Climatic variability, carbon exchange and vegetation vulnerability in Amazonia

1
Climatic variability, carbon exchange and
vegetation vulnerability in Amazonia Lucy R.
Hutyra1, J.William Munger1, Scott R. Saleska2,
Plinio B. de Camargo3, Steven C. Wofsy1 1Dept.
of Earth Planetary Sciences, Harvard
University 2University of Arizona 3CENA/USP,
Piracicaba, SP
Seasonality of Carbon and Water exchange in
Amazonia
SCIENTIFIC GOAL The stability of Amazonian
equatorial forests, and the fate of their immense
stores of organic carbon, depend on the ecosystem
response to climate and weather. This study
presents 4 years of eddy covariance measurements
of carbon and water fluxes and their response to
environmental conditions in an Amazonian
old-growth tropical forest.
METHODS The study site is located in the Tapajós
National Forest, Para, Brazil (TNF, Figures 4).
Eddy-flux measurements of CO2 and H2O were made
at a height of 57.8 m using a sonic anemometer
(CSAT-3) and a closed-path infrared gas analyzer
(LI-6262). Net ecosystem exchange of CO2 (NEE)
was calculated as the sum of CO2 flux above the
canopy and CO2 storage flux. Ecosystem
respiration (R) was determined based on nighttime
NEE measurements during well-mixed periods. Gross
ecosystem exchange (GEE) was calculated as the
difference between NEE and R during daytime
periods.
X
CONSTRAINING ECOSYSTEM RESPIRATION
ESTIMATES Biases in the day/night measurements
of CO2 flux can affect estimates of carbon
exchange due to the prevalence of weak vertical
mixing during the nighttime hours. To constrain
our R estimates we used three independent
approaches (a) u filtering to correct for
underestimation of nighttime fluxes (b) seasonal
light response relationships between PAR and NEE
(c) Radon-222 (Rn) derived nighttime NEE
estimates. We expected that the respiration
should be largely independent of the turbulence,
nevertheless, measured NEE decreased in calm
conditions suggesting that there was lost flux.
Approximately 57 of the nighttime hours at this
site were calm, with ult0.22 m s-1. We corrected
for lost flux by filtering calm night periods and
replacing the data with the mean value of nearby
well mixed time periods. We examined the
NEE-light relationship (Figure 5) using a
nonlinear least squares approximation (hyperbolic
function) (3) fitted NEE and PAR binned by 1
µmol m-2 s-1. The intercept, a1, of this overall
model provides an independent estimate of the
mean ecosystem R and agrees very well with the
mean nighttime, u filtered, NEE measurements,
9.38 and 8.58 µmol m-2 s-1, respectively. Martins
et al. 2004 independently assess raw and u
corrected NEE measurements by comparing them to
CO2 exchange derived from Rn canopy
concentration, Rn soil flux measurement, and
profile Rn concentrations. Rn-derived nighttime
NEE was found to be 9.000.99 µmol m-2 s-1 for
the wet season (June-July 2001) and 6.390.59
µmol m-2 s-1 in the dry season (November-December
2001) and agrees very well with u filtered NEE
measurements during the same period (8.651.07
and 6.560.73, respectively) Martins et al.,
2004.
Figure 4. We measured CO2 and H2O fluxes and
profile concentration a 64 m tall eddy flux tower.
Climatic variability and vegetation vulnerability
in Amazonia Hutyra et. al., 2005
SCIENTIFIC GOAL We assessed the vulnerability
and resilience of Amazonian vegetation to climate
change by analyzing observed climate-vegetation
relationships in a statistical framework using
climate data, observed vegetation distributions,
evapotranspiration rates based on eddy flux data
(ET), and water balances.
METHODS Data for water fluxes and temperature,
from January 2002 through November 2004, were
combined to develop a model of actual ET for
evergreen Amazonian tropical forest, denoted
forest evapotranspiration (FET) FET (mm day-1)
-6.7084 0.3764T (1) where
T is monthly mean temperature (C). When fit to 38
months of environmental measurements, equation
(1) explained 68of the total variance (Figure
1). We used the Climate Research Units (CRU)
100-year gridded (0.5o x 0.5o) time series for
temperature and precipitation Mitchell et al.,
2003 to model the FET across the Amazon (Figure
2). To derive a measure of drought occurrence, we
computed the quantity of soil water available to
trees (Plant Available Water, or PAW units mm
H2O), PAWi PAWi-1 Pi - FETi
(2) where i indexes the month of the 100 year
record. Values exceeding PAWmax were assumed lost
as runoff. The spatial distribution of PAWmax was
adapted from Kleidon 2004, who applied inverse
methods to a land surface model optimizing
photosynthesis. This PAWmax applies to current
vegetation assemblages, under current climate. A
drought was assessed at any grid cell where PAW
declined to less than 75 of PAWmax for 5 or more
months in a year, implying a dry period exceeding
6 or 7 months (Figure 3a). The spatial
distribution of estimated drought frequencies in
100 years was compared with vegetation in the
legal Brazilian Amazon, classified using Landsat
data from the early 1980s (prior to most forest
clearing, figure 3b).
Figure 5. NEE as a function of PAR,. The vertical
line denotes 0 µmol m-2 s-1 PAR. The horizontal
line is the mean nighttime NEE (8.58 µmol m-2
s-1, u0.22).

• SEASONAL CONTROLS ON LATENT HEAT, GEE, R, AND NEE
• Peak litterfall rates were observed August and
  September Rice et al., 2004 and leaf flush
  across the Basin
• occurs in (August-October) Figure 6,
  Huete et al., 2006. To quantity the phenology
  effects on GEE, we
• calculated the mean monthly GEE at a fixed
  light level, 800 75 µmol m-2 s-1, and compared
  the time series
• with that of leaf litterfall and the
  remotely sensed vegetation greenness parameter
  EVI at the TNF Huete et al.,
• 2006. Leaf litterfall rates explained
  76 of the observed variance in monthly GEE. EVI
  also correlated with
• GEE, explaining 41 of the observed
  variance, when lagged by 2 months. The lagged
  correlation in EVI is not
• surprising since it takes time for the
  leaves to fully elongate. The mean GEE, across
  all light levels, also
• correlated well with litterfall and EVI,
  explaining about 40 of the observed variance,
  but by looking at a
• narrower light window we can remove the
  variance due to seasonal difference in sunlight.
• This forest does not show signs of water
  limitation on growth (Figure 8)
• On short (hourly to daily) time scales, there was
  no significant relation between temperature and
  ecosystem
• respiration. On monthly time scales, respiration
  could be well predicted by temperature and
  precipitation patterns (figure 9)
• Annual carbon balance, particularly ecosystem
  respiration, is very sensitive to weather
  conditions during the
• dry to wet seasonal transition (Figures
  8,9)

Figure 1. (a) Observed and modeled forest ET for
the Santarem study site, R2 0.68 (b) time
series for measured forest ET, the FET model (1),
and potential evapotranspiration (PET)
Thornwaite, 1948.
RESULTS Our values for drought frequency (Figure
3a) are highest along the southern and eastern
edges of the legal Amazon, but less frequent
droughts occurred in the central basin. Areas
with high drought frequency are associated with
regional precipitation minima and/or high
temperature variability. The current distribution
of vegetation (Figure 3b) strikingly follows
drought frequency, with savanna replacing forest
and transitional vegetation as drought
frequencies increase (Figure 3c). Our study
supports the view that forests in areas of high
drought frequency (gt45 drought probability)
could shift to transition forests or savanna, if
aridity increases as predicted by climate change
models Cox et al., 2004. Potentially at risk
are over 600,000 km2 of forest (Figure 3), gt11
of the total area. Our maps show that increased
aridity may lead to bisection of Amazonian
equatorial forests.
Figure 6. Monthly mean GEE at PAR of 800 75
µmol m-2 s-1, open circles. Monthly mean leaf
litterfall rate, July 2000 May 2005, closed
circles. Monthly mean Enhanced Vegetation Index
(EVI), 2000-2005, triangles Huete et al. 2006.
Latitude
Latitude
-20 -10 0
10
-20 -10 0
10
Figure 7. 5-day mean time series for (a) latent
heat flux(b) sensible heat flux (c) net
radiation (d) precipitation dry season
indicated by blue shading. The annual mean
fraction of water lost through LE and the
precipitation inputs was approximately 0.53
(1116mm/2111mm), 0.64 (1114/1740), 0.49
(1137/2311), 0.51 (1123/2201) for 2002-2005,
respectively. The dry season LE was insensitive
to dry season precipitation and nearly constant
across years even as dry season precipitation
varied by 40. During the dry season the ratios
of evaporation to precipitation were 1.81
(503mm/279mm), 1.16 (522/448), 1.28 (514/402),
1.40 (536/383), respectively.
Figure 8. The annual carbon balance at km 67 has
shown a mean net loss of 939 kg C ha-1 yr-1
(observed range of -221 (uptake) to 2677 (loss)
kg C ha-1 yr-1). During the wet season and early
dry season, R dominates GEE and the ecosystem is
a net carbon source. The dominance switches by
September when R becomes moisture limited.
Overall, GEE maximizes in the middle of the dry
season. There may be a trend of increasing carbon
uptake. Annual ecosystem C losses were
decreasing between 2002-2004, but a weather
anomaly in late 2005 resulted in net carbon loss.
Figure 9. Time series for 4 years of monthly
mean GEE and R. There is a drop off in GEE
around May, before the start of the dry season,
as the ecosystem begins leaf senescence. Leaf
litterfall peaks in Aug. and Sept. when mean GEE
is at its minimum. The new flush of leafs begins
to emerge around September at the same time as
GEE rates begin to increase. High aerosol levels
increase the diffuse light and photosynthetic
efficiency, aerosol levels are highest between
September and October. Respiration decreases in
the dry season due to moisture limitation and
shows the greatest variability at the dry-to-wet
seasonal transition.
-70 -60 -50
-40
-70 -60 -50
-40
Longitude
Longitude
Figure 3. (a) Observed drought frequency (
years) (b) distribution of savanna, transitional
vegetation, and forest across the legal Amazon
(c) land area (km2) of vegetation types with
given drought frequency (), forest land area is
multiplied by 0.1 for scaling.
IMPLICATIONS/CONCLUSIONS Forest areas with high
climate variability are vulnerable to loss of
forest with either increased mean temperature, or
increased variability in temperature and/or
precipitation. Our analysis provides a physical
quantity (PAW deficit) to predict vegetation type
indicating that the seasonality of soil moisture
is a critical factor determining forest-savanna
boundaries. The critical links between fire,
climate, and land use are highly uncertain in
current coupled climate-vegetation models. In
order to assess vegetation vulnerability to
climate change, models must capture variability
of climate, the non-linear, hysteretic behavior
of vegetation response to rising drought
frequency, the synergistic effect of forest
fragmentation and development, and the occurrence
of landscape-changing fires.
IMPLICATIONS/CONCLUSIONS Contrary to
expectations, this forest does not show signs of
seasonal water limitation on growth despite a
5-month dry season. CO2 uptake responds primarily
to light on hourly time scales, but
photosynthesis overall maximizes in the middle of
the dry season, responding to ecophysiological
(flushing of new leaves) and atmospheric (high
aerosol loading) changes. Leaf phenology is the
major control on photosynthesis, but EVI lags the
phenological response by 2 months and explains
only 41 of the observed monthly variance. Annual
carbon balance was very sensitive to weather
anomalies, particularly the timing of the
dry-to-wet seasonal transition, with mean net
loss of 939 kg C ha-1 yr-1 (observed range of
-221 (uptake) to 2677 (loss) kg C ha-1 yr-1).
The climatic sensitivity has significant
implications for Amazonian carbon balances on
annual to decadal time scales.