<%BANNER%>

Effects of Episodic Droughts and Fire on the Carbon Cycle of Amazonian Vegetation

Permanent Link: http://ufdc.ufl.edu/UFE0041595/00001

Material Information

Title: Effects of Episodic Droughts and Fire on the Carbon Cycle of Amazonian Vegetation Field Research and Modeling of a Near-Term Forest Dieback
Physical Description: 1 online resource (133 p.)
Language: english
Creator: Brando, Paulo
Publisher: University of Florida
Place of Publication: Gainesville, Fla.
Publication Date: 2010

Subjects

Subjects / Keywords: amazon, bark, cambium, carbon, climate, conservation, drought, evi, fire, modis, mortality, phenology, tree, tropical
Interdisciplinary Ecology -- Dissertations, Academic -- UF
Genre: Interdisciplinary Ecology thesis, Ph.D.
bibliography   ( marcgt )
theses   ( marcgt )
government publication (state, provincial, terriorial, dependent)   ( marcgt )
born-digital   ( sobekcm )
Electronic Thesis or Dissertation

Notes

Abstract: Climate change may cause decreased rainfall in Amazonia and move forests to large-scale dieback events. To inform debates about this issue, I evaluated the impacts of drought and fire on Amazonian vegetation. Using a throughfall exclusion experiment, I found that wood production was the most sensitive component of above-ground net primary productivity to drought and that aboveground carbon stocks declined by 32.5 Mg ha-1 over 6 years of the treatment. These results indicate that severe droughts can substantially reduce forest C stocks in the Amazon. Based on a process-based model (CARLUC) that incorporates results from that partial throughfall exclusion experiment, I showed that total C stocks tended to increase between 1995-2004 (dry season) over the Amazon, corresponding to increases in photosynthetic active radiation. These results could partially explain the increases in biomass observed in the field over decades in the Amazon. To clarify how the gross productivity of Amazonian is influenced by drought, I used improved MODIS-enhanced vegetation index measurements (EVI) to show that in densely forested areas of the Amazon, no climatic variable tested contributed to explaining the high EVI inter-annual variability. Instead, I found evidences that suggest that production of new leaves could play an important role in the inter-annual EVI variability, but not necessarily in changes in vegetation productivity. These findings suggest that it still is unclear whether EVI can be employed as a tool to measure the vulnerability of tropical forests to drought. Finally, I studied the mechanisms associated with fire-induced tree mortality in a tropical forests in the southern Amazon. I found that the presence of water in the bark reduced the differential in temperature between the fire and vascular cambium due to boiling water in the bark. Also, I show that experimental fires influenced mortality through several processes unrelated to cambium insulation, suggesting that several plant traits that reduce the negative and indirect effects of fire should also be considered in the assessments of tropical forests' vulnerability to fire. In conclusion, I found evidence that droughts and fires exert strong influences on the terrestrial carbon cycling, but no evidences for imminent regime shifts in the region driven by climate change alone.
General Note: In the series University of Florida Digital Collections.
General Note: Includes vita.
Bibliography: Includes bibliographical references.
Source of Description: Description based on online resource; title from PDF title page.
Source of Description: This bibliographic record is available under the Creative Commons CC0 public domain dedication. The University of Florida Libraries, as creator of this bibliographic record, has waived all rights to it worldwide under copyright law, including all related and neighboring rights, to the extent allowed by law.
Statement of Responsibility: by Paulo Brando.
Thesis: Thesis (Ph.D.)--University of Florida, 2010.
Local: Adviser: Putz, Francis E.
Electronic Access: RESTRICTED TO UF STUDENTS, STAFF, FACULTY, AND ON-CAMPUS USE UNTIL 2011-02-28

Record Information

Source Institution: UFRGP
Rights Management: Applicable rights reserved.
Classification: lcc - LD1780 2010
System ID: UFE0041595:00001

Permanent Link: http://ufdc.ufl.edu/UFE0041595/00001

Material Information

Title: Effects of Episodic Droughts and Fire on the Carbon Cycle of Amazonian Vegetation Field Research and Modeling of a Near-Term Forest Dieback
Physical Description: 1 online resource (133 p.)
Language: english
Creator: Brando, Paulo
Publisher: University of Florida
Place of Publication: Gainesville, Fla.
Publication Date: 2010

Subjects

Subjects / Keywords: amazon, bark, cambium, carbon, climate, conservation, drought, evi, fire, modis, mortality, phenology, tree, tropical
Interdisciplinary Ecology -- Dissertations, Academic -- UF
Genre: Interdisciplinary Ecology thesis, Ph.D.
bibliography   ( marcgt )
theses   ( marcgt )
government publication (state, provincial, terriorial, dependent)   ( marcgt )
born-digital   ( sobekcm )
Electronic Thesis or Dissertation

Notes

Abstract: Climate change may cause decreased rainfall in Amazonia and move forests to large-scale dieback events. To inform debates about this issue, I evaluated the impacts of drought and fire on Amazonian vegetation. Using a throughfall exclusion experiment, I found that wood production was the most sensitive component of above-ground net primary productivity to drought and that aboveground carbon stocks declined by 32.5 Mg ha-1 over 6 years of the treatment. These results indicate that severe droughts can substantially reduce forest C stocks in the Amazon. Based on a process-based model (CARLUC) that incorporates results from that partial throughfall exclusion experiment, I showed that total C stocks tended to increase between 1995-2004 (dry season) over the Amazon, corresponding to increases in photosynthetic active radiation. These results could partially explain the increases in biomass observed in the field over decades in the Amazon. To clarify how the gross productivity of Amazonian is influenced by drought, I used improved MODIS-enhanced vegetation index measurements (EVI) to show that in densely forested areas of the Amazon, no climatic variable tested contributed to explaining the high EVI inter-annual variability. Instead, I found evidences that suggest that production of new leaves could play an important role in the inter-annual EVI variability, but not necessarily in changes in vegetation productivity. These findings suggest that it still is unclear whether EVI can be employed as a tool to measure the vulnerability of tropical forests to drought. Finally, I studied the mechanisms associated with fire-induced tree mortality in a tropical forests in the southern Amazon. I found that the presence of water in the bark reduced the differential in temperature between the fire and vascular cambium due to boiling water in the bark. Also, I show that experimental fires influenced mortality through several processes unrelated to cambium insulation, suggesting that several plant traits that reduce the negative and indirect effects of fire should also be considered in the assessments of tropical forests' vulnerability to fire. In conclusion, I found evidence that droughts and fires exert strong influences on the terrestrial carbon cycling, but no evidences for imminent regime shifts in the region driven by climate change alone.
General Note: In the series University of Florida Digital Collections.
General Note: Includes vita.
Bibliography: Includes bibliographical references.
Source of Description: Description based on online resource; title from PDF title page.
Source of Description: This bibliographic record is available under the Creative Commons CC0 public domain dedication. The University of Florida Libraries, as creator of this bibliographic record, has waived all rights to it worldwide under copyright law, including all related and neighboring rights, to the extent allowed by law.
Statement of Responsibility: by Paulo Brando.
Thesis: Thesis (Ph.D.)--University of Florida, 2010.
Local: Adviser: Putz, Francis E.
Electronic Access: RESTRICTED TO UF STUDENTS, STAFF, FACULTY, AND ON-CAMPUS USE UNTIL 2011-02-28

Record Information

Source Institution: UFRGP
Rights Management: Applicable rights reserved.
Classification: lcc - LD1780 2010
System ID: UFE0041595:00001


This item has the following downloads:


Full Text





EFFECTS OF EPISODIC DROUGHTS AND FIRE ON THE CARBON CYCLE OF
AMAZONIAN VEGETATION: FIELD RESEARCH AND MODELING OF A NEAR-TERM
FOREST DIEBACK


















By

PAULO MONTEIRO BRANDO


A DISSERTATION PRESENTED TO THE GRADUATE SCHOOL
OF THE UNIVERSITY OF FLORIDA IN PARTIAL FULFILLMENT
OF THE REQUIREMENTS FOR THE DEGREE OF
DOCTOR OF PHILOSOPHY

UNIVERSITY OF FLORIDA

2010

































2010 Paulo Monteiro Brando
































To my parents, my brother, and my friends









ACKNOWLEDGMENTS

I am most thankful to my advisor for his major influence on my intellectual

development during my PhD. I also thank Daniel Nepstad for challenging every finding,

word, and result I present here. Most of all, I thank all of those that rendered this thesis

a synthesis of years of collaborative field research and modeling efforts. Among this

very special group of people from several research groups I want to name a few: Eric

Davidson, Mike Coe, David Ray, Britaldo Soares-Filho, Rafaella Silvestrini, Herman

Rodrigues, Paul Lefebvre, Paulo Moutinho, Pieter Beck, Alessandro Baccini, Oswaldo

Carvalho, Claudia Stickler, Scott Goetz, Wendy, Kingerlee, Marcos Costa, Ane Alencar,

Adilson Coelho, Osvaldo Portella, Oswaldo Carvalho, Raimundo Quintino, Val, Osmar,

Wanderley Rocha, William Riker, Darlisson Nunes, Bati, Sandro, Bibal, Marcia Macedo.

Especial thanks to Ben Bolker, Mary Christman, Leda Kobziar, Wendell Cropper, Ted

Schuur, Yadvinder Malhi, Scott Saleska, and Jennifer Balch for their various

contributions to my dissertation. I thank Allie Shenkin, Fay Ray, and all my friends in

Gainesville for their support during all these years and helping to make my experience

in the United States an enjoyable one. The Instituto de Pesquisa Ambiental da

Amazonia (IPAM) and Woods Hole Research Center (WHRC) provided essential

institutional support for my PhD.









TABLE OF CONTENTS

page

A C KNOW LEDG M ENTS ......... .................................. .......................... ............... 4

LIST OF TABLES ......... ............... ..................... ....... ............... 8

LIS T O F F IG U R E S .................................................................. 9

A BSTRACT ........... ....... ..... .......... ..... ...................... ......... 11

CHAPTER

1 INTRODUCTION ........................... .......... ......... 13

2 DROUGHT EFFECTS ON LITTERFALL, WOOD PRODUCTION, AND
BELOWGROUND CARBON CYCLING IN ANAMAZON FOREST: RESULTS
FROM A PARTIAL THROUGHFALL REDUCTION EXPERIMENT .................... 16

Introduction .................... ............ ............... 16
M methods .............. ........................... ....... ............... 18
Site Description and Treatment Design ....... ........ ......................... 18
Aboveground NPP, Tree Mortality, and Soil CO2 Efflux ........... ............... 18
S tatistica l A na lysis ........................................................................ ......... 19
Results ................ ...... ............ ............ ................... 20
Soil Water...................................... 20
Mortality and Necromass.............. ........ .................... 20
Stem Growth and W ood Production .................................. .. ............ ....... 21
Litterfall and Carbon Dynamics ............. ..... ...... ......................... 21
ANPP ................................... ............... 22
Correlates of ANPP and LAI .................................... 22
Soil CO2 Efflux ................. ........ ....... ........... 22
Discussion .......................... ........ ..... ............... 23
Response and Recovery of ANPP ....... .......................... 23
Belowground Carbon Accumulation ............. ............................... 26
Carbon Stock Accumulation and Forest Recovery .............. ............. 28

3 EVALUATING THE RELATIONSHIPS BETWEEN INTERANNUAL CLIMATIC
VARIABILITY AND ABOVEGROUND CARBON STOCKS IN AMAZONIA ............ 38

Introduction ........................... ............... 38
Methods ................. ............. .................................... 40
Model Description....................... .................... ............... 40
Input Data .......................... ............... 41
Experimental Runs ................. ................... ......... 42
Analysis ....... ... .................... ....................... 42
Results ............................. .................................... 43









CARLUC Predictions for TNF...................... ....... ................. 43
Experimental Runs: Tree Mortality and NPP Allocation ............................... 43
Specific Regions and Field Plots ....................................................... 44
Total Carbon Stocks ............... .. ........ ................. 44
C lim ate a nd C a rbo n P o o ls....................................................... ... .. ............... 4 5
D is c u s s io n .............. ..... ............ ................. ............................................. 4 5

4 SEASONAL AND INTERANNUAL VARIABILITY OF CLIMATE AND
VEGETATION INDICES ACROSS THE AMAZON.......................................... 57

Introduction ........................... ............... 57
M e th o d s ..................................................................................................... 6 0
MODIS Data ...................................................................... .. ........................... 60
EVI Surrounding Meteorological Stations .......................... ...................... 61
S ite-specific A na lysis .................................................................... ......... 6 1
NBAR-EVI and EVI-Product ..................... ........... ............... 62
C lim ate ............. ........ ... .. .................................. 63
Statistical Analysis of EVI and Climatic Variables ................ .......... ........ 63
Basin-wide (mixed vegetation types) ........... ..... .... ................ ... 63
Densely forests areas ............. ... .. ...... .................. ............ 64
H ierarchica l partitioning .................. ... .... .... .......................... 64
Relationship of the variability of EVI with PAW ................. ................ 65
R results ................ .......................................................................................... 65
Spatial-temporal Patterns of EVI ............ ........... .......... .. ............. 65
Basin-wide Climate........................................... ............... 66
EVI and Climate Variables................................................... ....... 67
Spatial analysis across the Amazon Basin............................. ...... ..... 67
Spatial-temporal analysis in densely forested areas.............. .......... 70
S ite-specific A nalysis.......................................... ................... 71
D is c u s s io n ............. ......... .. .. ......... .. .. ......... .................................... 7 2

5 POST-FIRE TREE MORTALITY IN A TROPICAL FOREST OF BRAZIL: THE
ROLES OF BARK TRAITS, TREE SIZE, AND WOOD DENSITY ...................... 83

Introduction ............................. ............... 83
M eth od s ........... .... ..... .... ...................................................... ... ............ ... 86
General Approach and Site Description ........ .... ..................... ............... 86
M odeling Cam bium Insulation ............................ .. ......... ................. 86
Linking Tree Mortality and Cambium Insulation (R)..................................... 88
M modeling Tree M ortality............. ..................... .................... .... .......... 89
Tree mortality and cambium insulation ................................................... 89
Other correlates of fire-induced tree mortality.................... ........... 91
R results ..................... ............ ................... ...... .. .... ...... .......... 91
Cam bium Insulation (R ) ................................................................ 91
Cam bium Insulation and Tree Mortality............... .................... ............... 93
Other Correlates of Tree M ortality ......... ........ ......... ......... .................. 94
D discussion ........................................................ 95


6









Cambium Insulation .................. ...... ......... ...... 95
Bark Correlates with Tree Mortality ...... .................................. 96
Forest Resistance to Fire ........................................................................ ........ 98

6 CONCLUSION ............... ......... .................. 106

METEOROLOGICAL STATIONS............... ................................. 111

VARIABILITY IN VEGETATION INDEX............................... ............... 112

EQUATIONS ON HEAT TRANSFER................................... ............... 114

F IR E EX P E R IM E N T ............................................. ...... .............. 116

F LA M E H E IG H T.................................................... .................. 117

BARK THICKNESS AND TREE DBH ................................... 118

M O D E L 1 ........ ............ ................................. ....... ........... ..... 1 19

M O D E L 2 ........... ...... ........................................ ............................ 12 0

M O D E L 3 .............. ..... .................. .... ........................................... 12 1

BIOGRAPHICAL SKETCH ........... ... .................. ......... ............... 133









LIST OF TABLES


Table page

4-1 Slopes of precipitation, photosynthetic active radiation (PAR), and vapor
pressure deficit over time. ..... ........ ....................................................... 67

4-2 ANOVA table for all predictors of EVI in non-forested areas from a linear
model (M1). Note that we ran one model for each year of the study ............... 69

4-3 ANOVA table for all predictors of EVI in dense forested areas from a general
linear model (M2).............. ............................... 70

5-1 Model comparison using corrected Akaike Information Criteria (AICc).............. 92

F-1 Model 1. Period: 2004-2005 ....... .......... ........... ..... 119

G-1 Model 2. Period: 2004-2007 ............. .......... ............. ... ..... .......... 120

H-1 Model 3. Period: 2007-2008 ............ ................... ... .... ........... 121









LIST OF FIGURES


Figure page

2-1 Soil volumetric water content (VWC) integrated from 0 to 200 cm (A) and
from 200 to 1100 cm (B) and daily precipitation (C)..................... ............... .. 30

2-2 Annual rainfall (control plot) and effective rainfall. ............. ........... ........... 31

2-3 Annual percent of mortality measured on an individual basis for all individuals
2 10 cm dbh in the exclusion and control plots........................... ... ............... 32

2-4 Annual trends in the exclusion and control plots of: A) litterfall; B) wood
production; C) above-ground net primary productivity (ANPP); and D)
necromass production. ............................................................... ............ 33

2-5 Stem growth of individuals in the exclusion and control plots by different size
c la sse s : ......................................................................................... 3 4

2-6 Annual trends in the exclusion and control plots of leaf area index (LAI).. ........ 35

2-7 Relationships between annual mean VWC measured in the upper soil profile
(0 200 cm) and A) above-ground net primary productivity (ANPP) and B)
leaf area index (LA I). ........................................................................... 36

2-8 Measurements of surface flux of carbon dioxide, 613C and A14C in the
exclusion (grey) and control plots (black).. ............. ...... ................. 37

3-1 Conceptual diagram of the CARLUC ecosystem model.............................. 50

3-2 Annual rates of leaf mortality (top panel), stem mortality (mid panel), and
NPP allocation (lower panel) as a function of PAW................. ........... ..... 51

3-3 Temporal patterns in carbon stocks predicted by CARLUC in leaves (upper
panel) and stems (lower panel) for the Tapajos National Forest ..................... 52

3-4 Temporal patterns in average aboveground C stocks in regions with low,
m edium and high levels of PAW ............... ................................................ 53

3-5 Temporal patterns in total aboveground carbon stocks averaged by dry and
wet season and per year. .................................... ...... .. .... ............... 54

3-6 Climate patterns during the dry and wet seasons......... .................... ......... 55

3-7 Total aboveground carbon stocks simulated by CARLUC for the entire
Amazon and for three regions of varying PAW ......... .. ...... .. .............. 56

4-1 Average dry-season EVI across central South America ................................. 76









4-2 Spatial-temporal patterns of dry-season EVI across the Amazon for areas
w ith high canopy cover ................................................................................ 77

4-3 Average coefficient of variation of EVI for the entire Amazon, based on data
at 500 x 500 m resolution (only for areas with canopy cover 2 70%)................ 78

4-4 Monthly climate patterns over the Amazon region based on data from 280
meteorological stations distributed across the basin. ................................... 79

4-5 Temporal patterns of environmental and biophysical correlates of EVI.............. 80

4-6 Site Specific analysis on the relationships between monthly EVI-NBAR and
monthly field measurements .......... ............ .. .. ........... .. ..... ......... 81

4-7 Site Specific analysis on the relationships between monthly EVI product and
monthly field measurements ................ ........... ......... ............ ......... 82

5-1 Map showing the spatial distribution of trees located in three 50 ha plots........ 100

5-2 Overall results from the bark heating experiment................... ......... ..... 101

5-3 Relationships between rates of change in cumulative heating (R) and A) bark
thickness, B) bark density, and C) bark moisture content.............................. 102

5-4 Survival probabilities of trees with dbh 2 10 cm dbh as a function of Rhat,........ 103

5-5 Relative influence of Rhat, tree height, dbh, and wood density on tree
mortality for different fire treatments ............... .................. ......... ........... 104

5-6 Probability of tree survival as a function of standardized tree height (left
panel), dbh (middle panel), and wood density (right panel)........... .............. 105









Abstract of Dissertation Presented to the Graduate School
of the University of Florida in Partial Fulfillment of the
Requirements for the Degree of Doctor of Philosophy

EFFECTS OF EPISODIC DROUGHTS AND FIRE ON THE CARBON CYCLE OF
AMAZONIAN VEGETATION: FIELD RESEARCH AND MODELING OF A NEAR-TERM
FOREST DIEBACK.

By

Paulo Monteiro Brando

August 2010

Chair: Francis Edward Putz
Major: Interdisciplinary Ecology

Climate change may cause decreased rainfall in Amazonia and move forests to

large-scale dieback events. To inform debates about this issue, I evaluated the impacts

of drought and fire on Amazonian vegetation. Using a throughfall exclusion experiment,

I found that wood production was the most sensitive component of above-ground net

primary productivity to drought and that aboveground carbon stocks declined by 32.5

Mg ha-1 over 6 years of the treatment. These results indicate that severe droughts can

substantially reduce forest C stocks in the Amazon.

Based on a process-based model (CARLUC) that incorporates results from that

partial throughfall exclusion experiment, I showed that total C stocks tended to increase

between 1995-2004 (dry season) over the Amazon, corresponding to increases in

photosynthetic active radiation. These results could partially explain the increases in

biomass observed in the field over decades in the Amazon.

To clarify how the gross productivity of Amazonian is influenced by drought, I used

improved MODIS-enhanced vegetation index measurements (EVI) to show that in

densely forested areas of the Amazon, no climatic variable tested contributed to









explaining the high EVI inter-annual variability. Instead, I found evidences that suggest

that production of new leaves could play an important role in the inter-annual EVI

variability, but not necessarily in changes in vegetation productivity. These findings

suggest that it still is unclear whether EVI can be employed as a tool to measure the

vulnerability of tropical forests to drought.

Finally, I studied the mechanisms associated with fire-induced tree mortality in a

tropical forests in the southern Amazon. I found that the presence of water in the bark

reduced the differential in temperature between the fire and vascular cambium due to

boiling water in the bark. Also, I show that experimental fires influenced mortality

through several processes unrelated to cambium insulation, suggesting that several

plant traits that reduce the negative and indirect effects of fire should also be considered

in the assessments of tropical forests' vulnerability to fire.

In conclusion, I found evidence that droughts and fires exert strong influences on

the terrestrial carbon cycling, but no evidences for imminent regime shifts in the region

driven by climate change alone.









CHAPTER 1
INTRODUCTION

There is an urgent need for in-depth assessments of tropical forest vulnerability to

global changes, a critical step for understanding and predicting ecological trajectories in

the near future. Currently, it is clear that human-induced climate change are likely to

cause dramatic changes in the world's tropical forests (Christensen et al. 2007). Many

climate models predict that moderate to intense climate change will lead to large-scale

dieback of forests by the end of the 21st Century (e.g., Cox et al. 2004). Large scale

deforestation and widespread forest fires could exacerbate the negative effects of

climate change on ecosystem services of Amazonian forests (e.g., water cycling)

(Nepstad et al. 2008), causing the dieback of large portions of these forests much

earlier than currently predicted.

The question of how much climate change is too much for the maintenance of

tropical forest integrity remains unanswered (Malhi et al. 2008) but in my dissertation, I

provide some of the information needed. Because global climate change coupled with

deforestation is predicted to cause reductions in rainfall over portions of the Amazon

(Malhi et al. 2008), I studied the effects of experimentally reduced rainfall on the

aboveground net productivity (i.e., how much carbon is fixed by plants) of an eastern

Amazonian forest in Chapter 2 of my dissertation. Although shortage of soil water

availability will ultimately reduce the abilities of trees to fix carbon, based on currently

available science, the thresholds at which reductions in soil water availability will cause

major changes in forest functioning are not at all clear, nor are the mechanisms

associated with disruption in forest functioning (e.g., carbon allocation to different parts

of plants, tree mortality, soil respiration).









In Chapter 3, I modify an ecosystem model to include forest responses to drought

that were observed in field experiments, but that are lacking in most (if not all) models

employed to predict future trajectories of Amazonian forests (e.g., Cramer et al. 2001,

Foley et al. 1998). For example, droughts are expected to alter allocation of C among

leaves, trunks, and roots, which, in turn, can influence forest capacity to cycle carbon or

to cope with drought-induced stress. But we know very little how this process will

influence C-cycling during periods of severe droughts in the Amazon. Also, while most

models assume that droughts will affect C assimilation only via reduced photosynthesis,

there is strong evidences that episodic, severe droughts kill trees. After evaluating how

these two processes may affect carbon C stocks in the Amazon, I evaluate the potential

responses of Amazonian vegetation to climate oscillations during 1995-2005, in an

attempt to reconcile field-based measurements (Phillips et al. 2009) and model

predictions.

Given the report that the Amazon became "greener" during one of the strongest

droughts over the last century and the consequent inferences about forest vulnerability

to drought (Saleska et al. 2007), I found that there was a need to investigate what

'greener' means in terms of forest functioning. In Chapter 4 of my dissertation, I

evaluated how a vegetation index used to measure the greening of the Amazon, derived

from satellite-based measurements, correlated with variations in vegetation dynamics

as measured by leaf area index, litterfall, vegetation phenology, and several climatic

variables. I then investigated how this vegetation index varied across the Amazon and

how these variations were associated with oscillations in climate.









The likelihood of trees being exposed to fire is increasing in the tropics due to

synergistic interactions among droughts, forest degradation and fragmentation, and

increased fire ignition (Alencar et al. 2006, Aragao et al. 2007, Nepstad et al. 2008), yet

little is known about the characteristics that determine which trees survive and which

succumb to understory fires. For example, while bark thickness appears to be the most

important plant trait in preventing fire-induced damage (Barlow et al. 2003), its

importance in preventing fire-induced tree mortality has not been widely tested for

tropical forest trees and may be modulated by several other traits. In Chapter 5 of my

dissertation, I use field experiments and model-based analyses to evaluate the

influences of bark traits, plant size, and wood density on tree mortality following

recurrent understory fires of varying intensity in a transitional forest of Mato Grosso.

While the Amazon may experience severe degradation due to synergistic

interactions between drought and fire, the extent of these degradations is still poorly

understood. Thus, the field experiments, data analyses, and model simulations run

across different scales I employ in this dissertation should help to clarify how the

Amazon region responds to drought and fire and how these responses will be

expressed in the future under increasing drought conditions.









CHAPTER 2
DROUGHT EFFECTS ON LITTERFALL, WOOD PRODUCTION, AND
BELOWGROUND CARBON CYCLING IN ANAMAZON FOREST: RESULTS FROM A
PARTIAL THROUGHFALL REDUCTION EXPERIMENT.

Introduction

The fate of approximately 86-140 Pg of carbon contained in the forests of the

Amazon (Saatchi et al. 2007) will depend, in part, upon the effect of drought episodes

on the amount and allocation of forest biomass production. Several lines of evidences

suggest that Amazon drought episodes may be more common and more severe in a

warming world (Li et al. 2006).

One of the most important ways in which severe droughts influence tropical

forest carbon stocks is through tree mortality. Droughts associated with ENSO events

elevated tree mortality in East Kalimantan from a background of 2% to 26% yr1 (Van

Nieuwstadt and Sheil 2005), in central Amazonia from 1.1% to 1.9% (Williamson et al.

2000), Panama from 2% to 3% (Condit et al. 1995), and Sarawak from 0.9% to 4.3-

6.4% (Nakagawa et al. 2000). (In one Panama forest, ENSO had no effect on tree

mortality (Condit et al. 2004)). Experimentally-induced drought in one hectare of an

east-central Amazonian forest permitted the identification of a threshold of declining

plant-available soil moisture and cumulative canopy water stress beyond which canopy

tree mortality rose from 1.5% to 9% (Nepstad et al. 2007).

Responses of net primary productivity (NPP) and its allocation during drought

may have important influences on carbon stocks, but remain poorly understood in moist

tropical forests (Houghton 2005). Seasonal and inter-annual variation in the NPP of

moist tropical forests is controlled mostly by changes in light and soil moisture (Clark

and Clark 1994). Hence, NPP increases during ENSO events, when reduced cloud









cover increases photosynthetically-active radiation without provoking soil moisture

deficits large enough to inhibit photosynthesis. When soil moisture is in short supply,

NPP may decline and its relative allocation among leaves, stems and roots may be

affected, but our understanding of these responses is poor. Stem growth is expected to

be the most sensitive component of NPP to drought because it is low on the carbon

allocation hierarchy (Chapin et al. 1990).

The response of root production to drought is less clear, in part because of

substantial methodological challenge of mensuration (Trumbore et al. 2006). Soil

respiration, which integrates C02 production from all belowground sources, including

root respiration and leaf litter decomposition, tends to be lower in the dry season

compared to the wet season (Davidson et al. 2004, 2000, Saleska et al. 2003); this is

probably due largely to seasonal variation of heterotrophic respiration in the litter layer.

Responses to long-term drought treatments may also include changes in root

production.

A seven-year, partial throughfall exclusion experiment was conducted in east-

central Amazonia to examine forest responses to a 35-41% effective reduction in annual

rainfall. I quantified forest LAI litterfall, wood production, soil respiration, and the 613C

and A14C composition of C02 emitted from the soil in one-hectare treatment and control

plots to test the following predictions: wood production is the most responsive

component of aboveground NPP to drought, declining rapidly with the onset of drought

stress; litterfall increases initially through drought-induced leaf shedding, then declines

to below non-treatment levels as LAI stabilizes at a lower level; and, soil C02 efflux is









affected by responses of multiple processes of heterotrophic and autotrophic

respiration, which could have either accumulating or canceling effects.

Methods

Site Description and Treatment Design

The experiment was carried out in the Tapaj6s National Forest, Para, Brazil

(2.8970S,54.9520W). Annual precipitation ranges from 1700-3000 mm and averages ~

2000 mm, with a 6 mo dry season (Jul-Dec) when rainfall rarely exceeds 100 mm mo1.

Mean annual temperature is 280 C (min 220C, max 280C). The soil is deeply weathered

Haplustox. The depth to water table at a similar site 12 km away was >100 m, which

means that the stocks of soil water availability to trees is recharged by rainfall.

The throughfall exclusion experiment was carried out in two structurally and

floristically similar one-ha plots: an 'exclusion plot' (treatment) and a 'control plot'. In the

exclusion plot, approximately 34-40% of annual incoming precipitation was diverted

from the soil (70% of throughfall) during the 6-mo wet season from 2000 to 2004.

Details of the experimental design can be found in Nepstad et al. (2002). Please see

(Hurlbert 1984) for more information on large-scale unreplicated experiments.

Soil water to 11-m depth was measured monthly in five deep soil shafts per plot

between 1999 and 2005 using time domain reflectometry (TDR; (Jipp et al. 1998)).

Here I present volumetric water content (VWC) for two depth intervals: 0-200 cm (upper

soil profile); and 200-1100 cm (deep soil profile).

Aboveground NPP, Tree Mortality, and Soil CO2 Efflux

Dendrometer bands were installed on trees 2 10 cm dbh (diameter at 1.3 m or

above buttresses) and vines > 5 cm dbh in the exclusion (342 individuals) and control

plots (409 individuals). Diameter increment was monitored monthly between 2000 and









2005 using an electronic caliper from which relative stem diameter growth was

calculated. Wood production at the plot-level was calculated by summing biomass

estimates for trees and lianas determined using the allometric equations from

(Chambers et al. 2001) and (Gerwing and Farias 2000), respectively. I also calculated

individual annual stem growth averaged for small (10 to 20 cm dbh (trees) and 5 to 20

cm dbh (lianas)) and large (>20 cm dbh) stems. Ingrowth stems were added as they

exceeded the lower dbh threshold. Mortality of stems > 10 cm dbh in both plots was

assessed on an individual basis monthly between 1999 and 2005 and is described in

(Nepstad et al. 2007). Litterfall was collected bi-weekly from 1999 to 2005 from ~ 64

mesh traps (0.5 m2) on a systematic grid in the understory of both plots. Leaf area index

(LAI) was estimated indirectly monthly at these same points using two LiCor 2000 Plant

Canopy Analyzers (LI-COR, 1992) operating in differential mode. Above-ground NPP

was calculated as the sum of wood production and litterfall biomass in each year of the

study (2000 to 2005) according to the methodology of Clark et al. (2001). A manual

dynamic chamber system was used to measure soil CO2 efflux on 18 chambers per

treatment per date, as described by Davidson et al. (2004). Carbon isotope

measurements were determined for soil respiration on 5 occasions from 2000 to 2005.

Statistical Analysis

I used a repeated measures design to test for year-to-year differences between

plots for stem growth (using initial dbh as a covariate), LAI, litterfall, and soil CO2 efflux.

Changes in mortality were assessed using a generalized linear model with binomial

errors. The relationships among, VWC, ANPP, and LAI, were tested using linear

regressions.









Results


Soil Water

Prior initiation of treatment, VWC was 25 and 62 mm lower in the exclusion plot

than the control plot for the upper (0-200 cm) and deeper (200-1100 cm) soil profiles,

respectively (Figs. 2-1, 2-2). After three periods of partial throughfall exclusion (2002),

VWC was 97 and 272 mm lower in the exclusion plot than the control plot in the upper

and deeper soil profiles, respectively. The difference in VWC between plots did not

continue to increase in 2003 for two reasons. First, natural precipitation was well below

normal that year (Fig. 2-2), so that soil water was not fully recharged in the control plot.

Second, drought-induced mortality in 2003 (see Mortality and Necromass section

below) may have reduced transpiration in the exclusion plot.

Mortality and Necromass

From 2000 to 2005, the annual mortality rate of individuals 2 10 cm dbh was

significantly higher in the exclusion plot (5.7% yr-1) than in the control plot (2.4% yr-);

2001 was an exception to this general trend (Fig. 2-3A). The most dramatic treatment

effects were observed in 2002 and 2003, when the annual rate of mortality in the

exclusion plot reached 7.2% yr1 and 9.5% yr-1, respectively, as reported in Nepstad et

al. 2007. Mortality was also significantly higher in the exclusion (7.2 yr-) than the control

plot (3.1% yr-) after the termination of the experimental throughfall exclusion, in 2005.

The spike in treatment induced mortality in 2002 and 2003 led to a 25%

reduction in above-ground live biomass (AGLB) in the exclusion plot relative to the

control plot (Fig. 2-3B), a difference which remained stable after the exclusion treatment

was terminated in August 2004 and through the end of 2005. From 2000 through 2005,









47 Mg ha-1 more necromass [dead biomass] was produced in the exclusion compared

to the control plot (Fig. 2-4D).

Stem Growth and Wood Production

The treatment caused a reduction in relative stem diameter growth of large

individuals (P<0.001) yet had no effect on smaller stems (Fig. 2-5A,B,C), which were

growing less in the exclusion plot than the control plot before the treatment was applied

(P=0.322; Fig. 2-5B). Later, following the pulse in mortality of large trees in 2003, I

observed increased stem growth in small individuals in the exclusion plot relative to the

control plot (P<0.001). During this same period (2004 to 2005), stem growth was

declining in the control plot (Fig. 2-5A,B,C), perhaps as a lagged response to low annual

precipitation in 2003 (Figs. 2-1 C, 2-2).

Total wood production during the first exclusion period (2000) was 13% lower in

the exclusion (4.87 Mg ha-1 yr-1) than in the control plot (5.61 Mg ha-1 yr-1). This

difference increased over the treatment period to 46%, 63%, and 58% in 2001, 2002,

and 2003, respectively (Fig. 2-4B). After the drought treatment was removed in 2005,

the difference in biomass production between plots declined to 23% due to an increase

in individual tree growth (mostly small trees) in the exclusion plot and a decrease in

growth in the control plot.

Litterfall and Carbon Dynamics

The drought treatment also reduced LAI over time (Fig. 2-6A; P<0.001).

Differences between plots in LAI were small during the two first years of the study,

followed by a larger drop in 2002 and remaining 21% to 26% lower than the control-plot

through 2005.









Litterfall, which was higher in the exclusion plot than the control prior to the

initiation of the treatment, declined in the exclusion plot compared to the control.

However, it was lower in the exclusion plot than the control plot only in 2003 (23%,

P<0.001; Fig. 2-4A). During the last year of the treatment (2004) differences in litterfall

between plots were reduced to 10%, and, after the drought stress was removed (2005),

litterfall was 2% higher in the exclusion plot than the control plot, indicating a relative

increase in litterfall in the exclusion plot compared to the control (P<0.001).

ANPP

ANPP progressively declined in the exclusion plot relative to the control from 2000

to 2003: 12%, 30%, and 41% lower in 2001, 2002, and 2003, respectively (Fig. 2-4C). In

the last year of the treatment (2004) and in 2005, when the exclusion treatment was

removed, ANPP differences between plots were diminished to 30% and 10% of the

control, respectively. The average reduction in ANPP from 2000 to 2005 was 21%,

corresponding to a cumulative difference between plots of 17 Mg ha1.

Correlates of ANPP and LAI

Annual variations in VWC in the upper soil profile (0-200 cm) explained 51% of

the variation in ANPP over the 6-yr study period (Fig. 2-7A). Annual average LAI

showed a strong positive linear relationship with VWC (R2: 92%; P<0.001; Fig. 2-7B),

and with ANPP (R2: 49%; P=0.006; In(ANPP) = -0.9995 + 0.8362 In(LAI)).

Soil C02 Efflux

The effect of throughfall exclusion treatment on soil C02 efflux was not consistent

among years or seasons. Although there was a significant treatment-by-year interaction

(p < 0.001), the treatment-by-year-by-season interaction was not significant (p = 0.65).

When large pulses were measured, which were probably related with recent wetting









events, they tended to be higher in the control plot (Fig. 2-8A). The exclusion plot had

somewhat higher C02 efflux rates on most of the other dates during the exclusion

treatment years. These differences largely cancelled, resulting in nearly identical

estimates of average annual C02 efflux from the two plots during the 5-year exclusion

period (12.8 1.0 Mg C ha-1 yr1 in the exclusion plot and 12.8 1.3 Mg C ha- yr1 in the

control plot).

The 13C signature of the soil C02 efflux revealed a divergence between treatments

in 2004, when the 13C02 increased (became less negative) in the exclusion plot (Fig. 2-

8B). A similar difference in foliar 613C had been observed for some species in 2002 (J.

Ometto and J. Ehleringer, pers. comm.). I also observed a similar difference in 513C02

measured within the soil profile in 2003 (not shown), which would reflect allocation of

fractionated C substrates to root respiration.

There were no differences between treatment plots in 14C02 signatures on any

date (Fig. 2-8C), which suggests that the relative age of respired C substrates was not

altered by the exclusion treatment. These radiocarbon values are 10-25 %o greater than

values for C02 in ambient air sampled at the same time. This indicates that the C being

respired is a mixture of modern C substrates with substrates that are 5-40 years old and

are more enriched in 14C due to the signal of atmospheric testing of nuclear weapons in

the 1950s and 1960s (Trumbore et al. 2006).

Discussion

Response and Recovery of ANPP

The most striking result of this study was the differential response of wood

production and litterfall to throughfall exclusion. Wood production in the exclusion plot

was 13 to 62% below that of the control plot during each of the five treatment years,









while litterfall was substantially lower than the control plot (23%) only during the year of

lowest rainfall, in 2003. The allocation of ANPP to wood and litterfall was therefore

highly sensitive to soil moisture availability. At the beginning of the experiment (2000)

wood and litterfall comprised 40 and 60%, respectively, of the total ANPP, while in the

third year of the exclusion (2002), they were 29 and 71% of the ANPP, respectively.

This finding provides an important exception to the widely-held assumption of drought-

insensitive allocation of ANPP between litterfall and wood, as represented in many

ecosystem models (e.g., Hirsch et al. 2004, Potter et al. 2001, Tian et al. 1998). These

results also run contrary to the positive response of LAI to short-term drought that has

been postulated for Amazon forests based upon remotely-sensed (MODIS) estimates of

LAI (Huete et al. 2006, Myneni et al. 2007, Saleska et al. 2007). If such a positive

response is real and widespread, it may be overridden as drought severity grows.

However, it is important to remember that our throughfall exclusion treatment did not

simulate the higher radiation that is typically associated with lower rainfall.

Litterfall and wood production recovered rapidly following cessation of the drought.

Despite the death of 9% of all trees 210 cm dbh in 2003, litterfall recovered fully during

the first post-treatment year (2005) and wood production, which was 42% of the control

plot in 2003, climbed to 77% that of the control plot in 2005. This rapid recovery is

partially explained by the positive growth responses observed for residual, surviving

trees following the death of neighboring individuals, perhaps attributable to reduced

competition for light and soil moisture resources. LAI declined in the exclusion plot in

2003 and 2004, increasing light availability in the understory; the difference in VWC

between exclusion and control plots also diminished, in part because of reduced









transpiration associated with tree mortality A second litterfall recovery mechanism was

the establishment of fast-growing trees in the exclusion plot understory (P. Brando, D.

Nepstad, unpublished data). Increased soil nutrient availability associated with tree

mortality may have also contributed to the accelerated growth of surviving trees and the

establishment of fast-growing trees.

The rapid recovery of litterfall in 2004 and after the treatment ended in 2005 was,

surprisingly, not accompanied by a recovery of LAI. The relationship between LAI and

litterfall was similarly unexpected throughout the experiment since litterfall remained

high, with a small decline in 2003, despite the sharp, sustained decline of LAI beginning

in the third year of the treatment (2002). LAI declined in response to drought with little

effect on litterfall either through a lower specific leaf area or through a shortening of leaf

longevity or both.

Drought-induced tree mortality and reductions in wood production resulted in a 63

Mg ha-1 reduction in AGLB that persisted through the first post-treatment year. The

long-term recovery of pre-treatment AGLB will depend upon the balance between wood

production and tree mortality. The latter declined to the level of the control plot during

the post-treatment year, and could decline further if the drought treatment accelerated

the death of trees that were already senescent and prone to die in the near future. Post-

treatment, tree-level increases in stem diameter growth contributed to an acceleration of

wood production that, if it continued, could also reduce the time required to recovery

pre-treatment AGLB.

In sum, the experimental drought caused a rapid and drastic reduction in

aboveground wood production, while litterfall was less strongly affected. By the third









year of the exclusion treatment in 2003, a reduction in litterfall coincided with increased

mortality of large trees in the exclusion plot relative to the control plot, suggesting that

available soil water had fallen below a minimum threshold required to sustain minimal

physiological processes (Nepstad et al. 2007). Although the response of plants to soil

water depletion may have been initially delayed due to the several coping strategies

plants use to avoid or tolerate drought (Oliveira et al. 2005), increased mortality and

reductions in forest growth resulted in a 24% reduction in ANPP by the end of the

experiment, which was highly correlated with a decline in volumetric water content and

LAI.

Belowground Carbon Accumulation

Heterotrophic soil respiration can be inhibited by drought either directly through

microbial drought stress or indirectly through substrate limitation (Davidson et al. 2006),

which is consistent with observations of lower soil CO2 efflux rates during the dry

season in Amazonian forests (Davidson et al. 2000, 2004; Saleska et al. 2003). On the

other hand, it is also possible that increased allocation of C to fine roots could be an

adaptation to explore for deep water resources (Nepstad et al. 1994), which might

increase belowground inputs of C as a response to drought. The depletion of soil

moisture proceeds from the litter and upper mineral soil layers, where root systems are

concentrated, to deeper (>8 m in about 1/3 of the Amazon) soil layers (Jipp et al. 1998;

Nepstad et al. 1994). New root growth therefore occurs at progressively deeper soil

layers during drought episodes and may be inhibited near the soil surface by the lack of

available moisture. These plausible responses that differ in sign may help explain the

lack of a consistent and significant treatment effect on soil CO2 efflux. Either there were

no changes in belowground C cycling processes, or the changes that did occur









cancelled with respect to C02 production, thus resulting in nearly identical average

annual C02 efflux rates.

The isotope data shed some light on these possibilities. First, the 13C data

demonstrate that C fixed by drought-stressed leaves, which is more enriched (less

negative) in 13C than in the control plot, was transported to the soil and/or forest floor by

litterfall and/or root respiration and root inputs. Hence the heavier 13C signatures of C

respired in 2004 could reflect the decomposition of enriched-13C leaf litter produced the

1-2 years previously, a contribution of heavier 13C in root respiration reflecting drought

stress effects on photosynthetic fractionation, or both. However, the two-year lag

between fractionation of fresh foliage in 2002 and the first detection of that isotopic

signal in surface C02 efflux in 2004 suggests that decomposition of aboveground

litterfall rather than root respiration (which would reflect the isotopic signature of current

year's photosynthate) dominates the surface C02 efflux.

The 14C data demonstrate that there was no significant shift in the relative age of

respired C02 as a result of the drought treatment. In other words, the relative

contributions of respiration from young substrates (root respiration and microbial

respiration of exudates and fresh leaf and root litter) and older substrates (root, leaf, and

wood litter and soil organic matter that has persisted 5 years) did not change

appreciably. Rhizosphere respiration typically has a 14C signature that matches the

atmospheric 14C02 of the same year, so that an increase in allocation to roots in an

amount necessary to offset a reduction in C02 derived from older litter decomposition in

the exclusion plot would have reduced 14C values overall. Likewise, had there been

significant changes in decomposition rates of soil carbon stocks with higher 14C values









(leaf and root litter and soil organic matter; Trumbore et al. 2006), we should have

detected a change in the radiocarbon signature. Hence, we see little evidence of

significant changes in belowground C cycling processes induced by the drought

treatment. Responses to natural seasonal droughts, which temporarily reduce

respiration (Saleska et al. 2003), should not be confused with responses to longer-term

droughts, which may or may not alter annual rates of C inputs to and turnover times

within the soil. These results highlight the danger of extrapolating short-term metabolic

responses to longer-term ecosystem level responses that potentially involve changes in

C allocation, mortality, and growth.

Carbon Stock Accumulation and Forest Recovery

Partial throughfall exclusion induced reduction in AGLB points to the potential for

severe and intense droughts predicted for the Amazon basin to increase carbon

emissions to the atmosphere. Even mild droughts, such as those induced during the

first year of exclusion, can inhibit wood production and, hence, the long-term pattern

carbon storage in this forest ecosystem. This finding has important implications

regarding reports that many Amazon forest appear to be undergoing net carbon uptake

(Baker et al. 2004, Grace et al. 1995, Phillips et al. 1998, Phillips et al. 2009)- a finding

that has been met with some skepticism (Clark 2002, Saleska et al. 2003). Reductions

of rainfall in regions of strong seasonal drought, comprising nearly half of the Amazon's

forests (Nepstad et al. 2004), may reduce wood production, diminishing or reversing the

net uptake of carbon from these ecosystems.

The long-term effect of the 63 Mg ha-1 reduction in AGLB on aboveground carbon

stocks will depend upon the complex balance between ABLG regrowth and the rate at

which trees killed by the drought decompose. Many factors influence ABLG re-









accumulation, including tree mortality rate, wood production rate, and the wood density

of the trees that eventually replace those killed by drought stress. Forests recovering

from slash-and-burn agriculture in the Venezuelan Amazon required approximately two

centuries to recover pre-disturbance carbon stocks in part because of shifts in species

composition (Saldarriaga et al. 1988). The decline of LAI and AGLB observed in this

study also suggest that the positive feedback between drought and fire may be stronger

than previously hypothesized (Nepstad et al. 2001), which could have large long-term

implications for carbon storage. Hence, even this relatively long-term, 5-year

experimental manipulation was not long enough to investigate the full effects of climate

change on carbon stocks of Amazonian forests. Nevertheless, we have demonstrated

that multi-year drought induced substantial losses of aboveground biomass and

productivity.










* Control
Exclusion


0
0 .
a 0 .0


*

: **
** 4
0! 001 0.


E


0*
** 0
0 0
*0
.0
-
*



C






LLLM~Uii~LLL~Liil Li


2001


2002


Figure 2-1. Soil volumetric water content (VWC) integrated from 0 to 200 cm (A) and
from 200 to 1100 cm (B) and daily precipitation (C). Measurements were
taken from August 1999 to December 2004.
















E Exclusion


S.--;;I I


I l ',


Figure 2-2. Annual rainfall (control plot) and effective rainfall (annual rainfall minus water
excluded by the plastic panels during the wet season). The partial throughfall
exclusion treatment was carried during the six-month wet season each year
beginning in January 2000 and ending in August 2004.


I _


III













































0 -I--


incliiiduais, > 10 cm cibh


* Contol
Excdusion
,- ,


Figure 2-3. Upper panel: annual percent of mortality measured on an individual basis for
all individuals 2 10 cm dbh in the exclusion and control plots from December
2000 to December 2005. Lower panels: above-ground standing live biomass
(AGB Live) of all individuals 2 10 cm dbh for the exclusion and control plots
during the same period.












































_o a omass Contol Neromass- -
C


0_A





S- Bi omass- Control Necromass -
Biomass- Exclusion i] Necromass -Exciusi
I o D






2001

Figure 2-4. Annual trends in the exclusion and control plots of: A) litterfall; B) wood
production; C) above-ground net primary productivity (ANPP); and D)
necromass production. Measurements were taken from January 2000 to
December 2005. Units are Mg ha-1 yr-1. Error bars ( 1 SE) are presented
only for litterfall.













A) All individuals

Control












...1:. I. :i d: ii lOto2O0cm i' '-- i,-: 5to20cmdbh :. ....















Large Individuals: > 20 cm dbh

Control













2001 2002 2003 2004 2005
2001 2002 2003 2004 2005


0)
E
E



E




S,











C 3
T
d

m


5. Stem growth of individuals in the exclusion and control plots by different size
classes: A) > 20 cm dbh; B) between 5 (lianas) 10 (trees) and 20 cm dbh;
and C) > 5 cm dbh (lianas) and > 10 cm dbh (trees). Error bars indicate SE
(N is given by the number individuals). Measurements were taken from
December 1999 to December 2005. Annual stem growth was calculated as
the difference in dbh measured between two sampling intervals (December of
each year).


(



















Exclusion


20i-


Figure 2-6. Annual trends in the exclusion and control plots of leaf area index (LAI).

Measurements were carried out from January 2000 to December 2005. Error

bars for LAI indicate SE.


1"--






(03--






tO-


m -


0 O


-r


_P
r

N
C
r
V


"'l'l: '{'


'' ': ~.



















InY) = -5.8271 + 1.2966 In(V) Y = -1.96336 + 0.012206 VW
( "4













6.10 61.25 6.30 6.35 6.40 6.45 6.50 500 550 600 650
VWC (mm); natural-log scale VWC (mm)

Figure 2-7. Relationships between annual mean VWC measured in the upper soil profile
(0 200 cm) and A) above-ground net primary productivity (ANPP) and B)
leaf area index (LAI). Data represents annual means from January 2000 to
December 2004.
December 2004.


























- .-:.Flux'I-










......

T _
i_


2004


2005


Figure 2-8. Measurements of surface flux of carbon dioxide, 613C and A14C in the
exclusion (grey) and control plots (black). The error bars represents standard
errors of the mean.









CHAPTER 3
EVALUATING THE RELATIONSHIPS BETWEEN INTERANNUAL CLIMATIC
VARIABILITY AND ABOVEGROUND CARBON STOCKS IN AMAZONIA

Introduction

Climate change is projected to have major impacts on tropical forest metabolism

and carbon stocks (Bonan 2008). In Amazonia, increases in atmospheric CO2

concentrations may stimulate vegetation productivity (Lloyd & Farquhar, 2008), but the

drier and warmer conditions resulting from these increases may ultimately render the

region susceptible to late-century diebacks, according to predictions of climate-dynamic

vegetation global models (Cox et al. 2000, Cramer et al. 2001, Nobre et al. 1991). If

predicted diebacks do occur, a large amount of the 80-120 Gt of C stored in forests of

the Amazon will be released to the atmosphere (Betts et al. 2004, Cox et al. 2004).

Moreover, this occurrence may lead to a positive feedback loop, hastening further

climate warming and drying (Costa and Foley 2000).

The responses of Amazonian forests to drier and warmer climatic conditions

remains controversial. One outstanding issue relates to the large uncertainties about the

the intensity of drought needed to provoke widespread forest dieback across the

Amazon Basin (Malhi et al., 2009). While field-based studies have shown that episodic

droughts associated with climatic cycles are capable of causing increased mortality of

large trees in the tropics (Condit et al. 1995, Nakagawa et al. 2000, Van Nieuwstadt and

Sheil 2005, Williamson et al. 2000), recent satellite-based studies suggest that

Amazonian forests could be more drought resistant than previously thought (Saleska et

al. 2007). Indeed, there is strong evidence that forest photosynthesis (gross primary

productivity, GPP) increases at least during early stages of drought because of higher









photosynthetically active radiation, but decreases when thresholds in plant available soil

moisture cause increased tree mortality and reduced tree growth (Poulter et al. 2009).

Findings from two throughfall exclusion experiments carried out in moist tropical

forest systems of the Brazilian Amazon (Tapajos and Caxiuana National Forests) have

helped identify drought thresholds that could cause major changes in vegetation

metabolism and structure (Fisher et al. 2007, Meir et al. 2009, Nepstad et al. 2002). In

both experiments, net primary productivity (NPP), light saturated photosynthetic rates,

and leaf area index (LAI) all declined in response to drought treatment (i.e., 50% wet-

season throughfall exclusion) of 2-3 years. Following protracted water stress,

substantial increases in tree mortality rates were observed, corresponding to reductions

in plant available water (Brando et al. 2008, Nepstad et al. 2007). If the general

responses of Amazonian forests to drought are similar to those observed in the

Caixuana and Tapajos Forests, a drier future climate is likely to have major effects on

the physiology and carbon stocks.

In 2005, the southwestern Amazon experienced one of the most severe droughts

of the last century (Aragao et al. 2007, Marengo et al. 2008, Zeng et al. 2008). This

exceptionally dry and hot period, caused by an anomalously warm tropical North

Atlantic, provided an opportunity to observe forest resistance to severe drought at a

large-scale, similar to those predicted in the future. Based on results from a network of

permanent plots spread across the Amazon, Phillips et al. (2009) found that several

Amazonian forests were sensitive to the 2005 drought with mortality increasing at rates

that were high enough to reverse a trend in C accumulation of 0.45 PgC yr-1 to a -1.2-

1.6 PgC annual net emissions. Although the results of this study have received some









criticism because of the limited number of plots (Chambers et al. 2009), Gloor et al.

(2009) suggest that the RAINFOR plot network capture most of the variation in tree

mortality and, hence, on short-term, inter-annual oscillations in C stocks.

Here I evaluate the potential responses of Amazonian vegetation to drought from

1995 to 2005 using a process-based, ecosystem model, CARLUC (Carbon Land Use

Change; Hirsch et al. 2004). In particular, I estimate the effects of climate variability on

total carbon stocks and fluxes of Amazonian vegetation across the Basin and of forests

where permanent plots from RAINFOR are located (Phillips et al. 2009). Based on

previous findings of increased rates of carbon accumulation over the tropics (Baker et

al. 2004, Phillips et al. 1998), I expect CARLUC simulations to describe C accumulation

over the Amazon from 1998 to 2004, but C losses during the 2005 drought.

Given that CARLUC was parameterized using data from a rainfall exclusion

experiment representing a small area in Tapajos National Forest (TNF), it has

limitations in reproducing vegetation patterns in other regions. Despite this limitation,

simulations using CARLUC should provide insights into forest vulnerability to

oscillations in climate, given that it integrates four representative climatic variables of

NPP into a single predictor, potential carbon stocks.

Methods

Model Description

I assessed the effects of climate on carbon stocks of Amazonian forests between

1995 and 2005 using the CARLUC ecosystem model (described in detail in Hirsch et al.

2004), which borrows its basic structure from the 3-PG model (Landsberg and Waring

1997). Briefly, CARLUC is a process-based, dynamic model that estimates net primary

productivity based on characteristics of the vegetation [quantum efficiency (qe), and









carbon use efficiency (fnr)] and on climatic variables [vapor pressure deficit (VPD),

temperature (TEMP), photosynthetic active radiation (PAR), and plant available water

(PAW)] (Fig. 3-1). As the leaf area index increases, NPP saturates according to the

light- extinction function derived from Beer's law (eq. 1; note that all climatic variables

were scaled to range from 0 to 1). In the version of the model presented here, the

allocation of NPP to leaves, stems, and roots is represented as a function of PAW and

based on results from the TNF throughfall exclusion experiment (Fig. 3-2). Tree

mortality is modeled as being dependent on soil water availability: as PAW declines to

5 40%, the fraction of live biomass declines exponentially with PAW (Fig 3-2). As for 3-

PG, CARLUC was designed to integrate process-based phenomena with empirical

relationships derived from field experiments.

NPP = fnr qe-PAR TEIMP-min(PAW,VPD).- (-e07 SIAf ) (3-1)

Input Data

The model was initiated using interpolated meteorological climate data (~280

stations spread across the Amazon Basin) at a resolution of 2 km x 2 km (described in

detail in Nepstad et al. 2004). Because PAR was not available from these

meteorological stations, estimates were obtained using the GOES-8 satellite product

(Nepstad et al., 2004). The external variable PAW (i.e., independent from vegetation

dynamics) was generated based on the difference between soil field capacity (SWfc)

and wilting point (SWpwp). Briefly, texture data samples collected in the field were

combined with empirical relationships between soil texture and SWfc and SW wp.

Equations relating soil texture and carbon content to field capacity and permanent

wilting point were developed by Tomasella and Hodnett (1998) using soil water versus









matric potential curves for Amazon soils. Following, estimates of PAWmax (in mm) were

generated for each meter of the soil horizon from 0 to 10; PAWmax among horizons were

summed to estimate PAWmax for the entire sampled profile.

Experimental Runs

I performed four simulations to evaluate the potential effects of drought on the

carbon stocks of Amazonian forests.

* The mortality and NPP allocation functions were held constant by keeping PAW
fixed at 100% (see Fig. 3-2).

* The mortality and NPP allocation functions were allowed to vary spatially and
temporally with PAW.

* The mortality function was allowed to vary with PAW, while the NPP allocation
function was held fixed.

* The mortality function was held fixed according to PAW, while the NPP allocation
function was allowed to vary with PAW.

Analysis

Based on the runs of CARLUC in which mortality and NPP allocation were allowed

to vary as a function of PAW, I assessed spatial-temporal carbon stocks across the

Amazon as follows:

* First, I extracted CARLUC-predicted aboveground stocks only for the TNF and
compared these predictions with field measured data from Brando et al. (2008).

* Second, I summed all 13 carbon pools from CARLUC for each of the 132 months
of the simulations. In this analysis I evaluated intra- and inter-annual variability in
C pools (referred as "Total C stocks").

* Third, I chose three regions of to represent three different levels of dryness (i.e.,
PAW), and averaged the aboveground carbon pools for each region to evaluate
temporal patterns in C storage.

* Fourth, I evaluated the average aboveground carbon stocks of Amazonian forests
from 1995 to 2005 where field permanent plots are located (Phillips et al. 2009) to









assess how CARLUC predictions relate to field-plot measurements (referred as
"Field Plots").

Results

CARLUC Predictions for TNF

At the Tapajos National Forest, predicted average aboveground carbon stocks

(140 MgC ha1 ) were within ranges observed in the field. The temporal dynamics of

simulated C, however, diverged from on-site measurements. While predicted average

annual leaf carbon stocks (Cleaf) tended to increase from 2000 to 2005 at a rate of

0.0055 kgC ha-1 year-' ( 0.001; p=0.0038) (Fig. 3-3), Brando et al. (2008) observed

minor reductions in LAI from 2000 to 2004 (with the exception of 2002, when LAI was

higher than average, Brando et al. 2008). In 2005, however, both predicted Cleaf and

observed LAI were above the 2000-2004 average. Whereas predicted C stocks in

stems (Cstem) also tended to increase from 2002 to 2005, these increases were

relatively low when compared to total standing stocks (Fig. 3-3), which is in agreement

with field measurements of aboveground C stocks up to 2004. In 2005, however, there

was an decrease in biomass observed in the field, probably resulting from a lag-

response of vegetation to reduced plant available water, while an increase in predicted

C stocks in Cstems.

Experimental Runs: Tree Mortality and NPP Allocation

Carbon stocks predicted by CARLUC for the entire Amazon Basin indicated

vegetation sensitivity to drought as expressed by the effects of PAW on tree mortality.

Whereas CARLUC runs with PAW having no effect on mortality reached equilibrium at

108 Gt (total stocks for the entire Amazon), this equilibrium was reached at ~ 100 Gt

when PAW affected both mortality and NPP allocation. This result suggests a potential









interaction between the functions representing mortality and NPP allocation, given that

increased mortality may have reduced NPP. In contrasts, when PAW was allowed in the

model to affect either mortality or NPP allocation alone, CARLUC reached equilibria at

107 and 105 Gt, respectively.

Specific Regions and Field Plots

Given that simulated C stocks in CARLUC were sensitive to PAW (when mortality

and NPP allocation were allowed to vary as a function of PAW), drier regions of the

Amazon not only tended to reach equilibrium at lower C stocks, but also showed greater

temporal variability in C stocks than wetter regions (Fig. 3-4). For instance, monthly

analyses of C stocks in three regions of the Amazon (low, medium, and high levels of

PAW) showed that the driest region was predicted to lose carbon in 2004-2005, while

aboveground live C pools were predicted to increase in all three regions from 1995 to

2003. These predicted losses suggest that thresholds in PAW were crossed in drier

regions of the Amazon between 2004 and 2005.

CARLUC simulations of carbon stocks from regions where permanent plots are

located (e.g., Phillips et al. 2009) (Fig. 3-4) carried out on a monthly basis revealed little

change C pools over time (Figure 3-5). From 1995 to late 2003, live aboveground

biomass increased, but non-significantly, and then began to increase at low rates in late

2003.

Total Carbon Stocks

Based on temporal analyses among years and between seasons for the entire

Amazon Basin using CARLUC, I found that total wet season carbon stocks tended to

increase between 1995-1998 but to decrease subsequently, from 1999 to 2005 (Fig. 3-

5). In contrast, during the dry seasons, C stocks declined between 1995-2000, and









then increased between 2001-2004. During the major drought of 2005, there was a

minor shift in this trend of increments at the Basin scale, with an overall net reduction of

0.06 GtC year-. In sum, these contrasting wet and dry season behaviors resulted in a

net increase in C stocks between 1995 and 2004. Patterns of between season

increases and decreases in C stocks were associated with contrasting temporal

patterns in climate over the Basin. Whereas increases in dry season PAR contributed

positively to vegetation productivity after 2000, decreases in PAW (associated with

lower wet-season PPT and higher VPD) reduced productivity during the wet seasons,

particularly after 1998.

Climate and Carbon Pools

Although CARLUC indicated that PAW accounted for most variability in C stocks,

PAR had important effects in some cases as well. In regions with no soil water deficits,

higher PAR led to higher C stocks. It follows that by dividing the Amazon into the

relatively drier East and wetter West (following methodology in Malhi et al. 2009), there

was a clear difference in PAR, temperature, and VPD between these two regions

(higher in the East than the Western Amazon). In other words, where PAR was high and

PAW high enough to prevent mortality, productivity tended to be higher (Figs 3-6, 3-7).

Discussion

Based on the process-based coupled vegetation-atmosphere model CARLUC, I

evaluated the potential effects of climate variability on carbon dynamics of Amazonian

vegetation from 1995 to 2005. Predicted C stocks in leaves showed a slight trend of

increase from 1995 to 1998 (during the wet seasons) and from 1999 to 2003 (during dry

seasons), which suggest favorable weather conditions for forest growth from 1995 to

2003 (i.e., higher NPP) due to increasing PAR over time. These findings are partially









supported by results from Hashimoto et al. (2009), whose study on vegetation

responses to climate variability in the western Amazon (during 1984-2002) showed that

increases in NDVI were associated with increases in short-wave radiation between

August and December.

Despite the apparently favorable weather condition for C accumulation, I observed

a steady decrease in wet-season precipitation (and hence PAW) subsequent to the

drought of 1998. These reductions tended to lower predicted-C pools during the wet

seasons, exacerbating differences between dry and wet seasons over time. This

conspicuous seasonal pattern may be because portions of the Amazon reached levels

of PAW that were low enough to cause tree mortality (most variability in CARLUC-

based predictions of C stocks was associated with tree mortality). Note, however, that

this interpretation is based on the assumption that all forests of the Amazon would

respond to drought in ways similar to the site in Tapajos National Forest on which the

drought-induced tree mortality function was developed.

Losses of standing stocks of C were also observed following the drought of 2005.

In this year, there was a reversal in C accumulation compared to 2004 (shown in the

Basin-wide analysis), with a small net loss of C. While the general trend of increasing C

pools during the dry season until 2004, followed by a reduction in 2005, corroborates

the findings of Phillips et al. (2009), the magnitude of the reduction observed here was

less pronounced than in their study. At least four explanations may account for this

difference between studies. First of all, because CARLUC predicted modest increases

in C pools for RAINFOR plots whereas Phillips et al. (2009) reported actual reductions

for the same plots based on field measurements, it is reasonable to infer that the









Tapajos Forest (where the mortality function in CARLUC was developed) was more

resistant to drought than RAINFOR forests on average. This potential difference in

sensitivity of Amazonian forests to drought is probably a result of long-term adaptations

to cope with periods of low soil moisture. These results reinforce the need for long-term,

large-scale measurements of tree response to oscillations in climate.

Second, PAW driving CARLUC may have been overestimated, reducing predicted

drought-induced tree mortality. For example, CARLUC has no representation of

feedbacks between PAW and vegetation canopy cover or photosynthesis; PAW is an

external variable. Therefore, any potential increase in evapotranspiration in 2005

associated with vegetation dynamics (e.g., increased photosynthesis) could have no

effect on modeled PAW (see Nepstad et al., 2004 for more details on PAW

calculations). As a result, the model may have overestimated PAW, which could explain

the lower predicted C losses here compared to Phillips et al. (2009). In addition, PAW

was modeled based on the assumption that the top 2 m of the soil profile represented

depths down to 10 m at which point roots are exceedingly scarce (Nepstad et al. 1994).

In contrast, the scant available data on soil physics below 2 m (Jipp et al. 1998) indicate

that PAW maximum declines with depth (probably only for regions with very deep water

table); hence the estimates used to run CARLUC may have been overestimated. On the

other hand, higher drought-induced tree mortality in 2005 could have lowered ET, thus

increasing modeled PAW compared to actual PAW.

The plot network used by Phillips et al. (2009) to infer C patterns over the Amazon

Basin may not represent the climatic variability over the region. Although RAINFOR

plots are widely distributed, there are very few plots located between Peru and Manaus,









a region that CARLUC predicted to have had high productivity in 2005. Therefore, this

high productivity, associated with high levels of PAR and low mortality, could counter

balance C losses from mortality occurring in other parts of the Amazonia Note,

however, that while the climate data used here are from 280 widely distributed stations,

there are still several large geographical gaps.

Finally, CARLUC does not represent some intrinsic processes in vegetation

dynamics that are captured in field-based studies. CARLUC is an envelope of climatic

variables alone used to predict carbon stocks; hence, interspecific competition and

other important ecosystem processes are missing. These intrinsic dynamics, which are

difficult to capture in in models that do not represent competition among individuals or

species (e.g., 'big-leaf' models), could explain the differences between the model and

the field-based data.

While predictions suggest that droughts may become more common and intense

in the near future, apparently no model used to predict future vegetation trajectories in

the Amazon has yet incorporated the effects of drought-induced tree mortality on carbon

stocks; instead, droughts are assumed to have affect only photosynthesis (e.g., Foley et

al. 1996). Similarly, the allocation of NPP to different plant parts is often ignored, which

contrast with results from a partial throughfall exclusion experiment in which

aboveground wood production was drastically reduced in response to the throughfall

exclusion, while litterfall and soil C02 efflux showed minor or no response to the

throughfall exclusion, respectively (Brando et al. 2008). Without field-based, large-scale

data on carbon fluxes and stocks across the region measured over time, it is still difficult









to evaluate the future of forests of the Amazon at the basin scale under different climate

change scenarios.








































Figure 3-1. Conceptual diagram of the CARLUC ecosystem model. Carbon pools are controlled by climate (via its effects
on NPP) and PAW (via its effects on NPP, mortality, and NPP allocation). Once tissue dies, part of the carbon is
transferred to pools with low and high residence times. The other portion of the carbon resulting from dead
tissues is transferred directly to the atmosphere.


50











20 -





o

4 14-
_J

I 12-
0

1 -
I I I II I


12 Y = 0.009 + 0.49548 e


01
E 8-
0

6

4-

2-

0-
I I I I I I
c 1.0
.0- Prop. Stem Y=0.4(1-e ...)

S0.8 -O Prop. Root Y=0.28+0.72e-( PAW
0 Prop. Leaf- Y=1-Prop Stem-Prop Root


z



tL
0c



0 20 4L 60 80 100
Plant Available Water (% of Max)

Figure 3-2. Annual rates of leaf mortality (top panel), stem mortality (mid panel), and
NPP allocation (lower panel) as a function of PAW. These relationships were
derived from data from a large-scale partial throughfall exclusion experiment
(Nepstad et al. 2002, 2007; Brando et al. 2008).
















/0
S*

*





I I I I I
1996 1998 2000 2002 2004


1996 1998 2000 2002 2004
1996 1998 2000 2002 2004


Figure 3-3. Temporal patterns in carbon stocks predicted by CARLUC in leaves (upper
panel) and stems (lower panel) for the Tapajos National Forest.











0O
Region 1
Region 2
Region 3
Rainfor Plots









(Y)









O










1996 1998 2000 2002 2004 2006



Figure 3-4. Temporal patterns in average aboveground C stocks in regions with low,
medium and high levels of PAW. The black line represents C behavior
estimated by CARLUC in regions where permanent plots are used to infer
about aboveground carbon dynamics.









0)
S10 -*- wet-season
dry-season
co
4D-
Co



i .


0 r^ /\


U)
/ \















Figure 3-5. Temporal patterns in total aboveground carbon stocks averaged by dry and
wet season and per year. The dry season was represented by the months
August, September, and October, while the wet season by the months
between January and May. Note that the CARLUC run with both mortality and
NPP allocation varying as a function of PAW were used to generate the pools
of C.
uiI *








1996 1998 2000 2002 2004 2006


Figure 3-5. Temporal patterns in total aboveground carbon stocks averaged by dry and
wet season and per year. The dry season was represented by the months
August, September, and October, while the wet season by the months
between January and May. Note that the CARLUC run with both mortality and
NPP allocation varying as a function of PAW were used to generate the pools
of C.









Dry Season I


Figure 3-6. Climate patterns during the dry (July-October) and wet (January-April)
seasons. Each climatic variable was averaged per season from 1995 to 2005.


Wet Season


















./ "
.? ,


IN




4;-




.jr,,; j


'I


Legend
* RAINFORFPlts
Region 1 (Low)
Region 2 (High)
SRegion 3 (Medium)
ABG C Stocks (kg m )
I 13


V. .

0 27, 5 ,,"r
0 75 0 .. 550 Iometers
O 5 0 550 Kilometers


Figure 3-7. Total aboveground carbon stocks simulated by CARLUC for the entire
Amazon and for three regions of varying PAW. Stars represent RAINFOR
sites. Note that in this simulations the functions representing drought-induced
change in NPP allocation and mortality varied according to oscillations in
PAW.


; ;

~I









CHAPTER 4
SEASONAL AND INTERANNUAL VARIABILITY OF CLIMATE AND VEGETATION
INDICES ACROSS THE AMAZON

Introduction

The accumulation of heat-trapping gases in the atmosphere may subject large

areas of the Amazon Basin and other tropical forest formations to a greater frequency

and severity of drought in the coming decades (Christensen et al. 2007, Malhi et al.

2008). This trend may interact synergistically with regional inhibition of rainfall driven by

deforestation (Costa and Foley 2000, Nobre et al. 1991, Werth and Avissar 2002) and

more frequent sea surface temperature anomalies (e.g., El Nino Southern Oscillation

and North Atlantic Tropical Oscillation) (Cox et al. 2008, Marengo et al. 2008, Nepstad

et al. 2004) to move these tropical forest regions toward large-scale forest dieback

events. Drier and warmer climate in the region favors the persistence of grasses and

shrubs over trees, in a process that is reinforced by recurring fire (Nepstad et al. 2008).

Large scale forest diebacks would transfer to the atmosphere a substantial fraction of

the 250 Gt C contained in tropical forest trees (Christensen et al. 2007).

It is difficult to assess the drought thresholds beyond which forest dieback might

occur in part because of conflicting evidence regarding the response of forest

photosynthesis to drought (Fisher et al. 2006, Betts et al. 2008, Huntingford et al. 2008,

Mayle and Power 2008, Lloyd and Farquhar 2008). Some studies suggest that forest

photosynthesis (gross primary productivity, GPP) increases during the early stages of

drought because of higher photosynthetically active radiation (PAR) (Graham et al.

2003, Hutyra et al. 2007, Saleska et al. 2003). In contrast, two partial throughfall

exclusion experiments conducted in the Amazon region found that proxies of forest

productivity all declined under mild drought conditions, with tree mortality increasing









under high cumulative canopy water stress resulting from limited plant available water

(PAW) [reviewed by (Meir et al. 2009)]. These impacts were observed only after 2-3

years of simulated drought, with the lag probably due to plant adaptations to reduced

soil water availability (Oliveira et al. 2005) .

The apparently contradictory findings about drought effects on plants are

particularly striking in 2005, when the most severe drought of the last 100 years affected

the southwestern Amazon (Marengo et al. 2008c) This dry and warm period, linked to

an anomalously warm tropical North Atlantic, provided an opportunity to evaluate forest

resistance to drought over a greater spatial extent than the partial throughfall

experiments noted above. Two studies reporting on this drought event diverged in their

findings. Phillips et al. (Phillips et al. 2009) found evidence that intact forests of the

Amazon Basin were drought-sensitive, accumulating 1.2-1.6 Pg less carbon during the

drought period of 2004-2005 than in previous years, and concluded that Amazonian

forests were negatively affected by the drought of 2005. Conversely, Saleska et al.

(Saleska et al. 2007) found that intact Amazon forests experiencing anomalously low

precipitation (PPT) in 2005 had higher photosynthetic activity, as indicated by the

enhanced vegetation index (EVI), a canopy reflectance metric developed to minimize

the attenuating influences of background and atmospheric effects and thus remain

effective even in areas of high biomass and chlorophyll content (Huete et al. 2002). The

authors of the latter study concluded that these forests were more drought-resistant

than previously thought. A possible resolution of this conflict was provided by a recent

paper by Samanta et al. (2010) that suggested that the higher EVI observed in 2005 by









Saleska et al. (2007) could be attributable to atmospherically induced variations

associated with aerosol loadings.

We conducted a study to better understand the responses of tropical forests to

climate extremes, particularly the processes that permit forests to sequester carbon

during drought conditions. First, we evaluated the temporal patterns of climatological

variables from 280 meteorological stations (1995-2005), PAR from GOES, and an

improved MODIS EVI data set over the dry seasons of 2000-2008 across the Amazon

Basin. Second, we examined statistical relationships between EVI during the dry

season (July to September) and three integrative climate variables (VPD, PPT, and

modeled PAR). We analyzed the EVI climate variable relationships for both the entire

Amazon Basin (intra-annually) as well as just for the densely forested areas (inter-

annually). For the Basin-wide analysis, we predicted that EVI within a given year would

demonstrate greater sensitivity to drier climatic conditions (e.g., high VPD and low PPT)

across a gradient of increasing canopy closure. In densely forested areas (270%

canopy cover), we predicted that inter-annual EVI variability would be positively

correlated with modeled PAR in sites with high precipitation history (i.e., average

monthly dry-season precipitation for 1995-2005 > 100 mm). We also predicted that EVI

would be negatively correlated with VPD in locations with low PPT history (< 100 mm

per dry season month), where dry-season VPD is likely to limit productivity, thus

reducing leaf production and leaf area. Finally, using data at relatively high spatial (0.25

km2) and temporal resolution, we evaluated the EVI relationships not only on the basis

of these same climate variables, but also with monthly field measurements of LAI,

litterfall, and new leaf production in a dense forest near Santarem, Brazil.









measurements of LAI, litterfall, and new leaf production in a dense forest near

Santarem, Brazil.

Methods

MODIS Data

I computed the enhanced vegetation index (EVI) at a spatial resolution of 500 m

from the MCD43A4 (collection 5) MODIS Nadir Bidirectional Reflectance Distribution

Function (BRDF) adjusted reflectance (NBAR). The NBAR data were standardized to a

nadir view geometry and solar angle, and processed to limit the influence of cloud

cover, thereby limiting the influence of seasonal variations in acquisition conditions

(Schaaf et al. 2002). I further screened the NBAR reflectance quality by using the QA

flags provided for the NBAR product and we generated a monthly and seasonal

composite of "best quality" reflectance (only reflectance derived from full or magnitude

inversion) to calculate the EVI. Because of the high number of observations influenced

by cloud cover and atmospheric contamination during the wet seasons of 2000-2008, I

focused our Basin-wide analysis on the average of the driest months of the year, July-

September (referred to as dry season EVI; see appendix A), thus allowing direct

comparisons with Saleska et al. (2007). If EVI in one of these three months was

missing, the dry-season average EVI was based on the average of two other months. If

EVI was missing in two months, the dry-season EVI was based on a single value. The

NBAR EVI requires ~ 5-7 good looks every 16 days (which typically occurs only in the

dry season), while for the standard EVI product a single good clear day every 16 days is

sufficient. As a result, NBAR-EVI is a less noisy temporallyy variable) data set. This

procedure thereby assured the largest number of observations of the highest possible

data quality.









Because I found that the choice of those months could potentially influence the

spatial patterns of EVI from North to South of the Basin (see Appendix B) I did not focus

our analysis on geographical gradients. Also, I included in all statistical analysis

longitude and latitude as covariates (see statistical analysis for more details).

EVI Surrounding Meteorological Stations

For the comparison of EVI with in-situ climatic measurements, I averaged the EVI

data over an area of 8 km x 8 km (256 pixels) surrounding each meteorological station.

Following, I estimated canopy cover of the vegetation surrounding these meteorological

stations as of 2005 (Hansen et al. 2003). This analysis included both low (e.g., <70%)

and high (e.g., >70%) canopy cover. The former includes a mixture of vegetation types.

In contrast, densely-forested areas included only cells with high canopy cover (2 70%).

Site-specific Analysis

The credibility of MODIS-EVI in capturing vegetation dynamics was tested in the

Tapajos National Forest, Para, Brazil (2.8970, 54.9520W) (km-67) using data from both

a 1-ha plot (a "control plot" of a rainfall exclusion experiment; details in (Brando et al.

2008, Nepstad et al. 2002)) and an eddy covariance flux tower (Hutyra et al. 2007).

First, volumetric soil water content (VWC) was measured each month from 2000 to

2004 using a series of paired Time Domain Reflectrometry probes situated in five soil

pits in the 1 ha plot (Nepstad et al. 2002). The VWC measurements were used to derive

estimates of plant available water (PAW) for two interval depths: 0-2 m (PAW-2m); and

2-11 m (PAW-11 m). Second, leaf area index (LAI) was measured monthly from January

2000 to December 2005 at 100 grid points systematically distributed in the 1 ha control-

plot. We used two LiCor 2000 Plant Canopy Analyzers in differential mode (LI-COR,

1992) (Nepstad et al. 2002). Third, from January 2000 to December 2005, litterfall was









collected every 15 days using 100 screened traps (0.5 m2 each) elevated 1 m from

ground level (Brando et al. 2008) and located in the same grid points of LAI. Fourth,

visual assessments of the presence of new foliage were conducted monthly between

August 1999 and August 2004 for all individuals >= 10 cm in diameter at breast height

in 1 ha plot (480 individuals). Here I present percent of individuals with new leaves at a

given census. Finally, PAR was measured from 2002 to 2006 using an LiCor 190-SA at

63.6 m in height (Hutyra et al. 2007). Note that in this site: annual precipitation ranges

from 1700-3000 mm with a mean of ~ 2000 mm; 2) during the dry season (July through

December), rainfall rarely exceeds 75 mm mo-1; soils are deeply weathered oxisol clays

of the Haplustox group; and, the permanent water table at a similar site 12 km from the

research plot was >100 m deep.

NBAR-EVI and EVI-Product

For the site-specific analysis, I used both the EVI derived from the NBAR product

(500 m resolution) and the standard-EVI product (collection 5) composite each 16

days, with a spatial resolution of 250 m vs. 250 m. As noted earlier, the NBAR-EVI data

limit the attenuating effects of the atmosphere and viewing conditions (Schaaf et al.

2002), but required temporal compositing to maximize data quality. I screened the

NBAR reflectance quality using the QA flags provided for the NBAR product, generating

a monthly and seasonal composite of "best quality" reflectance (Schaaf et al. 2002). The

MODIS standard EVI product was also screened based on the MODIS product quality

flags (e.g., reliability). No corrections for solar or viewing conditions are made in the

standard EVI product. Thus, differences between the NBAR-EVI and standard EVI

products could be related either to the BRDF model used to solve for surface

reflectance, to procedures used to screen for cloud contamination and the normalization









for solar illumination angle, or to all of these factors. Moreover, I compared EVI through

repeated measurements of individual pixels, thus corrections for the influences of

viewing conditions and solar angle were potentially quite important.

Climate

Monthly data from ~280 meteorological stations (Nepstad et al. 2004) were used

to derive dry and wet season averages for the Amazon (from 1995 to 2005). In the

same locations where the meteorological stations were located, PAR was estimated

using GOES observed cloud cover and a clear sky solar radiation model. I assessed

seasonal averages of climatic variables through time for each meteorological station,

using a linear mixture model that accounted for spatial and temporal autocorrelation

(i.e., random effects of space and time) and a parameter to attenuate the effects of dry

years (i.e., 1998 and 2005) (Bates and Sarkar 2006). Finally, the slopes of these

regressions were tested for differences against the null model of zero change over time

(Baayen 2008), which included only random effects. For each one of the 280

meteorological stations, I calculated PPT history, i.e., the average dry season

precipitation based on data from 1996 through 2005.

Statistical Analysis of EVI and Climatic Variables

Basin-wide (mixed vegetation types)

A general linear model (M1) of EVI in mixed vegetation (i.e. not just densely

forested) areas was comprised of three components: 1) the predictor variables VPD,

PAR, PPT, canopy cover (CC), and the interaction of CC with the other covariates; 2)

error terms assumed to be spatially autocorrelated according to an exponential spatial

structure [see (Pinheiro et al. 2000.)]; and, 3) covariates of longitude and latitude, to

capture the large scale spatial gradient of EVI. I fitted this model to the data for each









year of the study (2000-2005) and, therefore, do not account for inter-annual variability

in mixed vegetation areas, which is complex because of its association land use

change. It does allow for separate regression coefficients and autocorrelation

coefficients for each year.

Densely forests areas

The general linear statistical models (M2) of EVI in forested areas were comprised

of three components. First, covariates of PAR, VPD, and PPT (each term interacted

with PPT history class), were included in order to assess the effects of climate on EVI in

three different climatic regions: seasonally very dry regions (PPT history class: 0 mm to

65 mm); seasonally dry regions (PPT history class: 66 mm to 100 mm); and non-

seasonal regions (PPT history class: > 100 mm). Second, longitude and latitude, were

included as covariates to capture the large scale spatial gradient of EVI. Third, the error

terms were assumed to have a autoregressive correlation structure of order 1 (Pinheiro

et al. 2000) to account for temporal autocorrelation among observations from the same

locations over several years. Note that for forested areas we did not find evidence of

spatial autocorrelation in the residuals from the fitted model.

Hierarchical partitioning

In both mixed vegetation and densely forested areas, I used hierarchical

partitioning (Murray and Conner 2009) as a complementary statistical method to

evaluate each covariate's contribution to EVI. In this method, the variance in the

response variable (EVI) shared by two predictors can be partitioned into the variance of

EVI attributable to each predictor uniquely. For mixed vegetation areas, for each year

of the study, I used hierarchical partitioning to evaluate model 1 (Ml), but without a

spatial structure. For densely forested areas, I evaluated a model that included all









climatic variables, the location of each meteorological station, longitude, latitude, and

year; referred to throughout as a variation of M2. Because PPT was dependent on

location of the meteorological station, I did not include this variable in the hierarchical

partitioning analysis.

Relationship of the variability of EVI with PAW

I first calculated for each MODIS pixel the coefficient of variation of the EVI for the

period 2000-2008, but retained only pixels with > 70% of canopy cover and with data for

6 or more years. The Amazon Basin was then stratified by classes of PAW (0-0 m in

depth): class 1: 470-985 mm; class 2: 986-1280 mm; class 3: 1281-1500 mm; class 4

1500-2100 mm, see Nepstad et al. (2004) for the calculation of PAW. Finally the

coefficient of variation of the EVI was compared between the four PAW classes visually.

Results

Spatial-temporal Patterns of EVI

EVI varied spatially and temporally across the Amazon Basin during 2000-2008

(Fig. 4-1). Generally, EVI varied most where soil moisture availability (PAW) was the

most variable. There was a strong gradient in mean annual EVI from the western to the

central portion of the Basin, with associated gradients in the variability of EVI and in the

estimated average annual PAW from 0 m to 10 m depth.

In areas with high (>70%) tree cover, roughly defining the area of intact forest

across the Basin (Fig. 4-2), EVI was below average in 2000, 2001, 2004, and 2007;

close to average in 2003 and 2008; and above average in 2005 and 2006. Pronounced

shifts in the annual anomalies (deviations from 2000-2008 mean values) were observed

over short time periods. For example, anomalies in 2002 were 151% higher than in

2001; in 2005 they were 168% higher than in 2004; and, in 2007 they were 257% lower









than 2006. Analysis of the spatial covariation of EVI and PAW revealed that inter-annual

EVI variability in dense forests was slightly higher in drier areas, with gradually

decreasing EVI variability as average PAW increased (Fig. 4-3).

Basin-wide Climate

The Amazon Basin experienced a decline in annual rainfall and an increase in

PAR from 1995 through 2005. Based on data from ~280 meteorological stations

distributed across the Amazon (Fig. 4-4), I found that annual wet-season PPT

decreased at a rate of 5.31 0.68 mm yr1 (Fig. 4-4). Dry-season PPT also tended to

decrease over this time period, but not significantly (-1.018 0.92 mm yr1). In contrast,

PAR increased over the period 1995-2005, primarily after 2002 (especially during the

wet-season), but the rate was only statistically significant for the dry-season (15.8

3.67 moles m-2 month- yr1). I could not detect strong linear temporal pattern in VPD

over the entire time period; on the other hand, it appears that VPD increased

substantially after 2002 (Fig.4-4). For example, while the rate of increase of average

VPD during 1995-2000 was modest (0.0029 kPa yr-), it increased to 0.43kPa yr-

during 2002-2005. As a result of decreased PPT and increased VPD, modeled PAW

from 0 to 10 m depth decreased significantly over the time period at a rate of 2.03

0.22% during the wet season and 2.21 0.11% year- during the dry season [see also

Nepstad et al. (2004)]. Similar patterns were observed for areas with high fraction of

canopy cover (Table 4-1).









Table 4-1. Slopes of precipitation, photosynthetic active radiation (PAR), and vapor
pressure deficit over time. All linear models between these climatic variables
and time were fit using a linear mixed model with random effects of space and
time, except for PAW. PAW was averaged over spaced and regressed over
time.
Meteorological stations surrounded Meteorological stations surrounded
by all fractions of forest cover only by high fraction of forest cover
Wet Season Dry Season Wet Season Dry Season
Precipitation (mm) -5.31 0.68*** -0.35 0.91 -7.791.62*** -1.351.13
PAR 10.38 7.64 12.22 4.51** 12.448.21 11.234.45**
(moles m-2 mo-1)
VPD 0.00 0.00 0.00 0.01 0.000.00 0.000.00
(kPa)
PAW -2.03 0.22*** -2.21 0.11***
(% of Max)


EVI and Climate Variables

Spatial analysis across the Amazon Basin

In areas with mixtures of vegetation types (e.g., pastures, cerrado, secondary and

primary forests), it was predicted that EVI would decrease with (i) increasing VPD and

(ii) decreasing canopy cover and PPT. Based on spatial analysis of EVI data from 256

cells (each 64 km2) surrounding each meteorological station, these predictions were

generally supported. There was a strong and significant interaction between canopy

cover and VPD in 4 out of 6 years of the study (Table 4-2). Thus, EVI tended to

decrease as VPD increased in areas of low canopy cover. Similarly, PPT interacted with

canopy cover in 3 out of 6 years the study, but only marginally (p<0.1); as PPT

decreased in areas of sparse canopy cover, EVI also decreased. In contrast, PAR

showed a marginally significant interaction with canopy cover only in one year (p =

0.054). Thus only in 2004 did EVI increase as PAR increased in areas of high canopy

cover. In addition to general linear models, I used hierarchical partitioning (Murray and

Conner 2009) to estimate the relative importance of each predictor on EVI, accounting









for collinearity among predictors in a linear model (see methods). Based on this

analysis, canopy cover, VPD and latitude explained, respectively, 55%, 17%, and 11%

of the total variation associated with the full model (average R2 of the full model, M1:

~56%; see methods). Overall these results indicate a strong effect of VPD and a weaker

effect of PPT and PAR on EVI from 2000 to 2005. These results also suggest that

spatial gradients, independent of variations in climate, had important effects on EVI.









Table 4-2. ANOVA table for all predictors of EVI in non-forested areas from a linear model (M1). Note that we ran one
model for each year of the study.
2000 2001 2002 2003 2004 2005
F p F pl F p F p F p F p
Intercept 3741 <.001 4724 <.001 7069 <.001 1874 <.001 8012 <.001 3165 <.0001
Canopy
Cover (CC) 66.481 <.001 83.257 <.001 125 <.001 96.3 <.001 123.0 <.001 138.7 <.0001
VPD 8.035 0.006 22.43 <.001 31 <.001 6.73 0.011 40.33 <.001 9.974 0.0021
PAR 0.882 0.350 0.028 0.868 3.7 0.055 0.93 0.336 0.773 0.3813 0 0.9935
PPT 6.834 0.010 7.060 0.009 2.6 0.106 2.21 0.136 0.01 0.9211 0.093 0.7604
Longitude 0.949 0.332 2.056 0.158 4.3 0.040 2.08 0.152 1.262 0.2637 6.376 0.013
Latitude 3.77 0.055 1.931 0.168 10.8 0.001 7.30 0.008 19.633 <.0001 5.551 0.0203
CC*VPD 0.863 0.355 2.034 0.157 7.8 0.006 4.00 0.048 11.975 0.0008 6.737 0.0108
CC*PAR 0.718 0.399 0.419 0.519 0.3 0.574 4.25 0.042 0.046 0.831 0.085 0.7715
CC*PPT 0.196 0.659 1.570 0.213 2.8 0.092 3.06 0.083 9.837 0.0022 0.113 0.738









Spatial-temporal analysis in densely forested areas

In densely forested areas (i.e., all cells with canopy cover 2 70%), there was no

support for the hypothesis that inter-annual EVI variability was associated with temporal

variation in PAR, VPD, or PPT across all classes of PPT history (Table 4-2). Rather, the

only significant predictors of EVI were longitude (p=0.02) and PPT history alone

(p=0.108) (Table 4-2). Using hierarchical partitioning analysis to assess the importance

of each predictor of a linear model on EVI (M2; see methods), I found that spatial

gradients accounted for most of the EVI variability. For example, of the 77% of the

variation explained by the linear model, spatial variability alone accounted for 91% of

this variation, whereas year, VPD, and PAR accounted for only 4.2%, 0.5%, 0.45%,

respectively. Overall, these results show that, by considering the effects of spatial

gradients and temporal autocorrelation in our analysis, no climatic variable could

meaningfully explain the EVI inter-annual variability.


Table 4-3. ANOVA table for all predictors of EVI in dense forested areas from a general
linear model (M2).
F p
Intercept 26359.68 <.0001
PPT History 2.253 0.1081
PAR 1.111 0.2933
VPD 0.276 0.6000
PPT 0.225 0.6359
Latitude 1.826 0.1783
Longitude 5.008 0.0265
PPT History*PAR 0.073 0.9300
PPT History*VPD 0.043 0.9577
PPT History*PPT 0.319 0.7274









Site-specific Analysis

At the intensively studied Tapajos site (near Santarem; 80% tree cover), I found

that monthly EVI was highly seasonal and thus positively and strongly correlated with

field measurements of the proportion of trees (2 10 cm in diameter at breast height) with

new leaves (R2=53%; p<0.01). EVI was also positively correlated with field

measurements of PAR (R2=35%; p<<0.01), but inversely correlated with PAW from 0 to

2 m depth (R2=46%; p<0.01), which indicates no soil water constraints on

photosynthesis (as 0-2m PAW decreased, EVI increased) (Figs. 4-5, 4-6). Hence, EVI

was most sensitive to production of new leaves and associated light levels (the

individual effects of PAR and leaf phenology on EVI could not be decoupled in our

analysis). This interpretation is reinforced by the findings that field measured LAI varied

little between seasons and was poorly correlated with EVI (R2=17%, p=0.05) (Fig. 4-6).

Although other environmental variables may also have affected EVI indirectly (for

example, via lagged effects on vegetation processes), their direct relationships with EVI

were not evident (Fig. 4-6).

In contrast to EVI derived from the MODIS corrected reflectance products (see

methods), the MODIS "standard collection 5 product", screened using standard quality

control flags, was better correlated with correlated with these same measured seasonal

environmental and biophysical variables (Fig. 4-7). These results indicate that the EVI

product used for the analyses (produced from bi-directionally corrected reflectance data

sets) was better correlated with field conditions than the standard MODIS EVI product,

which could be a result of correction of viewing angle conditions, screening procedures

for cloud cover, or both. While we cannot ensure the results at the Tapajos sites extend

across the entire Amazon Basin, particularly because this site was not affected by









severe drought during the study period, confidence in our regional results was

supported by these more local scale observations.

Discussion

Increases in solar radiation during dry periods may boost tropical forest

productivity (Graham et al. 2003, Saleska et al. 2003, Wright et al. 1999), but prolonged

and severe droughts ultimately limit this effect by inducing stomatal closure and even

tree mortality (Brando et al. 2008, Nepstad et al. 2007). The thresholds at which

drought starts to reduce productivity (as opposed to increasing it) are still not well

known. While satellite-based vegetation indices allow insight into potential

environmental thresholds (Huete et al. 2006, Myneni et al. 2007, Saleska et al. 2007,

Xiao et al. 2005), they have provided an ambiguous measure of vegetation responses

to drought in the Amazon Basin. Here, I demonstrate that spatial variations of an

improved EVI metric, corrected for bi-directional reflectance variations and other

potential attenuating influences (Schaaf et al. 2002), were associated with gradients of

PAW and VPD. This indicates that EVI captured complex spatial patterns of

photosynthetic responses to environmental variables across the Amazon. In particular, I

show contrasting responses of EVI to the interaction between tree canopy cover and

drought, over a wide range of environments (e.g., pastures, secondary forests,

cerrados). These results reinforce findings from previous studies that demonstrated that

where tree canopy cover is high, trees are better buffered against drought because of

their deep root systems (Nepstad et al. 1994). Further, these results indicate that this

buffering mechanism may not be restricted to the central Amazon (Huete et al. 2006).

In Tapajos's dense forest, even subtle changes in EVI captured complex seasonal

ecosystem dynamics. In particular, the NBAR-EVI increased with the number of canopy









trees with new leaves, while LAI varied little between seasons. This finding show that

leaf flushing (but also PAR) was an important driver of EVI and therefore of GPP

(Doughty and Goulden 2008). The association between EVI and PAR was more

apparent at Tapajos forest (R2=35%) than in our Basin-wide analysis of densely

forested areas, in which EVI was generally not responsive to any specific climatic

variable. This finding raises the possibility that other mechanisms other than PAR could

be driving the EVI inter-annual variability. I propose three possible mechanisms for this

observation.

First, given the strong correlation between EVI and the number of trees with new

leaves observed at Tapajos, it is reasonable to expect that leaf flushing could have

played an important role on the Basin-wide, inter-annual variability in EVI. Whereas leaf

flushing usually coincides with periods of increased radiation (Wright and van Schaik

1994), leaf bud break is not necessarily cued by radiation, but by gradual changes in

daylength (Rivera et al. 2002). Once bud break occurs, however, leaf development is

strongly controlled by water availability for cell expansion. Therefore, changes in tree

water status modified by precipitation events (or even by leaf shedding) during the dry

season of dry years is likely to synchronize bud development and, consequently, leaf

flushing. Because inter-annual EVI variability in dense forests during the dry season

was greatest in regions of lower PAW, I hypothesize that drought could increase EVI by

synchronizing leaf flushing via its effects on leaf bud development and tree water status.

This hypothesis could explain, in part, the high anomalies during dry years (e.g., 2005).

I suggest that this is a potentially important area of research.









Second, the lack of a relationship between PAR and basin-wide EVI could be

associated with long-term adaption for herbivory avoidance. It has been suggested that

the timing of leaf flushing of tropical trees is the result of selective processes to coincide

with periods of lowest insect activity, presumably during the peak of the dry season

(Aide 1993). Based on this hypothesis, however, I could not explain the EVI inter-annual

variability observed in this study.

Finally, the results presented here may be partly related to the sample size used in

the spatial-temporal analysis. Although I used one of the best available data set, there

were just 40 meteorological stations available for assessing significance in the densely

forested parts of the Basin. As a result of this relatively small sample size, I could not

further test the effects of PAR on the inter-annual variability of EVI nor the hypothesis of

increased productivity (GPP) during dry periods of increased PAR.

I thus hypothesize that the apparently conflicting observations between Phillips et

al. (2009) and Saleska et al. (2007) regarding the drought event of 2005 were related to

several mechanisms operating simultaneously. GPP (expressed as EVI) appears to

have increased due to the production of new leaves and increased PAR (Saleska et al.

2007), whereas aboveground net primary productivity (ANPP) measured in the field

concurrently decreased (Phillips et al. 2009) because of higher tree mortality and

increased respiration associated with lower PAW and high temperatures, respectively.

Moreover, the allocation of non-structural carbohydrates to belowground processes may

have increased, given that EVI (and possibly GPP) was higher and ANPP lower during

the drought of 2005.









While it was observed important oscillations in weather over the Amazon from

1995 to 2005 (e.g., a 5.3 mm yr-1 reduction in PPT), these oscillations were not clearly

related to EVI inter-annual variability in densely forested areas at the Basin scale. Thus,

there is a need for additional analyses that couple field measurements with satellite

observations in order to clarify how the Amazon region responds to drought, how those

responses will be expressed in the future under increasing drought conditions, and to

what extent those responses are captured in satellite observations of canopy

photosynthesis.























f EVI: coef.var. '00-'08

* I* y *
\..-^^^4'


PAW: mean '95-'05


PAW(%)





Figure 4-1. Upper panel: Average dry-season EVI across central South America for the
period 2000-2008. Overlaid circles represent average vapor pressure deficit
measured at 280 meteorological stations across the region for the period
1995-2005. Center panel: Coefficient of variation in annual EVI for the period
2000-2008. Lower panel: Average annual plant available water at 10m depth,
expressed as a percentages of the maximum, for the period 1995-2005.










SJ mean '00-'08 anomaly '00

compared
EVI 4-, with mean
>0.6 > +o003
0.55
<0.5 -.________ < -0.03
anomaly '01 % anomaly '02







anomaly '03 anomaly '04







Sanomaly'05 anomaly'06

I r 7





anomaly '07 anomaly'08




v,," V' '
-,-



Figure 4-2. Spatial-temporal patterns of dry-season EVI across the Amazon for areas
with high canopy cover (EVI > 0.4 and MODIS tree canopy cover product >
70%). The first panel shows average EVI from 2000-200 and the following
panels show the EVI anomaly for each year, calculated as EViI EVImean.

















CO





00
co

> -
UJ



C0




CO


1 2 3 4

PAW

Figure 4-3. Average coefficient of variation of EVI for the entire Amazon, based on data
at 500 x 500 m resolution (only for areas with canopy cover 2 70%).
















0

E
Eo _
- 04
.-_
0





Co
U0)
ic 0
0n
0o


1998 2000 2002


2004 2006
2004 2006


1996 1998 2000


CO


0





Cq

1996


2002 2004 2006


1998 2000 2002 2004 2006
1998 2000 2002 2004 2006


Figure 4-4. Monthly climate patterns over the Amazon region based on data from 280

meteorological stations distributed across the basin. (A) Monthly averages of

Precipitation, (B) Monthly modeled photosynthetic active radiation (modeled

PAR), (C) Modeled monthly plant available water at 10m depth (PAW), (D)

Monthly vapor pressure deficit (VPD). Blue lines represent a smooth curve

based on a loess method, and red lines represent a local regression model

(spline).


1996 1998 2000 2002 2004 2006


0-






0,



o-
s
(0


























1000


cn 800

E
*0 600
E

S400


200


x LAI + Leaf Phenology PAW(0-2m)
EVI-Product Litterfai A PAW(2-11m)




- 5





A+ 1 + + 4



7+
+ 0 0 "(13. M



*


S0.7


-0.6

EVI
0.5


0.4


0.3
120

100

80



40 _.
cO
20

0 -
S200

E
1- 50 E


100

so

0


Figure 4-5. Temporal patterns of environmental and biophysical correlates of EVI.
Upper panel: EVI derived from MODIS NBAR, EVI derived from the Collection
5 MODIS product screened to include only good or best quality control flags,
and field-measured leaf area index (LAI) (see methods). Center panel:
monthly photosynthetically active radiation (PAR), and bi-monthly litterfall and
new leaf production (see methods). Lower panel: plant available water (% of
maximum) at two depths (0-2m and 2-11 m) and daily precipitation (mm).



















80




















075



065-



S055 -



045-



0.35-




075-


0 65
*" '


0 0



0 0
S o
o *








All Months: R2=0 44 P=0

02 0.3 0.4 0.5 06 0.7
Individuals with new leaves (%)


20 40 600 8C
PAW(% of Max) 2-11 m


0.75-




S ,.

o7o5
*~sa .^
05 0



0.55 .. 0

*

0.45-



0.35 All Months: R=0.37, P=0001
600 700 800 900 1000
PAR (I mol m s 1)

0.75-



065 5- 7 0
O


0.55 *

4* 0* A

0.45-



0 35 All Months: R2=0.24, P=0 001


S 100 0


075 Dry-Season: R2-0.36, P-0.008


*
065 00 0
S00
065 0



...' O
055. ,, CO

*i o

045



0 35 All Months: R'=0.24, P=0.001
6 6 10 12
Litterfall (Mg ha yr 1)

075-



D 65- 0
0 O
0 .*
.* 0


0
D.55 o



0.45



D35


0.75


0




O o
0.55- '; O *

S*
0 .


0.45-



0 35 All Months: R2=0.4. P=0
20 40 60 80 100
PAW (% of Max) 0-2 m


EVI NBAR:
o Dry Season (Jul-Oct)
Wet Season (Nov-Jun)




- Regression Dry Season
- Regression -All Months
Smooth Line loesss)


100 200 300 400 50 55 610
PPT (nmmmonth) LAI (m2 m 2)


Figure 4-6. Site Specific analysis on the relationships between monthly EVI-NBAR and monthly field measurements of

individuals with new leaves (%), PAR, litterfall, shallow (0-2m) and deep (2-11m) PAW, PPT, and LAI. The solid

and dashed lines represent straight lines fitted using a standard linear regression models for all months and dry

season; respectively. The dotted line represents a smooth line.











81


*
0 0 0


... ..... 0


0* 0


0.55-



0.45-



035-


I































4b 0
0



%

*0

All Months: R=0.15, P=0.009

0.2 03 04 0.5 0.6
Individuals with new leaves


075



065-



L055-



045-



035-




075-



065-



0,55-



045-



035-




0 o
o
b.. .. .
o +"'


30 40 50 60 70 80 90 100
PAW(% of Max) 2-11 m


0













600 700 800 900 1000
PAR (A mol rnm s 1)






*)
09
~ *

o ..O -
*










S

















0 100 0 300 400 500
PPT (tm/month)
PPT (mm/mont1h)


Dry-Season: R'=0.38, P=0.035






0
* .oP. ,.





,













*

5.0 6 8 10T







Ltt (Mg m ')
*






*






5.0 5.5 60 65 70
LAI (m2 m 2)


0.75



0.65 -



0.55-] .......... o O
s* *


0.45 *



0.35-
12 20 40 60 80 10&
PAW (% of Max) 0-2 m


EVI Product:
o Dry Season (Jul-Oct)
Wet Season (Nov-Jun)



- -- Regression Dry Season
- Regression All Months
...... Smooth Line loesss)


Figure 4-7. Site Specific analysis on the relationships between monthly EVI product and monthly field measurements of

individuals with new leaves (%), PAR, litterfall, shallow (0-2m) and deep (2-11m) PAW, PPT, and LAI. The solid

and dashed lines represents straight lines fitted using a standard linear regression models for all months and

dry season; respectively. The dotted line represents a smooth line.


**


* *









CHAPTER 5
POST-FIRE TREE MORTALITY IN A TROPICAL FOREST OF BRAZIL: THE ROLES
OF BARK TRAITS, TREE SIZE, AND WOOD DENSITY

Introduction

The synergistic effects of droughts, forest degradation by logging, landscape

fragmentation, and increased fire ignition have already rendered many Amazonian

forests fire-prone (Alencar et al. 2006, 2004, Aragao et al. 2007, Aragao et al. 2008,

Cochrane and Laurance 2008, Cochrane 2003, Nepstad et al. 2001, 2004, Nepstad et

al. 2008). If current predictions of drier and warmer conditions (Cox et al. 2008, Cox et

al. 2000) hold true, even large forested areas in Amazonia may become vulnerable to

fire in the near future (Golding and Betts 2008). For example, 16 of 23 climate models

predicted substantial reductions in precipitation by the mid 21st Century (Malhi et al.

2008). While this change is anticipated to have substantial effects on forest carbon

storage and biodiversity (Barlow et al. 2002, Barlow and Peres 2008, 2006, Nepstad et

al. 1999), there are few data available to evaluate the vulnerability of forest trees to fire.

One of the most obvious ways that understory forest fires affect tropical forests is

by killing trees (Barlow et al. 2003, Cochrane and Laurance 2002, Pinard and Huffman

1997, Pinard et al. 1999, Uhl and Kauffman 1990). The characteristic low-intensity

surface fires of closed canopy tropical wet and moist forest spread slowly through the

understory and heat the vascular cambia of trees. When cambial cells die around the

entire circumference of a tree's bole, it is expected to die (Michaletz and Johnson 2007).

Based on this basic observation, a wide range of models have been proposed to

measure the vulnerability of tropical trees -and hence tropical forests to fire. The

simplest (and so far most used) models assume that bark thickness is the most

important trait in avoiding fire-induced cambium damage and, consequently, tree









mortality. Uhl and Kauffman (1990), for instance, studied a moist Eastern Amazonian

forest and found that a simulated fire increased cambium temperatures to lethal levels

of most trees with thin bark. Based on this finding, the authors speculated that the

occurrence of surface fires would kill most standing trees. Pinard and Huffman (1997)

used a similar approach to suggest that only individuals with bark > 18 mm thick would

resist a fire in a dry forest of Bolivia.

Although back thickness has been shown to correlate well with post-fire tree

mortality (Barlow et al. 2003), other traits that reduce the indirect (and negative)

influences of fire on the physiology and structure of tree trunks could better represent

the complex processes involved in post-fire tree mortality (Michaletz and Johnson

2007). For example, trees wounded during surface fires may experience wood decay

ranging from severe to moderate depending on their capacity to compartmentalize wood

decay, a process that is intrinsically associated with wood density (Romero and Bolker

2008). Likewise, simply by being large in diameter, trees may reduce the indirect effects

of fire on tree mortality by being resistant to wind throw (Larjavaara and Muller-Landau

2010), less likely to be killed by branch and tree fall from other trees, and less likely to

have their circumference heated during fires (Gutsell and Johnson 1996). Thus, large

individuals may be more likely to survive surface fires than smaller trees with equally

thick bark. Similarly, tall trees may disproportionally survive fires because their crowns

are less likely to be damaged by fire.

Given the predictions of increasing fire frequencies in the tropics, it is important to

elucidate the factors that render trees susceptible to fire-induced mortality. Here, I

conducted two sets of related experiments to improve our understanding of the roles of









plant traits in avoiding tree mortality. I first evaluate which bark traits best predict rates

of temperature change (R) of the vascular cambium during experimental bole heating. I

use the exact solution of Newton's Law of cooling to calculate R (Appendix C for the

reasoning behind the choice of Equation 3-1), a measure of how much conductive and

convective heat is transferred through bark over time (the inverse of cambium

insulation). My predictions are that: 1) bark thickness is inversely related to R and is the

most important bark trait in preventing the cambium temperature from rising to lethal

levels; 2) that R increases as a function of bark water content due to water's high

conductivity; and, 3) that bark density is negatively associated with R, given that the

greater the density, the lower diffusivity (Dickinson 2002; see Appendix C also).

While in the first set of analyses mentioned above I evaluate the important bark

traits for cambium insulation, in the second set of analyses I test which plant traits best

predict likelihood of fire-induced tree mortality, including R, wood density, tree height,

and diameter at breast height (dbh). I use data from a large-scale fire experiment

located in the southern Amazon to test the following predictions: 1) increasing tree dbh

and height decrease the likelihood of fire-induced tree mortality, not only because larger

individuals tend to have thicker bark, but also because they have traits that reduce the

indirect effects of fire; 2) decreasing R (higher cambium insulation) increases tree

survivorship; and, 3) fire treatments lead to higher tree mortality close to the forest edge

compared to the forest interior, given that fire intensity is expected to be higher at the

edge relative to the interior (Cochrane and Laurance 2002).









Methods

General Approach and Site Description

The study was conducted in a transitional forest in the southern part of the

Amazon Basin (13004'S, 52023') (Appendix D), where average annual precipitation is

1,800 mm and the dry season extends from May to September (Balch et al. 2008). To

evaluate the importance of different bark traits in preventing fire-induced tree mortality, I

conducted a series of field experiments. First, I heated the outer bark of trees with a

propane torch to relate heat transfer rates (R; inversely related to cambium insulation,

with units of C second-1) to bark traits. Second, I measured the bark thickness of 1,000

individuals representing 24 of the most common species in the study region to develop

a predictive model of bark thickness as a function of tree diameter at breast height

(dbh). Third, based on predicted bark thickness, I estimated R (referred as Rhat) for each

tree growing in the area of a large-scale fire experiment. Fourth, I tested several

statistical models of tree mortality as a function of Rhat under different experimental fire

regimes. Finally, to test the importance of different plant traits in predicting fire-induced

tree mortality, I compared the slope of models that included only Rhat with slopes from

models that included tree height, dbh, or wood density (derived from the literature;

Chave et al., 2006).

Modeling Cambium Insulation

During the dry season of 2008, cambium vulnerability to fire-induced damage was

estimated as the rate of change (R; units of C second-1) in cambium temperature (CT)

as a function of the air temperature adjacent to the outer bark, which was heated using

a torch. I first removed a 5 cm x 5 cm sample of bark at 25 cm above the ground from

each of 105 randomly sampled individuals, representing 18 species and ranging from









10-121 cm in dbh. These samples were taken to the lab for measurements of thickness

(BT), density (pb), and moisture content (BMC), following the procedures in Uhl and

Kauffman (1990). Immediately following bark removal, a thermocouple was inserted

between the inner bark and xylem (i.e., in the vascular cambium) and a second

thermocouple was placed on the surface of the outer bark. The air adjacent to the outer

bark of each tree was then heated using a propane torch coupled with a metal plate

placed in front of the flame to prevent direct thermocouple contact. Temperatures in the

vascular cambium and in the air adjacent to the outer bark were recorded every 10

seconds for 13 minutes, which allowed detection of changes in cambium temperatures

in even thick-barked species. Using an asymptotic regression model (eq. 1; see also

Appendix E), I estimated the rate of change (R in eq. 1) in CT as a function of

cumulative air temperature at the outer bark, which integrates the colinear variables of

time and air temperature adjacent to the outer bark into a single variable. To identify the

most important predictor of R, I compared models with predictors of BT, BMC, and pb

using the Akaike Information Criterion (AIC, Burnham and Anderson, 2002) and

hierarchical partitioning analysis. The latter method partitions the variance relative to R

shared by two predictors into the variance of R attributable to each predictor uniquely

(Murray and Conner 2009). Once the best model relating R and bark traits was

selected, it was used to predict R for each individual in the fire experiment, referred to

as Rhat. Note that the higher Rhat, the lower cambium insulation and the thinner the bark

thickness.



CT = Asymptote + (CT Asymptote)ecFeR (5-1)











Linking Tree Mortality and Cambium Insulation (R)

Data from a large-scale fire experiment was used to assess probabilities of fire-

induced tree mortality as a function of Rhat. In this experiment (Fig. 5-1), three plots of

50 ha each were established in 2004 in Mato Grosso, Brazil (details in Balch et al.

submitted): a control with no signs of recent fires; a plot that was experimentally burned

twice (2004 and 2007; plot B2); and, a plot that was experimentally burned four times

from 2004 to 2007 (plot B4). Within each of these plots, I conducted yearly mortality

censuses for individuals 10-19 cm dbh (N=6 transects of 10 m x 500 m per plot) and 20-

39 cm dbh (N= 6 transects of 20 m x 500 m per plot). All trees with dbh 2 40 cm were

assessed annually for mortality in all 50 ha plots. In total, 6,570 individuals of 97 tree

species were sampled for dbh (see Fig. 5-1). Height was measured for a subset of

these, with 4,071 individuals sampled. Wood density was derived from the literature for

87 species (Chave et al. 2006). Note that the northern edge of the experimental area

was adjacent to an open field (Appendix D).

Because some sampled trees were not charred during the experimental fires due

to burn heterogeneity across years (details in Balch et al. 2008), they were grouped into

the following categories, as summarized in Fig. 5-1:

Never charred and located in the control plot, in plot B2 (Charred Ox B2), or in
plot B4 (Charred Ox B4). Note that mortality in the control was expected to
be lower than in the burned plots even for trees not directly affected by the
experimental fire (i.e., a strong treatment effect) due to the high number of
falling trees in plots B2 and B4 and the associated changes in forest
microclimate and canopy damage.

Charred once (either in 2004 or 2007) and located in plot B2 or charred once at
any time between 2004 and 2007 and located in plot B4 (Charred 1x B4).
Note that the fires of 2004 (Charred 04) and 2007 (Charred 07) apparently









differed in intensity (see Appendix E), justifying their separation into two
distinct groups. In plot B4, however, this separation into groups was not
possible because of the large number of fire treatments and the small
sample size within each group for each species.

Charred twice and located either in plot B2 or in plot B4 (Charred 2x).

Charred three times and located in plot B4 (Charred 3x).

Charred four times and located in plot B4 (Charred 4x).



Given that BT could not be measured on all sampled individuals for logistical

reasons and because the measurements could increase mortality rates, I developed

predictive models of bark thickness (BT) for the 24 most common species in the region

based on dbh of 1,000 trees sampled in an adjacent area to the fire experiment (see

Appendix F). From BT, I then estimated a value for R for each individual of the 24 focal

species in the fire experiment area (N=3,752).

Modeling Tree Mortality

Tree mortality and cambium insulation

To model the probability of tree mortality as a function of Rhat (and therefore

assess the importance of Rhat in avoiding fire-induced mortality), I used logistic

regression mixed models with fixed and random effects of Rhat and species,

respectively. Because plot B2 was burned in 2004 and 2007, while plot B4 burned

annually from 2004 to 2007, mortality assessments were carried out at different times

among plots and over time. To deal with this complexity in the data set, I developed four

models based on a subset of the trees located in the fire experiment (Fig. 5-1):

Model 1 included trees that were potentially affected by one low-intensity fire and
were assessed for mortality after a period of one year following the
experimental fire of 2004 (period: 2004-2005; plots B1 and B4). Probability









of mortality was modeled as a function of Rhat interacting with five
treatments [Control, Charred Ox (plot B2 or B4), Charred 04 (plot B2 or B4)].

Model 2 included trees that were potentially affected by one low-intensity fire and
were assessed for mortality after a period of three years following the
experimental fire of 2004 (period: 2004-2007; plot B2). Probability of
mortality was modeled as a function of Rhat interacting with three treatments
(Control, Charred Ox, Charred 04)

Model 3 included trees that were potentially affected by at least one moderate- to
high-intensity fire and were assessed for mortality after a period of one year
following the experimental fire of 2007 (period: 2007-2008; plot B2). Note
that this group of trees included only those that were alive prior to the fires
of 2007 and that were not charred by the fires of 2004. Probability of
mortality was modeled as a function of Rhat interacting with four treatments
(Control, Charred Ox, Charred 07, Charred 2x).

Model 4 included trees that were potentially affected by several fires and were
assessed for mortality after a period ranging from 1 to 4 years following a
given experimental fire (period: 2004-2008). Probability of mortality was
modeled as a function of Rhat interacting with twelve treatments

To avoid bias due to spatial autocorrelation, I compared parameter estimates from

the models with and without a spatial error structure (glmer function in R Bates and

Maechler 2009; and glmmPQL function in R, Venables and Ripley 2002). Because the

inclusion of the spatial structure did not change the results, I present here only the

estimates derived from the glmerfunction (which is less likely to be biased Bolker et al.

2009). To account for the possible impact of proximity to an adjacent open field, the

distance from the edge was included as a co-variate in all models for each individual.

Given that Rhat was estimated from tree dbh, its association with mortality could simply

be due to a demographic effect in which smaller individuals (with high R) die at higher

rates than larger individuals, even in the absence of fire. To account for this potential

demographic effect in models, the influence of R on mortality was evaluated based on









the slope of the regression for mortality in the fire treatments contrasted with the slope of the

control. Therefore, if slopes for mortality in the control and fire treatments were equally

steep, R was considered to have no effect on fire-induced tree mortality.

Other correlates of fire-induced tree mortality

To compare the influences of Rhat, plant size (height and dbh), and wood density

on tree mortality, slopes of Model 4 were compared to the slopes of models that

included each plant trait as a single covariate. More specifically, the relative importance

of a given variable was based on the magnitude of the difference between slopes for

mortality of the control and burned treatments. All covariates of mortality were

standardized by two times the standard deviation of the entire population (Gelman and

Hill, 2007), making it possible to compare the magnitude of slopes across models and

treatments (note that slopes associated with Rhat are presented in absolute terms to

facilitate comparison with slopes of other models). The potential effects of spatial

autocorrelation were evaluated for all models but did not change the results (see section

above on Modeling Mortality). Note also that covariates of mortality were not included in

a single model because they were multi-collinear and had different sample sizes; when

all covariates were included in the same model, the degrees of freedom were reduced

substantially.

Results

Cambium Insulation (R)

Cambium temperatures (CT) were perhaps naively expected to increase steadily

until they reached an asymptote at the torch temperature (i.e., temperature adjacent to

the outer bark), but temperatures never exceeded 100"C because all the fire energy was

absorbed in the process of boiling water (i.e., latent heat of vaporization, Fig. 5-2).









Consequently, the presence of moisture in the bark not only prevented CT from rising

above 100 C, but also reduced the gradient in temperature (or introduced a step)

between the air adjacent to the outer bark and the cambium. For instance, the air

adjacent to the outer bark heated to 4000C for 10 minutes was expected to have

created a differential in temperature with the vascular cambium of 400"C minus the

initial cambium temperature (CTi). The presence of water in the bark, however,

constrained this differential to 100'C minus CTi. Despite the constraint imposed by

water, the amount of water in bark showed only a weak positive relationship with R (Fig.

5-2C). Bark thickness alone was the best predictor of R; as bark got thicker, the rates of

change in temperature through the bark decreased exponentially (Fig. 5-3). Inclusion of

pb and BMC in predictive models of R was not justified (Table 5-1), as the model with

only BT was the most parsimonious.

Table 5-1. Model comparison using corrected Akaike Information Criteria (AICc). In
this table, we also report the number of parameters (K), the adjusted R2, and
the AICc weight for each model. BT, pbT, BMC represent bark thickness,
wood specific gravity, and moisture content, respectively.
Models K Adj. R AAICc Weight

BT pb BMC 5 0.83 0.0 0.5
BT 3 0.83 1.3 0.3
BT + Pb 4 0.83 2.5 0.1

BT +BMC 4 0.83 2.8 0.1
BMC 3 0.14 146.5 0.0
Pb 3 0.03 156.7 0.0

Intercept 2 0.00 158.5 0.0
Intercept









Cambium Insulation and Tree Mortality

The relationships between Rhat and tree mortality varied among years and

treatments (Fig. 5-4). In the year following the low-intensity fire of 2004 (Model 1),

mortality was weakly associated with Rhat (and hence BT) and was similar between the

control and the burned plots (Fig. 5-4 and Appendix F; note that both standardized Rhat

and the associated BT are presented in this figure). Three years following the 2004 fires

(Model 2), survival in the control was higher than in the burned plots, but Rhat was still a

poor predictor of mortality (Fig. 5-4 and appendix G). These results indicate that the

fires of 2004 had minor effects on tree mortality compared to the control and that R was

a good predictor of fire-induced tree mortality.

While fire-induced tree mortality was poorly associated with Rhat following a single

low- to moderate-intensity fire, mortality was highly associated with Rhat during the year

following the fire of 2007 (Model 3; which was followed by the fire of 2004): as Rhat

increased, the probability of survival decreased (Fig. 5-4 and appendix H). This pattern

was strong and significant for trees charred twice but also for non-charred trees,

perhaps because smaller trees (with low cambium insulation) were more likely to be

damaged by fire-induced branch and tree fall than large trees (with high cambium

insulation) in the burned plots compared to the control. Hence, there was a strong

treatment effect in which tree mortality in the control plot was substantially lower than

mortality in the burned plots. However, there was a lack of relationship between Rhat

and mortality for trees charred by the fires of 2007, suggesting that one year after a fire

is perhaps too short an interval to evaluate fire-induced tree mortality. As expected,

mortality in the control plot showed no relationship with cambium insulation.









When the effects on mortality of all of the 12 fires were included in a single model

(Model 4), there was a positive relationship between tree mortality and R for trees

charred once, twice, and three times (Fig. 5-4). In other words, charred trees showed

strong, inverse relationships between mortality and cambium insulation for the period

2004-2008, but some new patterns emerged from this analysis. First, non-charred trees

and trees charred only in 2004 suffered high mortality across the entire range of R,

indicating a strong treatment effect of recurrent fires, but a minor effect of R for this

group of trees, which could be a result of an increased number of falling trees in the

burned plots killing non-charred trees. Second, trees that were charred 4 times showed

a high probability of surviving (e.g., similar to the control) and showed no relationship

with Rhat. This could be attributed to the fact that only the highly fire-resistant trees

survived four fires (see Balch et al. submitted).

The inclusion of distance from the edge was highly significant in all models. This

indicates that fire severity was higher at the forest edge adjacent to an open field than in

the forest interior.

Other Correlates of Tree Mortality

Overall, trees with higher wood density and larger dbh showed lower probability of

fire-induced mortality (Fig. 5-6), given that models with either wood density or dbh as a

covariate had steeper slopes for the fire treatments compared to the control (Fig. 5-5).

Further, the highest deviations from the control were observed for the fire treatments

Charred 1, 2, and 3x. Tree height was strongly and negatively associated with mortality

(Figs. 5-6), but mainly for non-charred trees and for trees in the control plot. In other

words, likelihood of survival increased as a function of increasing tree height at similar

rates between trees in the control and fire treatments. The exception to this pattern was









for trees burned during the fires of 2004 in plot B2, which had steeper slopes than trees

in the control as a function of increasing height (Fig. 5-6). In absolute terms, Rhat and

dbh were equally important in avoiding fire-induced tree mortality, while wood density

and tree height (in particular) were apparently less important. Overall, it is difficult to

separate the contribution of each variable to mortality, given that they are collinear and

have different samples sizes.

Discussion

Cambium Insulation

The findings from the heating experiment support the widely-accepted notion that

bark thickness is the single most important trait in insulating the vascular cambium and

thus in preventing fire-induced cambium damage (Dickinson, 2002; Uhl and Kauffman,

1990; Pinard and Huffman, 1997; Michaletz and Johnson, 2007). Bark thickness alone

accounted for 82% of the rate of change in cambium temperature (expressed as R) as a

function of cumulative air temperature on the outer bark. Nevertheless, the presence of

water in the bark also had substantial and potentially contrasting influences on R. On

one hand, water in the bark constrained bark temperatures to an asymptote at 100 C,

even when the temperature of the air adjacent to the outer bark was much higher than

100"C. This finding indicates that theoretical models of fire-induced cambial damage

that do not account for latent heat of vaporization in the bark (e.g., Dickinson, 2002)

could underestimate the differential in temperature between the fire and the cambium

and, in turn, underestimate the time before fire-induced cambial damage occurs. On the

other hand, the high conductivity of the water in the bark could increase bark diffusivity

compared to dry bark tissues, thus increasing R (i.e., faster increases in CT during

fires). In fact, it is suggested that the smaller the proportion of living tissues in the bark









(i.e., live inner vs. dead outer bark), the greater the cambium insulation, because dry,

corky bark has numerous air spaces that provide better fire insulation (i.e., lower heat

conductivity) than wet, dense phloem (Devet 1940 cited in Hare 1965). It is noteworthy

that the opposite result for trees in a mixed conifer forest has been reported (van

Mantgem and Schwartz 2003): the greater the proportion of bark composed by the

phloem, the lower CT, presumably due to the high heat capacity of the phloem. Based

on fig 2 in that study (p.346), nevertheless, it is reasonable to speculate that the latent

heat of vaporization was the major factor influencing the rates of increase in CT for

smaller trees (e.g., those with greater proportion of the bark composed by the phloem),

as opposed to heat conductivity or heat capacity [see Reifsnyder et al. (1967) for a

detailed discussion on the effects of BMC on heat capacity]. I suggest that the roles of

bark water content are important and should be studied in greater depth.

Bark Correlates with Tree Mortality

Although BT appears to be the most important plant trait in preventing fire-induced

cambial damage, its importance in preventing fire-induced tree mortality has not been

widely tested for tropical forest trees and may be modulated by other traits. For

example, during low intensity fires, heat is often distributed unevenly around the boles

of trees, causing only partial damage to the cambium around a small proportion of the

circumference of tree boles (Michaletz and Johnson, 2007). In such cases, plant traits

that reduce the indirect (and negative) influences of fire on the physiology and structure

of tree trunks may be even more important in avoiding tree mortality than BT alone,

which provides cambium protection against the direct effects of fire. Here, I show that

Rhat (and hence BT) was an important predictor of direct fire-induced tree mortality,

particularly for moderate to high-intensity fires. Trees experiencing the fire of 2007









(which followed 1, 2, or 3 fires) were extremely fire-resistant for BT 2 21 mm (i.e., 75-

100% probability of surviving) and fire-sensitive for BT 5 5.3 mm (0-25% probability of

survivorship). This threshold is in the same range as that suggested by Pinard and

Huffman (1997), who found that trees with BT 2 18 mm exhibited high resistance to

understory forest fires.

Despite the generally strong relationship between tree mortality and bark

thickness, I found no meaningful relationship between tree mortality and cambium

insulation for trees charred only once in either the 2004 or 2007 fires. Furthermore,

there were two common thin-barked species, Aspidosperma excelsum and Sloanea

escheri, that exhibited high survivorship even when charred (Balch et al. submitted).

These findings indicate that mechanisms associated with plant traits other than

cambium insulation could also influence fire-induced tree mortality. For example, wood

density has been shown to slow the progress of wood decay of damaged living trees

(Romero and Bolker 2008), which should reduce the likelihood of mortality after

cambium-exposing damage. Although our data suggest that wood density is important

for post-fire survival, the evidence is not compelling. Similarly, I found that tree mortality

decreased with increasing dbh. Although this relationship could result solely from the

strong and positive relationship between BT and tree size, an alternate (or concurrent)

explanation is that large trees are less likely to be killed by other falling trees and that

the likelihood of the entire circumference being exposed to cambium-killing

temperatures decreases with bole size (Gutsell and Johnson 1996). However, in this

study I was unable to separate the individual effects of dbh from bark thickness on

mortality. Finally, while tree height was apparently less important than other traits in









avoiding fire-induced mortality, height is expected to become more important during

intense fires that have the potential to damage tree crowns.

Forest Resistance to Fire

Given the minor effect of the 2004 fires on post-fire tree mortality, it is reasonable

to infer that this forest is relatively resistant to non-recurrent, low-intensity fires (as

measured by fire-induced tree mortality in individuals with dbh 2 10 cm; see Balch et al.

submitted for a discussion about smaller trees). This resistance may result from long-

term tree adaptations to cope with fire in the region. Although I am unaware of formal

estimates for fire return interval, transitional forests of Mato Grosso are flammable

during most dry seasons due to elevated vapor pressure deficits (Balch et al. 2008; see

also Blate 2005). Furthermore, they have been exposed to human-associated fire

ignition for millennia, given the presence of indigenous groups in the region who make

use of fire as a management tool (e.g., for hunting and agriculture). It is nevertheless

apparent that the fire-resistance of this tropical forest was reduced substantially by

recurrent fires, particularly near the forest edge adjacent to an open field. This edge

effect is likely related to higher fire intensity caused by drier fuels, to higher density of

small with low resistance to fire-induced cambium damage, or both. Also, changes in

forest microclimate alone could lead to increased fire-induced tree mortality (Cochrane

and Laurance 2002).

Predicted decreases in rainfall over portions of the tropics may cause extensive

forest dieback in the next 50 to 100 years, particularly in the eastern Amazon (Cox et al.

2000; Malhi et al. 2008). Synergistic effects of recurrent fires and forest degradation,

however, could drive this process much more rapidly (Nepstad et al. 2008). In

predicting the rate of forest loss due to recurrent fires, it is evident that while cambial









insulation is an important determinant of fire-induced tree mortality, other plant traits

also play important roles in avoiding fire-related death. Furthermore, the evidence that

fire-related mortality occurs even when tree boles do not exhibit physical charring

reminds us that fires influence several processes at the stand level. Examples of these

processes may include proximate fire-induced branch or tree fall, fire damage to roots

or leaves, or various indirect processes triggered by fire (e.g., disease, drought-related

death). I conclude that in-depth assessments of fire-related mortality in tropical forests

are needed for predicting the trajectories of Amazonian forests in the face of expected

increases in fire associated with frontier expansion and episodic droughts related to

climate change.













9 Transects Trees Affected by Different Number of Fires: Dbh size classes (cm)
(trees 10-39 cm dbh) Control Charred04(B) Charred2x(B) Charred 1x(C) Charred 3x(C)
Charred Ox(B) Charred 07(B) Charred Ox(C) Charred 2x(C) Charred 4x(C) 40 55 70 85 100 >15
Aoo
o *.'. .. O a -, e !1*
*.0 -9- 9 60
9* .9* 8 t o*




t. q ..,.P,, V "' +.^ o .i1 ,. ,. -. ". .
*1 00 099 a *9*10 baa,
I9 ,,. ,, .- .* 4 t 0 o l

e.: e y 9 4- < ,



S,,-,. oro,, -.I I, .' *r. 0,oe +**" + +r.. ,,'* .o 4
W 0- 9 vs* *:g *9 I # *
@741 ^- _* -., .. 0--ra e,2, i.- ** ,; **L',* 0 4.. 9'.
,.. ..* tr. ol --,,, 'o *0 e 9* *++e .,,-+ *," *' .. t. "I
Control* ~ U *lot B
9a a.. 0





0.g 5 00 10 150
o2.y004.$.. .90 a*n 27 and, 9o B4 (e yc




0m Z .Ae, o thaa a 0 t
S", .." a ". t -** .-. ., o










Curr Pl 82 Poat 84
(No Recent Fires) (Prescribed Fires in 2004 & 2007) (Prescribed Fires from 2004 to 2007! ----- T







times a given tree was charred. The size of symbols represents the dbh of each tree. Note that only trees 40
-,,. ., ,, ,..,, ., .I"_ -,-' "i, ,.. .., -.. .. ,., .,,.p .,.. .. ',
,"' O ",,o, '-,'+ I Or ., ,. .,.. ,, "'.
P r "- r e. Mr l *". .' ; ,, 'i. ['-"'' *'"
t-. ".- ]r, .., o +j.. .+;+ ,- +: e& .:' .,,, "..6
a,[ C ..oO"*% + ,'. +. + O ++ ._ "' +





500t 1000 1500*.
2~ ~Mete

Figure ~~9 5-1 Ma hwn h pta itrbto fteslctdi t hree 50 hapots plot A (Cotrl) pltBbundi
2004 an 07;ad ltB bre erybten204ad20) ahclrrpeet hubro
tims gve teewa chrrd.Th szeofsymol rprsetsth db o echtre.Noe tatony res 4
cm dbh wer censused acosteetr 5 a re 0c ee ape ln rnet bakarw)


100













100------ --------------

400
80-


o 300 a
S60-
CU)E

9200- 2. 4
E


I 00 -- --" ............. 20

Cambium Temp
-- Simulation 1
Simulation 2
0 0 -- Simulation 3

0 200 400 600 800 0 5000 15000 25000
Time (seconds) Cumulative HeatingC
Figure 5-2. Overall results from the bark heating experiment: A) air temperature
adjacent to the outer bark heated by a propane torch (black lines) and
cambium temperature (gray lines) as a function of time; B) cambium
temperature as a function of cumulative air temperature adjacent to the outer
bark (CTF). The dashed red lines in both figures represent 100 OC.
Simulations 1-3 represent three examples of R: lower quantile, median, and
upper quantile.


101













-0 -0 -0 O
A) R2: 82% B) o o
C0 0 0 0 0

-10- -10

-11 9 10 -
0 0 0 00
0 oo 0
-1 o o ao 0 80-10 o
-11 0 -11 o 0
0 o0 0
-12- ln(R)=-8.16566 -0.11971*BT -12

5 10 15 20 25 30 0.5 0.6 0.7 C
Bark Thickness (mm) Bark Density (g cm-3)
-8 C) o o o o a 100

idc-9-tes eie .effe of vral80-o
o o20o o




-11 o -
0-10 o o 0-20 60-


o O" I
0.4 0.6 0.8 1.0 1.2 1.4 Bark Thickness Bark Density Bark H20 (%)
Bark H20 (%)


Figure 5-3. Relationships between rates of change in cumulative heating (R) and A)
bark thickness, B) bark density, and C) bark moisture content. Figure D
indicates the independent effect of each variable on R.


102










Model 1 (2004-2005)
7TIM I


Model 2 (2004-2007) Model 3 (2007-2008)


Model 4 (2004-2008)


1001 -

901

70 "



60- -




harr d 7(B)
Legend Rt o

B- Charred 04(B) B i
c repres C 3?iB)Ch07nB dfen) te s Te f
30

Charred 2x(C)
20 Charred 3x(C)
Charred4x(C)
10

0-
0 IIln llllHI II EIUI Eg J HII IIImIIIInmml i imI mu m IIinI glnellmllll waIh I i amml m
-15 -1 0 -0.5 00 05 1.0 -1.5 -10 -0.5 00 05 10 -15 -1.0 -0.5 00 0.5 1.0 -1.5 -1.0 -0.5 00 0.5 10
Standardized Rate of Change Standardized Rate of Change Standardized Rate of Change Standardized Rate of Change

35 25 15 5 35 25 15 5 35 25 15 5 35 25 155
Bark Thickness (mm) Bark Thickness (mm) Bark Thickness (mm) Bark Thickness (mm)


Figure 5-4. Survival probabilities of trees with dbh 2 10 cm dbh as a function of Rhat, a proxy of cambium insulation
derived from the heating experiment. Each panel represents one logistic model (Models 1, 2, 3, and 4). Each
color represents the number of times a given trees was charred in different treatments. The first x-axis
represents the standardized rate of change (Rhat),while the second x-axis indicates the bark thickness
associated with a given Rhat (note that R and bark thickness are inversely related).


103


F T 111111


,,












-' I T


T Ii T liT1 IT
_IT' : tt i II tt
- -- : ----- --- ----r--I -


Height
-U-- db[
---Wo d cersity


Figure 5-5. Relative influence of Rhat, tree height, dbh, and wood density on tree
mortality for different fire treatments. On the x-axis, the control (A) is
represented by green, B2 by orange, and B4 by blue. Note that the influence
of each variable was inferred from on the magnitude of the difference
between slopes of the control and fire treatments. Note also that all co-
variates of mortality were standardized (centered and then divided by 2 times
the standard deviation) and that R is presented in absolute terms for
comparison with other plant traits.


104














Woo Denstiy


Treatments
m Control
M Charred x(B)
m Charred 04(B)
m Charred 07(B)
SCharred 2I(B)
S Charred Ox(C)
Charred 1 x(C)
m Charred 2x(C)
m Charred 3x(C)
I Charred 4x(C)


J/


Standardized Variable


Figure 5-6. Probability of tree survival as a function of standardized tree height (left
panel), dbh (middle panel), and wood density (right panel). Each color
represents a different fire treatment.


105


Trea Height









CHAPTER 6
CONCLUSION

The synergistic effects of drought, forest degradation by logging, landscape

fragmentation, and increased fire ignition already exert strong influences on the stability

of tropical forests. In combination with climate change, they may cause large portions

of tropical forests to suffer dieback events, with important implications for forest carbon

stocks, product yields, and biodiversity. Yet, very little is known about the thresholds in

environmental stresses beyond which regime shifts may occur. In my dissertation, I

assessed the vulnerability of Amazonian forests to severe drought and fire events, with

focus on the C cycle.

I first report on findings from a partial throughfall experiment located in a moist,

eastern Amazonian forest. In particular, I show that wood increment, net primary

productivity (NPP), and leaf area index (LAI) all declined in response to the drought

treatment of 2-3 years. Following protracted water stress, substantial increases in

mortality of large trees were observed, corresponding to sustained reductions in plant

available water. The results from this experiment suggest that, if the general response

of Amazonian forests to drought is similar to those recorded at the Tapajos Forest, a

drier future climate is likely to have major consequences to carbon stocks of Amazonian

forests (e.g., reductions of 25% of standing live biomass). This view receives support

from findings of another throughfall exclusion experiment located in a moist, eastern

Amazon forest (see Meir et al. 2009 for a review of these experimentss, where rates of

drought-induced tree mortality following 3-4 years of throughfall exclusion were reported

within the ranges I found in my dissertation.


106









Whereas the throughfall experiments mentioned above provide new and important

insights into drought-vegetation relations in Amazonia, they cover a relatively small area

and may not represent the spatial variability of forests' susceptibility to drought across

the Amazon. To deal with this limitation of field experiments, several studies took

advantage of the 2005 drought (one of the most severe droughts over the past 100

years) to evaluate how forests respond drought over a greater spatial extent. Based on

results from a network of permanent plots spread across the Amazon, Phillips et al.

(2009) found that several forests were sensitive to the 2005 drought, with mortality

increasing at rates that were high enough to reverse a trend in C accumulation of 0.45

PgC yr1 to a -1.2-1.6 PgC annual source. The results in Phillip et al. (2009) appear to

substantiate the notion that Amazonian forests are, overall, sensitive to severe, episodic

droughts. Also, these results indicate that, if droughts become more intense and

frequent in the future, large amounts of carbon will be released to the atmosphere.

Despite the apparently agreement among studies that Amazonian forests are

drought-sensitive, there is still no agreement about the thresholds of drying of the basin

that would cause regime shifts. In a controversial article, Samanta et al. (2010) claimed

to debunk some conclusions from most recent IPCC report that portions of dense

forests could "react drastically" to "slight reductions" in precipitation in the near future

resulting from climate change. More specifically, Samanta et al. 2010 used satellite-

based measurements of vegetation photosynthetic activity from MODIS to show that

Amazonian vegetation remained relatively irresponsive to the drought encountered in

2005. These results contrast with previous studies that assessed forest responses to

the drought of 2005; studies that showed that vegetation over Amazonia either became


107









more productive (Saleska et al. 2007) or experience elevated tree mortality (Phillips et

al. 2009). Given these apparently conflicting results among different studies of the

climate-productivity relationship, and the importance of these results for understanding

the impacts of climate change on the terrestrial carbon cycle, there was clearly a need

to properly quantify the mechanisms driving the variability of vegetation productivity of

the world's tropical forests. Thus, in an effort to reconcile results from these field- and

satellite-based studies, I conducted two experiments.

First, using a CARLUC process-based model, I estimated the effects of climate

variability on total carbon stocks and fluxes of vegetation across the Basin and of

forests where permanent plots from RAINFOR are located. I hypothesized that

CARLUC simulations, which were driven by climate variability alone, would describe C

accumulation over the Amazon from 1998 to 2004, but C losses during the 2005

drought (Phillips et al. 2009). While CARLUC simulated increases in C pools during the

dry season from 1995 to 2004, followed by a reduction in 2005, the magnitude of the

reduction reported in my dissertation was less pronounced than in Phillips et al. (2009).

One possible explanation for this difference between studies is that the Tapajos Forest

(where the mortality function in CARLUC was developed) was more drought resistance

than RAINFOR forests on average.

Second, I conducted a study that combined analysis of climate data, field

measurements of forest photosynthesis, and an improved satellite-based measure of

forest photosynthetic activity (EVI). From this study, I show that none of the climatic

variables tested (precipitation, vapor pressure deficit, photosynthetically active radiation)

contributed to explaining inter-annual variability of EVI across the Amazon. Instead, I


108









found evidence that suggest that production of new leaves could play an important role

in Basin-wide, inter-annual EVI variability, but not necessarily in changes in GPP or

NPP. These results suggest that it is still unclear whether EVI can be employed as a

tool to measure the vulnerability of tropical forests to drought. Also, they suggest that

several mechanisms could operate simultaneously to explain the contrasting results

between field-based studies and those that use satellite-based measurements to

assess forests' vulnerability to drought. For example, GPP (expressed as EVI) appears

to increase due to the production of new leaves and increased PAR during droughts,

whereas aboveground net primary productivity (ANPP) measured in the field

concurrently decreased because of higher tree mortality and increased respiration

associated with lower PAW and higher temperatures. Moreover, the allocation of non-

structural carbohydrates to belowground processes may have increased, given that

GPP was higher and ANPP lower during the drought of 2005. It is also possible that

MODIS cannot detect deaths of the large trees that store large amounts of carbon but

that account for only a small portion of the canopy. This should be the topic of further

research.

After evaluating the vulnerability of Amazonian forests to drought, I studied the

mechanisms associated with fire-induced tree mortality in a tropical forest in the

southern Amazon. Based on this study, I present two novel findings that improve our

ability to characterize tropical forest vulnerability to fire. First, I found that the presence

of water in the bark reduced the differential in temperature between the fire and

vascular cambium due to boiling water in the bark. This result is important because it

shows that theoretical models of fire-induced damage that do not account for the effects


109









of latent heat vaporization (boiling water) in the bark substantially underestimate the

time before fire-induced cambial damage occurs. Second, I show that the experimental

fires influenced mortality through several processes that were not directly related to

cambium protection against fire-induced damage, suggesting that plant traits that

reduce the negative and indirect effects of fire on post-fire tree mortality should also be

considered in the assessments of tropical forests' vulnerability to fire.

In conclusion, I found evidence that droughts and fires exert strong influences on

the terrestrial carbon cycling, but no evidences for imminent regime shifts in the region

driven by climate change alone. However, human activities that reduce forest resistance

to fire and in turn increase the likelihood of exotic grass invasions into forests may

change these conclusions. Poor timber harvest practices, for example, can substantially

increase canopy openness of tropical forests, which elevates forest flammability by i)

promoting sunlight-induced drying of fuel loads and ii) increasing rates of fuel load

accumulation on the forest floor. A thin canopy can also support the establishment of

invasive C4 grasses from forest-neighboring pastures, which can out-compete native

vegetation and elevate forest flammability permanently, potentially causing a shift in

stable state from tree to grass dominance. Because the two most common of the

intensive land-use practices in the Amazon, cattle ranching and swidden agriculture,

both use fire as a management tool, ignition sources for forest fires are common

(Nepstad et al. et al. 2001). In an extreme scenario, these feedbacks among fire,

vegetation, and climate can drive large portions of tropical forests to impoverished

systems.


110











APPENDIX A
METEOROLOGICAL STATIONS


Sg
U)
C

[0

E
z:

0


0 50 100 150 200
Classes of Precipitation for July-September (mm)


Figure A-1 Upper panel: Number of meteorological stations per classes of precipitation
(average between July and September); the dashed-red line separates the
number of mestations with (left) and without (right) a dry season. Note that
according to the standard definition of dry season as the average precipitation <
100 mm month-1, the period July-September chosen in this study captured the dry
season in 76% of the meteorological stations (212 of 280). For the 68 remaining
stations (24%), the average precipitation from July to September was not
representative of the driest months of the year in 37 of them (13% of total)
concentrated in the Northern Amazon (black stars).


111









APPENDIX B
VARIABILITY IN VEGETATION INDEX

Here I assessed whether the choice of months (July-Sept) could have influenced

the spatial patterns of EVI (West-East and North-South), given the potential for lags in

vegetation phenology (e.g., leaf flushing) across the wide ranges of latitude and

longitude reported in this study. For instance, one could hypothesize that leaf flushing

occurs predominantly in July-October in the East Amazon, October-November in the

Central Amazon, and November/December in the Western Amazon; temporal-spatial

patterns that could create trends of lower EVI from East to West of the Amazon.

Xiao et al. (2006) mapped the months with highest EVI (calculated from MODIS

collection 5 reflectances) across the Amazon in 2002 (Fig. 5b in Xiao et al. 2006). If

spatial patterns of EVI were closely associated with gradients in vegetation phenology,

the results from Xiao et al. (2006) are likely to show a consistent shift in the month of

highest EVI from West to East and/or South to North of the Amazon. Based on Fig.5b

from that study, I detected no evidence of gradients in vegetation phenology that could

explain the strong gradients found in our study.

Because Xiao et al. (2006) mapped peak EVI in only one year, I repeated their

analysis using the GIMMS AVHRR NDVI data set from 1981 to 2008 (Figure A2a; upper

panel) (Tucker et al. 2005). The GIMMS data generally confirmed the lack of a clear

spatial pattern in the timing of peak EVI/NDVI in the Amazon with one exception: there

was a potential gradient in the month with highest NDVI from South (earlier in the dry

season) to North (later in the dry season/early in the wet season). To make sure that

this potential gradient did not influence our conclusions, all statistical models between

EVI and climatic variables, for both forest and non-forested areas (Figure Alb; lower


112










panel), included longitude and latitude as co-variates. While we do not exclude the

possibility that the focus on the July-September period may have had some effect on

the spatial patterns of EVI, we are confident that the gradients in NBAR-EVI from West

to East were not merely an artifact of spatial phenological gradients. Also, the inclusion

of latitude and longitude in the statistical models between EVI and climatic variables

minimized any potential effect of the choice of month in our conclusions.


-II early Jan m early Jul
EI late Jan | j i
early Feb I early Aug
late Feb L late Aug
early Mar L early Sep
late Mar late Sep
earlyApr I early Oct
lateApr J

Slate May late Nov
Early Jun early Dec
Slate Jun late Dec


Figure A-2 The month of peak NDVI as recorded in the GIMMS-NDVI dataset between 1981
and 2008 for areas of high canopy cover (A) and by a wide range fractions of canopy cover
(B).


113


} a)





.' ^ '
.-1 ,.
, ,''; ';" ,, J... o' -




I
C Zu. ",.


'\ '^
i2



-n *









APPENDIX C
EQUATIONS ON HEAT TRANSFER

While the equation representing conductive heat transfer through bark (one-

dimensional model; equation C1) can be useful for predicting the rates of change in

cambium temperature (a proxy of cambium insulation) as a function of fire temperature,

this equation requires information about diffusivity, which is very difficult to measure

under field conditions and goes beyond of the scope of this work. Also, equation 1A

requires a constant heat source, which is challenging under field conditions due to air

movement during heating experiments. Thus, I decided to use Newton's Law of Cooling

to calculate the rate of change in cambium temperature given a gradient in temperature

(equation C2). In so doing, I could not separate the individual effects of conductivity and

convection. Note that the exact solution of Newton's Law of Cooling (equation 1C) is an

asymptotic model (i.e., the maximum temperature the cambium can reach is equal to

the hot air temperature on the outer bark). However, I used basic theory of conductive

heat transfer to drive my predictions (1, 2, and 3).

equation C1:

Ta Tf re ( BT
Toap- T 'fire 0)

where:

Tcamb = Cambium temperature (CT)

Tfire = Temperature of the flame

Tcambi = Initial temperature at the cambium

BT= Bark Thickness

a diffusivityy): heat conduction/(heat capacity x bark density)


114









t = time

Based on this relationship, the time for mortality at different temperatures (Tf) can

be predicted for an individual:

time = constant*Bt2



equation C2

dt
= I- R(CT-Ty)

whree:

CT: Cambium temperature

R: Rate of change in temperature

Tf: Hot air temperature adjacent to the bark heated using a torch.

equation 1C (exact solution of equation C2):

CT = Asymnptote + (CT Asymptote)e-CFTeR

Where:

Asymptote: Tf in equation 1 B

CTF: cumulative air temperature adjacent to the bark


115









APPENDIX D
FIRE EXPERIMENT


Figure A-3 Aerial photo of the experimental fire. On the right corner of the figure I show
the location of this experiment, Querencia, Mato Grosso State, Brazil.


116











APPENDIX E
FLAME HEIGHT


0
CD



0)
01U 0-
Cq
E
(U
LI-

aC-




0 -


2004 (B2) 2004 (B4)


2005 (B4) 2006 (B4) 2007 (B2) 2007 (B4)
Fire Treatments


Figure A-4 Box-plots for flame height in plot B2 (grey) and B4 from 2004 and 2007.
Measurements were taken along the fire line and thus were spread across the
experimental area in a pattern that depended on the fire behavior.


117


I I


_I I
1
















APPENDIX E

BARK THICKNESS AND TREE DBH


Amaoua guianens s

SA 'A R2=0.04
AAA. -& A A |




10 15 20 25
dbh (cm)
Mtcropholis egensis



A
A i



10 12 14 16 18
dbh (cm)
Trafftnmckia glaziovi

R =0.64 A
A A




10 20 30 40 50
dbh (cm)
Sacoglotts guianensis

R2=0.4 A


A A


10 15 20 25 30 35
dbh (cm)
Termmala fetraphylla

R =0.66





20 30 40 50
dbh (cm)


Ocotea acutangula

12 R =0.43 AA
10 A

0 -- t t t A


20 40 60 80 100
dbh (cm)
Tramnnickia bursert/ola


20 30 40 50 60
dbh (cm)
Dacryodes micfrcarpa

30 R=0.5 AA
20 z 4

10- a-


10 15 20 25 30 35 40


dbh (cm)
Micona punctata
16- E




0 E
6 2


6A
8 i ------ i iS
15 20 25 30 35 40
dbh (cm)
Nectandra cuspidata
40 E
3 R2=0.81
30


10 A ~

20 40 60 80 100 120
dbh (cm)


E 9
Aspidospenna exce-sum
8
7


SA
3A

10 11 12 13 14 15 16
dbh (cm)
Sloanea etchlen
8 ----



=A
F6 6
|2 4 AA aA A


10 15 20 25 30 35
dbh (cm)
Xylopla amazomca
E 40
8 35 R2=0.23 A
S30
25




10 20 30 40 50 60


dbh (cm)
Chaetocarpus schomburgkranus


6 AA -

4 tA A


10 15 20 25 30
dbh (cm)
Mrcona pynfolia
20
1 R2=0.73

105 A


10 20 30 40 50 60
dbh (cm)


10 15 20 25
dbh (cm)
Tratinmckia fhofolka
EF40 F
E 35 A

=30 =
|25
S20 A
l5 A1 a A

10 15 20 25
dbh (cm)
Myrca multiflora
E E








dbh (cm)
Mounn bgchyanthora
20 R2=0.417
S -1 ,
10 A

i 4-' i






10 12 14 1 1 20
dbh (cm)
Casouna brachyanera

20 R2-0.64




15
10


10 20 20 40






dbh (cm)
I 15A A^ *n

10 20 30 40


dbh (cm)


Tapitra gumanensis
14
12 R2=A31

8 A A --


4
10 20 30 40 50
dbh (cm)
Pouteta ramitfora

R2=0.58
15 A A

10



10 15 20 25
dbh (cm)
Vochysma vismnfolma
30
25 R2=0.64 A
20

150- A
5-

10 15 20 25 30 35 40
dbh (cm)
Sclerolobium paniculatum
251 R=0.74 A
20





10 20 30 40 50 60 70
dbh (cm)


Figure A-5 Relationships between dbh and bark thickness. Note that when there was no relationship between dbh and

BT for a given species, predicted BT was based on the intercept of the model.







118









APPENDIX F
MODEL 1



Table F-1. Model 1. Period: 2004-2005 (plot B2 and B4)


Intercept
R
Charred.Ox (B2)
Charred.04 (B3)
Charred.Ox (B4)
Charred.04 (B4)
Edge Distance
R:Charred.Ox (B2)
R:Charred.04 (B2)
R:Charred.Ox (B4)
R:Charred.04 (B4)


Estimate Std. Error z value
3.461 0.208 16.670
-0.615 0.373 -1.647
-0.610 0.219 -2.791
-0.812 0.184 -4.403
-0.351 0.225 -1.560
-0.597 0.196 -3.052
0.655 0.142 4.604
0.460 0.463 0.993
-0.435 0.438 -0.991
-0.364 0.503 -0.724
-0.923 0.457 -2.017


119


Pr(> z|)
0.000
0.100
0.005
0.000
0.119
0.002
0.000
0.320
0.322
0.469
0.044









APPENDIX G
MODEL 2

Table G-1. Model 2. Period: 2004-2007 (plot B2)


Estimate Std. Error zvalue Pr(>lzl)
Intercept 2.526 0.222 11.391 0.000
R 0.242 0.267 0.907 0.364
Charred.0x(B2) -0.340 0.162 -2.103 0.035
Charred.04(B2) -0.609 0.133 -4.594 0.000
EdgeDistance 0.476 0.133 3.582 0.000
R: Charred. 0x(B2) -0.250 0.332 -0.755 0.450
R: Charred. 04(B2) -0.389 0.293 -1.327 0.185


120









APPENDIX H
MODEL 3

Table H-1. Model 3. Period: 2007-2008 (plot B2)


Estimate Std. Error zvalue Pr(>lzl)
Intercept 3.257 0.233 13.965 0.000
R -0.494 0.338 -1.461 0.144
Charred.0x(B2) -1.289 0.425 -3.035 0.002
Charred.07(B2) -2.190 0.170 -12.885 0.000
Charred.2x(B2) -2.308 0.154 -14.962 0.000
Edge Distance 1.303 0.138 9.462 0.000
R:Charred. x(B2) -1.703 1.007 -1.691 0.091
R: Charred. 07(B2) -0.468 0.372 -1.257 0.209
R:Charred.2x(B2) -1.339 0.366 -3.656 0.000


121









LIST OF REFERENCES


Aide, T. M. 1993. Patterns of Leaf Development and Herbivory in a Tropical Understory
Community. Ecology 74:455-466.

Alencar, A., D. Nepstad, and M. C. Diaz. 2006. Forest understory fire in the Brazilian Amazon
in ENSO and non-ENSO years: area burned and committed carbon emissions. Earth
Interactions 10:1-17.

Alencar, A. A., L. A. Sol6rzano, and D. C. Nepstad. 2004. Modeling forest understory fires in an
eastern Amazonian landscape. Ecological Applications 14:139-149.

Aragao, L. E. O. C., Y. Malhi, R. M. Roman-Cuesta, S. Saatchi, L. O. Anderson, and Y. E.
Shimabukuro. 2007. Spatial patterns and fire response of recent Amazonian droughts.
Geophysical Research Letters 34:-. doi: Artn L07701 Doi 10.1029/2006g1028946.

Aragao, L., Y. Malhi, N. Barbier, A. Lima, Y. Shimabukuro, L. Anderson, and S. Saatchi. 2008.
Interactions between rainfall, deforestation and fires during recent years in the Brazilian
Amazonia. Philosophical Transactions of the Royal Society B: Biological Sciences
363:1779-1785.

Baayen, H. 2008. Analyzing linguistic data: A practical introduction to statistics using R.
Cambridge University Press.

Baker, T. R., O. L. Phillips, Y. Malhi, S. Almeida, L. Arroyo, A. Di Fiore, T. Erwin, N. Higuchi,
T. J. Killeen, S. G. Laurance, W. F. Laurance, S. L. Lewis, A. Monteagudo, D. A. Neill,
P. N. Vargas, N. C. A. Pitman, J. N. M. Silva, and R. V. Martinez. 2004a. Increasing
biomass in Amazonian forest plots. Philos. Trans. R. Soc. Lond. Ser. B-Biol. Sci.
359:353-365.

Baker, T. R., O. L. Phillips, Y. Malhi, S. Almeida, L. Arroyo, A. Di Fiore, T. Erwin, N. Higuchi,
T. J. Killeen, S. G. Laurance, W. F. Laurance, S. L. Lewis, A. Monteagudo, D. A. Neill,
P. Nfifez Vargas, N. C. A. Pitman, J. N. M. Silva, and R. Vasquez Martinez. 2004b.
Increasing biomass in Amazonian forest plots. Philosophical Transactions of the Royal
Society B: Biological Sciences 359:353-365. doi: 10.1098/rstb.2003.1422.

Balch, J., D. Nepstad, P. Brando, L. Curran, O. Portela, O. de Carvalho, and P. Lefebvre. 2008.
Negative fire feedback in a transitional forest of southeastern Amazonia. Global Change
Biology 14:2276-2287.

Barlow, J., T. Haugaasen, and C. A. Peres. 2002. Effects of ground fires on understorey bird
assemblages in Amazonian forests. Biological Conservation 105:157-169.

Barlow, J., B. O. Lagan, and C. A. Peres. 2003. Morphological correlates of fire-induced tree
mortality in a central Amazonian forest. Journal of Tropical Ecology 19:291-299.


122









Barlow, J., and C. A. Peres. 2008. Fire-mediated dieback and compositional cascade in an
Amazonian forest. Philosophical Transactions of the Royal Society B: Biological
Sciences 363:1787.

Barlow, J., and C. A. Peres. 2006. Effects of single and recurrent wildfires on fruit production
and large vertebrate abundance in a central Amazonian forest. Biodiversity and
Conservation 15:985-1012.

Bates, D., and M. Maechler. 2009. lme4: Linear mixed-effects models using S4 classes. R
package version 0.999375-31.

Bates, D., and D. Sarkar. 2006. lme4: Linear mixed-effects models using S4 classes. URL
http://CRAN. R-project. org, R package version 0.99875-8.

Betts, R. A., P. M. Cox, M. Collins, P. P. Harris, C. Huntingford, and C. D. Jones. 2004. The role
of ecosystem-atmosphere interactions in simulated Amazonian precipitation decrease and
forest dieback under global climate warming. Theoretical and Applied Climatology 78.
doi: 10.1007/s00704-004-0050-y.

Betts, R., M. Sanderson, and S. Woodward. 2008. Effects of large-scale Amazon forest
degradation on climate and air quality through fluxes of carbon dioxide, water, energy,
mineral dust and isoprene. Philosophical Transactions of the Royal Society B: Biological
Sciences 363:1873-1880.

Blate, G. M. 2005. Modest trade-offs between timber management and fire susceptibility of a
Bolivian semi-deciduous forest. Ecological Applications 15:1649-1663.

Bolker, B. M., M. E. Brooks, C. J. Clark, S. W. Geange, J. R. Poulsen, M. H. Stevens, and J. S.
White. 2009. Generalized linear mixed models: a practical guide for ecology and
evolution. Trends in Ecology & Evolution 24:127-135.

Bonan, G. B. 2008. Forests and climate change: forcing, feedbacks, and the climate benefits of
forests. Science 320:1444.

Brando, P. M., D. C. Nepstad, E. A. Davidson, S. E. Trumbore, D. Ray, and P. Camargo. 2008b.
Drought effects on litterfall, wood production and belowground carbon cycling in an
Amazon forest: results of a throughfall reduction experiment. Philosophical Transactions
of the Royal Society B: Biological Sciences 363:1839-1848.

Burnham, K. P., and D. R. Anderson. 2002. Model selection and multimodel inference: a
practical information-theoretic approach. Springer Verlag.

Chambers, J. Q., J. Santos, R. J. Ribeiro, and N. Higuchi. 2001. Tree damage, allometric
relationships, and above-ground net primary production in central Amazon forest. Forest
Ecology and Management 152:73-84.


123









Chapin III, F. S., E. Schulze, and H. A. Mooney. 1990. The ecology and economics of storage in
plants. Annual Review of Ecology and Systematics 21:423-447.

Chave, J., H. C. Muller-Landau, T. R. Baker, T. A. Easdale, H. Steege, and C. O. Webb. 2006.
Regional and phylogenetic variation of wood density across 2456 neotropical tree
species. Ecological Applications 16:2356-2367.

Christensen, J. H., B. Hewitson, A. Busuioc, A. Chen, X. Gao, I. Held, R. Jones, R. K. Kolli, W.
T. Kwon, R. Laprise, and others. 2007. Climate change 2007: the physical science basis.
Contribution of Working Group I to the Fourth Assessment Report of the
Intergovernmental Panel on Climate Change. Cambridge University Press Cambridge,
New York.

Clark, D. A. 2002. Are tropical forests an important carbon sink? Reanalysis of the long-term
plot data. Ecological Applications 12:3-7.

Clark, D. A., and D. B. Clark. 1994. Climate-induced annual variation in canopy tree growth in a
Costa Rican tropical rain forest. Journal of Ecology 82:865-872.

Clark, D. A., S. Brown, D. W. Kicklighter, J. Q. Chambers, J. R. Thomlinson, and J. Ni. 2001.
Measuring net primary production in forests: concepts and field methods. Ecological
Applications 11:356-370.

Cochrane, M. A., and W. F. Laurance. 2002. Fire as a large-scale edge effect in Amazonian
forests. Journal of Tropical Ecology 18:311-325.

Cochrane, M. A., and W. F. Laurance. 2008. Synergisms among fire, land use, and climate
change in the Amazon. AMBIO: A Journal of the Human Environment 37:522-527.
Cochrane, M. A. 2003. Fire science for rainforests. Nature 421:913-919.

Cochrane, M. A., and W. F. Laurance. 2002. Fire as a large-scale edge effect in Amazonian
forests. Journal of Tropical Ecology 18. doi: 10.1017/S0266467402002237.

Condit, R., S. Aguilar, A. Hernandez, R. Perez, S. Lao, G. Angehr, S. P. Hubbell, and R. B.
Foster. 2004. Tropical forest dynamics across a rainfall gradient and the impact of an El
Nino dry season. Journal of Tropical Ecology 20:51-72.

Condit, R., S. P. Hubbell, and R. B. Foster. 1995. Mortality rates of 205 neotropical tree and
shrub species and the impact of a severe drought. Ecological Monographs:419-439.

Costa, M. H., and J. A. Foley. 2000b. Combined effects of deforestation and doubled
atmospheric C02 concentrations on the climate of Amazonia. Journal of Climate 13:18-
34.


124









Cox, P. M., R. A. Betts, M. Collins, P. P. Harris, C. Huntingford, and C. D. Jones. 2004.
Amazonian forest dieback under climate-carbon cycle projections for the 21st century.
Theoretical and Applied Climatology 78. doi: 10.1007/s00704-004-0049-4.

Cox, P. M., P. P. Harris, C. Huntingford, R. A. Betts, M. Collins, C. D. Jones, T. E. Jupp, J. A.
Marengo, and C. A. Nobre. 2008. Increasing risk of Amazonian drought due to
decreasing aerosol pollution. Nature 453:212-215.

Cox, P. M., R. A. Betts, C. D. Jones, S. A. Spall, and I. J. Totterdell. 2000. Acceleration of
global warming due to carbon-cycle feedbacks in a coupled climate model. Nature
408:184-187. doi: 10.1038/35041539.

Cramer, W., A. Bondeau, F. I. Woodward, I. C. Prentice, R. A. Betts, V. Brovkin, P. M. Cox, V.
Fisher, J. A. Foley, and A. D. Friend. 2001. Global response of terrestrial ecosystem
structure and function to CO2 and climate change: results from six dynamic global
vegetation models. Global Change Biology 7:357-373.

Davidson, E. A., F. Y. Ishida, and D. C. Nepstad. 2004. Effects of an experimental drought on
soil emissions of carbon dioxide, methane, nitrous oxide, and nitric oxide in a moist
tropical forest. Global Change Biology 10:718-730.

Davidson, E. A., L. V. Verchot, J. H. Cattanio, I. L. Ackerman, and J. E. M. Carvalho. 2000.
Effects of soil water content on soil respiration in forests and cattle pastures of eastern
Amazonia. Biogeochemistry 48:53-69.

Davidson, E. A., I. A. Janssens, and Y. Luo. 2006. On the variability of respiration in terrestrial
ecosystems: moving beyond Q10. Global Change Biology 12:154-164.

Dickinson, M. & Johnson, E. (2001) Fire effects on trees. In: Forest fires: behavior and
ecological effects, Academic Press, 2001, 477-521

Doughty, C. E., and M. L. Goulden. 2008. Seasonal patterns of tropical forest leaf area index and
CO2 exchange. J Geophys Res 113.

Fisher, R. A., M. Williams, A. L. da Costa, Y. Malhi, R. F. da Costa, S. Almeida, and P. Meir.
2007. The response of an Eastern Amazonian rain forest to drought stress: results and
modelling analyses from a throughfall exclusion experiment. Global Change Biology
13:2361-2378.

Fisher, R., M. Williams, R. Lobo Do Vale, A. Lola Da Costa, and P. Meir. 2006. Evidence from
Amazonian forests is consistent with isohydric control of leaf water potential. Plant, cell
and environment(Print) 29:151-165.

Foley, J. A., S. Levis, I. C. Prentice, D. Pollard, and S. L. Thompson. 1998. Coupling dynamic
models of climate and vegetation. Glob. Change Biol. 4:561-579.


125









Foley, J. A., I. C. Prentice, N. Ramankutty, S. Levis, D. Pollard, S. Sitch, and A. Haxeltine.
1996. An integrated biosphere model of land surface processes, terrestrial carbon balance,
and vegetation dynamics. Glob. Biogeochem. Cycle 10:603-628.

Gerwing, J. J., and D. L. Farias. 2000. Integrating liana abundance and forest stature into an
estimate of total aboveground biomass for an eastern Amazonian forest. Journal of
tropical ecology 16:327-335.

Gloor, M., O. L. Phillips, J. J. Lloyd, S. L. Lewis, Y. Malhi, T. R. Baker, G. L6pez-Gonzalez, J.
Peacock, S. Almeida, A. C. de OLIVEIRA, and others. 2009. Does the disturbance
hypothesis explain the biomass increase in basin-wide Amazon forest plot data? Global
Change Biology 15:2418-2430.

Golding, N., and R. Betts. 2008. Fire risk in Amazonia due to climate change in the HadCM3
climate model: Potential interactions with deforestation. Global Biogeochemical Cycles
22.

Grace, J., J. Lloyd, J. McIntyre, A. C. Miranda, P. Meir, H. S. Miranda, C. Nobre, J. Moncrieff,
J. Massheder, and Y. Malhi. 1995. Carbon dioxide uptake by an undisturbed tropical rain
forest in southwest Amazonia, 1992 to 1993. Science 270:778.

Graham, E. A., S. S. Mulkey, K. Kitajima, N. G. Phillips, and S. J. Wright. 2003b. Cloud cover
limits net C02 uptake and growth of a rainforest tree during tropical rainy seasons.
Proceedings of the National Academy of Sciences 100:572-576.

Gutsell, S. L., and E. A. Johnson. 1996. How fire scars are formed: coupling a disturbance
process to its ecological effect. Canadian Journal of Forest Research 26:166-174.

Hansen, M. C., R. S. DeFries, J. R. G. Townshend, M. Carroll, C. Dimiceli, and R. A. Sohlberg.
2003. Global percent tree cover at a spatial resolution of 500 meters: First results of the
MODIS vegetation continuous fields algorithm. Earth Interactions 7:1-15.

Hare, R. C. 1965. Contribution of bark to fire resistance of southern trees. Journal of Forestry
63:248-251.

Hashimoto, H, F. Melton, F, K. Ichii, K, C. Milesi, C, W. Wang, W, and R. R. Nemani, R. 2009.
Evaluating the impacts of climate and elevated carbon dioxide on tropical rainforests of
the western Amazon basin using ecosystem models and satellite data. Global Change
Biology.

Hirsch, A. I., W. S. Little, R. A. Houghton, N. A. Scott, and J. D. White. 2004. The net carbon
flux due to deforestation and forest re-growth in the Brazilian Amazon: analysis using a
process-based model. Global Change Biology 10:908-924. doi: 10.1111/j.1529-
8817.2003.00765.x.


126









Houghton, R. A. 2005. Aboveground Forest Biomass and the Global Carbon Balance. Global
Change Biology 11:945-958. doi: 10.1111/j.1365-2486.2005.00955.x.

Huete, A., K. Didan, T. Miura, E. P. Rodriguez, X. Gao, L. G. Ferreira, and others. 2002.
Overview of the radiometric and biophysical performance of the MODIS vegetation
indices. Remote Sensing of Environment 83:195-213.

Huete, A. R., K. Didan, Y. E. Shimabukuro, P. Ratana, S. R. Saleska, L. R. Hutyra, W. Yang, R.
R. Nemani, and R. Myneni. 2006a. Amazon rainforests green-up with sunlight in dry
season. Geophysical Research Letters 33.

Huntingford, C., R. A. Fisher, L. Mercado, B. B. Booth, S. Sitch, P. P. Harris, P. M. Cox, C. D.
Jones, R. A. Betts, Y. Malhi, G. R. Harris, M. Collins, and P. Moorcroft. 2008. Towards
quantifying uncertainty in predictions of Amazon diebackk'. Philosophical Transactions
of the Royal Society B: Biological Sciences 363:1857-1864.

Hurlbert, S. H. 1984. Pseudoreplication and the design of ecological field experiments.
Ecological monographs 54:187-211.

Hutyra, L. R., J. W. Munger, S. R. Saleska, E. Gottlieb, B. C. Daube, A. L. Dunn, D. F. Amaral,
P. B. De Camargo, and S. C. Wofsy. 2007. Seasonal controls on the exchange of carbon
and water in an Amazonian rain forest. J. Geophys. Res 112.

Jipp, P. H., D. C. Nepstad, D. K. Cassel, and C. Reis de Carvalho. 1998. Deep soil moisture
storage and transpiration in forests and pastures of seasonally-dry Amazonia. Climatic
Change 39:395-412.

Landsberg, J. J., and R. H. Waring. 1997. A generalised model of forest productivity using
simplified concepts of radiation-use efficiency, carbon balance and partitioning. Forest
Ecology and Management 95:209-228.

Larjavaara, M., and H. C. Muller-Landau. 2010. Rethinking the value of high wood density.
Functional Ecology.

Li, W., R. Fu, and R. E. Dickinson. 2006. Rainfall and its seasonality over the Amazon in the
21st century as assessed by the coupled models for the IPCC AR4. Journal of
Geophysical Research 111:D02111.

Lloyd, J., and G. D. Farquhar. 2008. Effects of rising temperatures and [CO2] on the physiology
of tropical forest trees. Philosophical Transactions of the Royal Society B: Biological
Sciences 363:1811-1817.

Malhi, Y., E. Luiz, D. Aragao, C. Huntingford, R. Fisher, P. Zelazowski, S. Sitch, C.
McSweeney, and P. Meir. 2008a. Exploring the likelihood and mechanism of a climate-
change-induced dieback of the Amazon rainforest. Proceedings of the National Academy
of Sciences.


127









Malhi, Y., J. T. Roberts, R. A. Betts, T. J. Killeen, W. Li, and C. A. Nobre. 2008b. Climate
change, deforestation, and the fate of the Amazon. Science 319:169.

Malhi, Y., L. E. O. C. Aragdo, D. Galbraith, C. Huntingford, R. Fisher, P. Zelazowski, S. Sitch,
C. McSweeney, and P. Meir. 2009. Special Feature: Exploring the likelihood and
mechanism of a climate-change-induced dieback of the Amazon rainforest. Proceedings
of the National Academy of Sciences of the United States of America. doi:
10.1073/pnas.0804619106.

van Mantgem, P., and M. Schwartz. 2003. Bark heat resistance of small trees in Californian
mixed conifer forests: testing some model assumptions. Forest Ecology and Management
178:341-352.

Marengo, J. A., C. A. Nobre, J. Tomasella, M. F. Cardoso, and M. D. Oyama. 2008a. Hydro-
climatic and ecological behaviour of the drought of Amazonia in 2005. Philosophical
Transactions of the Royal Society B: Biological Sciences 363:1773-1778.

Marengo, J. A., C. A. Nobre, J. Tomasella, M. D. Oyama, G. Sampaio de Oliveira, R. de
Oliveira, H. Camargo, L. M. Alves, and I. F. Brown. 2008c. The Drought of Amazonia in
2005. Journal of Climate 21:495. doi: 10.1175/2007JCLI1600.1.

Mayle, F. E., and M. J. Power. 2008. Impact of a drier Early-Mid-Holocene climate upon
Amazonian forests. Philosophical Transactions of the Royal Society B: Biological
Sciences 363:1829-1838.

Meir, P., P. Brando, D. Nepstad, Vasconcelos, S, A. Costa, E. A. Davidson, S. Almeida, R. A.
Fisher, E. Sotta, D. Zarin, and G. Cardinot. 2009. The Effects of Drought on Amazonian
rain forests. In: Large-Scale Biosphere-Atmosphere Project in Amazonia: A synthesis

Michaletz, S., and E. Johnson. 2007. How forest fires kill trees: A review of the fundamental
biophysical processes. Scandinavian Journal of Forest Research 22:500-515. doi:
10.1080/02827580701803544.

Murray, K., and M. M. Conner. 2009. Methods to quantify variable importance: implications for
the analysis of noisy ecological data. Ecology 90:348-355.

Myneni, R. B., W. Yang, R. R. Nemani, A. R. Huete, R. E. Dickinson, Y. Knyazikhin, K. Didan,
R. Fu, R. I. Negr6n Juarez, and S. S. Saatchi. 2007a. Large seasonal swings in leaf area of
Amazon rainforests. Proceedings of the National Academy of Sciences 104:4820.

Nakagawa, M., K. Tanaka, T. Nakashizuka, T. Ohkubo, T. Kato, T. Maeda, K. Sato, H. Miguchi,
H. Nagamasu, K. Ogino, and others. 2000a. Impact of severe drought associated with the
1997-1998 El Nino in a tropical forest in Sarawak. Journal of Tropical Ecology 16:355-
367.


128









Nepstad, D., G. Carvalho, A. Cristina Barros, A. Alencar, J. Paulo Capobianco, J. Bishop, P.
Moutinho, P. Lefebvre, U. Lopes Silva, and E. Prins. 2001. Road paving, fire regime
feedbacks, and the future of Amazon forests. Forest Ecology and Management 154:395-
407.

Nepstad, D., P. Lefebvre, U. Lopes da Silva, J. Tomasella, P. Schlesinger, L. Solorzano, P.
Moutinho, D. Ray, and J. Guerreira Benito. 2004a. Amazon drought and its implications
for forest flammability and tree growth: a basin-wide analysis. Global Change Biology
10:704-717.

Nepstad, D. C., C. R. de Carvalho, E. A. Davidson, P. H. Jipp, P. A. Lefebvre, G. H. Negreiros,
E. D. da Silva, T. A. Stone, S. E. Trumbore, and S. Vieira. 1994. The role of deep roots in
the hydrological and carbon cycles of Amazonian forests and pastures.

Nepstad, D. C., I. M. Tohver, D. Ray, P. Moutinho, and G. Cardinot. 2007a. Mortality of large
trees and lianas following experimental drought in an Amazon forest. Ecology 88:2259-
2269.

Nepstad, D. C., P. Moutinho, M. B. Dias-Filho, E. Davidson, G. Cardinot, D. Markewitz, R.
Figueiredo, N. Vianna, J. Chambers, and D. Ray. 2002a. The effects of partial throughfall
exclusion on canopy processes, aboveground production, and biogeochemistry of an
Amazon forest. Journal of Geophysical Research 107:8085.

Nepstad, D. C., C. M. Stickler, B. S. Filho, and F. Merry. 2008. Interactions among Amazon land
use, forests and climate: prospects for a near-term forest tipping point. Philosophical
Transactions of the Royal Society B: Biological Sciences 363:1737-1746.

Nepstad, D. C., A. Verissimo, A. Alencar, C. Nobre, E. Lima, P. Lefebvre, P. Schlesinger, C.
Potter, P. Moutinho, and E. Mendoza. 1999. Large-scale impoverishment of Amazonian
forests by logging and re. Nature 398:505-508.

Nobre, C. A., P. J. Sellers, and J. Shukla. 1991a. Amazonian deforestation and regional climate
change. Journal of Climate 4:957-988.

Oliveira, R. S., T. E. Dawson, S. S. O. Burgess, and D. C. Nepstad. 2005. Hydraulic
redistribution in three Amazonian trees. Oecologia 145:354-363. doi: 10.1007/s00442-
005-0108-2.

Phillips, O. L., Y. Malhi, N. Higuchi, W. F. Laurance, P. V. Nfifez, R. M. Vasquez, S. G.
Laurance, L. V. Ferreira, M. Stern, and S. Brown. 1998b. Changes in the carbon balance
of tropical forests: evidence from long-term plots. Science 282:439.


129









Phillips, O. L., L. E. O. C. Aragao, S. L. Lewis, J. B. Fisher, J. Lloyd, G. Lopez-Gonzalez, Y.
Malhi, A. Monteagudo, J. Peacock, C. A. Quesada, G. van der Heijden, S. Almeida, I.
Amaral, L. Arroyo, G. Aymard, T. R. Baker, O. Banki, L. Blanc, D. Bonal, P. Brando, J.
Chave, A. C. A. de Oliveira, N. D. Cardozo, C. I. Czimczik, T. R. Feldpausch, M. A.
Freitas, E. Gloor, N. Higuchi, E. Jimenez, G. Lloyd, P. Meir, C. Mendoza, A. Morel, D.
A. Neill, D. Nepstad, S. Patino, M. C. Penuela, A. Prieto, F. Ramirez, M. Schwarz, J.
Silva, M. Silveira, A. S. Thomas, H. T. Steege, J. Stropp, R. Vasquez, P. Zelazowski, E.
A. Davila, S. Andelman, A. Andrade, K. Chao, T. Erwin, A. Di Fiore, E. H. C., H.
Keeling, T. J. Killeen, W. F. Laurance, A. P. Cruz, N. C. A. Pitman, P. N. Vargas, H.
Ramirez-Angulo, A. Rudas, R. Salamao, N. Silva, J. Terborgh, and A. Torres-Lezama.
2009. Drought Sensitivity of the Amazon Rainforest. Science 323:1344-1347. doi:
10.1126/science. 1164033.

Pinard, M. A., and J. Huffman. 1997. Fire resistance and bark properties of trees in a seasonally
dry forest in eastern Bolivia. Journal of Tropical Ecology:727-740.

Pinard, M. A., F. E. Putz, and J. C. Licona. 1999. Tree mortality and vine proliferation following
a wildfire in a subhumid tropical forest in eastern Bolivia. Forest Ecology and
Management 116:247-252.

Pinheiro, J., D. Bates, S. DebRoy, and D. Sarkar. (2000). the R Core team (2008), nlme: Linear
and Nonlinear Mixed Effects Models. R package version:3-1.

Potter, C., E. Davidson, D. Nepstad, and C. R. de Carvalho. 2001. Ecosystem modeling and
dynamic effects of deforestation on trace gas fluxes in Amazon tropical forests. For. Ecol.
Manage. 152:97-117.

Poulter, B., U. Heyder, and W. Cramer. 2009. Modeling the Sensitivity of the Seasonal Cycle of
GPP to Dynamic LAI and Soil Depths in Tropical Rainforests. Ecosystems 12:517-533.
doi: 10.1007/s10021-009-9238-4.

Reifsnyder, W. E., L. P. Herrington, and K. W. Spalt. 1967. Thermophysical properties of bark
of shortleaf, longleaf, and red pine. Yale University School of Forestry Bulletin 70:1-40.

Rivera, G., S. Elliott, L. Caldas, G. Nicolossi, V. Coradin, and R. Borchert. 2002. Increasing
day-length induces spring flushing of tropical dry forest trees in the absence of rain.
Trees Structure and Function 16:445-456. doi: 10.1007/s00468-002-0185-3.

Romero, C., and B. M. Bolker. 2008. Effects of stem anatomical and structural traits on
responses to stem damage: an experimental study in the Bolivian Amazon. Canadian
Journal of Forest Research 38:611-618.

Saatchi, S. S., R. A. HOUGHTON, R. Alvala, J. V. Soares, and Y. YU. 2007. Distribution of
aboveground live biomass in the Amazon basin. Glob. Change Biol. 13:816-837.


130









Saldarriaga, J. G., D. C. West, M. L. Tharp, and C. Uhl. 1988. Long-term chronosequence of
forest succession in the upper Rio Negro of Colombia and Venezuela. The Journal of
Ecology:938-958.

Saleska, S. R., K. Didan, A. R. Huete, and H. R. da Rocha. 2007b. Amazon forests green-up
during 2005 drought. Science 318:612-612. doi: DOI 10.1126/science. 1146663.

Saleska, S. R., S. D. Miller, D. M. Matross, M. L. Goulden, S. C. Wofsy, H. R. da Rocha, P. B.
De Camargo, P. Crill, B. C. Daube, and H. C. de Freitas. 2003b. Carbon in Amazon
forests: unexpected seasonal fluxes and disturbance-induced losses. American
Association for the Advancement of Science.

Samanta, A., S. Ganguly, H. Hashimoto, S. Devadiga, E. Vermote, Y. Knyazikhin, R. R.
Nemani, and R. B. Myneni. 2010. Amazon forests did not green-up during the 2005
drought. Geophysical Research Letters 37:L05401.

Schaaf, C. B., F. Gao, A. H. Strahler, W. Lucht, X. Li, T. Tsang, N. C. Strugnell, X. Zhang, Y.
Jin, and J. P. Muller. 2002. First operational BRDF, albedo nadir reflectance products
from MODIS. Remote Sensing of Environment 83:135-148.

Tian, H., J. M. Melillo, D. W. Kicklighter, A. D. McGuire, J. V. K. Helfrich, B. Moore, and C. J.
Vorosmarty. 1998. Effect of interannual climate variability on carbon storage in
Amazonian ecosystems. Nature 396:664-667. doi: 10.1038/25328.

Tomasella, J., and M. G. Hodnett. 1998. Estimating soil water retention characteristics from
limited data in Brazilian Amazonia. Soil science 163:190.

Trumbore, S., E. S. Da Costa, D. C. Nepstad, B. De Camargo, L. A. Martinelli, D. Ray, T.
Restom, and W. Silver. 2006. Dynamics of fine root carbon in Amazonian tropical
ecosystems and the contribution of roots to soil respiration. Global Change Biology
12:217-229.

Tucker, C. J., J. E. Pinzon, M. E. Brown, D. A. Slayback, E. W. Pak, R. Mahoney, E. F.
Vermote, and N. El Saleous. 2005. An extended AVHRR 8-km NDVI dataset compatible
with MODIS and SPOT vegetation NDVI data. International Journal of Remote Sensing
26:4485-4498.

Uhl, C., and J. B. Kauffman. 1990. Deforestation, fire susceptibility, and potential tree responses
to fire in the eastern Amazon. Ecology:437-449.

Van Nieuwstadt, M. G. L., and D. Sheil. 2005a. Drought, fire and tree survival in a Borneo rain
forest, East Kalimantan, Indonesia. JOURNAL OF ECOLOGY 93:191-201. doi:
10.1111/j. 1365-2745.2004.00954.x.


131









Venables, W. N., and B. D. Ripley. 2002. Modern applied statistics with S. Springer verlag.
Werth, D., and R. Avissar. 2002. The local and global effects of Amazon deforestation. J.
Geophys. Res 107.

Williamson, G. B., W. F. Laurance, A. A. Oliveira, P. Delamonica, C. Gascon, T. E. Lovejoy,
and L. Pohl. 2000. Amazonian Tree Mortality during the 1997 El Nino Drought.
Conservation Biology 14:1538.

Wright, S. J., and C. P. van Schaik. 1994. Light and the phenology of tropical trees. The
American Naturalist 143:192-199.

Wright, S. J., C. Carrasco, O. Calder6n, and S. Paton. 1999. The El Nino Southern Oscillation,
variable fruit production, and famine in a tropical forest. Ecology 80:1632-1647.

Xiao, X., Q. Zhang, S. Saleska, L. Hutyra, P. De Camargo, S. Wofsy, S. Frolking, S. Boles, M.
Keller, and B. Moore. 2005. Satellite-based modeling of gross primary production in a
seasonally moist tropical evergreen forest. Remote Sensing of Environment 94:105-122.

Xiao, X. M., S. Hagen, Q. Y. Zhang, M. Keller, and B. Moore. 2006. Detecting leaf phenology
of seasonally moist tropical forests in South America with multi-temporal MODIS
images. Remote Sensing of Environment 103:465-473. doi: 10.1016/j.rse.2006.04.013.

Zeng, N., J. Yoon, J. A. Marengo, A. Subramaniam, C. A. Nobre, A. Mariotti, and J. D. Neelin.
2008. Causes and impacts of the 2005 Amazon drought. Environmental Research Letters
3:014002.


132









BIOGRAPHICAL SKETCH

Paulo Brando was born in the town of Assis, five hundred kilometers from the most

populated state capital in Brazil, Sao Paulo. Early in his career, he decided to work on

topics related to conservation in Amazonia. With this goal in mind, he joined the

University of Sao Paulo to study forestry in 1999. After three years in college learning

more about pulp production than conservation, he was fortunate to join the Forest

Institute for a 2-year internship on restoration of savannas (cerrados). After developing

some skills as an ecologist during this internship, Paulo had the opportunity to study at

McGill University, Canada, where he was exposed to a vibrant scientific community

working on future scenarios for the Amazon. Excited by this approach, he contacted a

Brazilian institution, IPAM (Amazon Environmental Research Institute), to find out about

the possibilities of conducting similar research upon his return to Brazil. Two weeks

following this first contact with IPAM, Paulo arrived in Santarem, Para State, where he

spent the following three and half years studying the effects of droughts on the carbon

and water cycling of tropical forests. During this period he worked closely with local

community members, from whom he learned more about conservation than he had in

school or anywhere else. Paulo left Santarem in 2006 to join the Ph.D. program at the

University of Florida. In the near future, Paulo hopes for a career in which he can

continue to explore the vulnerability of tropical forests to repeated disturbances and

prolonged degradation. He intends to continue to inform the public and policy-makers

about potentially dangerous future trends of Amazonian forests in response to large-

scale deforestation, forest fires, and climate change. Perhaps more importantly, Paulo

Brando aims to develop an interdisciplinary career that will allow him to develop

workable approaches to large-scale conservation.


133





PAGE 1

1 EFFECTS OF EPISODIC DROUGHTS AND FIRE ON THE CARBON CYCLE OF AMAZONIAN VEGETATION: FIELD RESEARCH AND MODELIN G OF A NEAR TERM FOREST DIEBACK By PAULO MONTEIRO BRANDO A DISSERTATION PRESENTED TO THE GRADUATE SCHOOL OF THE U NIVERSITY OF FLORIDA IN PARTIAL FULFILLMENT OF THE REQUIREMENTS FOR THE DEGREE OF DOCTOR OF PHILOSOPHY UNIVERSITY OF FLORIDA 2010

PAGE 2

2 2010 Paulo Monteiro Brando

PAGE 3

3 To my parents, my brother, and my friends

PAGE 4

4 ACKNOWLEDGMENTS I am most thankful t o my advisor for his major influence on my intellectual development during my PhD. I also thank Daniel Nepstad for challenging every finding, word, and result I present here. Most of all, I thank all of those that rendered this thesis a synthesis of years of collaborative field research and modeling efforts Among this very special group of people from several research groups I want to name a few : Eric Davidson, Mike Coe, David Ray, Britaldo Soares Filho, Rafaella Silvestrini, Herman Rodrigu es, Paul Le f e b v re, Paulo Moutinho, Pieter Beck, Alessandro Baccini, Oswaldo Carvalho, Claudia Stickler, Scott Goetz Wendy, Kingerlee, Marcos Costa, Ane Alencar, Adilson Coelho, Osvaldo Portella, Oswaldo Carvalho, Raimundo Quintino, Val, Osmar, Wanderley Roc ha, William Riker Darlisson Nunes, Bati, Sandro, Bibal, Marcia Macedo Es pecial thank s to Ben Bolker, Mary Christman, Leda Kobziar, Wendell Cropper, Ted Schuur Yadvinder Malhi, Scott Saleska, and Jennifer Balch for their various contribution s to my dissertation I thank Allie Shenkin Fay Ray and al l my friends in Gainesville for their support during all these years and helping to make my experience in the United States an enjoyable one T he Instituto de Pesquisa Ambiental da Amazonia (IPAM) and Woods Hole Research Center (WHRC) pro vided essential institution al support f or my PhD.

PAGE 5

5 TABLE OF CONTENTS page ACKNOWLEDGMENTS ................................ ................................ ................................ .. 4 LIST OF TABLES ................................ ................................ ................................ ............ 8 LIST OF FIGURES ................................ ................................ ................................ .......... 9 ABSTRACT ................................ ................................ ................................ ................... 11 CHAPTER 1 INTRODUCTION ................................ ................................ ................................ .... 13 2 DROUGHT EFFECTS ON LITTERFALL, WOOD PRODUCTION, AND BELOWGROUND CARBON CYCLING IN ANAMAZON FOREST: RESULTS FROM A PARTIAL THROUGHFALL REDUCTION EXPERIME NT. ....................... 16 Introduction ................................ ................................ ................................ ............. 16 Methods ................................ ................................ ................................ .................. 18 Site Description and Treatment Design ................................ ............................ 18 Aboveground NPP, Tree Mortality, and Soil CO 2 Efflux ................................ ... 18 Statistical Analysis ................................ ................................ ............................ 19 Results ................................ ................................ ................................ .................... 20 Soil Water ................................ ................................ ................................ ......... 20 Mortality and Necromass ................................ ................................ .................. 20 Stem Growth and Wood Production ................................ ................................ 2 1 Litterfall and Carbon Dynamics ................................ ................................ ........ 21 ANPP ................................ ................................ ................................ ................ 22 Correlates of ANPP and LAI ................................ ................................ ............. 22 Soil CO2 Efflux ................................ ................................ ................................ 22 Discussion ................................ ................................ ................................ .............. 23 Response and Recovery of ANPP ................................ ................................ ... 23 Belowground Carbon Accumulation ................................ ................................ 26 Carbon Stock Accumulation and For est Recovery ................................ ........... 28 3 EVALUATING THE RELATIONSHIPS BETWEEN INTERANNUAL CLIMATIC VARIABILITY AND ABOVEGROUND CARBON STOCKS IN AMAZONIA ............ 38 Introduction ................................ ................................ ................................ ............. 38 Methods ................................ ................................ ................................ .................. 40 Model Description ................................ ................................ ............................. 40 Input Data ................................ ................................ ................................ ......... 41 Experimental Runs ................................ ................................ ........................... 42 Analysis ................................ ................................ ................................ ............ 42 Results ................................ ................................ ................................ .................... 43

PAGE 6

6 CARLUC Predictions for TNF ................................ ................................ ........... 43 Experimental Runs: Tree Mortality and NPP Allocation ................................ ... 43 Specific Regions and Field Plots ................................ ................................ ...... 44 Total Carbon Stocks ................................ ................................ ......................... 44 Climate and Carbon Pools ................................ ................................ ................ 45 Discussion ................................ ................................ ................................ .............. 45 4 SEASONAL AND INTERANNUAL VARIABILITY OF CLIMATE AND VEGETATION INDICES ACROSS THE AMAZON ................................ ................. 57 Introduction ................................ ................................ ................................ ............. 57 Methods ................................ ................................ ................................ .................. 60 MODIS Data ................................ ................................ ................................ ..... 60 EVI Surrounding Meteorological Stations ................................ ......................... 61 Site specific Analysis ................................ ................................ ........................ 61 NBAR EVI and EVI Product ................................ ................................ ............. 62 Climate ................................ ................................ ................................ ............. 63 Statistical Analysis of EVI and Climatic Variables ................................ ............ 63 Basin wide (mixed vegetation types) ................................ ......................... 63 Densely forests areas ................................ ................................ ................ 64 Hierarchical partitioning ................................ ................................ ............. 64 Relationship of the variability of EVI with PAW ................................ .......... 65 Results ................................ ................................ ................................ .................... 65 Spatial temporal Patterns of EVI ................................ ................................ ...... 65 Basin wide Climate ................................ ................................ ........................... 66 EVI and Climate Variables ................................ ................................ ................ 67 Spatial analysis across the Amazon Basin ................................ ................. 67 Spatial temporal analysis in densely forested areas ................................ .. 70 Site specific Analysis ................................ ................................ ........................ 71 Discussion ................................ ................................ ................................ .............. 72 5 POST FIRE TREE MORTALITY IN A TROPICAL FOREST OF BRAZIL: THE ROLES OF BARK TRAITS, TREE SIZE, AND WOOD DENSITY .......................... 83 Introduction ................................ ................................ ................................ ............. 83 Methods ................................ ................................ ................................ .................. 86 General Approach and Site Description ................................ ........................... 86 Modeling Cambium Insulation ................................ ................................ .......... 86 Linking Tree Mortality and Cambium Insulation (R) ................................ .......... 88 Modeling Tree Mortality ................................ ................................ ...................... 89 Tree mortality and cambium insulation ................................ ...................... 89 Other cor r elates of fi r e induced t r ee mortality ................................ ............. 91 Results ................................ ................................ ................................ .................... 91 Cambium Insulation ( R ) ................................ ................................ .................... 91 Cambium Insulation and Tree Mortality ................................ ............................ 93 Other Correlates of Tree Mortality ................................ ................................ .... 94 Discussion ................................ ................................ ................................ .............. 95

PAGE 7

7 Cambium Insulation ................................ ................................ .......................... 95 Bark Correlates with T r ee Mortality ................................ ................................ .. 96 Forest Resistance to Fire ................................ ................................ ................. 98 6 CONCLUSION ................................ ................................ ................................ ...... 106 METEOROLOGICAL STATIONS ................................ ................................ ................ 111 VARIABILITY IN VEGETATION INDEX ................................ ................................ ...... 112 EQUATIONS ON HEAT TRANSFER ................................ ................................ .......... 114 FIRE EXPERIMENT ................................ ................................ ................................ .... 116 FLAME HEIGHT ................................ ................................ ................................ .......... 117 BARK THICKNESS AND TREE DBH ................................ ................................ ......... 118 MODEL 1 ................................ ................................ ................................ .................... 119 MODEL 2 ................................ ................................ ................................ .................... 120 MODE L 3 ................................ ................................ ................................ .................... 121 BIOGRAPHICAL SKETCH ................................ ................................ .......................... 133

PAGE 8

8 LIST OF TABLES Table page 4 1 Slopes of precipitatio n, photosynthetic active radiation (PAR), and vapor pressure deficit over time. ................................ ................................ .................. 67 4 2 ANOVA table for all predictors of EVI in non forested areas from a linear model (M1). Note that we ran o ne model for each year of the study. ................. 69 4 3 ANOVA table for all predictors of EVI in dense forested areas from a general linear model ( M2 ). ................................ ................................ ............................... 70 5 1 Model comparison using corrected Aka ike Information Criteria (AICc) .............. 92 F 1 Model 1. Period: 2004 2005 ................................ ................................ ............. 119 G 1 Model 2. Period: 2004 2007 ................................ ................................ ............. 120 H 1 Model 3. Period: 2007 2008 ................................ ................................ ............. 121

PAGE 9

9 LIST OF FIGURES Figure page 2 1 Soil volumetric water content (VWC) integrated from 0 to 200 cm (A) and from 200 to 1100 cm (B) and daily precipitation (C).. ................................ ......... 30 2 2 Annual rainfall (control plot) an d effective rainfall.. ................................ ............. 31 2 3 A nnual percent o f mortality measured on an individual basis for all individuals ................................ ................. 32 2 4 Annual trends in the exclusion and control plots of: A) litter fall; B) wood production; C) above ground net primary productivity (ANPP); and D) necromass production.. ................................ ................................ ...................... 33 2 5 Stem growth of individuals in the exclusion and control plots by differ ent size classes:. ................................ ................................ ................................ ............. 34 2 6 Annual trends in the exclusion and control plots of leaf area index (LAI).. ......... 35 2 7 R elationships between annual mean VWC measured in the upper soil profile (0 200 cm) and A) above ground net primary productivity (ANPP) and B) leaf area index (LAI). ................................ ................................ .......................... 36 2 8 Measureme 13 14 C in the exclusion (grey) and control plots (black).. ................................ ......................... 37 3 1 Conceptual diagram of the CARLUC ecosystem model.. ................................ ... 50 3 2 Annual rates of leaf mortality (top panel), stem mortality (mid panel), and NPP allocation (lower panel) as a function of PAW.. ................................ .......... 51 3 3 Temporal patterns in carbon stocks predicted by CARLUC in leaves (upper panel) and stems (lower panel) for the Tapajos National Forest. ....................... 52 3 4 Temporal patterns in average aboveground C stocks in regions with low, medium and high levels of PAW.. ................................ ................................ ....... 53 3 5 Temporal patterns in total aboveground carbon stocks averaged by dry and wet season and per year.. ................................ ................................ .................. 54 3 6 Cli mate patterns during the dry and wet seasons.. ................................ ............. 55 3 7 Total aboveground carbon stocks simulated by CARLUC for the entire Amazon and for three regions of varying PAW.. ................................ ................. 56 4 1 A verage dry season EVI across central South America.. ................................ ... 76

PAGE 10

10 4 2 Spatial temporal patterns of dry season EVI across the Ama zon for areas with high canopy cover ................................ ................................ ..................... 77 4 3 Average coefficient of variation of EVI for the entire Amazon, based on data at 500 x 500 m resolution (only for areas with canopy cover ................. 78 4 4 Monthly climate patterns over the Amazon region based on data from 280 meteorological stations distributed across the basin. ................................ ...... 79 4 5 Temporal patterns of environmental and biophysical correlates of EVI. ............ 80 4 6 Site Specific analysis on the relationships between monthly EVI N BAR and monthly field measurements.. ................................ ................................ ............. 81 4 7 Site Specific analysis on the relationships between monthly EVI product and monthly field measurements.. ................................ ................................ ............. 82 5 1 Map showing the spatial distribution of trees located in three 50 ha plots.. ...... 100 5 2 Overall results f rom the bark heating experiment .. ................................ ........... 101 5 3 Relationships between rates of change in cumulative heating ( R) and A) bark thickness, B) bark density, and C) bark moisture content. ................................ 102 5 4 hat ,.. ...... 103 5 5 Relative influence of R hat tree height, dbh, and wood de nsity on tree mortality for different fire treatments.. ................................ ............................... 104 5 6 Probability of tree survival as a function of standardized tree height (left panel), dbh (middle panel), and wood densi ty (right panel).. ............................ 105

PAGE 11

11 Abstract of Dissertation Presented to the Graduate School of the University of Florida in Partial Fulfillment of the Requirements for the Degree of Doctor of Philosophy EFFECTS OF EP ISODIC DROUGHTS AND FIRE ON THE CARBON CYCLE OF AMAZONIAN VEGETATION: FIELD RESEARCH AND MODELING OF A NEAR TERM FOREST DIEBACK. By Paulo Monteiro Brando August 20 10 Chair: Francis Edward Putz Major: Interdisciplinary Ecology Climate change may cause de creased rainfall in Amazonia and move forests to large scale dieback events. To inform debates about this issue, I evaluated the impacts of drought and fire on Amazonian vegetation. Using a throughfall exclusion experiment, I found that wood production was the most sensitive component of above ground net primary productivity to drought and that a boveground carbon stocks declined by 32.5 Mg ha 1 over 6 years of the treatment These results indicate that severe drought s can substantially reduce forest C stock s in the Amazon. Based on a process based model (CARLUC) that incorporates results from that partial throughfall exclusion experiment, I showed that total C stocks tended to increase between 1995 2004 (dry season) over the Amazon, corresponding to increas es in photosynthetic active radiation. These results could partially explain the increases in biomass observed in the field over d ecades in the Amazon To clarify how the gross productivity of Amazonian is influenced by drought, I used improved MODIS enhanced vegetation index measu rements (EVI) to show that in densely forested areas of the Amazon, no climatic variable tested contributed to

PAGE 12

12 explaining the high EVI inter annual variability. Instead, I found evidences that suggest that production of new leaves could play an important role in the inter annual EVI variability, but not necessarily in changes in vegetation productivity. These findings suggest that it still is unclear whether EVI can be employed as a tool to measure the vulnerability of tropical forests to drought. Finally, I studied the mechanisms associated with fire induced tree mortality in a tropical forests in the southern Amazon. I found that the presence of water in the bark reduced the differential in temperature between the fire and vascular cambium due to boiling water in the bark. Also, I show that experimental fires influenced mortality through several processes unrelated to cambium insulation, suggesting that several plant traits that reduce the negative and indirect effects of fire should also be considered in the assessments of tropical forests' vulnerability to fire. In conclusion, I found evidence that droughts and fire s exert strong influences on the terrestrial carbon cycling but no evidence s for imminent regime shift s in the region driven by climate change alone

PAGE 13

13 CHAPTER 1 INTRODUCTION There is an urgent need for in depth assessments of tropical forest vulnerability to global changes, a critical step for understanding and predicting ecological trajectories in the near fu ture. Currently, it is clear that human induced climate change are likely to cause dramatic changes in the world's tropical forests (Christensen e t al. 2007) Many climate models predict that moderate to intense climate change will lead to large scale dieback of forests by the end of the 21 st Century (e.g., Cox et al. 2004) Large scale deforestation and widespread forest fires could exacerbate the negative effects of climate change on ecosystem services of Amazonian forests (e.g., water cycling) (Nepstad et al. 2008) causing the dieback of large portions of these forests much earlier than currently predicted. The question of how much climate change is too much for the maintenance of tropical forest integrity remains unanswered (Malhi et al. 2008) but in my dissertation, I provide some of the information needed. Because global climate change coupled with deforestation is predicted to cause reductions in rainfall over portions of the Amazon (Malhi et al. 2008) I studied the effects of experimentally reduced rainfall on the aboveground net productivity (i.e., how much carbon is fixed by pl ants) of an eastern Amazonian forest in Chapter 2 of my dissertation. Although shortage of soil water availability will ultimately reduce the abilities of trees to fix carbon, based on currently available science, the thresholds at which reductions in soil water availability will cause major changes in forest functioning are not at all clear, nor are the mechanisms associated with disruption in forest functioning (e.g., carbon allocation to different parts of plants, tree mortality, soil respiration).

PAGE 14

14 In C hapter 3 I modify an ecosystem model to include forest responses to drought that were observed in field experiments, but that are lacking in most (if not all) models employed to predict future trajectories of Amazonian forests (e.g., Cramer et al. 2001, Foley et al. 1998) For example droughts are expected to alter allocation of C among leaves, trunks, and roots, which, in turn, can influence forest capacity to cycle carbon or to cope with drought induced stress. But we know very little how this process will influence C cycling during periods of severe droughts in the Amazon. Also, while most mo dels assume that droughts will a ffect C assimilation only via reduced photosynthesis, there is strong evidences that episodic, severe droughts kill trees. After evaluating how these two process es may affect carbon C stocks in the Amazon, I evaluate the potential responses of Amazonian vege tation to climate oscillations d uring 1995 2005, in an attempt to reconcile field based measurements (Phillips et al. 2009) and model predictions. droughts over the last century and the consequent inferences about forest vuln erability to drought (Saleska et al. 2007) I found that there was a need to investigate what 'greener' means in terms of forest functioning. In Chapter 4 of my dissertation, I evaluated how a vegetation index used to measure the greening of the Amazon, derived from satellite based measurements, correlated with variations in vegetation dynamics as measured by leaf a rea index, litterfall, vegetation phenology, and several climatic variables. I then investigated how this vegetation index varied across the Amazon and how these variations were associated with oscillations in climate.

PAGE 15

15 T he likelihood of trees being expose d to fire is increasing in the tropics due to synergistic interactions among droughts, forest degradation and fragmentation, and increased fire ignition (Alencar et al. 2006, Aragao et al. 2007, Nepsta d et al. 2008) yet little is known about the characteristics that determine which trees survive and which succumb to understory fires. For example, w hile bark thickness appears to be the most important plant trait in preventing fire induced damage (Barlow et al. 2003) its importance in preventing fire induced tree mortality has not been widely tested for tropical forest trees and may be modulate d by several other traits In Chapter 5 of my dissertation, I use field experiments and model based analyses to evaluate the influences of bark traits, plant size, and wood density on tree mortality following recurrent understory fires of varying inte nsit y in a transitional forest of Mato Grosso. While the Amazon may experience severe degradation due to synergistic interactions betw een drought and fire, the extent of these degradations is still poorly understood. Thus, the field experiments, data analyses and model simulations run across different scales I employ in this dissertation should help to clarify how the Amazon region responds to drought and fire and how these responses will be expressed in the future under increasing drought conditions.

PAGE 16

16 CHAPT ER 2 DROUGHT EFFECTS ON LITTERFALL, WOOD PRODUCTION, AND BELOWGROUND CARBON CYCLING IN ANAMAZON FOREST: RESULTS FROM A PARTIAL THROUGHFALL REDUCTION EXPERIMENT. Introduction The fate of approximately 86 140 Pg of carbon contained in the forests of the Amaz on (Saatchi et al. 2007) will depend, in part, upon the effect of drought episodes on the amount and allocation of forest biomass production. Sev eral lines of evidences suggest that Amazon drought episodes may be more common and more severe in a warming world (Li et al. 2006) One of the most important ways in which severe droughts influence tropical forest carbon stocks is through tree mortality. Droughts associated with ENSO events elevated tree mortality in East Kalimantan from a background of 2% to 26% yr 1 (Van Nieuwstadt and Sheil 2005) in central Amazonia from 1.1% to 1.9% (Williamson et al. 2000) Panama from 2% to 3% (Condit et al. 1995) and Sarawak from 0.9% to 4.3 6.4% (Nakagawa et al. 2000) (In one Panama forest, ENSO had no effect on tree mortality (Condit et al. 2004) ). Experimentally induced drought in one hectare of an east central Amazonian forest permitted the identification o f a threshold of declining plant available soil moisture and cumulative canopy water stress beyond which canopy tree mortality rose from 1.5% to 9% (Nepstad et al. 2007) Responses of net primary productivity (NPP) and its allocation during drought may have important influences on carbon stocks, but remain poorly understood in moist tropical forests (Houghton 2005) Seasonal and inter annual variation in the NPP of moist tropical forests is controlled mostly by changes in light and soil moisture (Clark and Clark 1994) Hence, NPP increases during ENSO events, when reduced cloud

PAGE 17

17 cover increases photosynthetically active radiation without provoking soi l moisture deficits large enough to inhibit photosynthesis. When soil moisture is in short supply, NPP may decline and its relative allocation among leaves, stems and roots may be affected, but our understanding of these responses is poor. Stem growth is e xpected to be the most sensitive component of NPP to drought because it is low on the carbon allocation hierarchy (Chapin et al. 1990) The response of root production to drought is less clear, in part because of substantial methodological challenge of mensuration (Trumbore et al. 2006) Soil respiration, which integrates CO 2 production from all belowground sources, including root respiration and leaf litter decomposition, tends to be lower in the dry season compared to the w et season (Davidson et al. 2004, 2000, Saleska et al. 2003) ; this is probably due largely to seasonal variation of heterotrophic respiration in the litter layer. Responses to long term drought treatments may also include chang es in root production. A seven year, partial throughfall exclusion experiment was conducted in east central Amazonia to examine forest responses to a 35 41% effective reduction in annual rainfall. I quantified forest LAI litterfall, wood production, soil r espiratio 13 C 14 C composition of CO 2 emitted from the soil in one hectare treatment and control plots to test the following predictions: wood production is the most responsive component of aboveground NPP to drought, declining rapidly with the onse t of drought stress; litterfall increases initially through drought induced leaf shedding, then declines to below non treatment levels as LAI stabilizes at a lower level; and, soil CO 2 efflux is

PAGE 18

18 affected by responses of multiple processes of heterotrophi c and autotrophic respiration, which could have either accumulating or canceling effects. Methods Site Description and Treatment Design The experiment was carried out in the Tapajs National Forest, Par, Brazil (2.897 o S,54.952 o W). Annual precipitation ra nges from 1700 3000 mm and averages ~ 2000 mm, with a 6 mo dry season (Jul Dec) when rainfall rarely exceeds 100 mm mo 1 Mean annual temperature is 28 o C (min 22 o C, max 28 o C). The soil is deeply weathered Haplustox. The depth to water table at a similar s ite 12 km away was >100 m, which means that the stocks of soil water availability to trees is recharged by rainfall. The throughfall exclusion experiment was carried out in two structurally and floristically similar one exclusion plot, approximately 34 40% of annual incoming precipitation was diverted from the soil (70% of throughfall) during the 6 mo wet season from 2000 to 2004. Details of the experimental design can be found in Nepstad et al. (2002) Please see (Hurlbert 1984) for more information on large scale unreplicated experiments. Soil water to 11 m depth was measured monthly in five deep soil shafts per plot between 1999 and 2005 using time domain reflectometry (TDR; (Jipp et al. 1998) ) Here I present volumetric water content (VWC) for two depth intervals: 0 200 cm (upper soil profile); and 200 1100 cm (deep soil prof ile). Aboveground NPP, Tree Mortality and Soil CO 2 Efflux above buttresses) and vines > 5 cm dbh in the exclusion (342 individuals) and control plots (409 individuals). Diameter i ncrement was monitored monthly between 2000 and

PAGE 19

19 2005 using an electronic caliper from which relative stem diameter growth was calculated. Wood production at the plot level was calculated by summing biomass estimates for trees and lianas determined using th e allometric equations from (Chambers et al. 2001) and (Gerwing and Farias 2000) respectively. I also calculated individual annual stem growth averaged for small (10 to 20 cm dbh (trees) and 5 to 20 cm dbh (lianas)) and large (>20 cm dbh) stems. Ingrowth stems were added as they exceeded the lower dbh threshold. Mortality of stems > 10 cm dbh in both plots was assessed on an individual basis monthly between 1999 and 2005 and is described in (Nepstad et al. 2007) Litterfal l was collected bi weekly from 1999 to 2005 from ~ 64 mesh traps (0.5 m 2 ) on a systematic grid in the understory of both plots. Leaf area index (LAI) was estimated indirectly monthly at these same points using two LiCor 2000 Plant Canopy Analyzers (LI COR, 1992) operating in differential mode. Above ground NPP was calculated as the sum of wood production and litterfall biomass in each year of the study (2000 to 2005) according to the methodology of Clark et al. (2001) A manual dynamic chamber system was used to measure soil CO 2 efflux on 18 chambers per treatment per date, as described by Davidson et al. (2004). Carbo n isotope measurements were determined for soil respiration on 5 occasions from 2000 to 2005. Statistical Analysis I used a repeated measures design to test for year to year differences between plots for stem growth (using initial dbh as a covariate), LAI, litterfall, and soil CO 2 efflux. Changes in mortality were assessed using a generalized linear model with binomial errors. The relationships among, VWC, ANPP, and LAI, were tested using linear regressions.

PAGE 20

20 Results Soil Water Prior initiation of treatment, VWC was 25 and 62 mm lower in the exclusion plot than the control plot for the upper (0 200 cm) and deeper (200 1100 cm) soil profiles, respectively (Figs. 2 1, 2 2). After three periods of partial throughfall exclusion (2002), VWC was 97 and 272 mm lower in the exclusion plot than the control plot in the upper and deeper soil profiles, respectively. The difference in VWC between plots did not continue to increase in 2003 for two reasons. First, natural precipitation was well below normal that year (Fig. 2 2), so that soil water was not fully recharged in the control plot. Second, drought induced mortality in 2003 (see Mortality and Necromass section below) may have reduced transpiration in the exclusion plot Mortality and Necromass From 2000 to 2005, the significantly higher in the exclusion plot (5.7% yr 1 ) than in the control plot (2.4% yr 1 ); 2001 was an exception to this general trend (Fig. 2 3A). The most dramatic treatment effects were observed i n 2002 and 2003, when the annual rate of mortality in the exclusion plot reached 7.2% yr 1 and 9.5% yr 1 respectively, as reported in Nepstad et al. 2007. Mortality was also significantly higher in the exclusion (7.2 yr 1 ) than the control plot (3.1% yr 1 ) after the termination of the experimental throughfall exclusion, in 2005. The spike in treatment induced mortality in 2002 and 2003 led to a 25% reduction in above ground live biomass (AGLB) in the exclusion plot relative to the control plot (Fig. 2 3B ), a difference which remained stable after the exclusion treatment was terminated in August 2004 and through the end of 2005. From 2000 through 2005,

PAGE 21

21 47 Mg ha 1 more necromass [dead biomass] was produced in the exclusion compared to the control plot (Fig 2 4D). Stem Growth and Wood Production The treatment caused a reduction in relative stem diameter growth of large individuals (P<0.001) yet had no effect on smaller stems (Fig. 2 5A,B,C), which were growing less in the exclusion plot than the control pl ot before the treatment was applied (P=0.322; Fig. 2 5B). Later, following the pulse in mortality of large trees in 2003, I observed increased stem growth in small individuals in the exclusion plot relative to the control plot (P<0.001). During this same p eriod (2004 to 2005), stem growth was declining in the control plot (Fig. 2 5A,B,C), perhaps as a lagged response to low annual precipitation in 2003 (Figs. 2 1C, 2 2). Total wood production during the first exclusion period (2000) was 13% lower in the exc lusion (4.87 Mg ha 1 yr 1 ) than in the control plot (5.61 Mg ha 1 yr 1 ). This difference increased over the treatment period to 46%, 63%, and 58% in 2001, 2002, and 2003, respectively (Fig. 2 4B). After the drought treatment was removed in 2005, the differ ence in biomass production between plots declined to 23% due to an increase in individual tree growth (mostly small trees) in the exclusion plot and a decrease in growth in the control plot. Litterfall and Carbon Dynamics The drought treatment also reduced LAI over time (Fig. 2 6A; P<0.001). Differences between plots in LAI were small during the two first years of the study, followed by a larger drop in 2002 and remaining 21% to 26% lower than the control plot through 2005.

PAGE 22

22 Litterfall, which was higher in t he exclusion plot than the control prior to the initiation of the treatment, declined in the exclusion plot compared to the control. However, it was lower in the exclusion plot than the control plot only in 2003 (23%, P<0.001; Fig. 2 4A). During the last y ear of the treatment (2004) differences in litterfall between plots were reduced to 10%, and, after the drought stress was removed (2005), litterfall was 2% higher in the exclusion plot than the control plot, indicating a relative increase in litterfall in the exclusion plot compared to the control (P<0.001). ANPP ANPP progressively declined in the exclusion plot relative to the control from 2000 to 2003: 12%, 30%, and 41% lower in 2001, 2002, and 2003, respectively (Fig. 2 4C). In the last year of the trea tment (2004) and in 2005, when the exclusion treatment was removed, ANPP differences between plots were diminished to 30% and 10% of the control, respectively. The average reduction in ANPP from 2000 to 2005 was 21%, corresponding to a cumulative differen ce between plots of 17 Mg ha 1 Correlates of ANPP and LAI Annual variations in VWC in the upper soil profile (0 200 cm) explained 51% of the variation in ANPP over the 6 yr study period (Fig. 2 7A). Annual average LAI showed a strong positive linear relat ionship with VWC (R 2 : 92%; P<0.001; Fig. 2 7B), and with ANPP (R 2 : 49%; P=0.006; ln(ANPP) = 0.9995 + 0.8362 ln(LAI)). Soil CO2 Efflux The effect of throughfall exclusion treatment on soil CO 2 efflux was not consistent among years or seasons. Although th ere was a significant treatment by year interaction (p < 0.001), the treatment by year by season interaction was not significant (p = 0.65). When large pulses were measured, which were probably related with recent wetting

PAGE 23

23 events, they tended to be higher i n the control plot (Fig. 2 8A). The exclusion plot had somewhat higher CO 2 efflux rates on most of the other dates during the exclusion treatment years. These differences largely cancelled, resulting in nearly identical estimates of average annual CO 2 effl ux from the two plots during the 5 year exclusion period (12.8 1.0 Mg C ha 1 yr 1 in the exclusion plot and 12.8 1.3 Mg C ha 1 yr 1 in the control plot). The 13 C signature of the soil CO 2 efflux revealed a divergence between treatments in 2004, when th e 13 CO 2 increased (became less negative) in the exclusion plot (Fig. 2 13 C had been observed for some species in 2002 (J. 13 CO 2 measured w ithin the soil profile in 2003 (not shown), which would reflect allocation of fractionated C substrates to root respiration. There were no differences between treatment plots in 14 CO 2 signatures on any date (Fig. 2 8C), which suggests that the relative age of respired C substrates was not altered by the exclusion treatment. These radiocarbon values are 10 values for CO 2 in ambient air sampled at the same time. This indicates that the C being respired is a mixture of modern C substrates wit h substrates that are 5 40 years old and are more enriched in 14 C due to the signal of atmospheric testing of nuclear weapons in the 1950s and 1960s (Trumbore et al. 2006). Discussion Response and R ecovery of ANPP The most striking result of this study was the differential response of wood production and litterfall to throughfall exclusion. Wood production in the exclusion plot was 13 to 62% below that of the control plot during each of the five treatment years,

PAGE 24

24 while litterfall was substantially lower than the control plot (23%) only during the year of lowest rainfall, in 2003. The allocation of ANPP to wood and litterfall was therefore highly sensitive to soil moisture availability. At the beginning of the experiment (2000) wood and litterfall comprised 40 and 60%, respectively, of the total ANPP, while in the third year of the exclusion (2002), they were 29 and 71% of the ANPP, respectively. This finding provides an important exception to the widely held assumption of drought insensitive allocation of ANPP between litterfall and wood, as represented in many ecosystem models (e.g., Hirsch et al. 2004, Potter et al. 2001, Tian et al. 1998) These results also run contrary to the positive response of LAI to short te rm drought that has been postulated for Amazon forests based upon remotely sensed (MODIS) estimates of LAI (Huete et al. 2006, Myneni et al. 2007, Saleska et al. 2007) If such a positive response is real and widesp read, it may be overridden as drought severity grows. However, it is important to remember that our throughfall exclusion treatment did not simulate the higher radiation that is typically associated with lower rainfall. Litterfall and wood production recov ered rapidly following cessation of the drought. the first post treatment year (2005) and wood production, which was 42% of the control plot in 2003, climbed to 77% that of the control plot in 2005. This rapid recovery is partially explained by the positive growth responses observed for residual, surviving trees following the death of neighboring individuals, perhaps attributable to reduced competition for light and s oil moisture resources. LAI declined in the exclusion plot in 2003 and 2004, increasing light availability in the understory; the difference in VWC between exclusion and control plots also diminished, in part because of reduced

PAGE 25

25 transpiration associated wit h tree mortality A second litterfall recovery mechanism was the establishment of fast growing trees in the exclusion plot understory (P. Brando, D. Nepstad, unpublished data). Increased soil nutrient availability associated with tree mortality may have also contributed to the accelerated growth of surviving trees and the establishment of fast growing trees. The rapid recovery of litterfall in 2004 and after the treatment ended in 2005 was, surprisingly, not accompanied by a recovery of LAI. The relation ship between LAI and litterfall was similarly unexpected throughout the experiment since litterfall remained high, with a small decline in 2003, despite the sharp, sustained decline of LAI beginning in the third year of the treatment (2002). LAI declined i n response to drought with little effect on litterfall either through a lower specific leaf area or through a shortening of leaf longevity or both. Drought induced tree mortality and reductions in wood production resulted in a 63 Mg ha 1 reduction in AGLB that persisted through the first post treatment year. The long term recovery of pre treatment AGLB will depend upon the balance between wood production and tree mortality. The latter declined to the level of the control plot during the post treatment year, and could decline further if the drought treatment accelerated the death of trees that were already senescent and prone to die in the near future. Post treatment, tree level increases in stem diameter growth contributed to an acceleration of wood producti on that, if it continued, could also reduce the time required to recovery pre treatment AGLB. In sum, the experimental drought caused a rapid and drastic reduction in aboveground wood production, while litterfall was less strongly affected. By the third

PAGE 26

26 ye ar of the exclusion treatment in 2003, a reduction in litterfall coincided with increased mortality of large trees in the exclusion plot relative to the control plot, suggesting that available soil water had fallen below a minimum threshold required to sus tain minimal physiological processes (Nepstad et al. 2007) Although the response of plants to soil water depletion may have been initially delayed due to the several coping strategies plants use to avoid or tolerate drought (Oliveira et al. 2005) increased mortality and reductions in forest growth resulted in a 24% reduction in ANP P by the end of the experiment, which was highly correlated with a decline in volumetric water content and LAI. Belowground Carbon Accumulation Heterotrophic soil respiration can be inhibited by drought either directly through microbial drought stress or i ndirectly through substrate limitation (Davidson et al. 2006) which is consistent with observations of lower s oil CO 2 efflux rates during the dry season in Amazonian forests (Davidson et al. 2000, 2004; Saleska et al. 2003) On the other hand, it is also possible that increased allocation of C to fine roots could be an adaptation to explore for deep water resources (Nepstad et al. 1994) which might increase belowground inputs of C as a response to drought. The depletion of soil moisture proceeds from the litter and upper mineral soil layers, where root systems are concentrated, to deeper (>8 m in about 1/3 of the Amazon) s oil layers (Jipp et al. 1998; Nepstad et al. 1994) New root growth therefore occurs at progressively deeper soil layers during drought episodes and may be inhibited near the soil surface by the lack of available moisture. These plausible responses that differ in sign may help explain the lack of a consistent and significant treatment effect on soil CO 2 efflux. Either there were no changes in belowground C cycling processes, or the changes that did occur

PAGE 27

27 cancelled with respect to CO 2 production, thus resulting in nearly identical average annual CO 2 efflux rates The isotope data shed some light on these possibilities. First, the 13 C data demonstrate that C fixed by drought stressed leaves, which is more enriched (less negative) in 13 C than in the control plot, was transported to the soil and/or forest floor by l itterfall and/or root respiration and root inputs. Hence the heavier 13 C signatures of C respired in 2004 could reflect the decomposition of enriched 13 C leaf litter produced the 1 2 years previously, a contribution of heavier 13 C in root respiration refle cting drought stress effects on photosynthetic fractionation, or both. However, the two year lag between fractionation of fresh foliage in 2002 and the first detection of that isotopic signal in surface CO 2 efflux in 2004 suggests that decomposition of abo veground litterfall rather than root respiration (which would reflect the isotopic signature of current 2 efflux. The 14 C data demonstrate that there was no significant shift in the relative age of respired CO 2 as a result of the drought treatment. In other words, the relative contributions of respiration from young substrates (root respiration and microbial respiration of exudates and fresh leaf and root litter) and older substrates (root, leaf, and wood litter and soil organic matter that has persisted 5 years) did not change appreciably. Rhizosphere respiration typically has a 14 C signature that matches the atmospheric 14 CO 2 of the same year, so that an increase in allocation to roots in an amount necessary to offset a reduction in CO 2 derived from older litter decomposition in the exclusion plot would have reduced 14 C values overall. Likewise, had there been significant changes in decomposition rates of soil carbon stocks with higher 14 C values

PAGE 28

28 (leaf and root litter and soil organic matter; Trumbore et al. 2006), we should have detected a change in the radiocarbon signature. Hence, we see little evidence of significant changes in belowground C cycling processes induced by the drought treatment. Responses to nat ural seasonal droughts, which temporarily reduce respiration (Saleska et al. 2003) should not b e confused with responses to longer term droughts, which may or may not alter annual rates of C inputs to and turnover times within the soil. These results highlight the danger of extrapolating short term metabolic responses to longer term ecosystem level responses that potentially involve changes in C allocation, mortality, and growth. Carbon Stock Accumulation and Forest Recovery Partial throughfall exclusion induced reduction in AGLB points to the potential for severe and intense droughts predicted for t he Amazon basin to increase carbon emissions to the atmosphere. Even mild droughts, such as those induced during the first year of exclusion, can inhibit wood production and, hence, the long term pattern carbon storage in this forest ecosystem. This findin g has important implications regarding reports that many Amazon forest appear to be undergoing net carbon uptake (Baker et a l. 2004, Grace et al. 1995, Phillips et al. 1998, Phillips et al. 2009) a finding that has been met with some skepticism (Clark 2002, Saleska et al. 2003) Reductions forests (Nepstad et al. 2004) may reduce wood production, diminishing or reversing the net uptake of carbon from these ecosystems. The lon g term effect of the 63 Mg ha 1 reduction in AGLB on aboveground carbon stocks will depend upon the complex balance between ABLG regrowth and the rate at which trees killed by the drought decompose. Many factors influence ABLG re

PAGE 29

29 accumulation, including tr ee mortality rate, wood production rate, and the wood density of the trees that eventually replace those killed by drought stress. Forests recovering from slash and burn agriculture in the Venezuelan Amazon required approximately two centuries to recover pre disturbance carbon stocks in part because of shifts in species composition (Saldarriaga et al. 1988) The decline of LAI and AGLB observed in this study also suggest that the positive feedback between drought and fire may be stronger than previously hypothesized (Nepstad et al. 2001) which could have large long term implications for carbon storage. Hence, even this relatively long term, 5 year experimental manipulation was not long enough to investigate the full effects of climate change on carbon sto cks of Amazonian forests. Nevertheless, we have demonstrated that multi year drought induced substantial losses of aboveground biomass and productivity.

PAGE 30

30 Figure 2 1. Soil volumetric water content (VWC) integrated from 0 to 200 cm (A) and from 200 to 11 00 cm (B) and daily precipitation (C). Measurements were taken from August 1999 to December 2004.

PAGE 31

31 Figure 2 2. Annual rainfall (control plot) and effective rainfall (annual rainfall minus water excluded by the plastic panels during the wet season). The p artial throughfall exclusion treatment was carried during the six month wet season each year beginning in January 2000 and ending in August 2004.

PAGE 32

32 Figure 2 3. Upper panel: annual percent of mortality measured on an individual basis for 2000 to December 2005. Lower panels: above ground standing live biomass during the same period.

PAGE 33

33 Fi gure 2 4. Annual trends in the exclusion and control plots of: A) litterfall; B) wood production; C) above ground net primary productivity (ANPP); and D) necromass production. Measurements were taken from January 2000 to December 2005. Units are Mg ha 1 yr 1 Error bars ( 1 SE) are presented only for litterfall.

PAGE 34

34 Figure 2 5. Stem growth of individuals in the exclusion and control plots by different size classes: A) > 20 cm dbh; B) between 5 (lianas) 10 (trees) and 20 cm dbh; and C) > 5 cm dbh (lianas) and > 10 cm dbh (trees). Error bars indicate SE (N is given by the number individuals). Measurements were taken from December 1999 to December 2005. Annual stem growth was calculated as the difference in dbh measured between two sampling intervals (Dece mber of each year).

PAGE 35

35 Figure 2 6. Annual trends in the exclusion and control plots of leaf area index (LAI). Measurements were carried out from January 2000 to December 2005. Error bars for LAI indicate SE.

PAGE 36

36 Figure 2 7. Relationships between annual mea n VWC measured in the upper soil profile (0 200 cm) and A) above ground net primary productivity (ANPP) and B) leaf area index (LAI). Data represents annual means from January 2000 to December 2004.

PAGE 37

37 Figure 2 8. Measurements of surface flux of carbon d i 13 14 C in the exclusion (grey) and control plots (black). The error bars represents standard errors of the mean.

PAGE 38

38 CHAPTER 3 EVALUATING THE RELATIONSHIPS BETWEEN INTERANNUAL CLIMATIC VARIABILITY AND ABOVEGROUND CARBON STOCKS IN AMAZONIA In troduction Climate change is projected to have major impacts on tropical forest metabolism and carbon stocks (Bonan 2008) In Amazonia, increases in atmospheric CO 2 concentrations may stimulate vegetation productivity (Lloyd & Farquhar, 2008) but the dr ier and warmer conditions resulting from these increases may ultimately render the region susceptible to late century diebacks, according to predictions of climate dynamic vegetation global models (Cox et al 2000, Cramer et al. 2001, Nobre et al. 1991) If predicted diebacks do occur, a large amount of the 80 120 Gt of C stored in forests of the Amazon will be released to the atmosphere (Betts et al. 2004, Cox et al. 2004) Moreover, this occurrence may lead to a positive feedback lo op, hastening further climate warming and drying (Costa and Foley 2000) The responses of Amazonian forests to drier and warmer climatic conditions remains controversial. One outstanding issue relates to the large uncertainties about the the intensity of drought needed to provoke widespread forest dieback across the Amazon Basin (Malhi et al., 2009) While field based studies have shown that episodic droughts associated with climatic cycles are capable of causing increased m ortality of large trees in the tropics (Condit et al. 1995, Nakagawa et al. 2000, Van Nieuwstadt and Sheil 2005, Williamson et al. 2000) recent satellite based studies suggest that Amazonian forests could be more drought resistant than previously thought (Saleska et al. 2007) Indeed, there is strong evidence that forest photosynthesis (gross primary productivity, GPP) increases at least during early stages of drought because of higher

PAGE 39

39 photosynt hetically active radiation, but decreases when thresholds in plant available soil moisture cause increased tree mortality and reduced tree growth (Poulter et al. 2009) Findings from two throughfall exclusion experiments carried out in moist tropical forest systems of the Brazilian Amazon (Tapajos and Caxiuana National Forests) have helped identify drought thresholds that could cause major changes in vegetation metabolism and structure (Fisher et al. 2007, Meir et al. 2009, Nepstad et al. 2002) In both experiments, net primary productivity (NPP), light saturated photosynthetic rates, and leaf area index ( LAI) all declined in response to drought treatment (i.e., 50% wet season throughfall exclusion) of 2 3 years. Following protracted water stress, substantial increases in tree mortality rates were observed, corresponding to reductions in plant available wat er (Brando et al. 2008, Nepstad et a l. 2007) If the general response s of Amazonian forests to drought are similar to those observed in the Caixuana and Tapajos Forests, a drier future climate is likely to have major effects on the physiology and carbon stocks. In 2005, the southwestern Ama zon experienced one of the most severe droughts of the last century (Aragao et al. 2007, Marengo et al. 2008, Zeng et al. 2008) This exceptionally dry and hot period, caused by an anomalously warm tropi cal North Atlantic, provided an opportunity to observe forest resistance to severe drought at a large scale, similar to those predicted in the future. Based on results from a network of permanent plots spread across the Amazon, Phillips et al. (2009) found that several Amazonian forests were sensitive to the 2005 drought with mo rtality increasing at rates that were high enough to reverse a trend in C accumulation of 0.45 PgC yr 1 to a 1.2 1.6 PgC annual net emissions. Although the results of this study have received some

PAGE 40

40 criticism because of the limited number of plots (Chambers et al. 2009) Gloor et al. (2009) suggest that the RAINFOR plot network capture most of the variation in tree mortality and, hence, on short term, inter annual oscillations in C stocks Here I evaluate the potential responses of Amazonian vegetation to drou ght from 1995 to 2005 using a process based ecosystem model, CARLUC (Carbon Land Use Change; Hirsch e t al. 2004) In particular, I estimate the effects of climate variability on total carbon stocks and fluxes of Amazonian vegetation across the Basin and of forests where permanent plots from RAINFOR are located (Phillips et al. 2009) Based on previous findings of increased rates of carbon accumulation over the tropics (Baker et al. 2004, Phillips et al. 1998) I expect CARLUC simulati ons to describe C accumulation over the Amazon from 1998 to 2004, but C losses during the 2005 drought. Given that CARLUC was parameterized using data from a rainfall exclusion experiment representing a small area in Tapajos National Forest (TNF), it has limitations in reproducing vegetation patterns in other regions Despite this limitation simulations using CARLUC should provide insights into forest vulnerability to oscillations in climate, given that it integrates four representative climatic variables of NPP into a single predictor, potential carbon stocks Methods Model Description I assessed the effects of climate on carbon stocks of Amazonian forests between 1995 and 2005 using the CARLUC ecosystem model (described in detail in Hirsch et al. 2004) which borrows its basic structure from the 3 PG model (Landsberg and Waring 1997) Briefly, CARLUC is a process based, dynamic model that estimates net primary productivity based on characteristics of the vegetation [quant um efficiency ( qe ), and

PAGE 41

41 carbon use efficiency ( fnr )] and on climatic variables [vapor pressure deficit (VPD), temperature (TEMP), photosynthetic active radiation (PAR), and plant available water (PAW)] (Fig. 3 1). As the leaf area index increases, NPP satu rates according to the light ; note that all climatic variables were scaled to range from 0 to 1 ). In the version of the model presented here, the allocation of NPP to leaves, stems, and roots is represent ed as a function of PAW and based on results from the TNF throughfall exclusion experiment (Fig. 3 2). Tree mortality is modeled as being dependent on soil water availability: as PAW declines to ith PAW (Fig 3 2). As for 3 PG, CARLUC was designed to integrate process based phenomena with empirical relationships derived from field experiments. ( 3 1) Input Data The model was initiated using interpolated meteorological climate data (~280 stations spread across the Amazon Basin) at a resolution of 2 km x 2 km (described in detail in Nepstad et al. 2004) Because PAR was not available from these meteorological stations, estimates were obtained using the GOES 8 satellite product (Nepstad et al., 2004) The external variable PAW (i.e., independent from vegetation dynamics) was generated based on the difference between soil field capacity (SW fc ) and wilt ing point (SW pwp ). Briefly, texture data samples collected in the field were combined with empirical relationships between soil texture and SW fc and SW pwp Equations relating soil texture and carbon content to field capacity and permanent wilting point we re developed by Tomasella and Hodnett (1998) using soil water versus

PAGE 42

42 matric potential curves for Amazon s oils. Following, estimates of PAW max (in mm) were generated for each meter of the soil horizon from 0 to 10; PAW max among horizons were summed to estimate PAW max for the entire sampled profile. Experimental Runs I performed four simulations to evaluate th e potential effects of drought on the carbon stocks of Amazonian forests. The mortality and NPP allocation functions were held constant by keeping PAW fixed at 100% (see Fig. 3 2). The mortality and NPP allocation functions were allowed to vary spatially and temporally with PAW. The mortality function was allowed to vary with PAW, while the NPP allocation function was held fixed. The mortality function was held fixed according to PAW, while the NPP allocation function was allowed to vary with PAW. Analysi s Based on the runs of CARLUC in which mortality and NPP allocation were allowed to vary as a function of PAW, I assessed spatial temporal carbon stocks across the Amazon as follows: First, I extracted CARLUC predicted aboveground stocks only for the TNF a nd compared these predictions with field measured data from Brando et al. (2008) Second, I summed all 13 carbon pools from CARLUC for each of the 132 months of the simulations. In this analysis I evaluated intra and inter annual variability in C pools (referred as "Total C stocks"). Third, I chose three regions of to represent three different levels of dryness (i. e., PAW) and averaged the aboveground carbon pools for each region to evaluate temporal patterns in C storage Fourth, I evaluated the average aboveground carbon stocks of Amazonian forests from 1995 to 2005 where field permanent plots are located (Phillips et al. 2009) to

PAGE 43

43 assess how CARLUC predictions relate to field plot measurements (referred as Field Plots"). Results CARLUC Predictions for TNF At the Tapajos National Forest, predicted average aboveground carbon stocks ( 140 MgC ha 1 ) were within ranges observed in the field. The temporal dynamics of simulated C, however, diverged from on site mea surements. While predicted average annual leaf carbon stocks ( C leaf) tended to increase from 2000 to 2005 at a rate of 0.0055 k gC ha 1 year 1 ( 0.001; p=0.0038) (Fig. 3 3), Brando et al. (2008) observed minor reductions in LAI from 2000 to 2004 (with the exception of 2002, when LAI was higher than average Brando et al 2008). In 2005, however, both predicted C leaf and observed LAI were above the 2000 2004 average. Whereas predicted C stocks in stems ( C stem) also tended to increase from 2002 to 2005, these increases were relatively low when compared to total standing stocks (Fig. 3 3), which is in agreement with field measurements of aboveground C stocks up to 2004. In 2005, however, there was an decrease in biomass observed in the field, probably resulting from a lag response of vegetation to reduced plant available water while an increase in predicted C stocks in Cstems Experimental Runs: Tree Mortality and NPP Allocation Carbon stocks predicted by CARLUC for the entire Amazon B asin indicated vegetation sensitivity to drought as expressed by the effects of PAW on tree mortality. Whereas CARLUC runs with PAW having no effect on mortality reached equilibrium at 108 Gt (total stocks for the entire Amazon), this equilibrium was reached at ~ 100 Gt when PAW af fected both mortality and NPP allocation. This result suggest s a potential

PAGE 44

44 interaction between the functions representing mortality and NPP allocation given that increased mortality may have reduced NPP In contrasts, w hen PAW was allowed in the model to affect either mortality or NPP allocation alone CARLUC reached equilibri a at 107 and 105 Gt, respectively. Specific Regions and Field Plots Given that simulated C stocks in CARLUC were sensitive to PAW (when mortality and NPP allocation were allowed to va ry as a function of PAW) drier regions of the Amazon not only tended to reach equilibrium at lower C stocks, but also showed greater temporal variability in C stocks than wetter regions (Fig. 3 4). For instance, monthly analys es of C stocks in three regi ons of the Amazon (low, medium, and high levels of PAW) showed that the driest region was predicted to lose carbon in 2004 2005, while aboveground live C pools were predicted to increase in all three regions from 1995 to 2003. These predicted losses sugges t that thresholds in PAW were crossed in drier regions of the Amazon between 2004 and 2005. CARLUC simulations of carbon stocks from regions where permanent plots are located (e.g., Phillips et al. 2009) (Fig. 3 4) carried out on a monthly basis revealed little change C pools over time (Figure 3 5). From 1995 to late 2003, live aboveground biomass increased, but non significantly, and then began to increase at low rates in late 2003 Total Carbon Stocks Based on temporal analys e s among years and between s easons for the entire Amazon Basin using CARLUC I found that total wet season carbon stocks tended to increase between 1995 1998 but to decrease subsequently, from 1999 to 2005 (Fig. 3 5). In contrast, during the dry seasons, C stock s declined between 1 995 2000, and

PAGE 45

45 then increase d between 2001 2004. During the major drought of 2005, there was a minor shift in this trend of increments at the Basin scale, with an overall net reduction of 0.06 GtC year 1 In sum, these contrasting wet and dry season behavi ors resulted in a net increase in C stocks between 1995 and 2004. Patterns of between season increases and decreases in C stocks were associated with contrasting temporal patterns in climate over the Basin. Whereas increases in dry season PAR contributed positively to vegetation productivity after 2000, decreases in PAW (associated with lower wet season PPT and higher VPD) reduced productivity during the wet seasons, particularly after 1998. Climate and Carbon Pools Although CARLUC indicated that PAW accou nted for most variability in C stocks, PAR had important effects in some cases as well. In regions with no soil water deficits, higher PAR led to higher C stocks. It follows that by dividing the Amazon into the relatively drier East and wetter West (follo wing methodology in Malhi et al. 2009) there was a clear difference in PAR, temperature, and VPD betw een these two regions (higher in the East than the Western Amazon). In other words, where PAR was high and PAW high enough to prevent mortality, productivity tended to be higher (Figs 3 6, 3 7). Discussion Based on the process based coupled vegetation atmo sphere model CARLUC I evaluated the potential effects of climate variability on carbon dynamics of Amazonian vegetation from 1995 to 2005. P redicted C stocks in leaves showed a slight trend of increase from 1995 to 1998 (during the wet season s ) and from 1 999 to 2003 (during dry season s ), which suggest favorable weather condition s for forest growth from 1995 to 2003 (i.e., higher NPP) due to increasing PAR over time. These findings are partially

PAGE 46

46 supported by results from Hashimoto et al. (2009) whose study on vegetation responses to climate variability in the western Amazon (during 1984 2002) showed that increases in NDVI were associated with increases in short wave radiation between August and December. Despite the apparently favorable weather condition for C accumulation, I observed a steady decrease in wet season precipitation (and hence PAW) subsequent to the drought of 199 8. These reductions tended to lower predicted C pools during the wet season s exacerbating differences between dry and wet seasons over time. This conspicuous seasonal pattern may be because portions of the Amazon reached levels of PAW that were low enough to cause tree mortality (m ost variability in CARLUC based predictions of C stocks was associated with tree mortality ). Note, however, that this interpretation is based on the assumption that all forests of the Amazon would respond to drought in ways simil ar to the site in Tapajos National Forest on which the drought induced tree mortality function was developed. Losses of standing stocks of C were also observed following the drought of 2005. In this year, there was a reversal in C accumulation compared to 2004 (shown in the Basin wide analysis), with a small net loss of C While the general trend of increasing C pools during the dry season until 2004, followed by a reduction in 2005, corroborates the findings of Phillips et al. (2009), the magnitude of the reduction observed here was less pronounced than in their study At least four explanations may account for this difference between studies. First of all because CARLUC predicted modest increases in C pools for RAINFOR plots whereas Philli ps et al. (2009) reported actual reductions for the same plots based on field measurements, it is reasonable t o infer that the

PAGE 47

47 Tapajos F orest (where the mortality function in CARLUC was developed) was more resistant to drought than RAINFOR forests on average This potential difference in sensitivity of Amazonian forests to drought is probably a result of long term adaptations to cope with periods of low soil moisture. These results reinforce the need for long term, large scale measurements of tree response to oscillations in climate. Second, PAW driving CARLUC may have been overestimated, reducing predicted drought induced tree mortality. For example, CARLUC has no representation of feedbacks between PAW and vegetation canopy cover or photosynthesis; PAW is an external variable. Therefore, any potential increase in evapotranspiration in 2005 associated with vegetati on dynamics (e.g., increased photosynthesis) could have no effect on modeled PAW (see Nepstad et al., 2004 for more details on PAW calculations). As a result, the model may have overestimated PAW, which could explain the lower predicted C losses here compa red to Phillips et al. (2009). In addition, PAW was modeled based on the assumption that the top 2 m of the soil profile represented depths down to 10 m at which point roots are exceedingly scarce (Nepstad et al. 1994) In contrast the scant available data on soil physics below 2 m (Jipp et al. 1998) indicate that PAW maximum declines with depth (probably only for regions with very de ep water table ) ; hence the estimates used to run CARLUC may have been overestimated. On the other hand, higher drought induced tree mortality in 2005 could have lowered ET, thus increasing modeled PAW compared to actual PAW. The plot network used by Philli ps et al. (2009) to infer C patterns over the Amazon Basin may not represent the climatic variability over the region Although RAINFOR plot s are widely distributed there are very few plots located between Peru and Manaus,

PAGE 48

48 a region that CARLUC predicted t o have had high productivity in 2005. Therefore, this high productivity, associated with high levels of PAR and low mortality, could counter balance C losses from mortality occurring in other parts of the Amazonia Note, however, that while the climate da ta used here are from 280 widely distributed stations, the re are still several large geographical gaps. Finally, CARLUC does not represent some intrinsic processes in vegetation dynamics that are captured in field based studies. CARLUC is an envelope of climatic variables alone used to predict carbon stocks; hence, inter speci fic competition and other important ecosystem processes are missing. These intrinsic dynamics, which are difficult to capture in in models that do not represent competition among indi viduals or species (e.g., 'big leaf' models) could explain the differences between the model and the field based data. While predictions suggest that droughts may become more common and intense in the near future, apparently no model used to predict futur e vegetation trajectories in the Amazon has yet incorporated the effects of drought induced tree mortality on carbon stocks; instead, droughts are assumed to have affect only photosynthesis (e.g., Foley et al. 1996) Similarly, the allocation of NPP to different plant parts is often ignored, which contrast with results from a partial throughfall exclusion experi ment in which aboveground wood production was drastically reduced in response to the throughfall exclusion, while l itterfall and s oil CO 2 efflux showed minor or no response to the throughfall exclusion, respectively (Brando et al. 2008). Without field base d, large scale data on carbon fluxes and stocks across the region measured over time, it is still difficult

PAGE 49

49 to evaluate the future of forests of the Amazon at the basin scale under different climate change scenario s

PAGE 50

50 Figure 3 1. Conceptual diagram of the CARLUC ecosystem model. Carbon pools are controlled by climate (via its effects on NPP) and PAW (via its effects on NPP, mortality, and NPP allocation). Once tissue dies, part of the carbon is transferred to pools with low an d high residence times. The other portion of the carbon resulting from dead tissues is transferred directly to the atmosphere.

PAGE 51

51 Figure 3 2. Annual rates of leaf mortality (top panel), stem mortality (mid panel), and NPP allocation (lower panel) as a fun ction of PAW. These relationships were derived from data from a large scale partial throughfall exclusion experiment (Nepstad et al. 2002, 2007; Brando et al. 2008).

PAGE 52

52 Figure 3 3. Temporal patterns in carbon stocks predicted by CARLUC in leaves (upper pa nel) and stems (lower panel) for the Tapajos National Forest.

PAGE 53

53 Figure 3 4. Temporal patterns in average aboveground C stocks in regions with low, medium and high levels of PAW. The black line represents C behavior estimated by CARLUC in regions where pe rmanent plots are used to infer about aboveground carbon dynamics.

PAGE 54

54 Figure 3 5. Temporal patterns in total aboveground carbon stocks averaged by dry and wet season and per year. The dry season was represented by the months August, September, and October while the wet season by the months between January and May. Note that the CARLUC run with both mortality and NPP allocation varying as a function of PAW were used to generate the pools of C.

PAGE 55

55 Figure 3 6. Climate patterns during the dry (July October) and wet (January April) seasons. Each climatic variable was averaged per season from 1995 to 2005.

PAGE 56

56 Figure 3 7. Total aboveground carbon stocks simulated by CARLUC for the entire Amazon and for three regions of varying PAW. Stars represent RAINFOR sites Note that in this simulations the functions representing drought induced change in NPP allocation and mortality varied according to oscillations in PAW.

PAGE 57

57 CHAPTER 4 SEASONAL AND INTERANNUAL VARIABILITY OF CLIMATE AND VEGETATION INDICES ACROSS THE AMAZ ON Introduction The accumulation of heat trapping gases in the atmosphere may subject large areas of the Amazon Basin and other tropical forest formations to a greater frequency and severity of drought in the coming decades (Christensen et al. 2007, Malhi et al. 2008) This trend may interact synergistically with regional inhibition of rainfall driven by deforestation (Costa and Foley 2000, Nobre et al. 1991, Werth and Avissar 2002) and more frequent sea surface tempe rature anomalies (e.g., El Nino Southern Oscillation and North Atlantic Tropical Oscillation) (Cox et al. 2008, Marengo et al. 2008, Nepstad et al. 2004) to move these tropical forest regions toward large scale fores t dieback events. Drier and warmer climate in the region favors the persistence of grasses and shrubs over trees, in a process that is reinforced by recurring fire (Nepstad et al. 2008) Large scale forest diebacks would transfer to the atmosphere a substantial fraction of the 250 Gt C contained in tropical forest trees (Christensen et al. 2007) It is difficult to assess the drought thresholds beyond which forest dieback might occur in part because of conflicting evidence regardin g the response of forest photosynthesis to drought (Fisher et al. 2006, Betts et al. 2008, Huntingford et al. 2008, Mayle and Power 2008, Lloyd and Farquhar 2008 ) Some studies suggest that forest photosynthesis (gross primary productivity, GPP) increases during the early stages of drought because of higher photosynthetically active radiation (PAR) (Graham et al. 2003, Hutyra et al. 2007, Saleska et al. 2003) In contrast, two partial throughfall exclusion experiments conducted in the Amazon region found that proxies of forest productivity all declined under mild drought conditions, with tree mortality increa sing

PAGE 58

58 under high cumulative canopy water stress resulting from limited plant available water (PAW) [reviewed by (Meir et al. 2009) ]. These impacts were observed only after 2 3 years of simulated drought, with the lag probably due to plant adaptations to reduced soil water availability (Oliveira et al. 2005) The apparently contradictory findings about drought effects on plants are particularly striking in 2005, when the most severe drought of the last 100 years affected the southwestern Amazon (Marengo et al. 2008c) This dry and warm period, linked to an anomalously warm tropical North Atlantic, provided an opportunity to evaluate forest resistance to drought over a g reater spatial extent than the partial throughfall experiments noted above. Two studies reporting on this drought event diverged in their findings. Phillips et al. (Phillips et al. 2009) found evidence that intact forests of the Amazon Basin were drought sensitive, accumulating 1.2 1.6 Pg less carbon during the drought period of 2004 2005 than in previous years, and concluded that Ama zonian forests were negatively affected by the drought of 2005. Conversely, Saleska et al. (Saleska et al. 2007) found that intact Amazon forests experiencing anomalously low precipitation (PPT) in 2005 had higher photosynthetic activity, as indicated by the enhanced vegetation index (EVI), a canopy reflectance metric developed to minimize the attenuati ng influences of background and atmospheric effects and thus remain effective even in areas of high biomass and chlorophyll content (Huete et al. 2002) The authors of the latter study concluded that these forests were more drought resistant than previously thou ght. A possible resolution o f this conflict was provided by a recent paper by Samanta et al. (2010) that suggested that the higher EVI observed in 2005 by

PAGE 59

59 Saleska et al. (2007) could be attributable to atmospherically induced variations associated with aerosol loadings. We conducted a study to better understand the responses of tropical forests to climate extremes, particularly the processes that permit forests to sequester carbon during drought conditions. First, we evaluated the temporal patterns of climatological variables from 280 meteorological stations (1995 2005), PAR from GOES, and an impro ved MODIS EVI data set over the dry seasons of 2000 2008 across the Amazon Basin. Second, we examined statistical relationships between EVI during the dry season (July to September) and three integrative climate variables (VPD, PPT, and modeled PAR). We an alyzed the EVI climate variable relationships for both the entire Amazon Basin (intra annually) as well as just for the densely forested areas (inter annually). For the Basin wide analysis, we predicted that EVI within a given year would demonstrate grea ter sensitiv ity to drier climatic conditions (e.g., high VPD and low PPT) canopy cover), we predicted that inter annual EVI variability would be positively correlated with mode led PAR in sites with high precipitation history (i.e., average monthly dry season precipitation for 1995 2005 > 100 mm). We also predicted that EVI would be negatively correlated with VPD in locations with low PPT history (< 100 mm per dry season month), where dry season VPD is likely to limit productivity, thus reducing leaf production and leaf area. Finally, using data at relatively high spatial (0.25 km 2 ) and temporal resolution, we evaluated the EVI relationships not only on the basis of these same cli mate variables, but also with monthly field measurements of LAI, litterfall, and new leaf production in a dense forest near Santarem, Brazil.

PAGE 60

60 measurements of LAI, litterfall, and new leaf production in a dense forest near Santarem, Brazil. Methods MODIS D a ta I computed the enhanced vegetation index (EVI) at a spatial resolution of 500 m from the MCD43A4 (collection 5) MODIS Nadir Bidirectional Reflectance Distribution Function (BRDF) adjusted reflectance (NBAR). The NBAR data were standardized to a nadir view geometry and solar angle, and processed to limit the influence of cloud cover, thereby limiting the influence of seasonal variations in acquisition conditions (Schaaf et al. 2002) I further screened the NBAR reflectance quality by using the QA flags provided for the NBAR product and we generated a monthly and seasonal or magnitude inversion) to calculate the EVI. Because of the high number of observations influenced by cloud cover and atmospheric contamination during the wet seasons of 2000 2008, I focused our Basin wide analysis on the average of the driest months of the year, July September (referred to as dry season EVI; see appendix A), thus allowing direct comparisons with Saleska et al. (2007). If EVI in one of these three months was missing, the dry season average EVI was based on the average of two other months. If EVI was missing in two months, the dry season EVI was based on a single value. The NBAR EVI requires ~ 5 7 good looks every 16 days (which typically occurs only in the dry season), while for the standard EVI product a single good clear day every 16 day s is sufficient. As a result, NBAR EVI is a less noisy (temporally variable) data set. This procedure thereby assured the largest number of observations of the highest possible data quality.

PAGE 61

61 Because I found that the choice of those months could potentiall y influence the spatial patterns of EVI from North to South of the Basin (see Appendix B) I did not focus our analysis on geographical gradients. Also, I included in all statistical analysis longitude and latitude as covariates (see statistical analysis fo r more details). EVI S urrounding M eteorological S tations For the comparison of EVI with in situ climatic measurements, I averaged the EVI data over an area of 8 km x 8 km (256 pixels) surrounding each meteorological station. Following, I estimated canopy cover of the vegetation surrounding these meteorological stations as of 2005 (Hansen et al. 2003) This analysis included both low (e.g., <70%) a nd high (e.g., >70%) canopy cover. The former includes a mixture of vegetation types. In contrast, densely Site specific A nalysis The credibility of MODIS EVI in capturing vegetation dynami cs was tested in the Tapajos National Forest, Para, Brazil (2.897 o 54.952 o W) (km 67) using data from both a 1 (Brando et al. 2008, Nepstad et al. 2002) ) and an eddy covariance flux tower (Hutyra et al. 2007) First volumetric soil water content (VWC) was measured each month from 2000 to 2004 using a series of paired Tim e Domain Reflectrometry probes situated in five soil pits in the 1 ha plot (Nepstad et al. 2002). The VWC measurements were used to derive estimates of plant available water (PAW) for two interval depths: 0 2 m (PAW 2m); and 2 11 m (PAW 11m). Second leaf area index (LAI) was measured monthly from January 2000 to December 2005 at 100 grid points systematically distributed in the 1 ha control plot. We used two LiCor 2000 Plant Canopy Analyzers in differential mode (LI COR, 1992) (Nepstad et al. 2002). Third from January 2000 to December 2005, litterfall was

PAGE 62

62 collected every 15 days using 100 screened traps (0.5 m 2 each) elevated 1 m from ground level (Brando et al. 2008) and located in the same grid points of LAI. Fourth visual assessments of the presence of new foliage were conducted monthly between August 1999 and August 2004 for all individuals >= 10 cm in diameter at breast height in 1 ha plot (480 individuals). Here I present percent of individuals with new leaves at a given census. Finally PAR was measured from 2002 to 2006 using an LiCor 190 SA at 63.6 m in height (Hutyra et al. 2007) Note that in this site: annual precipitation ranges from 1700 3000 mm with a mean of ~ 2000 mm; 2) during the dry season (July through December), rainfall rarely exceeds 75 mm mo 1 ; soils are deeply weathered oxisol clays of the Haplustox group; and, the permanent water table at a similar site 12 km from the research plot was >100 m deep NBAR EVI and EVI Product For the site specific analysis, I used both the EVI derived from the NBAR product (500 m resolution) and the standard EVI product (collection 5) composited each 16 days, with a spatial resolution of 250 m vs. 250 m. As noted earlier, the NBAR EVI data limit the attenuating effects of the atmosphere and viewing conditions (Schaaf et al. 2002) but required temporal compositing to maximize data quality. I screened the N BAR reflectance quality using the QA flags provided for the NBAR product, generating MODIS standard EVI product was also screened based on the MODIS product quality fl ags (e.g., reliability). No corrections for solar or viewing conditions are made in the standard EVI product. Thus, differences between the NBAR EVI and standard EVI products could be related either to the BRDF model used to solve for surface reflectance, to procedures used to screen for cloud contamination and the normalization

PAGE 63

63 for solar illumination angle, or to all of these factors. Moreover, I compared EVI through repeated measurements of individual pixels, thus corrections for the influences of viewing conditions and solar angle were potentially quite important. Climate Monthly data from ~280 meteorological stations (Nepstad et al. 2004) were used to derive dry and wet season averages for the Amazon (from 1995 to 2005). In the same locations where the meteorological stations were located, PAR was estimated using GOES observed cloud cover and a clear sky solar radiation model. I assessed seasonal averages of climatic variables through time for each meteorological station, using a linear mixture model that accounted for spatial and temporal autocorrelation (i.e., random effects of space and time) and a parameter to attenuate the effects of dry years (i.e., 1998 and 2005) (Bates and Sarkar 2006) Finally, the slopes of these regressions were te sted for differences against the null model of zero change over time (Baayen 2008) which included only random effects. For each one of the 280 m eteorological stations, I calculated PPT history, i.e., the average dry season precipitation based on data from 1996 through 2005. Statistical Analysis of EVI and Climatic Variables Basin wide (mixed vegetation types) A general linear model ( M1 ) of EVI in mixed vegetation (i.e. not just densely forested) areas was comprised of three components: 1) the predictor variables VPD, PAR, PPT, canopy cover (CC), and the interaction of CC with the other covariates; 2) error terms assumed to be spatially autocorrelat ed according to an exponential spatial structure [see (Pinheiro et al. 2000.) ]; and, 3) covariates of longitud e and latitude, to capture the large scale spatial gradient of EVI. I fitted this model to the data for each

PAGE 64

64 year of the study (2000 2005) and, therefore, do not account for inter annual variability in mixed vegetation areas, which is complex because of it s association land use change. It does allow for separate regression coefficients and autocorrelation coefficients for each year. Densely forests a reas The general linear statistical models ( M2 ) of EVI in forested areas were comprised of three components. First, covariates of PAR, VPD, and PPT (each term interacted with PPT history class), were included in order to assess the effects of climate on EVI in three different climatic regions: seasonally very dry regions (PPT history class: 0 mm to 65 mm); season ally dry regions (PPT history class: 66 mm to 100 mm); and non seasonal regions (PPT history class: > 100 mm). Second, longitude and latitude, were included as covariates to capture the large scale spatial gradient of EVI. Third, the error terms were assum ed to have a autoregressive correlation structure of order 1 (Pinheiro et al. 2000 ) to account for temporal autocorrelation among ob servations from the same locations over several years. Note that for forested areas we did not find evidence of spatial autocorrelation in the residuals from the fitted model. Hierarchical p artitioning In both mixed vegetation and densely forested areas, I used hierarchical partitioning (Murray and Conner 2009) as a complementary statistical method to evaluate ea ch covariate's contribution to EVI. In this method, the variance in the response variable (EVI) shared by two predictors can be partitioned into the variance of EVI attributable to each predictor uniquely. For mixed vegetation areas, for each year of the study, I used hierarchical partitioning to evaluate model 1 ( M1 ), but without a spatial structure. For densely forested areas, I evaluated a model that included all

PAGE 65

65 climatic variables, the location of each meteorological station, longitude, latitude, and y ear; referred to throughout as a variation of M2 Because PPT was dependent on location of the meteorological station, I did not include this variable in the hierarchical partitioning analysis. Relationship of the variability of EVI with PAW I first calcul ated for each MODIS pixel the coefficient of variation of the EVI for the period 2000 2008, but retained only pixels wi 6 or more years. The Amazon Basin was then stratified by classes of PAW ( 0 0 m in depth ): class 1: 470 985 mm ; class 2: 986 1280 mm ; class 3: 1281 1500 mm ; class 4 1500 2100 mm, see Nepstad et al. (2004) for the c alculation of PAW. Finally the coefficient of variation of the EVI was compared between the four PAW classes visually. Results Spatial temporal P atterns of EVI EVI varied spatially and temporally across the Amazon Basin during 2000 2008 (Fig. 4 1). Gener ally, EVI varied most where soil moisture availability (PAW) was the most variable. There was a strong gradient in mean annual EVI from the western to the central portion of the Basin, with associated gradients in the variability of EVI and in the estimate d average annual PAW from 0 m to 10 m depth. In areas with high (>70%) tree cover, roughly defining the area of intact forest across the Basin (Fig. 4 2), EVI was below average in 2000, 2001, 2004, and 2007; close to average in 2003 and 2008; and above ave rage in 2005 and 2006. Pronounced shifts in the annual anomalies (deviations from 2000 2008 mean values) were observed over short time periods For example, anomalies in 2002 were 151% higher than in 2001; in 2005 they were 168% higher than in 2004; and, i n 2007 they were 257% lower

PAGE 66

66 than 2006 Analysis of the spatial covariation of EVI and PAW revealed that inter annual EVI variability in dense forests was slightly higher in drier areas, with gradually decreasing EVI variability as average PAW increased (Fi g. 4 3). Basin wide C limate The Amazon Basin experienced a decline in annual rainfall and an increase in PAR from 1995 through 2005. Based on data from ~280 meteorological stations distributed across the Amazon (Fig. 4 4), I found that annual wet season PP T decreased at a rate of 5.31 0.68 mm yr 1 (Fig. 4 4). Dry season PPT also tended to decrease over this time period, but not significantly ( 1.018 0.92 mm yr 1 ). In contrast, PAR increased over the period 1995 2005, primarily after 2002 (especially duri ng the wet season), but the rate was only statistically significant for the dry season (15.8 3.67 moles m 2 month 1 yr 1 ). I could not detect strong linear temporal pattern in VPD over the entire time period; on the other hand, it appears that VPD increa sed substantially after 2002 (Fig.4 4). For example, while the rate of increase of average VPD during 1995 2000 was modest (0.0029 kPa yr 1 ), it increased to 0.43kPa yr 1 during 2002 2005. As a result of decreased PPT and increased VPD, modeled PAW from 0 to 10 m depth decreased significantly over the time period at a rate of 2.03 0.22 % during the wet season and 2.21 0.11 % year 1 during the dry season [see also Nepstad et al. (2004)]. Similar patterns were observed for areas with high fraction of cano py cover (Table 4 1).

PAGE 67

67 Table 4 1. Slopes of precipitation, photosynthetic active radiation (PAR), and vapor pressure deficit over time. All linear models between these climatic variables and time were fit using a linear mixed model with random effects of s pace and time, except for PAW. PAW was averaged over spaced and regressed over time. Meteorological stations surrounded by all fractions of forest cover Meteorological stations surrounded only by high fraction of forest cover Wet Season Dry Season Wet Season Dry Season Precipitation (mm) 5.31 0.68*** 0.35 0.91 7.79 1.62*** 1.351.13 PAR ( moles m 2 mo 1 ) 10.38 7.64 12.22 4.51** 12.448.21 11.234.45 ** VPD (kPa) 0.00 0.00 0.00 0.01 0.000.00 0.000.00 PAW (% of Max) 2.03 0.22*** 2.21 0.11*** EVI and Climate Variables Spat ial analysis across the Amazon B asin In areas with mixtures of vegetation types (e.g., pastures, cerrado, secondary and primary forests), it was predicted that EVI would decrease with (i) increasing VPD an d (ii) decreasing canopy cover and PPT. Based on spatial analysis of EVI data from 256 cells ( each 64 km 2 ) surrounding each meteorological station, these predictions were generally supported. There was a strong and significant interaction b etween canopy co ver and VPD in 4 out of 6 years of the study ( Table 4 2 ). Thus, EVI tended to decrease as VPD increased in areas of low canopy cover. Similarly, PPT interacted with canopy cover in 3 out of 6 years the study, but only marginally (p<0.1); as PPT decreased i n areas of sparse canopy cover, EVI also decreased. In contrast, PAR showed a marginally significant interaction with canopy cover only in one year (p = 0.054). Thus only in 2004 did EVI increase as PAR increased in areas of high canopy cover. In addition to general linear models, I used hierarchical partitioning (Murray and Conner 2009) to estimate the relative importance of each pred ictor on EVI, accounting

PAGE 68

68 for collinearity among predictors in a linear model (see methods). Based on this analysis, canopy cover, VPD and latitude explained, respectively, 55%, 17% and 11% of the total variation associated with the full model (average R 2 of the full model, M1 : ~56%; see methods). Overall these results indicate a strong effect of VPD and a weaker effect of PPT and PAR on EVI from 2000 to 2005. These results also suggest that spatial gradients, independent of variations in climate, had impor tant effects on EVI.

PAGE 69

69 Table 4 2 ANOVA table for all predictors of EVI in non forested areas from a linear model (M1). Note that we ran one model for each year of the study. 2000 2001 2002 2003 2004 2005 F p F pl F p F p F p F p Intercept 3741 <.001 4724 <.001 7069 <.001 1874 <.001 8012 <.001 3165 <.0001 Canopy Cover (CC) 66.481 <.001 83.257 <.001 125 <.001 96.3 <.001 123.0 <.001 138.7 <.0001 VPD 8.035 0.00 6 22.43 <.001 31 <.001 6.73 0.01 1 40.33 <.001 9.974 0.0021 PAR 0.882 0.350 0.028 0.868 3.7 0 .05 5 0.93 0.33 6 0.773 0.3813 0 0.9935 PPT 6.834 0.010 7.06 0 0.009 2.6 0.106 2.21 0.13 6 0.01 0.9211 0.093 0.7604 Longitude 0.949 0.332 2.056 0.158 4.3 0.040 2.08 0.152 1.262 0.2637 6.376 0.013 Latitude 3.77 0.055 1.931 0.16 8 10.8 0.001 7.30 0.008 19.633 <.0001 5.551 0.0203 CC*VPD 0.863 0.355 2.034 0.157 7.8 0.00 6 4.00 0.04 8 11.975 0.0008 6.737 0.0108 CC*PAR 0.718 0.399 0.419 0.519 0.3 0.57 4 4.25 0.04 2 0.046 0.831 0.085 0.7715 CC*PPT 0.196 0.65 9 1.57 0 0.213 2.8 0.09 2 3.06 0.08 3 9.837 0.0022 0.113 0.738

PAGE 70

70 Spatial temporal analysis in densely forested areas support for the hypothesis that inter annual EVI variability was associated with temporal variation in PAR, VPD, or PPT across all classes of PPT history (Table 4 2). Rather, the only significant predictors of EVI were longitude (p=0.02) and PPT history alone (p=0.108) (Table 4 2). Using hierarchical partitioning analysis to assess the importance of each predictor of a lin ear model on EVI ( M2 ; see methods), I found that spatial gradients accounted for most of the EVI variability. For example, of the 77% of the variation explained by the linear model, spatial variability alone accounted for 91% of this variation, whereas yea r, VPD, and PAR accounted for only 4.2%, 0.5%, 0.45%, respectively. Overall, these results show that, by considering the effects of spatial gradients and temporal autocorrelation in our analysis, no climatic variable could meaningfully explain the EVI inte r annual variability. Table 4 3 ANOVA table for all predictors of EVI in dense forested areas from a general linear model ( M2 ). F p Intercept 26359.68 <.0001 PPT History 2.253 0.1081 PAR 1.111 0.2933 VPD 0.276 0.6 000 PPT 0.225 0.6359 Latitude 1. 826 0.1783 Longitude 5.008 0.0265 PPT History*PAR 0.073 0.93 00 PPT History*VPD 0.043 0.9577 PPT History*PPT 0.319 0.7274

PAGE 71

71 Site specific A nalysis At the intensively studied Tapajos site (near Santarem; 80% tree cover), I found that monthly EVI was hig hly seasonal and thus positively and strongly correlated with new leaves (R 2 =53%; p<0.01). EVI was also positively correlated with field measurements of PAR (R 2 =35%; p<<0.01), but inversely correlated with PAW from 0 to 2 m depth (R 2 =46%; p<0.01), which indicates no soil water constraints on photosynthesis (as 0 2m PAW decreased, EVI increased) (Figs. 4 5, 4 6). Hence, EVI was most sensitive to production of new leaves and associated light levels (the individual effects of PAR and leaf phenology on EVI could not be decoupled in our analysis). This interpretation is reinforced by the findings that field measured LAI varied little between seasons and was poorly correlated with EVI (R 2 =17%, p=0.05) (Fig. 4 6). Although other environmental variables may also have affected EVI indirectly (for example, via lagged effects on vegetation processes), their direct relationships with EVI were not evident (Fig. 4 6). In contrast to EVI derived from the MODIS corrected reflectance products (see control flags, was better correlated with correlated with these same measured seasonal environmental and bio physical variables (Fig. 4 7). These results indicate that the EVI product used for the analyses (produced from bi directionally c orrected reflectance data sets) was better correlated with field conditions than the standard MODIS EVI product which could b e a result of correction of viewing angle conditions, screening procedures for cloud cover, or both While we cannot ensure the results at the Tapajos sites extend across the entire Amazon Basin particularly because this site was not affected by

PAGE 72

72 severe d rought during the study period, confidence in our regional results was supported by these more local scale observations. Discussion Increases in solar radiation during dry periods may boost tropical forest productivity (Graham et al. 2003, Saleska et al. 2003, Wright et al. 1999) but prolonged and severe droughts ultimately limit this effect by inducing stomatal closure and even tree mortality (Brando et al. 2008, Nepstad et al. 2007) The thresholds at which drought starts to reduce productivity (as opposed to increasing it) are still not well known. While satellite based vegetation indices allow insight into potential environmental thresholds (Huete et al. 2006, Myneni et al. 2007, Saleska et al. 2007, Xiao et al. 2005) they have provided an ambiguous measure of vegetation responses to drought in the Amazon Basin. Here, I demonstrate that spatial variations of an impr oved EVI metric, corrected for bi directional reflectance variations and other potential attenuating influences (Schaaf et al. 2002) were associated with gradients of PAW and VPD. This indicates that EVI captured complex spatial patterns of photosynthetic responses to environmental variables across the Amazon. In particular, I show contrasting responses of EVI to the interaction betwe en tree canopy cover and drought, over a wide range of environments (e.g., pastures, secondary forests, cerrados). These results reinforce findings from previous studies that demonstrated that where tree canopy cover is high, trees are better buffered agai nst drought because of their deep root systems (Nepstad et al. 1994) Further, these results indica te that this buffering mechanism may not be restricted to the central Amazon (Huete et al. 2006) In Tapajos's dense forest, even subtle changes in EVI captured complex seasonal ecosystem dynamics. In particular, the NBAR EVI increased with the number of canopy

PAGE 73

73 trees with new leaves, while LAI varied little between seasons. This finding show that leaf flu shing (but also PAR) was an important driver of EVI and therefore of GPP (Doughty and Goulden 2008) The association between EVI and PAR was more apparent at Tapajos forest (R 2 =35%) than in our Basin wide analysis of densely forested areas, in which EVI was generally not responsive to any specific climatic variable. This finding raises the possibility that other mechanisms other than PAR could be d riving the EVI inter annual variability. I propose three possible mechanisms for this observation. First, given the strong correlation between EVI and the number of trees with new leaves observed at Tapajos, it is reasonable to expect that leaf flushing c ould have played an important role on the Basin wide, inter annual variability in EVI. Whereas leaf flushing usually coincides with periods of increased radiation (Wright and van Schaik 1994) leaf bud break is not necessarily cued by radiation, but by gradual changes in daylength (Rivera et al. 2002) Once bud break occurs, however, leaf development is strongly controlled by water availability for cell expansion. Therefore, changes in tree water status modified by precipitation events (or even by leaf shedding) during the dry season of dr y years is likely to synchronize bud development and, consequently, leaf flushing. Because inter annual EVI variability in dense forests during the dry season was greatest in regions of lower PAW, I hypothesize that drought could increase EVI by synchroniz ing leaf flushing via its effects on leaf bud development and tree water status. This hypothesis could explain, in part, the high anomalies during dry years (e.g., 2005). I suggest that this is a potentially important area of research.

PAGE 74

74 Second, the lack o f a relationship between PAR and basin wide EVI could be associated with long term adaption for herbivory avoidance. It has been suggested that the timing of leaf flushing of tropical tree s is the result of selective processes to coincide with periods of l owest insect activity, presumably during the peak of the dry season (Aide 1993) Based on this hypothesis, however, I could not explain the EVI i nter annual variability observed in this study. Finally, the results presented here may be partly related to the sample size used in the spatial temporal analysis. Although I used one of the best available data set, there were just 40 meteorological stati ons available for assessing significance in the densely forested parts of the Basin. As a result of this relatively small sample size, I could not further test the effects of PAR on the inter annual variability of EVI nor the hypothesis of increased produc tivity (GPP) during dry periods of increased PAR. I thus hypothesize that the apparently conflicting observations between Phillips et al. (2009) and Saleska et al. (2007) regarding the drought event of 2005 were related to several mechanisms operating sim ultaneously. GPP (expressed as EVI) appears to have increased due to the production of new leaves and increased PAR (Saleska et al. 2007), whereas aboveground net primary productivity (ANPP) measured in the field concurrently decreased (Phillips et al. 200 9) because of higher tree mortality and increased respiration associated with lower PAW and high temperatures, respectively. Moreover, the allocation of non structural carbohydrates to belowground processes may have increased, given that EVI (and possibly GPP) was higher and ANPP lower during the drought of 2005.

PAGE 75

75 While it was observed important oscillations in weather over the Amazon from 1995 to 2005 (e.g., a 5.3 mm yr 1 reduction in PPT), these oscillations were not clearly related to EVI inter annual va riability in densely forested areas at the Basin scale. Thus, there is a need for additional analyses that couple field measurements with satellite observations in order to clarify how the Amazon region responds to drought, how those responses will be expr essed in the future under increasing drought conditions, and to what extent those responses are captured in satellite observations of canopy photosynthesis.

PAGE 76

76 Figure 4 1. Upper panel: Average dry season EVI across central South America for the period 2000 2008. Overlaid circles represent average vapor pressure deficit measured at 280 meteorological stations across the region for the period 1995 2005. Center panel: Coefficient of variation in annual EVI for the period 2000 2008.Lower panel: Average annual p lant available water at 10m depth, expressed as a percentages of the maximum, for the period 1995 2005.

PAGE 77

77 Figure 4 2. Spatial temporal patterns of dry season EVI across the Amazon for areas with high canopy cover (EVI > 0.4 and MODIS tree canopy cover pro duct > 70%). The first panel shows average EVI from 2000 200 and the following panels show the EVI anomaly for each year, calculated as EVI i EVI mean

PAGE 78

78 Figure 4 3. Average coefficient of variation of EVI for the entire Amazon, based on data at 500 x 500

PAGE 79

79 Figure 4 4. Monthly climate patterns over the Amazon region based on data from 280 meteorological stations distributed across the basin. (A) Monthly averages of Precipitation, (B) Monthly mod eled photosynthetic active radiation (modeled PAR), (C) Modeled monthly plant available water at 10m depth (PAW), (D) Monthly vapor pressure deficit (VPD). Blue lines represent a smooth curve based on a loess method, and red lines represent a local regres sion model (spline).

PAGE 80

80 Figure 4 5. Temporal patterns of environmental and biophysical correlates of EVI. Upper panel : EVI derived from MODIS NBAR, EVI derived from the Collection 5 MODIS product screened to include only good or best quality control flag s, and field measured leaf area index (LAI) (see methods). Center panel : monthly photosynthetically active radiation (PAR), and bi monthly litterfall and new leaf production (see methods). Lower panel: plant available water (% of maximum) at two depths (0 2m and 2 11m) and daily precipitation (mm).

PAGE 81

81 Figure 4 6. Site Specific analysis on the relationships between monthly EVI NBAR and monthly field measurements of individuals with new leaves (%), PAR, litterfall, shallow (0 2m) and deep (2 11m) PAW, PP T, and LAI. The solid and dashed lines represent straight line s fitted using a standard linear regression model s for all months and dry season; respectively. The dotted line represents a smooth line.

PAGE 82

82 Figure 4 7. Site Specific analysis on the relations hips between monthly EVI product and monthly field measurements of individuals with new leaves (%), PAR, litterfall, shallow (0 2m) and deep (2 11m) PAW, PPT, and LAI. The solid and dashed lines represents straight lines fitted using a standard linear regr ession models for all months and dry season; respectively. The dotted line represents a smooth line.

PAGE 83

83 CHAPTER 5 POST FIRE TREE MORTALITY IN A TROPICAL FOREST OF BRAZIL: THE ROLES OF BARK TRAITS, TREE SIZE, AND WOOD DENSITY Introduction The synergistic ef fects of droughts, forest degradation by logging, landscape fragmentation, and increased fire ignition have already rendered many Amazonian forests fire prone (Alencar et al. 2006, 2004, Aragao et al. 2007, Aragao et al. 2008, Cochrane and Laurance 2008, Cochrane 2003, Nepstad et al. 2001, 2004, Nepstad et al. 2008) I f current predictions of drier and warmer conditions (Cox et al. 2008, Cox et al. 2000) hold true even large forested areas in Amazonia may become vulnerable to fire in the near future (Golding and Betts 2008) For example, 16 of 23 climate models predicted substantial reductions in precipitat ion by the mid 21 st Century (Malhi et al. 2008) While this change is anticipated to have substantia l effects on forest carbon storage and bio diversity (Barlow et al. 2002, Barlow and Peres 2008, 2006, Nepstad et al. 1999) there are few data avail able to evaluate the vulnerability of forest trees to fire. One of the most obvious ways that understory forest fires affect tropical forests is by killing trees (Barlow et al. 2003, Cochrane and Laurance 2002, Pinard and Huffman 1997, Pinard et al. 1999, Uhl and Kauffman 1990) The cha racteristic low intensity surface fires of closed canopy tropical wet and moist forest spread slowly through the understory and heat the vascular cambia of trees. When cambial cells die around the entire e (Michaletz and Johnson 2007) Based on this basic observation a wide range of models have been propose d to measure the vulnerability of tropical trees and hence tropical forests to fire. The simplest (and so far most used) models assume that bark thickness is the most important trait in avoiding fire induced cambium damage and, consequently, tree

PAGE 84

84 mortal ity. Uhl and Kauffman (1990) for instance, stud ied a moist Eastern Amazonian forest and found t hat a simulated fire increased cambium temperatures to lethal levels of most trees with thin bark. Based on this finding, the authors speculated that the occurrence of surface fires would kill most standing trees. Pinard and Huffman (1997) used a similar approach to suggest that only individuals with bark > 18 mm thick would resist a fire in a dry f orest of Bolivia. Although back thickness has been shown to correlate well with post fire tree mortality (Barlow et al. 2003), other traits that reduce the indirect (and negative) influences of fire on the physiology and structure of tree trunks could bet ter represent the complex processes involved in post fire tree mortality (Michaletz and Johnson 2007) Fo r example, t rees wounded during surface fires may experience wood decay ranging from severe to moderate depending on their capacity to compartmentalize wood decay a process that is intrinsically associated with wood density (Romero and Bolker 2008) Likewise, simply by being large in diameter, trees may reduce the indirect effects of fire on tree mortality by being resistant to wind throw (Larjavaara and Muller Landau 2010) less likely to be killed by branch and tree fall from other trees, and less likely to have the ir circumference heat ed during fires (Gutsell and Johnson 1996) Thus large individuals may be more likely to survive surface f ires than smaller trees with equally thick bark Similarly, tall trees may disproportionally survive fires because their crowns are less likely to be damaged by fire. Given the predictions of increasing fire frequenc ies in the tropics, it is important to e lucidate the factors that render trees susceptible to fire induced mortality. Here, I conducted two sets of related experiments to improve our understanding of the roles of

PAGE 85

85 plant traits in avoiding tree mortality. I first evaluate which bark traits best pr edict rates of temperature change (R) of the vascular cambium during experimental bole heating. I use the exact solution of Newton's Law of cooling to calculate R (Appendix C for the reasoning behind the choice of E quation 3 1 ), a measure of how much condu ctive and convective heat is transferred through bark over time ( the inverse of cambium insulation). My predictions are that : 1) bark thickness is inversely related to R and is the most important bark trait in preventing the cambium temperature from rising to lethal levels ; 2) that R increases as a function of bark water content high conductivity; and, 3) that bark density is negatively associated with R given that the greater the density, the lower diffusivity (Dickinson 2002; see Appendix C also) While in the first set of analyses mentioned above I evaluate the important bark trait s for cambium insulation, in the second set of analyses I test which plant traits best predict likelihoods of fire induced tree mortality including R, wood density, tree height, and diameter at breast height (dbh). I use data from a large scale fire exper iment located in the southern Amazon to test the following predictions: 1) increasing tree dbh and height decrease the likelihood of fire induced tree mortality, not only because larger individuals tend to have thicker bark, but also because they have trai ts that reduce the indirect effects of fire; 2) decreasing R (higher cambium insulation) increases tree survivorship; and, 3) fire treatments lead to higher tree mortality close to the forest edge compared to the forest interior, given that fire intensity is expected to be higher at the edge relative to the interior (Cochrane and Laurance 2002)

PAGE 86

86 Methods Gener al Approach and Site Description The study was conducted in a transitional forest in the southern part of the Amazon Basin ( 13 0 04'S, 52 0 23') (Appendix D ), where average annual precipitation is 1,800 mm and the dry season extends from May to September (Balch et al. 2008) To evaluate the importance of different bark traits in preventing fire induced tree mortality, I conducted a series of field experiments. First, I h eated the outer bark of trees with a propane torch to relate heat transfer rates (R; inversely related to cambium insulation with units of o C second 1 ) to bark traits. Second, I measured the bark thicknes s of 1,000 individuals representing 24 of the most common species in the study region to develop a predictive model of bark thickness as a function of tree diameter at breast height (dbh). Third, based on predicted bark thickness, I estimated R (referred a s R hat ) for each tree growing in the area of a large scale fire experiment. Fourth, I tested several statistical models of tree mortality as a function of R hat under different experimental fire regimes. Finally, to test the importance of different plant traits in predicting fire induced tree mortality, I compared the slope of models that included only R hat with slopes from models that included tree height, dbh, or wood density (derived from the literature; Chave et al., 2006) Modeling Cambium Insulation During the dry season of 2008, cambium vulnerability to fire induced damage was es timated as the rate of change (R; units of o C second 1 ) in cambium temperature (CT) as a function of the air temperature adjacent to the outer bark, which was heated using a torch. I first removed a 5 cm x 5 cm sample of bark at 25 cm above the ground from each of 105 randomly sampled individuals representing 18 species and ranging from

PAGE 87

87 10 121 cm in dbh These samples were taken to the lab for measurements of thickness following the procedures in Uhl and Kauf fman (1990) Immediately following bark removal, a thermocouple was inserted between the inner bark and xylem (i.e., in the vascular cambium) and a second thermocouple was placed on the surface of the outer bark. The air adjacent to the outer bark of each tree was then heated using a propane torch coupled with a metal plate placed in front of the flame to prevent direct thermocouple contact. Temperatures in the vascular cambium and in the air adjacent to the outer bark were recorded every 10 seconds for 13 minutes, which allowed detection of changes in cambium temperatures in even thick barked species. Using an asymptotic regression model (eq. 1; see also Appendix E ), I estimated the rate of change (R in eq. 1) in CT as a function of cumulative air temperatu re at the outer bark, which integrates the colinear variables of time and air temperature adjacent to the outer bark into a single variable. To identify the most important predictor of R, I using the Akaike Information Criteri on (AIC Burnham an d Anderson, 2002) and hierarchical partitioning analysis The latter method partitions the variance relative to R shared by two predictors into the variance of R attributable to each predictor uniquely (Murray and Conner 2009) Once the best model relating R and bark traits was selected, it was used to predict R for each individual in the fire exper iment, referred to as R hat Note that the higher R hat the lower cambium insulation and the thinner the bark thickness ( 5 1)

PAGE 88

88 Linking Tree Mortality and Cambium Insulation (R) Data from a large scale fire experiment was used to assess probabilities of fire induced tree mortality as a fu nction of R hat In this experiment ( Fig. 5 1) three plots of 50 ha each were established in 2004 in Mato Grosso, Brazil (details in Balch et al. submitted ) : a control with no signs of recent fires; a plot that was experimentally burned twice (2004 and 2007; plot B2); and, a plot that was experimentally burned four times from 2004 to 2007 (pl ot B4). Within each of these plots, I conducted yearly mortality censuses for individuals 10 19 cm dbh (N=6 transects of 10 m x 500 m per plot) and 20 assessed annually fo r mortality in all 50 ha plots. In total, 6,570 individuals of 97 tree species were sampled for dbh (see Fig. 5 1). Height was measured for a subset of these, with 4,071 individuals sampled. Wood density was derived from the literature for 87 species (Chave et al. 2006) Note that the northern edge of the experimental area was adjacent to an open field (Appendix D ). Because some sample d trees were not charred during the experimental fires due to burn heterogeneity across years (details in Balch et al. 2008), they were grouped into the following categories, as summarized in Fig. 5 1: N ever charred and located in the control plot, in plot B2 (Charred 0x B2), or in plot B4 (Charred 0x B4). Note that mortality in the control was expected to be lower than in the burned plots even for trees not directly affected by the experimental fire (i.e., a strong treatment effect) due to the high number of falling trees in plots B2 and B4 and the associated changes in forest microclimate and canopy damage C harred once (either in 2004 or 2007) and located in plot B2 or charred once at any time between 2004 and 2007 and located in plot B4 (Charred 1x B4 ). Note that the fires of 2004 (Charred 04) and 2007 (Charred 07) apparently

PAGE 89

89 differed in intensity (see Appendix E ), justifying their separation into two distinct groups. In plot B4, however, this separation into groups was not possible because of the large n umber of fire treatments and the small sample size within each group for each species. C harred twice and located either in plot B2 or in plot B4 (Charred 2x). C harred three times and located in plot B4 (Charred 3x). C harred four times and located in plot B 4 (Charred 4x). Given that BT could not be measured on all sampled individuals for logistical reasons and because the measurements could increase mortality rates, I developed predicti ve models of bark thickness (BT ) for the 24 most common species in the r egion based on dbh of 1,000 trees sampled in an adjacent area to the fire experiment (see Appendix F ). From BT, I then estimated a value for R for each individual of the 24 focal species in the fire experiment area (N=3,752). Modeling Tree Mortality Tree mortality and c ambium i nsulation To model the probability of tree mortality as a function of R hat ( and therefore assess the importance of R hat in avoiding fire induced mortality) I used logistic regression mixed models with fixed and random effects of R hat and species, respectively. Because plot B2 was burned in 2004 and 2007, while plot B4 burned annually from 2004 to 2007, mortality assessments were carried out at different times among plots and over time. To deal with this complexity in the data set, I developed four models based on a subset of the trees located in the fire experiment (Fig. 5 1): Model 1 included trees that were potentially a f fected by one l o w intensity fire and were assessed for mortality after a period of one year foll o wing the e xpe rimental fire of 2004 (period: 2004 2005; plots B1 and B4 ). Probability

PAGE 90

90 of mortality w as modeled as a function of R hat interacting with f i v e treatments [ Cont r ol Char r ed 0x (plot B2 or B4 ), Char r ed 04 (plot B2 or B4 )]. Model 2 included trees that were po tentially a f fected by one l o w intensity fire and were assessed for mortality after a period of three years foll o wing the e xperimental fire of 2004 (period: 2004 2007; plot B2 ). Probability of mortality w as modeled as a function of R hat interacting with three treatments ( Cont r ol, Char r ed 0x, Char r ed 04 ) Model 3 included trees that were potentially a f fected by at least one moderate to high intensity fire and were assessed for mortality after a period of one year foll o wing the e xperimental fire of 2007 ( period: 2007 2008; plot B2 ). Note that this group of trees included only those that were alive prior to the fires of 2007 and that were not charred by the fires of 2004. Probability of mortality w as modeled as a function of R hat interacting with four trea tments ( Cont r ol, Char r ed 0x, Char r ed 07, Char r ed 2 x ). Model 4 included trees that were potentially a f fected by s e v eral fires and were assessed for mortality after a period ranging from 1 to 4 years foll o wing a g i v en e xperimental fire (period: 2004 2008). P robability of mortality w as modeled as a function of R hat i nteracting with twel v e treatments T o avoid bias due to spatial autocorrelation, I compared parameter estimates from the models with and without a spatial error structure ( glmer function in R Bates and Maechler 2009; and glmmPQL function in R, Venables and Ripley 2002) Because the inclusion of the spatial structure did not change the results, I present here only the estimates der i v ed from the glmer function (which is less li k ely to be biased Bolker et al. 2009) T o account for the possible impact of proximity to an adj acent open field, the distance from the edge w as included as a co v ariate in all models for each individual. G i v en that R hat was estimated from tree dbh, its association with mortality could simply be due to a demographic e f fect in which smaller ind i vidua ls (with high R ) die at higher rates than la r ger ind i viduals, e v en in the absence of fire. T o account for this potential demographic e f fect in models, the influence of R on mortality w as ev aluated based on

PAGE 91

91 the slope of the regression for mortality in the f ire treatments contrasted with the slope of the control. Therefore, if slopes for mortality in the control and fire treatments were equally steep, R was considered to have no effect on fire induced tree mortality. Other cor r elates of fi r e induced t r ee mor tality To compare the influences of R hat plant size (height and dbh), and wood density on tree mortality, slopes of Model 4 were compared to the slopes of models that included each plant trait as a single covariate. More specifically, the relative import ance of a given variable was based on the magnitude of the difference between slopes for mortality of the control and burned treatments. A ll covariates of mortality were standardized by two times the standard deviation of the entire population (Gelman and Hill, 2007), making it possible to compar e the magnitude of slopes across models and treatments (note that slopes associated with R hat are presented in absolute terms to facilitate comparison with slopes of other models). The potential effects of spatial a utocorrelation were evaluated for all models but did not change the results (see section above on Modeling Mortality). Note also that covariates of mortality were not included in a single model because they were multi collinear and had different sample siz es; when all covariates were included in the same model, the degrees of freedom were reduced substantially. Results Cambium Insulation ( R ) Cambium temperatures (CT ) were perhaps naively expected to increase steadily until they reached an asymptote at the t orch temperature (i.e., temperature adjacent to the outer bark), but temperatures never exceeded 100 C because all the fire energy was absorbed in the process of boiling water (i.e., latent heat of vaporization Fig. 5 2).

PAGE 92

92 Consequently, the presence of moisture in the bark not only prevented CT from rising above 100 C, but also reduced the gradient in te mperature (or introduced a step) between the air adjacent to t he outer bark and the cambium. For instance, the air adjacent to the outer bark heated to 400 o C for 10 minutes was expected to have created a differential in temperature with the vascular cambiu m of 4 0 0 C mi n us the initial cambium temperature (CT i ). The presence of water in the bark, however, constrained this differential to 100 C minus CT i Despite the constraint imposed by water, the amount of water in bark showed only a weak positive relationship wit h R (Fig. 5 2C). B ark thickness alone was the best predictor of R; as bark got thicker, the rates of change in temperature through the bark decreased exponentially (Fig. 5 3). I nclusion of b and B MC in predictive models of R was not justified (Table 5 1), as the model with only BT was the most parsimonious. T able 5 1. Model comparison using corrected Akaike Information Criteria (AICc) In this table, we also report the number of parameters (K), the adjusted R 2 and BMC represent bark thickness, wood specific gravity, and moisture content, respectively Models K Adj. R 2 AICc Weight B T + b + B M C 5 0.83 0.0 0. 5 B T 3 0.83 1.3 0. 3 B T + b 4 0.83 2.5 0.1 B T + B M C 4 0.83 2.8 0.1 BM C 3 0.14 146.5 0.0 b 3 0.03 156 .7 0.0 I n t e r c e p t 2 0.00 158.5 0.0

PAGE 93

93 Cambium I nsulation and T ree M ortality The relationships between R hat and tree mortality varied among years and treatments (Fig. 5 4). In the year following the low intensity fire of 2004 ( Model 1 ), mortality was weakl y associated with R hat (and hence BT) and was similar between the control and the burned plots ( Fig. 5 4 and A ppendix F ; note that both standardized R hat and the associated BT are presented in this figure). Three years following the 2004 fires ( Model 2 ), survival in the control was higher than in the burned plots, but R hat was still a poor predictor of mortality (Fig. 5 4 and appendix G ). These results indicate that the fires of 2004 had minor effects on tree mortality compared to the control and that R wa s a good predictor of fire induced tree mortality. While fire induced tree mortality was poorly associated with R hat following a single low to moderate intensity fire, mortality was highly associated with R hat during the year following the fire of 2007 ( M odel 3 ; which was followed by the fire of 2004): as R hat increased the probability of survival decreased (Fig. 5 4 and appendix H ). This pattern was strong and significant for trees charred twice but also for non charred trees, perhaps because smaller tre es (with low cambium insulation) were more likely to be damaged by fire induced branch and tree fall than large trees (with high cambium insulation) in the burned plots compared to the control. Hence there was a strong treatment effect in which tre e morta lity in the control plot was substantially lower than mortality in the burned plots. However, there was a lack of relationship between R hat and mortality for trees charred by the fires of 2007, suggesting that one year after a fire is perhaps too short an interval to evaluate fire induced tree mortality As expected, mortality in the control plot showed no relationship with cambium insulation.

PAGE 94

94 When the effects on mortality of all of the 12 fires were included in a single model ( Model 4 ), there was a positiv e relationship between tree mortality and R for trees charred once, twice, and three times ( Fig 5 4). In other words, charred trees showed strong, inverse relationships between mortality and cambium insulation for the period 2004 2008, but some new patter ns emerged from this analysis. First, non charred trees and trees charred only in 2004 suffered high mortality across the entire range of R indicating a strong treatment effect of recurrent fires, but a minor effect of R for this group of trees which cou ld be a result of an increased number of falling trees in the burned plots killing non charred trees Second, trees that were charred 4 times showed a high probability of surviv ing (e.g., similar to the control) an d showed no relationship with R hat This c ould be attributed to the fact that only the highly fire resistant trees survived four fires ( see Balch et al. submitted ). The inclusion of distance from the edge was highly significant in all models This indicates that fire severity was higher at the for est edge adjacent to an open field than in the forest interior Other C orrelates of T ree M ortality Overall, trees with higher wood density and larger dbh showed lower probability o f fire induced mortality (Fig. 5 6), given that models with either wood dens ity or dbh as a covariate had steeper slopes for the fire treatments compared to the control (Fig. 5 5). Further, the highest deviations from the control were observed for the fire treatments Charred 1 2 and 3 x T ree height was strongly and negatively as sociated with mortality (Figs. 5 6), but mainly for non charred trees and for trees in the control plot. In other words, likelihood of survival increased as a function of increasing tree height at similar rates between trees in the control and fire treatme nts. The exception to this pattern was

PAGE 95

95 for trees burned during the fires of 2004 in plot B2 which had steeper slopes than trees in the control as a funct ion of increasing height (Fig. 5 6). In absolute terms, R hat and dbh were equally important in avoidin g fire induced tree mortality, while wood density and tree height (in particular) were apparently less important. Overall, it is difficult to separate the contribution of each variable to mortality, given that they are collinear and have different samples sizes. Discussion Cambium Insulation The findings from the heating experiment support the widely accepted notion that bark thickness is the single most important trait in insulating the vascular cambium and thus in preventing fire induced cambium damage ( Dickinson, 2002; Uhl and Kauffman, 1990; Pinard and Huffman, 1997; Michaletz and Johnson, 2007). Bark thickness alone accounted for 82% of the rate of change in cambium temperature (expressed as R) as a function of cumulative air temperature on the outer b ark Nevertheless, the presence of water in the bark also had substantial and potentially contrasting influence s on R On one hand, water in the bark constrained bark temperatures to an asymptote at 100 C, even when the temperature of the air adjacent to the outer bark was much higher than 100 C. This finding indicates that theoretical models of fire induced cambial damage that do not account for latent heat of vaporization in the bark (e.g., Dickinson, 2002) could underestimate the differential in temperature between the fire and the cambium and, in turn, underestimat e the time before fire induced cambial damage occurs. On the other hand, the high conductivity of the water in the bark could increase bar k diffusivity compared to dry bark tissues thus increasing R (i.e., faster increases in CT during fires). In fact, it is suggested that the smaller the proportion of living tissues in the bark

PAGE 96

96 (i.e., live inner vs. dead outer bark), the greater the cambiu m insulation, because dry, corky bark has numerous air spaces that provide better fire insulation (i.e., lower heat conductivity) than wet, dense phloem (Devet 1940 cited in Hare 1965) It is noteworthy that the opposite result for trees in a mixed conifer forest has been reported (van Mantgem and Schwartz 2003) : the greater the proportion of bark composed by the phloem, the lower CT, presumably due to the high heat capacity of the phloem. Based on fig 2 i n that study (p.346), nevertheless, it is reasonable to speculate that the latent heat of vaporization was the major factor influencing the rates of increase in CT for smaller trees (e.g., those with greater proportion of the bark composed by the phloem), as opposed to heat conductivity or heat capacity [see Reifsnyder et al. (1967) for a detailed discussion on t he effects of BMC on heat capacity] I suggest that the roles of bark water c ontent are important and should be studied in greater depth. Bark Correlates with T r ee Mortality Although BT appears to be the most important plant trait in preventing fire indu ced cambial damage, its importance in preventing fire induced tree mortality has not been widely tested for tropical forest trees and may be modulated by other traits For example, during low intensity fires, heat is often distributed unevenly around the b oles of trees, causing only partial damage to the cambium around a small proportion of the circumference of tree boles (Michaletz and Johnson, 2007). In such cases, plant traits that reduce the indirect (and negative) influences of fire on the physiology a nd structure of tree trunks may be even more important in avoiding tree mortality than BT alone, which provides cambium protection against the direct effects of fire Here, I show that R hat (and hence BT) was an important predictor of direct fire induced tree mortality particularly for moderate to high intensity fires Trees experiencing the fire of 2007

PAGE 97

97 (which followed 1, 2, or 3 fires) were extremely fire resistant for B 100% probability of surviving) and fire sensitive for BT 5.3 mm (0 25% probability of survivorship). This threshold is in the same range as that suggested by Pinard and Huffman (1997) who found that trees with BT 18 mm exhibited h igh resistance to understory forest fires. Despite the generally strong relationship between tree mortality and bark thickness, I found no meaningful relationship between tree mortality and cambium insulation for trees charred only once in either the 2004 or 2007 fires Furthermore, there were two common thin barked species, Aspidosperma excelsum and Sloanea escheri that exhibited high survivorship even when charred (Balch et al. submitted ). These findings indicate that mechanisms associated with plant tr aits other than cambium insulation could also influence fire induced tree mortality. For example, wood density has been shown to slow the progress of wood decay of damaged living trees (Romero and Bolker 2008) which should reduce the likelihood of mortality after cambium exposing damage. Although our data suggest that wood density is important for post fire survival, the evidence is not compelling. Similarly, I found that tree mortality decreased with increasing dbh. Although this relationship could result solely from the strong and positive relationship between BT and tree size, an alternate (or concurrent) explanation is that large trees are less likely to be killed by other falling trees and that the likelihood of the entire circumference being exposed to cambium killing temperatures decreases with bole size (Gutsell and Johnson 1996). However, in this study I was unable to separate the individual effects of dbh from bark thickness on mortality. Finally, while tree height was apparently less important than other traits in

PAGE 98

98 avoiding fire induced mortality, height is expected to become more important during intense fires that have the p otential to damage tree crowns. Forest Resistance to Fire Given the minor effect of the 2004 fires on post fire tree mortality, it is reasonable to infer that this forest is relatively resistant to non recurrent, low intensity fires (as measured by fire in submitted for a discussion about smaller trees). This resistance may result from long term tree adaptations to cope with fire in the region. Although I am unaware of formal estimates fo r fire return interval, transitional forests of Mato Grosso are flammable during most dry seasons due to elevated vapor pressure deficits (Balch et al. 2008; see also Blate 2005) Furthermore, they have been exposed to human associated fire ignition for millennia, given the pre sence of indigenous groups in the region who make use of fire as a management tool (e.g., for hunting and agriculture). It is nevertheless apparent that the fire resistance of this tropical forest was reduced substantially by recurrent fires, particularly near the forest edge adjacent to an open field. This edge effect is likely related to higher fire intensity caused by drier fuels, to higher density of small with low resistance to fire induced cambium damage, or both. Also, changes in forest microclimate alone could lead to increased fire induced tree mortality (Cochrane and Laurance 2002) Pred icted decreases in rainfall over portions of the tropics may cause extensive forest dieback in the next 50 to 100 years particularly in the eastern Amazon (Cox et al. 2000; Malhi et al. 2008) Synergistic effects of recurrent fires and forest degradation, however, could drive this process much more rapidly (Nepstad et al. 2008) In predicting the rate of forest loss due to recurrent fires, it is evident that while cambial

PAGE 99

99 insulation is an important determinant of fire induced tree mortality other plant traits also play important roles in avoiding fire related death. Furthermore, the evidence that fire related mortality occurs even when tree boles do not exhibit physical charring reminds us that fires influence several processes at the stand level. Examples of these processes may include proximate fire induced branch or tree fall, fire damage to roots o r leaves, or various indirect processes triggered by fire (e.g., disease, drought related death). I conclude that in depth assessments of fire related mortality in tropical forests are needed for predicting the trajectories of Amazonian forests in the face of expected increases in fire associated with frontier expansion and episodic droughts related to climate change.

PAGE 100

100 Figure 5 1. Map showing the spatial distribution of trees located in three 50 ha plots: plot A (Control); plot B2 (burned in 2004 and 2007 ); and, plot B4 (burned yearly between 2004 and 2007). Each color represents the number of times a given tree was charred. The size of symbols represents the dbh of each tree. Note that only trees 40 cm dbh were censused across the entire 150 ha. Trees < 40 cm were sampled along 6 transects (black arrows).

PAGE 101

101 Figure 5 2 Overall results from the bark heating experiment: A) air temperature adjacent to the outer bark heated by a propane torch (black lines) and cambium temperature (gray lines) as a function of time; B) cambium temperature as a function of cumulative air temperature adjacent to the outer bark ( CTF ). The dashed red lines in both figures represent 100 o C. Simulations 1 3 represent three e xamples of R: lower quantile, median, and upper quantile.

PAGE 102

102 Figure 5 3. Relationships between rates of change in cumulative heating ( R) and A) bark thickness, B) bark density, and C) bark moisture content. Figure D indicates the independent effect of each variable on R

PAGE 103

103 Figure 5 4. Survival p robabilit ies of trees with dbh 10 cm dbh as a function of R hat a proxy of cambium insulation derived from the heating experiment Each panel represents one logistic model (Models 1, 2, 3, and 4). Each color re presents the number of times a given trees was charred in different treatments The first x axis represents the standardized rate of change (R hat ),while the second x axis indicates the bark thickness associated with a given R hat (note that R and bark thick ness are inversely related).

PAGE 104

104 Figure 5 5 Relative influence of R hat tree height, dbh, and wood density on tree mortality for different fire treatments. On the x axis, the control (A) is represented by green, B2 by orange, and B4 by blue. Note that the influence of each variable was inferred from on the magnitude of the difference between slopes of the control and fire treatments. Note also that all co variates of mortality were standardized (centered and then divided by 2 times the standard deviation) and that R is presented in absolute terms for comparison with other plant traits.

PAGE 105

105 Figure 5 6. Probability of tree survival as a function of standardized tree height (left panel ), dbh (middle panel ), and wood density (right panel ). Each color represent s a different fire treatment.

PAGE 106

106 CHAPTER 6 CONCLUSION The synergistic effects of drought forest degradation by logging, landscape fragmentation, and increased fire ignition already exert strong influences on the stability of tropical forests. In combination with climate change, they may cause large portions of tropical forests to suffer dieback events with important implications for forest carbon stocks, product yields, and biodiversity Yet, very little is known about the thresholds in environmental stresses beyond which regime sh ifts may occur. In my dissertation, I assessed the vulnerability of Amazonian forests to severe drought and fire events, with focus on the C cycle. I first report on findings from a partial throughfall experiment located in a moist eastern Amazonian fores t. In particular, I show that wood increment, net primary productivity (NPP), and leaf area index (LAI) all declined in response to the drought treatment of 2 3 years. Following protracted water stress, substantial increases in mortality of large tree s were observed, corresponding to sustained reductions in plant available water. The results from this experiment suggest that, if the general response of Amazonian forests to drought is similar to those recorded at the Tapajos Forest, a drier future clim ate is likely to have major consequences to carbon stocks of Amazonian forests (e.g., reductions of 25% of standing live biomass). This view receives support from findings of another throughfall exclusion experiment located in a moist, eastern Amazon fores t (see Meir et al. 2009 for a review of these experiements ) where rates of drought induced tree mortality following 3 4 years of throughfall exclusion were reported within the ranges I found in my dissertation

PAGE 107

107 Whereas the throughfall experiments mentioned above provide new and important insights into drought v egetation relations in Amazonia, they cover a relatively small area and may not represent the spatial variability of forests' susceptibility to drought across the Amazon. To deal with this limitation of field experiments, several studies too k advantage of the 2005 drought (one of the most severe droughts over the past 100 years) to evaluate how forests respond drought over a greater spatial extent. Based on results from a network of permanent plots spread across the Amazon, Phillips et al. (2009) found that several forests were sensitive to the 2005 drought w ith mortality increasing at rates that were high enough to reverse a trend in C accumulation of 0.45 PgC yr 1 to a 1.2 1.6 PgC annual source. The results in Phillip et al. (2009) appear to substantiate the notion that Amazonian forests are, overall, sens itive to severe, episodic droughts. Also these results indicate that, if droughts become more intense and frequent in the future, large amounts of carbon will be released to the atmosphere. Despite the apparently agreement among studies that Amazon ian forests are drought sensitive, there is still no agreement about the thresholds of drying of the basin that would cause regime shifts. In a controversial article, Samanta et al. ( 2010 ) claimed to debunk some conclusions from most recent IPCC report th at portions of dense forests could "react drastically" to "slight reductions" in precipitation in the near future resulting from climate change More specifically, Samanta et al. 2010 used satellite based measurements of vegetation photosynthetic activity from MODIS to show that Amazonian vegetation remained relatively irresponsive to the drought encountered in 2005 These results contrast with previous studies that assessed forest responses to the drought of 2005; studies that showed that vegetation over A mazonia either became

PAGE 108

108 more productive (Saleska et al. 2007) or experience elevated tree mortality (Phillips et al. 2009) Given the se apparently conflicting results among different studies o f the climate productivity relationship, and the importance of th ese results for understanding the impacts of climate change on the terrestrial carbon cycle, there was clearly a need to properly quantify the mechanisms driving the variability of vegetation productivity of effort to reconcile results from these field and satellite based studies, I conducted two experiments First, using a CARLUC process based model, I estimated the effects of climate variability on total carbon stocks and fluxes of vegetation across t he Basin and of forests where permanent plots from RAINFOR are located. I hypothesized that CARLUC simulations, which were driven by climate variability alone, would describe C accumulation over the Amazon from 1998 to 2004, but C losses during the 2005 dr ought (Phillips et al. 2009) W hile CARLUC simulated increases in C pools during the dry season from 1995 to 2004, followed by a reduction in 2005, the magnitude of the reduction reported in my dissertation was less pronounced than in Phillips et al. (2009 ) One possible explanation for this difference between studies is that the Tapajos Forest (where the mortality function in CARLUC was developed) was more drought resistance than RAINFOR forests on average. Second, I conducted a study that combined analysis of climate data, field measurements of forest photosynthesis, and an improved satellite based measure of forest photosynthetic activity (EVI). From this study, I show that none of the climatic variables tested (precipitation, vapor pressure defici t, photosynthetically active radiation) contributed to explaining inter annual variability of EVI across the Amazon Instead, I

PAGE 109

109 found evidence that suggest that production of new leaves could play an important role in Basin wide, inter annual EVI variabil ity, but not necessarily in changes in GPP or NPP. These results suggest that it is still unclear whether EVI can be employed as a tool to measure the vulnerability of tropical forests to drought Also, they suggest that several mechanisms could operate si multaneously to explain the contrasting results between field based studies and those that use satellite based measurements to assess forests' vulnerability to drought For example, GPP (expressed as EVI) appears to increase due to the production of new le aves and increased PAR during droughts whereas aboveground net primary productivity (ANPP) measured in the field concurrently decreased because of higher tree mortality and increased respiration associated with lower PAW and high er temperatures. Moreover, the allocation of non structural carbohydrates to belowground processes may have increased, given that GPP was higher and ANPP lower during the drought of 2005. It is also possible that MODIS cannot detect deaths of the large trees that store large amounts of carbon but that account for only a small portion of the canopy. This should be the topic of further research. After evaluating the vulnerability of Amazonian forests to drought, I studied the mechanisms associated with fire induced tree m ortality in a tropical forest in the southern Amazon Based on this study, I present two novel findings that improve our ability to characterize tropical forest vulnerability to fire First, I found that the presence of water in the bark reduced the differential in temperature between the fire and vascular cambium due to boiling water in the bark. This result is important because it shows that theoretical models of fire induced damage that do no t account for the effects

PAGE 110

110 of latent heat vaporization (boiling water) in the bark substantially underestimate the time before fire induced cambial damage occurs. Second, I show that the experimental fires influenced mortality through several processes tha t were not directly related to cambium protection against fire induced damage, suggesting that plant traits that reduce the negative and indirect effects of fire on post fire tree mortality should also be considered in the assessments of tropical forests' vulnerability to fire. In conclusion, I found evidence that droughts and fire s exert strong influences on the terrestrial carbon cycling but no evidence s for imminent regime shift s in the region driven by climate change alone However, human activit ies that reduce forest resistance to fire and in turn increase the likelihood of exotic grass invasions into forests may change these conclusions. Poor timber harvest practices, for example, can substantially increase canopy openness of tropi cal forests, which elevates forest flammability by i) promoting sunlight induced drying of fuel loads and ii) increasing rates of fuel load accumulation on the forest floor. A thin canopy can also support the establishment of invasive C 4 grasses from fores t neighboring pastures, which can out compete native vegetation and elevate forest flammability permanently, potentially causing a shift in stable state from tree to grass dominance. Because t he two most common of the intensive land use practices in the Am azon, cattle ranching and swidden agriculture, both use fire as a management tool, ignition sources for forest fires are common (Nepstad et al. et al. 2001) In an extreme scenario, these feedbacks among fire, vegetation, and climate can drive large portions of tropical forests to impoverished systems.

PAGE 111

111 APPENDIX A METEOROLOGICAL STATI ONS Figure A 1 Upper panel: Number of meteorological stations per classes of precipitation (average between July and September); the dashed red line separates the number of mestations with (left) and without (right) a dry season. Note that according to the standard definition of dry season as the average precipitation < 100 mm month 1, the period July September chosen in this study captured the dry season in 76% of the meteorological stations (212 of 280). For the 68 remaining stations (24%), the average precipitation from July to September was not representative of the driest months of the year in 37 of them (1 3% of total) concentrated in the Northern Amazon (black stars).

PAGE 112

112 APPENDIX B VARIABILITY IN VEGET ATION INDEX Here I assessed whether the choice of months (July Sept) could have influenced the spatial patterns of EVI (West East and North South), given the potential for lags in vegetation phenology (e.g., leaf flushing) across the wide ranges of latitude and longitude reported in this study. For instance, one could hypothesize that leaf flushing occurs predominantly in July October in the East Amazon, October November in the Central Amazon, and November/December in the Western Amazon; temporal spatial patterns that could create trends of lower EVI from East to West of the Amazon. Xiao et al. (2006) mapped the months with highest EVI (calculated from MODIS collection 5 reflectances) across the Amazon in 2002 (Fig. 5b in Xiao et al. 2006 ). If spatial patterns of EVI were closely associated with gradients in vegetation phenology, the results from Xiao et al. (2006) are likely to show a consistent shift in the month of highest EVI from West to East and/or South to North of the Amazon. Based on Fig.5 b from that study, I detected no evidence of gradients in vegetation phenology that could explain the strong gradients found in our study. Because Xiao et al. (2006) mapped peak EVI in only one year, I repeated their analysis using the GIMMS AVHRR NDVI data set from 1981 to 2008 (Figure A2a; upper panel ) (Tucker et al. 2005) The GIMMS data generally confirmed the lack of a clear spa tial pattern in the timing of peak EVI/NDVI in the Amazon with one exception: there was a potential gradient in the month with highest NDVI from South (earlier in the dry season) to North (later in the dry season/early in the wet season). To make sure that this potential gradient did not influence our conclusions, all statistical models between EVI and climatic variables, for both forest and non forested areas ( Figure A1b; lower

PAGE 113

113 panel ), included longitude and latitude as co variates. While we do not exclude the possibility that the focus on the July September period may have had some effect on the spatial patterns of EVI, we are confident that the gradients in NBAR EVI from West to East were not merely an artifact of spatial phenological gradients. Also, the inclusion of latitude and longitude in the statistical models between EVI and climatic variables minimized any potential effect of the choice of month in our conclusions. Figure A 2 The month of peak NDVI as recorded in the GIMMS NDVI dataset between 1981 and 2008 for areas of high canopy cover (A) and by a wide range fractions of canopy cover (B)

PAGE 114

114 APPENDIX C EQUATIONS ON HEAT TRANSFER While the equation representing conductive heat transfer through bark (one dimensional mode l; equation C1) can be useful for predicting the rates of change in cambium temperature (a proxy of cambium insulation) as a function of fire temperature, this equation requires information about diffusivity, which is very difficult to measure under field conditions and goes beyond of the scope of this work. Also, equation 1A requires a constant heat source, which is challenging under field conditions due to air movement during heating experiments. Thus, I decided to use Newton's Law of Cooling to calculate the rate of change in cambium temperature given a gradient in temperature (equation C2). In so doing, I could not separate the individual effects of conductivity and convection. Note that the exact solution of Newton's Law of Cooling (equation 1C) is an a symptotic model (i.e., the maximum temperature the cambium can reach is equal to the hot air temperature on the outer bark). However, I used basic theory of conductive heat transfer to drive my predictions (1, 2, and 3). equation C1: where: T camb = Cam bium temperature (CT) T fire = Temperature of the flame T cambi = Initial temperature at the cambium BT = Bark Thickness

PAGE 115

115 t = time Based on this relationship, the time for mortality at different temperatures (T f ) can be predicted for an individual: time = constant*Bt 2 equation C2 whree: CT: Cambium temperatu re R: Rate of change in temperature T f : Hot air temperature adjacent to the bark heated using a torch. equation 1C (exact solution of equation C2) : Where: Asymptote: Tf in equation 1B CTF: cumulative air temperature adjacent to the bark

PAGE 116

116 APPENDIX D FIRE EXPERIMENT \ Figure A 3 Aerial photo of the experimental fire. On the right corner of the figure I show the location of this experiment, Querencia, Mato Grosso State, Brazil.

PAGE 117

1 17 APPENDIX E FLAME HEIGHT Figure A 4 Box plots for flame height in plot B2 (grey) and B4 from 2004 and 2007. Measurements were taken along the fire line and thus were spread across the experimental area in a pattern that depended on the fire behavior.

PAGE 118

118 APPENDIX E BARK THICKNESS AND T REE DBH Figure A 5 Rela tionships between dbh and bark thickness. Note that when there was no relationship between dbh and BT for a given species, predicted BT was based on the intercept of the model.

PAGE 119

119 APPENDIX F MODEL 1 Table F 1. Model 1. Period: 2004 2005 (plot B2 a nd B4 ) Estimate Std. Error z v alue Pr( > | z | ) Intercept 3.461 0.208 16.670 0.000 R 0.615 0.373 1.647 0.100 Charred.0x (B 2 ) 0.610 0.219 2.791 0.005 Charred.04 (B 3 ) 0.812 0.184 4.403 0.000 Charred.0x ( B4 ) 0.351 0.225 1.560 0.119 Charred.04 ( B4 ) 0.597 0.196 3.052 0.002 Edge Distance 0.655 0.142 4.604 0.000 R:Charred.0x (B 2 ) 0.460 0.463 0.993 0.320 R:Charred.04 (B 2 ) 0.435 0.438 0.991 0.322 R:Charred.0x ( B4 ) 0.364 0.503 0.724 0.469 R:Charred.04 ( B4 ) 0.923 0.457 2.017 0.044

PAGE 120

120 APPEND IX G MODEL 2 Table G 1. Model 2. Period: 2004 2007 (plot B2 ) Estimate Std. Error z v alue Pr( > | z | ) Intercept 2.526 0.222 11.391 0.000 R 0.242 0.267 0.907 0.364 Charred.0x(B 2 ) 0.340 0.162 2.103 0.035 Charred.04(B 2 ) 0.609 0.133 4.594 0.000 Edge Distance 0.476 0.133 3.582 0.000 R:Charred.0x(B 2 ) 0.250 0.332 0.755 0.450 R:Charred.04(B 2 ) 0.389 0.293 1.327 0.185

PAGE 121

121 APPENDIX H MODEL 3 Table H 1. Model 3. Period: 2007 2008 (plot B2 ) Estimate Std. Error z v alue Pr( > | z | ) Intercept 3.257 0.233 13.965 0.000 R 0.494 0.338 1.461 0.144 Charred.0x(B 2 ) 1.289 0.425 3.035 0.002 Charred.07(B 2 ) 2.190 0.170 12.885 0.000 Charred.2x(B 2 ) 2.308 0.154 14.962 0.000 Edge Distance 1.303 0.138 9.462 0.000 R:Cha rred.0x(B 2 ) 1.703 1.007 1.691 0.091 R:Charred.07(B 2 ) 0.468 0.372 1.257 0.209 R:Charred.2x(B 2 ) 1.339 0.366 3.656 0.000

PAGE 122

122 LIST OF REFERENCES Aide, T. M. 1993. Patterns of Leaf Development and Herbivory in a Tropical Understory C ommunity. Ecology 74:455 466. Alencar, A., D. Nepstad, and M. C. Diaz. 2006. Forest understory fire in the Brazilian Amazon in ENSO and non ENSO years: area burned and committed carbon emissions. Earth Interactions 10:1 17. Alencar, A. A., L. A. Solrzan o, and D. C. Nepstad. 2004. Modeling forest understory fires in an eastern Amazonian landscape. Ecological Applications 14:139 149. Aragao, L. E. O. C., Y. Malhi, R. M. Roman Cuesta, S. Saatchi, L. O. Anderson, and Y. E. Shimabukuro. 2007. Spatial pattern s and fire response of recent Amazonian droughts. Geophysical Research Letters 34: doi: Artn L07701 Doi 10.1029/2006gl028946. Aragao, L., Y. Malhi, N. Barbier, A. Lima, Y. Shimabukuro, L. Anderson, and S. Saatchi. 2008. Interactions between rainfall, de forestation and fires during recent years in the Brazilian Amazonia. Philosophical Transactions of the Royal Society B: Biological Sciences 363:1779 1785. Baayen, H. 2008. Analyzing linguistic data: A practical introduction to statistics using R. Cambridg e University Press. Baker, T. R., O. L. Phillips, Y. Malhi, S. Almeida, L. Arroyo, A. Di Fiore, T. Erwin, N. Higuchi, T. J. Killeen, S. G. Laurance, W. F. Laurance, S. L. Lewis, A. Monteagudo, D. A. Neill, P. N. Vargas, N. C. A. Pitman, J. N. M. Silva, an d R. V. Martinez. 2004a. Increasing biomass in Amazonian forest plots. Philos. Trans. R. Soc. Lond. Ser. B Biol. Sci. 359:353 365. Baker, T. R., O. L. Phillips, Y. Malhi, S. Almeida, L. Arroyo, A. Di Fiore, T. Erwin, N. Higuchi, T. J. Killeen, S. G. Laura nce, W. F. Laurance, S. L. Lewis, A. Monteagudo, D. A. Neill, P. Nez Vargas, N. C. A. Pitman, J. N. M. Silva, and R. Vsquez Martnez. 2004b. Increasing biomass in Amazonian forest plots. Philosophical Transactions of the Royal Society B: Biological Scie nces 359:353 365. doi: 10.1098/rstb.2003.1422. Balch, J., D. Nepstad, P. Brando, L. Curran, O. Portela, O. de Carvalho, and P. Lefebvre. 2008. Negative fire feedback in a transitional forest of southeastern Amazonia. Global Change Biology 14:2276 2287. B arlow, J., T. Haugaasen, and C. A. Peres. 2002. Effects of ground fires on understorey bird assemblages in Amazonian forests. Biological Conservation 105:157 169. Barlow, J., B. O. Lagan, and C. A. Peres. 2003. Morphological correlates of fire induced tre e mortality in a central Amazonian forest. Journal of Tropical Ecology 19:291 299.

PAGE 123

123 Barlow, J., and C. A. Peres. 2008. Fire mediated dieback and compositional cascade in an Amazonian forest. Philosophical Transactions of the Royal Society B: Biological Sci ences 363:1787. Barlow, J., and C. A. Peres. 2006. Effects of single and recurrent wildfires on fruit production and large vertebrate abundance in a central Amazonian forest. Biodiversity and Conservation 15:985 1012. Bates, D., and M. Maechler. 2009. lm e4: Linear mixed effects models using S4 classes. R package version 0.999375 31. Bates, D., and D. Sarkar. 2006. lme4: Linear mixed effects models using S4 classes. URL http://CRAN. R project. org, R package version 0.99875 8. Betts, R. A., P. M. Cox, M. Collins, P. P. Harris, C. Huntingford, and C. D. Jones. 2004. The role of ecosystem atmosphere interactions in simulated Amazonian precipitation decrease and forest dieback under global climate warming. Theoretical and Applied Climatology 78. doi: 10.1007 /s00704 004 0050 y. Betts, R., M. Sanderson, and S. Woodward. 2008. Effects of large scale Amazon forest degradation on climate and air quality through fluxes of carbon dioxide, water, energy, mineral dust and isoprene. Philosophical Transactions of the R oyal Society B: Biological Sciences 363:1873 1880. Blate, G. M. 2005. Modest trade offs between timber management and fire susceptibility of a Bolivian semi deciduous forest. Ecological Applications 15:1649 1663. Bolker, B. M., M. E. Brooks, C. J. Clark, S. W. Geange, J. R. Poulsen, M. H. Stevens, and J. S. White. 2009. Generalized linear mixed models: a practical guide for ecology and evolution. Trends in Ecology & Evolution 24:127 135. Bonan, G. B. 2008. Forests and climate change: forcings, feedbacks, and the climate benefits of forests. Science 320:1444. Brando, P. M., D. C. Nepstad, E. A. Davidson, S. E. Trumbore, D. Ray, and P. Camargo. 2008b. Drought effects on litterfall, wood production and belowground carbon cycling in an Amazon forest: results of a throughfall reduction experiment. Philosophical Transactions of the Royal Society B: Biological Sciences 363:1839 1848. Burnham, K. P., and D. R. Anderson. 2002. Model selection and multimodel inference: a practical information theoretic approach. S pringer Verlag. Chambers, J. Q., J. Santos, R. J. Ribeiro, and N. Higuchi. 2001. Tree damage, allometric relationships, and above ground net primary production in central Amazon forest. Forest Ecology and Management 152:73 84.

PAGE 124

124 Chapin III, F. S., E. Schul ze, and H. A. Mooney. 1990. The ecology and economics of storage in plants. Annual Review of Ecology and Systematics 21:423 447. Chave, J., H. C. Muller Landau, T. R. Baker, T. A. Easdale, H. Steege, and C. O. Webb. 2006. Regional and phylogenetic variati on of wood density across 2456 neotropical tree species. Ecological Applications 16:2356 2367. Christensen, J. H., B. Hewitson, A. Busuioc, A. Chen, X. Gao, I. Held, R. Jones, R. K. Kolli, W. T. Kwon, R. Laprise, and others. 2007. Climate change 2007: the physical science basis. Contribution of Working Group I to the Fourth Assessment Report of the Intergovernmental Panel on Climate Change. Cambridge University Press Cambridge, New York. Clark, D. A. 2002. Are tropical forests an important carbon sink? Re analysis of the long term plot data. Ecological Applications 12:3 7. Clark, D. A., and D. B. Clark. 1994. Climate induced annual variation in canopy tree growth in a Costa Rican tropical rain forest. Journal of Ecology 82:865 872. Clark, D. A., S. Brown, D. W. Kicklighter, J. Q. Chambers, J. R. Thomlinson, and J. Ni. 2001. Measuring net primary production in forests: concepts and field methods. Ecological Applications 11:356 370. Cochrane, M. A., and W. F. Laurance. 2002. Fire as a large scale edge effec t in Amazonian forests. Journal of Tropical Ecology 18:311 325. Cochrane, M. A., and W. F. Laurance. 2008. Synergisms among fire, land use, and climate change in the Amazon. AMBIO: A Journal of the Human Environment 37:522 527. Cochrane, M. A. 2003. Fire science for rainforests. Nature 421:913 919. Cochrane, M. A., and W. F. Laurance. 2002. Fire as a large scale edge effect in Amazonian forests. Journal of Tropical Ecology 18. doi: 10.1017/S0266467402002237. Condit, R., S. Aguilar, A. Hernandez, R. Perez S. Lao, G. Angehr, S. P. Hubbell, and R. B. Foster. 2004. Tropical forest dynamics across a rainfall gradient and the impact of an El Nino dry season. Journal of Tropical Ecology 20:51 72. Condit, R., S. P. Hubbell, and R. B. Foster. 1995. Mortality rat es of 205 neotropical tree and shrub species and the impact of a severe drought. Ecological Monographs:419 439. Costa, M. H., and J. A. Foley. 2000b. Combined effects of deforestation and doubled atmospheric CO2 concentrations on the climate of Amazonia. Journal of Climate 13:18 34.

PAGE 125

125 Cox, P. M., R. A. Betts, M. Collins, P. P. Harris, C. Huntingford, and C. D. Jones. 2004. Amazonian forest dieback under climate carbon cycle projections for the 21st century. Theoretical and Applied Climatology 78. doi: 10. 1007/s00704 004 0049 4. Cox, P. M., P. P. Harris, C. Huntingford, R. A. Betts, M. Collins, C. D. Jones, T. E. Jupp, J. A. Marengo, and C. A. Nobre. 2008. Increasing risk of Amazonian drought due to decreasing aerosol pollution. Nature 453:212 215. Cox, P M., R. A. Betts, C. D. Jones, S. A. Spall, and I. J. Totterdell. 2000. Acceleration of global warming due to carbon cycle feedbacks in a coupled climate model. Nature 408:184 187. doi: 10.1038/35041539. Cramer, W., A. Bondeau, F. I. Woodward, I. C. Pren tice, R. A. Betts, V. Brovkin, P. M. Cox, V. Fisher, J. A. Foley, and A. D. Friend. 2001. Global response of terrestrial ecosystem structure and function to CO2 and climate change: results from six dynamic global vegetation models. Global Change Biology 7: 357 373. Davidson, E. A., F. Y. Ishida, and D. C. Nepstad. 2004. Effects of an experimental drought on soil emissions of carbon dioxide, methane, nitrous oxide, and nitric oxide in a moist tropical forest. Global Change Biology 10:718 730. Davidson, E. A ., L. V. Verchot, J. H. Cattnio, I. L. Ackerman, and J. E. M. Carvalho. 2000. Effects of soil water content on soil respiration in forests and cattle pastures of eastern Amazonia. Biogeochemistry 48:53 69. Davidson, E. A., I. A. Janssens, and Y. Luo. 200 6. On the variability of respiration in terrestrial ecosystems: moving beyond Q10. Global Change Biology 12:154 164. Dickinson, M. & Johnson, E. (2001) Fire effects on trees In: Forest fires: behavior and ecological effects, Academic Press, 2001 477 521 Doughty, C. E., and M. L. Goulden. 2008. Seasonal patterns of tropical forest leaf area index and CO2 exchange. J Geophys Res 113. Fisher, R. A., M. Williams, A. L. da Costa, Y. Malhi, R. F. da Costa, S. Almeida, and P. Meir. 2007. The response of an Ea stern Amazonian rain forest to drought stress: results and modelling analyses from a throughfall exclusion experiment. Global Change Biology 13:2361 2378. Fisher, R., M. Williams, R. L obo D o V ale A. L ola D a C osta and P. Meir. 2006. Evidence from Amazoni an forests is consistent with isohydric control of leaf water potential. Plant, cell and environment(Print) 29:151 165. Foley, J. A., S. Levis, I. C. Prentice, D. Pollard, and S. L. Thompson. 1998. Coupling dynamic models of climate and vegetation. Glob. Change Biol. 4:561 579.

PAGE 126

126 Foley, J. A., I. C. Prentice, N. Ramankutty, S. Levis, D. Pollard, S. Sitch, and A. Haxeltine. 1996. An integrated biosphere model of land surface processes, terrestrial carbon balance, and vegetation dynamics. Glob. Biogeochem. Cy cle 10:603 628. Gerwing, J. J., and D. L. Farias. 2000. Integrating liana abundance and forest stature into an estimate of total aboveground biomass for an eastern Amazonian forest. Journal of tropical ecology 16:327 335. Gloor, M., O. L. Phillips, J. J. Lloyd, S. L. Lewis, Y. Malhi, T. R. Baker, G. Lpez Gonzlez, J. Peacock, S. Almeida, A. C. de OLIVEIRA, and others. 2009. Does the disturbance hypothesis explain the biomass increase in basin wide Amazon forest plot data? Global Change Biology 15:2418 24 30. Golding, N., and R. Betts. 2008. Fire risk in Amazonia due to climate change in the HadCM3 climate model: Potential interactions with deforestation. Global Biogeochemical Cycles 22. Grace, J., J. Lloyd, J. McIntyre, A. C. Miranda, P. Meir, H. S. Mira nda, C. Nobre, J. Moncrieff, J. Massheder, and Y. Malhi. 1995. Carbon dioxide uptake by an undisturbed tropical rain forest in southwest Amazonia, 1992 to 1993. Science 270:778. Graham, E. A., S. S. Mulkey, K. Kitajima, N. G. Phillips, and S. J. Wright. 2 003b. Cloud cover limits net CO2 uptake and growth of a rainforest tree during tropical rainy seasons. Proceedings of the National Academy of Sciences 100:572 576. Gutsell, S. L., and E. A. Johnson. 1996. How fire scars are formed: coupling a disturbance process to its ecological effect. Canadian Journal of Forest Research 26:166 174. Hansen, M. C., R. S. DeFries, J. R. G. Townshend, M. Carroll, C. Dimiceli, and R. A. Sohlberg. 2003. Global percent tree cover at a spatial resolution of 500 meters: First r esults of the MODIS vegetation continuous fields algorithm. Earth Interactions 7:1 15. Hare, R. C. 1965. Contribution of bark to fire resistance of southern trees. Journal of Forestry 63:248 251. Hashimoto, H, F. Melton, F, K. Ichii, K, C. Milesi, C, W. Wang, W, and R. R. Nemani, R. 2009. Evaluating the impacts of climate and elevated carbon dioxide on tropical rainforests of the western Amazon basin using ecosystem models and satellite data. Global Change Biology Hirsch, A. I., W. S. Little, R. A. Houg hton, N. A. Scott, and J. D. White. 2004. The net carbon flux due to deforestation and forest re growth in the Brazilian Amazon: analysis using a process based model. Global Change Biology 10:908 924. doi: 10.1111/j.1529 8817.2003.00765.x.

PAGE 127

127 Houghton, R. A 2005. Aboveground Forest Biomass and the Global Carbon Balance. Global Change Biology 11:945 958. doi: 10.1111/j.1365 2486.2005.00955.x. Huete, A., K. Didan, T. Miura, E. P. Rodriguez, X. Gao, L. G. Ferreira, and others. 2002. Overview of the radiometri c and biophysical performance of the MODIS vegetation indices. Remote Sensing of Environment 83:195 213. Huete, A. R., K. Didan, Y. E. Shimabukuro, P. Ratana, S. R. Saleska, L. R. Hutyra, W. Yang, R. R. Nemani, and R. Myneni. 2006a. Amazon rainforests gre en up with sunlight in dry season. Geophysical Research Letters 33. Huntingford, C., R. A. Fisher, L. Mercado, B. B. Booth, S. Sitch, P. P. Harris, P. M. Cox, C. D. Jones, R. A. Betts, Y. Malhi, G. R. Harris, M. Collins, and P. Moorcroft. 2008. Towards qu of the Royal Society B: Biological Sciences 363:1857 1864. Hurlbert, S. H. 1984. Pseudoreplication and the design of ecological field experiments. Ecological monographs 5 4:187 211. Hutyra, L. R., J. W. Munger, S. R. Saleska, E. Gottlieb, B. C. Daube, A. L. Dunn, D. F. Amaral, P. B. De Camargo, and S. C. Wofsy. 2007. Seasonal controls on the exchange of carbon and water in an Amazonian rain forest. J. Geophys. Res 112. Ji pp, P. H., D. C. Nepstad, D. K. Cassel, and C. Reis de Carvalho. 1998. Deep soil moisture storage and transpiration in forests and pastures of seasonally dry Amazonia. Climatic Change 39:395 412. Landsberg, J. J., and R. H. Waring. 1997. A generalised mod el of forest productivity using simplified concepts of radiation use efficiency, carbon balance and partitioning. Forest Ecology and Management 95:209 228. Larjavaara, M., and H. C. Muller Landau. 2010. Rethinking the value of high wood density. Functiona l Ecology. Li, W., R. Fu, and R. E. Dickinson. 2006. Rainfall and its seasonality over the Amazon in the 21st century as assessed by the coupled models for the IPCC AR4. Journal of Geophysical Research 111:D02111. Lloyd, J., and G. D. Farquhar. 2008. Eff ects of rising temperatures and [CO2] on the physiology of tropical forest trees. Philosophical Transactions of the Royal Society B: Biological Sciences 363:1811 1817. Malhi, Y., E. Luiz, D. Aragao, C. Huntingford, R. Fisher, P. Zelazowski, S. Sitch, C. M cSweeney, and P. Meir. 2008a. Exploring the likelihood and mechanism of a climate change induced dieback of the Amazon rainforest. Proceedings of the National Academy of Sciences.

PAGE 128

128 Malhi, Y., J. T. Roberts, R. A. Betts, T. J. Killeen, W. Li, and C. A. Nobre 2008b. Climate change, deforestation, and the fate of the Amazon. Science 319:169. Malhi, Y., L. E. O. C. Arago, D. Galbraith, C. Huntingford, R. Fisher, P. Zelazowski, S. Sitch, C. McSweeney, and P. Meir. 2009. Special Feature: Exploring the likelihoo d and mechanism of a climate change induced dieback of the Amazon rainforest. Proceedings of the National Academy of Sciences of the United States of America. doi: 10.1073/pnas.0804619106. van Mantgem, P., and M. Schwartz. 2003. Bark heat resistance of sm all trees in Californian mixed conifer forests: testing some model assumptions. Forest Ecology and Management 178:341 352. Marengo, J. A., C. A. Nobre, J. Tomasella, M. F. Cardoso, and M. D. Oyama. 2008a. Hydro climatic and ecological behaviour of the dro ught of Amazonia in 2005. Philosophical Transactions of the Royal Society B: Biological Sciences 363:1773 1778. Marengo, J. A., C. A. Nobre, J. Tomasella, M. D. Oyama, G. Sampaio de Oliveira, R. de Oliveira, H. Camargo, L. M. Alves, and I. F. Brown. 2008c The Drought of Amazonia in 2005. Journal of Climate 21:495. doi: 10.1175/2007JCLI1600.1. Mayle, F. E., and M. J. Power. 2008. Impact of a drier Early Mid Holocene climate upon Amazonian forests. Philosophical Transactions of the Royal Society B: Biologi cal Sciences 363:1829 1838. Meir, P., P. Brando, D. Nepstad, Vasconcelos, S, A. Costa, E. A. Davidson, S. Almeida, R. A. Fisher, E. Sotta, D. Zarin, and G. Cardinot. 2009. The Effects of Drought on Amazonian rain forests. In: Large Scale Biosphere Atmosph ere P roject in Amazonia: A synthesis Michaletz, S., and E. Johnson. 2007. How forest fires kill trees: A review of the fundamental biophysical processes. Scandinavian Journal of Forest Research 22:500 515. doi: 10.1080/02827580701803544. Murray, K., and M. M. Conner. 2009. Methods to quantify variable importance: implications for the analysis of noisy ecological data. Ecology 90:348 355. Myneni, R. B., W. Yang, R. R. Nemani, A. R. Huete, R. E. Dickinson, Y. Knyazikhin, K. Didan, R. Fu, R. I. Negrn Jure z, and S. S. Saatchi. 2007a. Large seasonal swings in leaf area of Amazon rainforests. Proceedings of the National Academy of Sciences 104:4820. Nakagawa, M., K. Tanaka, T. Nakashizuka, T. Ohkubo, T. Kato, T. Maeda, K. Sato, H. Miguchi, H. Nagamasu, K. Ogino, and others. 2000a. Impact of severe drought associated with the 1997 1998 El Nino in a tropical forest in Sarawak. Journal of Tropical Ecology 16:355 367.

PAGE 129

129 Nepstad, D., G. Carvalho, A. Cristina Barros, A. Alencar, J. Paulo Capobianco, J. Bishop, P. Moutinho, P. Lefebvre, U. Lopes Silva, and E. Prins. 2001. Road paving, fire regime feedbacks, and the future of Amazon forests. Forest Ecology and Management 154:395 407. Nepstad, D., P. Lefebvre, U. Lopes da Silva, J. Tomasella, P. Schles inger, L. Solorzano, P. Moutinho, D. Ray, and J. Guerreira Benito. 2004a. Amazon drought and its implications for forest flammability and tree growth: a basin wide analysis. Global Change Biology 10:704 717. Nepstad, D. C., C. R. de Carvalho, E. A. Davids on, P. H. Jipp, P. A. Lefebvre, G. H. Negreiros, E. D. da Silva, T. A. Stone, S. E. Trumbore, and S. Vieira. 1994. The role of deep roots in the hydrological and carbon cycles of Amazonian forests and pastures. Nepstad, D. C., I. M. Tohver, D. Ray, P. Mou tinho, and G. Cardinot. 2007a. Mortality of large trees and lianas following experimental drought in an Amazon forest. Ecology 88:2259 2269. Nepstad, D. C., P. Moutinho, M. B. Dias Filho, E. Davidson, G. Cardinot, D. Markewitz, R. Figueiredo, N. Vianna, J Chambers, and D. Ray. 2002a. The effects of partial throughfall exclusion on canopy processes, aboveground production, and biogeochemistry of an Amazon forest. Journal of Geophysical Research 107:8085. Nepstad, D. C., C. M. Stickler, B. S. Filho, and F. Merry. 2008. Interactions among Amazon land use, forests and climate: prospects for a near term forest tipping point. Philosophical Transactions of the Royal Society B: Biological Sciences 363:1737 1746. Nepstad, D. C., A. Verissimo, A. Alencar, C. Nobre E. Lima, P. Lefebvre, P. Schlesinger, C. Potter, P. Moutinho, and E. Mendoza. 1999. Large scale impoverishment of Amazonian forests by logging and re. Nature 398:505 508. Nobre, C. A., P. J. Sellers, and J. Shukla. 1991a. Amazonian deforestation and re gional climate change. Journal of Climate 4:957 988. Oliveira, R. S., T. E. Dawson, S. S. O. Burgess, and D. C. Nepstad. 2005. Hydraulic redistribution in three Amazonian trees. Oecologia 145:354 363. doi: 10.1007/s00442 005 0108 2. Phillips, O. L., Y. M alhi, N. Higuchi, W. F. Laurance, P. V. Nez, R. M. Vsquez, S. G. Laurance, L. V. Ferreira, M. Stern, and S. Brown. 1998b. Changes in the carbon balance of tropical forests: evidence from long term plots. Science 282:439.

PAGE 130

130 Phillips, O. L., L. E. O. C Aragao, S. L. Lewis, J. B. Fisher, J. Lloyd, G. Lopez Gonzalez, Y. Malhi, A. Monteagudo, J. Peacock, C. A. Quesada, G. van der Heijden, S. Almeida, I. Amaral, L. Arroyo, G. Aymard, T. R. Baker, O. Banki, L. Blanc, D. Bonal, P. Brando, J. Chave, A. C. A. de Oliveira, N. D. Cardozo, C. I. Czimczik, T. R. Feldpausch, M. A. Freitas, E. Gloor, N. Higuchi, E. Jimenez, G. Lloyd, P. Meir, C. Mendoza, A. Morel, D. A. Neill, D. Nepstad, S. Patino, M. C. Penuela, A. Prieto, F. Ramirez, M. Schwarz, J. Silva, M. Silve ira, A. S. Thomas, H. T. Steege, J. Stropp, R. Vasquez, P. Zelazowski, E. A. Davila, S. Andelman, A. Andrade, K. Chao, T. Erwin, A. Di Fiore, E. H. C., H. Keeling, T. J. Killeen, W. F. Laurance, A. P. Cruz, N. C. A. Pitman, P. N. Vargas, H. Ramirez Angulo, A. Rudas, R. Salamao, N. Silva, J. Terborgh, and A. Torres Lezama. 2009. Drought Sensitivity of the Amazon Rainforest. Science 323:1344 1347. doi: 10.1126/science.1164033. Pinard, M. A., and J. Huffman. 1997. Fire resistance and bark properties of trees in a seasonally dry forest in eastern Bolivia. Journal of Tropical Ecology:727 740. Pinard, M. A., F. E. Putz, and J. C. Licona. 1999. Tree mortality and vine proliferation following a wildfire in a subhumid tropical forest in eastern Bolivia. Forest Ecol ogy and Management 116:247 252. Pinheiro, J., D. Bates, S. DebRoy, and D. Sarkar. (2000). the R Core team (2008), nlme: Linear and Nonlinear Mixed Effects Models. R package version:3 1. Potter, C., E. Davidson, D. Nepstad, and C. R. de Carvalho. 2001. Ec osystem modeling and dynamic effects of deforestation on trace gas fluxes in Amazon tropical forests. For. Ecol. Manage. 152:97 117. Poulter, B., U. Heyder, and W. Cramer. 2009. Modeling the Sensitivity of the Seasonal Cycle of GPP to Dynamic LAI and Soil Depths in Tropical Rainforests. Ecosystems 12:517 533. doi: 10.1007/s10021 009 9238 4. Reifsnyder, W. E., L. P. Herrington, and K. W. Spalt. 1967. Thermophysical properties of bark of shortleaf, longleaf, and red pine. Yale University School of Forestry Bulletin 70:1 40. Rivera, G., S. Elliott, L. Caldas, G. Nicolossi, V. Coradin, and R. Borchert. 2002. Increasing day length induces spring flushing of tropical dry forest trees in the absence of rain. Trees Structure and Function 16:445 456. doi: 10.100 7/s00468 002 0185 3. Romero, C., and B. M. Bolker. 2008. Effects of stem anatomical and structural traits on responses to stem damage: an experimental study in the Bolivian Amazon. Canadian Journal of Forest Research 38:611 618. Saatchi, S. S., R. A. HOU GHTON, R. Alvala, J. V. Soares, and Y. YU. 2007. Distribution of aboveground live biomass in the Amazon basin. Glob. Change Biol. 13:816 837.

PAGE 131

131 Saldarriaga, J. G., D. C. West, M. L. Tharp, and C. Uhl. 1988. Long term chronosequence of forest succession in the upper Rio Negro of Colombia and Venezuela. The Journal of Ecology:938 958. Saleska, S. R., K. Didan, A. R. Huete, and H. R. da Rocha. 2007b. Amazon forests green up during 2005 drought. Science 318:612 612. doi: DOI 10.1126/science.1146663. S aleska, S. R., S. D. Miller, D. M. Matross, M. L. Goulden, S. C. Wofsy, H. R. da Rocha, P. B. De Camargo, P. Crill, B. C. Daube, and H. C. de Freitas. 2003b. Carbon in Amazon forests: unexpected seasonal fluxes and disturbance induced losses. American Association for the Advancement of Science. Samanta, A., S. Ganguly, H. Hashimoto, S. Devadiga, E. Vermote, Y. Knyazikhin, R. R. Nemani, and R. B. Myneni. 2010. Amazon forests did not green up during the 2005 drought. Geophysical Research Letters 37:L05401. Schaaf, C. B., F. Gao, A. H. Strahler, W. Lucht, X. Li, T. Tsang, N. C. Strugnell, X. Zhang, Y. Jin, and J. P. Muller. 2002. First operational BRDF, albedo nadir reflectance products from MODIS. Remote Sensing of Environment 83:135 148. Tian, H., J. M. Melillo, D W. Kicklighter, A. D. McGuire, J. V. K. Helfrich, B. Moore, and C. J. Vorosmarty. 1998. Effect of interannual climate variability on carbon storage in Amazonian ecosystems. Nature 396:664 667. doi: 10.1038/25328. Tomasella, J., and M. G. Hodnett. 1998. Estimating soil water retention characteristics from limited data in Brazilian Amazonia. Soil science 163:190. Trumbore, S., E. S. Da Costa, D. C. Nepstad, B. De Camargo, L. A. Martinelli, D. Ray, T. Restom, and W. Silver. 2006. Dynamics of fine root carb on in Amazonian tropical ecosystems and the contribution of roots to soil respiration. Global Change Biology 12:217 229. Tucker, C. J., J. E. Pinzn, M. E. Brown, D. A. Slayback, E. W. Pak, R. Mahoney, E. F. Vermote, and N. El Saleous. 2005. An extended A VHRR 8 km NDVI dataset compatible with MODIS and SPOT vegetation NDVI data. International Journal of Remote Sensing 26:4485 4498. Uhl, C., and J. B. Kauffman. 1990. Deforestation, fire susceptibility, and potential tree responses to fire in the eastern Am azon. Ecology:437 449. Van Nieuwstadt, M. G. L., and D. Sheil. 2005a. Drought, fire and tree survival in a Borneo rain forest, East Kalimantan, Indonesia. JOURNAL OF ECOLOGY 93:191 201. doi: 10.1111/j.1365 2745.2004.00954.x.

PAGE 132

132 Venables, W. N., and B. D. Ripley. 2002. Modern applied statistics with S. Springer verlag. Werth, D., and R. Avissar. 2002. The local and global effects of Amazon deforestation. J. Geophys. Res 107. Williamson, G. B., W. F. Laurance, A. A. Oliveira, P. Delamonica, C. Gascon, T. E. Lovejoy, and L. Pohl. 2000. Amazonian Tree Mortality during the 1997 El Nino Drought. Conservation Biology 14:1538. Wright, S. J., and C. P. van Schaik. 1994. Light and the phenology of tropical trees. The American Naturalist 143:192 199. Wright, S. J., C. Carrasco, O. Caldern, and S. Paton. 1999. The El Nino Southern Oscillation, variable fruit production, and famine in a tropical forest. Ecology 80:1632 1647. Xiao, X., Q. Zhang, S. Saleska, L. Hutyra, P. De Camargo, S. Wofsy, S. Frolking, S. Boles, M Keller, and B. Moore. 2005. Satellite based modeling of gross primary production in a seasonally moist tropical evergreen forest. Remote Sensing of Environment 94:105 122. Xiao, X. M., S. Hagen, Q. Y. Zhang, M. Keller, and B. Moore. 2006. Detecting leaf phenology of seasonally moist tropical forests in South America with multi temporal MODIS images. Remote Sensing of Environment 103:465 473. doi: 10.1016/j.rse.2006.04.013. Zeng, N., J. Yoon, J. A. Marengo, A. Subramaniam, C. A. Nobre, A. Mariotti, and J D. Neelin. 2008. Causes and impacts of the 2005 Amazon drought. Environmental Research Letters 3:014002.

PAGE 133

133 BIOGRAPHICAL SKETCH P aulo Brando was born in the town of Assis, five hundred kilometers from the most populated state capital in Brazil Sao Paulo Early in his career, h e decided to work on topics related to conservation in Amazon ia With this goal in mind, he joined the Un iversity of Sao Paulo to study f orestry in 1999 After three years in college learning more about pulp production than conserva tion he was fortunate to join the Forest Institute for a 2 year internship on restoration of savannas (cerrado s ) After developing some skills as a n ecologist during this internship Paulo had the opportunity to study at McGill University Canada where h e was exposed to a vibrant scientific community working on future scenarios for the Amazon Excited by this approach he contacted a Brazilian institution IPAM (Amazon Environmental Research Institute) to find out about the possibilities of conducting si milar research upon his return to Brazil Two weeks following this first contact with IPAM Paulo arrived in Santarem, Para State, where he spent the following three and half years stud ying the effects of droughts on the carbon and water cycling of tropica l forests During this period he worked closely with local community members from whom he learned more about conservation than he had in school or anywhere else Paulo left Santarem in 200 6 to join the Ph.D. program at the University of Florida. In the ne ar future, Paulo hopes for a career in which he can continue to explore the vulnerability of tropical forests to repeated disturbances and prolonged degradation. He intends to continue to inform the public and policy makers about potentially dangerous fu ture trends of Amazonian forests in response to large scale deforestatio n forest fires, and climate change. Perhaps more importantly, Paulo Brando aims to develop an interdisciplinary career that will allow him to develop workable approaches to large sca le conservation.