Global Change Biology (2010) 16, 255–271, doi: 10.1111/j.1365-2486.2009.01921.x
Evaluating the impacts of climate and elevated carbon
dioxide on tropical rainforests of the western Amazon
basin using ecosystem models and satellite data
H I R O F U M I H A S H I M O T O *w , F O R R E S T M E L T O N *w , K A Z U H I T O I C H I I z, C R I S T I N A
M I L E S I *w , W E I L E W A N G *w and R A M A K R I S H N A R . N E M A N I w
*California State University, Monterey Bay, Seaside, CA 93955, USA, wNASA Ames Research Center, Moffett Field, CA 94035,
USA, zFaculty of Symbiotic Systems Science, Fukushima University, Fukushima, Japan
Abstract
Forest inventories from the intact rainforests of the Amazon indicate increasing rates of
carbon gain over the past three decades. However, such estimates have been questioned
because of the poor spatial representation of the sampling plots and the incomplete
understanding of purported mechanisms behind the increases in biomass. Ecosystem
models, when used in conjunction with satellite data, are useful in examining the carbon
budgets in regions where the observations of carbon flows are sparse. The purpose of this
study is to explain observed trends in normalized difference vegetation index (NDVI)
using climate observations and ecosystem models of varying complexity in the western
Amazon basin for the period of 1984–2002. We first investigated trends in NDVI and
found a positive trend during the study period, but the positive trend in NDVI was
observed only in the months from August to December. Then, trends in various climate
parameters were calculated, and of the climate variables considered, only shortwave
radiation was found to have a corresponding significant positive trend. To compare the
impact of each climate component, as well as increasing carbon dioxide (CO2) concentrations, on evergreen forests in the Amazon, we ran three ecosystem models (CASA,
Biome-BGC, and LPJ), and calculated monthly net primary production by changing a
climate component selected from the available climate datasets. As expected, CO2
fertilization effects showed positive trends throughout the year and cannot explain the
positive trend in NDVI, which was observed only for the months of August to December.
Through these simulations, we demonstrated that the positive trend in shortwave
radiation can explain the positive trend in NDVI observed for the period from August
to December. We conclude that the positive trend in shortwave radiation is the most
likely driver of the increasing trend in NDVI and the corresponding observed increases
in forest biomass.
Keywords: Amazon, ecosystem model, NDVI, NPP
Received 19 November 2008 and accepted 23 January 2009
Introduction
The dynamics of the evergreen forests in the Amazon
basin have a substantial impact on worldwide climate
systems and play an important role in the regulation of
atmospheric carbon dioxide (CO2) concentrations.
Therefore, understanding the response of the AmazoCorrespondence: Hirofumi Hashimoto, California State University,
Monterey Bay, Seaside, CA 93955, USA, tel. 1 1 650 604 6446, fax
1 1 650 604 6569, e-mail: hirofumi.hashimoto@gmail.com
nian evergreen forests to climate change is critical for
predicting the future global climate (Cramer et al., 2004).
Early numerical modeling studies for this region focused on the impact of deforestation on regional climate
change and predicted that deforestation would have
irreversible effects (by decreasing both precipitation
and evapotranspiration) on the basin’s hydrologic cycle
(Salati & Vose, 1984). Along with the impacts of deforestation, it is expected that the response of the Amazonian forests to climate change accompanied by elevated
CO2 concentrations is the most important factor for
Published 2009
This article is a US Government work and is in the public domain in the USA
255
256 H . H A S H I M O T O et al.
predicting the future carbon balance of these ecosystems (Malhi, 2008). To assess the effect of these anthropogenic and climatological influences on future
Amazon ecosystems, it is crucial to understand the
mechanisms regulating the current state of the carbon
balance and hydrologic cycles of the Amazon basin.
Despite large uncertainty due to the insufficient number
of observations over such a large region, a number of
studies of the Amazon carbon balance have shown that
these tropical evergreen forests have been sequestering
carbon for the last two decades (see the review paper by
Ometto et al., 2005). An intercomparison study of atmospheric CO2 transportation models [the Atmospheric
Tracer Transport Model Intercomparison Project (Transcom)] showed, although with large uncertainty in the
estimates, that during the 1990s the carbon balance of
tropical South America was almost neutral, while South
America as a whole served as a small source of CO2
(Gurney et al., 2002). This result implied that the evergreen forest ecosystems of the Amazon must function as
carbon sinks to offset carbon emissions resulting from
land-use change and fire (Houghton et al., 2000). Other
plot-based studies provided proof that forest biomass
has been increasing in the Amazon basin for the last few
decades (Baker et al., 2004; Lewis et al., 2004b; Phillips
et al., 2004), indicating that these evergreen forest ecosystems sequestered carbon in most of the Amazon
basin. Also, measures of net ecosystem exchange using
eddy covariance techniques revealed that carbon uptake was occurring at many sites [e.g. Reserva Jaru
(Grace et al., 1995), Reserva Biológica do Cuieiras (Malhi
et al., 1998), and Caxiuaña (Carswell et al., 2002), except
for Tapajós (Saleska et al., 2003)].
Optical remote sensing and ecosystem model intercomparison are viable tools for deriving estimates of the
amount of vegetation, and clarifying the driving mechanism and flows of carbon in regions where a limited
number of carbon exchange observations are available.
Satellite-based measures of vegetation density over vast
areas can be derived through vegetation indexes such as
the normalized difference vegetation index (NDVI)
from the Advanced Very High Resolution Radiometer
(AVHRR) satellite instrument, for which a global longterm record is available since 1981. However, optical
remote sensing observations of tropical evergreen forests are complicated by residual cloud contamination
(even after monthly compositing procedures are applied) and by the saturation of the band reflectance
for regions with very dense vegetation, and require
independent validation through the use of ecosystem
models or field measurements.
Although a number of intercomparison studies of
ecosystem carbon models have been performed at the
global scale (Cramer et al., 1999), there are only few
regional modeling studies for the tropics which are
based on individual models. Early ecosystem modelbased studies of the Amazon basin found that net
primary production (NPP, the net amount of carbon
fixed in vegetation by the photosynthetic process) was
sensitive to changes in soil moisture resulting from
variability in precipitation, and that heterotrophic respiration increased with temperature (Tian et al., 1998).
Recently, both diagnostic (Nemani et al., 2003; Hicke,
2005) and prognostic (Ichii et al., 2005) ecosystem models reported that the carbon budget of the Amazon
basin is more sensitive to the interannual variation in
shortwave radiation. For example, the Biome-BGC results of Ichii et al. (2005) indicated that the interannual
variability in gross primary production (GPP, the
amount of carbon absorbed by photosynthesis discounted for carbon lost through respiration) was correlated with the interannual variability in shortwave
radiation. Others suggest that along with increases in
solar radiation and air temperature, increases in atmospheric CO2 concentrations have also contributed to
accelerated growth of tropical forests in the Amazon
basin (Lewis et al., 2004a, b).
The goal of this study is to understand whether the
observed carbon sequestration in evergreen forest
ecosystems of the Amazon basin could be explained by changes in climate alone, or whether other
potential drivers, such as elevated CO2 concentrations
must also be considered. To answer this question, we
used multiple ecosystem models in combination
with multiple climate datasets to simulate and account
for the observed trend in the NOAA/AVHRR NDVI
data.
Materials and methods
Study area
For our modeling analysis, we selected a study region of
the western Amazon extending between 0–101S and
60–701W (Fig. 1b). This is the most humid region of
the Amazon basin, with mean annual precipitation
42000 mm, and is characterized by smaller rates of
deforestation compared with regions in the eastern
and southern parts of the Amazon basin (Asner et al.,
2005), eliminating the need to model deforestation.
Also, in this region atmospheric contamination from
aerosols caused by biomass burning is small compared
with the southeastern Amazon region (Kobayashi &
Dye, 2005). Huete et al. (2006) observed that the seasonal
dynamics of the enhanced vegetation indices (EVI)
measured using the MODerate Resolution Imaging
Spectroradiometer (MODIS) differ in the eastern and
western parts of the Amazon basin. Myneni et al. (2007)
Published 2009
This article is a US Government work and is in the public domain in the USA, Global Change Biology, 16, 255–271
C L I M AT E I M PA C T S O N T H E W E S T E R N A M A Z O N E C O S Y S T E M
257
Atmospheric Administration satellite series (Tucker
et al., 2005). We used the GIMMS data to analyze the
interannual variability in vegetation productivity over
the western Amazon in response to changes in climatic
variables and atmospheric CO2 concentrations. The
GIMMS NDVI long-term dataset has been georeferenced, atmospherically corrected, and composited to
15-day maximum NDVI values. To correct for the solar
zenith angle perturbation, GIMMS used the empirical
mode decomposition (EMD) (Pinzón et al., 2005). Correction for stratospheric aerosols was made for the two
anomalously high aerosol burden periods resulting
from emissions from two major volcano eruptions
(El Chichon, April 1982 to December 1984; and Mt
Pinatubo, June 1991 to December 1994). To further
reduce cloud contamination in the study area, we
aggregated the GIMMS data into monthly composites
by selecting, for each pixel of each month, the maximum value from the 15-day composite GIMMS data.
Intercomparison and trend analysis of climate datasets
We compared several datasets for each climate variable
that is required as an input for the ecosystem models.
The details for each datasets utilized are listed in Table
1. For regional studies in locations that have a low
density of meteorological stations, such as the Amazon
basin, it is preferable to use satellite-derived climate
Fig. 1 (a) Mean monthly climatology of the International
Satellite Cloud Climatology Project (ISCCP) shortwave radiation
(solid line in upper panel), Global Precipitation Climatology
Project (GPCP) precipitation (bars in the lower panel), and
CRU temperature (solid line in lower panel) within the study
region. (b) Map of GPCP mean annual precipitation (1984–2002).
The square area surrounded by thick lines is the study area.
also reported a good correlation between the dynamics
of satellite observed leaf area index (LAI) and shortwave radiation, rather than precipitation, in the western
Amazon. However, the western Amazon has fewer
observations of both climate and vegetation compared
with the Eastern Amazon. These factors led us to
further analyze the carbon balance in the western
Amazon using ecosystem models.
Table 1 Descriptions for each dataset
Spatial
Parameter
Time
Time
Dataset resolution
resolution
period
CRU
UD
NCEP
0.51
0.51
ca. 2.51
Monthly
Monthly
6-Hourly
1901–2002
i
1950–1999
i
1948–present m
0.51
2.51
ca. 2.51
0.51
Monthly
Monthly
6-Hourly
Monthly
1901–2002
1979–present
1948–present
1950–1999
280 km
Monthly
NCEP
ca. 2.51
6-Hourly
1983/1987– m
2004
1948–present m
ISCCP
280 km
Monthly
NCEP
ca. 2.51
6-Hourly
Method
Temperature
Precipitation
CRU
GPCP
NCEP
UD
Shortwave radiation
ISCCP
i
s,i
m
i
Cloud cover
Trend analysis of NDVI datasets
We used a 1982–2003 subset of the 8 km resolution
NDVI data from the Global Inventory Modeling and
Mapping Studies (GIMMS) group. The GIMMS data
record is derived from observations collected by the
AVHRR instrument onboard the National Oceanic and
1983/1987– m
2004
1948–present m
Note that the descriptions are based on the available data. The
characters ‘i’, ‘s,’ and ‘m’ in the method column indicate the
method used to produce the gridded data set: interpolation (i),
satellite data (s), and climate model (m).
Published 2009
This article is a US Government work and is in the public domain in the USA, Global Change Biology, 16, 255–271
258 H . H A S H I M O T O et al.
data to capture the heterogeneity in meteorological
observations for the overall region. Some surface-level
climate variables, however, are difficult to measure
directly from satellite sensors (e.g. incident shortwave
radiation), and thus we need to use datasets that are
interpolated from ground observations, or modeled
datasets as alternatives to satellite-measured climate
datasets. The uncertainty in the interpolated datasets
depends on the density of meteorological stations in the
network, and measurement errors at a single station can
have a substantial influence on interpolated estimates
for regions with sparse station coverage (Malhi &
Wright, 2004). Modeled datasets can also have large
systematic errors, and thus the results deduced from the
modeled climate datasets (e.g. reanalysis datasets) can
be biased (Zhao et al., 2006). Therefore, our approach is
to first identify common trends in the available datasets,
and then compare the trends among datasets to obtain
reliable climate trends.
Temperature. We used three temperature datasets: CRU
TS 2.1 (Mitchell & Jones, 2005), the Willmott and
Matsuura datasets (Legates & Willmott, 1990b) from
the University of Delaware (Willmott & Matsuura,
2001), and the NCEP NCAR Reanalysis datasets
(Kalnay et al., 1996) (hereafter, referred to as CRU,
UD, and NCEP, respectively). Both CRU and UD are
interpolated datasets derived from data records from
ground observation networks using different
interpolation algorithms. The NCEP dataset is the
product of a global reanalysis that ingests datasets
including satellite data, sonde data, aircraft data,
surface observations, sea surface temperature data,
and other observations. NCEP has many output
parameters with varying degrees of accuracy. Kalnay
et al. (1996) ranked the variables by the extent of model
dependency and ranked surface air temperature as the
variable on which both models and observations have
the most influence, and is therefore the most reliable
parameter within the NCEP datasets.
Precipitation. We used the Global Precipitation
Climatology Project (GPCP) Version 2 combined
precipitation dataset (Adler et al., 2003) (hereafter,
refer to as GPCP), as well as precipitation data from
the CRU, UD (Legates & Willmott, 1990a), and NCEP
datasets. GPCP combines several satellite and gauge
network datasets to calculate a monthly gridded
precipitation data product. NCEP does not ingest
surface observations of precipitation from gauges, and
thus was categorized as a model-derived variable by
Kalnay et al. (1996), which means that it is less reliable
relative to the other NCEP parameters. Both CRU and
UD precipitation data are interpolated (as with
temperature) from ground-based observation records.
Shortwave radiation and cloud cover. Currently, there is no
method to directly measure the downward shortwave
radiation fluxes at the land surface level using satellites.
There are, however, two model-derived datasets
available for long-term analysis: the International
Satellite Cloud Climatology Project (ISCCP) datasets
(ISCCP-FD) (Zhang et al., 2004), and the NCEP
shortwave radiation dataset. ISCCP provides incident
shortwave radiation flux data at the surface level from
1984 to present. ISCCP shortwave radiation data are
calculated using the radiative transfer code from the
GISS Global Circulation Model (GCM), which ingests
multiple global climate datasets. The NCEP downward
shortwave radiation dataset is also categorized by
Kalnay et al. (1996) as a model-derived variable,
similarly to NCEP precipitation. The comparison,
reviews, and validation for the ISCCP and NCEP
datasets were thoroughly addressed by Hicke (2005).
We also used cloud cover datasets from ISCCP and
NCEP in place of shortwave radiation as an input for
the LPJ ecosystem model. As with the downward
shortwave radiation flux, cloud cover in the NCEP
dataset is categorized as a model-derived variable. We
did not use cloud cover data from CRU, as none of the
ground stations used for interpolation were located
inside the study region. Furthermore, the CRU cloud
cover data are derived from cloud cover observations
collected from 1971 to 1995, and sunshine duration
measurements from 1996 to 2001, with no overlapping
period between the two inputs (Mitchell & Jones, 2005).
Since the detection of trends in cloud cover from the
ISCCP dataset is prone to artifacts at the edges of the
field of view of the satellites from which these
observations are derived (Evan et al., 2007), we used
the Earth Radiation Budget Experiment (ERBE) S-10N
NF Edition 3 dataset (Barkstrom, 1984) and Clouds and
the Earth’s Radiant Energy System (CERES) ES-4
datasets (Wielicki et al., 1996) to validate the results of
the trend analysis of cloud cover data from ISCCP and
NCEP datasets. The ERBE S-10N NF Edition 3 datasets
(hereafter, referred to ERBE) includes top-of-theatmosphere (TOA) reflected shortwave radiation,
albedo, and total irradiance data measured by the
nonscanner instrument carried by the Earth Radiation
Budget Satellite (ERBS) spacecraft, and are available
from 1984 to 1999. We calculated monthly downward
shortwave radiation from the 51 grid of ERBE data
measured by the middle field-of-view (MFOV)
detector. CERES is the project that succeeded ERBE in
measuring the global radiation budget. CERES ES-4
datasets represent one of the radiation flux products
Published 2009
This article is a US Government work and is in the public domain in the USA, Global Change Biology, 16, 255–271
C L I M AT E I M PA C T S O N T H E W E S T E R N A M A Z O N E C O S Y S T E M
Table 2
259
Comparison of components of the three models used in this study (CASA, Biome-BGC, and LPJ)
CASA
Biome-BGC
LPJ
Model type
Diagnostic model
Prognostic model
Climate input
Air temperature
Precipitation
Shortwave radiation
Photosynthesis model
Monteith-type LUE
model (Monteith,
1972)
Maximum air temperature
Minimum air temperature
Precipitation
Shortwave radiation
VPD
Farquhar biochemical model
(Farquhar et al., 1980) with
Jarvis-type empirical stomatal
conductance model (Jarvis, 1976)
Prognostic model with dynamic
vegetation
Air temperature
Precipitation
Cloud cover
Wet day
emax or Vcmax*
Predefined biomespecific emax
Vcmax from leaf nitrogen and
Rubisco activity
Parameters constraining
photosynthesis
Shortwave radiation
CO2 concentration
(optional)
Air temperature
Soil water content
FPAR
Evapotranspiration model
De Marsily (1986)
Autotrophic respiration
model
Not calculated
Shortwave radiation
CO2 concentration
Soil water content
Air temperature
VPD
LAI
Leaf nitrogen content
Penman–Monteith equation with
Jarvis stomatal conductance
model (Jarvis, 1976)
Maintenance respiration calculated
by Q10 model regulated by the
nitrogen contents.
Growth respiration of prescribed
ratio of carbon storage rate
Reference
Potter et al. (1993)
Thornton et al. (2002)
Combination of biochemical model
(Haxeltine & Prentice, 1996)
based on Farquhar et al. (1980)
and Ball-type empirical stomatal
conductance model (Ball et al.,
1987)
Vcmax from optimal nitrogen
allocation with canopy
(Haxeltine & Prentice, 1996)
Shortwave radiation
CO2 concentration
Soil water content
Air temperature
FPAR
LAI
Equilibrium evapotranspiration
restricted by water availability
Maintenance respiration calculated
by Q10 model accounting for
acclimation with prescribed
coefficient for each plant
functional type.
Growth respiration of 25% of the
residual of GPP subtracted from
maintenance respiration
Sitch et al. (2003) and Gerten et al.
(2004)
*emax is the canopy-level maximum light use efficiency. Vcmax is the maximum carboxylation rate of Rubisco.
LAI, Leaf Area Index; VPD, Vapor Pressure Deficit; FPAR, fraction of absorbed photosynthetically active radiation by vegetation.
derived from CERES instruments carried by TRMM,
Terra, and Aqua spacecraft, and were calculated using
the same algorithm used for ERBE data. We averaged
the monthly CERES data from the products observed by
the available sensors (Terra-FM1, Terra-FM2, AquaFM3, and Aqua-FM4).
Ecosystem model experiments
NPP variations are generally representative of corresponding variations in NDVI (Schloss et al., 1999). We
calculate NPP from three ecosystem carbon models:
CASA (version 2003.04.29) (Potter et al., 1993), BiomeBGC (version 4.2) (Thornton et al., 2002), and LPJ
(version 1.2) (Sitch et al., 2003; Gerten et al., 2004).
Detailed features of the three models are summarized
in Table 2.
CASA is a diagnostic model that requires NDVI data
as an input, from which LAI and fraction of absorbed
photosynthetically active radiation (PAR) by vegetation
(FPAR) are estimated using empirical equations. In
CASA, NPP is calculated as the product of maximum
Light Use Efficiency (emax), PAR, FPAR, and climatedriven regulation factors, which are functions of air
temperature and soil water content. Optionally, CASA
can use CO2 concentration data to simulate the CO2
fertilization effect on NPP. However, since the modeling
of the CO2 fertilization effect in CASA is very simple,
Published 2009
This article is a US Government work and is in the public domain in the USA, Global Change Biology, 16, 255–271
260 H . H A S H I M O T O et al.
with photosynthesis linearly increasing with CO2 concentrations, we did not use this feature of CASA in our
study.
Biome-BGC is a prognostic biogeochemical model
driven by climate data for the prescribed land cover.
Biome-BGC uses the Farquhar biochemical photosynthesis model (Farquhar et al., 1980) to calculate GPP, and
estimates NPP as the remainder of GPP subtracted from
autotrophic respiration, which is a function of temperature and biomass. Stomatal conductance in Biome-BGC
is modeled using a Jarvis-type model (Jarvis, 1976) as
the product of predefined maximum stomatal conductance and climate regulation factors [shortwave radiation, air temperature, soil water potential, and vapor
pressure deficit (VPD)]. Globally applicable models
sometimes require tuning for regional applications
(Thornton et al., 2002). In this study, the main model
and ecophysiological parameters of Biome-BGC were
from Ichii et al. (2007), which adjusted the submodels
and parameters to match the Tapajós flux data (Saleska
et al., 2003). Also, to match the seasonal variation in GPP
measured at the Tapajós flux site, we added a submodel
of seasonal variations in percent of leaf nitrogen in
Rubisco, following Ichii et al. (2007). A combined satellite and model study (Ichii et al., 2007) and an in situ plot
study (Nepstad et al., 1994) both suggested that very
deep root systems (deeper than 5 m) exist in the Amazon tropical forests. Thus, we set the rooting depth at
5 m (vs. the default value of 1 m).
LPJ is also a prognostic model, but includes a dynamic biogeography submodel, which determines the
land cover implicitly from climate data. The plant
functional types in the Amazon were mostly categorized as tropical broad-leaved evergreen by the LPJ
dynamic biogeography submodel. As with BiomeBGC, LPJ uses the Farquhar biochemical photosynthesis
model (Haxeltine & Prentice, 1996), and subtracts autotrophic respiration from GPP. Autotrophic respiration is
controlled by air temperature. Stomatal conductance in
LPJ is estimated from a Ball-type conductance model
(Ball et al., 1987). We did not adjust any of the default
parameters.
To obtain the initial allocation of carbon in an ecosystem (i.e. the allocation to leaves, stems, roots, soil, etc.),
carbon ecosystem process models require a spin-up
period, during which the model runs until it reaches
an equilibrium state for the desired climate conditions.
We used climate data from the run period (i.e.
1984–2002, except for the experiments that used UD
data from 1984 to 1999) for the spin-up for each model.
Using averages from the climate data for the entire
study period to spin-up the models would have led to
a value of 0 for the total net ecosystem exchange (NEE)
from 1984 to 2002.
To identify the climate variable(s) driving the observed trend in vegetation dynamics over the western
Amazon forests, we followed the methodology adopted
by Ichii et al. (2005), in which the ecosystem model runs
are driven by changing a single climate variable while
keeping the other climate variables fixed. The climate
variables that remain fixed during the modeling experiments are calculated as monthly averages for 1984–2002
from the selected datasets, including CRU for temperature, GPCP for precipitation, and ISCCP for radiation
and cloud cover. CRU temperature data were selected
because they are derived from interpolation of at least
20 stations in the study region from 1984 to 2002, and
therefore are deemed to reliably represent the monthly
temperature climatology for the western Amazon. An
intercomparison study of global precipitation datasets
indicated that the latitudinal average of GPCP showed
the most reasonable profile (with the least number of
outliers) relative to the other datasets (Fekete et al.,
2004). Shortwave radiation and cloud cover were selected from the ISCCP dataset, as Hicke (2005) showed
these data agreed more closely with observations than
NCEP radiation. Figure 1a shows the seasonal climatology of the study region from CRU temperature, GPCP
precipitation, and ISCCP shortwave radiation.
For Biome-BGC, which uses daily rather than
monthly climate inputs, we also calculated daily
anomalies. The daily anomalies were calculated as
deviations from the monthly average of temperature,
precipitation and solar radiation from the NCEP dataset, which is the only dataset that provides daily values.
For LPJ, we used the weather generator (Gerten et al.,
2004) to derive daily climate data from the monthly
climate inputs.
The observed trends in climate variables are sometimes only detectable during specific months, and analysis of yearly averages can be insufficient to
understand the phenomenon of climate change. For
example, Chagnon & Bras (2005) found a shift in the
seasonality of precipitation from a gauge record that
spanned 75 years. Since the responses of models to
climate change can vary depending on the season, for
all the modeling experiments we calculated both annual
and monthly trends in NPP.
To assess the influence of elevated CO2 concentration
on the modeled NPP, we used the CO2 concentration
data observed at Mauna Loa, Hawaii, as no continuous
data record was found at a more proximate location for
our study period. While there are almost certainly
differences in the absolute CO2 concentrations between
Mauna Loa and the Amazon basin, the relative change
in concentrations over time both at Mauna Loa and in
the Amazon basin is expected to be consistent. Thus, the
results of the simulations using the Mauna Loa CO2
Published 2009
This article is a US Government work and is in the public domain in the USA, Global Change Biology, 16, 255–271
C L I M AT E I M PA C T S O N T H E W E S T E R N A M A Z O N E C O S Y S T E M
261
concentration should be deemed appropriate for the
purpose of our study.
Tests for trend analysis
To detect trends in climate and carbon cycling, we used
the Mann–Kendall trend test (Kendall, 1938). Compared
with the linear regression trend test, the Mann–Kendall
trend test has the following advantages. First, the
Mann–Kendall trend test does not require the assumption of normality in the variance. Perturbation events to
the climate system such as the Mt. Pinatubo eruption or
ENSO events could invalidate the normal distribution
assumption. Second, because the Mann–Kendall trend
test relies on rank-based statistics, an anomalous outlier
on the edge of a time-series data record does not have a
strong effect on the test result.
The slope of the trends was calculated from the
Kendall–Theil robust line (Helsel & Hirsch, 1992). The
Kendall–Theil robust line is defined as the line whose
slope is the median of all possible pair-wise slopes of
the data, and is on the median of both variables. Hence,
the Kendall–Theil robust line is also relatively insensitive to outliers on the edge.
Results and discussion
NDVI trend
The regional average of the annual mean NDVI for the
study region shows a significant positive trend
(P-valueo0.05) (Fig. 2a). The large drop in annual mean
NDVI observed in 1991 was due to the eruption of
Mt. Pinatubo. A comparison of monthly average NDVI
between two different periods (1982–1990 and 1993–
2003) indicates that the positive trend in annual mean
NDVI was due mainly to decadal increases in NDVI
between the months of August and December, with
November and December showing especially strong
significant positive trends (P-valueo0.05) (Fig. 2b).
The biggest limitation of optical remote sensing analysis in the humid tropics is cloud contamination,
especially for data with moderately coarse resolution
such as AVHRR. Although we made monthly composite NDVI data to minimize the cloud contamination
effect, residual contamination may still be present and
could result in artificial trends. The saturation of NDVI
over dense forests also makes trend analysis difficult.
For example, during the months of June and July,
monthly NDVI values are close to 0.8, suggesting that
observed increases in NDVI are difficult to interpret
(Fig. 2b). However, a consistent shift in seasonal cycle is
apparent between the two decades (Fig. 2b), with strong
increases in NDVI during the dry season.
Fig. 2 (a) Interannual variability in the yearly mean Normalized Difference Vegetation Index (NDVI) for the study region
from 1982 to 2003. (b) Monthly average of NDVI from 1982 to
1990 (broken line) and from 1993 to 2003 (solid line). Values at
each time step are the average value of all grid cells inside the
study region.
Comparison of climate datasets
Temperature trends. Positive trends in temperature are
present in the interpolated datasets (CRU and UD) but
not in NCEP (Fig. 3). Victoria et al. (1998) also reported a
positive trend result in the Amazon basin since the
1970s, and found that temperature in the Amazon
basin has increased at a rate similar to the overall
warming rate for the southern hemisphere. The
average monthly temperature values for the
interpolated datasets (CRU and UD) are very close,
but higher than NCEP by approximately 1.5 1C. For
the study period from 1984 to 2002, only CRU showed
a statistically meaningful positive trend of 0.051 yr 1
Published 2009
This article is a US Government work and is in the public domain in the USA, Global Change Biology, 16, 255–271
262 H . H A S H I M O T O et al.
Fig. 3 Interannual variability in the yearly average temperature
for each climate dataset from 1984 to 2002 (CRU, NCEP, and UD).
Values at each time step are the average value of all grid cells
inside the study region.
Fig. 4 Interannual variability in the yearly average precipitation for each climate dataset 1984–2002 [CRU, NCEP, UD, and
Global Precipitation Climatology Project (GPCP)]. Values at each
time step are the average value of all grid cells inside the study
region.
Table 3
dataset
There are many publications that discuss trends in
precipitation in the Amazon basin. For example, the
northern Amazon basin has shown a negative trend in
precipitation since 1950, but a positive trend has been
observed in the southern Amazon basin (Marengo,
2004). Malhi & Wright (2004) investigated the CRU
precipitation dataset for the existence of trends in
rainfall in the pan-tropics, but found no significant
trend in the Amazon from 1960 to 1998. Results of
those publications are highly dependent on the
datasets, study period, and study region used, and are
complicated by the heterogeneity of precipitation in the
Amazon.
The trend in the yearly average of each climate
Parameter
Dataset
1
Temperature (deg. yr )
Precipitation (mm day
1
yr 1)
Shortwave radiation (W m
Cloud cover (% yr 1)
2
yr 1)
CRU
UD
NCEP
CRU
GPCP
NCEP
UD
ISCCP
NCEP
ISCCP
NCEP
Trend
0.05*
0.04
0.01
0.03
0.06*
0.10*
0.02
0.48*
0.48*
0.36*
0.13
*Indicates that the P-value of the Mann–Kendall trend test was
o0.05.
(Table 3). Malhi & Wright (2004) also found a positive
temperature trend in the CRU dataset for the period
1976–1998 in the central Amazon, which includes our
study region.
Precipitation trends. The precipitation trend for the study
period (1984–2002) was significantly positive in GPCP
(0.06 mm day 1 yr 1), but no significant trend was
detected in the CRU and UD datasets (Table 3). A
significant negative trend was detected in NCEP
precipitation data ( 0.10 mm day 1 yr 1). NCEP data
clearly deviate from the other three datasets (CRU,
UD, and GPCP) until 1992 (Fig. 4). Such a discrepancy
in NCEP precipitation data was also reported in
Marengo et al. (2008).
Shortwave radiation and cloud cover trend. For the study
period, ISCCP and NCEP showed similar significant
positive trends (0.48 W m 2 yr 1) in shortwave radiation
(Table 3), although the magnitudes of the monthly
averages are quite different from each other (Fig. 5a).
Consistent with the increasing trends in shortwave radiation, both ISCCP and NCEP cloud cover
data show significantly negative trends (ISCCP:
0.36% yr 1; NCEP: 0.13% yr 1). Cloud cover in the
two datasets also has very different monthly average
magnitudes (Fig. 6).
The ERBE and CERES data also showed increasing
trends in downward shortwave radiation in the 1980s
and 1990s (Fig. 5b). ERBE and CERES data represent the
radiation flux at the Top of the Atmosphere (TOA)
rather than at the surface. However, since the amount
of shortwave radiation absorbed by the atmosphere is
small compared with the amount absorbed by the land
surface, the positive trends found in the ERBE and
Published 2009
This article is a US Government work and is in the public domain in the USA, Global Change Biology, 16, 255–271
C L I M AT E I M PA C T S O N T H E W E S T E R N A M A Z O N E C O S Y S T E M
263
Fig. 6 Interannual variability in the yearly average cloud cover
for each climate dataset from 1984 to 2002 [NCEP and International Satellite Cloud Climatology Project (ISCCP)]. Values at
each time step are the average value of all grid cells inside the
study region.
Fig. 5 (a) Interannual variability in the yearly average shortwave radiation at the surface for each climate dataset from 1984
to 2002 [NCEP and International Satellite Cloud Climatology
Project (ISCCP]. (b) Interannual variability in the yearly average
downward shortwave radiation at the top of the atmosphere
from Earth Radiation Budget Experiment (ERBE) (solid line) and
Clouds and the Earth’s Radiant Energy System (CERES) (broken
line). Values at each time step are the average value of all grid
cells inside the study region.
CERES datasets also imply an increase in absorbed
shortwave radiation at the surface over the two
decades analyzed in this study.
Globally, while solar dimming has been observed
through the 1980s (Stanhill & Cohen, 2001), solar
radiation has been shown to be increasing since the
late 1980s (Wild et al., 2005). In the tropics, the
increasing trend in incoming shortwave radiation was
reported from satellite observations and attributed to
decadal time-scale strengthening of the tropical Hadley
and Walker circulations (Chen et al., 2002; Wielicki et al.,
2002). In addition to wide scale climate change, changes
in the regional-scale hydrologic cycles (shallow
cumulus cloud and precipitation patterns) were
reported as a result of deforestation (Chagnon et al.,
2004; Chagnon & Bras, 2005). These observations
strengthen the conclusion that the shortwave radiation
Fig. 7 Monthly mean net primary production (NPP) for each
model (CASA, Biome-BGC, and LPJ) from 1984 to 2002. Values
shown are the average value of all grid cells inside the study region.
budget for the Amazon basin is changing, as a
consequence of climate change and/or land use change.
Modeling experiments
Seasonal variation in simulated monthly NPP. For all three
models, Fig. 7 shows that the minimum mean monthly
NPP occurs in July, as does the minimum in solar
radiation (Fig. 1a). On the other hand, the month in
which the peak in mean monthly NPP occurs differs
among the three models, with the peak occurring in
October, January, and March for CASA, Biome-BGC,
Published 2009
This article is a US Government work and is in the public domain in the USA, Global Change Biology, 16, 255–271
264 H . H A S H I M O T O et al.
Fig. 8 Net primary production (NPP) time series for each model experiment for (a) changing CO2 concentrations, (b) changing
temperature, (c) changing precipitation, and (d) changing shortwave radiation or cloud cover. The solid, broken, dashed, and dash-dot
lines represent the different datasets listed under each panel. The unit for the vertical axes is g C m 2 yr 1. Note that we did not
implement the CO2 experiment for CASA because of its simplicity, and thus there is no result shown for CASA in (a), the changing CO2
experiment panel. Values at each time step are the average value of all grid cells inside the study region.
and LPJ, respectively. Overall, the seasonal cycle of NPP
from all three models peaks in the dry season, with
minimum values occurring during the wet season. This
pattern indicates that water availability did not limit
monthly NPP during the dry season, but shortwave
radiation did limit monthly average NPP during the
wet season. Thus, shortwave radiation appears to be the
primary driver of seasonal patterns in NPP. These
simulated seasonal cycles are similar to the observed
patterns in GPP at the Tapajós site (Hutyra et al., 2007).
Ichii et al. (2007) explained nonwater limitations using
Biome-BGC with a deep rooting soil depth (45 m).
Other than seasonal cycles of shortwave radiation, leaf
production in the dry season (Hutyra et al., 2007) and
increases in diffusive radiation caused by aerosols
resulting from biomass burning (Oliveira et al., 2007)
should also contribute to high monthly NPP in the dry
season, but none of the models used in this study
accounted for them.
Elevated CO2 concentration experiments. The modeling
experiments to examine the effect of increased CO2
concentrations using both Biome-BGC and LPJ show a
monotonous positive trend in NPP that corresponds
with the monotonous increase in atmospheric CO2
concentrations (Fig. 8a). The rate of increase in annual
NPP is 4.66 (g C m 2 yr 1) yr 1 for Biome-BGC and
2.67 (g C m 2 yr 1) yr 1 for LPJ, and is statistically
Published 2009
This article is a US Government work and is in the public domain in the USA, Global Change Biology, 16, 255–271
C L I M AT E I M PA C T S O N T H E W E S T E R N A M A Z O N E C O S Y S T E M
Table 4 The slope of the trend [(g C m 2 yr 1) yr 1] in the
yearly NPP calculated from each ecosystem model and each
climate dataset
Data set
CASA
CO2
Temperature
CRU
0.54*
NCEP
0.24*
UD
0.47
Precipitation
CRU
0.59
NCEP
1.20*
UD
0.98
GPCP
0.74
Shortwave radiation or cloud cover
NCEP
3.17*
ISCCP
2.18*
BGC
LPJ
4.66*
2.67*
4.01
0.17
1.18
3.49*
0.30
2.75
0.99
0.10
1.93*
1.12
0.88
2.81*
0.59
1.30
1.42
0.71
2.34
6.35*
*Indicates that the P-value of the Mann–Kendall trend test was
o0.05.
significant per the Mann–Kendall trend test (Table 4).
Note that the CO2 concentration experiment was not
implemented for CASA, although given the linear
relationship between CO2 concentrations and
photosynthesis in the model, one would expect the
results from CASA to show a linear increase in NPP
that tracks the increase in the CO2 concentration data.
Temperature sensitivity experiments. All models simulate
higher NPP when using temperature from NCEP rather
than from CRU and UD (Fig. 8b), primarily because the
lower air temperature in the NCEP data (Fig. 3) reduced
the amount of autotrophic respiration. Interannual
fluctuations in autotrophic respiration are a function
of air temperature and drive the interannual variation
in NPP simulated from Biome-BGC and LPJ (Table 2). In
contrast, CASA calculates NPP directly, as opposed to
via GPP, resulting in smaller interannual fluctuations in
NPP than for the Biome-BGC and LPJ models. In addition, the air temperature in the study region during the
period analyzed displayed small interannual and
seasonal variability, remaining close to the optimum
temperature for photosynthesis defined in the CASA
model (CASA defines the optimum temperature for
photosynthesis as the temperature recorded when
NDVI reaches the annual maximum). The higher
variability of temperature data in the UD dataset (Fig.
3) does not have a large impact on NPP variability for
all three models, because of reduced sensitivity to
temperature in the models. The variability in NPP is
similar in each model for all three of the temperature
datasets used in the modeling analysis.
265
Precipitation sensitivity experiments. The year-to-year
variations in NPP as a function of the precipitation
dataset are dependent on the model used, with the
variations in NPP simulated from LPJ being higher
than those from CASA and Biome-BGC (Fig. 8c).
Within each model, the interannual variation in NPP is
very similar for all of the precipitation datasets, except for
NCEP, which was an outlier in the CASA and BiomeBGC simulations. The trends are very different among
the models and climate datasets (Table 4). Significant
negative trends in annual NPP were found when using
the NCEP-CASA [ 1.20 (g C m 2 yr 1) yr 1] and NCEPLPJ [ 2.81 (g C m 2 yr 1) y 1], which responded to the
decreasing trend observed in the NCEP precipitation data
( 0.10 mm day 1 yr 1). On the other hand, the NPP
modeled by Biome-BGC showed a significant positive
trend only when using the UD precipitation dataset
[1.93 (g C m 2 yr 1) yr 1].
Shortwave radiation or cloud cover experiments. All modeling experiments showed positive trends in annual
NPP driven by increases in shortwave radiation
and/or declines in cloud cover (Table 4). A significant
positive trend in NPP was calculated in the simulations from NCEP-CASA [3.17 (g C m 2 yr 1) yr 1],
ISCCP-CASA [2.18 (g C m 2 yr 1) yr 1], and ISCCP-LPJ
[6.35 (g C m 2 yr 1) yr 1] (Fig. 8d). Biome-BGC was less
sensitive to shortwave radiation than CASA and LPJ.
Therefore, the increasing trend in shortwave radiation
was the only climate factor which clearly and consistently
drove the increasing trend in annual NPP.
Monthly trends of model outputs and climate datasets
The trends in monthly NPP, as calculated by the three
models, are summarized in Fig. 9. The increases in atmospheric CO2 concentration recorded over the study period had a positive impact on the monthly NPP trend for
all seasons in both Biome-BGC and LPJ (Fig. 9a). Thus,
the increasing trend in CO2 concentrations cannot explain
the positive trends in NDVI observed in specific months.
The temperature experiments with CRU data showed
the strongest temperature-related impact on NPP for all
three models. All the temperature experiments showed
small negative trends for the months from July to
October (Fig. 9b). So, none of temperature experiments
can explain the positive trend in NDVI observed during
these months.
Similar to the precipitation effects on annual NPP
(Fig. 8), the impacts of precipitation on the monthly
NPP trends varied considerably, depending on the
input dataset, even when using the same model for
the simulation (Fig. 9c). Among these experiments, only
the Biome-BGC experiment driven by UD precipitation
Published 2009
This article is a US Government work and is in the public domain in the USA, Global Change Biology, 16, 255–271
266 H . H A S H I M O T O et al.
Fig. 9 Trends in monthly net primary production (NPP) (in g C m 2 month 1 yr 1) for the three models (CASA, Biome-BGC, and LPJ) for the
period from 1984 to 2002 for each climate variable [(a) CO2, (b) temperature, (c) precipitation, (d) shortwave radiation or cloud cover]. The names
of the dataset used in each experiment are listed under each panel. Values shown are the average value of all grid cells inside the study region.
data, which showed a weak positive trend in monthly
NPP from September to February, could be a candidate
to explain the observed increases in NDVI.
The radiation/cloud cover experiments simulated
strong positive trends in NPP from July to December
with CASA and LPJ driven by the NCEP and ISCCP
datasets, and with Biome-BGC driven by the NCEP
(Fig. 9d). These results suggest that the increasing trend
in shortwave radiation is the most likely driver of the
positive NDVI trends observed from August to December, as opposed to changes in CO2 concentrations,
temperature, or precipitation.
Sensitivity test for model response to changes in climate
factors
Increasing shortwave radiation from NCEP and ISCCP
caused an increase in NPP from July to December in the
CASA and LPJ simulations, while only the NCEP
shortwave radiation data resulted in a positive
trend in the Biome-BGC simulations (Fig. 9d).
Based on the Biome-BGC simulations, the UD precipitation data could also explain the positive NDVI
trends from October to December (Fig. 9c). These differences in the effects of climate on simulated NPP
originated from the varying model sensitivities to the
climate factors.
The responses to variations in temperature were very
different among models (Fig. 10a). The NPP simulated
by CASA peaked around the optimum temperature for
photosynthesis of 26 1C. The NPP modeled by BiomeBGC also peaked at 26 1C, due to the difference in the
rate of increase in photosynthesis vs. autotrophic respiration as a function of higher temperatures. LPJ, in
contrast, showed a monotonous decrease in response to
increasing temperature because photosynthesis in-
Published 2009
This article is a US Government work and is in the public domain in the USA, Global Change Biology, 16, 255–271
C L I M AT E I M PA C T S O N T H E W E S T E R N A M A Z O N E C O S Y S T E M
267
Fig. 11 Sensitivity of yearly average of photosynthesis regulation functions from three models (CASA, Biome-BGC, and LPJ).
Values shown are the average value of all grid cells inside the
study region.
Fig. 10 Model sensitivity test of net primary production (NPP)
for three models (CASA, Biome-BGC, and LPJ) by changing
climate, (a) temperature, (b) precipitation, and (c) shortwave
radiation. Values shown are the average value of all grid cells
inside the study region.
creased more slowly compared with autotrophic respiration.
The models were most sensitive to variations in
precipitation (Fig. 10b). CASA NPP decreased gradually in response to lower precipitation. On the other
hand, Biome-BGC and LPJ showed steep drops below a
critical level of annual precipitation, with Biome-BGC
dropping at a precipitation level of 1800 mm or less, and
LPJ dropping at a level of 1000 mm or less, with the
difference due to the photosynthesis submodels within
these two models. All of the models have multiplier
functions in their photosynthesis submodels that limit
photosynthesis in response to lack of water, which is
represented by soil moisture deficit. Figure 11 shows the
dependency of such photosynthesis regulation functions on soil moisture levels, which in turn are driven
by precipitation levels such as those shown in Fig. 10b.
As compared with NPP, the photosynthesis regulation
function in Biome-BGC is more sensitive to precipitation, and regulates photosynthesis strongly when annual precipitation is below a critical threshold of
1300 mm. This critical annual precipitation threshold
regulating photosynthesis depends on the soil parameterization, indicating that that proper parameterization
of the rooting depth defined in Biome-BGC is very
important (Ichii et al., 2005). The average yearly amount
of evapotranspiration simulated for the study region
was 1486 mm in CASA, 2173 mm in Biome-BGC, and
823 mm in LPJ, when using CRU temperature, GPCP
precipitation, and ISCCP shortwave radiation. The
average yearly precipitation in the study region was
2369 mm in GPCP. Since observations at Manaus indicate that 54% of rainfall (1279 mm) was lost by evapotranspiration from September 1995 to August 1996
(Malhi et al., 2002), it is likely that the evapotranspiration calculated by Biome-BGC was too high, suggesting
that our Biome-BGC experiments adjusted for Tapajós
flux data are still too sensitive to precipitation for the
Published 2009
This article is a US Government work and is in the public domain in the USA, Global Change Biology, 16, 255–271
268 H . H A S H I M O T O et al.
western Amazon basin, resulting in a higher critical
precipitation point for Biome-BGC compared with the
other models. Gerten et al. (2004) showed good agreement between observed and LPJ-modeled annual runoff, but the seasonal variation of runoff from LPJ was
not consistent with the observed data. The peak in LPJmodeled runoff was in the wet season, while the
observed peak occurred during the dry season. For
the same reason described by Ichii et al. (2007), the
default soil depth in LPJ of 1.5 m was too shallow and
needed to be adjusted.
Hydraulic redistribution is another mechanism that
regulates soil moisture sensitivity in the Amazon forest
(Lee et al., 2005), explaining why satellite observations
did not show a clear impact on vegetation condition as a
result of the severe drought recorded in 2005 (Saleska
et al., 2003), when the rainfall recorded in the Solimões
river basin was 100 mm month 1 lower than normal
from May to September. However, none of the three
models provides a mechanism for hydraulic redistribution. This is another reason that the critical annual
precipitation amount should be set at o1500 mm yr 1
in the western Amazon.
Annual NPP increases linearly in response to increases in shortwave radiation both in Biome-BGC
and LPJ (Fig. 10c). It has been reported that the efficiency of photosynthesis is higher under diffuse radiation than direct radiation (Gu et al., 2003). But, using
model simulations, Alton et al. (2007) reported that
enhancement of LUE cannot compensate the decrease
of total amount of shortwave radiation for dense canopy forests. Thus, the conclusion that NPP increase were
caused by the increase of shortwave radiation cannot be
affected by considering fraction of diffusive irradiance
to total shortwave radiation. Furthermore, diffuse irradiance caused by the aerosols from biomass burning on
cloud free days (Oliveira et al., 2007) can support a
linearly increasing response in NPP to shortwave radiation. Compare with cloud scattering, aerosol scattering
make smaller change in the total amount of shortwave
radiation because of the characteristic of more forward
scattering (Alton et al., 2007).
CO2 fertilization effect vs. shortwave radiation
Nemani et al. (2003) explained an observed positive
trend in NPP over the Amazon region with a corresponding increase in shortwave radiation. Lewis et al.
(2004b) suggested that the biomass increase could be
due to the CO2 fertilization effect and/or an increase in
shortwave radiation, because the biomass increase was
observed everywhere at the continental scale, while
other climate variables did not show spatially consistent
trends at such a wide scale. Our results from the Biome-
BGC and LPJ simulations suggest that shortwave radiation, rather than the CO2 fertilization effect, increases in
temperature, or changes in precipitation, explain the
seasonal positive trends in NDVI observed from August
to December over the study period. In the model
simulations, the increase in NPP resulting from the
CO2 fertilization effect was higher than the radiationdriven increase. The increase in NPP caused by the CO2
fertilization effect can be overestimated because a
strong dependency of light use efficiency upon temperature is hypothesized in Farquhar models (Hickler
et al., 2008). In the Free Air CO2 Enrichment (FACE)
experiment, CO2 fertilization did not increase LAI,
although NPP increased as a result of enhanced light
use efficiency (Norby et al., 2005). While the FACE
experiment was not conducted in the tropics, it demonstrated that it is possible that increases in NPP resulting
from CO2 fertilization cannot be detected by NDVI.
Biome-BGC and LPJ did not consider such an allometric
change corresponding to elevated CO2 concentrations,
and it is possible that the modeled increase in NPP
resulting from CO2 fertilization did not show up in the
monthly trend in AVHRR NDVI because additional
carbon was allocated to nonleaf biomass. Therefore,
our results do not exclude the possibility of a CO2
fertilization effect on tropical rainforests in the Amazon.
However, some issues still remain that both BiomeBGC and LPJ cannot resolve. For example, both models
cannot explain some of the other effects mediated by
elevated atmospheric CO2 concentration, such as mineral nutrient availability, growth of lianas, and biodiversity effects, which are also critical for predicting
ecosystem response in a CO2-rich world (Körner,
2004). Furthermore, other nutrients (e.g. nitrogen, potassium, and phosphorus) can limit the growth of
vegetation in response to the CO2 fertilization effect,
complicating the full capture of nutrient cycles by
ecosystem models (Hungate et al., 2003). These factors
should be considered for future research and modeling
in the Amazon region.
Conclusions
In this study, we analyzed the trends in photosynthetic
activity in the western Amazon over the last two
decades to explain the observed NDVI trend through
the use of modeling experiments driven by multiple
climate datasets. The positive trend in GIMMS NDVI
was found to be primarily influenced by increases in
NDVI from August to December. We analyzed the
trends of climate variables from multiple datasets for
the period from 1984 to 2002, finding the trend in each
climatic parameter to vary depending on the dataset.
Among the temperature datasets, a significant positive
Published 2009
This article is a US Government work and is in the public domain in the USA, Global Change Biology, 16, 255–271
C L I M AT E I M PA C T S O N T H E W E S T E R N A M A Z O N E C O S Y S T E M
trend was found in the CRU dataset. Among the precipitation datasets, the GPCP dataset showed a significant positive trend and NCEP showed a significant
negative trend, while no significant trend could be
found in the CRU and UD datasets. Shortwave radiation data showed increasing trends, while cloud cover
showed negative trends in both the ISCCP and NCEP
datasets.
We then used the different climate datasets to drive
three ecosystem models (CASA, Biome-BGC, and LPJ)
to simulate the trends in GIMMS NDVI by modeling
NPP. CO2 fertilization effects were evenly distributed
over the course of an entire year, but NDVI did not
show such an overall increase. Our results do not
exclude the possibility that potential gains in productivity resulting from CO2 fertilization effects were not
likely distributed to leaf production; however, we suggest that changes in climate rather than CO2 fertilization
effects could explain the increasing trend in NDVI.
Specifically, a positive trend in shortwave radiation
and negative trend in cloud cover most strongly explain
the corresponding increase in NDVI, as our simulations
showed that these factors drove a simulated increase in
NPP for the same months (from August to December) in
which the increases in NDVI have been observed.
Acknowledgement
This research was funded by grants from NASA Earth Sciences
program.
References
Adler RF, Huffman GJ, Chang A et al. (2003) The version-2
Global Precipitation Climatology Project (GPCP) monthly
precipitation analysis (1979-present). Journal of Hydrometeorology, 4, 1147–1167.
Alton PB, North PR, Los SO (2007) The impact of diffuse sunlight
on canopy light-use efficiency, gross photosynthetic product
and net ecosystem exchange in three forest biomes. Global
Change Biology, 13, 776–787.
Asner GP, Knapp DE, Broadbent EN, Oliveira PJC, Keller M,
Silva JN (2005) Selective logging in the Brazilian Amazon.
Science, 310, 480–482.
Baker TR, Phillips OL, Malhi Y et al. (2004) Increasing biomass in
Amazonian forest plots. Philosophical Transactions of the Royal
Society, Series B, 359, 353–365.
Ball JT, Woodrow IE, Berry JA (1987) A model predicting
stomatal conductance and its contribution to the control of
photosynthesis under different environmental conditions. In:
Progress in Photosynthesis Research, Vol. 4 (ed. Biggins J),
pp. 221–224. Martinus Nijhoff, Dordrecht.
Barkstrom BR (1984) The Earth Radiation Budget Experiment
(ERBE). Bulletin of the American Meteorological Society, 65, 1170–
1185.
269
Carswell FE, Costa AL, Palheta M et al. (2002) Seasonality in CO2
and H2O flux at an eastern Amazonian rain forest. Journal of
Geophysical Research, 107, 8076.
Chagnon FJF, Bras RL (2005) Contemporary climate change in
the Amazon. Geophysical Research Letters, 32, L13703.
Chagnon FJF, Bras RL, Wang J (2004) Climatic shift in patterns of
shallow clouds over the Amazon. Geophysical Research Letters,
31, L24212.
Chen J, Carlson BE, Del Genio AD (2002) Evidence for strengthening of the tropical general circulation in the 1990s. Science,
295, 838–841.
Cramer W, Bondeau A, Schaphoff S, Luchet W, Smith B, Sitch S
(2004) Tropical forests and the global carbon cycle: impacts of
atmospheric carbon dioxide, climate change and rate of deforestation. Philosophical Transactions of the Royal Society, Series B,
359, 331–343.
Cramer W, Kichlighter DW, Bondeau A et al. (1999) Comparing
global models of terrestrial Net Primary Productivity
(NPP): overview and key results. Global Change Biology, 5,
1–15.
De Marsily G (1986) Quantitative Hydrogeology: Groundwater
Hydrology for Engineers. Academic Press, New York.
Evan AT, Heidinger AK, Vimont DJ (2007) Arguments against a
physical long-term trend in global ISCCP cloud amounts.
Geophysical Research Letters, 34, L04701.
Farquhar GD, von Caemmerer S, Berry JA (1980) A biochemical
model of photosynthetic CO2 assimilation in leaves of C3
species. Planta, 149, 78–90.
Fekete BM, Vörösmarty CJ, Roads JO, Willmott CJ (2004) Uncertainties in precipitation and their impacts on runoff estimates. Journal of Climate, 17, 294–304.
Gerten D, Schaphoff S, Haberlandt U, Lucht W, Sitch S (2004)
Terrestrial vegetation and water balance – hydrological evaluation of a dynamic global vegetation model. Journal of
Hydrology, 286, 249–270.
Grace J, Lloyd J, McIntyre J et al. (1995) Carbon dioxide uptake by
an undisturbed tropical rain forest in southwest Amazonia,
1992 to 1993. Science, 270, 778–780.
Gu L, Baldocchi DD, Wofsy SC, Munger JW, Michalsky JJ,
Urbanski SP, Boden TA (2003) Response of a deciduous forest
to the Mount Pinatubo eruption: enhanced photosynthesis.
Science, 299, 2035–2038.
Gurney KR, Law RM, Denning AS et al. (2002) Towards robust
regional estimates of CO2 sources and sinks using atmospheric
transport models. Nature, 415, 626–630.
Haxeltine A, Prentice IC (1996) A general model for the light-use
efficiency of primary production. Functional Ecology, 10,
551–561.
Helsel DR, Hirsch RM (1992) Statistical Methods in Water Resources. Elsevier, Amsterdam.
Hicke JA (2005) NCEP and GISS solar radiation data sets available for ecosystem modeling: description, differences, and
impacts on net primary production. Global Biogeochemical
Cycles, 19, GB2006.
Hickler T, Smith B, Prentice IC, Mjöfors K, Miller P, Arneth A,
Sykes MT (2008) CO2 fertilization in temperate FACE experiments not representative of boreal and tropical forests. Global
Change Biology, 14, 1531–1542.
Published 2009
This article is a US Government work and is in the public domain in the USA, Global Change Biology, 16, 255–271
270 H . H A S H I M O T O et al.
Houghton RA, Skole DL, Nobre CA, Hackler JL, Lawrence KT,
Chomentowski WH (2000) Annual fluxes of carbon from
deforestation and regrowth in the Brazilian Amazon. Nature,
403, 301–304.
Huete AR, Didan K, Shimabukuro YE et al. (2006) Amazon
rainforests green-up with sunlight in dry season. Geophysical
Research Letters, 33, L06405.
Hungate BA, Dukes JS, Shaw MR, Luo Y, Field CB (2003)
Nitrogen and climate change. Science, 302, 1512–1513.
Hutyra LR, Munger JW, Saleska SR et al. (2007) Seasonal controls
on the exchange of carbon and water in an Amazonian rain
forest. Journal of Geophysical Research, 112, G03008.
Ichii K, Hashimoto H, Nemani R, White M (2005) Modeling the
interannual variability and trends in gross and net primary
productivity of tropical forests from 1982 to 1999. Global and
Planetary Change, 48, 274–286.
Ichii K, Hashimoto H, White MA et al. (2007) Constraining
rooting depths in tropical rainforests using satellite data and
ecosystem modeling for accurate simulation of gross primary
production seasonality. Global Change Biology, 13, 67–77.
Jarvis PJ (1976) The interception of the variations in leaf water
potential and stomatal conductance found in canopies in the
field. Philosophical transactions of the Royal Society of London.
Series B, 273, 593–610.
Kalnay E, Kanamitsu M, Kistler R et al. (1996) The NCEP/NCAR
40-year reanalysis project. Bulletin of the American Meteorological Society, 77, 437–471.
Kendall MG (1938) A new measure of rank correlation. Biometrika, 30, 81–93.
Kobayashi H, Dye DG (2005) Atmospheric conditions for monitoring the long-term vegetation dynamics in the Amazon
using normalized difference vegetation index. Remote Sensing
of Environment, 97, 519–525.
Körner C (2004) Through enhanced tree dynamics carbon dioxide enrichment may cause tropical forests to lose carbon.
Philosophical Transactions of the Royal Society, Series B, 359,
493–498.
Lee J-E, Oliveira RS, Dawson TE, Fung I (2005) Root functioning
modifies seasonal climate. Proceedings of the National Academy
of Science of the United States of America, 102, 17576–17581.
Legates DR, Willmott CJ (1990a) Mean seasonal and spatial
variability in gauge-corrected, global precipitation. International Journal of Climatology, 10, 111–127.
Legates DR, Willmott CJ (1990b) Mean seasonal and spatial
variability in global surface air temperature. Theoretical and
Applied Climatology, 41, 11–21.
Lewis SL, Malhi Y, Phillips OL (2004a) Fingerprinting the impacts of global change on tropical forests. Philosophical Transactions of the Royal Society of London. Series B: Biological Sciences,
359, 437–462.
Lewis SL, Phillips OL, Baker TR et al. (2004b) Concerted changes
in tropical forest structure and dynamics: evidence from 50
South American long-term plots. Philosophical Transactions of
the Royal Society, Series B, 359, 421–436.
Malhi Y (2008) Climate change, deforestation, and the fate of the
Amazon. Science, 319, 169–172.
Malhi Y, Nobre AD, Grace J, Kruijt B, Pereira MGP, Culf A,
Scott S (1998) Carbon dioxide transfer over a central
Amazonian rain forest. Journal of Geophysical Research, 103,
31593–31612.
Malhi Y, Pegoraro E, Nobre AD, Pereira MGP, Grace J, Culf AD,
Clement R (2002) Energy and water dynamics of a central
Amazonian rain forest. Journal of Geophysical Research, 107,
8061.
Malhi Y, Wright J (2004) Spatial patterns and recent trends in the
climate of tropical rainforest regions. Philosophical Transactions
of the Royal Society, Series B, 359, 311–329.
Marengo JA (2004) Interdecadal variability and trends of rainfall
across the Amazon basin. Theoretical and Applied Climatology,
78, 79–96.
Marengo JA, Nobre CA, Tomasella J et al. (2008) The drought of
Amazonia in 2005. Journal of Climate, 21, 495–516.
Mitchell TD, Jones PD (2005) An improved method of constructing a database of monthly climate observations and associated
high-resolution grids. International Journal of Climatology, 25,
693–712.
Monteith JL (1972) Solar radiation and productivity in tropical
ecosystems. Journal of Applied Ecology, 9, 747–766.
Myneni RB, Yang W, Nemani RR et al. (2007) Large seasonal
swings in leaf area of Amazon rainforests. Proceedings of the
National Academy of Sciences, 104, 4820–4823.
Nemani RR, Keeling CD, Hashimoto H et al. (2003) Climate
driven increase in terrestrial net primary production from 1982
to 1999. Science, 300, 1560–1563.
Nepstad DC, de Carvalho CR, Davidson EA et al. (1994) The role
of deep roots in the hydrological and carbon cycles of Amazonian forests and pastures. Nature, 42, 61–65.
Norby RJ, DeLucia EH, Gielen B et al. (2005) Forest response to
elevated CO2 is conserved across a broad range of productivity. Proceedings of the National Academy of Sciences of the United
States of America, 102, 18052–18056.
Oliveira PHF, Artaxo P, Pires C et al. (2007) The effects of biomass
burning aerosols and clouds on the CO2 flux in Amazonia.
Tellus B, 59, 338–349.
Ometto JPHB, Nobre AD, Rocha HR, Artaxo P, Martinelli LA
(2005) Amazonia and the modern carbon cycle: lessons
learned. Oecologia, 143, 483–500.
Phillips OL, Baker TR, Arroyo L et al. (2004) Pattern and process
in Amazon tree turnover, 1976–2001. Philosophical Transactions
of the Royal Society, Series B, 359, 381–407.
Pinzón JE, Brown ME, Tucker CD (2005) EMD correction of
orbital drift artifacts in satellite data stream. In: The HilbertHuang Transform and Its Applications (eds Huang NE, Shen
SSP), pp. 167–186. World Scientific Publishers, Singapore.
Potter CS, Randerson JT, Field CB, Matson PA, Vitousek PM,
Mooney HA, Klooster SA (1993) Terrestrial ecosystem production: a process model based on global satellite and surface
data. Global Biogeochemical Cycles, 7, 811–841.
Salati E, Vose PB (1984) Amazon basin: a system in equilibrium.
Science, 225, 129–138.
Saleska SR, Miller SD, Matross DM et al. (2003) Carbon in
Amazon forests: unexpected seasonal fluxes and disturbance-induced losses. Science, 302, 1554–1557.
Schloss AL, Kicklighter DW, Kaduk J et al. (1999) Comparing
global models of terrestrial Net Primary Productivity
(NPP): comparison of NPP to climate and the Normalized
Published 2009
This article is a US Government work and is in the public domain in the USA, Global Change Biology, 16, 255–271
C L I M AT E I M PA C T S O N T H E W E S T E R N A M A Z O N E C O S Y S T E M
Difference Vegetation Index (NDVI). Global Change Biology, 5,
25–34.
Sitch S, Smith B, Prentice IC et al. (2003) Evaluation of ecosystem
dynamics, plant geography and terrestrial carbon cycling in
the LPJ dynamic vegetation model. Global Change Biology, 9,
161–185.
Stanhill G, Cohen S (2001) Global dimming: a review of the
evidence for a widespread and significant reduction in global
radiation with discussion of its probable causes and possible
agricultural consequences. Agricultural and Forest Meteorology,
107, 255–278.
Thornton PE, Law BE, Gholz HL et al. (2002) Modeling and
measuring the effects of disturbance history and climate on
carbon and water budgets in evergreen needleleaf forests.
Agricultural and Forest Meteorology, 113, 185–222.
Tian H, Melillo JM, Kicklighter DW, McGuire AD, Helfrich JVK
III, Moore B III, Vörösmarty CJ (1998) Effect of interannual
climate variability on carbon storage in Amazonian ecosystems. Nature, 396, 664–667.
Tucker CJ, Pinzon JE, Brown ME et al. (2005) An extended
AVHRR 8-km NDVI dataset compatible with MODIS and
SPOT Vegetation NDVI data. International Journal of Remote
Sensing, 26, 4485–4498.
Victoria RL, Martinelli LA, Moraes JM et al. (1998) Surface air temperature variations in the Amazon region
271
and its borders during this century. Journal of Climate, 11,
1105–1110.
Wielicki BA, Barkstrom BR, Harrison EF, Lee RB, Louis Smith G,
Cooper JE (1996) Clouds and the Earth’s Radiant Energy
System (CERES): an earth observing system experiment. Bulletin of the American Meteorological Society, 77, 853–856.
Wielicki BA, Wong T, Allan RP et al. (2002) Evidence for large
decadal variability in the tropical mean radiative energy
budget. Science, 295, 841–844.
Wild M, Gilgen H, Roesch A et al. (2005) From dimming to
brightening: decadal changes in solar radiation at earth’s
surface. Science, 308, 847–850.
Willmott CJ, Matsuura K (2001) Terrestrial air temperature and
precipitation: Monthly and annual time series (1950–1999) version
1.02. Available at http://climate.geog.udel.edu/ climate/
html_pages/README.ghcn_ts2.html (access date: 04/02/2009).
Zhang Y, Rossow WB, Lacis AA, Oinas V, Mishchenko MI (2004)
Calculation of radiative fluxes from the surface to top of
atmosphere based on ISCCP and other global data sets:
refinements of the radiative transfer model and the input data.
Journal of Geophysical Research, 109, D19105.
Zhao M, Running SW, Nemani RR (2006) Sensitivity of Moderate
Resolution Imaging Spectroradiometer (MODIS) terrestrial
primary production to the accuracy of meteorological reanalyses. Journal of Geophysical Research, 111, G01002.
Published 2009
This article is a US Government work and is in the public domain in the USA, Global Change Biology, 16, 255–271