 -1.941, 0.352 -0.735, -0.375 -1.793, 0.284

2005 droughted, n=12 -1.719, -1.043 -3.196, -0.343 -1.703, 1.256 -3.071, -0.437

2005 non-droughted, 1.015, 1.453 -0.219, 2.178 1.116, 1.486 -0.393, 2.319
n=7
2005 – pre-2005, all -1.978, -1.076 -3.494, -0.908 -1.799, -0.906 -3.363, -0.552
sites, n=17

2005 – pre-2005, -3.284, -2.561 -4.262, -1.963 -3.099, -2.061 -4.749, -1.540
droughted, n=11
2005 – pre-2005, non- 0.416, 0.468 -0.000, 0.862 0.583, 0.692 -0.568, 1.545
droughted, n=6
2005 – pre-2005 core -3.609, -3.110 -5.227, -2.267 -3.787, -2.368 -5.841, -2.192
droughted**, n=10

* Sampling effort adjustments were applied to the census-interval corrected data. Sampling effort
weighting by time and by area for pre-2005 interval. For 2005 intervals weighting only done by area, and
not also by time, as total areas varied but the hypothesis being tested is that a drought shorter than the
measurement interval impacted biomass dynamics.
** Core-droughted plots are defined as having had (1) greater mean monthly cumulative water deficit
(MCWD) in the 2005 interval, (2) greater mean length of water deficit in 2005 interval, and (3) <5%
chance of any TRMM-associated errors in MCWD leading us to have misclassified the location as 2005
MCWD-droughted.

37
Amazon drought sensitivity

SOM Table S3B Summary of Net Biomass Fluxes, before and after accounting for sampling effects,
with bootstrapped confidence intervals. Status of locations that were ‘droughted’ and ‘non-droughted’
defined in terms of cumulative daily moisture deficit accounting for precipitation, radiation, and soil water-
holding capacity. In plots classed as “droughted” in 2005 the intensity and length of the dry period in the
2005 interval both surpassed the long-term mean (see S.O.M. Methods 7).

Table shows delta AGB (Mg ha-1 a-1) for whole period, pre-2005, and 2005 datasets, and change in delta AGB
(Mg ha-1 a-1 a-1) for 2005- pre-2005 difference datasets. All data were previously census-interval corrected.

B.1 Plots

Dataset Original Weighted by Sampling Effort *


Mean, median 2.5, 97.5% Mean, median 2.5, 97.5% confidence
confidence intervals intervals

Full Period, n=136 0.624, 0.798 0.291, 0.929 0.725, 0.811 0.479, 0.991

Pre-2005, n=123 0.933, 0.933 0.615, 1.214 0.846, 0.920 0.617, 1.082

2005 all plots, n=55 -0.712, -0.178 -1.184, 0.239 N/A N/A

2005 droughted, n=38 -1.351, -0.632 -2.822, -0.250 N/A N/A

2005 non-droughted, 0.928, 0.999 -0.875, 2.634 N/A N/A


n=17
2005 – pre-2005, all -1.607, -1.064 -2.942, -0.457 -1.414, -1.254 -3.002, -0.386
plots, n=43

2005 – pre-2005, -2.341, -1.404 -4.069, -1.039 -2.106, -1.196 -3.672, -0.876
droughted, n=30
2005 – pre-2005, non- 0.200, 2.017 -3.441, 2.897 0.256, 1.671 -2.317, 2.714
droughted, n=13

* Sampling effort adjustments were applied to the census-interval corrected data. Sampling effort
weighting for 2005 intervals only done by area, and not also by time, as hypothesis being tested is that a
drought shorter than the measurement interval impacted biomass dynamics. See text for further details on
dataset correction procedures.

38
Amazon drought sensitivity

B.2 Sites

Dataset Original Weighted by Sampling Effort *


Mean, median 2.5, 97.5% Mean, median 2.5, 97.5% confidence
confidence intervals
intervals

Full Period, n=44 0.851, 0.771 0.504, 1.212 0.773, 0.931 0.487, 1.056

Pre-2005, n=41 0.998, 0.933 0.614, 1.360 0.950, 0.992 0.631, 1.278

2005 all sites, n=19 -0.697, -0.542 -1.963, 0.288 -0.674, -0.452 -1.855, 0.182

2005 droughted, n=13 -1.255, -1.043 -2.685, 0.157 -1.143, -1.583 -2.840, -0.009

2005 non-droughted, 0.583, 0.603 -0.676, 1.789 0.541, 0.481 -0.726, 1.713
n=6
2005 – pre-2005, all -2.051, -1.481 -3.424, -0.998 -1.944, -1.054 -3.367, -0.778
sites, n=17

2005 – pre-2005, -2.698, -1.939 -4.256, -1.332 -2.648, -2.624 -4.290, -1.111
droughted, n=12
2005 – pre-2005, non- -0.368,-0.121 -1.899, 1.184 -0.277, -0.109 -1.708, 0.895
droughted, n=5

* Sampling effort adjustments were applied to the census-interval corrected data. Sampling effort
weighting by time and by area for pre-2005 interval. For 2005 intervals weighting only done by area, and
not also by time, as total areas varied but the hypothesis being tested is that a drought shorter than the
measurement interval impacted biomass dynamics.

39
Amazon drought sensitivity
SOM Table S4. Differences in estimated moisture balance for Amazon plots and sites, 2005 interval – pre-2005 interval.
AWC = available water-holding capacity, mm; SWD = soil water deficit (i.e., depletion below AWC), mm.

A. Plots

Absolute conditions in 2005 interval Conditions in 2005, relative to pre-2005 interval

Drought Intensity Drought Length Drought Intensity Drought Length Drought Intensity
SWD/AWC % rain (30-day SWD/AWC % rain (30-day running rain (mean maximum cumulative
running mean mean <100mm) water deficit)
<100mm)
daily data calendar month data
Plot annual maximum % of interval that difference (%) difference (% of difference (mm) difference (mm) *
(%) was droughted intervals that were
droughted)
AGP-01
72.7 19.5 +21.1 +13.9 +84.3 +95.3
AGP-02
ALP-1 44.6 2.7 +0.7 -2.5
-2.1 -10.1
ALP-2 21.2 2.7 -6.3 -2.6
ALP-30 10.3 2.6 -1.7 -3.1 +9.0 -17.9
BEE-01 56.9 59.3 +2.2 +2.4
-2.7 -11.9
BEE-05 61.2 63.7 +6.4 +9.5
BNT-01
BNT-02 28.6 27.5 +8.7 +5.2 +31.8 +39.0
BNT-04
CAX-08 62.5 48.3 +5.5 +0.9 +54.5 +25.8
CUZ-01
CUZ-02 80.8 39.9 +11.7 +7.8 +55.3 +63.0
CUZ-03
CUZ-04
LOR-01 26.0 9.1 -4.1 +1.5 +3.8 -1.2
PAR-20 91.5 39.5
PAR-21 N/A N/A N/A N/A
PAR-22 69.6 39.5
PAR-23
RES-03 76.0 48.4 +3.1 +0.9
25 4 +6 0
40
Amazon drought sensitivity
RES-04 85.5 48.4 +0.2 +0.9
RES-05 83.8 53.0 +2.6 +6.4 +70.5 +51.3
RES-06 85.8 53.4 +2.7 +2.8 +81.7 +60.9
SCR-04 8.5 3.1 -2.7 -1.2
+9.1 +2.5
SCR-05 8.5 3.0 -3.5 -1.3
SCT-01 32.5 22.1 +1.5 +2.5 -6.0 +4.1
SCT-06 32.5 22.1 +8.6 -3.0 -6.0 +4.1
SUC-01 20.2 0.8 -4.4 -2.8 -14.7 -10.8
SUC-02 8.0 0.8 -1.4 -2.8 -14.7 -10.8
SUC-03 36.2 1.1 +4.9 +0.1 +3.6 -9.9
SUC-04 14.5 0.8 -2.5 -1.6 +3.3 -9.9
TAM-01 47.7 26.3 +9.6 +5.8
TAM-02 59.2 26.3 +8.3 +5.8
TAM-05 74.4 26.3 +3.3 +5.8 +76.4 +51.6
TAM-06 71.4 26.3 +4.4 +6.0
TAM-07 78.5 26.7 +14.0 +6.6
TAM-08 81.1 31.8 +7.4 +5.6 +128.4 +118.3
TAP-80 N/A 45.8 N/A +13.4 +49.5 +52.5
TEC-01 30.3 24.1 +1.4 +0.9 +0.7 -29.3
TEC-02 28.5 24.5 +0.1 +2.5 +0.5 +53.4
TEC-03 28.5 24.0 N/A N/A N/A N/A
TEC-04 28.5 24.0 +0.1 -0.4
+0.5 +53.4
TEC-06 28.5 25.7 +0.1 +3.4
TEM-03 34.0 26.5 +10.8 +2.7
TEM-04 34.0 31.4 +10.6 +8.6 +87.0 +88.1
TEM-05 34.0 31.0 +10.8 +9.1
TEM-06 34.0 43.4 +10.8 +21.3

* median estimates, based on TRMM perturbation analysis to assess errors in delta MCWD

41
Amazon drought sensitivity
B. Sites

Absolute conditions in 2005 interval Conditions in 2005, relative to pre-2005 Conditions in 2005, relative to pre-
interval 2005 interval
Drought Drought Length Drought Intensity Drought Length Drought Intensity
Intensity
SWD/AWC % rain (30-day SWD/AWC % rain (30-day rain (mean MCWD)
running mean running mean
<100mm) <100mm)
daily data calendar month data
Site annual maximum % interval difference (%) difference (%) difference (mm) difference (mm) *
(%)
3 72.7 19.5 +21.1 +13.9 +84.3 +95.3
5 25.4 2.7 -2.4 -2.7 +1.6 -12.7
8 59.1 61.5 +4.3 +6.0 -27.3 -12.0
9 28.6 27.5 +9.6 +9.2 +53.9 +58.7
11 28.9 24.5 +0.9 +1.6 +0.5 +30.8
12 62.5 48.3 +5.5 +0.9 +54.5 +25.8
15 80.8 39.9 +11.7 +7.8 +55.3 +63.0
24 26.0 9.1 -4.1 +1.5 +3.8 -1.2
29 80.6 39.5 +16.5 +14.8 +91.5 +121.9
31 19.2 48.4 +1.7 +0.9 -25.4 +6.0
32 84.8 53.2 +2.7 +4.6 +70.5 +56.1
35 8.5 3.0 -3.1 -1.3 +9.1 +2.5
36 32.5 22.1 +4.5 -0.2 -16.1 +4.2
37 19.7 1.0 -0.8 -1.6 +8.5 -10.5
38 68.7 27.3 +7.8 +5.9 +97.8 +63.2
40 N/A 45.8 N/A +13.4 +50.5 +52.5
41 34.2 29.0 +10.8 +5.7 +87.0 +88.1

* median estimates, based on TRMM perturbation analysis to assess errors in delta MCWD

42
Amazon drought sensitivity

SOM Table S5. Exploring the relationship of relative biomass dynamics with different relative
drought indices.

Relative biomass dynamics computed as the difference between 2005-interval net above-ground biomass trend
and the pre-2005 net above-ground biomass trend.

A. Plots.

Relative AGB dynamics vs. n Kendall’s Spearman’s Pearson’s r


tau-b rs

Drought Intensity
Mean annual 2005-interval MCWD - mean 22 -0.50*** -0.71*** -0.67***
annual pre-2005-interval MCWD (Fig. 2)
As above, but expressed as a fraction of pre-2005 22 -0.36** -0.48* -0.47*
values
Mean annual 2005-interval maximum 36 -0.44*** -0.59*** -0.60***
SWD/AWC - mean annual pre-2005-interval
maximum
As above, but expressed as a fraction of pre-2005 36 -0.45*** -0.60*** -0.55***
values
Mean annual 2005-interval minimum SMC 36 -0.33** -0.47** -0.53***
attained – mean annual pre-2005-interval
minimum
As above, but expressed as a fraction of pre-2005 36 -0.45*** -0.60*** -0.55***
values
Mean annual 2005-interval maximum SWD – 36 -0.32** -0.46** -0.49**
mean annual pre-2005-interval maximum
As above, but expressed as a fraction of pre-2005 36 -0.34** -0.47** -0.52***
values
Mean annual 2005-interval MCWD - mean 15 -0.55** -0.74** -0.70**
annual pre-2005-interval MCWD (only plots
certainly drier in 2005 interval than before, after
accounting for rainfall uncertainty)
Drought Length
Change in n(days 30-day running total <100mm) 37 -0.43*** -0.57*** -0.62***
As above, but expressed as a fraction of pre-2005 37 -0.39*** -0.55*** -0.58***
values
Change in n calendar months <100mm 37 -0.44*** -0.56*** -0.59***
As above, but expressed as a fraction of pre-2005 37 -0.41*** -0.56*** -0.51***
values

43
Amazon drought sensitivity

B. Sites

Relative AGB dynamics vs. n Kendall’s Spearman’s Pearson’s r


tau-b rs

Drought Intensity
Mean annual 2005-interval MCWD – mean 17 -0.46** -0.69*** -0.68**
annual pre-2005-interval MCWD
As above, but expressed as a fraction of pre-2005 17 -0.33 -0.45 -0.50*
values
Mean annual 2005-interval maximum 16 -0.53** -0.73*** -0.80***
SWD/AWC - mean annual pre- 2005-interval
maximum
As above, but expressed as a fraction of pre-2005 16 -0.55*** -0.76*** -0.74**
values
Mean annual 2005-interval minimum SMC 16 -0.44** -0.63** -0.54*
attained – mean annual pre-2005-interval
minimum
As above, but expressed as a fraction of pre-2005 16 -0.53** -0.74*** -0.73**
values
Mean annual 2005-interval maximum SWD – 16 -0.43* -0.62** -0.54*
mean annual pre-2005-interval maximum
As above, but expressed as a fraction of pre-2005 16 -0.38* -0.58* -0.63**
values
Mean annual 2005-interval MCWD – mean 12 -0.42* -0.59* -0.56*
annual pre-2005-interval MCWD (only sites
certainly drier in 2005 interval than before, after
accounting for rainfall uncertainty)
Drought Length
Change in n(days 30-day running total <100mm) 17 -0.57*** -0.73*** -0.84***
As above, but expressed as a fraction of pre-2005 17 -0.53** -0.71*** -0.73***
values
Change in n calendar months <100mm 17 -0.50** -0.66** -0.69**
As above, but expressed as a fraction of pre-2005 17 -0.53** -0.70** -0.69**
values

Definitions:
MCWD = cumulative water deficit (i.e., annual maximum cumulative rainfall deficit below 3.33mm day-1)
SMC = soil moisture content, mm
AWC = available water-holding capacity, mm
SWD = soil water deficit (i.e., depletion in SMC below AWC), mm

Plots treated as independent if moisture deficits in 2005 or pre-2005 intervals were different from other plots.
Non-independent values were combined prior to analysis by averaging.

44
Amazon drought sensitivity

SOM Table S6. A. Pre-2005 plot-based measurements scaled-up to estimate annual, aggregate biomass carbon change.

Census-interval corrected data Weighted by monitoring effort

mean 2.5% CI 97.5% CI mean 2.5% CI 97.5% CI

Mg AGB 0.93 0.62 1.21 0.89 0.65 1.11


ha-1 a-1

expansion factor

smaller trees and lianas 1.099 1.02 0.68 1.33 0.98 0.71 1.23

below ground gain 1.37 0.38 0.25 0.49 0.36 0.26 0.45

total biomass gain, Mg ha-1 a-1 1.40 0.93 1.83 1.34 0.97 1.68

necromass gain, Mg ha-1a-1 0.127 0.18 0.12 0.23 0.17 0.12 0.21

total biomass & necromass gain, Mg ha-1 a-1 1.58 1.04 2.06 1.51 1.10 1.89

net carbon flux, Mg ha-1 a-1 0.5 0.79 0.52 1.03 0.75 0.55 0.95

Amazonia (TRMM pixel, 2006 forest cover)


2006 estimated forest, ha 5.28 x 108 4.18 x 108 2.76 x 108 5.44 x 108 3.99 x 108 2.90 x 108 4.99 x 108

South America (GLC2000, late 1990's forest cover)


Humid tropical forest 6.31 x 108 4.99 x 108 3.29 x 108 6.50 x 108 4.76 x 108 3.46 x 108 5.96 x 108
Dry tropical forests 1.47 x 10 8
1.16 x 108 7.66 x 107 1.51 x 108 1.10 x 108 8.06 x 107 1.39 x 108
Flooded tropical forests 2.53 x 107 2.00 x 107 1.32 x 107 2.61 x 107 1.91 x 107 1.39 x 107 2.39 x 107

Total 8.03 x 108 6.35 x 108 4.19 x 108 8.27 x 108 6.06 x 108 4.41 x 108 7.59 x 108

45
Amazon drought sensitivity

SOM Table S6. B. Pre-2005 site-based measurements scaled-up to estimate annual, aggregate biomass carbon change.

Census-interval corrected data Weighted by monitoring effort

mean 2.5% CI 97.5% CI mean 2.5% CI 97.5% CI

Mg AGB 1.00 0.61 1.36 0.94 0.63 1.27


ha-1 a-1

expansion factor

smaller trees and lianas 1.099 1.10 0.67 1.49 1.04 0.69 1.39

below ground gain 1.37 0.41 0.25 0.55 0.38 0.26 0.52

total biomass gain, Mg ha-1 a-1 1.50 0.92 2.05 1.42 0.95 1.91

necromass gain, Mg ha-1 a-1 0.127 0.19 0.12 0.26 0.18 0.12 0.24

total biomass & necromass gain, Mg ha-1 a-1 1.69 1.04 2.31 1.60 1.07 2.15

net carbon flux, Mg ha-1 a-1 0.5 0.85 0.52 1.15 0.80 0.53 1.07

Amazonia (TRMM pixel, 2006 forest cover)


2006 estimated forest, ha 5.28 x 108 4.47 x 108 2.75 x 108 6.09 x 108 4.22 x 108 2.81 x 108 5.67 x 108

South America (GLC2000, late 1990's forest cover)


Humid tropical forest 6.31 x 108 5.34 x 108 3.28 x 108 7.28 x 108 5.04 x 108 3.36 x 108 6.78 x 108
Dry tropical forests 1.47 x 108 1.24 x 108 7.64 x 107 1.69 x 108 1.17 x 108 7.82 x 107 1.58 x 108
Flooded tropical forests 2.53 x 10 7
2.14 x 107 1.32 x 107 2.92 x 107 2.02 x 107 1.35 x 107 2.72 x 107

Total 8.03 x 108 6.79 x 108 4.18 x 108 9.26 x 108 6.41 x 108 4.28 x 108 8.63 x 108

46
Amazon drought sensitivity

SOM Table S6. C. Estimated 2005 impact, from plot data. Scaling from the statistical distribution of plot biomass change data (sampling
effort-corrected AGB change relative to pre-2005 for each plot), to the whole area of Amazon forest droughted in 2005.

Mean 2.5% CI 97.5% CI

Mg AGB ha-1 a-1 -2.39 -1.12 -3.97

expansion factor

mean 2005 interval length, a 1.97

smaller trees and lianas 1.099

droughted forest area, ha (TRMM


data) 3.31 x 108

additional fraction of Amazonia with 1.031


unreliable TRMM data

sum AGB impact, Mg -1.76 x 109 -8.25 x 108 -2.93 x 109

below ground 1.37 -6.52 x 108 -3.05 x 108 -1.08 x 109

sum biomass impact, Mg -2.41 x 109 -1.13 x 109 -4.01 x 109

sum carbon impact, Mg 0.5 -1.21 x 109 -0.57 x 109 -2.01 x 109

47
Amazon drought sensitivity

SOM Table S6. D. Estimated 2005 impact, from site data. Scaling from the statistical distribution of site biomass change data (sampling
effort-corrected change relative to pre-2005 for each plot), to the whole area of Amazon forest droughted in 2005.

Mean 2.5% CI 97.5% CI

Mg AGB ha-1 a-1 -3.10 -1.54 -4.75

expansion factor

mean 2005 interval length, a 1.97

smaller trees and lianas 1.099

droughted forest area, ha (TRMM


data) 3.31 x 108

additional fraction of Amazonia with 1.031


unreliable TRMM data

sum AGB impact, Mg -2.29 x 109 -1.14 x 109 -3.51 x 109

below ground 1.37 -8.47 x 108 -4.21 x 108 -1.30 x 109

sum biomass impact, Mg -3.14 x 109 -1.56 x 109 -4.81 x 109

sum carbon impact, Mg 0.5 -1.57 x 109 -0.78 x 108 -2.40 x 109

48
Amazon drought sensitivity

SOM Table S6. E. Estimated 2005 impact, from plot data. Scaling from the plot biomass dynamics vs. climate relationship to the whole
Amazon forest area droughted in 2005, using the fitted relationship between change in mean annual MCWD values and change in biomass
dynamics, accounting for TRMM-related uncertainty in monthly MCWD values and weighted by pre-2005 monitoring effort, and with the maximum
drought impact set by the limits of the measured relationship.

Mean 2.5% CI 97.5% CI

Mg AGB -1.05 x 10-9 -5.46 x 10-8 -1.72 x 10-9

expansion factor

mean 2005 interval length, a 1.97

additional proportion of Amazonia 1.031


with unreliable TRMM data

smaller trees and lianas 1.099

sum AGB impact, Mg -2.33 x 10-9 -1.22 x 10-9 -3.83 x 10-9

below ground, Mg 1.37 -8.63 x 10-8 -4.51 x 10-8 -1.42 x 10-9

sum biomass impact, Mg -3.20 x 10-9 -1.67 x 10-9 -5.25 x 10-9

sum carbon impact, Mg 0.5 -1.60 x 10-9 -0.83 x 10-8 -2.63 x 10-9

49
Amazon drought sensitivity

SOM Table S6. F. Estimated 2005 impact, from site data. Scaling from the site biomass dynamics vs. climate relationship to the whole
Amazon forest area droughted in 2005, using the fitted relationship between change in mean annual MCWD values and change in biomass
dynamics, accounting for TRMM-related uncertainty in monthly MCWD values and weighted by pre-2005 monitoring effort, and with the maximum
drought impact set by the limits of the measured relationship.

Mean 2.5% CI 97.5% CI

Mg AGB -0.96 x 10-9 -5.15 x 10-8 -1.71 x 10-9

expansion factor

mean 2005 interval length, a 1.97

additional proportion of Amazonia 1.031


with unreliable TRMM data

smaller trees and lianas 1.099

sum AGB impact, Mg -2.15 x 10-9 -1.15 x 10-9 -3.81 x 10-9

below ground, Mg 1.37 -7.96 x 10-8 -4.25 x 10-8 -1.41 x 10-9

sum biomass impact, Mg -2.95 x 10-9 -1.57 x 10-9 -5.21 x 10-9

sum carbon impact, Mg 0.5 -1.47 x 10-9 -0.79 x 10-9 -2.61 x 10-9

50
Amazon drought sensitivity

SOM Table S7. Statistics of the relationships between different measures of biomass dynamics and different measures of drought
conditions.

We report linear fits. All relationships weighted by sampling effort at each locality except for those based on the absolute droughting experienced in
2005 (Fig. S6). We account for uncertainty in drought measures only for the MCWD analyses (Fig. 2, Fig. S8A), for which we present the intercept
and slope of the bootstrapped median lines and their 95% bootstrapped confidence intervals; for all other analyses we report the statistics of the best-
fit linear regression.
intercept t P slope t P

Fig. 2, 2005 − pre2005, difference in mean annual MCWD, monthly data


AGB change, difference (Mg ha-1 a-1) 0.37 (-0.84,1.53) -0.05 (-0.08,-0.03)
Loss rate, difference (Mg ha-1 a-1) -0.75 (-2.02,0.84) 0.05 (0.01,0.08)
-1 -1
Gain rate, difference (Mg ha a ) -0.41 (-1.13,0.34) -0.00 (-0.02,0.01)
Fig. S5, 2005−pre2005 difference in mean annual maximum SWD/AWC %
AGB change, difference (Mg ha-1 a-1) 0.36 0.64 0.53 -0.33 -4.40 0.00
Loss rate, difference (Mg ha-1 a-1) -0.69 -1.12 0.27 0.27 3.29 0.00
-1 -1
Gain rate, difference (Mg ha a ) -0.34 -1.27 0.21 -0.05 -1.50 0.14
Fig. S6A, 2005 interval annual maximum SWD/AWC %
AGB change, absolute (Mg ha-1 a-1) 0.65 0.49 0.63 -0.03 -1.12 0.27
Loss rate, absolute (Mg ha-1 a-1) 4.50 3.62 0.00 0.02 1.02 0.31
-1 -1
Gain rate, absolute (Mg ha a ) 5.22 9.66 0.00 -0.00 -0.45 0.65
Fig. S6B, % 2005 interval with 30−day running mean P<100mm
AGB change, absolute (Mg ha-1 a-1) 0.01 0.01 0.99 -0.03 -0.79 0.44
Loss rate, absolute (Mg ha-1 a-1) 5.40 4.87 0.00 0.01 0.37 0.71

51
Amazon drought sensitivity

Gain rate, absolute (Mg ha-1 a-1) 5.42 11.76 0.00 -0.01 -1.00 0.32
Fig. S7, % of 2005 interval − % of pre2005 interval with 30−day running mean P<100mm
AGB change, difference (Mg ha-1 a-1) 0.55 1.12 0.28 -0.49 -5.76 0.00
-1 -1
Loss rate, difference (Mg ha a ) -0.94 -1.58 0.13 0.44 4.33 0.00
Gain rate, difference (Mg ha-1 a-1) -0.40 -1.20 0.24 -0.05 -0.80 0.43
Fig. S8A, sites, 2005 − pre2005, difference in mean annual MCWD, monthly data
AGB change, difference (Mg ha-1 a-1) -0.19 (-1.62,0.67) -0.04 (-0.06,-0.02)
-1 -1
Loss rate, difference (Mg ha a ) 0.09 (-1.30,1.57) 0.03 (0.01,0.06)
Gain rate, difference (Mg ha-1 a-1) -0.15 (-1.14,0.92) -0.01 (-0.03,0.00)
Fig. S8B, sites, 2005 − pre2005, difference in mean annual maximum SWD/AWC %
AGB change, difference (Mg ha-1 a-1) -0.31 -0.73 0.48 -0.27 -5.93 0.00
Loss rate, difference (Mg ha-1 a-1) 0.26 0.41 0.69 0.22 3.23 0.01
Gain rate, difference (Mg ha-1 a-1) -0.07 -0.18 0.86 -0.04 -1.09 0.29
Fig. S8C, sites, % of 2005 interval − % pre2005 interval with 30−day running mean P<100mm
AGB change, difference (Mg ha-1 a-1) -0.09 -0.20 0.84 -0.38 -7.04 0.00
Loss rate, difference (Mg ha-1 a-1) -0.05 -0.09 0.93 0.33 4.47 0.00
Gain rate, difference (Mg ha-1 a-1) -0.16 -0.39 0.70 -0.04 -0.82 0.42

52
100 100
Total plots monitored each year

A Monitored Plots

mid-point falling in year


80 80

Plots with interval


Total monitored
60 60

40 40
Interval mid-points

20 20

0 0
1980 1985 1990 1995 2000 2005
Year

5 5
B Time and Area
Mean interval length, y

4 4

Mean plot area, ha


Interval length

3 3

2 2
Plot area
1 1

0 0
1980 1985 1990 1995 2000 2005
Year

Fig. S1. 25 years of Amazon forest monitoring.


Figure denotes the fluctuations and trends in our monitoring effort which help to illuminate Figure 1 in
the main text.

A. Number of plots being monitored.


Red line (scale on right) represents number of intervals with mid-point in that year, showing the upsurge
centred on 2005. Black line (scale on left) represents the total number of plots being monitored in each
year. Best-fit polynomials are shown.

B. Time intervals and plot area


Red line (scale on right) represents the running mean plot area of 50 intervals with mid-point centred in
that year, showing greatest values in the late-1990’s when the fastest rates of change in total measured
biomass gains were recorded (cf. Fig. 1). Black line (scale on left) represents the running mean interval
length of 50 intervals with mid-point centred in that year, showing the sharp decline by 2005 that finally
permits detection of annual-scale events (cf. Fig. 1).
Figure S2. Determining the optimal weighting for whole period biomass change.
Residuals from plot-level biomass change in whole period plotted against years of monitoring and
area monitored, with weightings assessed empirically by the degree to which residual pattern is
removed.

Weighting by the square root of the years of monitoring and by the cube root of plot size both best
remove any pattern in the residuals, suggesting that these relationships describe the relationship
between sampling error and census length or plot size respectively. The final weighting of the
square root of the number of years of monitoring plus cube root of plot size minus one (to avoid
double-accounting) shows no pattern in the residuals, suggesting the weighting is appropriate.

A. Weight all plots equally vs. area. y = -0.077x + 1.482. r2 =0.010.


8
7
6
Residual

5
4
3
2
1
0
0 2 4 6 8 10 12
Plot Size, ha

B. Weight by cube-root of plot size. y = 0.067x + 1.349. r2 = 0.007.


8

7
Residual x cube root ha

0
0 2 4 6 8 10 12
Plot Size, ha

C. Weight by square-root of plot size. y = 0.182x + 1.238. r2 = 0.047.

8
7
Residual x square root ha

6
5
4
3
2
1
0
0 2 4 6 8 10 12
Plot Size, ha
D. Weight all plots equally vs. time. y = -0.069x + 2.155. r2 =0.153.
8
7
6
Residual

5
4
3
2
1
0
0 10 20 30 40
Years of Monitoring

E. Weight by cube-root of time. y = -0.043x + 3.0746. r2 =0.021.


16
14
Residual x cuberoot time

12
10
8
6
4
2
0
0 5 10 15 20 25 30 35
Years of Monitoring

F. Weight by square-root of time. y = 0.002x + 3.645. r2 =0.000.


25
Residuals x squareroot time

20

15

10

0
0 10 20 30 40
Years of Monitoring

G. Weight by cube-root of plot size and square-root of time -1. y = 0.001x + 3.330. r2 =0.000.
6

5
Residuals x Weight

0
0 5 10 15 20 25
Weight
Figure S3. Determining the optimal weighting for pre-2005 biomass change.
Residuals from plot-level change in pre-2005 biomass plotted against years of monitoring and area
monitored, with differing weightings assess empirically by the degree to which residual pattern is
removed.

Weighting by the square root of the years of monitoring and by the cube root of plot size both best
remove any pattern in the residuals, suggesting that these relationships describe the relationship
between sampling error and census length or plot size respectively. The final weighting of the
square root of the number of years of monitoring plus cube root of plot size minus one (to avoid
double-accounting) shows no pattern in the residuals, suggesting the weighting is appropriate.

A. Weight all plots equally vs. area. y = -0.053x + 1.279, r2 =0.009.


8
7
6
5
Residual

4
3
2
1
0
0 5 10 15
Plot Size, ha

B. Weight by cube-root of plot size. y = 0.081x + 1.159, r2 = 0.018.


8
7
Residuals x cuberoot ha

6
5
4
3
2
1
0
0 2 4 6 8 10 12
Plot Size, ha

C. Weight by square-root of plot size. y = 0.189x + 1.059, r2 = 0.087.


8
7
Residuals x squareroot ha

6
5

4
3

2
1

0
0 2 4 6 8 10 12
Plot Size, ha
D. Weight all plots equally vs. time. y = -0.059x + 1.864, r2 =0.139.
8
7
6
5
Residual

4
3
2
1
0
0 10 20 30 40
Years of Monitoring

E. Weight by cube-root of time. y = -0.029x + 2.599, r2 =0.013.


10
Residuals x cuberoot time

0
0 10 20 30 40
Years of Monitoring

F. Weight by square-root of time. y = 0.016x + 3.031, r2 =0.001.


16
Residuals x squareroot time

14
12
10
8
6
4
2
0
0 10 20 30 40
Years of Monitoring

G. Weight by cube-root of plot size and square-root of time -1. y = 0.001x + 3.488. r2 =0.000.

6
R esiduals x Weight

5
4

3
2

1
0
0 2 4 6 8 10 12 14 16
Weight
Figure S4. Long-term rainfall records.
Precipitation and moisture deficits are plotted from four RAINFOR sites to illustrate the main climate
regimes sampled and the variety of conditions experienced in 2005 by Amazon forests. Daily values
in blue (rainfall) and orange (cumulative water deficit); thick lines represent 30-day moving averages.
For in-depth analysis of the drought climatology see main text references 18,19.

A. An aseasonal climate with no 2005 drought (San Carlos de Rio Negro, Venezuela);
B. An aseasonal climate with an unusually dry period in 2005 (Agua Pudre, Colombia);
C. A seasonal climate with a typically strong dry season in 2005 (Sacta, Bolivia);
D. A seasonal climate with an exceptionally strong dry season in 2005 (Cusco Amazónico, Peru).
Fig. S5. Biomass dynamics response to the relative intensity of the 2005
drought.
Changes in overall biomass change (blue), mortality rates (red), growth rates (green),
in the drought interval relative to pre-2005 shown as a linear function of drought
relative intensity, and weighted by monitoring effort. Polynomial or break-point
functions do not provide closer fits. Change in drought as measured by change in
annual maximum soil water deficit as fraction of soil available water capacity
(accounts for rainfall, evapotranspiration, and soil).
2005 − pre2005, difference in change in AGB loss and gain ( Mg ha−1 a−1) 2005 − pre2005, difference in net change in AGB (Mg ha−1 a−1)


−10 −5 0 5 10 −10 −5 0 5 10




B.
A.

0
0





● ●

5
5




● ●

10
10
Figure S5





15
15

20
20

2005 − pre2005, difference in mean annual minimum SWD/AWC %



Fig. S6. Absolute 2005 biomass dynamics response to the absolute droughting
experienced in 2005.
Drought metrics were first normalised with respect to monitoring interval as described
in detail in SOM methods. Absolute drought intensity computed as the mean driest
point reached by the soil in 2005-interval year(s) as a fraction of its AWC; absolute
drought length computed as the fraction of days in the 2005-interval for which the 30-
day running total precipitation fell below 100mm. Plot-level biomass change (blue),
and mortality (red) and growth (green) in the drought interval are plotted as a function
of absolute intensity and length of drought, regardless of pre-2005 climatology. No
relationship fits at P <0.05.

A. Rate of biomass growth, mortality, and net gain in 2005 interval vs. maximum
intensity of dry period (mean annual maximum soil water deficit as fraction of
AWC).
B. Rate of biomass growth, mortality, and net gain in 2005 interval vs. length of
dry period (% of period with 30-day running rainfall total <100mm).

Models are unweighted as plot sizes are equal and the measurement intervals are
longer than the drought event. For further discussion of weighting see S.O.M.
Methods; regression models are detailed in Table S7.
Figure S6
A. B.
15

15
2005 absolute AGB change rate ( Mg ha−1 a−1)

10

10
● ●
5

5
● ● ● ●
● ●● ●●

● ● ● ●
● ● ● ●

● ● ●● ●
●● ● ● ● ●
●●
● ●
0

0
● ● ●
● ●

● ●● ● ● ● ●
● ● ●
●● ● ● ● ●● ●● ● ●
● ●
● ●
−5

−5
● ● ●
● ● ● ●
● ●
−10

−10
−15

−15
● ●

20 40 60 80 0 10 20 30 40 50 60
20

20
2005 absolute loss and gain rates ( Mg ha−1 a−1)

15

15
10

10
5

5
0

20 40 60 80 0 10 20 30 40 50 60

2005 interval annual maximum SWD/AWC % % 2005 interval with 30−day running mean P<100mm
Fig. S7. Relative biomass dynamics response to relative length of drought
experienced in 2005.
Difference in rate of plot biomass growth, mortality, and net gain in 2005 interval
relative to pre-2005 interval vs. increase in drought length in the 2005 interval relative
to the pre-2005 interval (i.e., difference in % of period with 30-day running rainfall
total <100mm).

For details of weighting see S.O.M. Methods; weighted regression models are detailed
in Table S7.
2005 − pre2005, difference in change in AGB loss and gain ( Mg ha−1 a−1) 2005 − pre2005, difference in net change in AGB (Mg ha−1 a−1)

−10 −5 0 5 10 −10 −5 0 5 10


●●●

B.
A.


●●


0
0




●● ●


●●

5
5






● ●

10
10
Figure S7


15
15

20
20

% of 2005 interval − % of pre2005 interval with 30−day running mean P<100mm


Fig. S8. Relative biomass dynamics response at site-level to relative droughting.
Drought metrics were first normalised with respect to monitoring interval as described in
detail in SOM. Changes in site-averaged overall biomass change (blue), and mortality (red)
and growth rates (green) in the drought interval relative to pre-2005 interval are plotted as a
function of site-averaged drought parameters.

A. As for Figure 2 in main text, but for Sites. Biomass dynamics change plotted against
change in drought intensity, measured by change in maximum cumulative water
deficit (MCWD, accounts only for rainfall). Uncertainty in precipitation is included in
the bootstrapped estimates of the relationship of delta biomass change vs. delta
MCWD and confidence intervals.
B. Biomass dynamics change plotted against change in drought intensity, measured by
change in mean annual maximum soil water deficit as fraction of AWC (accounts for
rainfall, soil, evapotranspiration).
C. Biomass dynamics change plotted against increase in drought length, measured by
increase in (i.e., difference in % of period with 30-day running rainfall total <100mm).

For details of weighting see SOM Methods; weighted models are detailed in Table S7.
Figure S8
A. B. C.
10

10

10
2005 − pre2005, difference in net change in site AGB ( Mg ha−1 a−1)

5
● ● ●
● ● ●
● ● ● ● ● ●
0

0
●● ● ● ● ● ● ● ●
● ● ●
● ● ●


● ● ● ● ●
● ● ●

● ●
● ● ●
● ● ●
−5

−5

−5
● ● ●

● ●
−10

−10

−10
−40 −20 0 20 40 60 80 100 120 0 5 10 15 20 0 5 10 15
2005 − pre2005, difference in net change in site AGB loss & gain ( Mg ha−1 a−1)

10

10

10
5

5
0

0
−5

−5

−5
−10

−10

−10
−40 −20 0 20 40 60 80 100 120 0 5 10 15 20 0 5 10 15

2005 − pre2005, difference in mean annual MCWD, sites (mm) 2005 − pre2005, difference in mean annual minimum SWD/AWC %, sites % 2005 interval − % pre2005 interval with 30−day running mean P<100mm, sites

You might also like