Modeling Neotropical Ecosystems During the Last Glacial Maximum by Hiromitsu Sato a Thesis Submitted in Conformity with the Requ

Modeling Neotropical Ecosystems during the Last Glacial Maximum

Hiromitsu Sato

A thesis submitted in conformity with the requirements for the degree of Doctor of Philosophy Graduate Department of Earth Sciences University of Toronto

Modeling Neotropical Ecosystems during the Last Glacial Maximum

Hiromitsu Sato Doctor of Philosophy Graduate Department of Earth Sciences University of Toronto 2019

Rich biodiversity and biogeochemical significance has long motivated research on Neotropical ecosystems, though regionally-focused modeling studies remain relatively sparse and limited in scope. Three individual projects were performed to address potential issues in the modeling of Neotropical ecosystems in past and present contexts with particular emphasis on the Last Glacial Maximum. The first project was focused on discerning the effects of low CO2 and temperature on the carbon exchange of a well- watered tropical forest using a canopy-scale ecophysiological model. The radiative transfer regime and canopy energy balance were used to interpret the effect of environmental variables on carbon fluxes, as well as the reproduction of an observed phenomena where leaf temperature drops below air temperature.

The second project was an analysis of a four-year data set of ecosystem ﬂuxes designed to assess the seasonal patterns of carbon uptake and the impacts of drought from a tropical dry forest site in Santa

Rosa National Park, Costa Rica. A hyperbolic light response function was used to partition net ecosystem exchange into gross primary productivty and ecosystem respiration, while extracting estimates of ecosystem-level photosynthetic radiation and saturated rates of uptake. Bursts of carbon dioxide were observed at the onset of the rainy season suggesting the occurrence of the ‘Birch Eﬀect’ within tropical dry forest ecosystems. The third project was a regional-scale modeling study of vegetation cover in the

Neotropics during the Last Glacial Maximum, focusing on the individual and interactive effects of fire and low CO2 on biome distribution and tree cover. Inclusion of fire and the effects of low CO2 improved agreement with pollen records and suggest the past prevalence of grassier, more open ecosystems. Mod- eling evidence was found to support the existence of hypothesized routes and barriers to dispersal during the Last Glacial Maximum, bolstering theories of range expansion and diversification over Pleistocene climatic oscillations.

ii Acknowledgements

First off, I would like to thank my supervisor Sharon Cowling for the freedom and support that helped me finally perform and enjoy original research. My time in the Cowling lab taught me how much boldness and personal interest could fertilize and sustain good research. Thanks to my labmates who shared this funky journey through deep time. I would like to thank my friends for making life fun and lovely even when I was very much preoccupied with compilation errors and how lizards speciate. Thank you to my frequent study buddies, Nem and Mario, who shared the coffees, quick meals and mundane grind. Also, I’d like to thank my committee, Charly, Sean, and Jörg,for the nudges and suggestions that sometimes made a world of difference. Oh yes, also thank you to the many profs and post-docs and experts who responded to my impromptu emails kindly and constructively. I’d also like to thank my examination committee for taking the time to read and edit my thesis, and helping me understand my results in a larger context. I’d like to thank Douglas Kelley and Stephen Mayor for their guidance and technical support, which was a godsend that made me realize how important collaboration can be. I’d like to thank my fiancéJackie for the love and confidence to follow my heart. Thank you to my family, for the love and encouragement and belief in me, even when the chips were down. Its been a long road baby, but we’ve only just begun.

iii Contents

0.1 List of Tables ...... vi 0.2 List of Figures ...... vi 0.3 List of Symbols and Abbreviations ...... vii

1 Introduction 1 1.1 Overview ...... 1 1.2 Earth Systems Models and Palaeoecology ...... 2 1.3 Motivation and Research Goals ...... 3

2 Glacial Amazonia at the Canopy-Scale 5 2.1 Abstract ...... 5 2.2 Introduction ...... 5 2.3 Methods ...... 6 2.3.1 Canopy Model ...... 6 2.3.2 Climate Data and Ecosystem Parameters ...... 8 2.4 Results and Discussion ...... 9 2.4.1 Validation and LGM Carbon Processes ...... 9 2.4.2 Mechanisms behind Carbon Uptake Enhancement ...... 14 2.4.3 Eﬀects on Glacial Amazonia ...... 16 2.4.4 Adaptation and Dry Forest ...... 17 2.4.5 Implications for Future Forests ...... 18 2.5 Conclusion ...... 18 2.6 Acknowledgments ...... 19

3 Interpreting Carbon Fluxes from a Costa Rican Tropical Dry Forest 20 3.1 Abstract ...... 20 3.2 Introduction ...... 20 3.3 Methods ...... 22 3.3.1 Study Site ...... 22 3.3.2 Eddy Covariance Measurements ...... 23 3.3.3 Flux Partitioning and Empirical Parameter Estimation ...... 24 3.4 Results and Discussion ...... 26 3.4.1 Phenology and the Impact of Drought from Ecosystem Carbon Fluxes ...... 26 3.4.2 Comparison with Tropical Rainforest ...... 26 3.4.3 The Birch Eﬀect in Tropical Dry Forest ...... 29

iv 3.4.4 Connections with Model Representations of Tropical Dry Forest ...... 30 3.5 Conclusion ...... 31

4 Amazonian Dry Corridors opened by Fire and Low CO2 33 4.1 Abstract ...... 33 4.2 Introduction ...... 33 4.3 Results ...... 35 4.3.1 Comparison of Model Reconstructions with Palynological Data ...... 35

4.3.2 Fire and Low CO2 Activation Drives Expansions of Grasslands and Reductions of Forest ...... 37

4.3.3 Fire and Low CO2 Open Dry Corridors ...... 38

4.3.4 Interactive Eﬀects of Fire and low CO2 on Tree Cover ...... 41

4.3.5 Low CO2 Intensiﬁes the Fire-Forced Bimodality of Tree Cover ...... 43 4.4 Discussion ...... 44 4.5 Methods ...... 46 4.5.1 Model Description and Protocol ...... 46 4.5.2 Model-Pollen Biome Correspondence ...... 47 4.5.3 Discrete Manhattan Metric ...... 49 4.5.4 Stein-Alpert Decomposition ...... 50 4.6 Acknowledgments ...... 53

5 Conclusion 54

v 0.1 List of Tables

• 4.1 Aﬃnity matrix for LPX biomes to compute ‘distance’ between biomes in trait space (p. 48)

• 4.2 Correspondance legend between pollen reconstructed and model assigned biomes (p. 49)

0.2 List of Figures

• 2.1 Flow of submodules used by CANOAK to compute carbon, energy, and microclimatic proﬁles (p. 6)

• 2.2 a) Daily averages of Net Ecosystem Exchange at hourly intervals computed by CANOAK b) Theoretical computations of carbon ﬂuxes (p. 9)

• 2.3 Regression test between model output driven by modern values and measured values of Net Ecosystem Exchange with the equation for the line of best ﬁt and regression coeﬃcient (p. 10)

• 2.4 The dependence and sensitivity (slope) of average hourly rates of canopy photosynthesis (carbon uptake) and respiration (carbon release) to air temperature (p. 11)

• 2.5 a) The dependence and sensitivity of sensible and latent energy uxes to air temperature. b) The dependence and sensitivity of sunlit and mean leaf temperature to air temperature (p. 12)

• 3.1 Eddy covariance system mounted on a tower in Santa Rosa National Park, Guanacaste, Costa Rica (p. 21)

• 3.2 Tropical dry forest in study site during the dry season (February, 2016) (p. 22)

• 3.3 Sample of ﬁts of hyperbolic light response function to NEE and PAR data (p. 23)

• 3.4 Time series for NEE, GPP, and Reco for the four year study period (p. 25)

• 3.5 Time series for photosynthetic eﬃciency and saturated uptake for the four year study period,with blue shaded regions representing the wet season and red shaded regions representing the dry season (p. 26)

• 3.6 a) Net ecosystem exchange and soil moisture measurements at half-hourly intervals spanning from April 26th, 2016 (0:00) to April 28th, 2016 (02:00) b) Precipitation measurements for the same time period (p. 28)

• 4.1 Summary of model-data comparison scenarios from the ﬁre and low CO2 factorial experiment for each LGM climate reconstruction (p. 35)

• 4.2 LPX model reconstruction of biome distributions in the Neotropics during the LGM for four scenarios (p. 38)

• 4.3 Whittaker plots showing occurrence of biomes in climate space for four scenarios designed by

the ﬁre/CO2 factorial experiment (p. 39)

• 4.4 Identiﬁcation of reconstructed biogeographical formations in the ensemble ﬁre and low CO2 scenario (p. 40)

vi • 4.5 Results from the Stein-Alpert Decomposition showing individual and interactive eﬀects of ﬁre

and low CO2 in terms of fractional changes in tree cover (p. 41)

• 4.6 Fractional tree cover against mean annual precipitation aggregated by runs driven by four AOGCM LGM reconstructions (p. 42)

• 4.7 Flow of model protocol from spin-up to biome assignment for each factorial experiment run (p. 45)

• 4.8 Diagram representing the biome assignment scheme (p. 47)

0.3 List of Symbols and Abbreviations

• LGM: Last Glacial Maximum

• GPP: Gross Primary Productivity

• Reco: Ecosystem Respiration

• NEE: Net Ecosystem Exchange

• α: Photosynthetic Eﬃciency

• A2000: Light-Saturated Rate of Uptake

• gs: Stomatal Conductance

• A: Rate of photosynthetic uptake

• Vc: Rate of carboxylation

• Vo: Rate of oxygenation

• Rd: Rate of dark respiration

• Wc: Rate of carboxylation when RuBP is saturated

• Wj: Rate of carboxylation when limited by electron transport

• PAR: Photosynthetically Active Radiation

• SRNP: Santa Rosa National Park

• AOGCM: Atmosphere-Ocean General Circulation Model

• DGVM: Dynamic Global Vegetation Model

• LPX: Land Processes and eXchanges model

vii Chapter 1

Introduction

1.1 Overview

The objective of my thesis was to perform the most advanced model reconstruction analysis of vegetation of the Neotropics during the Last Glacial Maximum (LGM), which I duly accomplished. Achieving this goal required a good amount of modeling skill as well as familiarity with tropical ecosystems, which were formidable tasks given my recent entry into this field of research. To gain modeling experience, I first experimented with a simpler ecosystem-scale model that simulated tropical canopies and explored the processes of ecophysiology, the study of a plant’s function in relation to its environment. To become more familiar with tropical ecosystems beyond the modeling sphere, I joined a collaboration that was performing continuous measurements of ecosystem-level carbon fluxes from a tropical dry forest site in Costa Rica. This gave me a more tangible understanding of empirical measurements of ecosystems, how data is analyzed, and translated into modeling. It also allowed me to experience a tropical forest directly during a field work trip in 2017. After completing these two projects, I had the basic skills and network of collaborators necessary to undertake the larger task of hindcasting an entire region’s past vegetation.

The first project I completed was focused on modeling the effects of temperature and CO2 on forest productivity in the context of Amazonia during the LGM, examining the mechanisms underlying forest robustness and the effects of low CO2 on carbon uptake. The primary tool of this study was a canopy- scale model (CANOAK) that simulated light-transport, energy balance, and carbon exchange [13]. The second project I completed was the analysis of a novel data set of carbon exchange from a tropical dry forest site in Costa Rica, where the seasonal patterns of carbon uptake and impacts of drought were examined from flux data derived from an eddy covariance tower. The final project I undertook was the reconstruction of vegetation cover in the Neotropics during the LGM, which tested the effects of low

CO2 and fire on forest distributions and the biogeographical implications. These three projects compose the subsequent three chapters of my thesis. In terms of linkages between these relatively independent projects: the first chapter used a small- scale canopy model to test how well a Farquhar-based model of carbon uptake could be applied toward a tropical rainforest and as a diagnostic tool to determine what factor likely caused losses in carbon uptake in glacial scenarios. The second chapter used empirical methods to examine the actual phenology of carbon uptake of a tropical dry forest to better estimate an important photosynthetic parameter (maximum rate of uptake), provide data that would be able to refine dry forest representation, as well as

1 Chapter 1. Introduction 2 aide in model selection for my final project. The third chapter is the actual reconstruction of vegetation in the Neotropics on a large-scale in a palaeoenvironmental context. The content of chapter 2 was published in Quaternary Science Reviews [205], sections of chapter 3 were published in Environmental Research Letters [52], and the content of chapter 4 has been prepared as a manuscript currently under review for Nature Geoscience. I was the first author and primary researcher for the manuscripts submitted to Quanternary Science Reviews and Nature Geoscience.I was third author for the manuscript submitted to Environmental Research Letters, where my role was geared towards analysis and interpretation of data as well as writing of the manuscript, which is what composes the contents of chapter 3. My collaborator, Saulo Castro (University of Alberta) performed installation of the eddy covariance tower and meteorological monitoring station, and processesing of raw flux data that was used for the study. Given that each chapter has an introduction that includes a literature review and background information specific to the project, such information will be omitted here in exchange for broader perspectives.

1.2 Earth Systems Models and Palaeoecology

Earth Systems models are simplified representations of complex physical systems, which can enhance understanding of system processes and provide a degree of predictive power. They serve as integrations of our understanding of the planet, using theory from a broad range of scientific fields and data taken using a diverse set of methods. Quantitative projections of future climate and reconstructions of past climate are almost exclusively generated by General Circulation Models (GCM), which are expressed as computer code that simulate atmospheric and oceanic processes [214]. In essence, the GCM simulates a liquid (the ocean) covered by a gas (the atmosphere) on a rotating sphere covered by continents and ice sheets that is energetically driven by radiation (the sun), represented by the Navier-Stokes equation [208][115]. Since there are no known analytic solutions to the Navier-Stokes equation, more laborious numerical methods must be employed to obtain approximations of climate scenarios [182]. Since humans alone do not have the computational power to run these models in a timely fashion, computers were employed to process these equations. Thus, modern Earth Systems models have intrinsically been tied to the computer since their inception. Earth Systems models perform important functions in palaeoecology, providing a theoretical complement to the empirical inferences derived from proxies [184]. Palaeorecords may suggest what climate and ecosystems existed at a point in time and space, but models provide means to discern relationships between environmental and biotic observables. Process-based models are commonly parametrized based on modern field and lab studies, allowing utilization of a body of ecological and ecophysiological theory to understand palaeoenvironments. Palaeoecological model-data comparison also provides one of the few means by which we can assess model performance over long timescales and our ability to estimate the impact of climatic changes on the terrestrial biosphere [3]. The usage of Earth Systems models in palaeoecology may play a fundamental role in moving toward a unified and consistent understanding of Earth Systems phenomena in modern and future contexts. The work in this thesis is a strong example of this. I provide not only robust palaeoenvironmental reconstructions, but also direct connections with modern ecology and theories on the potential impacts of future climate change. A number of shortcomings stymie the progress of vegetation modeling. Large-scale models often require a large number of parameters to function, which may be sourced from outdated or unsuitable Chapter 1. Introduction 3 studies. Global studies often have coarse charaterizations of ecosystems, which receive uneven attention during construction [229]. Hydrological factors such as ground water processes are often simplified or neglected, with greater emphasis put on atmospheric controls of vegetation [89]. While recent methods have been developed to better benchmark models against modern observation, there is little quantitative comparison of palaeovegetation reconstructions with terrestrial proxy data. Proxy data is often limited in both spatial and temporal resolution, giving relative inferences on past conditions rather than absolute measurement. This makes direct comparison with model output a complex task that is rarely undertaken. In 2001, Cowling et al.[63] presented the first regional model reconstructions of Amazonia during the LGM using the BIOME3 model driven by an LGM climate reconstruction generated by the National Center for Atmospheric Research (NCAR) GCM [177]. A key finding was the apparent robustness of Amazonia during the LGM, showing moist forest to be resilient to the glacial environmental conditions. This was consistent with recently published palynological studies of the inner basin but contrary to the influential ‘Refugia Hypothesis’, a theory of widespread savannafication of Amazonia during the LGM [92]. Refutation of the Refugia Hypothesis had implications beyond palaeoecology, as Haffer [92] posited savannafication of Amazonia to be a mechanism for diversification, particularly for Neotropical birds. Cowling et al. however suggested that while forests remained largely intact, heterogeneity in canopy density may have been sufficient to induce the necessary vicariance required for allopatric speciation [63]. This was a rare utilization of vegetation models in palaeoecological contexts outside of biogeochemistry, where the evolutionary history of the region was investigated using modeling techniques. Subsequent model reconstructions of the Neotropics were often part of global studies [57][176][111], focusing mostly on carbon and energy fluxes. Like Cowling et al.’s study, Amazonian forest was found to be generally resilient during the LGM, roughly resembling Pre-Industrial distributions. Unlike Cowling et al.’s study, the evolutionary history and biological implications of past land cover were generally not investigated in these studies, which were rarely referenced by the communities of palaeoecologists and evolutionary biologists. While statistical models have been recently employed to reconstruct past vegetation [51][7], there has been no regional-scale process-based reconstructions of the LGM Neotropics in over a decade [20]. Since then, key enhancements have been added to vegetation models, including inclusion of disturbance ranging from wildfire to herbivory [129][242]. A number of relevant palynological studies have also been published that challenge the past conclusion of a robust glacial Amazonia [68][86], suggesting the value of a comprehensive modeling study of LGM Neotropics with comparison against a meta-analysis of pollen studies.

1.3 Motivation and Research Goals

The motivating concept that guided my thesis work was the usage of vegetation models to solve evolutionary biological questions, inspired by Cowling’s applications to South American and African palaeoecological contexts [63][65]. The start of my research coincided with the establishment of a large NSF-funded collaboration of biologists, geologists, and modelers (Dimensions of Amazonian Biodiversity), who aimed to integrate their respective methods to answer a common question: ‘why is Amazonia so rich in biodiversity?’. Molecular genetic analysis of a species may give a rough estimate of diversiﬁcation dates, which can then be linked to events in Earth’s history through palaeoecological data, but it is Earth Systems models that can provide high-resolution and mechanistic analysis. Unfortunately, modeling research is rarely presented in a form accessible to non-specialists. A natural pre-requisite to modeling Amazonia Chapter 1. Introduction 4 during the LGM is knowledge of how to model tropical ecosystems in general, which became a second motivation of my work.

The questions that drove my research were:

• What caused decreases in carbon uptake in the Neotropics during the LGM?

• What were the large-scale ecological consequences of decreases in carbon uptake?

• How did these large-scale consequence inﬂuence macroecological events such as diversiﬁcation?

Much of the efforts of my thesis were also guided by the desire to generate stronger reconstructions of the Neotropics, which required identification of areas in need of improvement. I determined that ecophysiology and process-based analysis could have a larger role in palaeoecology, which would allow stronger connections with modern and future ecological contexts. In terms of ecosystem representation, tropical dry forest seemed to be neglected in model reconstructions of the Neotropics, despite its hypothesized palaeoecological importance [185]. Most surprisingly, I found very little comparison of model reconstructions of the Neotropics to terrestrial proxy data, as model-data agreement of past studies tended to be brief, qualitative, or even absent. I felt this to be a major issue within palaeoecology that could have ramifications beyond the subject and something that should be rectified. Some of these issues were amendable within the duration of my thesis. For example, I successfully developed a robust relationship between CO2, fire, and forest cover, which will likely shed light on environmental contexts outside of the LGM Neotropics. Other issues proved to be outside the scope of my current resources, such as developing a mechanical, moisture-driven representation of tropical dry forest. However, I firmly believe that the sum of my research offers sufficient theoretical, methodological, and conceptual contributions to improve the modeling of Neotropical ecosystems. Chapter 2

Glacial Amazonia at the Canopy-Scale

2.1 Abstract

A canopy-scale model (CANOAK) was used to simulate lowland Amazonia during the Last Glacial Maximum. Modeled values of Net Ecosystem Exchange driven by glacial environmental conditions were roughly half the magnitude of modern ﬂuxes. Factorial experiments reveal lowered CO2 to be the primary cause of reduced carbon ﬂuxes while lowered air temperatures enhance net carbon uptake. LGM temperatures are suggested to be closer to optimal for carbon uptake than modern temperatures, ex- plained through the canopy energy balance. Further analysis of the canopy energy balance and resultant leaf temperature regime provide viable mechanisms to explain enhanced carbon-water relations at lowered temperatures and forest robustness over glaciations. An ecophysiological phenomena known as the ‘cross-over’ point, wherein leaf temperatures sink below air temperature, was reproduced and found to demarcate critical changes in energy balance partitioning.

2.2 Introduction

A sizable body of palynological and palaeovegetative modeling studies provide a justifiable estimate of lowland Amazonian palaeoecology over glacial-interglacial cycles. Pollen records have determined Pleistocene floristic composition and corresponding palaeoclimates, providing grounds to reconstruct past biome distributions and characteristics [44][43][91][158][178]. Unfortunately, complete pollen records in the Amazonian lowlands that date back to the Last Glacial Maximum (LGM) are scarce and large-scale inferences must be made from a limited data set [161]. To compliment these empirical efforts, modelers have applied several process-based, regional scale models to answer similar questions while searching for underlying ecological mechanisms and feedbacks [154][155][20][63]. Synthesis of these two independent approaches has been fruitful but we are still far from a complete understanding of past climatic changes and ecological responses in Amazonia. The current consensus is that Amazonian forests retained a closed-canopy over glaciations, refuting Haffer’s Refugia Hypothesis that proposed dramatic biome fragmentation and the formation of savanna- enclosed patches of moist forest [92]. This theory is consistent with both pollen records from lowland

5 Chapter 2. Glacial Amazonia at the Canopy-Scale 6 evergreen forest and the Amazon fan, as well as regional modeling studies [46][60][63][154]. The simulated maintenance of forest cover was proposed to originate in improved carbon-water relations correlated with cooler air temperatures, allowing continued dominance of forests against encroaching grasslands. There is, however, evidence of expanded savannah in marginal regions of Amazonia despite general forest robustness, suggesting limits to the effects of enhanced carbon-water relations [43][158]. The biomass of glacial forests has also been suggested to have been significantly lower (∼ 50%) than that of pre-Industrial Amazonia [20]. Cowling also found glacial forest canopy density to be lower and more heterogeneous than that of modern lowland forests, attributing these differences to low atmospheric carbon dioxide rather than temperature or aridity. Conversely, at more extreme values, lower air temperature can also diminish forest cover as deduced by studies on Andean sites of much higher elevation [172]. A motivation for our study is to develop a more comprehensive and quantitative foundation behind these processes. After determining what happened to Amazonian forests during climatic change, it is our intention to address how and why it happened. To do this, we investigate the specific ecological processes that can account for the larger scale changes that comprise our current picture of Pleistocene Amazonia over glacial-interglacial cycles. Our tool of choice is a canopy-scale model (CANOAK), built to capture and quantify finer scale phenomena outside the scope of previous studies. This approach puts a greater emphasis on ecophysiology, the study of an ecosystem and its components’ physiological interactions with the environment, which connects closely with modern forestry and carbon modeling. With this method, we can also explore the broader implications of these ecosystem processes, connecting to topics such as biome stability, drivers of biodiversity, and the projection into the future of Amazonia.

2.3 Methods

2.3.1 Canopy Model

CANOAK is a canopy-scale biophysical model developed by Dennis Baldocchi that computes carbon, water, and energy exchange (fluxes) between the biosphere and atmosphere, integrating concepts from micrometeorology, biochemistry, and ecophysiology. It has been grounded and applied to a number of field sites against measured fluxes (eddy covariance systems) over a range of timescales, predominantly in temperate regions [9][10][11][12]. Rigorous validation of CANOAK against flux measurements suggests that modelers are correctly parameterizing ecosystem processes while indicating phenomena unaccounted for by the model. While plant physiology has evolved significantly, it is likely that the processes simulated are valid in palaeoecological settings or conversely, can suggest specific evolutionary adaptations that may have occurred to confound our assumptions.

The model is driven by meteorological data in hourly steps (air temperature, CO2, incoming radiation, etc.), simulates radiative transport through the canopy, to then compute the energy balance through multiple layers of leaf and soil to estimate fluxes of sensible and latent heat [183]. This also determines proportions of ‘sunlit’ and ‘shaded’ leaves and their respective leaf temperature profiles. A key ecosystem parametrization used in this study that is neglected in regional studies is the encoding of vertical resolution within the canopy, which has been shown to significantly improve the fidelity of computed fluxes [213]. A simplified flow of processes used by CANOAK is shown in fig. 2.1. The Farquhar model for photosynthesis (eq. 2.1-2.2) [79] was used in combination with the Ball- Chapter 2. Glacial Amazonia at the Canopy-Scale 7

Figure 2.1: Flow of submodules used by CANOAK to compute carbon, energy, and microclimatic proﬁles, adapted from [12].

Berry-Collatz model for stomatal regulation (eq. 2.3) [61]:

A = Vc − 0.5Vo − Rd, (2.1)

Vc − 0.5Vo = min(Wc,Wj)(1 − Γ/Ci), (2.2) where A is the rate of photosynthesis, Vc is the rate of carboxylation, Vo is the rate of oxygenation, and Rd is the rate of dark respiration. In eq. 2.2, min(Wc,Wj) is the minimum between Wc, the rate of carboxylation when Ribulose Biphosphate (RuBP) is saturated and Wj, the rate of carboxylation when limited by electron transport (low light conditions). The compensation point (Γ) is the CO2 mole fraction where carbon uptake equals carbon loss, and Ci is the intercellular CO2 mole fraction.

Stomatal conductance (gs) can be expressed as a linear function of photosynthesis (A) through eq.

2.3, for a given relative humidity (rh). The parameters m and g0 are the respective slope and intercept that are ﬁtted against leaf-level gas exchange experiments [61].

gs = mArh/Cs + g0 (2.3)

The system of equations coupling photosynthesis and stomatal conductance was embedded in a cubic equation and solved analytically within CANOAK [8]. Photosynthesis, respiration, and transpiration are calculated for each leaf layer and then summed for net ecosystem exchanges [8]. This computational flow is iterated for each time step until stable values of microclimate, carbon and energy fluxes are reached. Soil and bole respiration are treated using empirical Chapter 2. Glacial Amazonia at the Canopy-Scale 8 functions while intra-canopy mixing is driven by a turbulent transfer submodule. Soil respiration was set to a constant rate of 6 µmol/m2s based on recent studies of similar sites [149][1]. Comprehensive descriptions of the model and its underlying theory can be found in [11][13][170]. A central concept used in this study is the canopy energy balance [183]. The energy balance of an ecosystem describes the flows of incoming, outgoing, and stored energy, serving as the interface between the environment and the surface, which in this case is a forest canopy. Solar radiation drives this system and is either reflected, transmitted, or absorbed, depending on the canopy’s spectral properties. Whatever is absorbed is then partitioned into latent heat (evapotranspiration), sensible heat (advection), or longwave emission (blackbody radiation), dissipating energy that is otherwise stored primarily by raising surface temperatures. A small amount of solar radiation fuels biochemical reactions that store energy as carbohydrates, used to fuel physiological processes and effectively feed the entire ecosystem [49]. The integration of these physical and biological processes (ex. radiative transport, energy balance partitioning, photosynthesis) add depth and realism to the simulated canopy, which we show can be the root of certain glacial forest feedback mechanisms. A deeper understanding of these mechanisms allows us to better gauge potential consequences of climatic change as well as get more insight into potential biological and evolutionary implications of changing environmental pressures.

2.3.2 Climate Data and Ecosystem Parameters

Modern climate data and eddy covariance measurements of carbon, water and energy were collected from the Large Scale Biosphere-Atmosphere Experiment in Amazonia (LBA) [117][118]. One month of hourly aggregated data for February 2003 was taken from continuous measurements of climatic variables and fluxes from a meteorological tower in Tapajos forest in North Central Brazil. Measurements from this data set present a mean air temperature of 24.3 ± 1.9◦C, mean vapor pressure of 2.8 ± 0.7 kPa, a monthly cumulative precipitation of 225 mm, and CO2 concentration of 388 ± 12 ppm, reflecting hot and humid modern climate as well as a recent local measurement of atmospheric CO2. Estimates for diffuse photosynthetically active radiation (PAR) were computed through an algorithm by Conghe Song of University of North Carolina at Chapel Hill, based on work by Weiss & Norman and Spitters [217][234]. Leaf area index (LAI) was similarly taken from the LBA study while an estimate for carboxylation 2 capacity (Vc,max = 29 µmol/m s), which dictates photosynthetic rates, was taken from Kattge [127] based on the classification of tropical vegetation on oxisol soils. LAI is a dimensionless parameter defined as leaf area (single-sided) per unit ground area (m2/m2) and is a common measure for canopy-density and active biomass, and therefore integral in flux computations. Modifications made for LGM settings were largely based on estimates from Mayle et. al (2009) [161], which gave an overall assessment of likely vegetative changes during the LGM based on an assimilation of pollen and charcoal data with LGM modeling studies. Temperature was uniformly decreased by 5◦C

(-20%) relative to modern data. Atmospheric CO2 was set to 200 ppm. These values fall within the mid- range of Cowling’s extreme and conservative estimates of LGM conditions [63]. Though precipitation was estimated to be 50% lower, modiﬁcations were not made in this study as it does not impact ﬂux calculations in CANOAK. It is noteworthy that seasonal precipitation patterns, particularly the severity and length of the dry season have strong impact on the phenology and carbon cycle of tropical forests. Chapter 2. Glacial Amazonia at the Canopy-Scale 9

2.4 Results and Discussion

2.4.1 Validation and LGM Carbon Processes

To ground the model, we applied CANOAK to carbon fluxes in modern Amazonia, as studies on tropical ecosystem-level fluxes are limited relative to those on temperate ecosystems [70][210][165]. The primary model output being validated is net ecosystem exchange (NEE), which is defined as the gross primary productivity minus ecosystem respiration. Eddy covariance towers are equipped precisely to measure NEE at the ecosystem level. Our results will focus on the responses of the competing processes of respiration and photosynthesis to environmental conditions. Model predictions of of NEE tend to agree with eddy covariances measurements, estimating mean diurnal patterns within an assumed error of 20% of flux measurements (fig. 2.2a) [169] [11]. Model- measurement regression tests between computed and measured fluxes suggest that CANOAK accounts for approximately 86% of the variance in carbon fluxes (fig.2.3), exceeding the community standard of model-measurement agreement (r2 > 0.80) [12]. A significant portion of disagreement stems from nighttime fluxes, when there is no light to drive radiation and the ecosystem simply respires (upper right region of fig. 2.3). Though nighttime respiration will play a role in our results, emphasis will be put on modeling the photosynthetic responses that occur during daytime. To explore potential carbon fluxes during glacial periods, the model was then driven by data modified to reflect the colder, lower CO2 of LGM scenarios. CANOAK estimated a somewhat expected result; less net carbon is exchanged under glacial conditions. The average daytime rate of NEE driven by glacial settings is 56% of computed modern values (52% of measured values). Peak rates of glacially driven NEE are approximately 60% of modern values. A natural conclusion upon first inspection may be that modern tropical climates are closer to ideal conditions for plant growth and that downscaling biologically relevant climatic variables leads to a proportional reduction of net carbon uptake. We performed a factorial experiment with respect to the temperature and CO2 to investigate their impacts in isolation and in concert.

The effects of reduced CO2 and cooler temperatures act in opposition to one another with respect to carbon fluxes at the ecosystem level, according to our results (fig. 2.2b). Reducing the concentration of carbon dioxide from 388 ppm to 200 ppm reduces average daytime NEE to 40% of their modern values and average peak rates to 50% relative to modern rates. This implies that net rates of carbon uptake are indeed sensitive to CO2 within the tested range of values, which is unsurprising given the body of work from laboratory studies [200][148]. Reducing average air temperature from 24.3◦C (modern) to 19.3◦C (LGM scenario) resulted in an 8% increase in average daytime NEE. Hence our model results suggest that cooler LGM temperatures slightly enhance carbon sequestration. To find the origin of this effect, we used CANOAK to test the responses of canopy photosynthesis and canopy respiration to air temperature individually. A larger range of air temperatures were used to study the broader sensitivity of carbon processes to temperature. Computations of canopy respiration rates increased with average air temperature. Respiration typically doubles for a 10◦C increase in temperature or is scaled through an Arrhenius function (eq. 2.4) used to model the dependence of reaction rates with temperature [9][152]:

f(Tk) = f(298) · exp[(Tk − 298)Ha/RTk298], (2.4) where f(Tk) is the value of the temperature-dependent parameter at temperature Tk in Kelvin, R is the Chapter 2. Glacial Amazonia at the Canopy-Scale 10

(a)

(b)

Figure 2.2: a) Daily averages of Net Ecosystem Exchange at hourly intervals computed by CANOAK, driven by modern data and LGM-adjusted meteorological data, with reference to measured values from an eddy covariance system. b) Theoretical computations of carbon ﬂux driven by modern data with independently lowered (LGM) CO2 and temperature. Chapter 2. Glacial Amazonia at the Canopy-Scale 11

Figure 2.3: Regression test between model output driven by modern values and measured values of Net Ecosystem Exchange with the equation for the line of best ﬁt and regression coeﬃcient.

gas constant (0.00831 kJ/mol) and Ha is the activation energy (kJ/mol) for the parameter f. Compu- tations for photosynthesis produced a parabolic dependence on temperature, showing an increase with temperature until it peaks at approximately 19◦C. Beyond this, photosynthesis steadily decreases with higher air temperatures. The estimated relationship between air temperature and carbon uptake is largely a product of the Farquhar equations and the modeled temperature-dependence of photosynthetically relevant parameters by Arrhenius functions employed by CANOAK [209][9][103]. This pattern is consistent with other studies on the eﬀects of leaf temperature on photosynthesis, though optimal temperature and sensistivity varies with biomes, species, and even seasons [10][199].

The net effect of canopy photosynthesis and respiration as functions of temperature result again in a parabolic relationship where temperature positively correlates with net productivity, an inflection point or ‘plateau’ occurs, after which higher temperatures reduce net productivity (fig. 2.4). Our studies suggest that this inflection point in behaviour could potentially occur between glacial and modern tropical temperatures. While this may seem lower than typical values of peak productivity, it is important to note that air temperature only indirectly affects biochemical processes and it is leaf temperature that is more biologically relevant. It is the energy balance of the leaf that determines leaf temperature and helps elucidate the complex energetic and physiological processes of the leaf. Chapter 2. Glacial Amazonia at the Canopy-Scale 12

Figure 2.4: The dependence and sensitivity (slope) of average hourly rates of canopy photosynthesis (carbon uptake) and respiration (carbon release) to air temperature. Rates of canopy photosynthesis and respiration are averaged over the entire day, including nighttime values where rates of photosynthesis are zero. Note that we use the convention of photosynthesis to be a negative carbon ﬂux. Chapter 2. Glacial Amazonia at the Canopy-Scale 13

(a)

(b)

Figure 2.5: a) The dependence and sensitivity of sensible and latent energy ﬂuxes to air temperature. b) The dependence and sensitivity of sunlit and mean leaf temperature to air temperature. Note that the ‘cross-over’ point, when the mean leaf temperature sinks below air temperature, occurs near modern temperatures. Chapter 2. Glacial Amazonia at the Canopy-Scale 14

2.4.2 Mechanisms behind Carbon Uptake Enhancement

Canopy Energy Balance

The energy balance of the canopy is central in determining energy, water and carbon fluxes. To better understand the behaviour of photosynthesis and respiration rates in the air temperature range between glacial and modern scenarios, we investigated the underlying processes of latent and sensible heat exchange between the canopy and atmosphere. Sensible and latent heat daytime values were averaged for eight scenarios of modified temperatures, including the LGM and modern cases. Sensible heat tends to decrease with increasing average air temperature while latent heat tends to increase with equal but opposite magnitude and sensivitivity, obeying conservation of energy given a consistent proportion of longwave emission (fig. 2.5a). In the absence of latent and sensible heat fluxes (infinite resistances and zero conductances), CANOAK predicts leaf temperatures upward of 42◦C, which would be biologically destructive. We can then interpret latent and sensible heat fluxes as modes to which solar energy can be dissipated to mitigate plant temperature warming [151][166]. As radiation is absorbed by the canopy, leaf temperatures increase beyond the thermal equilibrium held with ambient air. This creates a thermal gradient between the air and leaves, inducing the flow of sensible heat exchange (H), which is proportional to the difference between leaf temperature (Tleaf ) and air temperature (Tair) such that,

Tleaf − Tair H = ρCp (2.5) rH where ρ is the air density, Cp is the heat capacity of air, and rH is the boundary layer resistance. Heat loss through long wave radiation also begins to increase dramatically due its quartic dependence on temperature. The remaining proportion of energy is released through latent heat to maintain energy balance. Three distinct ‘regions of sensitivity’ of latent and sensible heat to air temperature were observed (fig. 2.5a). The first region occurs in cooler temperatures (T < 18◦C), the second zone ranges between tropical glacial and modern temperatures (18◦C < T < 24◦C), and the third occurs in hot to very hot scenarios (T > 24◦C). Both latent and sensible heat show an increase in sensitivity as average air temperature increases, with all three zones in ranges relevant to modern, palaeo, and predictive contexts. As air temperature rises, sensible heat becomes increasingly suppressed, demanding an increasing amount of compensatory latent heat exchange. Latent heat (λE) is driven by the difference between saturation vapor pressure at leaf temperature (es(Tleaf )) and the saturation vapor pressure at air temperature (es(Tair)) such that,

ρC e (T ) − e (T ) λE = p s leaf s air (2.6) γ rw where γ is the psychometric constant and rw is the sum of boundary layer and stomatal resistances. Latent heat flux is thus expressed as the difference between two exponential functions. This function is non-linear, increasing within our range of values, and begins to dominate at higher ambient air temperatures, assuming a constant humidity. At these higher temperatures, stomata tend to close (decrease in stomatal conductance) to temper excessive evapotranspiration. This is an indirect negative effect on productivity [141], which is also dependent on stomatal conductance given that atmospheric carbon is drawn into the leaf through the same stomata used for evapotranspiration. The partitioning Chapter 2. Glacial Amazonia at the Canopy-Scale 15 of energy fluxes is encoded in the solution to the leaf energy balance equation [183] and is independent from tissue damage induced reductions in productivity that occurs at significantly higher temperatures. We propose that it is stomatal closure [141] caused by increased tropical temperatures (>19◦C) that causes net losses in canopy productivity within our model.

Sunlit and Shaded Leaf Temperatures

Leaf temperature, determined by the canopy energy balance, impacts fluxes in a number of ways more conspicuous when using a canopy-scale model such as CANOAK. The radiative transport module, which simulates light penetration through foliage, divides the canopy into ‘sunlit’ and ‘shaded’ leaves along with their respective temperature profiles. Incident solar radiation at each of the leaf layers is also computed, driving photosynthesis as well as energy balance and leaf temperature. Model results indicate that for a number of scenarios, mean leaf temperatures (average of sunlit and shaded leaves) are close to air temperature, while sunlit leaf temperatures tend to be 3–4◦C higher (fig 2.5b). Accurate estimation of leaf temperature is vital in flux models, as sunlit leaves are the chief source of photosynthesis, which is dependent on leaf temperature through scaling of biological parameters by Arrhenius functions. We propose that in glacial scenarios with cooler air temperatures, sunlit leaf temperatures and consequently carbon uptake, were bolstered closer to optimal levels through radiative heating. The daytime shaded fraction of leaves does not receive radiative heating and photosynthesizes at a lower rate relative to sunlit leaves, only receiving diffuse (non-direct) photosynthetically active radiation [170]. The entire nighttime canopy does not perform photosynthesis nor receive radiative heating. Both categories of shaded leaves equilibrate with the cooler air temperatures in glacial tropical settings and the accompanied lowered rates of respiration, buffering carbon losses. Analogous effects would also apply to other sources of carbon emission such as bole and soil respiration, which have shown strong temperature dependence. Another salient effect occurs when average monthly air temperature approaches modern values, where leaf temperature actually sinks below air temperature due to heavy rates of evapotranspiration. This phenomena has been observed and studied at diurnal scales in forests and crops largely in the context of plant thermoregulatory mechanisms [151][123][77]. Modeling results of this study interestingly reproduce ◦ this ‘crossover’ temperature point (fig. 2.5b) close to the range of observed values (Tair ≈ 24 C), predicting its presence at larger scales and offering insight into its causal mechanism. As air temperature rises, sensible heat exchange is reduced and latent heat exchange (evapotranspiration) increases in proportion, otherwise driving leaf temperatures to biologically deleterious levels. The ‘cross-over point’, where mean leaf temperature equals air temperature (fig. 2.5b), occurs near modern values of average air temperature. At higher air temperatures, sensible heat fluxes approach zero and become negative (fig. 2.5a), indicating a reversal of thermal gradient direction and heat flow. Thus, beyond the cross-over point, sensible heat begins to flow from the warmer air to the cooler leaves [4][123]. Sensible heat would then reverse its direction, now flowing from the hot air toward the cooler canopy. Between LGM and modern air temperatures, sensible heat and latent heat exchange intersect and become equal in magnitude (fig. 2.5a). The slopes of sensible and latent heat exchange increase in magnitude at modern air temperature values, indicating an increase in sensitivity of heat fluxes to air temperature, and corresponds to the air temperature where the crossover phenomena occurs. This effect Chapter 2. Glacial Amazonia at the Canopy-Scale 16 could be detected by eddy covariance or Bowen ratio systems and could serve as an ecophysiological alarm, as suggested by Dong et al.[77]. With rising air temperatures, the cross-over point signals an enhanced sensitivity of energy balance partitioning to temperature and increasingly heavy evapotranspirative demand.

2.4.3 Eﬀects on Glacial Amazonia

The dominant cause of productivity losses in tropical forests during glaciations is lowered CO2 according to our study. This has been suggested by global and regional scale modeling experiments [63][57][176] and has been estimated at the canopy-scale through our study by testing its eﬀects in isolation and interactively with temperature.

A marked decrease of NEE (∼50% relative to modern) associated with reductions in CO2, suggests high sensitivity of carbon uptake to substrate concentration over glacial-interglacial cycles. This estimated reduction in carbon fluxes adds to modeling results by Beerling and Mayle [20] that estimated glacial biomass to be approximately half of modern values. This result also aligns with Cowling’s regional modeling study of glacial Amazonia that suggested that canopy density, a likely proxy for the magnitude of carbon processes, showed considerable heterogeneity and lower LAI under lower LGM CO2 [63]. Thus, glacial Amazonia had a closed yet thinner and more variable canopy, with a significantly reduced carbon pool, and smaller carbon fluxes, relative to modern Amazonia. These effects could underlie glacial forests’ vulnerability to intruding grasslands in marginal areas and more subtle biome shifts to tropical dry forest throughout the basin [185][186]. The robustness of tropical forests over glacial cycles can be attributed to the effects of cooler air temperatures. The effects driven by air temperature are mediated by the canopy energy balance. Colder glacial air temperatures of tropical regions induce high rates of sensible heat exchange, reducing latent heat exchange and evapotranspirative losses considerably. This also reduces the demand for water from the roots, decreasing sensitivity to drought. Colder air temperatures simultaneously lead to reduced ecosystem respiration rates, while sunlit photosynthetically active leaf temperatures are bolstered by radiative heating. At higher interglacial air temperatures, sensible heat exchange decreases as latent heat increases. Stomata in turn respond by closing and consequently inhibit photosynthesis, resulting in a relatively low optimal ambient air temperature that maximizes canopy productivity. Air temperature, through its indirect effect on the stomatal conductance of the sunlit fraction of the canopy, limits photosynthesis in tropical forests which curbs uptake beyond approximately 20◦C. This value is somewhat lower than expected for tropical regions but is in agreement with studies in boreal system, which show photosynthesis to rise with temperature until an optimal value, followed by a distinct decrease at higher temperatures [71][72][73]. While optimal values may vary amongst biomes, the fundamental mechanism is consistent. In summary, colder air temperatures in isolation bring about a canopy with a slightly greater ability for carbon uptake while requiring less water and yield a higher water-use efficiency (WUE). This is a oft-used metric in agricultural and forestry that quantifies ecosystem carbon-water relations, defined as the ratio of units of water consumed to units of carbon sequestered. Given lower rates of precipitation and water availability, a higher WUE could be critical in maintaining carbon stores, thereby maintaining a closed canopy. Lower rates of precipitation during the LGM may have also resulted in less nutrient leaching and higher soil nutrition relative to modern soils, further aiding forest cover. These types of ecosystem level Chapter 2. Glacial Amazonia at the Canopy-Scale 17 feedbacks, more often of focus in ecophysiology and carbon modeling, likely have a collective effect on biome stability [137] with respect to environmental change, helping offset larger detrimental condition such as greatly decreased CO2. Thus, our results advance previous claims of improved plant carbon-water relations due to glacial cooling by establishing several underlying, ecophysiological mechanisms.

2.4.4 Adaptation and Dry Forest

Haffer’s Refugia theory and Cowling’s canopy-density hypothesis both suggest allopatric speciation as a mechanism for Amazonian biodiversity and explore the effects of respective biome and canopy-structure heterogeneity at regional scales [92][63][64]. Simulating the ecosystem at the canopy-scale can offer us sharper insight into evolutionary processes during the Pleistocene. Large scale climatic changes translate to more local environmental changes that are felt by biota. In the case of forests, these changes are largely mediated by the canopy, which feeds and houses its myriad of organisms. Modeling and phylogenetic studies have proposed the respective expansion and diversification of tropical dry forest over Quaternary climatic oscillations [20][197]. Tropical dry forest species tend to be a drought-tolerant subset of rain forest species. As the climate shifted to colder and drier conditions, these species could flourish and further adapt to aridity. These adaptations could include a deeper root system, defenses against herbivory, changes in stomatal density, an inverse phenology, or physiological changes to leaves in the canopy [203]. Based on our study, warmer air temperatures, as associated with interglacial periods, induce relatively high rates of evapotranspiration (latent heat exchange) and low rates of sensible heat exchange. These warmer air temperatures also correspond to lower rates of photosynthesis, with the net effect of less efficient water-carbon relations with respect to glacial tropical air temperatures. The thickness of the leaf boundary layer (aerodynamic resistance) is a function of leaf size and shape and ultimately moderates rates of sensible heat exchange. Assuming sunlit leaves are warmer than the ambient air, maximization of sensible heat exchange would buffer evapotranspirative losses. This form of thermal stress would select for thinner, smaller leaves, particularly in the hotter, heavily transpiring upper canopy. Conversely, during glacial periods, leaf morphology could potentially shift to optimize radiative heating effects to boost leaf temperatures closer to their optimal value. At the canopy-scale, empirical studies have argued that species diversity and more directly, functional diversity, has an inverse relationship with evaporation [15]. The species composition and structure of foliage in the canopy has also been proposed to optimize photosynthesis, favoring a diverse set of species [102][87][104]. This would lead to an increasingly robust and diverse canopy driven by glacial-interglacial cycles. Plants are capable of biochemical, physiological, and structural adjustments when exposed to ex- tended environmental change, which can significantly affect their ecophysiological relationships [215]. This process is called acclimation. Unlike adaptation, which occurs over multiple generations of taxa, acclimation can occur within an organism’s lifespan. The fertilization and downregulation of photosynthesis in response to elevated CO2 [136] and acclimation through variable stomatal sensitivity [140] may have a significant impact on carbon, water, and energy fluxes. Acclimation to temperature and CO2 were not included in this study and should be considered in forthcoming palaeoecological and future modeling experiments to better assess long-term ecosystem response to climatic change. Chapter 2. Glacial Amazonia at the Canopy-Scale 18

2.4.5 Implications for Future Forests

Our modeling results suggests that the air temperatures of modern tropical forests are higher than their optimal value for net carbon uptake. As sensible heat becomes increasingly suppressed by rising air temperatures, evapotranspiration must compensate to cool the canopy, requiring ample supply of water. The sensitivity of evapotranspiration to air temperature is highest in this environment, exceeding the aforementioned ‘cross-over’ point where sensible heat exchange becomes negative. This is accompanied by reduced carbon uptake due to stomatal closure that occurs to subdue excessive evapotranspirative losses. Fertilization due to increased atmospheric carbon dioxide may well-compensate for temperature-related losses, but the net response after acclimation is likely complex and worthy of further study [142]. Acclimation to rising air temperatures could also increase the optimal temperature for photosynthesis [215], potentially leading to more efficient plant carbon-water relations. In the case where there is insufficient water for cooling, the canopy may experience dessication and burning, leading to thinning and decreases in pool size. The severity of these effects would also be determined by changes in precipitation and would be naturally worsened by drought.

2.5 Conclusion

This study was able to successfully apply a new class of model to an important palaeoecological question: ‘How did Amazonia maintain forest cover through glaciations?’ Initial results are consistent with previous studies at larger scales, estimating carbon uptake to be lower under LGM conditions. Lower atmospheric carbon was found to be the sole cause of reduced ecosystem fluxes, while colder air temperature slightly enhances carbon uptake in the context of LGM Amazonia. This was suggested to be partially rooted in the ‘refrigeration’ of the shaded and night canopy by glacial cooling along with the solar heating of the sunlit, photosynthesizing portion of the daytime canopy. Furthermore, glacial air temperatures in the tropics were suggested to be close to ideal for photosynthesis, where higher temperatures result in stomatal closure to suppress water losses. Finally, analysis of the canopy energy balance through glacial-interglacial periods provides a probable mechanism for improved glacial carbon-water relations. An unexpected but fortuitous result was computation and development of the ‘cross-over’ point, where leaf temperatures sinks below air temperature. This effect may be significant for ecosystem thermoregulation and future response of tropical systems to temperature elevation and should be further investigated. To better characterize the increased aridity on the Amazonia lowland forests over glacial-interglacial cycles using a canopy-scale model, we recommend the inclusion of an integrated soil moisture and precipitation module. Another avenue of improvement would be the examination of vapor pressure changes due to precipitation and its effects on canopy processes. The significance of radiative transfer through the canopy has been assessed in vegetation models [114][146], but has yet to be applied in the context of palaeoecology. A Dynamic Global Vegetation Model (DGVM) with a vertically resolved canopy [56] could be used to assess the effects of impact of a thermally heterogeneous canopy on carbon fluxes in LGM Amazonia at the regional scale. CANOAK is the integration of a well-established body of theory, grounded against the strongest ecosystem flux measurements available. Using this model enables better usage of modern ecophysiology Chapter 2. Glacial Amazonia at the Canopy-Scale 19 and hopefully more communication between distinct scientific communities. While palaeo work benefits from records of both climatic change and ecological consequence, it is limited in terms of completeness and resolution. Modern experiments operate at a higher degree of experimental control, but are limited to relatively short timescales. We aim for further synthesis of these approaches to better comprehend the relationship between climates and ecosystems.

2.6 Acknowledgments

Special thanks to Dennis Baldocchi, Young-Lan Shin, Anna Phillips, Vasa Lukic, William Feng, Ting Zheng, Kinoko Sama and Wataru Yamori for helpful discussions and advice through this project. Thank you to my family and friends for love and support through this project. Funding: This research was supported by the Natural Sciences and Engineering Research Council of Canada (NSERC), Jeanne F. Goulding Fellowship, Centre for Global Change Science (CGCS) at the University of Toronto and a collaborative Dimensions of Biodiversity-BIOTA grant supported by FAPESP (2012/50260-6), NSF and NASA. Chapter 3

Interpreting Carbon Fluxes from a Costa Rican Tropical Dry Forest

3.1 Abstract

Tropical dry forest is a vast, biologically-rich ecosystem that is relatively understudied and threatened by land-use and climate change. Fluxes of carbon at the ecosystem-scale were measured continuously over a four year period from 2013-2016 using an eddy covariance system at a secondary tropical dry forest site in Santa Rosa National Park, Costa Rica. Two years within this period were characterized as El Niño-induced‘drought years’ (2014-2015), while the first and last years were deemed to be ‘normal years’ (2013, 2016). This allowed observation of both the typical phenology of tropical dry forest as expressed by ecosystem carbon exchange, as well as insight into the effects of drought on productivity. Fluxes of Net Ecosystem Exchange were separated into Gross Primary Productivity and ecosystem respiration through fitting flux data to photosynthetically active radiation, which was then used to estimate ecosystem-scale saturated uptake rates and photosynthetic efficiency. This allowed for extraction of a potentially valuable photosynthetic parameter used in global vegetation models. Patterns of fluxes were then compared to those of tropical rain forest, finding distinct differences in seasonal uptake patterns and maximal rates. A phenomena known as the Birch Effect was observed, where a large flux of carbon was released after the first rains following the 2016 dry season. The parametrization of respiration from tropical dry forest within vegetation models was briefly analyzed in light of observed patterns.

3.2 Introduction

Tropical dry forest currently occupies more than one million square kilometres globally, comprises roughly forty percent of the total amount of tropical forest, and has been described as the most threatened of all tropical forest types [167][173][119]. Tropical dry forest also experiences some of the highest rates of land- use related threats among tropical forests globally, including deforestation and fragmentation. Moreover, anthropogenic climate change has also been projected to have a signiﬁcant eﬀect on its distribution and function, particularly in the Americas, where exposure is highest [167]. Tropical dry forest currently and historically has played important roles in agricultural and ranching, including modern widespread subsistence farming in Africa and the establishment of South and Central American ‘Haciendas’ of the

20 Chapter 3. Interpreting Carbon Fluxes from a Costa Rican Tropical Dry Forest 21 nineteenth century [150][54]. Dry forest also occupies climate zones with the highest human population densities in the tropics, which has lead to sustained usage and exploitation [173]. Given its vast extent, socioeconomic importance, and role in biogeochemical cycles, tropical dry forest provides a strong case for conservation and management efforts [30]. Unfortunately, research on tropical dry forest is scarce, particularly in comparison to tropical moist and rain forest [202]. Tropical dry forest occurs in regions of warm to hot temperatures with precipitation that is distinctly seasonal, often occurring in transitional zones between grassier ecosystems such as savanna and moist forest [109]. The dry season is pronounced (prec. < 300mm/month) and typically lasts three or more months, with high interannual variability in terms of timing and intensity [202]. Moisture is a major control of phenological cycles, where growth, decay, and reproduction are synchronized with precipitation patterns. Relative to tropical moist forest, dry forest tend be shorter in stature, more open, and prone to clumping [173]. While less species rich compared to tropical moist forest, dry forest understo- ries tend to contain a wider array of shrubs and small trees [112]. Composition varies among sites but has been defined as forest where at least half of its trees are drought deciduous. While it is clear that water dynamics play a major role in dry forest ecology, the magnitude and patterns of seasonal fluxes in response to precipitation are still largely unknown. The development of eddy covariance measurement systems has led to long term, nearly continuous monitoring of net carbon, water and energy fluxes of ecosystems. Coupled with appropriate environmental measurements, these flux time-series allow for examination of the carbon cycle and biophysical phenology at the ecosystem scale, and identification of probable controls of fluxes. Our study presents a novel data set from one such eddy covariance system, located in the rarely studied tropical dry forest ecosystem, which we use to better understand its distinct moisture-driven ecology. Our four year data set derived from a Central American dry forest consists of two ‘normal’ years and two El Ninõdrought years, which allowed the respective examination of typical phenology and the impacts of drought on ecosystem fluxes. A secondary motivation for our project was to provide much needed data for Earth Systems models that rely on eddy covariance measurements for groundtruthing. Dynamic Global Vegetation Models (DGVMs), which simulate large-scale atmosphere-biosphere interactions, are a powerful tool in forecast- ing the impacts of climate change on terrestrial ecosystems. Projections of climate change generated by General Circulation Models have predicted changes in rainfall events including an increase in extreme events, shifts in seasonal sums, and changes in seasonality in the tropics [83]. Such projections feed directly in DGVMs, which can then project ecosystem response and vegetative feedbacks. However, ecosystems within these models are not parametrized with uniform attention, with a number of important ecosystems such as tropical dry forest having simplistic and expedient though potentially inaccurate representations. This may translate to significant error in estimating current and future fluxes and pools of carbon, water and energy at the regional and global scale. We first compare environmental triggers in parametrizations of dry forest in two makor DGVMs in light of observed patterns in our data set. Secondly, we assess for potential of parameter extraction for usage in these models. Chapter 3. Interpreting Carbon Fluxes from a Costa Rican Tropical Dry Forest 22

Figure 3.1: Eddy covariance system mounted on a tower in Santa Rosa National Park, Guanacaste, Costa Rica.

3.3 Methods

3.3.1 Study Site

The tropical dry forest under study is located in Santa Rosa National Park (SRNP), Guanacaste province, in Northwestern Costa Rica near the Nicaraguan border. SNRP was one of the first parks established in Costa Rica (∼1970), spans roughly 495 km2, and houses a diverse set of ecosystems including wetlands, forest, and savanna [38]. Though SRNP has experienced significant agricultural land use and was at a point largely composed of a mosaic of pasture with islands of tropical dry forest, restoration has allowed regeneration of larger areas of forest. The eddy covariance tower used for flux measurements is located at: 10 deg 44.206’ N, 85 deg 37.034’W at 290 masl, with prevailing winds coming from the northwest. The vegetation of the study site is a secondary tropical dry forest in intermediate stages of recovery with an average tree height of 13 metres with a significant presence of lianas [139]. SRNP experiences a tropical seasonal climate with a mean annual air temperature of 25 degrees Celsius and mean annual precipitation of 1575 mm. Precipitation is highly seasonal with significant interannual variability, with a wet season from May to November and a dry season from December to April [126]. Chapter 3. Interpreting Carbon Fluxes from a Costa Rican Tropical Dry Forest 23

Figure 3.2: Tropical dry forest in study site during the dry season (February, 2016).

3.3.2 Eddy Covariance Measurements

The eddy covariance technique is a micrometeorological method of measuring ecosystem fluxes of carbon, water, and energy at high temporal resolutions, utilizing the framework of turbulent transport theory. Air flow near the surface is assumed to be turbulent, composed of a large number of eddies: vortices of rotating air of varying size that transport mass and energy. High-frequency (>10 Hz) measurements of wind velocity, air temperature, pressure, humidity, and trace gas concentrations are then used to estimate the integrated eddy-driven vertical transport of air parcels and thus the average flux, typically averaged at half-hourly intervals [14]. Using a Reynold’s decomposition [84], ecosystem carbon flux (Fc) can be approximated as:

0 0 Fc ≈ ρaw s (3.1)

0 0 Where ρa is the density of air, w is the instantaneous variance in wind speed, s is the instantaneous variance in carbon dioxide concentration, and w0s0 is the ‘covariance’ between w0 and s0. An eddy covariance system was installed (S. Castro) on a 35 m tower in our tropical dry forest site in SRNP in early 2013 to measure carbon and energy ﬂuxes alongside measurements of environmental variables made by a nearby meteorological station. Measurements of photosynthetically active radiation (PAR), which were used in ﬂux partitioning calculations, were made by a radiometer also mounted Chapter 3. Interpreting Carbon Fluxes from a Costa Rican Tropical Dry Forest 24

Figure 3.3: Sample of ﬁt of hyperbolic light response function to NEE and PAR data for 2013-11-01 to 2013- 11-05. on the ﬂux tower. Detailed description of instrumentation and data-processing can be found in the supplementary information of [52].

3.3.3 Flux Partitioning and Empirical Parameter Estimation

To estimate gross primary productivity (GPP) and ecosystem respiration (Reco), Light Response Curves (LRC) were fitted against daytime Net Ecosystem Exchange (NEE) measurements [221] [175]. From low to medium levels of light, photosynthesis increases significantly, then saturates toward an asymptote at higher levels. Daytime NEE data was defined as PAR > 40 µmol/m2s, following the methods of [117]. Data was binned into 5-day intervals save for a small number of 4-day bins caused by gaps in data. Fig. 3.1 shows a sample LRC fit onto daytime flux data, which were performed using MATLAB and CurveExpert. Measurements of net ecosystem exchange (NEE) was partitioned into its two components as:

NEE = GPP + Reco (3.2)

2 where GPP is the gross primary production (µmol /m s ) and Reco is the total ecosystem respiration (µmol/m2s ). A light-response curve model, solely on characterizing carbon assimilation as a function of light, was used as a method to estimating Reco and GPP. This method typically uses a hyperbolic function that resembles the response of photosynthesis to radiation. There are a variety of light response Chapter 3. Interpreting Carbon Fluxes from a Costa Rican Tropical Dry Forest 25 curve models available, but the speciﬁc function used for this study was utilized by [117] for eddy covariance data in a tropical forest and expressed as:

a × PAR NEE(PAR) = + c = GPP + R , (3.3) b + PAR eco

a × PAR GPP(PAR) = , (3.4) b + PAR

Reco = c, (3.5) where net ecosystem exchange is a function of photosynthetically active radiation (PAR) with three parameters a, b, and c. The ﬁrst term represents the GPP as a function of photosynthetically active radiation (PAR) (3.4), while the second term, the constant c, represents ecosystem respiration for that interval (3.5).

Photosynthetic efficiency (α), defined as number of moles of carbon dioxide affixed through photosynthesis over the number of moles of incident photosynthetically active radiation (photons), was estimated by taking the initial slope of the light response curve function. To do this, we took the derivative of NEE with respect to PAR and evaluated at PAR = 0:

dNEE a(b + PAR) − (a × PAR) = (3.6) dP AR (b + PAR)2

dNEE ab a α ≡ = = . (3.7) dP AR P AR=0 b2 b

To get a measure of gross uptake rates near saturation, GPP at PAR = 2000 µmol/m2s was evaluated (3.8). This was judged to be more physical than taking the limit of NEE as PAR → ∞, which theoretically represents productivity at light saturation but is more sensitive to uncertainty in ﬁts.

a × 2000 A ≡ GPP(PAR = 2000) = (3.8) 2000 b + 2000

Analysis of parameters of the hyperbolic function used here were further developed for this study to produce mathematically consistent estimates of light saturated uptake and photosynthetic eﬃciency throughout the season.

In its initial application by [117], the constant parameter c was taken as ecosystem respiration while parameters a and b were not utilized. However, similar to methods used by [180], we chose to use a and b as uptake parameters. Nighttime fluxes were used as an independent estimate of ecosystem respiration, given the absence of photosynthesis at night. For eddy covariance fluxes to be interpreted as net ecosystem exchange, the surrounding air must be sufficiently turbulent for vertical mixing. Nighttime respiration was used when correlation between PAR and photosynthesis was beyond a cutoff of 0.60. A friction velocity cutoff of u* < 21 m/s was set to filter out periods of calm conditions where eddy covariance measurement techniques would not be reliable. Chapter 3. Interpreting Carbon Fluxes from a Costa Rican Tropical Dry Forest 26

3.4 Results and Discussion

3.4.1 Phenology and the Impact of Drought from Ecosystem Carbon Fluxes

Phenological cycles associated with tropical dry forest were generally well-reflected by ecosystem carbon fluxes measured over the four-year study period [173]). Fluxes during ‘normal’ years (2013 and 2016), particularly 2016, showed all phases of the growth season (fig. 3.4). Green-up was observed as a rapid increase in productivity, stabilizing about a seasonal maximum during maturity, to then gradually decline during senescence. Net ecosystem exchange quickly decreased during the dry season, either stabilizing at a low rate of uptake or becoming slightly positive. The impact of drought on carbon fluxes was substantial. Decreases in precipitation during the 2014 and 2015 drought months were accompanied by reductions in GPP, NEE, and Reco. Low drought productivity suggests that precipitation accumulated before intra-seasonal drought events (El Veranillo de San Juan) did not provide sufficient moisture to maintain peak productivity levels over the drought periods. Late dry-season periods following drought years acted as mean carbon sources. These episodes may demarcate the limits of drought-adapted vegetation, likely due to depletion of moisture reserves in plant tissue and soil layers. Our results suggest that secondary tropical dry forest productivity and respiration is sensitive to precipitation anomalies, particularly during El Ninõsevere drought events.

Ecosystem-level photosynthetic efficiency (α) and saturated rates (A2000) generally followed patterns of net ecosystem exchange, which in turn followed patterns of precipitation (fig. 3.5). At the onset of the wet season, efficiency and saturated rates tend to increase toward a stable maximum approximately mid-season, to then decline progressively during the dry season, reaching approximately zero by the onset of the subsequent wet season. Both also showed punctuated declines and recoveries during the mid-wet season drought events in 2015 and 2016. Growth and decline of leaf area likely underlie some of the variation in efficiency and saturated rates. However, physiological changes at the leaf-level driven by moisture availability, may also have a significant contribution [67]. Interestingly, photosynthetic efficiency showed markedly higher values in 2016, significantly greater than all three years, rapidly growing and stabilizing during the onset of the season. Saturated rates of photosynthesis also showed similarly pronounced values in 2016, though growth was more gradual, peak- ing in mid-season, and declining afterward. Given that precipitation was only roughly 200 mm greater than the previous ‘normal’ year of 2013, this phenomenon may be caused by the previous consecutive years of drought. Though fast response to sudden increases in soil moisture has been reported in tropical forest ecosystems [147], I hypothesize that these elevated rates of efficiency and uptake may be a form of post-drought ‘recovery’, though further ecological study at a finer-scale is required.

3.4.2 Comparison with Tropical Rainforest

Hutyra et al [117] performed a four-year study of tropical rainforest (Tapajos National Forest, Para, Brazil) fluxes as measured by an eddy covariance system, which we use to compare the seasonal patterns of carbon fluxes of these two major tropical forest ecosystems. Tapajos National Forest is located within Amazonia and is classified as primary rain forest, with distinct seasonal meteorology including an approximate five-month dry season (July–September). However, Tapajos rainforest (∼300–400 mm) still receives an order of magnitude more dry season precipitation than SRNP ( ∼20–30 mm). The severe SRNP dry season is reflected in our site by the decline of respiration and GPP throughout Chapter 3. Interpreting Carbon Fluxes from a Costa Rican Tropical Dry Forest 27

Figure 3.4: Time series for NEE, GPP, and Reco for the four year study period (modified figure from Castro et al. [52]) the dry season (fig. 3.6). NEE remained negative during the dry season for non-drought years and acted as a carbon sink, possibly due to the presence of evergreen trees with deep root systems. However, Chapter 3. Interpreting Carbon Fluxes from a Costa Rican Tropical Dry Forest 28

Figure 3.5: Time series for photosynthetic efficiency (α) and light-saturated rate of uptake (A2000( for the four year study period, with blue shaded regions representing the wet season and red shaded regions representing the dry season (modified figure from [52]).

NEE tapers off with progression into the end of the dry season, switching to a carbon source at the end of the season. In contrast, rain forests have been shown to sustain high levels of productivity during its dry season with highest uptake values occurring during its driest period, coinciding with peak levels of PAR [90][113]. Our studied dry forest of SRNP was also found to have higher peak rates of GPP (>13 µmol/m2s) than the Tapajos rain forest (∼8 µmol/m2s) during its wet season [117]. While environmental conditions differ making direct comparison difficult, tropical deciduous trees have been previously found to have higher photosynthetic capacity than moist forest trees [164].

During the wet season, both rain and dry forest (for non-drought years) follow a similar hill-shaped pattern of productivity where GPP increases at the onset of the rains, peaks in mid-season, then declines for the rest of the season. Respiration rates of rainforest have been found to decrease during the dry season, attributed to the drying of surface litter [117]. Thus, the aridity of the dry season in rainforest may be suﬃcient to suppress respiration, but insuﬃcient to inhibit productivity [117]. However, the more extreme aridity of the dry season in tropical dry forest drives senescence and an eventual cessation of carbon uptake. Chapter 3. Interpreting Carbon Fluxes from a Costa Rican Tropical Dry Forest 29

3.4.3 The Birch Eﬀect in Tropical Dry Forest

Sudden pulses of CO2 were observed following the first rains at the onset of the wet season. This was most apparent in 2016, after two consecutive El Ninõinduced years of drought (fig. 3.6). In the final

(a)

(b)

Figure 3.6: a) Net ecosystem exchange and soil moisture measurements at half-hourly intervals spanning from April 26th, 2016 (0:00) to April 28th, 2016 (02:00) at the onset of the first rains at the end of the dry season. b) Precipitation measurements for the same time period. Chapter 3. Interpreting Carbon Fluxes from a Costa Rican Tropical Dry Forest 30 days of the dry season, soil moisture tapered to steady low values concurrent with low but steady values of positive carbon fluxes (release). A sudden increase in soil moisture signaling the absorption of water from the first rain event was recorded at approximately 21:00 (April 26th, 2016), followed by a sudden burst of CO2 that quickly peaked and more gradually declined over a 9 hour period. Given that the delay between soil moisture increase and ecosystem response occurred at night, it is possible that carbon fluxes were simply not recorded due to the lack of turbulent transport. The origins of the observed pulse of CO2 are likely to be of heterotrophic respiration by microbial decomposition of stored organic matter, as little foliage remained active by the end of the dry season. The sudden wetting of dry soils and the subsequent pulse of CO2 release and nitrogen mineralization is a phenomena known as the ‘Birch effect’ [29], which was first observed in East African soils. The Birch effect has also been recently observed in another tropical dry forest site in Guanacaste, however on a smaller spatial scale using simulated precipitation and chamber-based measurements [232].

Our identification and characterization of the Birch Effect in SRNP closely resembles CO2 effluxes observed at the onset of the wet season in a dry forest site in northwestern Mexico [230]. Seasonal drought is a key characteristics of SRNP, suggesting that Birch Effect events may also be a regular function of this dry forest ecosystem with important consequences in the carbon cycle and soil nutrient processes.

3.4.4 Connections with Model Representations of Tropical Dry Forest

The Lund-Potsdam-Jena (LPJ) DGVM and Spatially Explicit Individual Based DGVM (SEIB-DGVM) are two advanced vegetation models that are commonly used to project the future impact of climate change on terrestrial ecosystems. Model processes (radiative transfer, photosynthesis, respiration, evapotranspiration. etc) are driven by climatological data and run on a spatial grid at daily, monthly, and annual time steps. Establishment, growth, competition, and mortality of individuals from plant function types (PFT) are simulated within each grid cell. The composition of PFTs within a grid cell are determined by atmospheric and edaphic conditions, which affect each PFT through their allometric, bioclimatic, and physiologic parameterizations. Biome classification of a grid cell is typically done during post-processing of raw model output, where the proportion of PFTs within the cell play a central role. The primary difference between the two models is SEIB-DGVM’s increased spatial dimensionality, modeling the canopy in three dimensions for more accurate modeling of radiative transfer with a consequently heavier computational load. For detailed descriptions of LPJ and SEIB-DGVM, see [211] and [204] respectively. LPJ and SEIB-DGVM use a similar characterization of the dry forest plant functional type (PFT), ‘tropical broadleaf raingreen’ (TrBR), originally used by the BIOME3 model [97]. The ‘raingreen phenology’ used by SEIB-DGVM for TrBR uses water status, a function of volumetric soil moisture, to control phenological status. A moving-average of water status is computed over a fixed interval (Dmax), set to 10 days in most studies. When this value falls below a threshold value, individuals of the TrBR pft shift into a drought-induced dormancy state, where growth is suspended and foliage is shed. When the average water status rises above said threshold value, TrBR PFTs resume their growth phase. Similarly, in LPJ, a water stress factor (ω), the ratio of evapotranspirative supply against demand, controls leaf phenology for the TrBR PFT. When ω falls below a threshold that indicates severe drought, leaves are immediately shed, and are immediately restored when ω returns to values above the threshold. Unlike water status in SEIB-DGVM, ω in LPJ is computed daily, causing a more sudden reaction to drought. Chapter 3. Interpreting Carbon Fluxes from a Costa Rican Tropical Dry Forest 31

In years with well-defined wet and dry seasons, both model parameterizations likely capture dry forest phenology with several caveats. Delays between initial precipitation events and green-up in SNRP range from two to seven days, while volumetric soil water content reacted to rainfall within hours, [80][52], suggesting that a ten-day (Dmax) moving average as used by SEIB-DGVM may be too large and that the single-day evaluation used in LPJ may be more appropriate. In contrast, transitions to dry-season dormancy tend to be gradual based on time-series of carbon uptake, photosynthetic efficiency, and light-saturated uptake (fig. 3.4,3.5), following patterns of soil moisture to drying. DGVMS such as LPJ and SEIB-DGVM typically use a simple, empirically-derived function to connect soil moisture and carbon uptake, which varies among models [189]. Ultimately, the abrupt phenology imposed on the TrBR should be replaced by a mechanistic hydrodynamic parametrization that models soil to leaf water transport [82], characterizing water transport with a more natural threshold that could also better model the effects of non-seasonal drought events. Evaluation of soil moisture relations to ecosystem metabolism may be critical in improving estimation of carbon fluxes and pools, given the importance of phenology in ecosystem-level carbon exchange [58]. Soil respiration parametrizations are typically controlled by temperature, possibly owing to the generally temperate ecosystem bias in validation sites. It is unlikely that the Birch Effect potentially observed at SRNP would be well-parametrized by the soil moisture algorithms used in DGVMs such as SEIB-DGVM and LPJ, both which apply methods from Foley’s litter decay model [85]. Though the soil decomposition rate used by these models are functions of both soil temperature and moisture, monthly rather than daily or hourly averages are used. Pulses of carbon stimulated by first rains of the wet season may occur over a small timescale but large spatial scale, considering the large extent of tropical dry forest. Large-scale, global models tend to be temporally coarse, limiting the detail in parametrizations of dry forest phenology, though fast phenomena like the Birch effect may be significant to annual carbon budgets [120]. Models that simulate both the diffusion of water into the soil substrate and moisture- dependent microbial activity at high temporal resolution (i.e. hourly) have recently been successful in reproducing patterns akin to the Birch effect [78], and though computationally more demanding, could be integrated into DGVMs. Values of α as estimated by our study can be implemented into DGVM through its PFT-specific parameters. Light use efficiency (LUE) is defined by SEIB DGVM as the potential maximum value of light use efficiency and is a primary PFT specific parameter that a number of internal photosynthetic parameters are contingent on. The peak value of α within our time series is 0.061, which occurred in the early peak of productivity during the 2016 wet season and was taken to be a potentially more accurate value of LUE for tropical dry forest. This is significantly greater than the value of 0.05 assigned to the LUE of TrBR and all interestingly all PFTs for the global validation of study of SEIB-DGVM. Modification of LUE would translate to changes in flux and pool estimates. Variance between PFT LUE could also translate to changes in competition between pfts within grid cells and the spatial distribution of PFTs.

3.5 Conclusion

The seasonal patterns of carbon fluxes were observed through an eddy covariance measurement system in this study, while a hyperbolic light-response function was used to partition NEE into GPP and Reco. Net uptake, productivity and respiration were both found to be larger during the rainy season following Chapter 3. Interpreting Carbon Fluxes from a Costa Rican Tropical Dry Forest 32 an approximate hill-shaped pattern to then decline in the dry season. El Niñoyears showed lower annual precipitation and mid-wet season drought events, which induced decreases in uptake, respiration, and productivity, as well as concurrent decreases in photosynthetic efficiency and saturated rates of uptake. Thus unlike tropical rain forest, moisture levels were found to inhibit rates of carbon assimilation at the ecosystem scale. Photosynthetic efficiency and saturated uptake were observed to be markedly higher than previous years during the wet season of the final full year of observation (2016), after two consecutive years of drought. A phenomena known as the ‘Birch Effect’ was potentially observed during our study period, particularly in 2016, when a large pulse of carbon dioxide was released following the first rain at the onset of the wet season. The results of this study give a quantitative basis to previously speculative theories of ecosystem function of tropical dry forest, enabling a general understanding of patterns of carbon uptake and release. However, for reliable annual sums, a storage term should be applied to account for accumulation or depletion of CO2 within levels of the vertical profile under the measurement system. In addition to the paucity of direct research on tropical dry forest, there is a lack of devoted trans- lation of existing research into computational models. Examination of the numerical representations of tropical dry forest in DGVMs suggest that basic parametrizations still needs to be performed, including but extending beyond refinement of parameter values. Perhaps most feasible is the integration of a mechanistic model linking soil moisture to ecosystem respiration and productivity within a land surface model. Chapter 4

Amazonian Dry Corridors opened by Fire and Low CO2: A Song of Ice and Fire

4.1 Abstract

We present integrated modeling evidence for a grassier, less forested Neotropics during the Last Glacial Maximum, congruent with palaeoecogical and biological studies, and grounded in the fire-mediated dynamics of forest-savanna boundaries. High-resolution model reconstructions of vegetation were generated using the Land Processes and eXchanges model, driven by four climate model reconstructions of LGM climate, and tested against a body of palynological data. A factorial experiment was performed to quantify the impact of fire and low atmospheric carbon dioxide on modeled vegetation and model-data agreement. Fire and low CO2 were both found to induce widespread expansion of savanna and grassland while improving model-data agreement. The interactive effects of fire and low CO2 were found to induce the greatest ‘savannafication’ of the Neotropics, providing strong modeling evidence for a number of biologically relevant open vegetation formations including two dry corridors; paths of savanna through and around Amazonia that facilitated major dispersal and diversification events. Our results suggest forest and savanna to be alternate stable states as fire was also found to induce the bimodality of tree cover, which was further enhanced by ‘CO2 deprivation’.

4.2 Introduction

The dynamics of vegetation cover over geologic time carry heavy evolutionary biological and biogeographical implications. A number of proposed mechanisms of biotic diversification in the Neotropics rely on broad changes in vegetation, which induce large-scale processes such as dispersal, vicariance, and speciation [81]. Perhaps most well-known is Haffer’s Refugia Hypothesis, which postulated that Ama- zonian rainforest was fragmented into disjoint ‘refugia’ by tracts of savanna during Pleistocene glacial periods. This fragmentation would be sufficient to inhibit gene flow between Refugia, leading to speciation. Forests would expand and reconnect during interglacials, leading to range expansion of the newly

33 Chapter 4. Amazonian Dry Corridors opened by Fire and Low CO2 34 diversified taxa [92]. Since its inception, a number of palaeoecological studies have refuted the Refugia Hypothesis [59][43][44], claiming that Amazonia remained resilient over Pleistocene climatic fluctuations. These studies are roughly consistent with past model reconstructions [63][57][176][111], which show Amazonian forest during the LGM to be similar in extent to its pre-Industrial status. Conversely, more recent palynological studies have found savanna-like vegetation during the late Pleistocene within current Amazonian forest [100][86] and re-evaluation of the Lake Pata records has also challenged the notion of stable moist forest even within the basin over glaciations, finding discontinuities and significant compositional changes during the last glacial period [68]. The degree of ‘savannafication’ of the glacial Neotropics is thus largely unknown due to the scarcity of pollen cores that date back to the LGM and absence of comparison against modeling output. Moreover, Haffer’s conceptual contributions regarding connections between cyclical Pleistocene climatic changes, large-scale biome shifts, and diversification events, may still be a valuable framework in understanding vegetation dynamics and diversification. Distant and disjoint distributions of species associated with semi-arid biomes suggest the past presence of open vegetation biomes in regions that are currently occupied by closed canopy tropical forest [194][190]. Three past savanna formations, referred to as ‘dry corridors’, have been hypothesized to explain past connectivity between the northern and southern savanna regions of South America [50][194]. The central Amazonian corridor would extend diagonally northwest to southeast from the savannas in the northern Amazonian border to those in central Brazil, along an extensive tract of forest that currently experiences a significant degree of seasonality in precipitation [241][47]. The circum-Amazonian corridor or Andean corridor would have existed along the Andes and western border of Amazonian forests, while the coastal corridor has been hypothesized to have existed along the Atlantic coast. Similarly, distributions of caatinga taxa provide evidence for a hypothetical past formation known as the ‘Pleistocene Dry Arc’ [190][185][206][228], an arc-shaped expanse of semi-arid caatinga vegetation extending east-west from southeastern Amazonia to the current caatinga of eastern Brazil. Palaeoecological studies often attribute large-scale changes in vegetation to the cooler and drier glacial climate of the LGM. However, natural disturbances such as fire have been shown to be strong regulators of forest-savanna boundaries and vegetation composition in both modern and palaeoecological contexts [174][107]. Additionally, the concentration of atmospheric carbon dioxide has been proposed to have marked effects on biome distribution through differential decreases in growth rates [122][96]. Given the potentially critical role of CO2 fertilization on vegetation in future climate scenarios, it is natural to consider converse scenarios where concentrations were significantly lower. Low CO2 as associated with glacial periods, has been proposed to have given a strong competitive advantage to grasses over trees, allowing savanna to encroach into forested regions [34][130]. Trees allocate large amounts of carbon to non-photosynthesizing structural biomass (ex. wood), which results in a slower post-fire recovery rate relative to grasses. Bond [33] postulated that low CO2 could slow tree recovery rates such that they are even further outcompeted by grasses. Slowed tree growth rates associated with low CO2 would also increase the time required for trees to grow to heights sufficient to survive surface fires, which would progressively decrease tree density if fires are frequent. Forest, savanna, and grassland have been proposed to be alternate stable states (biomes) for regions of intermediate precipitation that are heavily influenced by fire regime [157]. According to the alternative state theory, multiple distinct biomes could be supported by identical climatic and edaphic conditions, and posits that other processes such as fire may be important determinants of which state the region Chapter 4. Amazonian Dry Corridors opened by Fire and Low CO2 35 settles into [220]. Modern remote sensing analyses have shown consistent bimodality in tree cover against gradients of precipitation as opposed to a continuous spectrum, providing empirical support for alternate stable state theory [105][218]. These studies also deem the central Amazonian dry corridor to be bistable forest with relatively low resilience to fire-induced loss of tree cover and the east-west Pleistocene dry arc to be bistable regions of low tree cover, leaving only a small, core region in the inner lowlands to be climatically deterministic forest. The variability of CO2 over glacial-interglacial cycles adds another dimension of complexity. CO2 fertilization allows for lower stomatal conductance for similar rates of photosynthesis, which simultaneously mitigates water loss through transpiration (see ch. 2). Conversely, low CO2 as associated with glacial cycles could have lead to ‘CO2 deprivation’, inducing higher rates of stomatal conductance, lower water use efficiency, and an altered competition between trees and grasses.

Here, we test the effects of fire processes and atmospheric CO2 levels on Neotropical late-Pleistocene vegetation. A factorial experiment was conducted to elucidate these effects by constructing four scenarios, all driven by LGM climate:

• pre-Industrial CO2 (280 ppm) without ﬁre processes (control)

• pre-Industrial CO2 (280 ppm) with ﬁre processes (ﬁre only)

• LGM CO2 (180 ppm) without ﬁre processes (low CO2 only)

• LGM CO2 (180 ppm) with ﬁre processes (ﬁre and low CO2)

To account for variability among climate models, four distinct Atmosphere-Ocean General Circulation Model (AOGCM) reconstructions of LGM climate (MIROC3.2, HadleyCM3M2, FGOALS-1.0g, and CNRM-CM33) were used to drive the Land surface Processes and eXchanges (LPX) dynamic global vegetation model (DGVM) [193]. Each LGM climate model reconstruction was used to drive LPX in order to generate the four scenarios required for the ﬁre-CO2 factorial experiment. The outputs from all four of the LGM AOGCM reconstructions were also averaged to create an ensemble factorial experiment that was also subject to analysis. We then performed a model-data comparison for each of the four scenarios for the ﬁve climate model reconstructions, using a collection of pollen core studies (table 4.3) that date back to the LGM. After identifying potential evidence for past biogeographic formations, a

Stein-Alpert Decomposition was used to find the individual and interactive effects of fire and low CO2 on tree cover to better understand the drivers of savannafication in the Neotropics during the LGM.

4.3 Results

4.3.1 Comparison of Model Reconstructions with Palynological Data

The activation of fire and low atmospheric CO2, individually and in concert, resulted in small but consistent improvements in agreement between model reconstructions of LGM vegetation and palynological data. Expansion of open biomes and displacement of forest, induced by fire activation and low CO2, were deemed to underlie these improvements. Model-data agreement was improved by the simultaneous activation of fire and CO2 relative to control for all five experiments (four AOGCM LGM climate reconstructions + one ensemble), four of which were deemed to be statistically significant according to a paired student t-test. The activation of fire (fire only) and the imposition of low CO2 (low CO2 only) Chapter 4. Amazonian Dry Corridors opened by Fire and Low CO2 36

Figure 4.1: Summary of model-data comparison scenarios from the fire and low CO2 factorial experiment for each LGM climate reconstruction. Scenarios with solid black bars at bottom are not significantly different from Fire and CO2 ensemble scenario. Dashed Bars: mean of DMM point score/scenario scores Solid Bars: median of DMM point scores Boxes = Interquartile Range Whiskers: Minimum and Maximum or 1.5 × Interquartile Range in isolation both improved model-data agreement, though fire activation consistently improved DMM scores more than low CO2. However, the simultaneous inclusion of fire processes and low CO2 (fire and low CO2) resulted in the lowest average DMM and highest model-data agreement, implying the importance of their combined effects. In terms of intermodel variation, the ensemble runs had higher model-data agreement than any runs driven by individual AOGCM LGM climate reconstruction. Variations between LGM climate reconstructions were reflected in variations in model-data agreement, though almost all AOGCM runs showed similar trends within their factorial experiment (fig. 4.1). The control scenario, with Pre-

Industrial CO2 levels and fire processes disabled, had lowest data-model agreement, followed by fire only and low CO2 only scenarios, with fire and low CO2 having the highest relative model-data agreement save for FGOALS-1.0g. Thus, of the twenty model vegetation reconstructions, the ensemble fire-and-low

CO2 run agreed most with palynological data according to our methods. While a greater number of pollen cores that date back to the LGM would greatly aide groundtruthing of model reconstructions of Chapter 4. Amazonian Dry Corridors opened by Fire and Low CO2 37 terrestrial vegetation, such cores are limited by the rarity of suitable preservation sites in the largely fluvial Amazonian system. The margins of Amazonia, particular the southern border with the cerrado, is a notable region of contention between the model scenarios. The string of pollen sites in this region (Katira, Lago do Saci, and Serra Sul Carajas) provide a rough measure of the retreat of the southern Amazonian border, which varied significantly between models and within factorial experiment scenarios. The extent and fragmentation of Atlantic forests also play a significant role in variations in model-data agreement, as pollen records indicate a grassy, open landscape, while model reconstructions vary widely, showing sensitivity to fire and CO2. In the cases where the inclusion of fire and low CO2 resulted in decreased model-data agreement, model reconstructions tended to predict open or very open (ex. desert) vegetation where pollen indicated forest or open vegetation with some trees (ex. savanna). This occured in the factorial experiment driven by MIROC3.2 LGM climate reconstructions, where the activation of fire processes at Pre-Industrial CO2 decreased model data agreement (increased average DMM distance relative to control) by incorrectly reconstructing desert at the northern margins of Amazonia. This also occurred in the FGOALS-1.0g experiment, where savanna core points (Crominia, Laguna Bella Vista,

Chaplin, Lagoa de Caco) were correctly reconstructed in the low CO2 only scenario, but shifted to grassland in the fire and low CO2 scenario. However, no reductions in model-data agreements caused by fire or low CO2 were determined to be statistically significant.

4.3.2 Fire and Low CO2 Activation Drives Expansions of Grasslands and Reductions of Forest

Fire and low atmospheric CO2 had profound and widespread effects on model reconstructions of LGM vegetation. Relative to the control case of fire processes disabled and Pre-Industrial CO2, the activation of fire processes at pre-Industral CO2 (fire only) showed a general shift to biomes associated with aridity. In the fire only scenario, tropical moist forest was reduced by approximately 10% relative to control, which was replaced predominantly by tropical savanna (fig. 4.1 a, b). Warm temperate forest, most of which was from Atlantic forest, also showed reductions in area and were replaced by sclerophyll woodland. The area occupied by grassland more than doubled, replacing savanna in the currently semi- arid caatinga of Northeastern Brazil and woodland and parkland in and adjacent to the Pampas region in the south. In the north, in northern Colombia and Venezuela, fire was found to induce desert where savanna and grassland occupied in the control.

Interestingly, the effects of low CO2 appear to have a similar but weaker effect on biome distributions as fire: forest biomes are reduced in extent by intrusions of grassy biomes such as savanna and grassland

(ﬁg. 4.2 a, c). Relative to control, low CO2 (low CO2 only) induced shifts in biomes similar to that of the

fire only scenario with a few significant differences. Low CO2, unlike fire, removed the strip of tropical dry forest that bordered savannas of the cerrado region and Amazonian moist forest. This is likely owed the fire-tolerance (lower probability of mortality during fires) of the tropical broadleaf raingreen plant functional type (pft) that dominates dry forest biomes. Low CO2 induced expansions of grassland, but less than that of fire-processes. Desert area also remained mostly unchanged by the decrease in CO2.

The combined effect of fire and low CO2 had the most pronounced effect on biome distribution relative to the control scenario. Reductions in forest area were dramatic as were expansions of grassy biomes, including regions that were unaffected by either factor in isolation. Amazonian forest contracted significantly, particularly the southern margins which showed a more jagged, irregular shape relative Chapter 4. Amazonian Dry Corridors opened by Fire and Low CO2 38 to control. Moreover, savanna and grassland showed extensive intrusion into the central Amazonian corridor region, with small patches of isolated savanna of various size within the forest. There was roughly half the area of tropical moist forest in the fire and low CO2 scenario relative to the control scenario. Warm temperate forest, which represents Atlantic forests, occupied only 15% of the area relative to control. Conversely, grassland was expanded by over five-fold relative to control, occuring in large patches in the mosaic of open canopy biomes that comprise the majority of the fire and low

CO2 scenario. These general trends were true for all LGM model recosntructions driven by the four AOGCMS, though significant variation exists in biome distributions. It is noteworthy that Amazonian rainforests remained vast and continuous in scenarios in both fire only and low CO2 only scenarios, consistent with past modeling studies [63][57][176][111]. It is only with both effects simultaneously activated that savanna and grassland expansion manifestly contradicts previous reconstructions of LGM

South America, which did not explicitly test non-climatic factors such as fire and CO2. To examine the effects of atmospheric carbon dioxide concentration and fire on climate-vegetation relations, Whittaker plots for each model run scenario were generated (fig. 4.3). Similar to Holdridge’s lifezone plot [108], Whittaker plots [238] display the location of biomes in climate space (mean annual precipitation vs. mean annual temperature). Scenarios with fire processes disabled tend to have biomes that occupy clearly defined areas of climatic space, similar to the classic Whittaker plot. The activation of fire processes tends to obscure these well-defined boundaries, primarily by expanding the climate- space occupied by open biomes. The expansion of grassland is perhaps the most distinct effect of fire, which is well-reflected in its broad and speckled climate-space distribution. Low CO2 tends to reduce the area of climate-space of all major forest types, allowing savanna to encroach into the climate-space of tropical moist and dry forest and sclerophyll woodland to encroach into that of warm temperate forest.

The combined effect of low CO2 and activated fire processes compounds these two effects, resulting in a heavily reduced climate-space of forest biomes while expanding that of grassland, reflected in the

Whittaker plot for the ﬁre and low CO2 scenario relative to that of control.

4.3.3 Fire and Low CO2 Open Dry Corridors

A number of biologically significant formations of open vegetation, including two dry corridors, were identified in model reconstructions of the Neotropics during the LGM (fig. 4.4). Indications of a central Amazonian corridor appeared in a number of model runs, showing almost full connectivity in the ensemble fire and low CO2 scenario, which had the highest agreement with pollen records. In this reconstruction, the savannas of the cerrado significantly expanded through the southeast margins of Amazona to connect with the grassy biomes in the far north. A narrow, continuous tract of savanna and grassland resembling a circum-Amazonian dry corridor was also reconstructed in a number of runs, also most prominent in fire and low CO2 scenarios. Moist forest remained robust along the Atlantic coast in eastern Brazil, showing no indication of a trans-Amazonian Atlantic corridor at the biome level. However, tree cover and height were slightly reduced according to the ensemble vegetation reconstruction, suggesting a degree of openness in the region. Atlantic forests were found to be present but heavily reduced in size and restricted to the Brazilian coast, consistent with phylogeographic and statistical modeling work of Carnaval and Moritz [51]. The

Pernambuco and Bahia refuges were both present in the ﬁre and low CO2 reconstruction, separated from one another by tropical savanna. A third potential refuge composed of a thin strip of warm temperate forest south of the Doce river was also identiﬁed in this run, which was predicted to be stable based on Chapter 4. Amazonian Dry Corridors opened by Fire and Low CO2 39

Figure 4.2: LPX model reconstruction of biome distributions in the Neotropics during the LGM for four scenarios:

Thf = tropical humid forest, Tdf = tropical dry forest, Ts = tropical savanna, sw = sclerophyll woodland, tp = temperate parkland, bp = boreal parkland, dg = dry grass/shrubland, hd = hot desert, st = shrub tundra, t = tundra

phylogeographic data but unstable based on palaeomodeling techniques. The apparent fragmentation of Chapter 4. Amazonian Dry Corridors opened by Fire and Low CO2 40

Figure 4.3: Whittaker plots [238] showing occurrence of biomes in climate space for four scenarios designed by the ﬁre/CO2 factorial experiment. Points that are identical to control scenario are left white while coloured points represent shifts due to activation of ﬁre and low CO2.

Atlantic forest during the Pleistocene is hypothesized to be a primary driver of the high biodiversity of this biome [51]. The inclusion of fire also provides modeling evidence for a the Pleistocene Dry Arc. The ensemble low CO2 only scenario shows a distribution of dry grass-shrubland, savanna, and temperate parkland from the Eastern Brazil transitioning to the Paraguay and Argentina that generally agrees with modern relative locations of the caatinga, cerrado, and chaco biomes. In all LGM climate scenarios, fire changes the composition of these three open landscapes from well-defined distribution of biomes with smooth borders, to patchier mosaics, inducing large regions of drygrass-shrub where fire-off runs reconstructed savanna and temperate parkland, roughly corresponding to the Piedmont, Misiones, and caatinga nuclei Chapter 4. Amazonian Dry Corridors opened by Fire and Low CO2 41

Figure 4.4: Identiﬁcation of reconstructed biogeographical formations in the ensemble ﬁre and low CO2 scenario. Arrows indicate estimated location of formations (colour varies for reader visibility).

[190].

4.3.4 Interactive Eﬀects of Fire and low CO2 on Tree Cover

To estimate spatially-resolved individual and interactive effects of fire processes and CO2 on tree cover, a modified Stein-Alpert factor separation was performed on the four fire/CO2 scenarios driven by the ensemble LGM climate. Individually, fire and low CO2 had showed similar effects on tree cover, inducing large reductions in Central America and the north, south, and southeast of South America, while having Chapter 4. Amazonian Dry Corridors opened by Fire and Low CO2 42

Figure 4.5: Results from the Stein-Alpert Decomposition showing individual and interactive effects of fire and low CO2 in terms of fractional changes in tree cover. Purple regions show losses in tree cover due to a specific factor while green shows increases.

little to no effect on Amazonia (fig. 4.5 a,b). Low CO2 individually had broader effects on tree cover, showing small reductions in the central corridor regions of Amazonia and major reductions in southeast of Brazil (fig. 4.5 a, c). Interactions between fire and low CO2 result in enhanced tree cover loss in several regions, including the northern savanna of Colombia and Venezuela and large patches of Amazonia in central Brazil (fig. 4.5 a, d). It is notable that the broad savannafication required to open the central Amazonian dry corridor seems contingent on the interactive effects of fire and low CO2, while both the individual and interactive Chapter 4. Amazonian Dry Corridors opened by Fire and Low CO2 43

effects of fire and low CO2 induce opening of the circum-Amazonian corridor. These synergistic effects may underlie differences in results from past reconstructions of LGM Neotropics, given that scenarios without interative fire/CO2 effects also do not reconstruct a central Amazonian corridor or widespread expansion of grassy biomes.

4.3.5 Low CO2 Intensiﬁes the Fire-Forced Bimodality of Tree Cover

Figure 4.6: Fractional tree cover against mean annual precipitation aggregated by runs driven by four AOGCM LGM reconstructions. Darker shades of green indicate higher density of points.

Tree cover was plotted against precipitation for each scenario using output aggregated from all four model runs to better understand how fire and CO2 may mediate the relationship between moisture and tree cover (fig. 4.6). In the control scenario, a spectrum of tree cover occurred over regions of Chapter 4. Amazonian Dry Corridors opened by Fire and Low CO2 44 intermediate precipitation (0 - 2000 mm) (fig. 4.6 a). Intermediate tree cover (0.3 - 0.6) is present, showing little if any sign of bimodality of tree cover. The activation of low CO2 (low CO2 only) tends to reduce the density of very high tree cover (∼ 0.95), likely expressing reductions of tree cover even in regions that receive ample precipitation (fig. 4.6 c). The activation of fire (fire only) results in the increased presence of low tree cover even in regions of high precipitation relative to control (fig. 4.6 b). Fire also has another distinct effect on tree cover: intermediate tree cover is less frequent and the spectrum of tree cover is shaped into a more definite bimodality. The fire-forced bimodality of tree cover is further enhanced by the simultaneous activation of low CO2 and fire (fire and low CO2), showing more distinct reductions of intermediate tree cover over a broader range of precipitation (fig. 4.6 d).

4.4 Discussion

A number of methodological advances were made in the effort to perform a comprehensive modeling study of the Neotropics during the LGM. Four AOGCM models plus an ensemble average were used to account for variability in climate reconstructions and four scenarios were created for each driving model to account for the variable effects of low CO2 and fire, unlike past studies that typically rely on a single reconstruction and neglect non-climatic factors. A novel method was designed to compare model reconstructions against empirical data, which allowed measure of model-data agreement among driving AOGCM climate reconstructions and vegetation model scenarios. This is a key step toward reconciling the theoretical approaches of modeling with the empirical approaches of palaeoecology, which are both necessary to progressively assess and improve the skill of terrestrial vegetation models over long timescales. The activation of fire and inclusion of low atmospheric carbon dioxide improved data-model agreement for almost all climate model scenarios, suggesting that it may have had a significant role in determining vegetation in the LGM Neotropics. Fire inclusion generally expanded open vegetation biomes such as savanna and grassland while reducing forest, particularly Amazonian rainforest. Though reduced in size, Amazonia remained largely continuous with a stable western core consistent among various reconstruction scenarios. Grassland and savanna were also reconstructed in regions that have been hypothesized to have lower tree cover in the past, such as the central Amazonian and circum-Amazonian corridor regions. Thus, we conclude that vegetation in LGM South America may have been more open than previously thought, with significantly less forest and significantly more mosaics of grassy biomes. A major consequence of our study was the reconstruction of a number of biologically-significant formations including the central Amazonian corridor, circum-Amazonian corridor and Atlantic forest refugia. This provides strong modeling evidence for extensive grassy dispersal routes in South America, critical to understanding the processes that generated the high biodiversity of the region. An aim of this study was to provide robust model reconstructions to specialists outside the modeling realm, with intentions of integration with biological data. The mechanisms that generated such biogeographical formations were also explored in this study. The processes underlying the widespread savannafication were rooted in the effects of fire and low atmospheric carbon dioxide, which act as simultaneous and interactive stresses on trees. Similar to fire, low CO2 was shown to reduce tree cover and increase grass cover in large expanses. Within the context of the LGM, low CO2 gave grasses a competitive advantage over trees, compounded by the effects of fire, and worsened by aridity associated with glacial periods. These processes would better explain Chapter 4. Amazonian Dry Corridors opened by Fire and Low CO2 45 palaeoecological evidence of a largely grassier South America during the LGM, which was overall more vulnerable to intrusions by savanna. In the context of future projections, CO2 levels may well play have an active role in the distribution of future Neotropical ecosystems, which may be more sensitive to environmental changes than previously thought.

The opening of the central and to a lesser extent, circum-Amazonian dry corridor, was attributed to the interactive effects of fire and low CO2. Fire was found to be a driving factor in the bimodality of tree cover, with enhanced efficacy at lower levels of CO2; a potentially significant interaction between fire and ‘CO2 deprivation’. The results of our study thus support and expand upon two existing ecological theories: the fire-driven bimodality of tree cover and the expansion of grasses due to low atmospheric carbon dioxide. Our results also emphasize the oft-neglected but potentially critical role of non-climatic drivers on vegetation cover in palaeoecological contexts and evolutionary history in tropical regions.

If fire and CO2 are indeed important non-climatic determinants of terrestrial vegetation, there would be important consequences beyond the realm of modeling. Even when driven by identical climate scenarios, modification of atmospheric carbon dioxide concentration and fire regime can induce heavy variations in distributions of terrestrial vegetation. Relations between vegetation and climate are sensitive to non- climatic factors and dynamic through time due to fluctuating CO2, which is conceptually distinct from the works of Holdridge [108] and Whittaker [238]. Our results support the notion that interpretation of pollen spectra for palaeoclimate reconstructions be adjusted to account for important non-climatic processes such as fire and atmospheric CO2 (ex. transfer functions). Similarly, ecological niche modeling studies which also assume robust, stable relationships between vegetation and climate, may also face this issue if reconstructing vegetation over geologic time. Without accounting for fire and low CO2 in glacial periods, stability of forest biomes may be overestimated and vegetation-mediated diversification processes may be obscured.

A mechanistic understanding of glacial period savannafications could have important consequences, with conceptual similarities to Haffer’s model of Pleistocene differentiation [93][88]. During glacial periods, open vegetation corridors would emerge and induce dispersal and range expansion for savanna and grassland adapted taxa. During interglacials, forest would replace said open biomes and dry corridors would close, inducing vicariance and differentation. Conversely, a broad central Amazonian grassy corridor could feasibly act as a barrier and isolate populations of moist forest taxa at opposing sides of the corridor inducing differentiation. Closing of corridors during interglacials would then allow the newly diversified moist forest taxa to disperse and expand their ranges. This process would be cyclic, following Pleistocene climatic flucutations, and act as a two-way pump or ‘accordian’. This would allow consistent and rapid differentiation, given that each phase leads to vicariance of taxa of alternating forest and savanna niches. While the duration and extent of a corridor may not permit full speciation, this process could drive recent intraspecific genetic differentiation. This unique confluence of factors involving the geography of the Neotropics and Pleistocene climatic oscillations may have led to an accumulation of biological diversity, which emphasizes the value in protection of its ecosystems and conservation of its taxa. Chapter 4. Amazonian Dry Corridors opened by Fire and Low CO2 46

4.5 Methods

4.5.1 Model Description and Protocol

Dynamic Global Vegetation Models (DGVMs) simulate interactions between the atmosphere and terrestrial biosphere, computing spatially and temporally resolved estimates of potential vegetation and ecosystem fluxes and pools among other outputs. These models provide an integrated, process-based means to estimate the impacts of climate changes on terrestrial ecosystems [66][212]. A primary usage of these models is to hindcast the past impacts of climatic change on the Earth’s land surface to complement palaeoecological data, while groundtruthing models over long timescales [162]. The DGVM used for this study, Land surfaces Processes and eXchanges (LPX) [192], is a descendent of the widely used Lund-Potsdam-Jena (LPJ) model [211], and a more recent development of LPJ SPread and InTensity of FIRE (LPJ-SPITFIRE) [222]. The fire regime within LPX is based on the SPITFIRE model, where dry fuel generated dynamically, ignited by lighting strikes and spread according to the Rothermal equations [198]. LPX and its fire regime have been extensively groundtruthed against modern data [193][128][129], which made it a particularly suitable model for the purposes of this study. Comprehensive descriptions of LPX and its components can be found in [192][128][222].

LGM Climate Scenarios and Modeling Protocol

Figure 4.7: Flow of model protocol from spin-up to biome assignment for each factorial experiment run (LGM climate reconstruction + factorial experiment conditions) Chapter 4. Amazonian Dry Corridors opened by Fire and Low CO2 47

Inputs reflecting the LGM climate were derived from four atmosphere-ocean general circulation model (AOGCM) datasets (MIROC3.2, FGOALS-1.0g, HadCM3M2, CNRM-CM33) produced by the Palaeoclimate Modelling Intercomparison Project Phase II (PMIP2) [40]. These datasets were also used to drive previous global-scale reconstructions of LGM vegetation [192][48]. Atmospheric CO2 was set to 185 ppm in accordance with the PMIP2 protocol for all runs. Several steps were required to prepare these model scenarios for compatibility with LPX. ‘Anomalies’ for each input variable, defined as the differences between LGM climate scenarios and a Pre-Industrial Holocene baseline climate, were superimposed onto a higher resolution (0.5◦ × 0.5◦) Pre-Industrial climate scenario (Climate Research Unit Version 3.0, detrended data from 1900-1950). Climate data with superimposed anomalies was then extrapolated onto the exposed continental shelf and removed from regions covered by ice sheets characterisitic of the LGM. For all model simulations, ‘spin-ups’ from bare ground were ran for 4000 years and main runs for 1380 years. Equilibrium tests were performed on each data set to check for the temporal stability of output variables. If for the majority of grid cells, canopy density (Leaf Area Index - LAI) and foliage projective cover (fpc) showed less than 2% variation relative to the previous timestep, equilibrium was taken to have been reached. Model output for the last 138 years (length of base Pre-Industrial data set) was averaged and used for reconstructions of LGM vegetation.

LPX Biome Assignment

Biome assignment was implemented through post-processing of three LPX outputs: Growing Degree Days (GDD), vegetation height, and foliage projective cover (fpc). The biome assignment scheme used in this study is identical to that of [192] and [48], save a refinement to distinguish seasonal and evergreen forests. A threshold of mean annual GDD (above 5◦C) was set to 350◦C·days to separate cold biomes from their warm and tropical counterparts (ex. tundra vs desert). Height and fpc are then used to distinguish between bareground, grassland, savanna, and forest biomes. The presence and or dominance of pfts within a grid cell then determines if it is tropical, boreal or temperate. Forests are classified as either seasonal or evergreen based on the relative proportions of summergreen, raingreen, and evergreen forest pfts. An illustration of the biome assignment scheme is shown in figure 8, adapted from the original in [192] to include the distinction between deciduous and evergreen biomes.

4.5.2 Model-Pollen Biome Correspondence

The definitions of South American biomes, their correspondence to pollen spectra, and LGM biome reconstructions were based on a comprehensive meta-analyses by Mayle et al. [161] and Marchant et. al [153] in addition to the original studies (supplementary - core list). Comparison of model output to pollen spectra requires a number of ‘translations’ (fig. 2). Pollen spectra can be translated to biomes either by personal interpretation or a statistical categorization process such as biomisation or machine learning algorithms [191][216]. Vegetation models have various schemes in translating vegetative outputs to biomes. In practice, pollen-based biomes and the model- outputs rarely have a simple one-to-one correspondance, given that vegetation model biomes are often developed for global application, while pollen-based biome reconstructions are refined to more subtle regional or local definitions. Once a correspondance scheme is construced between the two sets of biomes, the issue of how to quantify model skills against discrete, categorical data points can then be Chapter 4. Amazonian Dry Corridors opened by Fire and Low CO2 48

Figure 4.8: Diagram representing the biome assignment scheme. a) Division of cold and warm-hot biomes according to GDD and general organization of biomes according by fpc and height. b) Classification into more specific biomes by presence and dominance of pfts. c) Further classification of forests into seasonal and evergreen categories based on pft proportions. approached. Interpretations of pollen spectra can be subtle and vary between researchers. Re-evaluation of existing cores can have considerable affect on the features of ones’ biome reconstructions. D’Apolito (2017) [68] argues that South American forest pollen may dominate spectra due to higher pollen productivity and efficient dispersal. Thus, smaller abundances of Poaceae and other savanna-associated pollen taxa cannot necessarily exclude the presence of a savanna biome [123]. Similarly, a strong savanna signals from South American pollen spectra may suggest relatively large extensions of the savanna biome [69]. Underestimation of past savanna cover could further support lines of evidence for past expanses of savanna, such as the Dry Corridor/Arc hypotheses. Model-data comparison was perfomed using the

Discrete Manhattan Metric (DMM - see methods) applied to the four ﬁre/CO2 scenarios for ﬁve climate Chapter 4. Amazonian Dry Corridors opened by Fire and Low CO2 49

Biome Dense Sparse Tall Short Hot Cold Seasonal Evergreen Thf 1 0 1 0 1 0 0 1 Tdf 1 0 1 0 1 0 1 0 wtf 1 0 1 0 2/3 1/3 1/2 1/2 tef 1 0 1 0 1/3 2/3 0 1 tdf 1 0 1 0 1/3 2/3 1 0 bef 1 0 1 0 0 1 0 1 bdf 1 0 1 0 0 1 1 0 Ts 2/3 1/3 0 1 1 0 2/3 1/3 sw 2/3 1/3 0 1 2/3 1/3 1/2 1/2 tp 2/3 1/3 0 1 1/3 2/3 1/2 1/2 bp 2/3 1/3 0 1 0 1 1/2 1/2 g 1/3 2/3 0 1 1 0 1/2 1/2 d 0 1 0 1 1 0 1/2 1/2 st 1/3 2/3 0 1 0 1 1/2 1/2 t 0 1 0 1 0 1 1/2 1/2

Table 4.1: Aﬃnity matrix for LPX biomes to compute ‘distance’ between biomes in trait space. Thf = Tropical humid forest, Tdf = Tropical dry forest, wtf = warm temperate forest, tef = temperate evergreen forest, tdf = temperate deciduous forest, bef = boreal evergreen forest, bdf = boreal deciduous forest, Ts = Tropical savanna, sw = sclerophyll woodland, tp = temperate parkland, bp = boreal parkland, g = dry grass/shrubland, d = desert, st = shrub tundra, t = tundra

reconstruction inputs (four AOGCM climate reconstructions and one ensemble) for a total of twenty cases. The DMM characterizes biomes through bioclimatic and ecophysiological traits, permitting a quantiﬁcation scheme to rate the ‘distance’ between two biomes. The scores for each scenario represents the average distance, with smaller values indicating higher average agreement between model and pollen reconstructions.

4.5.3 Discrete Manhattan Metric

Discrete Manhattan Metric

The ‘Discrete Manhattan Metric’ (DMM) was developed to quantify the distance or ‘closeness’ between biomes, permitting direct comparison between model output and pollen records and a measure of overall model-data agreement. Biomes were characterized by a set of core ecological traits such as productivity, aridity, vegetation height, and temperature. Each biome has an affinity toward each trait, represented by a discrete number, resulting in an affinity matrix (fig. 3). Given two biomes, each having an affinity for each trait represented by a number between zero and one, we find their ‘ecological distance’ by: 1) calculating the difference in their affinity scores for each trait 2) summing the magnitude of these differences 3) normalizing by the total number of traits.

The xij element of affinity matrix, XMN , is the affinity for the ith biome to the jth trait for N biomes and M traits. The distance, d(a, b) between two biomes (indexed by a and b) is the sum of the differences in traits, normalized by the total number of traits N (eq. 4.1).

N X d(a, b) = |xaj − xbj|/N (4.1) j=1 Chapter 4. Amazonian Dry Corridors opened by Fire and Low CO2 50

Pollen Reconstructed Biomes Model Assigned Biomes Tropical Rainforest Tropical Humid Forest Tropical Seasonal Forest Tropical Dry Forest Cerrado Tropical Savanna Caatinga Dry Grass-Shrublands Steppe Desert Hot Desert

Table 4.2: Correspondence legend between pollen reconstructed and model assigned biomes.

In the case where the biomes are the same, their distance between them is equal to 0. If the biomes are maximally different (ex. tropical rainforest and tundra), their distance is equal to 1. This process would be repeated for every pollen-core site, where the modeled biome reconstruction for that point would be tested against the pollen-based reconstruction. Model performance is thus the mean of the scores for each pollen site, multiplied by two to remain consistent with properties of the continuous Manhattan Metric used in modern benchmarking [128]. A paired student t-test was performed on all 20 model-data comparisons to determine the likelihood of equality between mean scores with a p-value cutoff of 0.05. The correspondence between the LPX biome assignment scheme and pollen-based biomes were made by the author’s judgement, based on meta-analysis by Mayle et al. [161], Marchant et al. [153], and the original palaeoecological studies. A number of major pollen-reconstructed biome under examination had natural correspondances with model assigned biomes, while others were more subtle and open to interpretation (table 2). For example, tropical rainforest and dry forest from pollen studies had a natural correspondance to tropical humid forest and tropical dry forest in LPX, while the various reconstructions of open, non-analogue vegetation sites were more difficult to categorize in terms of model biomes. While correspondances may be rudimentary, the impact of errors in categorization would be softened by the design of the DMM, as opposed to a direct binary metric. Moreover, our study developed and applied a scheme to quantify model-data agreement between DGVM output and pollen data, which is rarely attempted despite its importance. Further details and references to original studies for each core site are located in the ‘core list’ in the appendix.

4.5.4 Stein-Alpert Decomposition

A Stein-Alpert decomposition was designed to compute isolated and synergistic effects of factors within numerical simulations [219] . Though initially developed for atmospheric models, this factor separation scheme can be adapted for climatic and non-climatic factors within vegetation models. In our decomposition, f0, f1, f2, and f12 are fields composed of tree cover outputs from the ensemble climate reconstruction for the control, low CO2 only, fire only, and fire and low CO2 (eq. 2-5) scenarios respectively.

f0 : ﬁre oﬀ, Pre-Industrial CO2 (4.2)

f1 : ﬁre oﬀ, LGM CO2 (4.3)

f2 : ﬁre on, PI CO2 (4.4)

f12 : ﬁre on, LGM CO2 (4.5) Chapter 4. Amazonian Dry Corridors opened by Fire and Low CO2 51

Effects from the factors of fire and CO2 are calculated by addition and subtraction of the fields. The isolated effect of fire is the difference between the tree cover from the fire only and the control scenarios

(eq. 6). Similarly, the isolated effect of low CO2 is the difference between the low CO2 only and control scenario (eq. 7). The simultaneous effect of both fire and low CO2 is represented by < f3 >, which is not a component of a formal Stein-Alpert decomposition (eq. 4.8). The synergistic effects of fire and

CO2 are computed by subtracting both the tree cover from the ﬁre only and low CO2 only scenarios from the ﬁre and low CO2 scenario while adding the tree cover from the control scenario (eq. 4.9).

< f1 >= f1 − f0 (4.6)

< f2 >= f2 − f0 (4.7)

< f3 >= f12 − f0 (4.8)

< f12 >= f12 − (f1 + f2) + f0 (4.9)

A logarithmic transformation was performed on the tree cover ﬁelds fi to convert the bounded variable of tree cover (ranging from 0 to 1) to an unbounded variables (ranging from −∞ to ∞). For a bounded variable y, which in our case are values of tree cover within each grid cells of a scenario, equation 1 and 2 are applied to create an unbounded transformed variabley2, which undergoes the arithmetic of the Stein

Alpert decomposition. Afterward the arithmetic is performed, equation 3 is used to transform y2 back to the initial bounded variable y.

y1 = (99y + 0.5)/100 (4.10)

y2 = log(y1/(1 − y1)) (4.11)

y = 2/(1 + e−y2 ) − 1.0 (4.12) Chapter 4. Amazonian Dry Corridors opened by Fire and Low CO2 52

Site Name Latititude Longitude Country Pollen Biome Reference Lake Patsucuarco 19.6 -101.58 Mexico 9 [233] Chalco Lake 19.5 -99 Mexico 9 [75] Lake Texcoco 19.4 -99 Mexico 9 [76] Lake Quexil 16.3 -89.9 Guatemala 9 [138] El Valle 8.43 -79.8 Panama 8 [188] La Chonta 8 -82 Costa Rica 9 [110] Fuquene II 5.45 -73.77 Colombia 3 [226] Agua Blanca 5 -74.45 Colombia 12 [?] Herrera 5 -73.9 Colombia 3 [227] El Pinal 4.1 -70.4 Colombia 8 [23] Timbio 2.4 -76.6 Colombia 3 [239] Lagoa Das Patas 0.26 -66.7 Brazil 1 [178] Lake Pata 0.26 -66.1 Brazil 1 [44] Lagoa de Caco -2.97 -43.3 Brazil 8 [144][74] GeoB 3104-1 -3.67 -37.7 Ocean 11 [26] Ciudad Universitaria -4.75 -74.2 Colombia 3 [224] Serra Sul Carajas -5 -49.5 Brazil 8 [99] Katira -9 -63 Brazil 8 [225] Lago do Saci -9.1 -56.3 Brazil 8 [86] Laguna Junin -11 -76.2 Peru 12 [95] Laguna Bella Vista -13.6 -61.56 Bolivia 3 [158] Consuelo -13.95 -68.9 Peru 8 [223] Chaplin -14.5 -61.1 Bolivia 12 [43] Aguas Emendadas -15 -47.6 Brazil 12 [16] Titicaca -16.1 -69.2 Bolivia/Peru 12 [171] Lake Huinamimarca -16.5 -69 Bolivia 12 [172] Crominia -17.3 -49.4 Brazil 8 [201] Wasa Mayu -17.54 -65.9 Bolivia 12 [?] Siberia -17.8 -64.7 Bolivia 12 [172] Salitre -19 -46.8 Brazil 12 [143] Serra Negra -18.95 -46.85 Brazil 8 [178] GeoB 3229-2 -19.63 -38.7 Brazil 12 [26] Salar de Uyuni -20 -68 Bolivia 12 [53] Catas Altas -20.1 -43.4 Brazil 12 [22] GeoB 3202-1 -21.6 -39.9 Brazil 12 [25] Morro de Itapeva -22.8 -45.5 Brazil 12 [21] Colonia -23.9 -46.7 Brazil 12 [145] Curucutu -23.9 -46.7 Brazil 12 [187] Volta Velha -26.1 -48.6 Brazil 12 [24] Cambara Sol -29.1 -50.1 Brazil 12 [27] Sao Francisco -29.6 -55.3 Brazil 12 [28]

Table 4.3: List of original palynological studies used in conjunction with meta-analyses by [153] and [161]. 1:tropical humid forest, 2:tropical dry forest, 3: warm temperate forest, 4: temperate evergreen forest, 5: temperate deciduous forest, 6: boreal evergreen forest, 7: boreal deciduous forest, 8: tropical savanna, 9: sclerophyll woodland, 10: temperate parkland, 11: boreal parkland, 12: dry grass/shrubland 13: hot desert, 14: shrub tundra, 15: tundra Chapter 4. Amazonian Dry Corridors opened by Fire and Low CO2 53

4.6 Acknowledgments

We would like to thank Patrick Bartlein for providing the LGM climate data sets. We thank Lucia G. Lohmann, Joel L. Cracraft, John M. Bates, and Hazuki Arakida for constructive discussions throughout the research process. Funding: This research was supported by the Natural Sciences and Engineering Research Council of Canada (NSERC) and is a contribution of the Dimensions of Biodiversity US-Biota SãoPaulo program through the Fundacãode Amparo àPesquisa do Estado de SãoPaulo (FAPESP 2012/50260-6) and NSF and NASA (NSF DEB 1241056). Chapter 5

Conclusion

The first project of my thesis (ch. 2) was a study that employed a canopy-scale model to simulate the carbon and energy fluxes of a tropical forest during the LGM to discern the effects of cooler air temperatures and low atmospheric carbon dioxide. Low CO2 was found to reduce modeled net ecosystem exchange significantly, while cooler air temperatures were found to slightly increase carbon uptake. The differential response of ecosystem photosynthesis and respiration to temperature was hypothesized to cause this increase in uptake. Further analysis of radiative transfer and diurnal flux trends suggest the potential importance of vertical resolution and leaf temperature in flux modeling within vegetation models. The cross-over point, a phenomena where leaf temperature drops below air temperature, was reproduced and attributed to sensitivity changes in the canopy energy balance and the partitioning of sensible and latent heat. In terms of future directions, the sensitivity of carbon uptake, pools, and ecosystem distribution to vertical resolution of the canopy should be tested at larger scales. These smaller-scale processes may help us understand processes such as forest robustness and vulnerability to climatic change. The ratio of sensible to latent heat, a quantity of key importance in ch. 2, is known as the Bowen ratio and can be monitored through measurement systems that are more cost-effective and labor intensive than eddy covariance systems. Occurence of the cross-over point, which signals a decrease in water use efficiency, may be easily measured by these systems. The second project of my thesis (ch. 3) was an analysis a four-year data set (2013-2016) of carbon fluxes from a tropical dry forest site from Santa Rosa National Park, Costa Rica. Net ecosystem exchange tended to increase during the onset of the rainy season toward a mid-season plateau, to then decline into senescence into the dry season. Mid wet season drought events, as caused by El Niño,were found to cause punctuated decreases in net carbon exchange during 2014 and 2015. A hyperbolic light response curve was fitted to the entire data set to decompose net ecosystem exchange into gross primary productivity and ecosystem respiration, and estimate rates of ecosystem-level saturated uptake and photosynthetic efficiency. Both photosynthetic efficiency and rates of saturated uptake were found to be highest in 2016, the ‘normal’ years following two years of consecutive drought. The maximum value for photosynthetic efficiency was compared to previous estimates utilized in DGVMs and proposed to be a stronger estimate for Central American dry forests. Bursts of carbon dioxide were released at the onset of the first rains, which is likely an example of the ‘Birch Effect’. To reliably determine annual sums, a storage term should be calculated and applied to the entire data set. I also recommend re-evaluation of basic representation of moisture-controlled ecosystems. The methods used in my study to extract ecosystem-

54 Chapter 5. Conclusion 55 level photosynthetic efficiency and saturated rates of uptake could be utilized to determine key variables for vegetation models, which should ideally use a consistent parameter derivation scheme. The existing data from my four-year study could also be used to derive an empirical model of carbon uptake suited for moisture-sensitive periods, where net ecosystem exchange would be represented as a function of soil moisture and radiation. The third and final project of my thesis (ch. 4) was the modeling of vegetation cover of the Neotrop- ics during the LGM, centred around a factorial experiment designed to discern the impacts of fire and

CO2 on biome distribution, tree cover, and model-data agreement. Fire and CO2 were found to increase model-data agreement individually and in concert, by reducing tree cover and increasing the presence of grassy biomes. Both factors were found to have significant impact in the competition between forest and savanna through the enhancement of the competitive advantage of grasses against trees. To estimate the net impact of including a vertically-resolved canopy and associated canopy cooling (ch. 2) in addition to the effects of low CO2 and fire, a similar but more complex modeling experiment would need to be performed. Distributions of grassy, open biomes were identified to resemble a number of biogeo- graphically significant vegetation formations such as the central and circum Amazonian ‘dry corridors’. A natural continuation would be a study that compares model reconstructions of vegetation against a comprehensive body of palynological evidence at the global scale, which would test the effects of both climatic and non-climatic factors on model-data agreement. The effects of climatic and non-climatic factors on model-data agreement could then be assessed using a similar methodology to ch. 4, leading to stronger assessments of Earth Systems model skill over long time-scales and large spatial-scales. In- tegration of model reconstructions with biological data, such phylogeographic data, could lead to more definite understanding of large-scale vegetation mediated diversification events. A consistent theme of my thesis was the effort to take a process-based approach to palaeoecology. A body of proxy records may provide strong evidence for past environmental conditions, but models can provide means to understand the interlocking mechanisms that connect Earth Systems together. This also applies to conceptual models, which quantitative models are descendents of. A process-based approach also facilitates transmission of results between palaeoecological, modern and future contexts of Earth Systems studies. Analysis of my model reconstructions was often grounded in modern ecology and forestry studies, while my results on forest stability may be relevant in future contexts. The most significant discovery of my thesis research may have been reconstruction of a number of past savanna formations, which may be most relevant in evolutionary biology. One major limitation of global vegetation models such as LPX is the usage of overly simplistic representations of ecosystems, particularly those that are controlled by moisture. Though I provided an improved estimate of a key photosynthetic parameter, I did not find a suitable mathematical expres- sion for ecosystem photosynthesis and respiration as a function of soil moisture. Additionally, global vegetation models typically classify ecosystems based on those that exist in temperate regions, that may not correspond well to certain tropical ecosystems that exist along the grassland, savanna, and dry forest spectrum. This is an on-going issue that was recognized but not resolved within my thesis. And though my analysis may be the most thorough model reconstruction of the Neotropics to date, there are still many roads to improvement. Most obviously, a larger number of pollen cores would allow for more statistically significant tests of model-data agreements and ultimately, more accurate vegetation reconstructions. Also, though I only tested the roles of fire and CO2 on savannafication explicitly, a number of other climatic (ex. seasonality) and non-climatic factors (ex. herbivory) may be important Chapter 5. Conclusion 56 drivers. Biologists commonly asked if it was possible to generate reliable vegetation reconstructions for periods older than the LGM, given that genetic evidence suggests diversification events may have occured long before the Pleistocene. This would be difficult given the combined rarity of pollen records in South America that date back prior to the LGM and model reconstructions of palaeoclimates besides the LGM and mid-Holocene. This is a natural limitation to my approach that would be difficult to circumvent. However, better understanding of vegetation-climate-CO2 relationships would still enable stronger reconstruction and projections. A number of projects emerged from my thesis work that expand my thesis work in different directions. A biogeographical analysis of the pectoral sparrow, a songbird well distributed in the Neotropics, has indicated an evolutionary history consistent with my theory on Pleistocene forest-savanna dynamics. I am working with ornithologists to better develop a theory of Neotropical diversification that is simultaneously consistent with genetic, palaeoecological, and modeling evidence. I am also working with anthropologist Dr. Michaela Ecker on testing the role of CO2 on the competition between C3 and C4 grasses, which was previously hypothesized based on isotope data from bovidae teeth. This would further examine the idea that CO2 mediates competition between vegetation types, which may further elucidate vegetation dynamics over geologic time. Lastly, I aim to perform a comprehensive modeling analysis of Canadian ecosystems under a changing climate, assessing the stability of forests to climatic shifts, wildfire, and CO2 fertilization. Though this would be focusing on the boreal rather than the tropics and near-future rather than deep past, it would be methodologically and conceptually similar to the reconstruction work of my thesis and could serve as a strong example of how palaeoecological research can enhance projection studies. Bibliography

[1] Adachi Minaco, Atsushi Ishida, Sarayudh Bunyavejchewin, Toshinori Okuda and Hiroshi Koizumi (2009) Spatial and temporal variation in soil respiration in a seasonally dry tropical forest, Thailand. Journal of Tropical Ecology, 25

[2] Alizadeh, K., Matthias, I., Rodr´ıguez-Zorro,P. A., Hermanowski, B., da Costa, M. L., & Behling, H. (2017). Forest-savanna boundary shift on the plateau of Serra Sul dos Caraj´as(southeastern Amazonia) since the mid-Holocene; driving forces and limiting factors. Quaternary International, 449, 12-21.

[3] Anderson, N. J., Bugmann, H., Dearing, J. A., & Gaillard, M. J. (2006). Linking palaeoenvironmental data and models to understand the past and to predict the future. Trends in ecology & evolution, 21(12), 696-704.

[4] Andrews, P. K., Chalmers, D. J., & Moremong, M. (1992). Canopy-air temperature diﬀerences and soil water as predictors of water stress of apple trees grown in a humid, temperate climate. Journal of the American Society for Horticultural Science, 117(3), 453-458.

[5] Antonelli, A., Zizka, A., Carvalho, F. A., Scharn, R., Bacon, C. D., Silvestro, D., & Condamine, F. L. (2018). Amazonia is the primary source of Neotropical biodiversity. Proceedings of the National Academy of Sciences, 201713819.

[6] Argollo, J., & Mourguiart, P. (2000). Late Quaternary climate history of the Bolivian Altiplano. Quaternary International, 72(1), 37-51.

[7] Arruda, D. M., Schaefer, C. E., Fonseca, R. S., Solar, R. R., & Fernandes-Filho, E. I. (2018). Vegetation cover of Brazil in the last 21 ka: New insights into the Amazonian refugia and Pleistocenic arc hypotheses. Global Ecology and Biogeography, 27(1), 47-56.

[8] Baldocchi D.D. (1994) An analytical solution for coupled leaf photosynthesis and stomatal conductance models. Tree Physiology 14, 1069–1079

[9] Baldocchi, D.D. and P.C. Harley. (1995). Scaling carbon dioxide and water vapor exchange from leaf to canopy in a deciduous forest: model testing and application. Plant, Cell and Environment. 18: 1157-1173.

[10] Baldocchi, D.D. (1997) Measuring and modelling carbon dioxide and water vapour exchange over a temperate broad-leaved forest during the 1995 summer drought. Plant, Cell and Environment 20, 1108–1122

57 Bibliography 58

[11] Baldocchi, D.D and T.P. Meyers. (1998). On using eco-physiological, micrometeorological and biogeochemical theory to evaluate carbon dioxide, water vapor and gaseous deposition ﬂuxes over vegetation. Agricultural and Forest Meteorology 90: 1-26.

[12] Baldocchi, D.D. and K.B.Wilson. (2001). Modeling CO2 and water vapor exchange of a temperate broadleaved forest across hourly to decadal time scales. Ecological Modeling 142: 155-184

[13] Baldocchi, D. D., Wilson, K. B., & Gu, L. (2002). How the environment, canopy structure and canopy physiological functioning inﬂuence carbon, water and energy ﬂuxes of a temperate broadleaved deciduous forest—an assessment with the biophysical model CANOAK. Tree Physiology, 22(15-16), 1065-1077.

[14] Baldocchi, D. D. (2003). Assessing the eddy covariance technique for evaluating carbon dioxide exchange rates of ecosystems: past, present and future. Global change biology, 9(4), 479-492.

[15] Baldocchi, D. D. (2005). The role of biodiversity on the evaporation of forests. In Forest Diversity and Function (pp. 131-148). Springer Berlin Heidelberg.

[16] Barberi, M., Salgado-Labouriau, M. L., & Suguio, K. (2000). Paleovegetation and paleoclimate of “Vereda de Aguas´ Emendadas”, central Brazil. Journal of South American Earth Sciences, 13(3), 241-254.

[17] Barneby, R. C. (1991). Sensitivae censitae: a description of the genus Mimosa Linnaeus (Mi- mosaceae) in the New World (Vol. 65).

[18] Baudena, M., Dekker, S. C., van Bodegom, P. M., Cuesta, B., Higgins, S. I., Lehsten, V., ... & Zavala, M. A. (2015). Forests, savannas, and grasslands: bridging the knowledge gap between ecology and Dynamic Global Vegetation Models. Biogeosciences, 12(6), 1833-1848.

[19] Beckage, B., & Ellingwood, C. (2008). Fire feedbacks with vegetation and alternative stable states. Complex Systems, 18(1), 159.

[20] Beerling, D. J., & Mayle, F. E. (2006). Contrasting eﬀects of climate and CO2 on Amazonian ecosystems since the last glacial maximum. Global Change Biology, 12(10), 1977-1984.

[21] Behling, H. (1997). Late Quaternary vegetation, climate and ﬁre history from the tropical mountain region of Morro de Itapeva, SE Brazil. Palaeogeography, Palaeoclimatology, Palaeoecology, 129(3-4), 407-422.

[22] Behling, H., & Lichte, M. (1997). Evidence of dry and cold climatic conditions at glacial times in tropical southeastern Brazil. Quaternary research, 48(3), 348-358.

[23] Behling, H., & Hooghiemstra, H. (1999). Environmental history of the Colombian savannas of the Llanos Orientales since the Last Glacial Maximum from lake records El Pinal and Carimagua. Journal of Paleolimnology, 21(4), 461-476.

[24] Behling, H., & Negrelle, R. R. (2001). Tropical rain forest and climate dynamics of the Atlantic lowland, Southern Brazil, during the Late Quaternary. Quaternary Research, 56(3), 383-389.

[25] Behling, H. (2002). South and southeast Brazilian grasslands during Late Quaternary times: a synthesis. Palaeogeography, Palaeoclimatology, Palaeoecology, 177(1-2), 19-27. Bibliography 59

[26] Behling, H., Arz, H. W., P¨atzold,J., & Wefer, G. (2002). Late Quaternary vegetational and climate dynamics in southeastern Brazil, inferences from marine cores GeoB 3229-2 and GeoB 3202-1. Palaeogeography, Palaeoclimatology, Palaeoecology, 179(3-4), 227-243.

[27] Behling, H., Pillar, V. D., Orlóci,L., & Bauermann, S. G. (2004). Late Quaternary Araucaria forest, grassland (Campos), fire and climate dynamics, studied by high-resolution pollen, charcoal and multivariate analysis of the Cambarádo Sul core in southern Brazil. Palaeogeography, Palaeoclimatology, Palaeoecology, 203(3-4), 277-297.

[28] Behling, H., Pillar, V. D., & Bauermann, S. G. (2005). Late Quaternary grassland (Campos), gallery forest, ﬁre and climate dynamics, studied by pollen, charcoal and multivariate analysis of the S˜ao Francisco de Assis core in western Rio Grande do Sul (southern Brazil). Review of palaeobotany and palynology, 133(3-4), 235-248.

[29] Birch, H. F. (1958). The eﬀect of soil drying on humus decomposition and nitrogen availability. Plant and soil, 10(1), 9-31.

[30] Blackie, R., Baldauf, C., Gautier, D., Gumbo, D., Kassa, H., Parthasarathy, N., ...& Sunderland, T. (2014). Tropical dry forests: The state of global knowledge and recommendations for future research (Vol. 2). CIFOR.

[31] Bonal, D., Burban, B., Stahl, C., Wagner, F., & H´erault,B. (2016). The response of tropical rainforests to drought—lessons from recent research and future prospects. Annals of forest science, 73(1), 27-44.

[32] Bonan, G. B. (2008). Forests and climate change: forcings, feedbacks, and the climate beneﬁts of forests. science, 320(5882), 1444-1449.

[33] Bond, W. J., & Midgley, G. F. (2000). A proposed CO2-controlled mechanism of woody plant invasion in grasslands and savannas. Global Change Biology, 6(8), 865-869.

[34] Bond, W. J., Midgley, G. F., & Woodward, F. I. (2003). The importance of low atmospheric CO2 and ﬁre in promoting the spread of grasslands and savannas. Global Change Biology, 9(7), 973-982.

[35] Bond, W. J., & Keeley, J. E. (2005). Fire as a global ‘herbivore’: the ecology and evolution of ﬂammable ecosystems. Trends in ecology & evolution, 20(7), 387-394.

[36] Bond, W. J., Woodward, F. I., & Midgley, G. F. (2005). The global distribution of ecosystems in a world without ﬁre. New phytologist, 165(2), 525-538.

[37] Bowman, D. M., Murphy, B. P., & Banfai, D. S. (2010). Has global environmental change caused monsoon rainforests to expand in the Australian monsoon tropics?. Landscape ecology, 25(8), 1247- 1260.

[38] Boza, M. A. (1993). Conservation in action: past, present, and future of the national park system of Costa Rica. Conservation Biology, 7(2), 239-247.

[39] Bragg, F. J., Prentice, I. C., Harrison, S. P., Eglinton, G., Foster, P. N., Rommerskirchen, F., & Rullk¨oetter,J. (2013). Stable isotope and modelling evidence for CO2 as a driver of glacial-interglacial vegetation shifts in southern Africa. Bibliography 60

[40] Braconnot, P., Otto-Bliesner, B., Harrison, S., Joussaume, S., Peterchmitt, J. Y., Abe-Ouchi, A., ... & Kageyama, M. (2007). Results of PMIP2 coupled simulations of the Mid-Holocene and Last Glacial Maximum–Part 1: experiments and large-scale features. Climate of the Past, 3(2), 261-277.

[41] Braconnot, P., Otto-Bliesner, B., Harrison, S., Joussaume, S., Peterchmitt, J. Y., Abe-Ouchi, A., ... & Kageyama, M. (2007). Results of PMIP2 coupled simulations of the Mid-Holocene and Last Glacial Maximum–Part 2: feedbacks with emphasis on the location of the ITCZ and mid-and high latitudes heat budget. Climate of the Past, 3(2), 279-296.

[42] Bueno, M. L., Pennington, R. T., Dexter, K. G., Kamino, L. H. Y., Pontara, V., Neves, D. M., ... & Oliveira-Filho, A. T. (2017). Eﬀects of Quaternary climatic ﬂuctuations on the distribution of Neotropical savanna tree species. Ecography, 40(3), 403-414.

[43] Burbridge, R. E., Mayle, F. E., & Killeen, T. J. (2004). Fifty-thousand-year vegetation and climate history of Noel Kempﬀ Mercado National Park, Bolivian Amazon. Quaternary Research, 61(2), 215- 230.

[44] Bush, M. B., De Oliveira, P. E., Colinvaux, P. A., Miller, M. C., & Moreno, J. E. (2004). Amazonian paleoecological histories: one hill, three watersheds. Palaeogeography, palaeoclimatology, palaeoecology, 214(4), 359-393.

[45] Bush, M. B., & Silman, M. R. (2004). Observations on Late Pleistocene cooling and precipitation in the lowland Neotropics. Journal of Quaternary Science, 19(7), 677-684.

[46] Bush, M. B., & Oliveira, P. E. D. (2006). The rise and fall of the Refugial Hypothesis of Amazonian speciation: a paleoecological perspective. Biota Neotropica, 6(1), 0-0.

[47] Bush, M.B. (2017). Climate science: The resilience of Amazonian forests. Nature, 541(7636), 167.

[48] Calvo, M., & Prentice, I. C. (2015). Eﬀects of ﬁre and CO2 on biogeography and primary production in glacial and modern climates. New Phytologist, 208(3), 987-994.

[49] Campbell, G. S., & Norman, J. M. (2012). An introduction to environmental biophysics. Springer Science & Business Media.

[50] Cardoso Da Silva, J. M., & Bates, J. M. (2002). Biogeographic Patterns and Conservation in the South American Cerrado : A Tropical Savanna Hotspot: The Cerrado, which includes both forest and savanna habitats, is the second largest South American biome, and among the most threatened on the continent. AIBS Bulletin, 52(3), 225-234.

[51] Carnaval, A. C., & Moritz, C. (2008). Historical climate modelling predicts patterns of current biodiversity in the Brazilian Atlantic forest. Journal of Biogeography, 35(7), 1187-1201.

[52] Castro, S. M., Sanchez-Azofeifa, G. A., & Sato, H. (2018). Eﬀect of drought on productivity in a Costa Rican tropical dry forest. Environmental Research Letters, 13(4), 045001.

[53] Chepstow-Lusty, A., Bush, M. B., Frogley, M. R., Baker, P. A., Fritz, S. C., & Aronson, J. (2005). Vegetation and climate change on the Bolivian Altiplano between 108,000 and 18,000 yr ago. Qua- ternary Research, 63(1), 90-98. Bibliography 61

[54] Chidumayo, E. N. (2010). The dry forests and woodlands of Africa: managing for products and services. Earthscan.

[55] Clapperton, C. M. (1993). Nature of environmental changes in South America at the Last Glacial Maximum. Palaeogeography, palaeoclimatology, palaeoecology, 101(3-4), 189-208.

[56] Clark, D. B., Mercado, L. M., Sitch, S., Jones, C. D., Gedney, N., Best, M. J., ... & Boucher, O. (2011). The Joint UK Land Environment Simulator (JULES), model description-Part 2: carbon ﬂuxes and vegetation dynamics. Geoscientiﬁc Model Development, 4(3), 701.

[57] Claussen, M., Selent, K., Brovkin, V., Raddatz, T., & Gayler, V. (2013). Impact of CO2 and climate on Last Glacial maximum vegetation–a factor separation. Biogeosciences, 10, 3593-3604.

[58] Cleland, E. E., Chuine, I., Menzel, A., Mooney, H. A., & Schwartz, M. D. (2007). Shifting plant phenology in response to global change. Trends in ecology & evolution, 22(7), 357-365.

[59] Colinvaux, P. A., De Oliveira, P. E., Moreno, J. E., Miller, M. C., & Bush, M. B. (1996). A long pollen record from lowland Amazonia: forest and cooling in glacial times. Science, 274(5284), 85.

[60] Colinvaux, P. A., De Oliveira, P. E., & Bush, M. B. (2000). Amazonian and neotropical plant communities on glacial time-scales: the failure of the aridity and refuge hypotheses. Quaternary Science Reviews, 19(1), 141-169.

[61] Collatz, GJ, Ball JT, Grivet, C, Berry, JA, (1991) Regulation of stomatal conductance and transpiration: a physiological model of canopy processes. Global Change Biology 5, 45–46.

[62] Costa, L. P. (2003). The historical bridge between the Amazon and the Atlantic Forest of Brazil: a study of molecular phylogeography with small mammals. Journal of Biogeography, 30(1), 71-86.

[63] Cowling, S. A., Maslin, M. A., & Sykes, M. T. (2001). Paleovegetation simulations of lowland Amazonia and implications for neotropical allopatry and speciation. Quaternary Research, 55(2), 140-149.

[64] Cowling, S. A. (2004). Tropical forest structure: a missing dimension to Pleistocene landscapes. Journal of Quaternary Science, 19(7), 733-743.

[65] Cowling, S. A., Cox, P. M., Jones, C. D., Maslin, M. A., Peros, M., & Spall, S. A. (2008). Simulated glacial and interglacial vegetation across Africa: implications for species phylogenies and trans- African migration of plants and animals. Global Change Biology, 14(4), 827-840.

[66] Cramer, W., Bondeau, A., Woodward, F. I., Prentice, I. C., Betts, R. A., Brovkin, V., ... & Kucharik, C. (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(4), 357-373.

[67] Craven, D., Dent, D., Braden, D., Ashton, M. S., Berlyn, G. P., & Hall, J. S. (2011). Seasonal variability of photosynthetic characteristics inﬂuences growth of eight tropical tree species at two sites with contrasting precipitation in Panama. Forest Ecology and Management, 261(10), 1643- 1653. Bibliography 62

[68] D’Apolito, C., Absy, M. L., & Latrubesse, E. M. (2013). The Hill of Six Lakes revisited: new data and re-evaluation of a key Pleistocene Amazon site. Quaternary Science Reviews, 76, 140-155.

[69] D’Apolito, C., Latrubesse, E. M., & Absy, M. L. (2018). Results confirm a relatively dry setting during the last glacial (MIS 3 and LGM) in Carajás,Amazonia: A comment on Guimarãeset al. The Holocene, 28(2), 330-331.

[70] Dai, Y., Dickinson, R. E., & Wang, Y. P. (2004). A two-big-leaf model for canopy temperature, photosynthesis, and stomatal conductance. Journal of Climate, 17(12), 2281-2299.

[71] Dang, Q. L., Margolis, H. A., Coyea, M. R., Sy, M., & Collatz, G. J. (1997) Regulation of branch- level gas exchange of boreal trees: roles of shoot water potential and vapor pressure diﬀerence. Tree Physiology, 17(8-9), 521-535.

[72] Dang, Q. L., Margolis, H. A., Sy, M., Coyea, M. R., Collatz, G. J., & Walthall, C. L. (1997) Proﬁles of photosynthetically active radiation, nitrogen and photosynthetic capacity in the boreal forest: Implications for scaling from leaf to canopy. Journal of Geophysical Research: Atmospheres, 102(D24), 28845–28859.

[73] Dang, Q. L., Margolis, H. A., & Collatz, G. J. (1998). Parameterization and testing of a coupled photosynthesis–stomatal conductance model for boreal trees. Tree physiology, 18(3), 141-153.

[74] D’Apolito, C., Absy, M. L., & Latrubesse, E. M. (2013). The Hill of Six Lakes revisited: new data and re-evaluation of a key Pleistocene Amazon site. Quaternary Science Reviews, 76, 140-155.

[75] del Socorro Lozano-Garcia, M., & Ortega-Guerrero, B. (1994). Palynological and magnetic suscep- tibility records of Lake Chalco, central Mexico. Palaeogeography, Palaeoclimatology, Palaeoecology, 109(2-4), 177-191.

[76] del Socorro Lozano-Garc´ıa, M., & Ortega-Guerrero, B. (1998). Late Quaternary environmental changes of the central part of the Basin of Mexico; correlation between Texcoco and Chalco basins. Review of Palaeobotany and Palynology, 99(2), 77-93.

[77] Dong, N., Prentice, I.C., Harrison, S., Song, Q., & Zhang, Y. (2016) Biophysical homeostasis of leaf temperature: a neglected process for vegetation and land-surface modelling. Global Ecology and Biogeography (under review)

[78] Evans, S., Dieckmann, U., Franklin, O., & Kaiser, C. (2016). Synergistic effects of diffusion and microbial physiology reproduce the Birch effect in a micro-scale model. Soil Biology and Biochem- istry, 93, 28-37. Farrick, K. K., & Branfireun, B. A. (2014). Soil water storage, rainfall and runoff relationships in a tropical dry forest catchment. Water Resources Research, 50(12), 9236-9250.

[79] Farquhar GD, von Caemmerer S, Berry JA (1980) A biochemical model of photosynthetic CO2

assimilation in leaves of C3 species. Planta, 149, 78–90

[80] Farrick, K. K., & Branﬁreun, B. A. (2013). Left high and dry: a call to action for increased hydrological research in tropical dry forests. Hydrological Processes, 27(22), 3254-3262.

[81] Fine, P. V., & Lohmann, L. G. (2018). Importance of dispersal in the assembly of the Neotropical biota. Proceedings of the National Academy of Sciences, 115(23), 5829-5831. Bibliography 63

[82] Fisher, R. A., Williams, M., Da Costa, A. L., Malhi, Y., Da Costa, R. F., Almeida, S., & Meir, P. (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(11), 2361-2378.

[83] Feng, X., Porporato, A., & Rodriguez-Iturbe, I. (2013). Changes in rainfall seasonality in the tropics. Nature Climate Change, 3(9), 811.

[84] Foken, T., Aubinet, M., & Leuning, R. (2012). The eddy covariance method. In Eddy Covariance (pp. 1-19). Springer, Dordrecht.

[85] Foley, J. A. (1995). An equilibrium model of the terrestrial carbon budget. Tellus B: Chemical and Physical Meteorology, 47(3), 310-319.

[86] Fontes, D., Cordeiro, R. C., Martins, G. S., Behling, H., Turcq, B., Sifeddine, A., ... & Rodrigues, R. A. (2017). Paleoenvironmental dynamics in South Amazonia, Brazil, during the last 35,000 years inferred from pollen and geochemical records of Lago do Saci. Quaternary Science Reviews, 173, 161-180.

[87] Gamfeldt, L., Sn¨all,T., Bagchi, R., Jonsson, M., Gustafsson, L., Kjellander, P., ... & Mikusi´nski, G. (2013). Higher levels of multiple ecosystem services are found in forests with more tree species. Nature communications, 4, 1340.

[88] Garzón-Orduña,I. J., Benetti-Longhini, J. E., & Brower, A. V. (2014). Timing the diversification of the Amazonian biota: butterfly divergences are consistent with Pleistocene refugia. Journal of biogeography, 41(9), 1631-1638.

[89] Gerten, D., Schaphoﬀ, S., Haberlandt, U., Lucht, W., & Sitch, S. (2004). Terrestrial vegetation and water balance—hydrological evaluation of a dynamic global vegetation model. Journal of Hydrology, 286(1-4), 249-270.

[90] Goulden, M. L., Miller, S. D., Da Rocha, H. R., Menton, M. C., de Freitas, H. C., e Silva Figueira, A. M., & de Sousa, C. A. D. (2004). Diel and seasonal patterns of tropical forest CO2 exchange. Ecological Applications, 14(sp4), 42-54.

[91] Haberle, S. G., & Maslin, M. A. (1999). Late Quaternary vegetation and climate change in the Amazon Basin based on a 50,000 year pollen record from the Amazon Fan, ODP Site 932. Quaternary Research, 51(1), 27-38.

[92] Haﬀer, J. (1969). Speciation in Amazonian forest birds. Science, 165(3889), 131-137.

[93] Haﬀer, J. R. (1997). Alternative models of vertebrate speciation in Amazonia: an overview. Biodi- versity & Conservation, 6, 451-476.

[94] Hanson P.J., Wullschleger S.D., Bohlman S.A. & Todd D.E. (1993) Seasonal and topographic pat-

terns of forest ﬂoor CO2 ﬂux from an upland oak forest. Tree Physiology 13, 1–15.

[95] Hansen, B. C. S., Rodbell, D. T., Seltzer, G. O., Leon, B., Young, K. R., & Abbott, M. (2003). Late- glacial and Holocene vegetational history from two sites in the western Cordillera of southwestern Ecuador. Palaeogeography, Palaeoclimatology, Palaeoecology, 194(1-3), 79-108. Bibliography 64

[96] Harrison, S. P., & Prentice, C. I. (2003). Climate and CO2 controls on global vegetation distribution at the last glacial maximum: analysis based on palaeovegetation data, biome modelling and palaeoclimate simulations. Global Change Biology, 9(7), 983-1004.

[97] Haxeltine, A., & Prentice, I. C. (1996). BIOME3: An equilibrium terrestrial biosphere model based on ecophysiological constraints, resource availability, and competition among plant functional types. Global Biogeochemical Cycles, 10(4), 693-709.

[98] He Liming, Jing M. Chen, Jan Pisek, Crystal B. Schaaf & Alan H. Strahler (2012) Global clumping index map derived from the MODIS BRDF product. Remote Sensing of Environment Volume 119, 16 April 2012, Pages 118-130, ISSN 0034-4257

[99] Guimar˜aes,J. T. F., Rodrigues, T. M., Reis, L. S., de Figueiredo, M. M. J. C., da Silva, D. F., Alves, R., ... & Sahoo, P. K. (2017). Modern pollen rain as a background for palaeoenvironmental studies in the Serra dos Caraj´as,southeastern Amazonia. The Holocene, 27(8), 1055-1066.

[100] Hermanowski, B., da Costa, M. L., & Behling, H. (2012). Environmental changes in southeastern Amazonia during the last 25,000 yr revealed from a paleoecological record. Quaternary Research, 77(1), 138-148.

[101] Hermanowski, B., Da Costa, M. L., & Behling, H. (2015). Possible linkages of palaeoﬁres in southeast Amazonia to a changing climate since the Last Glacial Maximum. Vegetation history and archaeobotany, 24(2), 279-292.

[102] Hector, A. (2011). Ecology: diversity favours productivity. Nature, 472(7341), 45-46.

[103] Hikosaka, K., Ishikawa, K., Borjigidai, A., Muller, O., & Onoda, Y. (2006). Temperature acclimation of photosynthesis: mechanisms involved in the changes in temperature dependence of photosynthetic rate. Journal of experimental botany, 57(2), 291-302.

[104] Hikosaka, K., & Anten, N. P. (2012). An evolutionary game of leaf dynamics and its consequences for canopy structure. Functional Ecology, 26(5), 1024-1032.

[105] Hirota, M., Holmgren, M., Van Nes, E. H., & Scheﬀer, M. (2011). Global resilience of tropical forest and savanna to critical transitions. Science, 334(6053), 232-235.

[106] Hoﬀmann, W. A., Jaconis, S. Y., Mckinley, K. L., Geiger, E. L., Gotsch, S. G., & Franco, A. C. (2012). Fuels or microclimate? Understanding the drivers of ﬁre feedbacks at savanna–forest boundaries. Austral Ecology, 37(6), 634-643.

[107] Hoﬀmann, W. A., Geiger, E. L., Gotsch, S. G., Rossatto, D. R., Silva, L. C., Lau, O. L., ... & Franco, A. C. (2012). Ecological thresholds at the savanna-forest boundary: how plant traits, resources and ﬁre govern the distribution of tropical biomes. Ecology letters, 15(7), 759-768.

[108] Holdridge, L. R. (1947). Determination of world plant formations from simple climatic data. Sci- ence, 105(2727), 367-368.

[109] Holdridge, L. R. (1967). Life zone ecology. Life zone ecology., (rev. ed.)). Bibliography 65

[110] Hooghiemstra, H., Cleef, A. M., Noldus, C. W., & Kappelle, M. (1992). Upper Quaternary vegetation dynamics and palaeoclimatology of the La Chonta bog area (Cordillera de Talamanca, Costa Rica). Journal of Quaternary Science, 7(3), 205-225.

[111] Hopcroft, P. O., & Valdes, P. J. (2015). Last glacial maximum constraints on the Earth System model HadGEM2-ES. Climate Dynamics, 45(5-6), 1657-1672.

[112] Hubbell, S. P. (1979). Tree dispersion, abundance, and diversity in a tropical dry forest. Science, 203(4387), 1299-1309.

[113] Huete, A. R., Didan, K., Shimabukuro, Y. E., Ratana, P., Saleska, S. R., Hutyra, L. R., ... & Myneni, R. (2006). Amazon rainforests green-up with sunlight in dry season. Geophysical research letters, 33(6).

[114] Huntingford, C., Fisher, R. A., Mercado, L., Booth, B. B., Sitch, S., Harris, P. P., ... & Harris, G. R. (2008). Towards quantifying uncertainty in predictions of Amazon ‘dieback’. Philosophical Transactions of the Royal Society B: Biological Sciences, 363(1498), 1857-1864.

[115] Hurrell, J. W., Holland, M. M., Gent, P. R., Ghan, S., Kay, J. E., Kushner, P. J., ... & Lipscomb, W. H. (2013). The community earth system model: a framework for collaborative research. Bulletin of the American Meteorological Society, 94(9), 1339-1360.

[116] Hutyra, L.R. 2007. Carbon and water exchange in Amazonian rainforests. Ph.D. Thesis, Depart- ment of Earth and Planetary Sciences, Harvard University, Cambridge, Massachusetts.

[117] Hutyra, L., S. Wofsy and S. Saleska. (2007) LBA-ECO CD-10 CO2 and H2O Eddy Flux Data at km 67 Tower Site, Tapajos National Forest. Data set. Available on-line [http://daac.ornl.gov] from Oak Ridge National Laboratory Distributed Active Archive Center, Oak Ridge, Tennessee, U.S.A. doi:10.3334/ORNLDAAC/860

[118] Hutyra, L. R., Munger, J. W., Saleska, S. R., Gottlieb, E., Daube, B. C., Dunn, A. L., ... & Wofsy, S. C. (2007). Seasonal controls on the exchange of carbon and water in an Amazonian rain forest. Journal of Geophysical Research: Biogeosciences, 112(G3).

[119] Janzen, D. H. (1988). Tropical Dry Forests: The Most Endangered Major Tropical Ecosystem. Wilson (ed.) Biodiversity.

[120] Jarvis, P., Rey, A., Petsikos, C., Wingate, L., Rayment, M., Pereira, J., ... & Manca, G. (2007). Drying and wetting of Mediterranean soils stimulates decomposition and carbon dioxide emission: the “Birch eﬀect”. Tree physiology, 27(7), 929-940.

[121] Johnson, M. A., Saraiva, P. M., & Coelho, D. (1999). The role of gallery forests in the distribution of Cerrado mammals. Revista Brasileira de Biologia, 59(3), 421-427.

[122] Jolly, D., & Haxeltine, A. (1997). Eﬀect of low glacial atmospheric CO2 on tropical African montane vegetation. Science, 276(5313), 786-788.

[123] Jones, H. G., & Rotenberg, E. (2011). Energy, radiation and temperature regulation in plants. eLS. Bibliography 66

[124] Jones, H. G. (2013). Plants and microclimate: a quantitative approach to environmental plant physiology. Cambridge university press.

[125] Jones, H. T., Mayle, F. E., Pennington, R. T., & Killeen, T. J. (2011). Characterisation of Bolivian savanna ecosystems by their modern pollen rain and implications for fossil pollen records. Review of Palaeobotany and Palynology, 164(3-4), 223-237.

[126] Kalacska, M., Sanchez-Azofeifa, G. A., Calvo-Alvarado, J. C., Quesada, M., Rivard, B., & Janzen, D. H. (2004). Species composition, similarity and diversity in three successional stages of a seasonally dry tropical forest. Forest ecology and management, 200(1-3), 227-247.

[127] Kattge J, Knorr W, Raddatz T, Wirth C. (2009) Quantifying photosynthetic capacity and its relationship to leaf nitrogen content for global-scale terrestrial biosphere models. Global Change Biology 15:976–991.

[128] Kelley, D. I., Prentice, I. C., Harrison, S. P., Wang, H., Simard, M., Fisher, J. B., & Willis, K. O. (2013). A comprehensive benchmarking system for evaluating global vegetation models. Biogeo- sciences, 10, 3313-3340.

[129] Kelley, D. I., Harrison, S. P., & Prentice, I. C. (2014). Improved simulation of ﬁre–vegetation interactions in the Land surface Processes and eXchanges dynamic global vegetation model (LPX- Mv1). Geoscientiﬁc Model Development, 7(5), 2411-2433.

[130] Kgope, B. S., Bond, W. J., & Midgley, G. F. (2010). Growth responses of African savanna trees implicate atmospheric [CO2] as a driver of past and current changes in savanna tree cover. Austral Ecology, 35(4), 451-463.

[131] Kricher, J. C. (1998) A Neotropical Companion: an Introduction to the Animals, Plants, and Ecosystems of the New World Tropics.

[132] Kricher, J. (2011). Tropical ecology. Princeton University Press.

[133] Kutsch, W. L., Hanan, N., Scholes, R. J., McHugh, I., Kubheka, W., Eckhardt, H., & Williams, C. (2008). Response of carbon ﬂuxes to water relations in a savanna ecosystem in South Africa. Biogeosciences Discussions, 5(3), 2197-2235.

[134] Langevelde, F., Van De Vijver, C. A., Kumar, L., Van De Koppel, J., De Ridder, N., Van Andel, J., ... & Prins, H. H. (2003). Eﬀects of ﬁre and herbivory on the stability of savanna ecosystems. Ecology, 84(2), 337-350.

[135] Lavin, M. (1994, October). Origin, diversity and biogeography of neotropical Fabaceae. In Pro- ceedings of the VI Congreso Latinoamericano de botanica: Mar del Plata, Argentina (pp. 2-8).

[136] Leakey, A. D., Ainsworth, E. A., Bernacchi, C. J., Rogers, A., Long, S. P., & Ort, D. R. (2009). Elevated CO2 eﬀects on plant carbon, nitrogen, and water relations: six important lessons from FACE. Journal of experimental botany, 60(10), 2859-2876.

[137] Lenton, T. M., & Lovelock, J. E. (2001) Daisyworld revisited: quantifying biological eﬀects on planetary self-regulation. Tellus b, 53(3), 288-305. Bibliography 67

[138] Leyden, B. W. (1984). Guatemalan forest synthesis after Pleistocene aridity. Proceedings of the National Academy of Sciences, 81(15), 4856-4859.

[139] Li, W., Cao, S., Campos-Vargas, C., & Sanchez-Azofeifa, A. (2017). Identifying tropical dry forests extent and succession via the use of machine learning techniques. International Journal of Applied Earth Observation and Geoinformation, 63, 196-205.

[140] Lin, Y. S., Medlyn, B. E., & Ellsworth, D. S. (2012). Temperature responses of leaf net photosynthesis: the role of component processes. Tree Physiology, 32(2), 219-31.

[141] Lloyd, J., & Farquhar, G. D. (2008). Eﬀects of rising temperatures and [CO2] on the physiology of tropical forest trees. Philosophical Transactions of the Royal Society of London B: Biological Sciences, 363(1498), 1811-1817.

[142] Lloyd, J., & Farquhar, G. D. (2008). Eﬀects of rising temperatures and [CO2] on the physiology of tropical forest trees. Philosophical Transactions of the Royal Society B: Biological Sciences, 363(1498), 1811-1817.

[143] Ledru, M. P., Braga, P. I. S., Soubi`es,F., Fournier, M., Martin, L., Suguio, K., & Turcq, B. (1996). The last 50,000 years in the Neotropics (Southern Brazil): evolution of vegetation and climate. Palaeogeography, Palaeoclimatology, Palaeoecology, 123(1-4), 239-257.

[144] Ledru, M. P., Cordeiro, R. C., Dominguez, J. M. L., Martin, L., Mourguiart, P., Sifeddine, A., & Turcq, B. (2001). Late-Glacial cooling in Amazonia inferred from pollen at Lagoa do Ca¸c´o,Northern Brazil. Quaternary Research, 55(1), 47-56.

[145] Ledru, M. P., Mourguiart, P., & Riccomini, C. (2009). Related changes in biodiversity, insolation and climate in the Atlantic rainforest since the last interglacial. Palaeogeography, Palaeoclimatology, Palaeoecology, 271(1-2), 140-152.

[146] Loew, A., van Bodegom, P. M., Widlowski, J. L., Otto, J., Quaife, T., Pinty, B., & Raddatz, T. (2014). Do we (need to) care about canopy radiation schemes in DGVMs? Caveats and potential impacts. Biogeosciences, 11, 1873-1897.

[147] Lugo, A. E., Gonzalez-Liboy, J. A., Cintron, B., & Dugger, K. (1978). Structure, productivity, and transpiration of a subtropical dry forest in Puerto Rico. Biotropica, 278-291.

[148] Luo, Y., Sims, D. A., Thomas, R. B., Tissue, D. T., & Ball, J. T. (1996). Sensitivity of leaf photosynthesis to CO2 concentration is an invariant function for C3 plants: A test with experimental data and global applications. Global Biogeochemical Cycles, 10(2), 209-222.

[149] Luo, Y. and Zhou, X. (2006) Soil Respiration and the Environment. Elsevier/Academic Press, Amsterdam

[150] Maass, J. M., Balvanera, P., Castillo, A., Daily, G. C., Mooney, H. A., Ehrlich, P., ... & Mart´ınez- Yrizar, A. (2005). Ecosystem services of tropical dry forests: insights from long-term ecological and social research on the Paciﬁc Coast of Mexico. Ecology and society, 10(1).

[151] Mahan, J. R., & Upchurch, D. R. (1988). Maintenance of constant leaf temperature by plants—I. Hypothesis-limited homeothermy. Environmental and Experimental Botany, 28(4), 351-357. Bibliography 68

[152] Malhi, Y., Baldocchi, D. D., & Jarvis, P. G. (1999). The carbon balance of tropical, temperate and boreal forests. Plant, Cell & Environment, 22(6), 715–740.

[153] Marchant, R., Cleef, A., Harrison, S. P., Hooghiemstra, H., Markgraf, V., Van Boxel, J., ... & Behling, H. (2009). Pollen-based biome reconstructions for Latin America at 0, 6000 and 18 000 radiocarbon years ago. Climate of the Past, 5, 725-767.

[154] Marchant, R., Boom, A., Behling, H., Hooghiemstra, H., Melief, B., Geel, B. V., ... & Wille, M. (2004). Colombian vegetation at the Last Glacial Maximum: a comparison of model-and pollen-based biome reconstructions. Journal of Quaternary Science, 19(7), 721-732.

[155] Marchant, R., Berr´ıo,J. C., Behling, H., Boom, A., & Hooghiemstra, H. (2006). Colombian dry moist forest transitions in the Llanos Orientales—a comparison of model and pollen-based biome reconstructions. Palaeogeography, Palaeoclimatology, Palaeoecology, 234(1), 28-44.

[156] Martin Calvo, M., & Prentice, I. C. (2015). Eﬀects of ﬁre and CO2 on biogeography and primary production in glacial and modern climates. New Phytologist, 208(3), 987-994.

[157] Mayer, A. L., & Khalyani, A. H. (2011). Grass trumps trees with ﬁre. Science, 334(6053), 188-189.

[158] Mayle, F. E., Burbridge, R., & Killeen, T. J. (2000). Millennial-scale dynamics of southern Ama- zonian rain forests. Science, 290(5500), 2291-2294.

[159] Mayle, F. E. (2004). Assessment of the Neotropical dry forest refugia hypothesis in the light of palaeoecological data and vegetation model simulations. Journal of Quaternary Science, 19(7), 713- 720.

[160] Mayle, F. E., Beerling, D. J., Gosling, W. D., & Bush, M. B. (2004). Responses of Amazonian ecosystems to climatic and atmospheric carbon dioxide changes since the last glacial maximum. Philosophical Transactions of the Royal Society of London. Series B: Biological Sciences, 359(1443), 499-514.

[161] Mayle, F. E., Burn, M. J., Power, M. and Urrego, D. H. (2009) Vegetation and ﬁre at the Last Glacial Maximum in tropical South America. In: Sylvestre, F., Vimeux, F. and Khodri, M. (eds.) Past climate variability in South America and surrounding regions: from the Last Glacial Maximum to the Holocene. Developments in paleoenvironmental research (14). Springer, New York, pp. 89-112. ISBN 9789048126729

[162] McMahon, S. M., Harrison, S. P., Armbruster, W. S., Bartlein, P. J., Beale, C. M., Edwards, M. E., ... & Prentice, I. C. (2011). Improving assessment and modelling of climate change impacts on global terrestrial biodiversity. Trends in Ecology & Evolution, 26(5), 249-259.

[163] McMahon, T.A., Peel M.C., Lowe L., Srikanthan, R., & McVicar, T.R. (2013) Estimating actual, potential, reference crop and pan evaporation using standard meteorological data: a pragmatic synthesis. Hydrology and Earth System Sciences 17 1331–1363

[164] Medina, E., & Klinge, H. (1983). Productivity of tropical forests and tropical woodlands. In Phys- iological plant ecology IV (pp. 281-303). Springer, Berlin, Heidelberg. Bibliography 69

[165] Mercado, L. M., Lloyd, J., Dolman, A. J., Sitch, S., & Patino, S. (2009). Modelling basin-wide variations in Amazon forest productivity–Part 1: Model calibration, evaluation and upscaling functions for canopy photosynthesis. Biogeosciences Discussions, 6(2), 2965-3030.

[166] Michaletz, S. T., Weiser, M. D., Zhou, J., Kaspari, M., Helliker, B. R., & Enquist, B. J. (2015). Plant thermoregulation: energetics, trait–environment interactions, and carbon economics. Trends in ecology & evolution, 30(12), 714-724.

[167] Miles, L., Newton, A. C., DeFries, R. S., Ravilious, C., May, I., Blyth, S., ... & Gordon, J. E. (2006). A global overview of the conservation status of tropical dry forests. Journal of Biogeography, 33(3), 491-505.

[168] Mogni, V. Y., Oakley, L. J., & Prado, D. E. (2015). The distribution of woody legumes in neotropical dry forests: the Pleistocene Arc Theory 20 years on. Edinburgh Journal of Botany, 72(1), 35-60.

[169] Moncrieﬀ J.B., Malhi Y., & Leuning R. (1996) The propagation of errors in long term measurements of land atmosphere ﬂuxes of carbon and water. Global Change Biology 2, 231–240.

[170] Monson, R., & Baldocchi, D. (2014). Terrestrial biosphere-atmosphere ﬂuxes. Cambridge Univer- sity Press.

[171] Mourguiart, P., Argollo, J., & Wirrmann, D. (1995). Evoluci´ondel lago Titicaca desde 25000 anos BP.

[172] Mourguiart, P., & Ledru, M. P. (2003). Last glacial maximum in an Andean cloud forest environment (Eastern Cordillera, Bolivia). Geology, 31(3), 195-198.

[173] Murphy, P. G., & Lugo, A. E. (1986). Ecology of tropical dry forest. Annual review of ecology and systematics, 17(1), 67-88.

[174] Nelson, B. W. (1994). Natural forest disturbance and change in the Brazilian Amazon. Remote Sensing Reviews, 10(1-3), 105-125.

[175] Ogren, E., & Evans, J. R. (1993). Photosynthetic light-response curves. Planta, 189(2), 182-190.

[176] O’ishi, R., & Abe-Ouchi, A. (2013). Inﬂuence of dynamic vegetation on climate change and terrestrial carbon storage in the Last Glacial Maximum.

[177] Oliger, E., Wellck, E., Kasahara, A., & Washington, M. (1970). Description of NCAR global circulation model.

[178] Oliveira PE (1992) A palynological record of Late Quaternary vegetational and climatic change in southeastern Brazil. PhD dissertation, Ohio State University, Colombus

[179] Oliveira, P. S., & Marquis, R. J. (2002). The cerrados of Brazil: ecology and natural history of a neotropical savanna. Columbia University Press.

[180] Owen, K. E., Tenhunen, J., Reichstein, M., Wang, Q., Falge, E., Geyer, R., ... & Aubinet, M. (2007). Linking ﬂux network measurements to continental scale simulations: Ecosystem carbon dioxide exchange capacity under non-water-stressed conditions. Global Change Biology, 13(4), 734-760. Bibliography 70

[181] Pacchioni, R. G., Carvalho, F. M., Thompson, C. E., Faustino, A. L., Nicolini, F., Pereira, T. S., ... & Agnez-Lima, L. F. (2014). Taxonomic and functional proﬁles of soil samples from Atlantic forest and Caatinga biomes in northeastern Brazil. MicrobiologyOpen, 3(3), 299-315.

[182] Palmer, T. N., & Williams, P. D. (2008). Introduction. Stochastic physics and climate modelling.

[183] Paw U K.T. (1987) Mathematical analysis of the operative temperature and energy budgets. Journal of Thermal Biology 12, 227–233.

[184] Peng, C., Guiot, J., Wu, H., Jiang, H., & Luo, Y. (2011). Integrating models with data in ecology and palaeoecology: advances towards a model–data fusion approach. Ecology Letters, 14(5), 522-536.

[185] Pennington, R. T., Prado, D. E., & Pendry, C. A. (2000). Neotropical seasonally dry forests and Quaternary vegetation changes. Journal of Biogeography, 27(2), 261-273.

[186] Pennington, R. T., Lavin, M., Prado, D. E., Pendry, C. A., Pell, S. K., & Butterworth, C. A. (2004). Historical climate change and speciation: neotropical seasonally dry forest plants show patterns of both Tertiary and Quaternary diversiﬁcation. Philosophical Transactions of the Royal Society of London B: Biological Sciences, 359(1443), 515-538.

[187] Pessenda, L. C. R., De Oliveira, P. E., Mofatto, M., de Medeiros, V. B., Garcia, R. J. F., Aravena, R., ... & Etchebehere, M. L. (2009). The evolution of a tropical rainforest/grassland mosaic in southeastern Brazil since 28,000 14 C yr BP based on carbon isotopes and pollen records. Quaternary Research, 71(3), 437-452.

[188] Piperno, D. R., Bush, M. B., & Colinvaux, P. A. (1991). Paleoecological perspectives on human adaptation in central Panama. I. The Pleistocene. Geoarchaeology, 6(3), 201-226.

[189] Powell, T. L., Galbraith, D. R., Christoﬀersen, B. O., Harper, A., Imbuzeiro, H. M., Rowland, L., ... & Levine, N. M. (2013). Confronting model predictions of carbon ﬂuxes with measurements of Amazon forests subjected to experimental drought. New Phytologist, 200(2), 350-365.

[190] Prado, D. E., & Gibbs, P. E. (1993). Patterns of species distributions in the dry seasonal forests of South America. Annals of the Missouri Botanical Garden, 902-927.

[191] Prentice, C., Guiot, J., Huntley, B., Jolly, D., & Cheddadi, R. (1996). Reconstructing biomes from palaeoecological data: a general method and its application to European pollen data at 0 and 6 ka. Climate Dynamics, 12(3), 185-194.

[192] Prentice, I. C., Harrison, S. P., & Bartlein, P. J. (2011). Global vegetation and terrestrial carbon cycle changes after the last ice age. New Phytologist, 189(4), 988-998.

[193] Prentice, I. C., Kelley, D. I., Foster, P. N., Friedlingstein, P., Harrison, S. P., & Bartlein, P. J. (2011). Modeling ﬁre and the terrestrial carbon balance. Global Biogeochemical Cycles, 25(3).

[194] Quijada-Mascare˜nasAdrian, J., Ferguson, J. E., Pook, C. E., Salomao, M. D. G., Thorpe, R. S., & W¨uster,W. (2007). Phylogeographic patterns of trans-Amazonian vicariants and Amazonian biogeography: the Neotropical rattlesnake (Crotalus durissus complex) as an example. Journal of Biogeography, 34(8), 1296-1312. Bibliography 71

[195] Raich, J. W., and Potter, C.S. (1995) Global patterns of carbon dioxide emissions from soils. Tellus 44B, 81–99.

[196] Ratter, J. A., Ribeiro, J. F., & Bridgewater, S. (1997). The Brazilian cerrado vegetation and threats to its biodiversity. Annals of botany, 80(3), 223-230.

[197] Richardson, J. E., Pennington, R. T., Pennington, T. D., & Hollingsworth, P. M. (2001). Rapid diversiﬁcation of a species-rich genus of neotropical rain forest trees. Science, 293(5538), 2242-2245.

[198] Rothermel, R. C. (1972). A mathematical model for predicting ﬁre spread in wildland fuels.

[199] Roy, J., Mooney, H. A., & Saugier, B. (Eds.). (2001). Terrestrial global productivity. Academic Press.

[200] Sage, R. F., Sharkey, T. D., & Seemann, J. R. (1989) Acclimation of photosynthesis to elevated CO2 in ﬁve C3 species. Plant Physiology, 89(2), 590-596.

[201] Salgado-Labouriau, M. L., Barberi, M., Ferraz-Vicentini, K. R., & Parizzi, M. G. (1998). A dry climatic event during the late Quaternary of tropical Brazil. Review of Palaeobotany and Palynology, 99(2), 115-129.

[202] Sanchez-Azofeifa, G. A., Quesada, M., Rodriguez, J. P., Nassar, J. M., Stoner, K. E., Castillo, A., ... & Fajardo, L. (2005). Research priorities for Neotropical dry forests 1. Biotropica: The Journal of Biology and Conservation, 37(4), 477-485.

[203] Sanchez-Azofeifa, A., Powers, J. S., Fernandes, G. W., & Quesada, M. (Eds.). (2013). Tropical dry forests in the Americas: ecology, conservation, and management. CRC Press.

[204] Sato, H., Itoh, A., & Kohyama, T. (2007). SEIB–DGVM: A new Dynamic Global Vegetation Model using a spatially explicit individual-based approach. Ecological Modelling, 200(3-4), 279-307.

[205] Sato, H., & Cowling, S. A. (2017). Glacial Amazonia at the canopy-scale: Using a biophysical model to understand forest robustness. Quaternary Science Reviews, 171, 38-47.

[206] Schmidt, K. P., & Inger, R. F. (1951). Amphibians and reptiles of the Hopkins-Branner expedition to Brazil. Chicago Natural History Museum.

[207] Scholes, R. J., & Archer, S. R. (1997). Tree-grass interactions in savannas. Annual review of Ecology and Systematics, 28(1), 517-544.

[208] Sellers, W. D. (1969). A global climatic model based on the energy balance of the earth-atmosphere system. Journal of Applied Meteorology, 8(3), 392-400.

[209] Sharpe, P. J., & DeMichele, D. W. (1977). Reaction kinetics of poikilotherm development. Journal of Theoretical Biology, 64(4), 649-670.

[210] Simon, E., Meixner, F. X., Rummel, U., Ganzeveld, L., Ammann, C., & Kesselmeier, J. (2005). Coupled carbon-water exchange of the Amazon rain forest, II. Comparison of predicted and observed seasonal exchange of energy, CO 2, isoprene and ozone at a remote site in Rondˆonia.Biogeosciences Discussions, 2(2), 399-449. Bibliography 72

[211] Sitch, S., Smith, B., Prentice, I. C., Arneth, A., Bondeau, A., Cramer, W., ... & Thonicke, K. (2003). Evaluation of ecosystem dynamics, plant geography and terrestrial carbon cycling in the LPJ dynamic global vegetation model. Global Change Biology, 9(2), 161-185.

[212] Sitch, S., Huntingford, C., Gedney, N., Levy, P. E., Lomas, M., Piao, S. L., ... & Jones, C. D. (2008). Evaluation of the terrestrial carbon cycle, future plant geography and climate-carbon cycle feedbacks using ﬁve Dynamic Global Vegetation Models (DGVMs). Global Change Biology, 14(9), 2015-2039.

[213] Smith, W. K., Vogelmann, T. C., & Critchley, C. (Eds.). (2010). Photosynthetic adaptation: chloroplast to landscape (Vol. 178). Springer Science & Business Media.

[214] Smith, R. S., & Gregory, J. (2012). The last glacial cycle: transient simulations with an AOGCM. Climate dynamics, 38(7-8), 1545-1559.

[215] Smith, N. G., & Dukes, J. S. (2013). Plant respiration and photosynthesis in global-scale models: incorporating acclimation to temperature and CO2. Global Change Biology, 19(1), 45-63.

[216] Sobol, M. K., & Finkelstein, S. A. (2018). Predictive pollen-based biome modeling using machine learning. PloS one, 13(8), e0202214.

[217] Spitters, C. J. T., Toussaint, H. A. J. M., & Goudriaan, J. (1986). Separating the diﬀuse and direct component of global radiation and its implications for modeling canopy photosynthesis Part I. Components of incoming radiation. Agricultural and Forest Meteorology, 38(1-3), 217-229.

[218] Staver, A. C., Archibald, S., & Levin, S. A. (2011). The global extent and determinants of savanna and forest as alternative biome states. Science, 334(6053), 230-232.

[219] Stein, U., & Alpert, P. I. N. H. A. S. (1993). Factor separation in numerical simulations. Journal of the Atmospheric Sciences, 50(14), 2107-2115.

[220] Sutherland, J. P. (1974). Multiple stable points in natural communities. The American Naturalist, 108(964), 859-873.

[221] Tamiya, H., Hase, E., Shibata, K., Mituya, A., Iwamura, T., Nihei, T., & Sasa, T. (1953). Kinetics of growth of Chlorella, with special reference to its dependence on quantity of available light and on temperature. Algal culture from laboratory to pilot plant, 204-32.

[222] Thonicke, K., Spessa, A., Prentice, I. C., Harrison, S. P., Dong, L., & Carmona-Moreno, C. (2010). The influence of vegetation, fire spread and fire behaviour on biomass burning and trace gas emissions: results from a process-based model. Biogeosciences, 7(6), 1991.

[223] Urrego, D. H., Bush, M. B., & Silman, M. R. (2010). A long history of cloud and forest migration from Lake Consuelo, Peru. Quaternary Research, 73(2), 364-373.

[224] Van der Hammen, T., & Gonz´alez,E. (1960). Upper pleistocene and holocene climate and vegetation of the “Sabana de Bogota”(Colombia, South America). Leidse Geologische Mededelingen, 25(1), 261-315.

[225] Van der Hammen, T., & Absy, M. L. (1994). Amazonia during the last glacial. Palaeogeography, palaeoclimatology, palaeoecology, 109(2-4), 247-261. Bibliography 73

[226] van der Hammen, T., & Hooghiemstra, H. (2003). Interglacial–glacial Fuquene-3 pollen record from Colombia: an Eemian to Holocene climate record. Global and Planetary Change, 36(3), 181-199.

[227] Van Geel, B., & Van der Hammen, T. (1973). Upper Quaternary vegetational and climatic se- quence of the Fuquene area (Eastern Cordillera, Colombia). Palaeogeography, Palaeoclimatology, Palaeoecology, 14(1), 9-92.

[228] Vanzolini, P. E. (1974). Ecological and geographical distribution of lizards in Pernambuco, northeastern Brasil (Sauria). Museu de Zoologia, Universidade de S¯aoPaulo.

[229] Verbeeck, H., & Kearsley, E. (2016). The importance of including lianas in global vegetation models. Proceedings of the National Academy of Sciences, 113(1), E4-E4.

[230] Verduzco, V. S., Garatuza-Payán,J., Yépez, E. A., Watts, C. J., Rodr´ıguez,J. C., Robles-Morua, A., & Vivoni, E. R. (2015). Variations of net ecosystem production due to seasonal precipitation differences in a tropical dry forest of northwest Mexico. Journal of Geophysical Research: Biogeo- sciences, 120(10), 2081-2094.

[231] Wang, X., Edwards, R. L., Auler, A. S., Cheng, H., Kong, X., Wang, Y., ... & Chiang, H. W. (2017). Hydroclimate changes across the Amazon lowlands over the past 45,000 years. Nature, 541(7636), 204.

[232] Waring, B. G., & Powers, J. S. (2016). Unraveling the mechanisms underlying pulse dynamics of soil respiration in tropical dry forests. Environmental Research Letters, 11(10), 105005.

[233] Watts, W. A., & Bradbury, J. P. (1982). Paleoecological Studies at Lake Patzcuaro on the West- Central Mexican Plateau and at Chalco in the Basin of Mexico 1. Quaternary Research, 17(1), 56-70.

[234] Weiss, A., & Norman, J. M. (1985). Partitioning solar radiation into direct and diﬀuse, visible and near-infrared components. Agricultural and Forest meteorology, 34(2-3), 205-213.

[235] Werneck, F. P., & Colli, G. R. (2006). The lizard assemblage from Seasonally Dry Tropical Forest enclaves in the Cerrado biome, Brazil, and its association with the Pleistocenic Arc. Journal of Biogeography, 33(11), 1983-1992.

[236] Werneck, F. P. (2011). The diversiﬁcation of eastern South American open vegetation biomes: historical biogeography and perspectives. Quaternary Science Reviews, 30(13-14), 1630-1648.

[237] Werneck, F. P., Gamble, T., Colli, G. R., Rodrigues, M. T., & Sites Jr, J. W. (2012). Deep diversi- ﬁcation and long term persistence in the South American ‘Dry Diagonal’: Integrating continent-wide phylogeography and distribution modeling of geckos. Evolution, 66(10), 3014-3034.

[238] Whittaker, R. H., & Klomp, H. (1975). The design and stability of plant communities. In Unifying concepts in ecology (pp. 169-183). Springer, Dordrecht.

[239] Wille, M., Negret, J. A., & Hooghiemstra, H. (2000). Paleoenvironmental history of the Popay´an area since 27 000 yr BP at Timbio, southern Colombia. Review of Palaeobotany and Palynology, 109(1), 45-63. Bibliography 74

[240] Williams, M., Rastetter, E. B., Fernandes, D. N., Goulden, M. L., Wofsy, S. C., Shaver, G. R., ... & Nadelhoﬀer, K. J. (1996). Modelling the soil-plant-atmosphere continuum in a Quercus–Acer stand at Harvard Forest: the regulation of stomatal conductance by light, nitrogen and soil/plant hydraulic properties. Plant, Cell & Environment, 19(8), 911-927.

[241] Wuster, W., Ferguson, J. E., Quijada-Mascare˜nas,J. A., Pook, C. E., Da Graca Salomao, M., & Thorpe, R. S. (2005). Tracing an invasion: landbridges, refugia, and the phylogeography of the Neotropical rattlesnake (Serpentes: Viperidae: Crotalus durissus). Molecular Ecology, 14(4), 1095- 1108.

[242] Zhu, D., Ciais, P., Chang, J., Krinner, G., Peng, S., Viovy, N., ... & Zimov, S. (2018). The large mean body size of mammalian herbivores explains the productivity paradox during the Last Glacial Maximum. Nature ecology & evolution, 2(4), 640.