ABSTRACT JUST, MICHAEL G. on Fire and Vegetation Structure

ABSTRACT

JUST, MICHAEL G. On Fire and Vegetation Structure along Longleaf Pine Savanna – Wetland Ecotonal Gradients. (Under the direction of William A. Hoffmann).

Vegetation structure (i.e. species composition and physical structure) has an integral role in many ecological phenomena, including biodiversity, carbon storage, and microclimate.

Vegetation structure distributions are used to demarcate ecosystems boundaries; as such, understanding what controls these distributions is of ecological consequence. For example, vegetation structure defines the boundaries between fire-promoting (e.g. open-canopied) and fire-deterring (e.g. closed-canopy) ecosystems; these systems remain distinct through positive fire-vegetation feedbacks. I examined vegetation structure properties along fire- managed longleaf pine savanna – wetland ecotonal gradients in North Carolina, USA to better understand vegetation boundaries between fire-promoting and fire-deterring communities.

Fire only directly influences vegetation it actually burns; therefore, determining where fires stop between frequently and infrequently burned ecosystems is important for understanding vegetation boundary dynamics. To this end, I developed an empirically-based statistical model to predict flammability along the ecotonal gradient using vegetation and microclimate predictors. I found that C4 grass and photosynthetically active radiation promoted fire, and that evergreen shrubs deterred fire, and that vegetation was the most important factor for predicting flammability. My data suggests that the feedback for these gradients might have less control on vegetation boundaries as compared to other fire- frequency gradients. These results provide useful information on where fires extinguish along these ecotonal gradients, allowing better predictions to be made concerning how environmental change might affect vegetation structure distribution. Next, using this statistical model of flammability, I developed a cellular automaton to model fire spread and vegetation dynamics in response to variations in fire frequency. I found that vegetation structure quickly rebounded from fire events. This resilience suggests that the mechanism controlling vegetation distributions along this gradient includes factors beyond just fire (e.g. hydrology). However, I found a negative relationship between fire frequency and mean fire return interval, suggesting that prolonged periods of altered fire frequencies could result in vegetation structure boundary movement.

Then, I considered factors affecting the persistence of resprouting trees in frequently burned ecosystems. Some resprouting tree species that can coexist are instead partitioned between areas of high or low fire return intervals. Persistence of these individuals may be determined by their capacity to limit wood decay in belowground storage organs after fire- related trauma. I investigated the extent of root crown decay between a fire-resistant tree species and four species restricted to less frequently burned positions along the savanna – wetland gradient. I found that the fire-resistant species had significantly less decay, and I identified the greatest values for traits that confer decay resistance (wood density, lignin, extractable phenolics), as compared to other species. Persistence in frequently burned ecosystems is often attributed to the ability of resprouting woody plants to store and remobilize belowground carbohydrate reserves after fire damage; this work provides evidence to support a possibly overlooked alternative explanation of persistence: wood decay resistance.

Finally, I examined the invasibility of this fire-managed gradient by non-native, woody species. A successful savanna invasion by woody individuals can alter both vegetation structure and fire behavior – begetting a denser, less-frequently burned system – at the expense of open vegetation structure. I set-up an experiment to test the effects of fire and site conditions on the establishment of six non-native woody species along this gradient. For unburned sites, three species had individuals survive until the end of the study, and only one species had individuals survive fire after seedling emergence. These results suggest, that while resistant, this landscape is not immune to invasion by woody species, nor to the possible impacts they can have on vegetation structure.

by Michael G. Just

A dissertation submitted to the Graduate Faculty of North Carolina State University in partial fulfillment of the requirements for the degree of Doctor of Philosophy

Plant Biology

Raleigh, North Carolina

2016

APPROVED BY:

______William A. Hoffmann Ryan E. Emanuel Committee Chair

______Kevin Gross Thomas R. Wentworth

DEDICATION

For my mom, may she rest in peace.

BIOGRAPHY

Michael Just was born and raised in Chicagoland. He likes science. Since Michael has been aware of the concept of ‘working’, he has wanted to be a scientist. At that time, Michael thought scientists used beakers, made potions, and discovered things. Despite growing up in a very developed landscape, the neighborhoods where Michael lived were old enough to have mature trees. However, his interest in what he would later know as ecology was initially fostered by at least two distinct experiences.

One was his grandfather’s dairy farm in central Wisconsin, which served as the most

‘natural’ place he knew for quite some time. The farm provided time and space for out-of- doors discovery and reflection. Second, and long-ago in the time of Hypercolor T-Shirts, the content of Weekly Readers was rich with environmental science topics, such as recycling and concern for tropical deforestation and, importantly, these topics held Michael’s attention better than others. While these experiences may have been a bit obtuse, they allowed an adolescent Michael’s naïve ecological concerns to mature into a career path that has allowed him to remain (nearly) as quizzical as then.

Michael’s academic pursuits thus far have resulted in a bachelor’s and master’s degree from the University of Illinois in Natural Resources and Environmental Sciences.

Some of his pre-doctoral research experiences included projects involving agricultural pests, endangered bats, rare plants, aquatic invasive species, and prairie restoration. He currently lives in Raleigh, NC with his fiancée and their two dogs, and the occasional foster turtle.

iii

ACKNOWLEDGMENTS

First, I would like to acknowledge my family for convincing me to attend college. Little did I know that I would really enjoy it. Thank you.

I must extended thanks to my advisor, Dr. William A. Hoffmann, for providing me the opportunity to pursue this doctoral degree. Throughout the degree, Bill has not only provided me guidance, but also the latitude to make this dissertation my own. I would also like to thank my graduate advisory committee – Dr. Ryan E. Emanuel, Dr. Kevin Gross, and

Dr. Thomas R. Wentworth – for their support through this endeavor.

Thank you to my laboratory members, past and present, and my fellow graduate students in this department and others, whose camaraderie, support, and insights have contributed to my success. I would like to specifically acknowledge, Bradley Breslow, Dr.

Stephanie Hollingsworth, Dr. Alice Broadhead, Dr. Jennifer Schafer, Jacob Norton, Dr. Tim

Antonelli, Tyson Wepperich, Megan Thoemmes, Emily Meineke, Sean Giery, Dr. Rene

Marchin, Dr. Wade Wall, and Wyatt Sanders.

I would have not been able to complete these projects without the assistance of many fine laboratory and field assistants: A. Yousef Abuahmad, Alicia Ballard, Spencer Bell,

Samantha Byerley, Katie Fraboni, Camera Hedin, DeAnna Metiver, Ashley McGuigan,

Spencer Goyette, Patrick Mulvaney, Anna Parot, Megan Walz, and Anthony Whitehead.

Thank you.

Thank you to Dr. Robert R. Dunn for the additional mentorship and research opportunities. I am grateful for the fellowship and opportunities provided by the Southeast

Climate Science Center and its staff. Thank you to Janet Gray (Fort Bragg Endangered

Species Branch) for making sure I had access to my field sites, among other things. I must also thank my MS advisor, and continuing collaborator, Matthew G. Hohmann, for his initial and sustained support of my scientific pursuits.

I also owe thanks the Department of Plant and Microbial Biology staff, especially Sue

Vitello for shepherding me through all the details.

I’d like to acknowledge the following musical artists, who were the top 10 artists

(based on play time) providing the soundtrack to my dissertation: Aphex Twin, Chance the

Rapper, Prefuse 73, Tycho, Philip Glass, Ratatat, Tortoise, DJ Shadow, King Tubby, and DJ

Rashad.

A thank you to my friends for championing my scholastic and scientific endeavors and providing feedback and comic relief when needed.

Finally, thank you Lexie, you’ve been a great proofreader, scientific illustrator, and a very patient partner.

TABLE OF CONTENTS

LIST OF TABLES ...... viii LIST OF FIGURES ...... x Chapter 1: Introduction ...... 1 Chapter 2: Where fire stops: vegetation structure and microclimate influence fire spread along an ecotonal gradient ...... 12 Acknowledgements ...... 12 Abstract ...... 13 Introduction ...... 14 Methods...... 15 Results ...... 24 Discussion ...... 27 References ...... 33 Tables ...... 40 Figures...... 44 Chapter 3: Effects of fire frequency on flammability and vegetation structure along managed longleaf pine savanna – wetland gradients ...... 49 Abstract ...... 49 Introduction ...... 51 Materials and Methods ...... 54 Results ...... 61 Discussion ...... 65 References ...... 70 Figures...... 76 Chapter 4: Wood decay resistance and the persistence of resprouting, woody plants in pyrophilic ecosystems ...... 86 Abstract ...... 86 Introduction ...... 88 Materials and Methods ...... 90 Results ...... 95 References ...... 100 Tables ...... 106

Figures...... 108 Chapter 5: Invasibility of a fire-managed savanna-wetland ecotone by non-native, woody plant species ...... 111 Abstract ...... 111 Introduction ...... 112 Materials and methods ...... 115 Results ...... 120 Discussion ...... 125 References ...... 130 Tables ...... 137 Figures...... 143 Chapter 6: Summary ...... 148 Appendices ...... 151 Appendix A: Chapter 2 Online Resources ...... 152 Appendix B: Chapter 3 Supplementary Figures ...... 161 Appendix C: Chapter 5 Supplementary Tables ...... 169

vii

LIST OF TABLES

Table 2.1 Vegetation Structure functional types, microclimate, plot, and transect variables . 40 Table 2.2 Results of linear mixed-effect models evaluating vegetation structure functional type, or plot variable by relative gradient position ...... 41 Table 2.3 Comparison of savanna and wetland microclimate variable means (Welch’s unequal variance t-test). Collection period was May – August 2012 – 2014 ...... 42 Table 2.4 Model details for predicting the conditional probability of fire spread of a plot (i.e. the preceding plot within the transect was burned). Predictor with coefficient direction, marginal R2, variance influence factor (√VIF score), relative importance weights, and AIC score ...... 43 Table 4.1 Characteristics of the individuals included in the study. Ranges for height, basal diameter, and pre-coppicing volume represent the minimum and maximum mean value ...... 106 Table 4.2 Decay proportion (mean proportion of discolored wood volume to total wood volume to a belowground depth of 5.0 cm) for the five study species. Results of exact Wilcoxon-Mann-Whitney tests comparing decay proportion between 9- and 19-month harvest groups for each species are also presented ...... 107 Table 5.1 Woody species used in this experiment ...... 137 Table 5.2 Mean emergence, survival, and establishment for our six study species. Emergence is defined as the proportion of seedlings to seeds sown at the end of the first growing season. Two-year survival is defined as the proportion of emergent seedlings that survived until the end of a 2-year period beginning at the end of the first growing season. Establishment is defined as the proportion of seedlings at the end of the study period to seeds sown ...... 138 Table 5.3 Type II Wald χ2 test statistic values from generalized linear mixed-effect models (GLMM) evaluating the effect of vegetative community, seed-sowing treatment, or species identity on germination, survival, and emergence for our six study species and all species combined. Emergence is the proportion of seedlings to seeds sown at the end of the first growing season. Survival is the proportion of emergent seedlings that survived until the end of a 2-year period beginning at the end of the first growing season. Establishment is the proportion of seedlings at the end of the study period (Sept. 2015) to seeds sown...... 139 Table 5.4 Number and proportion (seedlings / seeds sown) of established seedlings by seed- sowing treatment (undisturbed [UD], litter removal and undisturbed soil [LRU], litter removal and covered with loose soil [LRS]) in unburned and burned sites ...... 140 Table 5.5 Mean soil properties and environmental variables of the study plots (Welch’s unequal variance t-test) ...... 141

viii

Table 5.6 Mean survival rates (proportion of seedlings that survived from beginning until the end of the survival period) considering all species ...... 142

LIST OF FIGURES

Fig. 2.1 Conceptual model of the vegetation structure along the savanna – wetland ecotonal gradient with a diagram of transect and plot layout. Each transect begins in savanna and ends in wetland and consists of 1 m2 plots (black squares) spaced every 10 m. Transect length is variable and is determined by the length of the natural gradient. Dashed lines represent an impermeable clay soil layer resulting in a perched water table ...... 44 Fig. 2.2 Mean cover (%) by relative gradient position for the C4 grass and evergreen shrub functional types included in our vegetation structure and microclimate fire spread probability model. Error bars are standard errors of the mean ...... 45 Fig. 2.3 Violin plots for mean understory photosynthetically active radiation (PAR) and mean coarse fuel moisture (%) between savanna and wetland. The box plots indicate data range, quartiles, and median. Dots are outliers. The density trace is shown in light gray. Data represent the mean values for burn days during the growing season (April – Sept, 10:00-16:00)...... 46 Fig. 2.4 (a) Expected fire return interval (FRI) by relative gradient position. (b) Proportion of plots (n=489) burned per relative gradient position. The dashed line represents the FRI for gradients with a shrub-dominant (shrubby) wetland and the solid line represents those with an herbaceous wetland ...... 47 Fig. 2.5 Lines are predictions of the conditional probability of a plot burning using the vegetation structure and microclimate generalized linear mixed model (Table 2.4). For each plot all predictors were held constant at their mean except for the variable listed on the horizontal axis ([a] C4 grass cover, [b] evergreen shrub cover, [c] photosynthetically active radiation), which varied ...... 48 Fig. 3.1 Lines are predictions of the conditional probability of a plot burning using a generalized linear model with C4 grass and evergreen shrub volume as predictors. For C4 grass (solid brown line) predictions, evergreen shrub volume was held constant at 3 zero, while C4 grass volume varied (0 – 2 m ), and vice versa for evergreen shrub (dashed green line) predictions. Vegetation type volume is represented on the horizontal axis as the proportion of a 2 m3 cell (i.e. vegetation volume / 2 m3) ...... 76 Fig. 3.2 Conceptual diagram of the stochastic cellular automaton model used in this study. The top row is a representative lattice (1x5 cells) used in our simulations of fire spread and vegetation response to burn status along longleaf pine savanna – wetland ecotonal gradients. The second row represents a time step in which a fire was initiated (Fires were initiated every 1, 3, or 9 years, or never). Fires were initiated in the savanna cell of the lattice and could spread downslope towards wetland. Fire spread was conditional, and a cell could only burn if the preceding, upslope cell burned. A probability of burning [P(fire)] was assigned to each cell based on a binomial logistic regression (developed with field-collected data) that used C4 Grass and evergreen

shrub volume as predictors. In this example, the first three (orange) of five cells burned. After each time step of the simulation, the vegetation structure volume for both types is updated (third row) based on burn status (i.e. burned, unburned [blue]). The probability of burning is then updated based on the new vegetation structure values. The process then continues on the updated lattice (fourth row in the diagram) until 200 time steps are reached ...... 77

Fig. 3.3 Field-collected (2012) mean values by transect and relative gradient position for C4 grass and evergreen shrub volume (C4 grass [solid brown line], evergreen shrub [dashed green line]; 100 transects comprised of 460 2 m3 plots). (A) The mean proportion of a plot occupied by vegetation structure type (e.g. C4 grass or evergreen shrub). (B) The mean proportion of plots by vegetation structure type with non- 3 3 negligible volume (i.e. C4 grass > 0.004 m , evergreen shrub > 0.019 m ). We used these gradients as templates for our model simulations...... 78

Fig. 3.4 The mean proportion of a cell occupied by either vegetation type (C4 grass [solid brown line], evergreen shrub [dashed green line]) type by gradient position and arranged by vegetation initialization scheme (columns) by fire frequency (rows). The proportion of cell occupation was calculated as the mean across lattices (n = 100) by cell gradient position. Cells in the Field 2012 scheme were initiated with our observed (2012) vegetation structure values. Cells in the High Grass scheme were initiated with 3 3 a high value for C4 grass (0.25 m ) and a low value for evergreen shrub (0.025 m ), and vice versa for the lattice cells of the High Shrub scheme. This plot depicts the results for our model instances that were simulated with field-based rules for updating flammability and vegetation response to burn status ...... 79

Fig. 3.5 The mean proportion of plots by vegetation type (C4 grass [solid brown line], evergreen shrub [dashed green line]) with non-negligible volume (i.e. C4 grass > 0.004 m3, evergreen shrub > 0.019 m3) by gradient position and arranged by vegetation structure initialization scheme (columns) by fire frequency (rows). The proportion of cells with non-negligible volume was the mean across lattices (n = 100) by cell gradient position. Cells in the Field 2012 scheme were initiated with our observed (2012) vegetation structure values. Cells in the High Grass scheme were 3 initiated with a high value for C4 grass (0.25 m ) and a low value for evergreen shrub (0.025 m3), and vice versa for the lattice cells of the High Shrub scheme. This plot depicts the results for our model instances that were simulated with field-based rules for updating flammability and vegetation response to burn status ...... 80 Fig. 3.6 The mean fire return interval (years) by simulation instance (vegetation structure initialization scheme, updating rules) by fire frequency (1-yr [solid orange line], 3-yr [dashed purple line], 9-yr [dotted green line]). The mean fire return interval was calculated as the mean across lattices (n = 100) by cell gradient position. Fire was initiated in the savanna gradient position of each lattice every 1, 3, or 9 time steps. The cells of the Field 2012 schemes were initiated with our observed (2012) vegetation structure values. The cells of the High Grass schemes were initiated with a 3 3 high value for C4 grass (0.25 m ) and a low value for evergreen shrub (0.025 m ), and

vice versa for the cells of the High Shrub schemes. This plot depicts the results for our model instances that were simulated with field-based rules for updating flammability and vegetation response to burn status, the bottom row depict rules that enhanced the fire-vegetation feedback...... 81 Fig. 3.7 Field collected (2012 – 2014) fire data. (A) Expected fire return interval (years) by relative gradient position, calculated as the inverse of the probability of burning multiplied by fire frequency (1-yr [solid orange line], 3-yr [dashed purple line], 9-yr [dotted green line]). (B) Proportion of plots burned per relative gradient position .... 82 Fig. 3.8 The mean proportion of burned cells by simulation instance (vegetation structure initialization scheme, model updating rules) by fire frequency (1-yr [solid orange line], 3-yr [dashed purple line], 9-yr [dotted green line]). The proportion of burned cells was calculated as the mean across lattices (n = 100) by cell gradient position. The cells of the Field 2012 schemes were initiated with our observed (2012) vegetation structure values. The cells of the High Grass schemes were initiated with a 3 3 high value for C4 grass (0.25 m ) and a low value for evergreen shrub (0.025 m ), and vice versa for the cells of the High Shrub schemes. This plot depicts the results for our model instances that were simulated with field-based rules for updating flammability and vegetation response to burn status, the bottom row depict rules that enhanced the fire-vegetation feedback...... 83

Fig. 3.9 The mean proportion of a cell occupied by either vegetation structure (C4 grass [solid brown line], evergreen shrub [dashed green line]) type by gradient position and arranged by vegetation initialization scheme (columns) by fire frequency (rows). The proportion of cell occupation was calculated as the mean across lattices (n = 100) by cell gradient position. Cells in the Field 2012 scheme were initiated with our observed (2012) vegetation structure values. Cells in the High Grass scheme were initiated with 3 3 a high value for C4 grass (0.25 m ) and a low value for evergreen shrub (0.025 m ), and vice versa for the lattice cells of the High Shrub scheme. This plot depicts the results for our model instances that were simulated with enhanced-feedback based rules for updating flammability and vegetation response to burn status ...... 84

Fig. 3.10 The mean proportion of plots by vegetation type (C4 grass [solid brown line], evergreen shrub [dashed green line]) with non-negligible volume (i.e. C4 grass > 0.004 m3, evergreen shrub > 0.019 m3) by gradient position and arranged by vegetation structure initialization scheme (columns) by fire frequency (rows). The proportion of cells with non-negligible volume was the mean across lattices (n = 100) by cell gradient position. Cells in the Field 2012 scheme were initiated with our observed (2012) vegetation structure values. Cells in the High Grass scheme were 3 initiated with a high value for C4 grass (0.25 m ) and a low value for evergreen shrub (0.025 m3), and vice versa for the lattice cells of the High Shrub scheme. This plot depicts the results for our model instances that were simulated with enhanced- feedback based rules for updating flammability and vegetation response to burn status ...... 85

xii

Fig. 4.1 Ratio of post-coppicing volume to pre-coppicing volume by harvest group per species. The ratio of post-:pre-coppicing volume recovery did not vary between species at 9 or 19 months after coppicing (Kruskal-Wallis; χ2 = 3.57, P = 0.468; χ2 = 4.30, P = 0.368, respectively). The box plots indicate data range, quartiles, and median. Dots are outliers. Asterisks denote mean. A single outlier for L. tulipifera for the 19- month harvest group is not shown (value = 3.37). Species are ordered by the position along the savanna-wetland gradient (Schafer and Just 2014) ...... 108 Fig. 4.2 (A) Wood density (g cm-3), (B) total lignin (proportion of lignin per [dry weight basis]), and (C) total phenolics (proportion of extractable phenolics per sample [dry weight basis]) per species. The box plots indicate data range, quartiles, and median. Dots are outliers. Asterisks denote mean. Letters within each plot represent species’ differences using Benjamini-Hochberg post hoc comparisons (α = 0.05) on Kruskal- Wallis analysis between species for the variable of interest. Species are ordered (left to right) by descending mean value for the variable of interest ...... 109 Fig. 4.3 Decay proportion, defined here as the mean proportion of discolored wood volume to total wood volume to a belowground depth ≤ 5.0 cm, by species for (A) the 9 and 19 month post-coppicing harvest groups combined, (B) the 9 month harvest group, and (C) the 19 month harvest group. The box plots indicate data range, quartiles, and median. Dots are outliers. Asterisks denote mean. Letters within each plot represent species’ differences using Benjamini-Hochberg post hoc comparisons (α = 0.05) on Kruskal-Wallis analysis of decay proportion between species. Species are ordered (left to right) by descending mean value for the variable of interest ...... 110 Fig. 5.1 Conceptual model of the longleaf pine savanna – wetland ecotonal gradient with paired community plot arrangement ...... 143 Fig. 5.2 Effect of seed-sowing treatment (undisturbed [green], litter removal and undisturbed soil [orange], litter removal and covered with loose soil [purple]) on seedling emergence, survival, and establishment. (a) Emergence is the proportion of seedlings to seeds sown at the end of the first growing season. (b) Two-year survival is the proportion of emergent seedlings that survived until the end of a 2-year period beginning at the end of the first growing season. (c) Establishment is the proportion of seedlings at the end of the study period (Sept. 2015) to seeds sown. Error bars represent the standard error of the mean. Significance levels for type II Wald χ2 test statistic: ***P < 0.001,**P < 0.01, *P < 0.05, NS = not significant ...... 144 Fig. 5.3 Predicted probability of P. calleryana seedling survival after a topkilling fire based soil humic matter content by seed-sowing treatment (undisturbed [green], litter removal and undisturbed soil [orange], litter removal and covered with loose soil [purple]) ...... 145 Fig. 5.4 Line is the predicted probability of a P. calleryana seedling surviving a topkilling fire based on its height. Diamonds represent observed data points ...... 146

xiii

Fig. 5.5 Proportion of P. calleryana individuals that are either < 0.45 or ≥ 0.45 cm3 for three dates (May 2013 was approximately one year after seed sowing). Individuals that are ≥ 0.45 cm3 have a > 50% probability of surviving a topkilling fire. Values indicate the number of individuals per size class ...... 147

xiv

Chapter 1: Introduction

The spatial distribution of species is a fundamental concern for ecologists and biologists, and endeavors to determine differences in species’ distributions have been ongoing since at least the writings of Georges-Louis Leclerc, Comte de Buffon (Leclerc 2004) during the

Enlightenment. A century later, Alfred Russel Wallace began delineating species distributions into discrete units (Wallace 1876) and delineations continue with contemporary efforts (e.g. Kreft and Jetz 2010, Ghiglione et al. 2012, Just et al. 2014). Ecologists have used these species distributions, along with physical characteristics of the environment to delineate the natural world at both small and large spatial extents (e.g. communities, biomes), and use these delineations, especially for plant taxa, as the foundations upon which we relate to and examine the natural world.

Arguably, the distribution of vegetation structure (i.e. species composition and physical structure) is one of the most important ecological phenomena. Vegetation structure plays an important role in ecological processes across many scales (Gosz and Sharpe 1989), including biodiversity (Tews et al. 2004, Stein et al. 2014), habitat complexity/availability for various organisms (Skowno and Bond 2003, Yates and Muzika 2006, Pringle and Fox-Dobbs

2008), nutrient cycling (Hobbie 1992), microclimatic conditions (D’Odorico et al. 2010), and resilience and resistance to disturbances (Barlow and Peres 2008, Condon et al. 2011,

Redmond et al. 2015).

What controls the distribution of vegetation structure? Climate-based (e.g. temperature, precipitation, water balance) predictions of the global distribution of vegetation

structure have been reported for quite some time (Whittaker 1975, Stephenson 1990), and the predicted distributions are rather accurate (Bond et al. 2005). However, these global, climate- based vegetation distribution predictions do tend to overestimate the extent of forests and underestimate the extent of C4 grasslands (Bond et al. 2005). The mismatch between predicted and extant distributions have been largely attributed to fire (Bond and Keeley

2005).

Fire is a re-occurring disturbance in many areas across the globe (Bowman et al.

2009), including South African savanna (Smit et al. 2010), North American oak (Peterson and Reich 2008) and pine savanna (Chapman 1932), Australian eucalypt woodland (Murphy et al. 2013), and Brazilian cerrado (Dantas et al. 2013). These periodic disturbances shape vegetation structure distributions and consequently ecosystem functioning (Habeck and

Mutch 1973, Brockway and Lewis 1997). For example, fire can have a severe effect upon vegetation structure, because in areas where fires are recurrent it tends to repeatedly topkill

(i.e. death of aerial biomass) woody plants of small and intermediate sizes (Hoffmann et al.

2009, Levick et al. 2012) – a phenomenon known as the fire trap – generally resulting in open-canopied ecosystems (e.g. Glitzenstein et al. 2003, Smit et al. 2010).

Frequently burned systems (i.e. pyrophilic), such as savanna and other grasslands, are typified by an open canopy and a dominant C4 grass understory (Staver et al. 2011a, Parr et al. 2014). These open systems are warmer and drier than closed-canopied systems

(Hennenberg et al. 2005, Ibanez et al. 2013), and these are favorable conditions for fire

(Hoffmann et al. 2012). Moreover, the open canopy benefits the shade-intolerant C4 grasses, which are also an excellent fuel. In other words, positive feedbacks exist between fire,

vegetation structure, and microclimate that not only maintain, but also reinforce vegetation structure patterns (Silva et al. 2013). Fire is common occurrence for many geographies worldwide, but not all ecosystems experience frequent fires (Krawchuk et al. 2009). For example, frequently-burned systems such as savanna may transition into adjacent, fire- deterring (i.e. pyrophobic) systems, such as forest (e.g. Staver et al. 2011b, Dantas et al.

2013). In contrast to savanna, forests are typified by closed canopies with a woody understory, resulting in a cooler and moister microclimate; these conditions result in positive feedbacks that deter fire. These contrasting positive feedbacks sharpen the vegetation structure boundaries between pyrophilic and pyrophobic communities (Wilson and Agnew

1992).

However, these feedbacks should not be considered absolute. For example, in frequently burned communities where fire has been suppressed or otherwise absent, conversion to woody vegetation structure commonly occurs (e.g. Gilliam and Platt 1999,

Moreira 2000, Bond and Keeley 2005). The same is true in the other direction, where increases in grass abundance are correlated with increased fire frequency and spread, which results in the sustained reduction of woody vegetation structure (e.g. D’Antonio and Vitousek

1992, Grigulis et al. 2005, Silvério et al. 2013). Therefore, understanding what determines the fire-vegetation feedback to transition from fire-promoting to fire-impeding is important for predicting vegetation structure patterns, especially in light of environmental change (e.g. altered fire regimes (Beckage et al. 2006, Nowacki and Abrams 2008), climate (Bedel et al.

2013), non-native invasive species (Brooks et al. 2004), N deposition (Henry et al. 2006),

land-use change (Duncan and Schmalzer 2004, Brudvig et al. 2014), and elevated CO2

(Poorter and Navas 2003, Wang et al. 2012)).

I investigated different ecological aspects of vegetation structure along fire-managed longleaf pine (Pinus palustris) savanna – wetland ecotonal gradients at Fort Bragg, a US

Army installation in North Carolina, USA. Longleaf pine – wiregrass (Aristida stricta) savannas are pyrophilic and have an estimated mean historical fire return interval of 2-3 years (Stambaugh et al. 2011). In portions of their range, these savannas are dissected by pyrophobic wetlands (Sorrie et al. 2006, Schafer et al. 2013), whose fire return interval has an estimated range of 7-50 years (Frost 1993). The longleaf pine ecosystem has been noted for high species diversity and endemism (Barnett 1999, Estill and Cruzan 2001, Noss et al.

2015), as a result of the complex landscape created from the interactions of physiography, vegetation structure, and frequent fires (Noss 1989, Brockway and Lewis 1997). Prior to

European settlement, the longleaf pine ecosystem was considered to be that with the greatest areal extent in North America (Landers et al. 1995). The extant coverage of the longleaf pine ecosystem is estimated to be 3% of its historical maximum (37 million ha), with losses attributed to land-use change and fire suppression (Frost 1993, Outcalt 2000). Prolonged fire suppression in many longleaf pine savannas has resulted in a change to more dense vegetation structure (Gilliam and Platt 1999). However, there is spirited interest in the conservation and restoration of these systems including active, fire-based management (e.g.

Van Lear et al. 2005, Varner et al. 2005, Aschenbach et al. 2010). These longleaf pine savanna – wetland gradients are also fire-return-interval gradients and as such are excellent locations to consider questions about fire and vegetation structure.

The overall goal of my dissertation research was to better understand the ecology of vegetation structure between fire-promoting and fire-deterring communities in an effort to strengthen our ability to predict both flammability and community boundary dynamics under changes in the environment or management. In Chapter 2, I examined vegetation structure and microclimate factors to uncover what regulates the extent of flammable conditions between pyrophilic and pyrophobic communities. Understanding what determines where fires stop is a necessary first step to creating predictions about community boundaries under change. In Chapter 3, I model flammability and vegetation structure dynamics under variations in fire frequency to assess the stability of community boundaries under potential environmental changes. Chapter 4 examines incipient woody decay in root collars of resprouting trees that differ in their abundances along this fire-frequency gradient. In doing so, I highlight a potentially overlooked mechanism, wood decay resistance, of woody plant persistence in frequently burned ecosystems. For chapter 5, I tested the potential of six non- native woody plant species to invade these gradients. I examined the effects of fire and site conditions on their survival and establishment success. Successful invasion by woody individuals increases the chances of fire-deterrence and woody vegetation structure encroachment in the savanna, and therefore the movement of community boundaries.

References

Aschenbach, T. A., B. L. Foster, and D. W. Imm. 2010. The initial phase of a longleaf pine- wiregrass savanna restoration: species establishment and community responses. Restoration Ecology 18:762–771.

Barlow, J., and C. A. Peres. 2008. Fire-mediated dieback and compositional cascade in an

Amazonian forest. Philosophical Transactions of the Royal Society B: Biological Sciences 363:1787–1794.

Barnett, J. P. 1999. Longleaf Pine Restoration. Journal of Sustainable Forestry 9:89–96.

Beckage, B., L. Gross, and W. Platt. 2006. Modelling responses of pine savannas to climate change and large-scale disturbance. Applied Vegetation Science 9:75–82.

Bedel, A. P., T. Mote, and S. Goodrick. 2013. Climate change and associated fire potential for the south-eastern United States in the 21st century. International Journal of Wildland Fire 22:1034–1043.

Bond, W. J., and J. E. Keeley. 2005. Fire as a global “herbivore”: the ecology and evolution of flammable ecosystems. Trends in Ecology & Evolution 20:387–394.

Bond, W. J., F. I. Woodward, and G. F. Midgley. 2005. The global distribution of ecosystems in a world without fire. New Phytologist 165:525–538.

Bowman, D. M. J. S., J. K. Balch, P. Artaxo, W. J. Bond, J. M. Carlson, M. A. Cochrane, C. M. D’Antonio, R. S. DeFries, J. C. Doyle, and S. P. Harrison. 2009. Fire in the Earth system. Science 324:481–484.

Brockway, D. G., and C. E. Lewis. 1997. Long-term effects of dormant-season prescribed fire on plant community diversity, structure and productivity in a longleaf pine wiregrass ecosystem. Forest Ecology and Management 96:167–183.

Brooks, M. L., C. M. D’Antonio, D. M. Richardson, J. B. Grace, J. E. Keeley, J. M. DiTomaso, R. J. Hobbs, M. Pellant, and D. Pyke. 2004. Effects of invasive alien plants on fire regimes. BioScience 54:677.

Brudvig, L. A., J. L. Orrock, E. I. Damschen, C. D. Collins, P. G. Hahn, W. B. Mattingly, J. W. Veldman, and J. L. Walker. 2014. Land-use history and contemporary management inform an ecological reference model for longleaf pine woodland understory plant communities. PloS one 9:e86604.

Chapman, H. H. 1932. Is the longleaf type a climax? Ecology 13:328–334.

Condon, L., P. J. Weisberg, and J. C. Chambers. 2011. Abiotic and biotic influences on Bromus tectorum invasion and Artemisia tridentata recovery after fire. International Journal of Wildland Fire 20:597–604.

D’Antonio, C. M., and P. M. Vitousek. 1992. Biological invasions by exotic grasses, the grass/fire cycle, and global change. Annual Review of Ecology and Systematics 23:63–

87.

D’Odorico, P., J. D. Fuentes, W. T. Pockman, S. L. Collins, Y. He, J. S. Medeiros, S. DeWekker, and M. E. Litvak. 2010. Positive feedback between microclimate and shrub encroachment in the northern Chihuahuan desert. Ecosphere 1:art17.

Dantas, V. D. L., M. A. Batalha, and J. J. G. Pausas. 2013. Fire drives functional thresholds on the savanna – forest transition. Ecology 94:2454–2463.

Duncan, B. W., and P. A. Schmalzer. 2004. Anthropogenic influences on potential fire spread in a pyrogenic ecosystem of Florida, USA. Landscape Ecology 19:153–165.

Estill, J., and M. Cruzan. 2001. Phytogeography of rare plant species endemic to the southeastern United States. Castanea 66:3–23.

Frost, C. 1993. Four centuries of changing landscape patterns in the longleaf pine ecosystem. Pages 17–43 Proceedings of the Tall Timbers fire ecology conference.

Ghiglione, J.-F., P. E. Galand, T. Pommier, C. Pedrós-Alió, E. W. Maas, K. Bakker, S. Bertilson, D. L. Kirchman, C. Lovejoy, and P. L. Yager. 2012. Pole-to-pole biogeography of surface and deep marine bacterial communities. Proceedings of the National Academy of Sciences 109:17633–17638.

Gilliam, F., and W. Platt. 1999. Effects of long-term fire exclusion on tree species composition and stand structure in an old-growth Pinus palustris (longleaf pine) forest. Plant Ecology 1920:15–26.

Glitzenstein, J., D. Streng, and D. Wade. 2003. Fire frequency effects on longleaf pine (Pinus palustris P. Miller) vegetation in South Carolina and northeast Florida, USA. Natural Areas Journal 23:22–37.

Gosz, J. R., and P. J. H. Sharpe. 1989. Broad-scale concepts for interactions of climate, topography, and biota at biome transitions. Landscape Ecology 3:229–243.

Grigulis, K., S. Lavorel, I. D. Davies, A. Dossantos, F. Lloret, and M. Vilà. 2005. Landscape- scale positive feedbacks between fire and expansion of the large tussock grass, Ampelodesmos mauritanica in Catalan shrublands. Global Change Biology 11:1042– 1053.

Habeck, J. R., and R. W. Mutch. 1973. Fire-dependent forests in the northern Rocky Mountains. Quaternary Research 3:408–424.

Hennenberg, K. J., D. Goetze, L. Kouamé, B. Orthmann, and S. Porembski. 2005. Border

and ecotone detection by vegetation composition along forest‐savanna transects in Ivory Coast. Journal of Vegetation Science 16:301–310.

Henry, H. A. L., N. R. Chiariello, P. M. Vitousek, H. A. Mooney, and C. B. Field. 2006. Interactive effects of fire, elevated carbon dioxide, nitrogen deposition, and precipitation on a California annual grassland. Ecosystems 9:1066–1075.

Hobbie, S. E. 1992. Effects of plant species on nutrient cycling. Trends in ecology & evolution 7:336–9.

Hoffmann, W. A., R. Adasme, M. Haridasan, M. T. de Carvalho, E. L. Geiger, M. A. B. Pereira, S. G. Gotsch, A. C. Franco, and M. T. de Carvalho. 2009. Tree topkill, not mortality, governs the dynamics of savanna-forest boundaries under frequent fire in central Brazil. Ecology 90:1326–1337.

Hoffmann, W. A., S. Y. Jaconis, K. L. McKinley, E. L. Geiger, S. G. Gotsch, and A. C. Franco. 2012. Fuels or microclimate? Understanding the drivers of fire feedbacks at savanna–forest boundaries. Austral Ecology 37:634–643.

Ibanez, T., C. Hély, and C. Gaucherel. 2013. Sharp transitions in microclimatic conditions between savanna and forest in New Caledonia: Insights into the vulnerability of forest edges to fire. Austral Ecology 38:680–687.

Just, M. G., J. F. Norton, A. L. Traud, T. Antonelli, A. S. Poteate, G. A. Backus, A. Snyder- Beattie, R. W. Sanders, and R. R. Dunn. 2014. Global biogeographic regions in a human-dominated world: the case of human diseases. Ecosphere 5:art143.

Krawchuk, M. A., M. A. Moritz, M. A. Parisien, J. Van Dorn, and K. Hayhoe. 2009. Global pyrogeography: the current and future distribution of wildfire. PLoS ONE 4:e5102.

Kreft, H., and W. Jetz. 2010. A framework for delineating biogeographical regions based on species distributions. Journal of Biogeography 37:2029–2053.

Landers, J. L., D. H. Van Lear, and W. D. Boyer. 1995. The longleaf pine forests of the Southeast: requiem or renaissance? Journal of Forestry 93:39–44.

Van Lear, D. H., W. D. D. Carroll, P. R. R. Kapeluck, and R. Johnson. 2005. History and restoration of the longleaf pine-grassland ecosystem: implications for species at risk. Forest Ecology and Management 211:150–165.

Leclerc, G.-L. 2004. Excerpts from: Natural history, general and particular. Pages 16–18 in M. Lomolino, D. Sax, and J. Brown, editors. Foundations of biogeography: classic paper with commentaries. University of Chicago Press, Chicago.

Levick, S. R., G. P. Asner, and I. P. J. Smit. 2012. Spatial patterns in the effects of fire on savanna vegetation three-dimensional structure. Ecological Applications 22:2110–2121.

Moreira, A. G. 2000. Effects of fire protection on savanna structure in central Brazil. Journal of Biogeography 27:1021–1029.

Murphy, B. P., R. A. Bradstock, M. M. Boer, J. Carter, G. J. Cary, M. A. Cochrane, R. J. Fensham, J. Russell-Smith, G. J. Williamson, and D. M. J. S. Bowman. 2013. Fire regimes of Australia: a pyrogeographic model system. Journal of Biogeography 40:1048–1058.

Noss, R. F. 1989. Longleaf pine and wiregrass: keystone components of an endangered ecosystem. Natural Areas Journal 9:211–213.

Noss, R. F., W. J. Platt, B. A. Sorrie, A. S. Weakley, D. B. Means, J. Costanza, and R. K. Peet. 2015. How global biodiversity hotspots may go unrecognized: lessons from the North American Coastal Plain. Diversity and Distributions 21:236–244.

Nowacki, G. J., and M. D. Abrams. 2008. The demise of fire and “mesophication” of forests in the Eastern United States. BioScience 58:123–138.

Outcalt, K. W. 2000. The longleaf pine ecosystem of the South. Native Plants Journal 1:43– 53.

Parr, C. L., C. E. R. Lehmann, W. J. Bond, W. A. Hoffmann, and A. N. Andersen. 2014. Tropical grassy biomes: misunderstood, neglected, and under threat. Trends in Ecology & Evolution 29:205–213.

Peterson, D. W., and P. B. Reich. 2008. Fire frequency and tree canopy structure influence plant species diversity in a forest-grassland ecotone. Plant Ecology 194:5–16.

Poorter, H., and M. L. Navas. 2003. Plant growth and competition at elevated CO2: on winners, losers and functional groups. New Phytologist 157:175–198.

Pringle, R. M., and K. Fox-Dobbs. 2008. Coupling of canopy and understory food webs by ground-dwelling predators. Ecology Letters 11:1328–1337.

Redmond, M. D., N. S. Cobb, M. J. Clifford, and N. N. Barger. 2015. Woodland recovery following drought-induced tree mortality across an environmental stress gradient. Global Change Biology 21:3685–3695.

Schafer, J. L., B. P. Breslow, M. G. Just, M. G. Hohmann, S. N. Hollingsworth, S. L.

Swatling-Holcomb, and W. A. Hoffmann. 2013. Current and historical variation in wiregrass (Aristida stricta) abundance and distribution is not detectable from soil δ13C measurements in longleaf pine (Pinus palustris) savannas. Castanea 78:28–36.

Silva, L. C. R., W. A. Hoffmann, D. R. Rossatto, M. Haridasan, A. C. Franco, and W. Horwath. 2013. Can savannas become forests? A coupled analysis of nutrient stocks and fire thresholds in central Brazil. Plant Soil 373:829–842.

Silvério, D. V, P. M. Brando, J. K. Balch, F. E. Putz, D. C. Nepstad, C. Oliveira-Santos, and M. M. C. Bustamante. 2013. Testing the Amazon savannization hypothesis: fire effects on invasion of a neotropical forest by native cerrado and exotic pasture grasses. Philosophical transactions of the Royal Society of London. Series B, Biological sciences 368:20120427.

Skowno, A. L., and W. J. Bond. 2003. Bird community composition in an actively managed savanna reserve, importance of vegetation structure and vegetation composition. Biodiversity and Conservation 12:2279–2294.

Smit, I. P. J., G. P. Asner, N. Govender, T. Kennedy-Bowdoin, D. E. Knapp, and J. Jacobson. 2010. Effects of fire on woody vegetation structure in African savanna. Ecological applications 20:1865–1875.

Sorrie, B. A., J. B. Gray, and P. J. Crutchfield. 2006. The vascular flora of the longleaf pine ecosystem of Fort Bragg and Weymouth Woods, North Carolina. Castanea 71:129–161.

Stambaugh, M. C., R. P. Guyette, and J. M. Marschall. 2011. Longleaf pine (Pinus palustris Mill.) fire scars reveal new details of a frequent fire regime. Journal of Vegetation Science 22:1094–1104.

Staver, A. C., S. Archibald, and S. Levin. 2011a. Tree cover in sub-Saharan Africa: rainfall and fire constrain forest and savanna as alternative stable states. Ecology 92:1063–1072.

Staver, A. C., S. Archibald, and S. A. Levin. 2011b. The global extent and determinants of savanna and forest as alternative biome states. Science 334:230–232.

Stein, A., K. Gerstner, and H. Kreft. 2014. Environmental heterogeneity as a universal driver of species richness across taxa, biomes and spatial scales. Ecology Letters 17:866–880.

Stephenson, N. 1990. Climatic control of vegetation distribution: the role of the water balance. American Naturalist 135:649–670.

Tews, J., U. Brose, V. Grimm, K. Tielboerger, M. C. Wichmann, M. Schwager, and F. Jeltsch. 2004. Animal species diversity driven by habitat heterogeneity/diversity: The

importance of keystone structures. Journal of Biogeography 31:79–92.

Varner, J. M., D. R. Gordon, F. E. Putz, and J. K. Hiers. 2005. Restoring fire to long- unburned Pinus palustris ecosystems: novel fire effects and consequences for long- unburned ecosystems. Restoration Ecology 13:536–544.

Wallace, A. R. 1876. The geographical distribution of animals: With a study of the relations of living and extinct faunas as elucidating the past changes of the earth’s surface. Cambridge University Press, Cambridge, England.

Wang, D., S. A. Heckathorn, X. Wang, and S. M. Philpott. 2012. A meta-analysis of plant physiological and growth responses to temperature and elevated CO 2. Oecologia 169:1–13.

Whittaker, R. 1975. Communities and ecosystems. Second. Macmillian Publishing Co., Inc., New York, NY.

Wilson, J. B., and A. D. Q. Agnew. 1992. Positive-feedback switches in plant communities. Pages 263–336 in M. Begon and A. Fitter, editors. Advances in Ecological Research. Academic Press London.

Yates, M., and R. Muzika. 2006. Effect of forest structure and fragmentation on site cccupancy of bat species in Missouri Ozark Forests. Journal of Wildlife Management 70:1238–1248.

Chapter 2: Where fire stops: vegetation structure and microclimate influence fire

spread along an ecotonal gradient

*Accepted for publication in Plant Ecology, the final publication is available at http://link.springer.com/article/10.1007/s11258-015-0545-x, with kind permission from

Springer Science+Business Media

Acknowledgements

We thank A. Ballard, S. Bell, B. Breslow, K. Fraboni, S. Goyette, R. Sanders, M. Walz, and

A. Whitehead for research assistance. We also thank the Fort Bragg Endangered Species and

Forestry branches for logistic support. This research was supported by a cooperative agreement between the US Army Engineer Research and Development Center and North

Carolina State University (W9132T-11-2-0007 to W.A.H.). M.G.J. received support from a

Southeast Climate Science Center fellowship.

Abstract

Positive feedbacks influenced by direct and indirect interactions between fire, vegetation, and microclimate can allow pyrophilic and pyrophobic ecosystems to co-occur in the same landscape, resulting in the juxtaposition of flammable and non-flammable vegetation. To quantify the drivers of these feedbacks, we combined measurements of vegetation, fuels, and microclimate with observations of fire spread along ecotonal gradients. We established 113 permanent transects (consisting of 532 plots), each traversing an ecotone between savanna and wetland in the Sandhills of North Carolina, USA. In each plot, we recorded cover of ten plant functional types. We collected surface fuels at a subset of our transects. We continuously monitored microclimate (nine meteorological variables) across 21 representative ecotones. Following prescribed fire, we measured fire spread along each transect. Vegetation structure and microclimate significantly predicted fire spread along the savanna-wetland ecotone. Fire spread was most influenced by vegetation structure, specifically C4 grass cover, which accounted for 67% of the variance explained by our model. We have identified the components of the fire, vegetation, and microclimate feedback that control where fires stop under current conditions, but their control should not be considered absolute. For example, when ignited in savanna, prescribed burns continued through wetland vegetation 43% of the time. The feedback operating within these systems may be relatively weak as compared to other savanna systems. Environmental changes may alter fire spread extent, and with it ecosystem boundaries, or even ecosystem states.

Introduction

Understanding the role of fire in controlling the distribution of vegetation is complicated by positive feedbacks that commonly exist between vegetation and fire (Murphy and Bowman

2012; Fill et al. 2015). Within these feedbacks, vegetation structure (i.e. physical structure and species composition) influences fire (Dwire and Kauffman 2003), while fire influences vegetation structure (Smit et al. 2010), making it difficult to disentangle cause and effect in the association between the two factors. Positive feedbacks tend to reinforce existing patterns of vegetation across landscapes (Silva et al. 2013), thereby sharpening boundaries between flammable (pyrophilic) and non-flammable (pyrophobic) vegetation (Wilson and Agnew

1992; Dantas et al. 2013). The origin and maintenance of such boundaries depends upon the factors that control this transition from flammable to non-flammable conditions.

The transition between flammable and non-flammable states, however, is difficult to predict.

Physical-based models have been well elaborated for characterizing fire spread in homogeneous vegetation (Morvan 2011), but these exhibit two limitations that make them inadequate for predicting the spread of fire into pyrophobic vegetation. First, these models are designed to simulate steady-state fire spread in homogeneous fuels; therefore, their performance is questionable for heterogeneous conditions. Second, the underlying Rothermel

(1972) equations simulate fire characteristics, but do not predict where fires extinguish.

Under conditions unable to sustain burning, the equations generate predictions of impossibly low fire intensities and rates of spread, rather than predicting where fires stop (Rothermel

1983; Cruz and Alexander 2013).

Modelling of fire spread within homogeneous ecosystems has been well described with simulation models (Cheney et al. 1993; Simeoni et al. 2011; Russo et al. 2014), and other modelling efforts have identified fire spread drivers (Hoffmann et al. 2002; Favier et al.

2004; Beckage et al. 2009). However, empirical identification of fire spread drivers from gradients with varying components (e.g. topography, hydrology) and real fires is lacking.

Without specific knowledge of what governs the extent of fire spread along ecotonal gradients, predictions of pyrogenic ecosystem distributions under environmental change are suspect.

In this study we examined the influence of vegetation structure and microclimate on the extent of fire spread along savanna – wetland ecotonal gradients to understand what controls fire spread along this transition. Our objectives were to determine the relative degree to which vegetation structure and microclimate influence fire spread. We recorded vegetation structure, microclimate, and fire spread extent (under prescribed-fire meteorological conditions) at 113 ecotonal gradients containing 532 plots. These data were used to model fire spread along the gradient. This work is an important step in further understanding the relationships between fire, vegetation, and microclimate in controlling ecosystem boundaries and is one of few to use information from observed fire events.

Methods

Study site

This study was conducted at the Fort Bragg US Army installation (73,469 ha), in the

Sandhills region of North Carolina, USA (35°07’N, 79°10’W). The Sandhills are remnant sand ridges of an ancient coastline and lie between the piedmont and the coastal plain from

North Carolina to Alabama. Mean annual precipitation is 1275 mm and mean temperature ranges from 6.9°C in the winter to 26°C in the summer (Sorrie et al. 2006). Elevation ranges from 43 to 176 m. The predominant vegetation community on Fort Bragg is longleaf pine

(Pinus palustris Mill.) – wiregrass (Aristida stricta Michx.) savanna (i.e. pine scrub oak sandhill sensu [Schafale 2012])) which is situated upon well-drained, sandy soils (Platt

1999). Wetlands, known locally as streamhead pocosins and sandhill seeps (Weakley and

Schafale 1991), are embedded within the savanna matrix. These wetlands arise from a perched water table on an impermeable clay soil layer that underlies the adjacent uplands

(Fig. 2.1). Fire history is a considerable determinant of the dominant vegetation of these wetlands (Sorrie et al. 2006). Wetlands that have experienced fires that are more frequent are dominated by pyrogenic, herbaceous vegetation such as switchcane (Arundinaria tecta

(Walter) Muhl.) or ferns (e.g. Osmunda cinnamomea L., Pteridium aquilinum (L.) Kuhn).

Whereas wetlands with less frequent fires are denser and dominated by shrubs and trees (e.g.

Clethra alnifolia L., Persea palustris (Raf.) Sarg., Acer rubrum (L.)).

Historically, longleaf pine ecosystems were maintained by frequent, low-intensity wildfires

(Christensen 2000), but fire suppression was widely practiced throughout much of the 20th century. Since the creation of Fort Bragg in 1918, military training has frequently ignited wildfires that helped maintain the pyrophilic, longleaf pine savanna. A prescribed burning

program began in the early 1990s and Fort Bragg was sectioned into 1,355 burn compartments mostly managed with a 3-year fire return interval, which follows estimates of mean pre-settlement fire frequencies (Stambaugh et al. 2011). Prescribed burns occur between February and August, employing techniques to maximize control; for example, burns are started with backing fires (Lashley et al. 2014) and performed under constrained meteorological conditions (e.g. typical growing season burn-day meteorological conditions

[10:00-16:00]: wind speed at 10 m 4.0-5.8 m s-1, relative humidity 42-48% [Gerdes 2011]).

Data collection

We established 113 permanent transects along ecotonal gradients in 73 burn compartments

(Fig. 2.1) and distributed over 230 km2 of Fort Bragg. We identified potential transect locations from Fort Bragg GIS plant community data. We chose burn compartments that contained both longleaf pine savanna and wetland communities. Each transect began in and contained at least 10 m of savanna and extended down the gradient into wetland vegetation.

The mean distance between a transect and its nearest neighbor was approximately 302 m.

Most transects (74) ended in closed, woody wetlands (generally dominated by evergreen shrubs) and the remaining (39) ended in open, herbaceous wetlands. At 10-m intervals along each transect we created 1 m2 plots for monitoring fuels, vegetation and fire spread. Transect length depended upon the abruptness of the gradient and mean transect length was 37.4 m, yielding 532 sampling plots. To account for differences in transect length, plot location along a transect was transformed to a relative scale (relative gradient position) which was

calculated as the distance of a plot’s center from the savanna plot divided by the total length of the transect.

Vegetation structure

To characterize vegetation structure we determined the cover of ten plant functional types

(Table 2.1). We visually estimated the cover of each functional type in each plot (Bonham

2013) to a height of 2 m. Cover was estimated in 5% increments starting at zero, with an additional class (1%) for presence with negligible cover.

In each plot, we recorded fuelbed depth, total plot cover, and canopy closure. Fuelbed depth

(cm) was measured near the four corners of the plot as the maximum height of continuous dead fuel (e.g. pine litter, grass) and recorded as the mean. We calculated total plot cover as the sum of the cover of each functional type, so plot cover was occasionally greater than

100%. The canopy closure (%) for strata greater than 2 m was measured for each plot using a concave spherical densiometer (Lemmon 1956). These variables were measured once per year (late May – early July) in 2012 – 2014.

Fuels

In the summers of 2012 and 2013, a subset of transects (n=23), were randomly chosen for more detailed measurements of surface fuels. We established a 0.5-m2 fuel plot (n=103) with center positioned 2 m laterally from the center of each vegetation plot in the transect and harvested all fuels (living and dead) present. Fuels were sorted into the same functional types

as vegetation structure, with additional classes for tree bark, pine cones and needles, and miscellaneous plant matter. Fuel samples were oven-dried at 60°C for at least 5 days and weighed. We calculated fuel bulk density (mass per volume) per plot as described in Online

Resource 1. We calculated a surface area to volume ratio for each class from fuel subsamples

(Online Resource 2).

Microclimate

To characterize microclimate across ecotones, we set up meteorological stations to monitor precipitation throughfall, soil moisture, air temperature, relative humidity, wind speed, wind direction, and understory photosynthetically active radiation (PAR) from 15 May 2012 – 30

September 2014. We chose three representative ecotones in 2012 and 2013 with paired microclimate stations in savanna and wetland (community determinations based on vegetation and GIS data). One pair of stations remained at the same ecotone for two years

(May 2012 – May 2014), while the others were moved after one year (May 2012 – 2013, and

May 2013 – 2014) because of planned burns. To better characterize site variation, in 2014, we selected 16 representative ecotones where stations were placed for approximately two weeks per ecotone (May – September 2014). Distance between stations was approximately

50 m. All sensors were installed at height of 1 m unless otherwise stated. Sensor measurements were taken every minute, and 10-minute means (except current relative humidity and total precipitation) were recorded onto a CR1000 data logger (Campbell

Scientific, Logan, UT, USA). Precipitation (mm) was measured with a TR-525M tipping rain bucket (Texas Electronics, Dallas, TX, USA) installed at a height of 0.75 m. Volumetric soil

water content (%) to a depth of 5 cm was measured with a EC-5 soil moisture probe

(Decagon Devices, Pullman, WA, USA). Air temperature (°C) and relative humidity were measured with a CS215 temperature/relative humidity sensor (Campbell Scientific) with solar radiation shielding. Wind speed (ms-1) was measured with a 014A cup anemometer

(MetOne Instruments, Grants Pass, OR, USA). Wind direction (°) was measured in savanna with a 024A wind vane (MetOne Instruments). PAR (µmol m-2 s-1) was measured with a SQ-

110 quantum sensor (Apogee Instruments, Logan, UT, USA). We calculated vapor pressure deficit (kPa) as the difference between vapor pressure at saturation and vapor pressure of the air (Lawrence 2005). We calculated the component of wind speed in the direction of the gradient as the cosine of the difference between transect bearing and wind direction

(Atkinson et al. 2010).

Our meteorological stations also included sensors to monitor coarse and fine fuel moisture and fuel temperature concurrently. Coarse fuel moisture (%) was measured with a CS506 fuel moisture sensor (Campbell Scientific) with a pine dowel that emulates fuel moisture of a

10-hour timelag fuel (Nelson Jr. 2000). These dead, woody fuels range in diameter from 0.6-

2.5 cm and 10 hours is the time required for the fuel particle moisture to equilibrate to changes in atmospheric moisture. Fine fuel moisture (%) was monitored with a modified

CS616 water content reflectometer (Campbell Scientific). The probe rods were shaped to hold a 1 × 10 cm Magnolia grandiflora L. leaf section (~0.25 mm thick) positioned at a height of 30 cm. Fine fuel moisture was estimated from author-calculated calibration curves

(adj. r2=0.864, p<0.001). Fuel temperature was measured with a constantan-copper thermocouple inserted within a 1.3 cm diameter × 15 cm long Betula wood dowel.

Fire

Our sites were subjected to prescribed fire, where fires are ignited in the upland and allowed to burn into wetlands. Following fire, we measured the extent (length [m]) of fire spread along each transect, recorded each plot as either burned or unburned, and calculated the proportion of the gradient burned. Over the study period, 113 (84 unique) transects were subjected to a fire event. Of the 523 (389 unique) plots at these transects, 421 (336 unique) plots were burned. We recorded maximum fire temperature (2013-2014) at the center of 161 vegetation plots using Tempilaq® temperature-indicating paints (Air Liquide America

Corporation, South Plainfield, NJ, USA). We applied ten paints, each with a different temperature indication (i.e. 93, 149, 204, 260, 316, 371, 427, 482, 538, or 593 °C), to 2 × 7 cm aluminum tags positioned 20 cm above ground level. There is discussion about the effectiveness of temperature indicating paints (e.g. Bova and Dickinson 2008) as well as the issue of melting aluminum tags (none of our tags melted). However, this approach is supported as an inexpensive means of quantifying fire temperature (Iverson et al. 2004). The fire history (number of fires since 1991 and years since fire) of each transect was obtained from Fort Bragg prescribed fire records.

Analyses:

Vegetation structure, fuels, fire temperature, and microclimate

We used linear mixed-effect models (LMM) to evaluate the relationships between vegetation structure functional types and gradient position (fixed effect) with an uncorrelated random intercept (site) and random slope (gradient position by site) with lme4 (Bates et al. 2015). We used Satterwaithe’s approximation of denominator degrees of freedom to calculate P values in lmerTest (Kuznetsova et al. 2012). Using the same LMM procedures, we evaluated fuel mass by relative gradient position and vegetation structure, as well as fire temperature by vegetation structure, gradient position, and microclimate. We compared microclimate variables between savannas and wetlands using Welch’s unequal variance t-test. All statistical analyses were performed in R (R Core Team 2015).

To determine the best predictors of fire spread along the gradient, we used mixed-effects logistic regression to evaluate the factors that best predict the probability of burning (with site as a random effect). The microclimate data from the reference transects were used to characterize microclimate at the date of burning. The mean value during midday (10:00 to

16:00) for each burn date was calculated for savanna and wetland endpoints, and these were used to linearly interpolate values at each intermediate plot location based on relative gradient position.

Prescribed fires at Fort Bragg are ignited in the savanna and are allowed to burn into wetlands, thereby defining the direction of fire spread along our transects. Therefore a plot is exposed to a probability of burning only if the previous plot on the transect had burned. This allowed us to analyze the conditional probability of burning (i.e. the probability of burning,

given that the preceding plot had burned). To do this, we first removed data for all transects not exposed to burning, which includes transects in burn compartments not subjected to prescribed fire, as well as a few cases in which prescribed fire failed to reach the transect. We then removed the first plot in each transect because the previous step constrained these to have 100% probability of burning. Finally, we removed all plots in which the preceding plot did not burn. We used the resulting data to assess the controls on fire spread along the gradient by identifying the suite of variables that best predict the probability of burning. We began by individually testing the ability of each vegetation or microclimate variable to predict the probability that a plot burned. We used the results of these models to reduce dataset dimensionality, and removed from subsequent analyses any variable which had a P- value of greater than 0.30 (Harrell, Jr. 2001). We used the remaining variables in a model selection procedure which utilized a ‘genetic algorithm’ to build mixed-effects logistic regression models (GLMM) with glmutli (Calcagno and de Mazancourt 2010). This approach achieves similar results to comparing all possible models, but is more computationally efficient. We employed three model selection schemes: one each that used only vegetation structure or microclimate variables as potential model terms, and a third where these variables were considered together. We used the information criterion AICc to score our candidate models (Hurvich and Tsai 1993). We performed ten iterations of the model selection procedure per scheme and calculated the consensus model-averaged importance of terms (Burnham and Anderson 2002). We identified the terms that had a model-averaged importance of at least 0.8, and used them to construct a model per each scheme. Using the predicted probability of fire spread, we calculated the expected fire return interval (years) per

plot as the current prescribed fire interval (i.e. 3 years) divided by the predicted probability of burning.

We calculated the marginal R2 to assess the fit of the best GLMM model using the method proposed by Johnson (2014). The marginal R2 is a measure of the variance explained by only the fixed effects (Nakagawa and Schielzeth 2013). We also calculated the relative- importance weight of each predictor variable, which is a measure of the amount of variation it accounts for in the model (Tonidandel and LeBreton 2011). We calculated the variance inflation factor (VIF) scores (i.e. √VIF) for each term per model to ensure that our models were not confounded by multicollinearity (O’Brien 2007).

Results

Vegetation structure

Vegetation composition was correlated with position along the gradient, despite substantial variability among transects. The strongest patterns were a decline in C4 grass cover (t=-11.6, marginal R2 =0.26, p<0.001) and an increase in evergreen shrub cover (t=9.6, marginal

R2=0.17, p<0.001) from savanna to wetland. These two functional types did not co-occur at high abundances (Fig. 2.2), illustrating the distinction between savanna and wetland for some vegetation structure functional types. Compared to the uplands, wetlands also had significantly higher cover of switchcane, deciduous shrubs, evergreen trees, and ferns, but lower cover of deciduous trees and herbaceous dicots (Table 2.2).

Microclimate:

Our analyses reveal, as expected, that savanna sites were windier and drier than the wetland sites, with greater insolation and warmer fuels. For example, PAR was greater in the savanna, contributing to the drier coarse fuel as compared to wetland (Fig. 2.3). There were significant differences in midday (May – August) microclimate between savanna and wetland (Table

2.3). We found differences in each microclimate variable except air temperature.

Fuel:

For the subset of 103 plots in which fuel mass was measured, we found weak relationships between fuel mass and each of the other variables (marginal R2≤0.11; Online Resources 3 and 4). Although total fine fuel mass did not vary consistently along the gradient (Online

Resource 6), the composition of fuels did exhibit significant trends. Dead pine needles which accounted for 31% of total fuel mass overall, declined along the gradient (marginal R2=0.37, p<0.001) from a mean of 50% in savanna to a mean of 20% in wetland (Online Resource 5).

Most other fuels exhibited no trend along the gradient (Online Resource 6), except for modest increases in deciduous shrub fuel mass (marginal R2=0.05, p=0.049). Neither fuel bulk density (marginal R2=0.04, Table 2.1) nor the surface area to volume ratio (not shown) varied significantly along the gradient.

Fire:

Of the prescribed fires ignited at our study sites, 57% terminated within the transect, with remaining fires continuing through the end of our transects (Fig. 2.4b). In sites with

herbaceous (open) wetlands, 44% of the fires stopped within the transect, compared with

67% in sites with shrubby (closed) wetlands, respectively. The mean proportion of the gradient burned differed between open (93%) and closed (85%) wetlands (t=-2.471, df=109, p=0.015). The estimated fire return interval was similar along the savanna and ecotone portions of the gradient (3.0 – 3.6 years), but more than doubled in the wetland (7.5 years).

Moreover, herbaceous wetlands burned more frequently (fire interval = 5.4 years) than shrubby wetland (fire interval = 9.9 years, Fig. 2.4a). We did not find any strong linear relationships between maximum fire temperature and vegetation structure, microclimate, plot, or transect variables (marginal R2≤0.16, Online Resources 7 and 8).

Modeling of fire spread:

Here we quantify fire spread using the conditional probability of a plot burning. Fire spread was modeled using alternate subsets of predictor variables (vegetation or microclimate), with the vegetation structure model outperforming the microclimate model (marginal R2 = 0.706 versus 0.484, Table 2.4). When both vegetation and microclimate were included in model selection, the resulting best model had the best fit (marginal R2=0.756) and lowest AIC score

(Table 2.4).

In our final model, C4 grass cover had the greatest relative influence on fire spread, accounting for approximately 67% of the total variation (Table 2.4). An increase in C4 grass cover of one percentage point (other factors held constant) increases the odds of burning by

30%. Moreover, once C4 grass cover surpasses 10% (other factors held constant) the

probability of burning asymptotes near 100% (Fig. 2.5a). Similarly, once PAR exceeds 800

µmol m-2 s-1 the probability of burning is effectively 100% (Fig. 2.5c). Of the three predictor variables included in the final model, C4 grass cover and PAR increased fire spread, while evergreen shrub cover decreased fire spread (Table 2.4, Fig. 2.5). Our VIF analysis indicated that the selected models did not suffer from multicollinearity (i.e. √VIF was less than two;

Table 2.4).

Discussion

Drivers of fire spread

Overall, vegetation had a stronger effect than microclimate in controlling fire spread along the savanna-wetland gradient. The best predictive model included just two vegetation variables (C4 grass and evergreen shrub cover) and one microclimate variable (PAR), yet the model discriminated between the two burn states (burned and unburned) of the study plots

2 quite well (marginal R =0.756), with C4 grass cover accounting for 67% of the explained variance in probability of burning (Table 2.4).

Our study system exhibits a strong hydrological gradient (Table 2.3), yet this does not have a clear, direct role in controlling fire spread. Specifically, soil moisture was not a good predictor of burning along the gradient, suggesting that the influence of hydrology on fire spread is indirect. This indirect effect of hydrology on fire is likely mediated through its effect on the abundance of functional types differing in flammability and the rate at which they recover after fire (Dwire and Kauffman 2003). Most notably, fire-promoting C4 grasses

were most abundant in the xeric savanna uplands, while fire-suppressing evergreen shrubs were most abundant in the wetlands (Fig. 2.2).

Among the microclimatic variables, PAR was the best predictor of fire spread, but its contribution to flammability is likely mediated through other variables that are correlated with PAR. Some of these are directly influenced by solar radiation (e.g. fuel temperature, fuel moisture, and C4 cover) while others may be indirectly correlated with PAR (e.g. relative humidity and precipitation). Moreover, transmitted PAR is strongly dependent on vegetation structure (Martens et al. 2000), which itself has multiple effects on flammability (Fig. 2.5;

Schwilk 2003; Baeza et al. 2006). Consequently, understory PAR is a single indicator that integrates multiple vegetation and meteorological variables that control fire spread.

Curiously, our best predictors of fire spread did not include any of the microclimate variables typically used in wildfire modeling (e.g. wind speed, fuel moisture, fuel temperature

[Rothermel 1972]). It is important to note, however, that prescribed fires at our gradients occurred during moderate meteorological conditions, thereby excluding extremes of fire weather. If we had included a wider range of weather conditions, we would have likely found a stronger influence of meteorological variables on fire spread and a weaker influence of vegetation. Thus, by avoiding prescribed fire during extreme meteorological events, the control of vegetation structure on probability of burning is likely reinforced (Ryan et al.

2013).

It is also noteworthy that bulk density was not selected as a predictor of fire spread despite having been identified as a primary driver of fire spread in similar systems (Hoffmann et al.

2012b; Varner et al. 2015). If we replace the C4 grass cover with bulk density as a predictor in our best model it does not perform as well (marginal R2 0.75 vs 0.47; AIC 259.4 vs 269.9), though we suspect the contribution to fire spread from C4 grass cover includes impacts from bulk density (Hoffmann et al. 2012b).

Implications for vegetation dynamics

Fire-vegetation feedbacks can arise in systems where 1) vegetation composition and/or structure strongly influence flammability (Beckage et al. 2009; Schertzer et al. 2014) and 2) burning fosters vegetation that promotes fire (Mutch 1970; Bond and Keeley 2005; Grigulis et al. 2005; Dantas et al. 2013; Parr et al. 2014). Here we have shown that the first of these conditions is met, and there is strong evidence by others that the second is true. In particular, frequent burning is essential for maintaining longleaf pine savanna; under fire exclusion, these savannas quickly revert to hardwood forest (Gilliam and Platt 1999), similar to changes observed in mesic savannas elsewhere across the globe.

Less attention, however, has been given to understanding fire feedbacks and alternate states in wetland systems, although there are examples in the southeast USA (Drewa et al. 2006;

Martin and Kirkman 2009). In our study system, there are two distinct wetland types that differ in vegetation structure and composition: evergreen shrub-dominated (shrubby) or switchcane-/fern-dominated (herbaceous). Shrubby wetland was less likely to burn than

herbaceous wetland (30.4% vs 55.8%). Consequently, under the current 3-year fire cycle in the uplands, shrubby wetland has a mean expected fire return interval of 9.9 years, almost double that of herbaceous wetland (Fig. 2.4a); these values are within previously estimated intervals of 2-3 years for longleaf pine savanna and 7-50 years for adjacent wetland (Frost

1998; Stambaugh et al. 2011). Moreover, when evergreen shrubs dominate the wetlands, they appear to effectively suppress more-flammable herbaceous species. Although shrubs tend to resprout vigorously after fire (Schafer and Just 2014), there is nevertheless at least a short- term reduction in shrub cover after fire (27% decline one year after fire, not shown).

The location of the flammability transition along the ecotonal gradient is not static, but, on average, results in distinct communities that either promote or hinder fire via feedbacks.

Feedbacks such as these have been noted to result in hysteresis (Wilson and Agnew 1992) in which the current state of a system is conditional on its past, resulting in relatively stable boundaries until an environmental threshold is crossed, for example, increased fire frequency can erode feedback control by influencing vegetation structure (e.g. Gagnon and Platt 2008).

Differences in the proportion of ecotone-extinguished fires between herbaceous (44%) and shrubby (67%) wetland demonstrates the potential for increased fire spread and that system boundaries are not deterministic. However, community boundaries maybe experience periods of stability under certain conditions, given the high resprouting capacity of these shrubs and limited colonization success of grasses (Drewa et al. 2006; Brewer et al. 2009).

In our system, there are three alternate ecotonal gradients that arise from fire-vegetation feedbacks: savanna – herbaceous wetland, savanna – shrubby wetland, or dry forest – shrubby wetland (Online Resource 9). The combined effects of hydrology and fire result in a complex mosaic of forested, shrubby, and herbaceous wetland within a matrix of uplands dominated by savanna interspersed with patches of xeric hardwood forest (known locally as hardwood inclusions). This vegetation heterogeneity illustrates the complex reality of savanna-forest systems that are sometimes described as simple, binary systems governed primarily by fire (Abades et al. 2014), but are commonly underlain by resource gradients

(Hoffmann et al. 2012a; Murphy et al. 2013).

These configurations of the savanna – wetland ecotonal gradient are likely influenced by some important differences between the longleaf pine and other savannas. Unlike the continuous C4 grass layer typical of most savannas (Parr et al. 2014), C4 grass cover was discontinuous in our savannas and does not produce sufficient biomass to leave the distinctive soil isotopic signature typical of savanna (Schafer et al. 2013). Unlike most savannas, however, which rely on C4 grass for the vegetation component of fire feedbacks, longleaf pine litter is also an abundant and highly flammable fuel (Rebertus et al. 1989;

Fonda 2001) that reinforces fire feedbacks (Beckage et al. 2009). Because longleaf pine is tall, relative to the length of the ecotonal gradient, pine leaf litter is dispersed well into the wetlands, enhancing flammability well beyond the extent of C4 grasses. We observed pine litter throughout the gradient (31% of overall fuel mass); its contribution declined from savanna to wetland from 50% to 20% of overall fuel mass, respectively.

Management considerations

In pyrogenic landscapes, the persistence of pyrophobic vegetation patches can present considerable challenges to land managers. Where objectives seek to reverse the effects of past fire suppression, efforts may focus on increasing pyrophilic vegetation (Aschenbach et al. 2010), while in other cases, pyrophobic vegetation may be promoted to provide refugia for fire-sensitive species (Trauernicht et al. 2012). In either case, the balance between pyrophilic and pyrophobic vegetation will be largely dependent upon the transition in flammability at the interface between the two vegetation states. Consequently, the ability to predict the extent of fire spread should prove to be a useful management tool. This is particularly important in the imperiled longleaf pine system (Landers et al. 1995), where biodiversity is a function of numerous distinct communities and ecotones (Kaeser and

Kirkman 2009; Noss et al. 2015). Effective anticipation of changes to pyrophilic-pyrophobic ecosystem boundaries requires better understanding of vegetation dynamics and feedbacks including, importantly, the impact of vegetation on fire spread.

References

Abades SRS, Gaxiola A, Marquet PA (2014) Fire, percolation thresholds and the savanna forest transition: a neutral model approach. J Ecol 102:1386–1393. doi: 10.1111/1365- 2745.12321

Aschenbach TA, Foster BL, Imm DW (2010) The initial phase of a longleaf pine-wiregrass savanna restoration: species establishment and community responses. Restor Ecol 18:762–771. doi: 10.1111/j.1526-100X.2009.00541.x

Atkinson D, Chladil M, Janssen V, Lucieer A (2010) Implementation of quantitative bushfire risk analysis in a GIS environment. Int J Wildl Fire 19:649–658. doi: 10.1071/WF08185

Baeza MJ, Raventós J, Escarré A, Vallejo VR (2006) Fire risk and vegetation structural dynamics in Mediterranean shrubland. Plant Ecol 187:189–201. doi: 10.1007/s11258- 005-3448-4

Bates D, Mächler M, Bolker B, Walker S (2015) Fitting Linear Mixed-Effects Models Using lme4. J Stat Softw. doi: 10.18637/jss.v067.i01

Beckage B, Platt WJ, Gross LJ (2009) Vegetation, fire, and feedbacks: a disturbance- mediated model of savannas. Am Nat 174:805–18. doi: 10.1086/648458

Bond WJ, Keeley JE (2005) Fire as a global “herbivore”: the ecology and evolution of flammable ecosystems. Trends Ecol Evol 20:387–394. doi: 10.1016/j.tree.2005.04.025

Bonham C (2013) Measurements for terrestrial vegetation, Second Ed. Wiley-Blackwell, Hoboken, NJ, USA

Bova AS, Dickinson MB (2008) Beyond “fire temperatures”: calibrating thermocouple probes and modeling their response to surface fires in hardwood fuels. Can J For Res 38:1008–1020. doi: 10.1139/X07-204

Brewer JS, Cunningham AL, Moore TP, Brooks RM, Waldrup JL (2009) A six-year study of fire-related flowering cues and coexistence of two perennial grasses in a wet longleaf pine (Pinus palustris) savanna. Plant Ecol 200:141–154. doi: 10.1007/s11258-008- 9440-z

Burnham K, Anderson D (2002) Model selection and multimodel inference: a practical information-theoretic approach. Springer-Verlag New York, Inc., New York, NY

Calcagno V, de Mazancourt C (2010) glmulti: an R package for easy automated model selection with (generalized) linear models. J Stat Softw 34:1–29. doi: 10.18637/jss.v034.i12

Cheney NP, Gould JS, Catchpole WR (1993) The influence of fuel, weather and fire shape variables on fire-spread in grasslands. Int J Wildl Fire 3:31–44. doi: 10.1071/WF9930031

Christensen NL (2000) Vegetation of the Southeastern Coastal Plain. In: Barbour M, Billings W (eds) North American terrestrial vegetation, Second Ed. Camrbridge University Press, New York, NY, pp 397–448

Cruz MG, Alexander ME (2013) Uncertainty associated with model predictions of surface and crown fire rates of spread. Environ Model Softw 47:16–28. doi: 10.1016/j.envsoft.2013.04.004

Dantas V de L, Batalha M, Pausas J (2013) Fire drives functional thresholds on the savanna- forest transition. Ecology 94:2454–2463.

Drewa PB, Thaxton JM, Platt WJ (2006) Responses of root-crown bearing shrubs to differences in fire regimes in Pinus palustris (longleaf pine) savannas: Exploring old- growth questions in second-growth systems. Appl Veg Sci 9:27. doi: 10.1111/j.1654- 109X.2006.tb00653.x

Dwire KA, Kauffman JB (2003) Fire and riparian ecosystems in landscapes of the western USA. For Ecol Manage 178:61–74. doi: 10.1016/S0378-1127(03)00053-7

Favier C, Chave J, Fabing A, Schwartz D, Dubois MA (2004) Modelling forest–savanna mosaic dynamics in man-influenced environments: effects of fire, climate and soil heterogeneity. Ecol Modell 171:85–102. doi: 10.1016/j.ecolmodel.2003.07.003

Fill JM, Platt WJ, Welch SM, Waldron JL, Mousseau TA (2015) Updating models for restoration and management of fiery ecosystems. For Ecol Manage 356:54–63. doi: 10.1016/j.foreco.2015.07.021

Fonda RW (2001) Burning characteristics of needles from eight pine species. For Sci 47:390–396.

Frost C (1998) Presettlement ﬁre frequency regimes of the United States: a ﬁrst approximation. In: Pruden T, Brennan L (eds) Fire in ecosystem management: shifting the paradigm from suppression to prescription. Tall Timbers Research Station, Tallahassee, FL, pp 70–81

Gagnon PR, Platt WJ (2008) Multiple disturbances accelerate clonal growth in a potentially monodominant bamboo. Ecology 89:612–618. doi: 10.1890/07-1255.1

Gerdes G (2011) Modeling the effects of prescribed burning on ozone precursors at Fort Bragg, NC. US Army Corp of Engineers, PWTB 200-1-82, Washington, DC.

Gilliam F, Platt W (1999) Effects of long-term fire exclusion on tree species composition and stand structure in an old-growth Pinus palustris (longleaf pine) forest. Plant Ecol 1920:15–26. doi: 10.1023/A:1009776020438

Grigulis K, Lavorel S, Davies ID, Dossantos A, Lloret F, Vilà M (2005) Landscape-scale positive feedbacks between fire and expansion of the large tussock grass, Ampelodesmos mauritanica in Catalan shrublands. Glob Chang Biol 11:1042–1053. doi: 10.1111/j.1365-2486.2005.00980.x

Harrell, Jr. FE (2001) Regression modeling strategies: with applications to linear models, logistic regression, and survival analysis. Springer-Verlag New York, Inc., New York, NY

Hoffmann WA, Geiger EL, Gotsch SG, Rossatto DR, Silva LCR, Lau OL, Haridasan M, Franco AC (2012a) Ecological thresholds at the savanna-forest boundary: how plant traits, resources and fire govern the distribution of tropical biomes. Ecol Lett 15:759–68. doi: 10.1111/j.1461-0248.2012.01789.x

Hoffmann WA, Jaconis SY, McKinley KL, Geiger EL, Gotsch SG, Franco AC (2012b) Fuels or microclimate? Understanding the drivers of fire feedbacks at savanna–forest boundaries. Austral Ecol 37:634–643.

Hoffmann WA, Schroeder W, Jackson RB (2002) Positive feedbacks of fire, climate, and vegetation and the conversion of tropical savanna. Geophys Res Lett 29:2052. doi: 10.1029/2002GL015424

Hurvich CM, Tsai C (1993) A corrected Akaike information criterion for vector autoregressive model selection. J Time Ser Anal 14:271–279. doi: 10.1111/j.1467- 9892.1993.tb00144.x

Iverson LR, Yaussy DA, Rebbeck J, Hutchinson TF, Long RP, Prasad AM (2004) A comparison of thermocouples and temperature paints to monitor spatial and temporal characteristics of landscape scale prescribed fires. Int J Wildl Fire 13:311–322. doi: 10.1071/WF03063

2 Johnson PCD (2014) Extension of Nakagawa & Schielzeth’s R GLMM to random slopes models. Methods Ecol Evol 5:944–946. doi: 10.1111/2041-210X.12225

Kaeser M, Kirkman L (2009) Estimating total plant species richness in depressional wetlands in the longleaf pine ecosystem. Wetlands 29:866–874. doi: 10.1672/08-179.1

Kuznetsova A, Brockhoff P, Christensen R (2012) lmerTest: tests for random and fixed effects for linear mixed effect models (lmer objects of lme4 package). R package version 2.0. http://CRAN.R-project.org/package=lmerTest

Landers JL, Van Lear DH, Boyer WD (1995) The longleaf pine forests of the Southeast: requiem or renaissance? J For 93:39–44.

Lashley MA, Chitwood MC, Prince A, Elfelt MB, Kilburg EL, Deperno CS, Moorman CE (2014) Subtle effects of a managed fire regime: A case study in the longleaf pine ecosystem. Ecol Indic 38:212–217. doi: 10.1016/j.ecolind.2013.11.006

Lawrence MG (2005) The relationship between relative humidity and the dewpoint temperature in moist air: a simple conversion and applications. Bull Am Meteorol Soc 86:225–233. doi: 10.1175/BAMS-86-2-225

Lemmon PE (1956) A spherical densiometer for estimating forest overstory density. For Sci 2:314–320.

Martens SN, Breshears DD, Meyer CW (2000) Spatial distributions of understory light along the grassland/forest continuum: effects of cover, height, and spatial pattern of tree canopies. Ecol Modell 126:79–93. doi: 10.1016/S0304-3800(99)00188-X

Martin K, Kirkman L (2009) Management of ecological thresholds to re-establish disturbance-maintained herbaceous wetlands of the south-eastern USA. J Appl Ecol 46:906–914. doi: 10.1111/j.1365-2664.2009.01659.x

Morvan D (2011) Physical phenomena and length scales governing the behaviour of wildfires: a case for physical modelling. Fire Technol 47:437–460. doi: 10.1007/s10694-010-0160-2

Murphy BP, Bowman DMJS (2012) What controls the distribution of tropical forest and savanna? Ecol Lett 15:748–758. doi: 10.1111/j.1461-0248.2012.01771.x

Murphy BP, Bradstock RA, Boer MM, Carter J, Cary GJ, Cochrane MA, Fensham RJ, Russell-Smith J, Williamson GJ, Bowman DMJS (2013) Fire regimes of Australia: a pyrogeographic model system. J Biogeogr 40:1048–1058. doi: 10.1111/jbi.12065

Mutch R (1970) Wildland fires and ecosystems--a hypothesis. Ecology 51:1046–1051. doi: 10.2307/1933631

Nakagawa S, Schielzeth H (2013) A general and simple method for obtaining R2 from generalized linear mixed-effects models. Methods Ecol Evol 4:133–142. doi: 10.1111/j.2041-210x.2012.00261.x

Nelson Jr. R (2000) Prediction of diurnal change in 10-h fuel stick moisture content. Can J For Res 30:1071–1087. doi: 10.1139/x00-032

Noss RF, Platt WJ, Sorrie BA, Weakley AS, Means DB, Costanza J, Peet RK (2015) How global biodiversity hotspots may go unrecognized: lessons from the North American Coastal Plain. Divers Distrib 21:236–244. doi: 10.1111/ddi.12278

O’Brien RM (2007) A caution regarding rules of thumb for variance inflation factors. Qual Quant 41:673–690. doi: 10.1007/s11135-006-9018-6

Parr CL, Lehmann CER, Bond WJ, Hoffmann WA, Andersen AN (2014) Tropical grassy biomes: misunderstood, neglected, and under threat. Trends Ecol Evol 29:205–213. doi: 10.1016/j.tree.2014.02.004

Platt W (1999) Southeastern pine savannas. In: Anderson R, Fralish J, Baskin J (eds) Savanna, barrens and rock outcrop plant communities of North America. Cambridge University Press, New York, NY, pp 23–51

R Development Core Team (2015) R: A language and environment for statistical computing. R Foundation for Statistical Computing, Vienna.

Rebertus A, Williamson G, Moser E (1989) Longleaf pine pyrogenicity and turkey oak mortality in Florida xeric sandhills. Ecology 70:60–70. doi: 10.2307/1938412

Rothermel RC (1983) How to predict the spread and intensity of forest and range fires. USDA Forest Service, Intermountain Forest and Range Experiment Station, INT-143, Ogden, UT

Rothermel RCR (1972) A mathematical model for predicting fire spread in wildland fuels. USDA Forest Service, Intermountain Forest and Range Experiment Station, INT-115, Ogden, UT

Russo L, Russo P, Vakalis D, Siettos C (2014) Detecting weak points of wildland fire spread: a cellular automata model risk assessment simulation approach. Chem Eng Trans 36:253–258. doi: 10.3303/CET1436043

Ryan KC, Knapp EE, Varner JM (2013) Prescribed fire in North American forests and woodlands: history, current practice, and challenges. Front Ecol Environ 11:e15–e24. doi: 10.1890/120329

Schafale M (2012) Guide to the natural communities of North Carolina: fourth approximation. Department of Environment, Health, and Natural Resources, North Carolina Natural Heritage Program, Division of Parks and Recreation, Raleigh, NC

Schafer JL, Breslow BP, Just MG, Hohmann MG, Hollingsworth SN, Swatling-Holcomb SL, Hoffmann WA (2013) Current and historical variation in wiregrass (Aristida stricta) abundance and distribution is not detectable from soil δ13C measurements in longleaf pine (Pinus palustris) savannas. Castanea 78:28–36. doi: 10.2179/12-021

Schafer JL, Just MG (2014) Size dependency of post-disturbance recovery of multi-stemmed resprouting trees. PLoS One 9:e105600. doi: 10.1371/journal.pone.0105600

Schertzer E, Staver AC, Levin S (2014) Implications of the spatial dynamics of fire spread for the bistability of savanna and forest. J Math Biol 329–341. doi: 10.1007/s00285- 014-0757-z

Schwilk DW (2003) Flammability is a niche construction trait: canopy architecture affects fire intensity. Am Nat 162:725–733. doi: 10.1086/379351

Silva LCR, Hoffmann WA, Rossatto DR, Haridasan M, Franco AC, Horwath W (2013) Can savannas become forests? A coupled analysis of nutrient stocks and fire thresholds in central Brazil. Plant Soil 373:829–842. doi: 10.1007/s11104-013-1822-x

Simeoni A, Salinesi P, Morandini F (2011) Physical modelling of forest fire spreading through heterogeneous fuel beds. Int J Wildl Fire 20:625–632. doi: 10.1071/WF09006

Smit IPJ, Asner GP, Govender N, Kennedy-Bowdoin T, Knapp DE, Jacobson J (2010) Effects of fire on woody vegetation structure in African savanna. Ecol Appl 20:1865– 1875. doi: 10.1890/09-0929.1

Sorrie BA, Gray JB, Crutchfield PJ (2006) The vascular flora of the longleaf pine ecosystem of Fort Bragg and Weymouth Woods, North Carolina. Castanea 71:129–161. doi: 10.2179/05-02.1

Stambaugh MC, Guyette RP, Marschall JM (2011) Longleaf pine (Pinus palustris Mill.) fire scars reveal new details of a frequent fire regime. J Veg Sci 22:1094–1104. doi: 10.1111/j.1654-1103.2011.01322.x

Tonidandel S, LeBreton JM (2011) Relative importance analysis: a useful supplement to regression analysis. J Bus Psychol 26:1–9. doi: 10.1007/s10869-010-9204-3

Trauernicht C, Murphy BP, Portner TE, Bowman DMJS (2012) Tree cover-fire interactions promote the persistence of a fire-sensitive conifer in a highly flammable savanna. J Ecol 100:958–968. doi: 10.1111/j.1365-2745.2012.01970.x

Varner JM, Kane JM, Kreye JK, Engber E (2015) The flammability of forest and woodland litter: a synthesis. Curr For Reports 1:91–99. doi: 10.1007/s40725-015-0012-x

Weakley AS, Schafale MP (1991) Classification of pocosins of the Carolina Coastal Plain. Wetlands 11:355–375. doi: 10.1007/BF03160756

Wilson JB, Agnew ADQ (1992) Positive-feedback switches in plant communities. In: Begon M, Fitter A (eds) Advances in Ecological Research. Academic Press London, pp 263– 336

Tables

Table 2.1 Vegetation Structure functional types, microclimate, plot, and transect variables

Vegetation Microclimate Fuels and vegetation Other composition structure Switchcane Air temperature Canopy closure Bearing Deciduous shrub Coarse fuel moisture Fuelbed depth No. of fires Deciduous tree Fine fuel moisture Maximum fire temperature Years since fire Evergreen shrub Fuel temperature Plot cover Evergreen tree Precipitation throughfall Relative gradient position Fern Relative humidity Woody Debris

C4 grass Soil moisture

C3 graminoids Photosynthetically active radiation Herbaceous dicot Vapor pressure deficit Wind direction Wind speed

Table 2.2 Results of linear mixed-effect models evaluating vegetation structure functional type, or plot variable by relative gradient position

Variable Estimate (SE) DF t-value Marginal R2 *** C4 grass cover (%) -16.70 (1.45) 108 -11.6 0.220 Evergreen shrub cover (%) 20.70 (2.15) 110 9.6*** 0.172 Canopy closure (%) 22.42 (2.02) 114 11.1*** 0.125 Plot cover (%) 28.93 (2.76) 113 10.5*** 0.120 Switchcane cover (%) 9.31 (1.18) 112 7.9*** 0.120 Fern cover (%) 12.89 (1.70) 111 7.6*** 0.079 Fuelbed depth (cm) 1.74 (0.21) 111 8.3*** 0.040 Bulk density (kg m-3) -6.07 (1.01) 110 -6.0*** 0.039 Deciduous tree cover (%) -7.42 (1.65) 113 -4.5*** 0.030 Herbaceous dicot cover (%) -2.51 (0.39) 112 -6.4*** 0.030 Deciduous shrub cover (%) 7.65 (1.89) 113 4.1*** 0.023 Evergreen tree cover (%) 3.00 (0.75) 112 4.0*** 0.018 Max. fire temperature (°C) -100.13 (58.86) 36 -1.7NS 0.015 Woody debris cover (%) 1.26 (0.41) 119 3.1** 0.009 NS C3 graminoid cover (%) -0.38 (0.30) 124 -1.3 0.001 ***P < 0.001, **P < 0.01, NS = not significant

Table 2.3 Comparison of savanna and wetland microclimate variable means (Welch’s unequal variance t-test). Collection period was May – August 2012 – 2014

Variable Savanna Wetland DF t-value Mean Mean Rain throughfall (mm d-1) 1.50 0.49 1088 3.9*** Wind (m s-1) 0.89 0.36 1641 40.7*** Air temperature (°C) 20.52 20.03 1690 1.2NS Fuel temperature (°C) 23.46 21.86 1675 3.5*** Relative humidity (%) 62.94 67.20 1695 -4.7*** Coarse fuel moisture (%) 17.60 22.76 1676 -12.4*** Photosynthetically active radiation 563.0 214.4 1038 25.7*** (µmol m-2 s-1) Vapor pressure deficit (kPa) 1.05 0.87 1652 5.5*** Soil moisture (%) 0.16 0.33 1682 -37.3*** Fine fuel moisture (%) 0.17 0.32 1548 -8.9*** ***P < 0.001, NS = not significant

Table 2.4 Model details for predicting the conditional probability of fire spread of a plot (i.e. the preceding plot within the transect was burned). Predictor with coefficient direction, marginal R2, variance influence factor (√VIF score), relative importance weights, and AIC score

Model Scheme Predictor Marginal R2 VIF RIW AIC

Vegetation structure + C4 grass cover (%) 0.706 1.02 0.827* 303.6

- Canopy closure (%) 1.02 0.072NS

- Evergreen shrub cover (%) 1.021 0.101NS

Microclimate + Photosynthetically active 0.484 1.658 0.885* 282.4 radiation (µmol m-2 s-1)

- Vapor pressure deficit (kPa) 1.658 0.115*

Vegetation structure & + C4 grass cover (%) 0.756 1.015 0.671* 259.4 microclimate

- Evergreen shrub cover (%) 1.02 0.068NS

+ Photosynthetically active 1.017 0.26NS radiation (µmol m-2 s-1)

*significant RIW (Tonidandel and LeBreton 2011), NS = RIW not significant

Figures

Fig. 2.2 Mean cover (%) by relative gradient position for the C4 grass and evergreen shrub functional types included in our vegetation structure and microclimate fire spread probability model. Error bars are standard errors of the mean

Fig. 2.3 Violin plots for mean understory photosynthetically active radiation (PAR) and mean coarse fuel moisture (%) between savanna and wetland. The box plots indicate data range, quartiles, and median. Dots are outliers. The density trace is shown in light gray. Data represent the mean values for burn days during the growing season (April – Sept, 10:00-16:00)

Fig. 2.4 (a) Expected fire return interval (FRI) by relative gradient position. (b) Proportion of plots (n=489) burned per relative gradient position. The dashed line represents the FRI for gradients with a shrub-dominant (shrubby) wetland and the solid line represents those with an herbaceous wetland

Fig. 2.5 Lines are predictions of the conditional probability of a plot burning using the vegetation structure and microclimate generalized linear mixed model (Table 2.4). For each plot all predictors were held constant at their mean except for the variable listed on the horizontal axis ([a] C4 grass cover, [b] evergreen shrub cover, [c] photosynthetically active radiation), which varied

Chapter 3: Effects of fire frequency on flammability and vegetation structure along

managed longleaf pine savanna – wetland gradients

Abstract

Positive fire-vegetation feedbacks are important ecological phenomena that maintain vegetation structure distributions between many fire-promoting and fire-deterring communities. Environmental perturbations (e.g. invasive species, land-use change, management) may alter fire regimes, and the ability of these feedbacks to maintain community boundaries under change is not widely known. We developed a cellular automaton that incorporated field-based predictions of flammability and vegetation responses to burning in order to simulate fire spread and vegetation dynamics along managed longleaf pine savanna – wetland ecotonal gradients. We investigated if modified fire frequencies (a proxy for environmental change) could cause disruption in these fire-vegetation feedbacks.

Specifically, we examined if an increase in fire frequency, relative to current fire management, would result in the expansion of fire-promoting vegetation and gradient flammability. Conversely, we tested if reduced fire frequencies would result in diminished flammability and woody encroachment of savanna. In general, vegetation structure rebounded quickly from fire events in our model. Although we found that increasing fire frequency to once every year generally reduced vegetation abundances along the gradient, it also increased gradient flammability. Moreover, we found that the initial starting conditions of vegetation were more important for vegetation structure boundaries than fire frequency.

We observed only modest changes in vegetation structure in response to modified fire

frequency, but our simulations still revealed an inverse relationship between fire frequency and gradient flammability. As such, we suspect that vegetation structure boundaries are mutable, especially under sustained fire frequency alterations.

Introduction

What governs vegetation structure (i.e. species composition and physical structure) distribution? This is one of the fundamental questions in terrestrial ecology; delineations of vegetation distributions (e.g. communities, biomes) are the cornerstones upon which many ecological questions are based. Globally, the use of climate alone has been largely successful in predicting biome distributions (Whittaker 1975, Stephenson 1990). More recent efforts have demonstrated a mismatch between extant global vegetation distributions and the distributions predicted from the climate’s potential, notably, the current extent of forest cover is less than climate alone predicts (Bond et al. 2005). One explanation for this mismatch is fire (Bond and Keeley 2005), an important driver of vegetation distribution in many ecosystems (Lehmann et al. 2014). For example, the current extent of savanna and other grassy ecosystems, are greater than would be expected based on climate only predictions

(Bond et al. 2005).

The vegetation structure of savanna ecosystems (i.e. open canopy and continuous herbaceous understory) is maintained by positive fire-vegetation feedbacks (e.g. Beckage et al. 2009, Hoffmann et al. 2012b). Fire only has direct control over the vegetation that it burns. Fire is able to maintain savanna vegetation in locations where burning is episodic, and in turn, completing the feedback, the vegetation structure of savanna promotes fire (Wilson and Agnew 1992, Murphy and Bowman 2012). However, many savannas are adjacent to, or transition into, closed-canopy, pyrophobic communities (e.g. Kirkman et al. 1998, Staver et al. 2011, Ibanez et al. 2013). Contrastingly, these pyrophobic vegetation communities (e.g. forest) exhibit positive feedbacks that hinder fire (Hoffmann et al. 2012b). Fire-vegetation

feedbacks maintain vegetation structure distributions between pyrophilic and pyrophobic communities (Zedler et al. 1983, Pausas et al. 1999). However, the control of these feedbacks is not absolute. Long-term fire suppression, for example, has altered vegetation structure along savanna – forest gradients across the globe (e.g. Van Langevelde et al. 2003, Varner et al. 2005, Müller et al. 2007, Hoffmann et al. 2012a), including the longleaf pine (Pinus palustris) ecosystems of the southeastern United States (Gilliam and Platt 1999).

Longleaf pine ecosystems are some of the most species-rich systems in North

America, but are presently imperiled (Noss et al. 2015). The current areal extent of longleaf pine ecosystems is estimated at 3% of their historical maximum (37 million ha), with reductions attributed to fire suppression and land-use change (Frost 1993, Simberloff 1998).

Longleaf pine savanna relies on episodic fire (Wells and Shunk 1931, Chapman 1932,

Landers et al. 1995) and associated positive fire-vegetation feedbacks for persistence (Fill et al. 2012). These savannas have an estimated historical fire return interval ranging from 1-12 years (Myers and White 1987, Stambaugh et al. 2011), whereas adjacent, closed-canopy wetland communities have been estimated to have burned every 7-50 years (Frost 1993). The ability of fire-vegetation feedbacks, under environmental perturbation, to maintain these savanna – wetland vegetation structural boundaries is not generally known (Just et al. in press).

There are numerous models that explore savanna – forest vegetation structure boundary dynamics (e.g. Beckage and Ellinwood 2008, Hoffmann et al. 2009, Hirota et al.

2011, Wakeling et al. 2011). Many of these models use differential equations (e.g. Beckage et al. 2009, Staver and Levin 2012), that utilize mass-action dynamics that assume all

components (e.g. fires, grasses, trees) interact equally; however, this assumption may be inappropriate for modeling the observed dynamics in longleaf pine savanna – wetland gradients (e.g. Drewa et al. 2002b, Thaxton and Platt 2006, Wenk et al. 2011), where patchiness in both vegetation structure and flammability are common (Fill et al. 2015). In addition, these models do not generally consider the heterogeneity (i.e. varying environmental features) of savanna-forest gradients and how this may affect fire-vegetation feedbacks (Just et al. in press). Furthermore, when fire spread is present in these modeling efforts, it is usually deterministic, although some stochastic examples exist (e.g. Beckage et al. 2011). These modeling structures are too coarse for longleaf pine savanna - wetland vegetation gradients, whose vegetation distributions differ from generalizations of abrupt transitions between savanna and forest due to their lack of vegetation structure extremes

(Kirkman et al. 1998); for example, they do not have a continuous grass layer as is common in many other savannas (Parr et al. 2014).

Due to the deficiencies of available models, we develop our own model to simulate fire-vegetation dynamics, using both fire regime (i.e. fire frequency) and vegetation as drivers, which are considered to be the most important factors for positive feedbacks in pine savannas (Fill et al. 2015). We investigate the effects of fire frequency on fire spread and vegetation community boundaries along a longleaf pine savanna – wetland ecotonal gradient, using a stochastic cellular automaton parameterized with field-based data. Fire frequency has been successfully used as a proxy of environmental change to model vegetation shifts in other pyrophilic systems (e.g. Pausas et al. 1999, King et al. 2013). We previously collected fire spread, vegetation structure, and microclimate data along this gradient and developed a

statistical model to predict the probability of fire spread (Just et al. in press). We address two questions: 1) Do increased fire frequencies result in both the downslope movement of savanna vegetation and increased fire spread?, and 2) Does a reduction in fire frequency result in the encroachment of savanna by woody vegetation, with a concomitant reduction in fire spread extent?

Materials and Methods

Study area

Field data were collected at Fort Bragg (73,469 ha), a US Army installation, situated in the

Sandhills physiographic region of North Carolina, USA (35°07’N, 79°10’W). The dominant vegetation community at Fort Bragg is longleaf pine – wiregrass (Aristida stricta) savanna

(Sorrie et al. 2006), typified by a longleaf pine canopy and wiregrass understory. Savanna occurs on upland, xeric ridges, compromised of sandy soils, remnants of pre-historical

Atlantic coastal features (Christensen 2000). Across Fort Bragg, the savanna is dissected in numerous locations by wetland (Schafer et al. 2013), known locally as streamhead pocosin and sandhill seep (Weakley and Schafale 1991). An impermeable clay soil layer exists beneath portions of the Sandhills, and these clay layers result in the lateral movement of ground water (Oliver 1978), which exits at clay outcroppings, or areas of topographic relief, as seepage, resulting in wetland vegetation communities (Brinson 1991). These relatively more productive wetlands are generally characterized by shade-tolerant woody species represented by a shrub understory and hardwood canopy (Ames et al. 2015). Although some

wetlands, particularly those with a fierier past, consist primarily of herbaceous vegetation

(e.g. switchcane [Arundinaria tecta]) and open canopies (Sorrie et al. 2006).

To facilitate fire management at Fort Bragg, managers have segmented the landscape into burn compartments. Prescribed fire is applied to approximately one-third of burn compartments annually; this 3-yr frequency is based on estimated mean historical fire return intervals for longleaf pine savanna (Stambaugh et al. 2011). Prescribed fires are ignited in savanna and burn downslope towards wetland (Lashley et al. 2014).

Field data

We established 113 permanent transects along savanna – wetland ecotonal gradients in 73 burn compartments distributed over 230 km2 of Fort Bragg. Each transect began in upland savanna and transitioned downslope into wetland vegetation communities. The mean minimum distance between transects was approximately 302 m. We established 2 m3 (n =

532) plots every (approximately) 10 m along each transect and recorded vegetation structure and fire spread. Total transect length varied due to natural gradient variation (mean length:

37.4 m), as such we recorded the position of each plot relative to the most upland plot (i.e. relative gradient position, range: 0 – 1). When using transects in aggregate we calculated average values using the following gradient position bins: 0.00 to 8.00%, 8.01 to 37.50%,

37.50 to 62.50%, 62.51 to 88.50%, and 88.51 to 100% We visually estimated the percent cover of ten vegetation structure functional types in 5% increments (0-100%), with an additional class (1%) for presence with negligible cover. Heights (cm) per functional type per plot were also recorded. Functional-type plot volume was calculated as the height × cover.

Transects were censused for vegetation structure cover once per year (May – July) from 2012

– 2014. We erected meteorological stations to characterize ten microclimatic variables across representative ecotonal gradients (2-3 gradients concurrently) from 15 May 2012 – 30

September 2014. Each representative ecotone had a paired microclimate stations, with a station in both savanna and wetland. Paired stations were approximately 50 m apart, and values for intermediate gradient positions were linearly interpolated. Following prescribed fire (2012 -2014), the extent of fire spread (length [m]) along each burned transect was measured. More details about our field campaign and data collection efforts can be found in

Just et al. (in press).

Gradient flammability model

We previously investigated the vegetation structure and microclimate variables that best described the conditional probability of burning (i.e. the probability of burning, given that the preceding, upslope plot had burned) along these study gradients. We used a genetic- algorithm (Calcagno and de Mazancourt 2010) to identify the model that best predicted flammability (generalized linear mixed-effect model [GLMM], with site as a random effect

[logistic regression with a logit link]). The random site effect was not significant and we procced here with a more parsimonious GLM. Additional details about the model construction and selection can found in Just et al. (in press). Our final model included three

-2 -1 factors, C4 grass cover (%) and photosynthetically active radiation (µmol m s ), which promoted fire spread, and evergreen shrub cover (%) which hindered fire spread. To better capture the size response of vegetation structure response to fire, we used vegetation volume

(m3) in our model simulations. Over 65% of variation in this model was explained by vegetation structure alone, and since PAR is modulated by vegetation structure (Martens et al. 2000), we chose to only include vegetation structure factors to assign burn probabilities

(Fig. 3.1) in our model simulations.

Model description

We simulated the dynamics of flammability and vegetation along savanna – wetland gradients under varying fire frequencies using a stochastic cellular automaton (Fig. 3.2).

Model instances were initialized with one of three vegetation structure schemes, and each were run with two sets of rules (i.e. field-based and enhanced feedback) that determined flammability and vegetation response to burning. We randomly selected 100 of our field transects to serve as the templates upon which the model operates, as such, we constructed

100 1 × c (range of c: 4-6) lattices, each cell (n = 460) within the lattice represented a discrete volume of 2 m3. Therefore, the first cell in each lattice represented savanna and final cell, wetland. We used the same spatial resolution as our field observations (i.e. 2 m3 plot,

~10 m apart) for our lattices, because our statistical models of flammability and vegetation responses were developed at that scale.

Each lattice was initialized in one of three vegetation structure schemes, which differed in starting volumes of both C4 grass and evergreen shrub. Under the Field 2012 scheme, lattices were initiated with our observed (2012) vegetation structure values. For the

3 High Grass scheme, all cells were initiated with a high value for C4 grass (0.25 m ) and a low value for evergreen shrub (0.025 m3), and vice versa for the High Shrub scheme.

Each cell was assigned a probability of burning (time t) derived from the GLM model, which used C4 grass and evergreen shrub volume as predictors. Flammability was stochastic and conditional: if a plot did not burn, the fire did not spread to remaining downslope cells in the lattice (Fig. 3.1). Fires were initiated in the first cell of a lattice; this design emulates the prescribed fires used to manage these landscapes where fires are ignited in the upland savanna and burn towards wetland (Ames et al. 2015). After fire spread was complete, the burn status (i.e. burned or unburned) of each cell was updated. For time steps (e.g. year) in the simulation model without fire, each cell was assigned an unburned status. Burn probabilities were updated after each time step for each cell based on vegetation structure values at time t+1 (Fig. 3.2).

After each time step, vegetation responses to burn status for each cell were updated

(Fig. 3.2). The models used to update vegetation values were developed from our field data

(2012-2014). For non-zero starting values, vegetation volumes (time t+1) were determined from a linear regression (one for each vegetation type and burn status [n = 4]), which used relative gradient position and vegetation volume (time t) as predictors. Each cell with a value of zero (time t) was assigned a probability of becoming non-zero in the next time step. Using our 2012-2014 field data, these probabilities were calculated for each vegetation structure type by burn status using logistic regression, where the probability of becoming non-zero was based on relative gradient position. These probabilities were assigned to each cell and were used throughout the model instance. The non-zero values for successful cells were derived from the linear regression described above. Finally, there was also a chance for a cell with a non-zero value for either vegetation structure type to become zero. The probability per cell,

for a given vegetation type, to be reduced to zero from a non-zero value was calculated with a logistic regression model like above, with the addition of second predictor term, vegetation structure volume at time t. The maximum volume of either vegetation structure type was 2 m3.

We did not observe C4 grasses and evergreen shrubs co-occurring at high abundances along our study ecotones (Fig. 3.3A), as is common when comparing vegetation communities along environmental gradients (McCune and Root 2015). To account for this in our model, we constructed constraints to preclude co-occurrence at high abundances within our cells.

Constraints were set for each vegetation structure type as the observed value at which the

3 other type did not occur above negligible volumes (i.e. C4 grass = 0.05 m , evergreen shrub =

0.2 m3). If at the end of time step a cell exceeded our thresholds for co-occurrence, the vegetation type that had reached the threshold value first retained its value, and the other was reduced to 60% of its threshold volume value.

Using the same model template already described, we again simulated the dynamics of flammability and vegetation structure, but with enhanced fire-vegetation feedbacks. These model instances were performed to ensure that the modeling framework could indeed result in the movement of vegetation boundaries and/or fire spread extent, given the aforementioned subdued vegetation structure extremes at these gradients (e.g. Kirkman et al.

1996). To this end, the feedback was enhanced by manipulating both flammability and the vegetation response to burn status. For model instances using the enhanced rules, there were three possible probabilities of burning for a given cell: 1) C4 grass and evergreen shrub = 0

3 m (probability of burning = 75%), 2) C4 grass ≥ evergreen shrub volume (99.99%), and 3)

evergreen shrub > C4 grass volume (50%). Vegetation responses after burning were enhanced

3 as follows: if C4 grass (time t) > 0 m , C4 grass (time t+1) = 1.25 × C4 grass (time t),

3 otherwise C4 grass (time t+1) = mean observed 2012 C4 grass value × 1.025 (i.e. 0.010 m ); evergreen shrub (time t+1) = 0.75 × evergreen shrub (time t). Vegetation responses to no

3 burning were enhanced as follows: if evergreen shrub (time t) > 0 m , evergreen shrub (time t+1) = 1.025 × evergreen shrub (time t), otherwise evergreen shrub (time t+1) = mean

3 observed 2012 evergreen shrub value × 1.025 (i.e. 0.019 m ); C4 grass (time t+1) = 0.975 ×

C4 grass (time t).

For each vegetation structure initialization scheme (n = 3), we performed model simulations using both our field-based and enhanced probability of fire and vegetation response rules (n = 2) and each scheme – rules pair was run with four fire frequencies, that is a fire was initiated in the savanna cell of each lattice every 1, 3, or 9 time steps, or never (n =

4). In total, we developed 24 (3 × 2 × 4) model instances, each ran for 200 time steps, and we conducted 10 simulation trials for each model instance. Model outputs were recorded as the mean of the trials. All analyses and model simulations were performed in R 3.2.2 (R Core

Team 2015).

Model output

Vegetation structure responses were recorded as the mean of the lattices by relative gradient position. First, we calculated the mean proportion of the cell occupied by either vegetation structure type. We also calculated the proportion of cells that contained greater

3 than negligible volumes of either vegetation structure type (i.e. C4 grass > 0.004 m ,

evergreen shrub > 0.019 m3). We report two flammability results, first, mean fire return interval, which was calculated as the number of time steps between fires by fire frequency and relative gradient position, and recorded as the mean of all lattices. We also calculated the proportion of burned cells per lattice by fire frequency and gradient position, and recorded value as the mean of all lattices.

Results

Field-based rules

The mean proportion of the cells occupied by C4 grass or evergreen shrub varied among model instances (Fig. 3.4). For example, C4 grass mean cell proportion ranged from 10.51% to 19.61% in the savanna cell, for the 1-yr fire frequency and no fire model instances, respectively. In both of these instances, the mean C4 grass proportion was approximately

0.1% in the wetland cell. Contrastingly, evergreen shrub mean cell proportion ranged from

18.54% to 32.68% in the wetland cell, for the 1-yr fire frequency and no fire model instances, respectively. For these instances, the mean evergreen shrub proportion of the savanna cell was 0.00% for the 1-yr fire frequency, and for no fire 1.07%. Between initialization schemes, the mean cell proportion of vegetation structure types along the gradient varied within fire frequency groups more so than by initialization scheme (Fig. 3.4). In general, a 1-yr fire frequency reduced the mean cell proportion that either vegetation structure type occupied, as compared to the other fire frequencies. Our observed 2012 values for the mean proportion of plot occupied by a vegetation structure type can be found in Fig. 3.3A.

The mean proportion of cells with non-negligible volume for vegetation structure types, followed similar patterns to the mean occupied proportion of the cell (Fig. 3.5). The maximum mean proportion of cells with non-negligible C4 grass volume ranged from

74.20% to 85.20% in the savanna gradient position, for the 1-yr fire frequency and no fire model instances, respectively, with proportions ≤ 0.1% in the wetland lattice position. For evergreen shrubs, the maximum mean proportion of cells with volume ranged from 48.15% to 55.62% in the wetland cell, for the 1-yr fire frequency and no fire simulation instances, respectively. For these instances, the mean proportion of cells with evergreen shrub volume in the savanna gradient position was 0.00% for the 1-yr fire frequency, and for no fire 2.25%.

Between initialization schemes, the mean proportion of cells with non-negligible vegetation volume along the gradient, again, varied within fire frequency groups more so than by scheme (Fig. 3.5). However, the High Grass scheme did have a greater mean value for non- negligible C4 grass volume, at all gradient positions, than the other two schemes. The proportion of cells with non-negligible evergreen shrub volume was generally greater than the other two schemes, except for the first two gradient positions for the 1-yr fire frequency instance. Our observed 2012 values for the mean proportion of plots with non-negligible volumes can be found in Fig. 3.3B.

The mean fire return interval varied along the gradient and by fire frequency (Fig.

3.6). In the savanna lattice position, the mean fire return interval by fire frequency was 1.05,

3.09, and 9.23 years for 1-yr, 3-yr, and 9-yr fire frequencies, respectively. Whereas the mean fire return interval for the wetland lattice position was approximately double (1.75, 5.60,

17.30 years; 1-, 3-, 9-yr fire frequencies, respectively). The mean fire return intervals

between Field 2012 and High Shrub by lattice initialization schemes and fire frequencies were very similar (Fig 6), but were reduced for the High Grass scheme (e.g. 12.94 years [9-yr fire frequency]) relative to the other schemes. The expected fire return interval from our observed data (2012-2014) is displayed in Fig. 3.7A.

The mean proportion of cells burned per gradient position varied by fire frequency

(Fig. 3.8). At the savanna position, the mean proportion of burned cells ranged from 95.14% under a 1-yr fire frequency to 10.73% under a 9-yr fire frequency. In the wetland, the mean proportion of burned cells ranged from 57.69 – 5.92% for 1- and 9-yr fire frequencies, respectively. The mean proportion of burned cells was approximately the same for the Field

2012 and High Shrub initialization schemes (Fig. 3.8). The High Grass scheme resulted in a greater mean proportion of cells burned across the gradient, for example, under a 1-yr fire frequency 74.37% of wetland cells burned. The mean proportion of cells burned from our observed data (2012-2014) is presented in Fig. 3.7B.

Enhanced rules

The results from model instances that used the enhanced-feedback rules for updating flammability and vegetation responses to burn status had results that differed from our model instances that used field-based rules. In general, vegetation responses under the enhanced feedback varied more by both initialization scheme and fire frequency than our field-based model rules (e.g. Figs. 3.4, 3.9). For example, increasing fire frequency resulted in greater mean proportions of C4 grass volume per cell (Figs. 3.9, 3.10), whereas mean evergreen shrub cell proportion generally had a negative relationship with fire frequency. At the

extremes of fire frequency, once per year or never, we saw marked differences from our field-based simulations. Namely, without fire, C4 grass was all but absent from the landscape.

With fires initiated every year, the mean proportion of C4 grass was greater at every gradient position as compared to our field-based simulations. Under enhanced feedbacks and 1-yr fire frequency, C4 grass was greater than evergreen shrub for each initialization scheme and gradient position, except for wetland within the enhanced Field 2012 scheme (i.e. C4 grass

6.17%, evergreen shrub 9.98%). The proportion of cells with non-negligible volume also differed between the enhanced and field-based model rules (Fig. 3.10). Notably, under the

High Shrub scheme with no fire, the mean proportion of plots with non-negligible evergreen shrub volume was greater than 90% at all gradient positions and C4 grass was absent. Under the High Grass scheme with a 1-yr fire frequency, the mean proportion of plots with C4 grass volume was greater than 96% for all positions along the gradient and evergreen shrub was absent from the landscape.

The mean fire return intervals along the gradient under enhanced-feedback model rules generally increased by both fire frequency and initialization scheme, as compared to output from our field-based model rules (Fig. 3.6). Among the schemes, enhanced High

Shrub had the longest mean fire return intervals, and enhanced High Grass the shortest, for example, under the 9-yr fire frequency the mean fire return interval in the wetland ranged from 50.0 - 62.1 years, for enhanced High Grass and Shrub schemes, respectively. Under the enhanced model rules, the mean proportion of plots burned followed similar patterns to that of the field-based model results (Fig. 3.7). Although there were some differences in magnitude, for example, the High Grass scheme increased the mean proportion of plots

burned under a 1-yr fire frequency, and a decrease was observed for the High Shrub scheme under a 9-yr fire frequency.

Discussion

We created a model to examine the effects of fire frequency on flammability and vegetation responses to burning along a longleaf pine savanna – wetland ecotonal gradient. We were interested in determining if increased fire frequencies would result in greater downslope abundances of C4 grass and increased gradient flammability. Our empirical based model simulations revealed that increasing fire frequency generally resulted in the reduction of C4 grasses (Fig. 3.4), as well as evergreen shrubs. The mean proportion of cells with non- negligible C4 volumes under the 1-yr fire frequency was slightly reduced along the downslope half of the gradient as compared to the other fire frequencies, whose values were nearly identical (Fig. 3.5). Increasing the fire frequency decreased the mean fire return interval at all positions along the gradient (Fig. 3.6), and, importantly, the mean proportion of cells burned was markedly increased along the entire gradient (Fig. 3.8). These results did not demonstrate an appreciable downslope increase in C4 grass abundance. Nevertheless, compared to evergreen shrub, relatively small volumes of C4 grass practically guarantee that a cell will burn (Fig. 3.1), and we did observe greater proportions of burned cells under more frequent fires (Fig. 3.8).

We also wanted to know if a reduction in fire frequency would result in decreased gradient flammability and the encroachment of savanna by evergreen shrubs. Decreased fire frequencies (including no fire) did result in slight increases in the mean proportion of

evergreen shrub volume per cell (Fig. 3.4), as well as slight increases in the mean proportion of cells with non-negligible evergreen shrub volume (Fig. 3.5), along the entire gradient.

Decreased fire frequencies also resulted in marked reductions in the mean proportion of burned cells. For example, an average 95% of cells burned in the savanna under a 1-yr fire frequency as compared to 11% under a 9-yr fire frequency (Fig. 3.8). A reduction in fire frequency did slightly increase abundances of evergreen shrub along the gradient, which included minor upslope encroachment, but the more prominent result was the reduced proportion of burned cells.

Using our field-based model rules, manipulating fire frequencies did not result in the profound movement of vegetation boundaries. Moreover, vegetation structure rebounded quickly from fire events (Figs. B.1-B.8), indicating that the distribution of vegetation structure along these ecotonal gradients is relatively robust to changes of fire frequency alone. Instead, it may be that the distribution of vegetation structure is determined by fire in conjunction with other environmental factors (e.g. soil type and soil moisture, Skinner and

Sorrie 2002, Peet 2006). Moreover, the vegetation structure stability we observed under these conditions may be due to the poor colonization success of grasses (Brewer et al. 2009) and the impressive resprouting ability of shrubs (Drewa et al. 2006) in longleaf pine savanna and adjacent wetlands. Unsolved by our models, then, is the scenario surrounding the ascension of wetlands with open canopies and herbaceous understories from woody, closed-canopy wetlands.

Canebrakes (i.e. switchcane monocultures) are an example of wetlands with open vegetation structure, and they require recurrent fires (Platt and Brantley 1997, Sorrie et al.

2006). Periodic fires repeatedly topkill resprouting woody plants, keeping them suppressed, but do not generally result in high levels of woody plant mortality (Olson and Platt 1995). In the absence of frequent fires, these herbaceous wetlands do revert to woody domination

(Hughes 1966) from the suppressed, but persistently resprouting shrubs (Drewa et al. 2002b).

While we did not observe the complete removal of evergreen shrub under increased fire frequencies in our Field 2012 initiated model instances, we do not expect that woody vegetation would persist indefinitely under prolonged periods of frequent (topkilling) fires.

While, we did not explicitly model switchcane, we know that grasses in the systems are poor colonizers, such that a sufficiently long period of shrub suppression may be necessary for grasses at low abundances to spread vegetatively to a newly opened wetland (e.g. Gagnon

2009).

As expected, the High Grass and High Shrub schemes each had higher abundances of

C4 grass and evergreen shrub, respectively. Importantly, our model results highlight the importance of initial vegetation conditions, as they, too, were relatively stable (Figs. B.1-

B.8), meaning that where you start is a good indication of where you end, at least under these moderate (i.e. mean-based) fire-management conditions (Figs. 3.9, 3.10). These results further suggest that the distribution of vegetation is a complex phenomenon along these gradients, and may include factors in addition to fire, such as land-use history (e.g. Veldman et al. 2014)

The results of the enhanced feedback model instances more closely resembled the modeling efforts of other savanna systems (e.g. Beckage et al. 2009, 2011), where changes in fire frequency do enhance the fire-vegetation feedback and result in switches of vegetation

structure/community (Wilson and Agnew 1992). For example, under a 1-yr fire frequency and Field 2012 initialization, C4 grass quickly dominates evergreen shrub at all gradient positions (Fig. B.1), except in the wetland where they are effectively equivalent in abundance with evergreen shrubs. The opposite is true under fire suppression, where evergreen shrub dominates (Fig. B.4). Initial vegetation structure conditions were important under the enhanced rules too, and they accentuated the already enhanced feedback, leading to quicker domination by respective vegetation types (Figs. B.1-B.8). The enhanced-feedback results indicate the our modeling framework performed as intended, and it also confirms that these gradients do not follow the same patterns that lead to vegetation structure extremes found along other savanna – forest gradients.

There may be differences between predicting vegetation response to burning and predicting flammability. In our model, predicting flammability was only based on C4 grass and evergreen shrub volume, because these were previously identified as the most important predictors. On the other hand, our vegetation structure responses to burning were based on relative gradient position and vegetation volume at time t. We did not previously find other environmental factors important for predicting flammability, but that does not rule out the possible existence of such factors (e.g. soil type, soil moisture) that we did not account for in this model. However, our observed field values (Fig. 3.3) were similar to our model (i.e.

Field 2012 scheme, field-based rules [Figs. 3.4, 3.5]) results, suggesting that this model is a reasonable approximation of reality. For example, within longleaf pine savannas, it has been noted that woody species’ distributions respond more to changes in fire frequency than do herbaceous species (Drewa et al. 2002a), and this is what we observed in our model output

(Figs. 3.4, 3.5). In terms of the distribution of these two vegetation types, the gradient appeared to be relatively stable, but even small changes in vegetation structure can have large impacts on fire spread (Just et al. in press), suggesting that environmental perturbations might disrupt the feedback, even if extreme changes in vegetation structure are not quickly observable.

References

Ames, G. M., D. L. Vineyard, S. M. Anderson, and J. P. Wright. 2015. Annual growth in longleaf (Pinus palustris) and pond pine (P. serotina) in the Sandhills of North Carolina is driven by interactions between fire and climate. Forest Ecology and Management 340:1–8.

Beckage, B., and C. Ellinwood. 2008. Fire feedbacks with vegetation and alternative stable states. Complex Systems 18:159–173.

Beckage, B., L. J. Gross, and W. J. Platt. 2011. Grass feedbacks on fire stabilize savannas. Ecological Modelling 222:2227–2233.

Beckage, B., W. J. Platt, and L. J. Gross. 2009. Vegetation, fire, and feedbacks: a disturbance-mediated model of savannas. The American Naturalist 174:805–18.

Bond, W. J., and J. E. Keeley. 2005. Fire as a global “herbivore”: the ecology and evolution of flammable ecosystems. Trends in Ecology & Evolution 20:387–394.

Bond, W. J., F. I. Woodward, and G. F. Midgley. 2005. The global distribution of ecosystems in a world without fire. New Phytologist 165:525–538.

Brewer, J. S., A. L. Cunningham, T. P. Moore, R. M. Brooks, and J. L. Waldrup. 2009. A six-year study of fire-related flowering cues and coexistence of two perennial grasses in a wet longleaf pine (Pinus palustris) savanna. Plant Ecology 200:141–154.

Brinson, M. M. 1991. Landscape properties of pocosins and associated wetlands. Wetlands 11:441–465.

Calcagno, V., and C. de Mazancourt. 2010. glmulti: an R package for easy automated model selection with (generalized) linear models. Journal of Statistical Software 34:1–29.

Chapman, H. H. 1932. Is the longleaf type a climax? Ecology 13:328–334.

Christensen, N. L. 2000. Vegetation of the southeastern coastal plain. Pages 397–448 in M. Barbour and W. Billings, editors. North American terrestrial vegetation. Second Ed. Camrbridge University Press, New York, NY.

Drewa, P. B., W. J. Platt, and E. B. Moser. 2002a. Community structure along elevation gradients in headwater regions of longleaf pine savannas.

Drewa, P. B., W. J. Platt, E. B. Moser, and B. Rouge. 2002b. Fire effects on resprouting of shrubs in headwaters of southeastern longleaf pine savannas. Ecology 83:755–767.

Drewa, P. B., J. M. Thaxton, and W. J. Platt. 2006. Responses of root-crown bearing shrubs to differences in fire regimes in Pinus palustris (longleaf pine) savannas: exploring old- growth questions in second-growth systems. Applied Vegetation Science 9:27.

Fill, J. M., W. J. Platt, S. M. Welch, J. L. Waldron, and T. A. Mousseau. 2015. Updating models for restoration and management of fiery ecosystems. Forest Ecology and Management 356:54–63.

Fill, J., S. Welch, J. Waldron, and T. Mousseau. 2012. The reproductive response of an endemic bunchgrass indicates historical timing of a keystone process. Ecosphere 3:1– 12.

Frost, C. 1993. Four centuries of changing landscape patterns in the longleaf pine ecosystem. Pages 17–43 Proceedings of the Tall Timbers fire ecology conference.

Gagnon, P. R. 2009. Fire in floodplain forests in the southeastern USA: Insights from disturbance ecology of native bamboo. Wetlands 29:520–526.

Hirota, M., M. Holmgren, E. H. Van Nes, and M. Scheffer. 2011. Global resilience of tropical forest and savanna to critical transitions. Science 334:232–5.

Hoffmann, W. A., E. L. Geiger, S. G. Gotsch, D. R. Rossatto, L. C. R. Silva, O. L. Lau, M. Haridasan, and A. C. Franco. 2012a. Ecological thresholds at the savanna-forest boundary: how plant traits, resources and fire govern the distribution of tropical biomes. Ecology letters 15:759–68.

Hoffmann, W. A., S. Y. Jaconis, K. L. McKinley, E. L. Geiger, S. G. Gotsch, and A. C. Franco. 2012b. Fuels or microclimate? Understanding the drivers of fire feedbacks at savanna–forest boundaries. Austral Ecology 37:634–643.

Hughes, R. 1966. Fire ecology of canebrakes. Pages 149–158 Proceedings Fifth Annual Tall Timbers Fire Ecology Conference. Tall Timbers Research, Inc., Tallahassee, FL.

Just, M. G., M. G. Hohmann, and W. A. Hoffmann. in press. Where fire stops: vegetation structure and microclimate influence fire spread along an ecotonal gradient. Plant Ecology.

King, D. A., D. M. Bachelet, and A. J. Symstad. 2013. Climate change and fire effects on a prairie-woodland ecotone: projecting species range shifts with a dynamic global vegetation model. Ecology and evolution 3:5076–97.

Kirkman, L., M. Drew, L. West, and E. Blood. 1998. Ecotone characterization between upland longleaf pine/wiregrass stands and seasonally-ponded isolated wetlands. Wetlands 18:346–364.

Kirkman, L., R. Lide, G. Wein, and R. Sharitz. 1996. Vegetation changes and land-use legacies of depression wetlands of the western coastal plain of South Carolina: 1951– 1992. Wetlands 16:564–576.

Landers, J. L., D. H. Van Lear, and W. D. Boyer. 1995. The longleaf pine forests of the Southeast: requiem or renaissance? Journal of Forestry 93:39–44.

Van Langevelde, F., C. A. D. M. Van De Vijver, L. Kumar, J. Van De Koppel, N. De Ridder, J. Van Andel, A. K. Skidmore, J. W. Hearne, L. Stroosnijder, W. J. Bond, H. H. T. Prins, and M. Rietkerk. 2003. Effects of fire and herbivory on the stability of savanna ecosystems. Ecology 84:337–350.

Lashley, M. A., M. C. Chitwood, A. Prince, M. B. Elfelt, E. L. Kilburg, C. S. Deperno, and C. E. Moorman. 2014. Subtle effects of a managed fire regime: A case study in the longleaf pine ecosystem. Ecological Indicators 38:212–217.

Lehmann, C. E. R., T. M. Anderson, M. Sankaran, S. I. Higgins, S. Archibald, W. A. Hoffmann, N. P. Hanan, R. J. Williams, R. J. Fensham, J. Felfili, L. B. Hutley, J. Ratnam, J. San Jose, R. Montes, D. Franklin, J. Russell-Smith, C. M. Ryan, G. Durigan, P. Hiernaux, R. Haidar, D. M. J. S. Bowman, and W. J. Bond. 2014. Savanna vegetation-fire-climate relationships differ among continents. Science 343:548–552.

Martens, S. N., D. D. Breshears, and C. W. Meyer. 2000. Spatial distributions of understory light along the grassland/forest continuum: effects of cover, height, and spatial pattern of tree canopies. Ecological Modelling 126:79–93.

McCune, B., and H. T. Root. 2015. Origin of the dust bunny distribution in ecological community data. Plant Ecology 216:645–656.

Müller, S. C., G. E. Overbeck, J. Pfadenhauer, and V. D. Pillar. 2007. Plant functional types of woody species related to fire disturbance in forest-grassland ecotones. Plant Ecology 189:1–14.

Murphy, B. P., and D. M. J. S. Bowman. 2012. What controls the distribution of tropical forest and savanna? Ecology Letters 15:748–758.

Myers, R. L., and D. L. White. 1987. Landscape history and changes in Sandhill vegetation in north-central and south-central Florida. Bulletin of the Torrey Botanical Club 114:21–32.

Oliver, C. D. 1978. Subsurface geologic formations and site variation upper sand hills of South Carolina. Journal of forestry 76:352–354.

Olson, M. S., and W. J. Platt. 1995. Effects of habitat and growing season fires on resprouting of shrubs in longleaf pine savannas. Vegetatio 119:101–118.

Parr, C. L., C. E. R. Lehmann, W. J. Bond, W. A. Hoffmann, and A. N. Andersen. 2014. Tropical grassy biomes: misunderstood, neglected, and under threat. Trends in Ecology & Evolution 29:205–213.

Pausas, J. G., G. Juli, C. De Estudios, M. Ceam, P. Tecnològic, and C. Darwin. 1999. Response of plant functional types to changes in the fire regime in Mediterranean ecosystems: a simulation approach. Journal of Vegetation Science 10:717–722.

Peet, R. K. 2006. Ecological classification of longleaf pine woodlands. Pages 51–93 in S. Jose, E. Jokela, and D. Miller, editors. The longleaf pine ecosystem: ecology silviculture, and restoration. Springer Science+Business Media, LLC, New York, NY.

Platt, S. S. G., and C. C. G. Brantley. 1997. Canebrakes: an ecological and historical perspective. Castanea 62:8–21.

R Core Team. 2015. R: A language and environment for statistical computing. R Foundation for Statistical Computing, Vienna, Austria.

Schafer, J. L., B. P. Breslow, M. G. Just, M. G. Hohmann, S. N. Hollingsworth, S. L.

Simberloff, D. 1998. Flagships, umbrellas, and keystones: is single-species management passé in the landscape era? Biological Conservation 83:247–257.

Skinner, M. W., and B. A. Sorrie. 2002. Conservation and ecology of Lilium pyrophilum, a new species of liliaceae from the Sandhills Region of the Carolinas and Virginia, U.S.A. Novon 12:94–105.

Sorrie, B. A., J. B. Gray, and P. J. Crutchfield. 2006. The vascular flora of the longleaf pine ecosystem of Fort Bragg and Weymouth Woods, North Carolina. Castanea 71:129–161.

Staver, A. C., S. Archibald, and S. Levin. 2011. Tree cover in sub-Saharan Africa: rainfall and fire constrain forest and savanna as alternative stable states. Ecology 92:1063–1072.

Staver, A. C., and S. A. Levin. 2012. Integrating theoretical climate and fire effects on savanna and forest systems. American Naturalist 180:211–224.

Stephenson, N. L. 1990. Climatic control of vegetation distribution: the role of the water balance. The American Naturalist 135:649–670.

Thaxton, J. M., and W. J. Platt. 2006. Small-scale fuel variation alters fire intensity and shrub abundance in a pine savanna. Ecology 87:1331–7.

Veldman, J. W., L. A. Brudvig, E. I. Damschen, J. L. Orrock, W. B. Mattingly, and J. L. Walker. 2014. Fire frequency, agricultural history and the multivariate control of pine savanna understorey plant diversity. Journal of Vegetation Science 25:1438–1449.

Wakeling, J. L., A. C. Staver, and W. J. Bond. 2011. Simply the best: the transition of savanna saplings to trees. Oikos 120:1448–1451.

Weakley, A. S., and M. P. Schafale. 1991. Classification of pocosins of the Carolina Coastal Plain. Wetlands 11:355–375.

Wells, B. W., and I. V. Shunk. 1931. The vegetation and habitat factors of the coarser sands of the North Carolina Coastal Plain: an ecological study. Ecological Monographs 1:465– 520.

Wenk, E. S., G. G. Wang, and J. L. Walker. 2011. Within-stand variation in understorey vegetation affects fire behaviour in longleaf pine xeric sandhills. International Journal of Wildland Fire 20:866–875.

Whittaker, R. 1975. Communities and ecosystems. Second. Macmillian Publishing Co., Inc., New York, NY.

Wilson, J. B., and A. D. Q. Agnew. 1992. Positive-feedback switches in plant communities. Pages 263–336 in M. Begon and A. Fitter, editors. Advances in Ecological Research. Academic Press London.

Zedler, P. H., C. R. Gautier, and G. S. McMaster. 1983. Vegetation change in response to extreme events: the effect of a short interval between fires in California chaparral and coastal scrub. Ecology 64:809–818.

Figures

Fig. 3.1 Lines are predictions of the conditional probability of a plot burning, using a generalized linear model with C4 grass and evergreen shrub volume as predictors. For C4 grass (solid brown line) predictions, evergreen 3 shrub volume was held constant at zero, while C4 grass volume varied (0 – 2 m ), and vice versa for evergreen shrub (dashed green line) predictions. Vegetation type volume is represented on the horizontal axis as the proportion of a 2 m3 cell (i.e. vegetation volume / 2 m3)

Fig. 3.2 Conceptual diagram of the stochastic cellular automaton model used in this study. The top row is a representative lattice (1x5 cells) used in our simulations of fire spread and vegetation response to burn status along longleaf pine savanna – wetland ecotonal gradients. The second row represents a time step in which a fire was initiated (Fires were initiated every 1, 3, or 9 years, or never). Fires were initiated in the savanna cell of the lattice and could spread downslope towards wetland. Fire spread was conditional, and a cell could only burn if the preceding, upslope cell burned. A probability of burning [P(fire)] was assigned to each cell based on a binomial logistic regression (developed with field-collected data) that used C4 Grass and evergreen shrub volume as predictors. In this example, the first three (orange) of five cells burned. After each time step of the simulation, the vegetation structure volume for both types is updated (third row) based on burn status (i.e. burned, unburned [blue]). The probability of burning is then updated based on the new vegetation structure values. The process then continues on the updated lattice (fourth row in the diagram) until 200 time steps are reached

Fig. 3.3 Field-collected (2012) mean values by transect and relative gradient position for C4 grass and evergreen shrub volume (C4 grass [solid brown line], evergreen shrub [dashed green line]; 100 transects comprised of 460 3 2 m plots). (A) The mean proportion of a plot occupied by vegetation structure type (e.g. C4 grass or evergreen shrub). (B) The mean proportion of plots by vegetation structure type with non-negligible volume (i.e. C4 grass > 0.004 m3, evergreen shrub > 0.019 m3). We used these gradients as templates for our model simulations

Fig. 3.4 The mean proportion of a cell occupied by either vegetation type (C4 grass [solid brown line], evergreen shrub [dashed green line]) type by gradient position and arranged by vegetation initialization scheme (columns) by fire frequency (rows). The proportion of cell occupation was calculated as the mean across lattices (n = 100) by cell gradient position. Cells in the Field 2012 scheme were initiated with our observed (2012) vegetation structure values. Cells in the High Grass scheme were initiated with a high value for C4 grass (0.25 m3) and a low value for evergreen shrub (0.025 m3), and vice versa for the lattice cells of the High Shrub scheme. This plot depicts the results for our model instances that were simulated with field-based rules for updating flammability and vegetation response to burn status

Fig. 3.6 The mean fire return interval (years) by simulation instance (vegetation structure initialization scheme, updating rules) by fire frequency (1-yr [solid orange line], 3-yr [dashed purple line], 9-yr [dotted green line]). The mean fire return interval was calculated as the mean across lattices (n = 100) by cell gradient position. Fire was initiated in the savanna gradient position of each lattice every 1, 3, or 9 time steps. The cells of the Field 2012 schemes were initiated with our observed (2012) vegetation structure values. The cells of the High Grass 3 3 schemes were initiated with a high value for C4 grass (0.25 m ) and a low value for evergreen shrub (0.025 m ), and vice versa for the cells of the High Shrub schemes. This plot depicts the results for our model instances that were simulated with field-based rules for updating flammability and vegetation response to burn status. The bottom row depict rules that enhanced the fire-vegetation feedback

Fig. 3.7 Field collected (2012 – 2014) fire data. (A) Expected fire return interval (years) by relative gradient position, calculated as the inverse of the probability of burning multiplied by fire frequency (1-yr [solid orange line], 3-yr [dashed purple line], 9-yr [dotted green line]). (B) Proportion of plots burned per relative gradient position

Fig. 3.8 The mean proportion of burned cells by simulation instance (vegetation structure initialization scheme, model updating rules) by fire frequency (1-yr [solid orange line], 3-yr [dashed purple line], 9-yr [dotted green line]). The proportion of burned cells was calculated as the mean across lattices (n = 100) by cell gradient position. The cells of the Field 2012 schemes were initiated with our observed (2012) vegetation structure 3 values. The cells of the High Grass schemes were initiated with a high value for C4 grass (0.25 m ) and a low value for evergreen shrub (0.025 m3), and vice versa for the cells of the High Shrub schemes. This plot depicts the results for our model instances that were simulated with field-based rules for updating flammability and vegetation response to burn status. The bottom row depict rules that enhanced the fire-vegetation feedback

Fig. 3.9 The mean proportion of a cell occupied by either vegetation structure (C4 grass [solid brown line], evergreen shrub [dashed green line]) type by gradient position and arranged by vegetation initialization scheme (columns) by fire frequency (rows). The proportion of cell occupation was calculated as the mean across lattices (n = 100) by cell gradient position. Cells in the Field 2012 scheme were initiated with our observed (2012) vegetation structure values. Cells in the High Grass scheme were initiated with a high value for C4 grass (0.25 m3) and a low value for evergreen shrub (0.025 m3), and vice versa for the lattice cells of the High Shrub scheme. This plot depicts the results for our model instances that were simulated with enhanced-feedback based rules for updating flammability and vegetation response to burn status

Fig. 3.10 The mean proportion of plots by vegetation type (C4 grass [solid brown line], evergreen shrub [dashed green line]) with non-negligible volume (i.e. C4 grass > 0.004 m3, evergreen shrub > 0.019 m3) by gradient position and arranged by vegetation structure initialization scheme (columns) by fire frequency (rows). The proportion of cells with non-negligible volume was the mean across lattices (n = 100) by cell gradient position. Cells in the Field 2012 scheme were initiated with our observed (2012) vegetation structure values. Cells in the 3 High Grass scheme were initiated with a high value for C4 grass (0.25 m ) and a low value for evergreen shrub (0.025 m3), and vice versa for the lattice cells of the High Shrub scheme. This plot depicts the results for our model instances that were simulated with enhanced-feedback based rules for updating flammability and vegetation response to burn status

Chapter 4: Wood decay resistance and the persistence of resprouting, woody plants in

pyrophilic ecosystems

Abstract

In pyrophilic ecosystems, woody plants are repeatedly injured or topkilled (i.e. aboveground tissue is killed) by frequent fires. Many woody species persist in these frequently burned systems through resprouting, a life-history strategy where new aboveground tissues are generated from belowground energy reserves and organs. The success (persistence) of resprouters has generally been attributed to the ability of these plants to store and remobilize carbohydrate reserves. However, some coexisting resprouting species differ in their abundances between areas of high and low fire frequency. For these species, persistence might not be determined by resource reserves, but rather by their ability to prevent wood decay and, consequently, maintain the integrity of belowground organs. We hypothesized that species with less ability to contain the extent of wood decay would be restricted to less frequently burned areas. To this end, we measured variables previously identified to confer decay resistance in woody plants, including plant size, wood density, and lignin and extractable phenolic concentrations in five woody species, one upland, fire-tolerant species and four species restricted to less frequently burned areas of the same landscape. We induced topkill by coppicing 20 individuals of each species. At 9 and 19 months after coppicing, we harvested the root crown of 10 individuals per species, after measuring resprout height and diameter, if present. We found that the fire-tolerant species (Quercus laevis) had the least amounts of wood decay and the greatest values for wood density, and lignin and phenolic

concentrations. Q. laevis also had the smallest increase in decay between the 9- and 19- month post-coppicing harvest groups. We present evidence for a potentially overlooked aspect of resprouting success, wood decay resistance.

Introduction

Woody individuals within or near frequently-burned (i.e. pyrophilic) community boundaries must endure repeated topkilling (i.e. causing death of aerial biomass) fire events to survive and many species do so by resprouting (Klimešová and Klimeš 2007). Resprouting is a mechanism by which new biomass is generated from pre-existing plant tissue after an injurious event (Bellingham and Sparrow 2000), and, in pyrophilic ecosystems, the location of plant tissue from which the resprouts of topkilled individuals originate is most often near or below ground (Bond and Midgley 2001). Individuals that employ resprouting for persistence must balance carbon allocation between aboveground and belowground tissues

(Schutz et al. 2009, Poorter et al. 2010). Trade-offs between aboveground and belowground carbon allocation have fitness costs and, for example, may reduce competitive ability of resprouters as compared to non-resprouters (Clarke et al. 2013). Yet, not all resprouting species are able to persist under conditions of frequent burning (Fensham et al. 2003). This presents a conundrum, because some species known to be eliminated by frequent fire are also quite vigorous resprouters (e.g. Nyssa aquatica, Liriodendron tulipifera; Grady and

Hoffmann 2012). This phenomenon suggests that energy reserves alone many not be sufficient to guarantee persistence.

To persist, resprouting species must not only have the ability to resprout, but also to maintain the integrity of belowground tissues following the repeated trauma of topkilling fires. A possible explanation for the abundance patterns of resprouting woody species between areas of contrasting fire frequencies is that those least able to prevent decay after topkill are restricted to locations burned less often. Wood decaying microorganisms are

ubiquitous (Lonsdale et al. 2008) in the natural environment and any trauma to plant tissue provides a possible initiation site of wood rot (Smith 2008). Reoccurring fires, in systems such as savanna, provide many opportunities for an individual to become inoculated with wood decaying microorganisms. Two principal strategies have been suggested as the mechanisms woody plants can employ to mitigate damages from stem wounds with specific regard to wood decay: wound closure and decay resistance (Romero et al. 2009). Decay resistance has been positively correlated with factors such as wood density (Larjavaara and

Muller-Landau 2010) and presence of secondary metabolites (Smith 1997). These factors facilitate compartmentalization (Vek et al. 2013, Smith 2015), a process where an injured plant separates healthy and wounded tissues, thus restricting decay to wounded, old wood

(Shigo 1984). The second strategy, wound closure, mitigates the damage from injury by closing the wound through callus formation or exudates (Guariguata et al. 1996, Romero et al. 2009). Wound closure has been demonstrated to be an effective strategy to reduce or prevent decay in fire-caused wounds that did not result in topkill (e.g. Sutherland and Smith

2000, Balch et al. 2011). However, the frequency and magnitude of damage following topkill in pyrophilic communities suggests that wound closure would be an inferior strategy to control wood rot (Romero and Bolker 2008), because the rates of wound closure would be exceeded by the extent of injury. Decay resistance, then, should be particularly important for resprouting woody plants in frequently burned landscapes where topkill is commonplace.

We hypothesized that species restricted to wetter, and less fire-prone portions of a savanna – wetland gradient would experience greater amounts of decay. To this end, we compared an upland, fire-tolerant species with four species that are generally restricted to

more mesic, downslope portions of the gradient that are not as frequently burned. To examine the extent of decay in belowground plant tissues following topkill, we coppiced aboveground stems to simulate fire-caused topkill (Goorman et al. 2011, Moreira et al.

2012). We measured the size of all individuals, and wood density and secondary metabolites for each species. We harvested the belowground tissues of topkilled individuals at two time points after coppicing. We recorded the amount of (incipient) wood decay for each harvested individual.

Materials and Methods

Our study was conducted at the Fort Bragg US Army Installation (73,469 ha), located within the Sandhills region of North Carolina, USA (35°07’N, 79°10’W), which are upland, xeric ridges of sandy soils that are remnants of a prehistoric Atlantic coastline and æolian processes (Christensen 2000, Ivester and Leigh 2003). Mean annual precipitation in this

Sandhills region is 1275 mm and mean temperature ranges from 6.9°C in the winter to 26°C in the summer (Sorrie et al. 2006). These ridges host the predominant vegetation community of Fort Bragg: longleaf pine (Pinus palustris Mill.) – wiregrass (Aristida stricta Michx.) savanna, known locally as Xeric Sandhill Scrub (Schafale 2012). The upland savanna community transitions to more mesic communities downslope (Schafer et al. 2013).

Wetlands, locally known as streamhead pocosins and sandhill seeps (Weakley and Schafale

1991), are embedded within the savanna matrix, and occur where subsurface water emerges as seepage at clay outcroppings (Brinson 1991) that lie underneath portions of the Sandhills

(Oliver 1978).

The mean historical fire frequency for longleaf pine savanna is estimated at 2.2 years

(range: 0.5-12 years; Stambaugh et al. 2011) and 7-50 years for adjacent wetlands (Frost

1998). The difference in fire return interval along these gradients has been attributed to changes in vegetation structure and microclimate (Just et al. in press), where the more open vegetation structure of savanna promotes fire, and the denser vegetation structure of wetland hinders fire. Longleaf pine savanna requires episodic fire events to prevent woody encroachment and maintain an open canopy (Wells and Shunk 1931, Gilliam and Platt 1999).

In an effort to replicate the mean historical fire frequency, the landscape at Fort Bragg has been segmented into burn compartments, of which one-third burn annually. Fort Bragg uses low-intensity prescribed fires that are ignited in savanna and burn towards wetland (Lashley et al. 2014). Most prescribed fires occur between April and September, congruent with peak lightning activity for this geography (Fowler and Konopik 2007).

We selected five tree species that differ in their abundances along a savanna-to- wetland ecotonal gradient (Table 4.1): Quercus laevis Walter (turkey oak), Liquidambar styraciflua L. (sweetgum), Liriodendron tulipifera L. (tuliptree), Persea palustris (Raf.) Sarg.

(swamp bay), Acer rubrum L. (red maple). At sites with active fire management, Q. laevis is found in savanna, L. styraciflua is most common in the ecotone, and A. rubrum, L. tulipifera, and P. palustris are most common in the wetland, but also occur in the ecotone (Schafer and

Just 2014). Each of the selected species has the capacity to resprout from basal or belowground buds following topkill (Schafer and Just 2014).

In October 2012, we identified and tagged 20 individuals of each species that varied in their height and diameter (Table 4.1). Study individuals were located in six burn

compartments that had last burned 3-4 years before our experiment began. Our study sites spanned across 33.38 km2 of the landscape at Fort Bragg, and the mean minimum distance between sites was 2.92 km. We selected individuals of each species across a range of sizes because individual size is positively correlated with both injury (Schoonenberg et al. 2003) and decay resistance (Smith et al. 1970, Lowell et al. 1992) for some woody species. We measured the basal diameter (2 cm above ground level) of each individual to the nearest 0.01 mm using digital calipers, and we measured height to the nearest cm. We coppiced all stems at ground level. Coppicing has been used to simulate topkilling fire events (Goorman et al.

2011, Schafer and Just 2014). Moreover, coppicing minimizes the variation in damage that would result from differences in fire intensity across the landscape.

We harvested belowground tissue (i.e. root crown) for 9-10 individuals for each species at both 9 and 19 months after coppicing (July 2013 and May 2014 respectively) using hand tools. The root crown was harvested until the plant material ended or to depth of at least

50 mm. A wildfire burned a portion of one our sites in June 2013, which reduced the sample size of L. tulipifera to nine for the 19-month harvest. Before harvest, we recorded the number of resprouts, if any, and their height and basal diameter. After harvesting, we used a band saw to horizontally section the belowground material (target section thickness was 5 mm) from the coppiced stem downward until the plant material ended or began to grow perpendicular to the axis we were sectioning. The thickness of each section was measured with digital calipers. The mean thickness across species and harvest groups was 6.90 ± 0.09 mm (SE). The number of sections per individual was dependent upon the length of the harvested belowground material.

For each individual, we used height and basal diameter measurements to calculate the aboveground conical volume for each stem prior to coppicing and at the time of harvest. We calculated the ratio of post-coppicing to pre-coppicing volume to quantify volume recovery after topkill.

We visually determined the extent of incipient wood decay based on wood discoloration. This method has been widely used previously (e.g. Shigo and Hillis 1973,

Whitney 1997, Deflorio et al. 2008, Watson 2008), and wood discoloration associated with decay can be observed in as a little as 15-60 days after wounding (e.g. Schoonenberg et al.

2003), making it appropriate for our study. To quantify the extent of incipient decay for each section we measured the wood surface area and the surface area of decay (i.e. wood discoloration or staining), if present. To accomplish this, we traced the outline of wood and decay onto transparent plastic film, made the traced area opaque with a permanent marking pen, and then determined area with a portable laser leaf area meter (model: CI-202, CID Inc.,

Camas, WA, USA). The volume (cm3) of total wood and decaying wood for each section was calculated as its surface area × thickness. For each individual, volumes for total wood and decaying wood were summed for all sections of each stem and then for all stems (for multi- stemmed individuals).

We determined wood density (g cm-3), a plant trait known to confer decay resistance

(Larjavaara and Muller-Landau 2010), as dried mass per fresh volume of 12-16 aboveground stem wood samples of each species. We determined fresh wood volume using water displacement. We then oven-dried wood samples at 60°C for five days before recording their mass using a digital balance.

We analyzed three composite belowground tissue samples per species for lignin and extractable soluble phenolics. Both of these types of phenolic compounds have been reported to be correlated with decay resistance in many woody plant species (e.g. Cowling 1961,

Scheffer and Cowling 1966, Smith 1997, Harju et al. 2003, Gierlinger et al. 2004, Witzell and Martín 2008, Skyba et al. 2013). Samples were oven-dried, ground to a fine powder, and sent to the University of British Columbia for analysis (Dr. S.D. Mansfield, Department of

Wood Science, University of British Colombia). Samples were Soxhlet extracted with hot acetone for 24 hours. Lignin content was determined from 0.2 g dry-weight extract-free tissue using a modified Klason method (Coleman et al. 2007), where acid-insoluble lignin content was determined gravimetrically (Coleman et al. 2009), and acid-soluble lignin content was determined spectrophotometrically (Technical association of the pulp and paper industry 1991). Total lignin was calculated as the sum of the soluble and insoluble lignin and total lignin concentration was recorded as the proportion of lignin per sample dry weight.

Total phenolic concentration was determined using the Folin-Ciocalteu method (Sánchez-

Rangel et al. 2013) and absorbance was determined spectrophotometrically at 760 nm against a reagent blank. The phenolic concentration was expressed as mg of catechin equivalents per

100 mg sample (dry weight), and is reported here as a proportion.

Statistical analysis

We analyzed differences among species in post-:pre-coppicing volume recovery, total lignin, total phenolics, wood density, and decay with a Kruskal-Wallis test; Benjamini-Hochberg post-hoc significance tests (α = 0.05) were used to adjust P-values for multiple comparisons

(Benjamini and Hochberg 1995) using R package agricolae (de Mendiburu 2015).

Comparisons of proportional decay between harvest groups for each species were tested with exact Wilcoxon-Mann-Whitney tests using R package coin (Hothorn et al. 2015). We assessed wood decay resistance following the complete loss of aboveground tissue by examining the correlation between decay proportion and pre-coppicing size using ordinary linear models. All analyses were performed in R 3.2.2 (R Core Team 2015).

Results

The proportion of volume recovered after coppicing did not vary between species at 9 or 19 months after coppicing (χ2 = 3.57, P = 0.468; χ2 = 4.30, P = 0.368, respectively; Fig. 4.1).

Wood density varied among the study species (χ2 = 44.60, P < 0.001), and ranged from 0.443

±0.014 g cm-3 in L. tulipifera to 0.748 ±0.028 g cm-3 in Q. laevis. Species sorted into two groups for wood density (Fig. 4.2A), with Q. laevis, representing the group with the densest wood.

The measured chemical properties of wood also varied by species (Figs. 4.2B-C).

Total lignin (%) varied among species (χ2 = 24.69, P < 0.001) and Q. laevis had the greatest mean value (27.2 ±0.3%) and P. palustris the lowest (20.2 ±0.8%). For total lignin, species sorted into four groups with Q. laevis representing the species with highest lignin concentrations, and L. tulipifera and P. palustris representing the species with the lowest lignin concentrations. Total phenolics (%) varied between species (χ2 = 27.46, P < 0.001); Q. laevis had the greatest mean total phenolics concentration at 4.1 ±0.3% which was 64.99% greater than the species with the next greatest concentration (A. rubrum). Based on mean

total phenolic concentration, the species sorted into five, single-species groups, with Q. laevis representing the group with the greatest phenolic concentration and L. tulipifera representing the group with the lowest phenolic concentration.

The decay proportion, defined here as the proportion of discolored wood volume to the total wood volume to a belowground depth ≤ 5.0 cm, varied among our five study species

(χ2 = 41.19, P < 0.001). When considering both harvest groups together, Q. laevis experienced the least decay (3.9 ±1.0%) and P. palustris (53.6 ±5.3%) the most (Fig. 4.3A).

Considering both harvest groups separately, decay proportion again varied between species for both the 9- (χ2 = 25.92, P < 0.001) and 19-month (χ2 = 22.38, P < 0.001) harvest groups

(Figs. 4.3B-C). Again, in both harvest groups, we observed the lowest decay proportion in Q. laevis. We observed an effect of harvest date for three of the five species (Table 4.2), A. rubrum, L. tulipifera, and Q. laevis, and each had greater proportional decay at 19 months after coppicing than at 9 months. We did not observe an effect of pre-coppicing aboveground volume on decay proportion for either harvest groups when considering species separately (P

> 0.17; not shown).

Discussion

The extent of decay in belowground tissue after coppicing varied substantially among our five study species along a longleaf pine – wetland ecotonal gradient in the Sandhills of North

Carolina. Concordant with our hypothesis, the species with greater abundances at the more mesic positions on the gradient where those with the greatest extent of decay (Table 4.2).

Quercus laevis, whose abundance was greatest in the xeric, fire-prone upland was also the

species with the least decay (Table 4.2). Notably, Q. laevis not only had the lowest proportions of decay, but we also possessed the greatest concentrations of lignin and phenolics, and the densest wood of the five species (Fig. 4.3).

All species resprouted after fire, and the recovery of aboveground plant volume after coppicing did not differ between species (Fig. 4.1). This result confirms that the individuals of these species in this system are not restricted to certain positions along this ecotonal / fire- frequency gradient by their inability to resprout after topkill (i.e., species are not restricted to the less-oft burned wetland because they are unable to resprout). Moreover, each of the species has also been documented within upland, savanna habitat (Monk 1968, Hartnett and

Krofta 1989, Sorrie et al. 2006, Taggart 2010), demonstrating their ability to survive under variations in resource availability. Instead, this result suggests that there is another mechanism responsible for partitioning capable resprouters along this gradient. A possible explanation for the distribution of these species is that those that are unable to contain wood rot after topkill are restricted to areas that burn less often. The ability to resprout alone may not grant long-term persistence in frequently burned landscapes.

We did not observe an effect of initial plant size on decay proportion for any of the study species. This result is different from other studies examining injuries to plant stems, where plant size was positively correlated with lower rates of decay for some species

(Cartwright 1941, Smith et al. 1970, Lowell et al. 1992, Amusant et al. 2004). Plant size, as it relates to bark thickness, has been documented to confer resistance to topkill by fire for some savanna species (Lawes et al. 2011a, 2011b). However, in frequently burned systems, such as ours, once an individual is topkilled, the likelihood of future topkill is increased (e.g.

Hoffmann and Solbrig 2003) and individuals suffering repeated injuries of this magnitude are unlikely to prevent wood decay through wound closure regardless of size. For areas with frequent topkilling fires, our results suggest that size is less important than other factors that contribute to wood decay resistance (e.g. phenolics, density).

Resprouting is a common life-history strategy among woody plants in pyrophilic ecosystems (Pausas and Keeley 2014). Resprouting remobilizes carbohydrate reserves to create new aboveground tissues (i.e. resprouts), which have the ability to photosynthesize and then contribute back to carbohydrate reserves to be used after the next topkilling event.

This is an important process for many woody plant species in disturbance-prone systems

(Bond and Midgley 2001) where topkilling can be commonplace. In our study system, many woody plants are able to resprout (Drewa et al. 2006) and have resource requirements that can be met at several positions along the savanna-to-wetland gradient, as evidenced by the upslope abundances of these species in the absence of fire (Gilliam and Platt 1999). Yet these species abundances are different under active fire management, with many resprouting species only commonly existing in the wetter reaches of the gradient (e.g. Schafer et al.

2015). Our results suggest that persistence is these systems is not only correlated with the ability of topkilled individuals to resprout, but also their ability to prevent or limit wood decay.

Wood decay resistance has been well studied in other contexts for living and dead individuals for both ecological and commercial reasons (e.g. Kaufert 1933, Scheffer and

Cowling 1966, Whitney 1997, Deflorio et al. 2008, Cornwell et al. 2009). However, there is a lack of research on the role of decay in resprouting plant persistence in environments with

frequent topkilling events. We selected these species because they differed in abundances along a savanna-to-wetland gradient. However, our results demonstrate only two distinct, contrasting groups of species: Q. laevis and the rest. As such, we cannot generalize these patterns of decay resistance beyond this suite of species in this landscape. Analysis of additional resprouting species in upland, fire-prone habitats is necessary to assess if these wood decay resistance properties are common. This work is an important first step in understanding how wood decay resistance within topkilled resprouting trees contributes to species persistence in pyrophilic ecosystems.

References

Amusant, N., J. Beauchene, M. Fournier, G. Janin, and M.-F. Thevenon. 2004. Decay resistance in Dicorynia guianensis Amsh.: analysis of inter-tree and intra-tree variability and relations with wood colour. Annals of Forest Science 61:373–380.

Balch, J. K., D. C. Nepstad, L. M. Curran, P. M. Brando, O. Portela, P. Guilherme, J. D. Reuning-Scherer, and O. de Carvalho. 2011. Size, species, and fire behavior predict tree and liana mortality from experimental burns in the Brazilian Amazon. Forest Ecology and Management 261:68–77.

Bellingham, P. J., and A. D. Sparrow. 2000. Resprouting as a life history strategy in woody plant communities. Oikos 89:409–416.

Benjamini, Y., and Y. Hochberg. 1995. Controlling the false discovery rate: a practical and powerful approach to multiple testing. Journal of the Royal Statistical Society 57:289– 300.

Bond, W. J., and J. J. Midgley. 2001. Ecology of sprouting in woody plants: the persistence niche. Trends in Ecology & Evolution 16:45–51.

Brinson, M. M. 1991. Landscape properties of pocosins and associated wetlands. Wetlands 11:441–465.

Cartwright, K. S. G. 1941. The variability in resistance to decay of the heartwood of home- grown western red cedar (Thuja plicata D. Don) and its relation to position in the log. Forestry 15:65–75.

Christensen, N. L. 2000. Vegetation of the southeastern coastal plain. Pages 397–448 in M. Barbour and W. Billings, editors. North American Terrestrial Vegetation. Second Ed. Camrbridge University Press, New York, NY.

Clarke, P. J., M. J. Lawes, J. J. Midgley, B. B. Lamont, F. Ojeda, G. E. Burrows, N. J. Enright, and K. J. E. Knox. 2013. Resprouting as a key functional trait: How buds, protection and resources drive persistence after fire. New Phytologist 197:19–35.

Coleman, H. D., T. Canam, K.-Y. Kang, D. D. Ellis, and S. D. Mansfield. 2007. Over- expression of UDP-glucose pyrophosphorylase in hybrid poplar affects carbon allocation. Journal of Experimental Botany 58:4257–4268.

Coleman, H. D., J. Yan, and S. D. Mansfield. 2009. Sucrose synthase affects carbon partitioning to increase cellulose production and altered cell wall ultrastructure. Proceedings of the National Academy of Sciences of the United States of America

100

106:13118–23.

Cornwell, W. K., J. H. C. Cornelissen, S. D. Allison, J. J. Bauhus, P. Eggleton, C. M. Preston, F. Scarff, J. T. Weedon, C. Wirth, and A. E. Zanne. 2009. Plant traits and wood fates across the globe: rotted, burned, or consumed? Global Change Biology 15:2431– 2449.

Cowling, E. 1961. Comparative biochemistry of the decay of sweetgum sapwood by white- rot and brown-rot fungi. Technical Bulletin No. 1258. USDA Forest Service, Forest Product Laboratory, Madison, WI.

Deflorio, G., C. Johnson, S. Fink, and F. W. M. R. Schwarze. 2008. Decay development in living sapwood of coniferous and deciduous trees inoculated with six wood decay fungi. Forest Ecology and Management 255:2373–2383.

Fensham, R. J., R. J. Fairfax, D. W. Butler, and D. M. J. S. Bowman. 2003. Effects of fire and drought in a tropical eucalypt savanna colonized by rain forest. Journal of Biogeography 30:1405–1414.

Fowler, C., and E. Konopik. 2007. The history of fire in the southern United States. Human Ecology Review 14:165–176.

Frost, C. 1998. Presettlement ﬁre frequency regimes of the United States: a ﬁrst approximation. Pages 70–81 in T. Pruden and L. Brennan, editors. Fire in ecosystem management: shifting the paradigm from suppression to prescription. Tall Timbers Research Station, Tallahassee, FL.

Gierlinger, N., D. Jacques, R. Wimmer, L. E. Paques, and M. Schwanninger. 2004. Heartwood extractives and lignin content of different larch species (Larix sp.) and relationships to brown-rot decay-resistance. Trees - Structure and Function 18:230–236.

Goorman, R., A. Bartual, S. Paula, and F. Ojeda. 2011. Enhancement of photosynthesis in post-disturbance resprouts of two co-occurring Mediterranean Erica species. Plant Ecology 212:2023–2033.

101

Grady, J. M., and W. A. Hoffmann. 2012. Caught in a fire trap: recurring fire creates stable size equilibria in woody resprouters. Ecology 93:2052–60.

Guariguata, M. R., G. S. Gilbert, S. Biotropica, and N. Mar. 1996. Interspecific variation in rates of trunk wound closure in a Panamanian lowland forest. Biotropica 28:23–29.

Harju, A., M. Venäläinen, and S. Anttonen. 2003. Chemical factors affecting the brown-rot decay resistance of Scots pine heartwood. Trees:263–268.

Hartnett, D. C., and D. M. Krofta. 1989. Fifty-five years of post-fire succession in a southern mixed hardwood forest. Bulletin of the Torrey Botanical Club 116:107–113.

Hoffmann, W. A., and O. T. Solbrig. 2003. The role of topkill in the differential response of savanna woody species to fire. Forest Ecology and Management 180:273–286.

Hothorn, T., K. van de Wiel, H. Winell, and A. Zeileis. 2015. coin: conditional inference procedures in a permutation test framework. R package version 1.1-2. http://coin.r- forge.r-project.org/.

Ivester, A. H., and D. S. Leigh. 2003. Riverine dunes on the Coastal Plain of Georgia, USA. Geomorphology 51:289–311.

Just, M. G., M. G. Hohmann, and W. A. Hoffmann. in press. Where fire stops: vegetation structure and microclimate influence fire spread along an ecotonal gradient. Plant Ecology.

Kaufert, F. H. 1933. Fire and decay injury in the southern bottomland hardwoods. Journal of Forestry 31:64–67.

Klimešová, J., and L. Klimeš. 2007. Bud banks and their role in vegetative regeneration–a literature review and proposal for simple classification and assessment. Perspectives in Plant Ecology, Evolution and Systematics 8:115–129.

Larjavaara, M., and H. C. Muller-Landau. 2010. Rethinking the value of high wood density. Functional Ecology 24:701–705.

Lawes, M. J., H. Adie, J. Russell-Smith, B. Murphy, and J. J. Midgley. 2011a. How do small savanna trees avoid stem mortality by fire? The roles of stem diameter, height and bark thickness. Ecosphere 2:art42.

102

Lawes, M. J., A. Richards, J. Dathe, and J. J. Midgley. 2011b. Bark thickness determines fire resistance of selected tree species from fire-prone tropical savanna in north Australia. Plant Ecology 212:2057–2069.

Lonsdale, D., M. Pautasso, and O. Holdenrieder. 2008. Wood-decaying fungi in the forest: conservation needs and management options. European Journal of Forest Research 127:1–22.

Lowell, E. C., S. A. Willits, and R. L. Krahmer. 1992. Deterioration of fire-killed and fire- damaged timber in the western United States. de Mendiburu, F. 2015. agricolae: statistical procedures for agricultural research. R package version 1.2-3. http://tarwi.lamolina.edu.pe/~fmendiburu.

Monk, C. D. 1968. Successional and environmental relationships of the forest vegetation of north central Florida. The American Midland Naturalist 79:441–457.

Moreira, B., J. Tormo, and J. G. Pausas. 2012. To resprout or not to resprout: factors driving intraspecific variability in resprouting. Oikos 121:1577–1584.

Oliver, C. D. 1978. Subsurface geologic formations and site variation upper sand hills of South Carolina. Journal of forestry 76:352–354.

Pausas, J., and J. Keeley. 2014. Evolutionary ecology of resprouting and seeding in fire- prone ecosystems. New Phytologist 204:55–65.

Poorter, L., K. Kitajima, P. Mercado, J. Chubiña, I. Melgar, and H. Prins. 2010. Resprouting as a persistence strategy of tropical forest trees: relations with carbohydrate storage and shade tolerance. Ecology 91:2613–2627.

R Core Team. 2015. R: A language and environment for statistical computing. R Foundation for Statistical Computing, Vienna, Austria.

Romero, C., and B. M. Bolker. 2008. Effects of stem anatomical and structural traits on responses to stem damage: an experimental study in the Bolivian Amazon. Canadian Journal of Forest Research 38:611–618.

Romero, C., B. M. Bolker, and C. E. Edwards. 2009. Stem responses to damage: the evolutionary ecology of Quercus species in contrasting fire regimes. New Phytologist 182:261–271.

Sánchez-Rangel, J. C., J. Benavides, J. B. Heredia, L. Cisneros-Zevallos, and D. A. Jacobo-

103

Velázquez. 2013. The Folin–Ciocalteu assay revisited: improvement of its specificity for total phenolic content determination. Analytical Methods 5:5990.

Schafale, M. 2012. Guide to the natural communities of North Carolina: fourth approximation. Department of Environment, Health, and Natural Resources, North Carolina Natural Heritage Program, Division of Parks and Recreation, Raleigh, NC.

Schafer, J. L., B. P. Breslow, M. G. Just, M. G. Hohmann, S. N. Hollingsworth, S. L. Swatling-Holcomb, and W. A. Hoffmann. 2013. Current and historical variation in wiregrass (Aristida stricta) abundance and distribution is not detectable from soil δ13C measurements in longleaf pine (Pinus palustris) savannas. Castanea 78:28–36.

Schafer, J. L., and M. G. Just. 2014. Size dependency of post-disturbance recovery of multi- stemmed resprouting trees. PloS one 9:e105600.

Scheffer, T. C., and E. B. Cowling. 1966. Natural resistance of wood to microbial deterioration. Annual Review of Phytopathology 4:147–168.

Schoonenberg, T., M. Pinard, and S. Woodward. 2003. Responses to mechanical wounding and fire in tree species characteristic of seasonally dry tropical forest of Bolivia. Canadian Journal of Forest Research 33:330–338.

Schutz, A. E. N., W. J. Bond, and M. D. Cramer. 2009. Juggling carbon: allocation patterns of a dominant tree in a fire-prone savanna. Oecologia 160:235–246.

Shigo, A. 1984. Compartmentalization: a conceptual framework for understanding how trees grow and defend themselves. Annual review of phytopathology 22:189–214.

Shigo, A., and W. Hillis. 1973. Heartwood, discolored wood, and microorganisms in living trees. Annual Review of Phytopathology:1971.

Skyba, O., C. J. Douglas, and S. D. Mansfield. 2013. Syringyl-rich lignin renders poplars more resistant to degradation by wood decay fungi. Applied and Environmental Microbiology 79:2560–71.

Smith, K. 1997. Phenolics and compartmentalization in the sapwood of broad-leaved trees. Pages 189–198 in W. Dashek, editor. Methods in plant biochemistry and molecular biology. CRC Press, Boca Raton, FL.

Smith, K. T. 2008. An organismal view of dendrochronology. Dendrochronologia 26:185– 193.

Smith, K. T. 2015. Compartmentalization, resource allocation, and wood quality. Current

104

Forestry Reports 1:8–15.

Smith, R., H. Craig, and D. Chu. 1970. Fungal deterioration of second-growth Douglas-fir logs in coastal British Columbia. Canadian Journal of Botany 48:1541–1551.

Sorrie, B. A., J. B. Gray, and P. J. Crutchfield. 2006. The vascular flora of the longleaf pine ecosystem of Fort Bragg and Weymouth Woods, North Carolina. Castanea 71:129–161.

Sutherland, E. K., and K. T. Smith. 2000. Resistance is not futile: The response of hardwoods to fire-caused wounding. Pages 111–115 in D. Yaussy, editor. Proceedings: Workshop on Fire, People, and the Central Hardwoods Landscape. Gen. Tech. Rep. NE-274. USDA, Forest Service, Northeastern Research Station, Newton Square, PA.

Taggart, J. B. 2010. The vascular flora of Sandy Run Savannas State Natural Area, Onslow and Pender Counties, North Carolina. Castanea 75:484–499.

Technical association of the pulp and paper industry. 1991. Tappi useful method UM-250: acid-soluble lignin in wood and pulp. Atlanta, GA.

Vek, V., P. Oven, and M. Humar. 2013. Phenolic extractives of wound-associated wood of beech and their fungicidal effect. International Biodeterioration and Biodegradation 77:91–97.

Watson, G. 2008. Discoloration and decay in severed tree roots. Arboriculture and Urban Forestry 34:260–264.

Weakley, A. S., and M. P. Schafale. 1991. Classification of pocosins of the Carolina Coastal Plain. Wetlands 11:355–375.

Wells, B. W., and I. V. Shunk. 1931. The vegetation and habitat factors of the coarser sands of the North Carolina Coastal Plain: an ecological study. Ecological Monographs 1:465– 520.

Whitney, R. D. 1997. Relationship between decayed roots and aboveground decay in three conifers in Ontario. Canadian Journal of Forest Research 27:1217–1221.

Witzell, J., and J. A. Martín. 2008. Phenolic metabolites in the resistance of northern forest trees to pathogens-past experiences and future prospects. Canadian Journal of Forest Research 38:2711–2727.

105

Tables

Table 4.1 Characteristics of the individuals included in the study. Ranges for height, basal diameter, and pre- coppicing volume represent the minimum and maximum mean value

No. individuals Basal Pre-coppicing Species Height (cm) coppiced diameter (mm) volume (cm3) Quercus laevis 20 52-330 6.40-61.18 88.69-32337.13 Liquidambar styraciflua 20 12-399 2.38-41.20 1.78-17731.11 Liriodendron tulipifera 20 30-418 3.66-34.22 19.00-17205.32 Persea palustris 20 55-228 4.85-25.33 51.82-3612.22 Acer rubrum 20 38-339 4.27-24.70 18.14-5861.85 Species are ordered by the position along the savanna-wetland gradient (Schafer and Just 2014)

106

Table 4.2 Decay proportion (mean proportion of discolored wood volume to total wood volume to a belowground depth of 5.0 cm) for the five study species. Results of exact Wilcoxon-Mann-Whitney tests comparing decay proportion between 9- and 19-month harvest groups for each species are also presented

Species Both harvests 9-month harvest 19-month harvest n Decay (%) BH n Decay (%) BH n Decay (%) BH z A. rubrum 20 43.59 (5.46) bc 10 30.46 (6.87) b 10 56.72 (6.34) a -2.34* L. tulipifera 19 32.50 (6.20) c 10 14.56 (5.46) c 9 52.42 (7.14) a -3.27*** L. styraciflua 20 33.50 (6.07) c 10 25.33 (6.50) bc 10 41.66 (9.93) a -0.98 P. palustris 20 53.56 (5.25) ab 10 51.21 (6.36) a 10 55.91 (8.64) a -0.98 Q. laevis 20 3.89 (0.99) d 10 1.18 (0.56) d 10 6.60 (1.48) b -2.8** Decay is presented as the mean value with standard error of the mean in parentheses. Different letters within harvest groups represent species’ differences using Benjamini-Hochberg (BH) post hoc comparisons (α = 0.05) on Kruskal-Wallis analysis of decay proportion between species. Symbols for the statistical comparison of decay proportion between harvest groups: ***P < 0.001, **P < 0.01, *P < 0.05

107

Figures

108

Fig. 4.2 (A) Wood density (g cm-3), (B) total lignin (proportion of lignin per [dry weight basis]), and (C) total phenolics (proportion of extractable phenolics per sample [dry weight basis]) per species. The box plots indicate data range, quartiles, and median. Dots are outliers. Asterisks denote mean. Letters within each plot represent species’ differences using Benjamini-Hochberg post hoc comparisons (α = 0.05) on Kruskal-Wallis analysis between species for the variable of interest. Species are ordered (left to right) by descending mean value for the variable of interest

109

Fig. 4.3 Decay proportion, defined here as the mean proportion of discolored wood volume to total wood volume to a belowground depth ≤ 5.0 cm, by species for (A) the 9 and 19 month post-coppicing harvest groups combined, (B) the 9 month harvest group, and (C) the 19 month harvest group. The box plots indicate data range, quartiles, and median. Dots are outliers. Asterisks denote mean. Letters within each plot represent species’ differences using Benjamini-Hochberg post hoc comparisons (α = 0.05) on Kruskal-Wallis analysis of decay proportion between species. Species are ordered (left to right) by descending mean value for the variable of interest

110

Chapter 5: Invasibility of a fire-managed savanna-wetland ecotone by non-native,

woody plant species

Abstract

Fire-promoting, open-canopy ecosystems are under threat of conversion to a fire-deterring, closed-canopy condition due to woody encroachment. This vegetation structure conversion has been fostered by fire suppression and introduced woody species. We investigated the impacts of prescribed fire and site conditions on the invasibility of a pine savanna by non- native, woody species. We performed a field experiment to study these impacts on the growth, survival, and establishment success of six invasive, woody species along a managed longleaf pine savanna – wetland ecotonal gradient in the Sandhills of North Carolina, USA.

Seeds of the six woody species were sown using three sowing methods across 18 study sites; each site contained paired plots located in savanna and savanna-wetland ecotone vegetation communities and were scheduled to be burned 1, 2, or 3 years after sowing. We identified fire as an important factor for preventing woody invasion, and it filtered 5 of 6 study species from the landscape. However, the landscape was not immune to invasion. At the end of the study period, three species had established individuals in unburned sites. In sites burned after seedling emergence, only one species, Pyrus calleryana, survived and established. We found

P. calleryana survival and establishment to be a function of seedling size, soil humic matter content, and seed-sowing treatment.

111

Introduction

Woody encroachment of open-canopy plant communities can alter ecosystem processes, species distributions, and biodiversity (Brudvig 2010). Possible causes of woody encroachment include CO2 enrichment, nitrogen deposition, fire suppression, overgrazing, and non-native species introductions (Van Auken 2009, Bond and Midgley 2012). Woody encroachment of open habitats modifies vegetation structure and, unchecked, eventually results in canopy closure. Canopy closure alters the understory environmental conditions, including insolation, temperature, and precipitation (Smith and Johnson 2004, Van Auken

2009), and these changes promote woody vegetation. Therefore, woody encroachment upon open-canopied plant communities can alter successional pathways of communities (Mandle et al. 2011). Once the transition is made from an open- to closed-canopy structure, it is often difficult to return to an open-canopy community even when reversing the conditions that created the transition, due to hysteresis (e.g. Wilson and Agnew 1992, Nowacki and Abrams

2008).

Periodic disturbances are important phenomena for maintaining open habitats, largely preventing woody individuals from reaching canopy heights. These disturbances present opportunities (e.g. nutrients, gaps) and challenges (e.g. damage, death) for plant recruitment

(Grubb 1988) within landscapes. In pyrophilic ecosystems, for example, seedling establishment is both directly and indirectly influenced by fire through effects on safe-sites, seed banks, light availability, nutrient recycling, and microclimates (Lamont et al. 1993,

Hoffmann 1996, Bond and Keeley 2005) as evidenced by the brief recruitment times after fire for many species in fire-prone systems (Grubb 1988, Denham et al. 2010). However,

112

beneficial or detrimental disturbance-mediated conditions are available to both native and non-native plant species alike.

Fire may have divergent effects on invasibility. Fire may enhance seedling establishment of invading species by creating favorable conditions for germination and early growth (Ne’eman et al. 2004, Glasgow and Matlack 2007, Emery et al. 2011). However, fire can kill seedlings (Huddle and Pallardy 1999, Gignoux et al. 2009, Green et al. 2010), even those of species that become fire tolerant later in their development (e.g. Grace and Platt

1995, Hoffmann et al. 2012). For such species, seedling growth rate should be particularly important because it will influence whether a seedling will survive the next fire. Some seedlings may avoid fire-induced topkill (i.e. loss of aerial biomass) by reaching sizes that confer fire-resistant traits (e.g. bark thickness; Lawes et al. 2011). Growth rate is important for resprouting species too. If seedlings reach sizes that store adequate belowground energy before topkill, they can resprout; fast growing resprouts also reach fire-resistant sizes quicker

(e.g. Stevens and Beckage 2010).

Woody individuals that reach sizes that escape topkill may alter fire behavior (e.g.

Briggs et al. 2002) and promote further woody encroachment (Ryan et al. 2013), increasing the potential for displacement of native species. Moreover, some non-native woody individuals are arriving in open-canopied landscapes already experiencing woody encroachment, and because many woody invaders are shade tolerant (Webster et al. 2006,

Brym et al. 2011), these may prove additional reprieve from fire. Wildfire suppression was widely practiced in the United States during the 20th century, and anthropogenically fragmented landscapes have further reduced the extent of burned areas (Duncan and

113

Schmalzer 2004). Repeated reductions in burned area of fire-prone, open-canopy communities has resulted increased woody encroachment and canopy closure (Nowacki and

Abrams 2008). Importantly, these changes have been accompanied by the arrival of invasive, woody species which can further alter vegetation structure and fire regimes (e.g. Mandle et al. 2011, Arianoutsou and Vilà 2012).

In landscapes where natural fire regimes have been altered or lost, prescribed fire has been used to conserve and restore fire-prone (i.e. pyrophilic) communities, including the important role of fire in preventing woody encroachment. However, the extent to which prescribed fire can inhibit non-native woody encroachment is not well defined (Mandle et al.

2011). The invasibility of fire-managed ecosystems by non-native plant species will be dependent on the traits of the invader, the current status of the ecosystem, and the details of the fire management practices (Keeley 2006), including fire frequency. Therefore, successful invaders need to be compatible with current fire management practices or have the ability to change fire behavior.

Here we test the invasibility of a fire-managed longleaf pine (Pinus palustris) savanna

– wetland ecotone in the southeastern USA by non-native, woody plants. Longleaf pine ecosystems are maintained by periodic fire disturbance (Chapman 1932) and are some of the most species-rich systems in the North America (Noss et al. 2015). The current areal extent of longleaf pine ecosystems is estimated at 3% (~1 million ha) of their historical maximum

(Frost 1993, Simberloff 1998), and much of this area is in an unhealthy condition with only half experiencing regular fire events (Outcalt 2000). However, healthy longleaf pine systems with minimal plant invasions exist (e.g. Sorrie et al. 2006), and are those under active fire

114

management (Glitzenstein et al. 2012). In this study, we performed a field experiment to test the effects of prescribed fire and environmental conditions on non-native species recruitment.

We examined the emergence, survival, and establishment of six woody, non-native, avian- dispersed species that were extant, but not abundant in a longleaf pine savanna landscape.

Materials and methods

Study site

This study was conducted at Fort Bragg Army installation (73,468 ha), located within the

Sandhills region of North Carolina, USA (35°07’N, 79°10’W). The dominant vegetation community at Fort Bragg is longleaf pine savanna (known locally as Xeric Sandhill Scrub;

Schafale 2012) which tends to inhabit upland ridges of sandy soils (Sorrie et al. 2006). These sandy-ridged xeric uplands are the remnants of ancient marine coastal features and aeolian processes (Christensen 2000, Ivester and Leigh 2003). Some areas of the Sandhills are perched atop a less permeable clay substrate, resulting in the lateral movement of ground water (Oliver 1978), which exits at clay outcroppings, or other areas of topographic relief, as seepage, resulting in wetlands (known locally as streamhead pocosins and sandhill seeps;

Weakley and Schafale 1991).

Mean annual precipitation in the study area is 1275 mm and mean temperature ranges from 6.9°C in the coldest months to 26°C in the warmest months (Sorrie et al. 2006).

Elevation ranges from 43 to 176 m. The landscape at Fort Bragg has been divided into discrete burn compartments of which one-third burn annually, approximating the estimated,

115

historic mean fire frequency (2.2 yrs) of longleaf pine savannas (Stambaugh et al. 2011).

Prescribed fires are started in savanna and burn towards wetland (Lashley et al. 2014).

This research was performed along ecotonal gradients between longleaf pine savanna and wetland (Fig. 5.1). Sites were identified from vegetation community and fire history GIS data. We first selected burn compartments that contained both longleaf pine savanna and wetland communities. From this subset, we selected sites that last had a prescribed fire 0, 1, or 2 years before our experiment. Next, we excluded sites that were scheduled to burn within

1 year after seed sowing. We finally selected sites (n = 18) that maximized spatial extent while distributing time since fire equally (n = 6 sites per years since fire category). Our sites were distributed over 134 km2 of Fort Bragg, and the mean distance between a site and its nearest neighboring site was approximately 692 m. Prescribed burns implemented during our study resulted in 5 burned sites in 2013 (1 yr after sowing), 4 burned sites in 2014 (2 yrs after sowing), and 8 burned sites in 2015 (3 yrs after sowing). In total, 13 sites were burned and 4 of these sites burned twice, once in 2013 and again in 2015.

At each site, we established one rectangular plot (2.5 × 5 m) within savanna and one within the savanna-wetland ecotone; the placement of these community plots was determined by vegetation (Fig. 5.1). Each community plot was divided into 18 circular subplots

(diameter: 30 cm) for 648 subplots (six species × three seed sowing treatments) in this experiment. We had three seed-sowing treatments 1) undisturbed, 2) litter removal and undisturbed soil, and 3) litter removal and covered with loose soil. The undisturbed treatment consisted of placing seeds atop litter (e.g. pine needles, broadleaves), emulating dispersal by birds. The litter removal and undisturbed soil treatment consisted of removing litter and duff

116

to bare soil, emulating fire litter removal prior to seed arrival. In the final treatment, we removed litter and then covered seeds with 1 cm of loose soil, emulating seed movement by granivores after avian dispersal. Assignment of species and sowing treatment to subplots was random. Krall et al. (2014) reported evidence of granivory for three of our study species at

Fort Bragg and; to reduce seed removal by animals, we erected steel mesh (1.27 cm2 mesh size) exclosures for each of our subplots.

Study species

We studied six avian-dispersed, non-native woody species that are generally considered invasive in the southeast USA: Elaeagnus umbellata Thunb., Ligustrum sinense Lour., Melia azedarach L., Nandina domestica Thunb., Pyrus calleryana Dcn., and Triadica sebifera (L.)

Small (Table 5.1). These species have been documented to occur in longleaf pine landscapes in both xeric and mesic communities (Herring and Judd 1995, Drew et al. 1998, Renne et al.

2002, Jenkins and McMillan 2009, Noss 2012). Moreover, each of these non-native species have been found at Fort Bragg, with the greatest abundances in the cantonment, where some species have been planted ornamentally. Each of the six species has been documented to resprout after a topkilling event (i.e. death or removal of aboveground tissue) as adults or established plants (Faulkner 1989, Grace 1998, Miller 2003, Culley and Hardiman 2007,

Herrero et al. 2015).

Seeds of each species were bulk-collected from on or near Fort Bragg during

November 2011 and then processed. E. umbellata fruits were macerated and the seeds were thoroughly rinsed, soaked in running water for 24 hours and air dried before cold

117

stratification (5° C) for 90 days. Ripe L. sinense fruits were collected at the beginning stages of desiccation. We further air-dried the fruits for one month after which they were rehydrated before we manually removed seeds. Seeds were air-dried and cold stratified (5° C) for 60 days. M. azedarach fruits were manually de-fleshed and stones were thoroughly rinsed and air-dried before storage at room temperature. N. domestica seeds were manually removed from fruit, thoroughly rinsed, and then cold stratified (5° C) under slightly moist conditions for 60 days. P. calleryana fruits were frozen (0° C) and then thawed (5° C) to facilitate flesh removal. After thawing, the fruits were macerated and seeds were manually removed and thoroughly rinsed. Seeds were soaked in water for 24 hours before cold stratification (5° C) for 60 days. T. sebifera capsules were collected after fruit splitting and were air-dried for one month. Seeds were then manually removed from capsules. Seeds were stored at room temperature. Prior to sowing, seeds were counted and packaged (Table 5.1). All seeds were sown in March 2012.

Field measurements

At the end of summer (September 2012), we recorded the number of emergent individuals per plot. Beginning approximately one year after sowing, we measured basal diameter and height of all individuals with digital calipers. Number of individuals and seedling size were censused annually during the growing season (May 2013 – 2015), and once at the end of the study period (September 2015). For analyses considering seedling size we calculated the conical volume (cm3) for each individual using height and diameter measurements.

118

In summer 2013 we sampled the surface soil (0-10 cm) of each community plot (n =

36). Soil samples were analyzed by the North Carolina Department of Agriculture’s

Agronomic Division. Soil pH was determined on a 1:1 soil to water volume ratio.

Exchangeable acidity (Ac; meq 100 cm-3) was determined using the Mehlich et al. (1976) method. Percent humic matter determination was made with a NaOH digestion with colorimetry (Mehlich 1984a). Soil amounts of Ca, Cu, K, Mg, Mn, Na, P, and Zn (mg dm-3) were determined with the Mehlich 3 soil text extractant method (Mehlich 1984b). Cation exchange capacity (CEC; meq 100 cm-3) was calculated as the summation of extractable Ca,

K, Mg, and Ac. Base saturation (BS; %) and soil bulk density (g cm-3) were also reported.

Volumetric soil moisture content (0-20 cm) was measured at 5 locations per community plot four times per year (2012-2015) with a HydroSense II soil-moisture sensor

(Campbell Scientific, Logan, UT, USA) and recorded as the mean. Canopy closure (%) was estimated with a concave spherical densiometer (Lemmon 1956) at 5 locations per community plot four times per year (2012-2015) and recorded as the mean.

Analyses

Statistical analyses were computed in R 3.2.2 (R Core Team 2015). Most of our analyses were designed to use subplots, not individual plants, as the sampling unit. Our emergence, survival, and establishment responses were proportions; as such, we employed generalized linear mixed-effect models (GLMM) with a binomial error structure with random and fixed effects to evaluate the impact of variables on our seed fates using R package lme4 (Bates et al. 2015). Linear mixed-effect models (LMM) were used for analyses concerning

119

relationships between seedling size (individual plants as sampling unit) and predictor variables, and we used Satterwaithe’s approximation of denominator degrees of freedom to calculate P values in R package lmerTest (Kuznetsova et al. 2012). Our mixed-effect models used the Laplace approximation of maximum likelihood, and we considered site as a random effect (random intercept) and habitat, treatment, environmental variables, and species as fixed effects. GLMM models were evaluated for overdispersion, and when the ratio of residual deviance to degrees of freedom was greater than 1.5 we added an observation-level random effect to account for the overdispersion (Harrison 2015). In cases where there too few observations across sites, the inclusion of site as a random factor became irrelevant and we analysed those relationships with generalized linear models (GLM).

To identify important variables in our 1-yr seedling survival models we used type II

Wald χ2 tests to evaluate full models versus reduced models. Full models were constructed with all variables previously identified as statistically significant (when considered individually) as predictors. We tested the full model versus a reduced model (i.e. one variable removed at a time); variables were removed if there was no difference in explanatory power between the full and reduced model.

Selected soil properties were summarized with a principal-component analysis (PCA) with R package caret (Kuhn 2008). Each selected soil property was Box-Cox transformed, centered, and standardized before the PCA was performed.

Results

Seedling emergence, survival, and establishment

120

Seedlings of all six species emerged by the end of the first growing season (Table 5.2).

Across all species, emergence, defined as the proportion of seedlings to seeds sown, had a mean of 6.1%. Community type had no effect on mean seedling emergence (χ2 = 0.50, P =

0.23). For individual species, emergence did not differ between savanna and ecotone plots, except for L. sinense (χ2 = 8.31, P = 0.003; savanna = 1.5%, ecotone = 0.6%), the species with the lowest emergence rate (Fig. 5.2). Across all study species, mean emergence differed among seed sowings treatments (χ2 = 101.78, P < 0.001) with the litter removal and loose soil treatment having the greatest emergence (Fig. 5.2). There was also an effect of sowing treatment on emergence (P < 0.001) when considering species separately (Table 5.3). Species identity had an effect on mean seedling emergence (χ2 = 52.82, P < 0.001).

The mean survival rate, considering all species, between the end of first growing season and the beginning of the second was 42 ± 6%. Two-year survival, defined here as the proportion of emerged seedlings that survived until the end of a 2-year period beginning at the end of the first growing season, across all species had a mean of 23 ± 3% (no L. sinense seedlings survived). Fire had a marked impact on seedling survival (Table 5.2); as such, we filtered the dataset and performed analyses for unburned sites and sites burned after emergence separately. Considering all species for sites that did not burn in 2013, mean 2-yr survival was 32.1%. We did not observe an effect of community type on survival (Table 5.3).

Testing for each species individually revealed an effect of community type on survival for N. domestica and P. calleryana (Table 5.3). At burned sites, only P. calleryana survived, and there was an effect of community type on survival (χ2 = 5.26, P = 0.02), 60.0 ± 24.49% mean survival in savanna and no survival in ecotone. For comparison, P. calleryana had a mean

121

survival rate of 21.82 ± 12.20% in burned sites, and 61.85 ± 9.71% in unburned sites.

Considering all species, we did not observe an effect of seed-sowing treatment on seedling survival in unburned subplots. Analyzing each species separately, there was an effect of treatment on survival for both unburned and burned (P. calleryana only) sites, save E. umbellata (Table 5.3). Species identity had a significant effect on survival for unburned (χ2 =

166.36, P <0.001) and burned (χ2 = 59.10, P < 0.001) subplots (Table 5.3).

Here we define establishment as the proportion of seeds that germinated and survived until the end of the study period. Across all study species, the overall mean establishment rate was 1.12 ± 0.2%, with 7.16 ± 1.8% in unburned subplots and 0.36 ± 0.2% in subplots burned after germination. Only three species (N. domestica, P. calleryana, T. sebifera) had individuals that survived until the end of the study, and P. calleryana was the only species with individuals remaining in burned sites (Table 5.3, Fig. 5.2). Across the three species, there was an effect of community type on establishment in unburned sites (χ2 = 10.50, P =

0.001), with a mean establishment rate of 8.7 ± 3% in savanna and 5.6 ± 2% in ecotone. The effect of community type upon establishment for either burned or unburned sites was not significant when considering species separately (P > 0.05). Across the established species, there was a significant effect of treatment on establishment for unburned (χ2 = 151.12, P <

0.001) subplots, with litter removal and loose soil subplots having the greatest mean establishment 19.0 ± 5% (undisturbed, 0.65 ± 0.4%; litter removal, 1.84 ± 1.8%). For each species, in both burned and unburned sites, the greatest rates of establishment occurred in litter removal and loose soil subplots (Table 5.4). There was a significant effect of species identity on mean establishment in unburned (χ2 = 22.81, P < 0.001) subplots (N. domestica

122

[3.87 ± 2.6%], P. calleryana [9.78 ± 3.6%], and T. sebifera [7.83 ± 3.5%]). The mean establishment rate of P. calleryana was 1.07 ± 0.5% in burned sites.

Environmental conditions, seedling size, and 1-yr survival

The following soil properties were summarized with a PCA: Ac, BS, Ca, CEC, Cu, K, Mg,

Mn, Na, P, pH, and Zn. The first two principal components accounted for 56.5% of the variance of these selected soil properties (Table C.1), and were used in the following seedling survival (1-yr) and size analyses. We observed no difference between savanna and ecotone plots in canopy closure, ecotone plots had greater humic matter, and soil moisture, and savanna plots had greater soil bulk density (Table 5.5).

Considering the five species with surviving individuals, subplot mean seedling volume at the beginning of the second growing season (May 2013) did not differ between savanna and ecotone community plots (t = 1.02, df = 85, P = 0.31). For each species, mean seedling volume (mean of all individuals per subplot) did not differ between savanna and ecotone plots (P > 0.10), E. umbellata was not tested as it had only six individuals contained in 1 savanna and 2 ecotone subplots. Mean seedling volume did not differ among seed sowing treatments (F2,96 = 0.13, P = 0.88) when considering all species, or when considering each species separately (P > 0.20), except for M. azedarach (F2,7 = 25.78, P < 0.001), which had the fewest subplots (n = 10) with established individuals, where 7 subplots where in the litter removal and loose soil treatment, which also had the largest M. azedarach individuals.

Seedling volume (May 2013) of extant individuals was not correlated with environmental covariates, except for a weak relationship with soil bulk density (P = 0.020,

123

marginal R2 = 0.02), when all species were considered. The strength and significance of the relationships between environmental covariates and seedling volume for each species was idiosyncratic (Table C.2), and only P. calleryana had a significant relationship with a marginal R2 ≥ 0.10 (humic matter; marginal R2 = 0.19). Seedling volume was not correlated with years since fire (Table C.2).

The 1-year seedling survival rate (May 2013 – 2014) in unburned sites for all species had only one significant, but weak, relationship with an environmental variable (PC2, P =

0.005, marginal R2 = 0.04). The 1-year seedling survival rate was not correlated with years since fire (Table C.3). These relationships for individual species were, again, idiosyncratic

(Table C.3). Moreover, we did not observe general patterns among the four species (E. umbellata had only six seedlings at the beginning of the 1-year period and was not individually analyzed) regarding survival and environmental covariates. For burned P. calleryana subplots, there was an effect of each environmental variable on survival except for soil moisture (Table C.4). The 1-year survival rate for unburned seedlings across all species was positively correlated with May 2013 seedling volume (P < 0.001, marginal R2 =

0.47). When examining species individually, seedling volume was positively correlated with survival for N. domestica, P. calleryana, and T. sebifera, but not M. azedarach (Table C.5).

The 1-year survival rate for burned P. calleryana seedlings was positively correlated with seedling volume (P = 0.02, marginal R2 = 0.64). Overall mean survival rates are given in

Table 5.6.

We further investigated the factors affecting P. calleryana survival and establishment as it was the only species to survive (i.e. resprout) after topkill. The subplot mean 1-yr

124

survival rate in burned sites was correlated with all but one (soil moisture) of the environmental variables (Table C.4). We constructed a model including these variables and seed-sowing treatment (i.e. the full model), where 1-yr survival = humic matter + soil bulk density + canopy closure + PC1 + PC2 + seed-sowing treatment. The final, reduced model included two predictors, humic matter and seed-sowing treatment (Fig. 5.3), and had a marginal R2 = 0.60. In both unburned and burned sites, seedling size was an important factor affecting P. calleryana seedling survival (Table C.5). The probability of a seedling surviving fire was greater than 50% once it surpassed a volume of 0.45 cm3 (Fig. 5.4). The percentage of P. calleryana seedlings at or above this volume were 8.4% (18 individuals) in May 2013,

9.6% (n = 13) in May 2014, and 16.8% (n = 16) in May 2015 (Fig. 5.5).

Discussion

Savanna, and other open-canopy ecosystems are under multiple conversion threats (Nowacki and Abrams 2008, Parr et al. 2014), including the conversion to closed-canopy systems from woody encroachment of native and non-native species alike. Historically, positive fire- vegetation feedbacks maintained the open vegetation structure of savanna, but humans have greatly altered fire regimes (through fire suppression, species introductions, etc.), disrupting the fire-vegetation feedback, which is a necessity for long-term maintenance of pyrophilic communities. Recent conservation efforts have used prescribed fire to prevent woody encroachment (e.g. Platt et al. 2015); however, the efficacy of prescribed fire at filtering invasive woody species from savanna landscapes is not well documented (Mandle et al.

2011). We investigated the role of fire and environmental conditions on recruitment success

125

of six woody, invasive plant species along a fire-managed savanna-wetland gradient. Our results demonstrated that seed-sowing treatment, species identity, seedling size, and fire were the most important factors for determining seedling survival and establishment success within this longleaf pine landscape. At the end of the study period, three of the six study species had established individuals (Fig. 5.2). Overall, there were 189 established seedlings from 21,060 sown seeds. Of these 189 seedlings, 164 established under the litter removed and loose soil seed-sowing treatment. Only P. calleryana, survived fire, with 25 (from

15,210 seeds sown) seedlings established in sites burned after emergence, 23 of these seedlings were in the litter removal and loose soil subplots (Table 5.4).

The Sandhills are nutrient and water limited (Wells and Shunk 1931, Hatchell and

Marx 1987), resulting in slow vegetation growth rates. Slow growth combined with periodic topkilling fires, create obstacles to seedling survival, and individuals in this environment may need prolonged periods to reach fire-resistant sizes, as compared less stressful environments.

Seedling size had a positive effect on the 1-yr survival rates for each of the three established species in unburned sites (Table C.5), and for P. calleryana in sites burned after emergence

(Fig. 5.4). Humic matter was the only environmental factor with a significant effect on seedling size, but only for P. calleryana (Table C.2). Humic matter was one of the two predictors included in our final model predicting P. calleryana survival after fire, and macro and micronutrient uptake, germination, and root growth have been positively correlated with humic matter (Trevisan et al. 2010, Jindo et al. 2012, Poblete et al. 2015). The survival and establishment success of P. calleryana after topkill may have been its increased ability to acquire resources due to increased root biomass from greater humic matter content and

126

stressful site conditions (Franco et al. 2011). For sites with more amenable growth conditions and faster growth rates, fire may be a less effective filter, as individuals may reach fire resistant sizes more quickly.

Fire return intervals, one aspect of fire regimes, affect encroachment and invasibility of pyrophilic ecosystems (Sankaran et al. 2005, Keeley 2006, Higgins et al. 2007). Frequent fires may exhaust energy reserves of topkilled seedlings or fires may occur before they are physiologically capable of resprouting (Ryan and Williams 2011). On the other hand, very infrequent fires may result in a landscape with unfavorable site conditions (Denham et al.

2010). However, managers for pyrophilic landscapes often employ an intermediate fire return interval (Ryan et al. 2013). For example, the current fire management of our study landscape utilizes a three-year fire return interval based on an estimated, historical average. This fire frequency appears to be largely effective at filtering five of these six non-native, woody species from the landscape (i.e. 2.80% vs 0.16% overall establishment rate in unburned and burned sites, respectively). Yet, this does not indicate that this landscape is immune to increased invasion rates by woody plants. For example, in unburned sites, individuals of three species became established during the study period (Fig. 5.2). In spite of the regular use of prescribed fire as a management tool, fire does not burn everywhere – landscapes are heterogeneous – resulting in burned and unburned patches. These unburned patches may provide emergent seedlings the time they need to establish. Moreover, managers are now considering conservation and restoration techniques for pyrophilic communities that are more temporally and spatially heterogeneous with respect to prescribed fire (Ryan et al. 2013).

127

While we did not observe great differences in emergence, survival, or establishment proportions between the savanna and ecotone communities, we suspect that differences in vegetation structure along the ecotonal gradient could play an important role in invasibility.

Previous work in this system has identified vegetation structure as an important factor for determining where prescribed fires stop spreading along this savanna – wetland gradient, with grasses promoting and evergreen shrubs deterring fire spread (Just et al. in press). For example, Just et al. (in press) reported an expected fire return interval ranging from 3 years in savanna to 7.5 years in wetland (based on fires started every 3 years in the savanna). This difference in expected fire return intervals demonstrates the potential for an invasive, woody seed to arrive at a recently burned wetland with an average of 7.5 years to develop before the next fire. Three of the 6 study species, E. umbellata (Brym et al. 2011), L. sinense (Webster et al. 2006), N. domestica (Knox and Wilson 2006), and T. sebifera (Jones and Mcleod 1989) are documented to be shade tolerant, and there are cases of non-native species increasing shade tolerance at introduced sites (e.g. Brym et al. 2011). Inter-fire periods may provide adequate time and space for non-native individuals to establish and reach sizes that reduce fire related mortalities, especially at the wetter end of the gradient. Moreover, larger woody individuals may deter fire spread through their effects on microclimate and surrounding vegetation (e.g. D’Odorico et al. 2010, Platt et al. 2015), strengthening the fire-deterring fire- vegetation feedback in the wetland.

Understanding the mechanisms that determine the success of an invasive seed after its arrival within this ecotonal gradient is a complex task. Fire (even prescribed fire) occurrence varies spatially and temporally, as do the other biotic and abiotic processes that determine

128

seed fates. Our observed low establishment rates indicate that the current fire management practices appear to be largely effective at filtering five of these six species from the landscape. Despite relatively low establishment rates, at least of three of the species still pose an invasion threat as established adults plants can produce many seeds (e.g. Webster et al.

2006). Moreover, we studied plants during their most vulnerable life history stages, suggesting that they would continue to persist, at least in unburned portions of the landscape.

For example, P. calleryana is a particular threat because it had a number of 1-year old seedlings survive topkill, and more information is needed to understand its response to multiple fires. Finally, we acknowledge the great ecological complexity surrounding invasion, which suggests that while this managed longleaf pine – wetland ecotonal gradient is resistant to invasion, it is not exempt from increased invasion rates from these or other species.

129

References

Arianoutsou, M., and M. Vilà. 2012. Fire and invasive plant species in the Mediterranean basin. Israel Journal of Ecology and Evolotiuon 58:195–203. van Auken, O. W. 2009. Causes and consequences of woody plant encroachment into western North American grasslands. Journal of Environmental Management 90:2931– 2942.

Bates, D., M. Mächler, B. Bolker, and S. Walker. 2015. Fitting Linear Mixed-Effects Models Using lme4. Journal of Statistical Software 67:1–48.

Bond, W. J., and J. E. Keeley. 2005. Fire as a global “herbivore”: the ecology and evolution of flammable ecosystems. Trends in Ecology & Evolution 20:387–394.

Bond, W. J., and G. F. Midgley. 2012. Carbon dioxide and the uneasy interactions of trees and savannah grasses. Philosophical Transactions of the Royal Society B: Biological Sciences 367:601–612.

Briggs, J. M., A. K. Knapp, and B. L. Brock. 2002. Expansion of woody plants in tallgrass prairie: a fifteen-year study of fire and fire-grazing interactions. The American Midland Naturalist 147:287–294.

Brudvig, L. A. 2010. Woody encroachment removal from Midwestern oak savannas alters understory diversity across space and time. Restoration Ecology 18:74–84.

Brym, Z. T., J. K. Lake, D. Allen, and A. Ostling. 2011. Plant functional traits suggest novel ecological strategy for an invasive shrub in an understorey woody plant community. Journal of Applied Ecology 48:1098–1106.

Chapman, H. H. 1932. Is the longleaf type a climax? Ecology 13:328–334.

Culley, T. M., and N. A. Hardiman. 2007. The beginning of a new invasive plant: a history of the ornamental callery pear in the United States. BioScience 57:956.

130

Denham, A. J., R. J. Whelan, T. D. Auld, and R. J. Denham. 2010. The coupling of recruitment and disturbance by fire in two resprouting Proteaceae species. Plant Ecology 212:471–481.

Drew, M. B., L. K. Kirkman, and A. K. Gholson. 1998. The vascular flora of Ichauway, Baker County, Georgia: a remnant longleaf pine / wiregrass ecosystem. Castanea 63:1– 24.

Duncan, B. W., and P. A. Schmalzer. 2004. Anthropogenic influences on potential fire spread in a pyrogenic ecosystem of Florida, USA. Landscape Ecology 19:153–165.

Emery, S. M., J. Uwimbabazi, and S. L. Flory. 2011. Fire intensity effects on seed germination of native and invasive eastern deciduous forest understory plants. Forest Ecology and Management 261:1401–1408.

Faulkner, J. L. 1989. Use of prescribed burning for managing natural and historic resources in Chickamauga and Chattanooga National Military Park, USA. Environmental Management 13:603–612.

Franco, J. A., S. Banon, M. J. Vicente, J. Miralles, and J. J. Martinez-Sanchez. 2011. Root development in horticultural plants grown under abiotic stress conditions - a review. Journal of Horticultural Science & Biotechnology 86:543–556.

Frost, C. 1993. Four centuries of changing landscape patterns in the longleaf pine ecosystem. Pages 17–43 Proceedings of the Tall Timbers fire ecology conference.

Gignoux, J., G. Lahoreau, R. Julliard, and S. Barot. 2009. Establishment and early persistence of tree seedlings in an annually burned savanna. Journal of Ecology 97:484– 495.

Glasgow, L. S., and G. R. Matlack. 2007. The effects of prescribed burning and canopy openness on establishment of two non-native plant species in a deciduous forest, southeast Ohio, USA. Forest Ecology and Management 238:319–329.

Glitzenstein, J. S., D. R. Streng, R. E. Masters, K. M. Robertson, and S. M. Hermann. 2012. Fire-frequency effects on vegetation in north Florida pinelands: another look at the long- term Stoddard Fire Research Plots at Tall Timbers Research Station. Forest Ecology and Management 264:197–209.

Grace, J. B. 1998. Can prescribed fire save the endangered coastal prairie ecosystem from Chinese tallow invasion? Endangered Species Update 15:70–76.

Grace, S. L., and W. J. Platt. 1995. Effects of adult tree density and fire on the demography

131

of pregrass stage juvenile longleaf pine (Pinus palustris Mill.). Journal of Ecology 83:75–86.

Green, S. R., M. A. Arthur, and B. A. Blankenship. 2010. Oak and red maple seedling survival and growth following periodic prescribed fire on xeric ridgetops on the Cumberland Plateau. Forest Ecology and Management 259:2256–2266.

Grubb, P. J. 1988. The uncoupling of disturbance and recruitment, two kinds of seed bank, and persistence of plant populations at the regional and local scales. Annals Zoologica Fennici 25:23–36.

Harrison, X. A. 2015. A comparison of observation-level random effect and beta-binomial models for modelling overdispersion in binomial data in ecology & evolution. PeerJ 3:e1114.

Hatchell, G. E., and D. H. Marx. 1987. Response of longleaf, sand, and loblolly pines to pisolithus ectomycorrhizae and fertilizer on a Sandhills site in South Carolina. Forest Science 33:301–315.

Herrero, M. L., R. C. Torres, and D. Renison. 2015. Do wildfires promote woody species invasion in a fire-adapted ecosystem? Post-fire resprouting of native and non-native woody plants in central Argentina. Environmental Management 57:308–317.

Herring, B. J., and W. S. Judd. 1995. A floristic study of Ichetucknee Springs State Park, Suwannee and Columbia Counties, Florida. Castanea 60:318–369.

Higgins, S. I., W. J. Bond, E. C. February, A. Bronn, I. Douglas, B. Enslin, N. Govender, L. Rademan, S. O. Regan, L. F. Andre, S. Scheiter, R. Sowry, L. Trollope, W. S. W. Trollope, S. Ecology, N. May, I. Higgins, J. Bond, C. February, S. O. Regan, and P. O. B. P. Elizabeth. 2007. Effects of four decades of fire manipulation on woody vegetation structure in savanna. Ecology 88:1119–1125.

Hoffmann, W. A. 1996. The effects of fire and cover on seedling establishment in a neotropical savanna. Journal of Ecology 84:383–393.

Hoffmann, W. A., E. L. Geiger, S. G. Gotsch, D. R. Rossatto, L. C. R. Silva, O. L. Lau, M. Haridasan, and A. C. Franco. 2012. Ecological thresholds at the savanna-forest boundary: how plant traits, resources and fire govern the distribution of tropical biomes. Ecology letters 15:759–68.

Huddle, J., and S. Pallardy. 1999. Effect of fire on survival and growth of Acer rubrum and Quercus seedlings. Forest ecology and management 118:49–56.

132

Ivester, A. H., and D. S. Leigh. 2003. Riverine dunes on the Coastal Plain of Georgia, USA. Geomorphology 51:289–311.

Jenkins, R. A., and P. D. McMillan. 2009. Vascular flora of Sandhill Research and Education Center, Richland County, South Carolina. Castanea 74:168–180.

Jindo, K., S. A. Martim, E. C. Navarro, F. Pérez-Alfocea, T. Hernandez, C. Garcia, N. O. Aguiar, and L. P. Canellas. 2012. Root growth promotion by humic acids from composted and non-composted urban organic wastes. Plant and Soil 353:209–220.

Jones, R. H., and K. W. Mcleod. 1989. Shade tolerance in seedlings of Chinese tallow tree, American sycamore, and cherrybark oak. Bulletin of the Torrey Botanical Club 116:371–377.

Just, M. G., M. G. Hohmann, and W. A. Hoffmann. in press. Where fire stops: vegetation structure and microclimate influence fire spread along an ecotonal gradient. Plant Ecology.

Keeley, J. E. 2006. Fire management impacts on invasive plants in the western United States. Conservation Biology 20:375–384.

Knox, G. W., and S. B. Wilson. 2006. Evaluating North and South Florida landscape performance and fruiting of ten cultivars and a wild-type selection of Nandina domestica, a potentially invasive shrub. Journal of Environmental Horticulture 24:137– 142.

Krall, J. S., M. G. Hohmann, and J. M. Fraterrigo. 2014. Contingent fire effects on granivore removal of exotic woody plant seeds in longleaf pine savannas. Biological Invasions 16:1055–1068.

Kuhn, M. 2008. Building predictive models in R using the caret package. Journal Of Statistical Software 28:1–26.

Kuznetsova, A., P. Brockhoff, and R. Christensen. 2012. lmerTest: tests for random and fixed effects for linear mixed effect models (lmer objects of lme4 package). R package version 2.0. http://cran.r-project.org/package=lmerTest.

Lamont, B. B., E. T. F. Witkowski, and N. J. Enright. 1993. Post-fire litter microsites: safe for seeds, unsafe for seedlings. Ecology 74:501–512.

Lashley, M. A., M. C. Chitwood, A. Prince, M. B. Elfelt, E. L. Kilburg, C. S. DePerno, and C. E. Moorman. 2014. Subtle effects of a managed fire regime: a case study in the longleaf pine ecosystem. Ecological Indicators 38:212–217.

133

Lawes, M. J., A. Richards, J. Dathe, and J. J. Midgley. 2011. Bark thickness determines fire resistance of selected tree species from fire-prone tropical savanna in north Australia. Plant Ecology 212:2057–2069.

Lemmon, P. E. 1956. A spherical densiometer for estimating forest overstory density. Forest Science 2:314–320.

Mandle, L., J. L. Bufford, I. B. Schmidt, and C. C. Daehler. 2011. Woody exotic plant invasions and fire: reciprocal impacts and consequences for native ecosystems. Biological Invasions 13:1815–1827.

Mehlich, A. 1984a. Photometric determination of humic matter in soils, a proposed method. Communications in Soil Science and Plant Analysis 15:1417–1422.

Mehlich, A. 1984b. Mehlich 3 soil test extractant: A modification of Mehlich 2 extractant. Communications in Soil Science and Plant Analysis 15:1409–1416.

Mehlich, A., S. S. Bowling, and A. L. Hatfield. 1976. Buffer pH acidity in relation to nature of soil acidity and expression of lime requirement. Communications in Soil Science and Plant Analysis 7:253–263.

Miller, J. 2003. Nonnative invasive plants of southern forests: a field guide for identification and control. General Technical Report SRS-XX. USDA Forest Service, Southern Research Station, Asheville, NC.

Ne’eman, G., S. Goubitz, and R. Nathan. 2004. Reproductive traits of Pinus halepensis in the light of fire - a critical review. Plant Ecology 171:69–79.

Noss, R. 2012. Forgotten grasslands of the South: natural history and conservation. Island Press, Washington, DC.

Nowacki, G. J., and M. D. Abrams. 2008. The demise of fire and “mesophication” of forests in the Eastern United States. BioScience 58:123–138.

Oliver, C. D. 1978. Subsurface geologic formations and site variation upper sand hills of South Carolina. Journal of forestry 76:352–354.

Outcalt, K. W. 2000. The longleaf pine ecosystem of the South. Native Plants Journal 1:43–

134

53.

Parr, C. L., C. E. R. Lehmann, W. J. Bond, W. A. Hoffmann, and A. N. Andersen. 2014. Tropical grassy biomes: misunderstood, neglected, and under threat. Trends in Ecology & Evolution 29:205–213.

Platt, W. J., A. K. Entrup, E. K. Babl, C. Coryell-Turpin, V. Dao, J. A. Hebert, C. D. LaBarbera, J. F. L. Noto, S. O. Ogundare, L. K. Stamper, and N. Timilsina. 2015. Short- term effects of herbicides and a prescribed fire on restoration of a shrub-encroached pine savanna. Restoration Ecology 23:909–917.

Poblete, H., M. Martinez, and R. Ortega. 2015. Integrated plant nutrition on the recovery of a pear (Pyrus communis “Packam”s Triumph’) orchard. Pages 179–186 in R. Ortega, M. Martinez, and P. Ospina, editors. Second international symposium on organic matter managment and compost use in horticulture. Leuven University Press., Leuven, Belgium.

R Core Team. 2015. R: A language and environment for statistical computing. R Foundation for Statistical Computing, Vienna, Austria.

Renne, I. J., W. C. Barrow Jr, L. A. Johnson Randall, and W. C. Bridges Jr. 2002. Generalized avian dispersal syndrome contributes to Chinese tallow tree (Sapium sebiferum, Euphorbiaceae) invasiveness. Diversity and Distributions 8:285–295.

Ryan, C. M., and M. Williams. 2011. How does fire intensity and frequency affect miombo woodland tree populations and biomass? Ecological Applications 21:48–60.

Ryan, K. C., E. E. Knapp, and J. M. Varner. 2013. Prescribed fire in North American forests and woodlands: history, current practice, and challenges. Frontiers in Ecology and the Environment 11:e15–e24.

Sankaran, M., N. P. Hanan, R. J. Scholes, J. Ratnam, D. J. Augustine, B. S. Cade, J. Gignoux, S. I. Higgins, X. Le Roux, F. Ludwig, J. Ardo, F. Banyikwa, A. Bronn, G. Bucini, K. K. Caylor, M. B. Coughenour, A. Diouf, W. Ekaya, C. J. Feral, E. C. February, P. G. H. Frost, P. Hiernaux, H. Hrabar, K. L. Metzger, H. H. T. Prins, S. Ringrose, W. Sea, J. Tews, J. Worden, and N. Zambatis. 2005. Determinants of woody cover in African savannas. Nature 438:846–849.

135

Simberloff, D. 1998. Flagships, umbrellas, and keystones: is single-species management passé in the landscape era? Biological Conservation 83:247–257.

Smith, D. L., and L. Johnson. 2004. Vegetation-mediated changes in microclimate reduce soil respiration as woodlands expand into grasslands. Ecology 85:3348–3361.

Sorrie, B. A., J. B. Gray, and P. J. Crutchfield. 2006. The vascular flora of the longleaf pine ecosystem of Fort Bragg and Weymouth Woods, North Carolina. Castanea 71:129–161.

Stevens, J. T., and B. Beckage. 2010. Fire Effects on Demography of the Invasive Shrub Brazilian Pepper (Schinus terebinthifolius) in Florida Pine Savannas. Natural Areas Journal 30:53–63.

Trevisan, S., O. Francioso, S. Quaggiotti, and S. Nardi. 2010. Humic substances biological activity at the plant-soil interface: from environmental aspects to molecular factors. Plant signaling & behavior 5:635–43.

Weakley, A. S., and M. P. Schafale. 1991. Classification of pocosins of the Carolina Coastal Plain. Wetlands 11:355–375.

Webster, C. R., M. A. Jenkins, and S. Jose. 2006. Woody invaders and the challenges they pose to forest ecosystems in the eastern United States. Journal of Forestry 104:366–374.

Wells, B. W., and I. V. Shunk. 1931. The vegetation and habitat factors of the coarser sands of the North Carolina Coastal Plain: an ecological study. Ecological Monographs 1:465– 520.

Wilson, J. B., and A. D. Q. Agnew. 1992. Positive-feedback switches in plant communities. Pages 263–336 in M. Begon and A. Fitter, editors. Advances in Ecological Research. Academic Press London.

136

Tables

Table 5.1 Woody species used in this experiment

Species Common name Family No. seeds per subplot Elaeagnus umbellata Thunb. autumn olive Elaeagnaceae 40 Ligustrum sinense Lour. Chinese privet Oleaceae 40 Melia azedarach L. Chinaberry Meliaceae 40 Nandina domestica Thunb. nandina Berberidaceae 25 Pyrus calleryana Dcn. Bradford pear Rosaceae 30 Triadica sebifera (L.) Small Chinese tallow tree Euphorbiaceae 20

137

Table 5.2 Mean emergence, survival, and establishment for our six study species. Emergence is defined as the proportion of seedlings to seeds sown at the end of the first growing season. Two-year survival is defined as the proportion of emergent seedlings that survived until the end of a 2-year period beginning at the end of the first growing season. Establishment is defined as the proportion of seedlings at the end of the study period to seeds sown

Species Emergence Two-year survival Establishment Unburned, burned Unburned, burned E. umbellata 3.56 (0.84) 3.04 (2.26), 0 (0) 0 (0), 0 (0) L. sinense 1.06 (0.32) 0 (0), 0 (0) 0 (0), 0 (0) M. azedarach 5.65 (0.84) 11.44 (4.99), 0 (0) 0 (0), 0 (0) N. domestica 9.56 (1.93) 49.21 (10.76), 0 (0) 3.87 (2.58), 0 (0) P. calleryana 8.24 (1.47) 61.85 (9.71), 21.82 (12.20) 9.78 (3.62), 1.07 (0.45) T. sebifera 8.61 (1.45) 43.21 (8.52), 0 (0) 7.83 (3.51), 0 (0) Standard errors of the mean are in parentheses

138

Table 5.3 Type II Wald χ2 test statistic values from generalized linear mixed-effect models (GLMM) evaluating the effect of vegetative community, seed-sowing treatment, or species identity on germination, survival, and emergence for our six study species and all species combined. Emergence is the proportion of seedlings to seeds sown at the end of the first growing season. Survival is the proportion of emergent seedlings that survived until the end of a 2-year period beginning at the end of the first growing season. Establishment is the proportion of seedlings at the end of the study period (Sept. 2015) to seeds sown

Significance levels for type II Wald χ2 test statistic: ***P < 0.001,**P < 0.01, *P < 0.05

139

Table 5.4 Number and proportion (seedlings / seeds sown) of established seedlings by seed-sowing treatment (undisturbed [UD], litter removal and undisturbed soil [LRU], litter removal and covered with loose soil [LRS]) in unburned and burned sites

Species UD LRU LRS Unburned Burned Unburned Burned Unburned Burned N. domestica 0, 0% 0, 0% 0, 0% 0, 0% 29, 11.60% 0, 0% P. calleryana 6, 2.00% 3, 0.37% 16, 5.33% 0, 0% 66, 22.00% 22, 2.93% T. sebifera 0, 0% 0, 0% 0, 0% 0, 0% 47, 23.50%, 0, 0%

140

Table 5.5 Mean soil properties and environmental variables of the study plots (Welch’s unequal variance t-test)

Variable Savanna plots Ecotone plots t-value Soil bulk density (g cm-3) 1.16 (0.01) 1.01 (0.01) -3.66** Humic matter (%) 0.96 (0.03) 1.47 (0.03) 2.71* Soil moisture (%) 7.36 (0.12) 10.58 (0.24) 2.73* Canopy closure (%) 69.21 (0.83) 71.25 (0.89) 0.38 Standard errors of the mean are in parentheses. Significance levels for t-test between savanna and ecotone plots: **P < 0.01, *P < 0.05

141

Table 5.6 Mean survival rates (proportion of seedlings that survived from beginning until the end of the survival period) considering all species

Survival period Overall Unburned Burned September 2012 – 42.10 (6.34) NA NA May 2013

May 2013 – 61.89 (2.03) 68.97 (2.09) 17.72 (4.32) May 2014

May 2014 – 56.27 (3.07) 59.04 (3.12) 7.14 (7.14) May 2015 May 2015 – 87.41 (2.87) 100.00 (0) 15.00 (8.19) September 2015 Standard errors of the mean are in parentheses

142

Figures

Fig. 5.1 Conceptual model of the longleaf pine savanna – wetland ecotonal gradient with paired community plot arrangement

143

Fig. 5.2 Effect of seed-sowing treatment (undisturbed [green], litter removal and undisturbed soil [orange], litter removal and covered with loose soil [purple]) on seedling emergence, survival, and establishment. (a) Emergence is the proportion of seedlings to seeds sown at the end of the first growing season. (b) Two-year survival is the proportion of emergent seedlings that survived until the end of a 2-year period beginning at the end of the first growing season. (c) Establishment is the proportion of seedlings at the end of the study period (Sept. 2015) to seeds sown. Error bars represent the standard error of the mean. Significance levels for type II Wald χ2 test statistic: ***P < 0.001,**P < 0.01, *P < 0.05, NS = not significant

144

Fig. 5.3 Predicted probability of P. calleryana seedling survival after a topkilling fire based soil humic matter content by seed-sowing treatment (undisturbed [green], litter removal and undisturbed soil [orange], litter removal and covered with loose soil [purple])

145

Fig. 5.4 Line is the predicted probability of a P. calleryana seedling surviving a topkilling fire based on its height. Diamonds represent observed data points

146

147

Chapter 6: Summary

The goal of this dissertation was to provide increased ecological understanding of some of the determinants and roles of vegetation structure along fire-promoting – fire-deterring ecotonal gradients, specifically the longleaf pine savanna – wetland gradients located in the

Sandhills physiographic region of North Carolina, USA. I accomplished this though a series of research questions and experiments, which are summarized in following text.

In Chapter 2, I identified the important vegetation structure and microclimate variables that determine flammability along the longleaf pine savanna – wetland gradients. I

-2 -1 found that C4 grass cover (%) and photosynthetically active radiation (µmol m s ) promoted gradient flammability, and that evergreen shrub cover (%) deterred it. I then developed a statistical model to predict flammability along the gradient using these identified drivers. Moreover, I identified C4 grass cover as the most important predictor of flammability along these gradients, explaining 67% of the variance explained by the model. Fire only has a direct effect on vegetation structure where burning actually occurs. Thus, identifying the factors that regulate flammability between fire-promoting and fire-deterring communities is necessary to make worthwhile predictions of vegetation distributions under environmental change.

Chapter 3 considers the stability of community boundaries by way of vegetation structure distributions and gradient flammability under altered fire frequencies. Using empirically parameterized models, I was able to simulate what might happen to these gradients if fire became more or less frequent (I modified fire frequencies as a proxy for environmental change). I found that the vegetation structure along these gradients was able to

148

rebound quickly after burning, suggesting that fire-vegetation feedbacks along these gradients may not control vegetation boundaries as strongly as other savanna – forest gradients. I observed an inverse relationship between fire frequency and gradient flammability, suggesting that these vegetation boundaries are likely to experience changes given an extended period of modified fire frequencies. However, I would not expect these structure changes to be as extreme as observed in other savanna – forest gradients.

In Chapter 4 I studied vegetation structure in light of resprouting woody individuals in frequently burned ecosystems. Many woody plant species in frequently burned ecosystems use resprouting as a strategy to endure repeated topkill. The persistence of these individuals is usually attributed to their ability, or lack thereof, to store and remobilize belowground energy reserves. However, given that there is sometimes partitioning of capable resprouters between areas of high or low fire activity, I set up an experiment to consider an alternative explanation – wood decay resistance – for persistence of resprouting species. I examined five resprouting tree species, of which four were restricted to less frequently burned portions of the gradient and the fifth was fire-resistant and located the upland. I found that the upland species (Quercus laevis) had the least root crown decay, and greatest values for traits known to confer decay resistance (i.e. wood density, lignin, and extractable phenolic concentrations) as compared to the other species. This study provides evidence for an alternative mechanism that contributes to the persistence of woody vegetation structure in frequently burned landscapes.

Chapter 5 examines the invasibility of this fire-managed ecotone by six non-native, woody species. I designed an experiment to investigate the effects of fire and site conditions

149

on the invasion success of these species. I observed individuals for three of the six species survive and establish by the end of the experiment in sites that did not burn. Only one species, Pyrus calleryana, had individuals establish in sites that had burned after seedling emergence. Size was positively correlated with survival, and this suggests that individuals that arrive at portions of the gradient with greater resource availability or longer fire return intervals would receive boosts to their growth and consequently, establishment success, as compared to seedling in other areas. Increases in the abundance of woody individuals can alter vegetation structure and fire behavior, such as encroachment upon the savanna, which can lead to the eventual exclusion of shade-intolerant species (e.g. C4 grass).

Overall, the scientific contribution of this dissertation is progress towards a more complete understanding of vegetation structure. This research not only increases our knowledge about present-day conditions, but also provides information required to make predictions about these gradients under environmental change. Moreover, it also provide additional baseline ecological knowledge about vegetation structure in the longleaf pine ecosystem, one of the most species rich and imperiled ecosystems in North America.

150

Appendices

151

Appendix A: Chapter 2 Online Resources

*Accepted for publication in Plant Ecology, the final publication is available at http://link.springer.com/article/10.1007/s11258-015-0545-x, with kind permission from Springer Science+Business Media

Online Resource 1 Plot fuel bulk density:

We calculated plot fuel bulk density (BD) as:

oven‐dried fuel mass (kg) 퐵퐷 = 푓푢푒푙 푣표푙푢푚푒 (푚3)

We found no significant relationships between fuel mass and relative gradient position, or vegetation structure predictors (Online Resource 3). As such we used the mean plot oven- dried fuel mass (0.704 [0.040 SE] kg) from our subset of fuel transects in our BD calculations. This fuel mass was calculated from the following six vegetation structure functional types: C4 grass, C3 graminoids, fern, herbaceous dicots, and switchcane. These are vegetation structure functional types that most readily burn along our ecotonal gradients.

We calculated the fuel volume for each plot (full dataset). First, we calculated the volume 3 (m ) for the following three vegetation structure functional types: C4 grass, fern, and switchcane. This was calculated as:

푣푒푔푒푡푎푡𝑖표푛 푠푡푟푢푐푡푢푟푒 푓푢푛푐푡𝑖표푛푎푙 푡푦푝푒 푣표푙푢푚푒 (푚3) = ℎ푒𝑖푔ℎ푡 (푚) × 푎푟푒푎 (푚2)

Area of the vegetation structure functional types was derived from the percent cover of 1 m2 plot that they occupied. We then computed the volume of the fuelbed of the plot where height was the fuelbed depth and area was calculated as the difference between the sum of the percent cover of the three vegetation structure functional types and 100%.

In our BD calculations plot fuel volume was calculated as the weighted average of the volume of the three vegetation structure functional types and the fuelbed.

152

Online Resource 2 Surface area to volume ratio (SV) calculations:

We used the surface area to volume ratio formulas proposed by van Wagtendonk (2006) for our analyses.

The SV of cylindrical fuels (e.g. pine needles, stems) was calculated as:

4 푆푉 = 푑𝑖푎푚푒푡푒푟 표푓 푓푢푒푙 푝푎푟푡𝑖푐푙푒

The SV of ‘flat’ fuels such as broadleaves was calculated as:

2 푆푉 = 푡ℎ𝑖푐푘푛푒푠푠 표푓 푓푢푒푙 푝푎푟푡𝑖푐푙푒

van Wagtendonk JW (2006) Fire as a physical process. In: Sugihara NG, van Wagtendonk JW, Shaffer KE, Fites-Kaufman J, Thode AE (eds) Fire in California’s ecosystems. University of California Press, Ltd., Berkley and Los Angeles, CA, pp 38–57

153

Online Resource 3 Results of linear mixed-effect models evaluating total plot fuel mass (kg) by vegetation structure functional type, or plot variable. Fixed effect: variable, Random effects: uncorrelated random intercept (site) and random slope (variable by site) Variable Estimate (SE) DF t-value Marginal R2 Years since fire 0.131 (0.049) 18 2.7* 0.112 No. of fires -0.066 (0.041) 20 -1.6NS 0.048 Switchcane cover (%) 0.006 (0.006) 3 1.0NS 0.027 Deciduous tree cover (%) -0.003 (0.002) 89 -1.4NS 0.018 Evergreen tree cover (%) -0.006 (0.004) 91 -1.3NS 0.016 Plot cover (%) -0.001 (0.001) 93 -0.9NS 0.008 Evergreen shrub cover (%) -0.002 (0.002) 11 -0.8NS 0.008 Relative gradient position -0.098 (0.111) 34 -0.9NS 0.007 NS C4 grass cover (%) 0.003 (0.005) 12 0.6 0.007 Fern cover (%) 0.001 (0.003) 4 0.6NS 0.003 Woody debris cover (%) -0.007 (0.012) 82 -0.5NS 0.003 Deciduous cover (%) -0.001 (0.002) 82 -0.4NS 0.001 Herbaceous dicot cover (%) -0.002 (0.007) 88 -0.3NS 0.001 NS C3 graminoid cover (%) 0.002 (0.011) 81 0.1 0.000 Canopy closure (%) 0.000 (0.001) 82 -0.1NS 0.000 Dead surface fuel depth (cm) 0.001 (0.015) 85 0.0NS 0.000 *P < 0.05, NS = not significant

154

Online Resource 4

Top Panel: Scatterplot of total plot fuel mass (kg) by relative gradient position. Bottom panel: Boxplots of total plot fuel mass by relative gradient position. The solid center bar represents the median and the asterisk depicts the mean. The letters represent a post-hoc Tukey’s HSD test. These letters are all the same indicating that relative gradient positions are not significantly different in their mean plot fuel mass.

155

Online Resource 5

Boxplots of the pine needle proportion of total dead fuel mass by relative gradient position. The solid center bar represents the median and the asterisk depicts the mean. The letters represent a post-hoc Tukey’s HSD test. Relative gradient positions with different letters are significantly different.

156

Online Resource 6 Results of linear mixed-effect models evaluating total class fuel mass per plot (kg) or class proportional fuel mass by relative gradient position. Fixed effect: relative gradient position, Random effects: uncorrelated random intercept (site) and random slope (relative gradient position by site) Fuel class Estimate (SE) DF t-value Marginal R2 Total class fuel mass per plot Herbaceous dicot -0.024 (0.014) 11 -1.7NS 0.197 Pinecone 0.226 (0.114) 6 2.0NS 0.172 Pine needles -0.191 (0.048) 30 -4.0*** 0.133 Deciduous tree 0.098 (0.082) 7 1.2NS 0.092 Evergreen tree 0.031 (0.022) 18 1.4NS 0.079 Fern 0.068 (0.054) 28 1.3NS 0.052 Deciduous shrub 0.053 (0.026) 44 2.0* 0.047 Switchcane 0.068 (0.056) 36 1.2NS 0.038 Fine woody fuels 0.083 (0.063) 50 1.3NS 0.025 NS C4 grass -0.029 (0.042) 29 -0.7 0.013 Evergreen shrub 0.033 (0.130) 32 0.3NS 0.002 Bark 0.006 (0.021) 51 0.3NS 0.001 NS C3 graminoids -0.003 (0.009) 7 -0.3 0.001 Miscellaneous plant matter 0.006 (0.042) 39 0.1NS 0.000 Proportion of total fuel mass per plot Pine needles -33.237 (4.477) 46 -7.4*** 0.366 Pinecone 12.751 (5.537) 5 2.3NS 0.225 Herbaceous dicot -3.775 (2.516) 11 -1.5NS 0.158 Deciduous tree 14.542 (12.478) 4 1.2NS 0.127 Evergreen tree 3.950 (2.793) 18 1.4NS 0.087 Switchcane 8.168 (5.588) 36 1.5NS 0.053 NS C4 grass -6.773 (4.822) 29 -1.4 0.051 Fern 5.333 (4.227) 28 1.3NS 0.051 Fine woody fuels 8.769 (4.582) 39 1.9NS 0.049 Deciduous shrub 9.369 (4.854) 53 1.9NS 0.046 NS C3 graminoids -1.176 (1.246) 7 -0.9 0.008 Evergreen shrub -4.281 (9.529) 32 -0.4NS 0.006 Bark -0.295 (1.735) 56 -0.2NS 0.000 Miscellaneous plant matter 0.104 (2.809) 68 0.0NS 0.000 ***P < 0.001, *P < 0.05, NS = not significant

157

Online Resource 7

Scatterplot of maximum fire temperature (°C; as indicated by temperature indicating paint at 20 cm above ground level) by relative gradient position

158

Online Resource 8 Results of linear mixed-effect models evaluating maximum fire temperature (°C; as indicated by temperature indicating paint at 20 cm above ground level) by vegetation structure functional type, site, microclimate, or plot variable. Fixed effect: variable, Random effects: uncorrelated random intercept (site) and random slope (variable by site) Variable Estimate (SE) DF t-value Marginal R2 Years since fire 43.63 (10.01) 35 4.4*** 0.160 Resultant 172.37 (59.13) 42 2.9** 0.062 No. of fires -19.14 (8.07) 37 -2.4* 0.056 Dead fuel depth (cm) 9.63 (4.34) 116 2.2* 0.037 Canopy closure (%) 0.97 (0.47) 104 2.1* 0.032 Evergreen tree cover (%) -3.27 (1.38) 149 -2.4* 0.032 * C4 grass cover (%) 1.63 (0.72) 146 2.3 0.029 Deciduous tree cover (%) 1.44 (0.81) 20 1.8NS 0.025 Switchcane cover (%) -2.45 (1.19) 144 -2.1* 0.024 Relative humidity (%) -1.69 (1.13) 78 -1.5NS 0.021 Wind speed (m s-1) 66.14 (45.17) 54 1.5NS 0.018 Deciduous shrub cover (%) 0.92 (0.54) 141 1.7NS 0.016 Bulk Density (kg m-3) -3.02 (1.98) 152 -1.5NS 0.015 Soil moisture (%) -160.84 (95.37) 133 -1.7NS 0.015 Coarse fuel moisture (%) -5.67 (4.20) 125 -1.3NS 0.015 Wood debris cover (%) -4.95 (3.52) 142 -1.4NS 0.011 Plot cover (%) 0.45 (0.38) 79 1.2NS 0.010 Air Temperature (°C) -3.59 (3.83) 50 -0.9NS 0.009 Precipitation throughfall (mm d-1) -15.47 (13.39) 152 -1.2NS 0.009 Wind direction (°) -0.09 (0.11) 21 -0.8NS 0.007 Fern cover (%) -0.66 (0.72) 148 -0.9NS 0.006 Evergreen shrub cover (%) -0.59 (0.72) 140 -0.8NS 0.004 Fuel temperature (°C) -1.80 (3.06) 81 -0.6NS 0.003 Vapor pressure deficit (kPa) 14.19 (24.70) 40 0.6NS 0.003 Photosynthetically active radiation (µmol m-2 s-1) 0.01 (0.03) 149 0.2NS 0.000 Herbaceous dicot cover (%) 0.47 (2.31) 6 0.2NS 0.000 ***P < 0.001, **P < 0.01, *P < 0.05, NS = not significant

159

Online Resource 9

Conceptual diagram of three possible ecotonal gradients (depicted by the solid lines) by gradient position and fire frequency at our study location. 1) If fires frequently burn the entire gradient, we expect savanna upland and herbaceous (e.g. switchcane, fern) wetland. 2) If fires reliably extinguish in the ecotone between the savanna and wetland, we expect savanna upland and shrubby wetland 3) If fires are infrequent or absent along the entire gradient, we expect dry forest upland and shrubby wetland.

160

Appendix B: Chapter 3 Supplementary Figures

Fig. B.1 The mean proportion of a cell occupied by either vegetation structure (C4 grass [solid brown line], evergreen shrub [dashed green line]) type by time steps under a 1-yr fire frequency arranged by gradient position (columns) and vegetation initialization scheme (rows). The proportion of cell occupation was calculated as the mean across lattices (n = 100) by cell gradient position. Cells in the Field 2012 schemes were initiated with our observed (2012) vegetation structure values. Cells in the High Grass schemes were initiated 3 3 with a high value for C4 grass (0.25 m ) and a low value for evergreen shrub (0.025 m ), and vice versa for the lattice cells of the High Shrub schemes. This plot depicts the results for our model instances that were simulated with field-based and enhanced-feedback rules for updating flammability and vegetation response to burn status.

161

Fig. B.2 The mean proportion of a cell occupied by either vegetation structure (C4 grass [solid brown line], evergreen shrub [dashed green line]) type by time steps under a 3-yr fire frequency arranged by gradient position (columns) and vegetation initialization scheme (rows). The proportion of cell occupation was calculated as the mean across lattices (n = 100) by cell gradient position. Cells in the Field 2012 schemes were initiated with our observed (2012) vegetation structure values. Cells in the High Grass schemes were initiated 3 3 with a high value for C4 grass (0.25 m ) and a low value for evergreen shrub (0.025 m ), and vice versa for the lattice cells of the High Shrub schemes. This plot depicts the results for our model instances that were simulated with field-based and enhanced-feedback rules for updating flammability and vegetation response to burn status.

162

Fig. B.3 The mean proportion of a cell occupied by either vegetation structure (C4 grass [solid brown line], evergreen shrub [dashed green line]) type by time steps under a 9-yr fire frequency arranged by gradient position (columns) and vegetation initialization scheme (rows). The proportion of cell occupation was calculated as the mean across lattices (n = 100) by cell gradient position. Cells in the Field 2012 schemes were initiated with our observed (2012) vegetation structure values. Cells in the High Grass schemes were initiated 3 3 with a high value for C4 grass (0.25 m ) and a low value for evergreen shrub (0.025 m ), and vice versa for the lattice cells of the High Shrub schemes. This plot depicts the results for our model instances that were simulated with field-based and enhanced-feedback rules for updating flammability and vegetation response to burn status

163

Fig. B.4 The mean proportion of a cell occupied by either vegetation structure (C4 grass [solid brown line], evergreen shrub [dashed green line]) type by time steps under fire suppression arranged by gradient position (columns) and vegetation initialization scheme (rows). The proportion of cell occupation was calculated as the mean across lattices (n = 100) by cell gradient position. Cells in the Field 2012 schemes were initiated with our observed (2012) vegetation structure values. Cells in the High Grass schemes were initiated with a high value 3 3 for C4 grass (0.25 m ) and a low value for evergreen shrub (0.025 m ), and vice versa for the lattice cells of the High Shrub schemes. This plot depicts the results for our model instances that were simulated with field-based and enhanced-feedback rules for updating flammability and vegetation response to burn status.

164

Fig. B.5 The mean proportion of plots by vegetation type (C4 grass [solid brown line], evergreen shrub [dashed green line]) with non-negligible volume (i.e. C4 grass > 0.004 m3, evergreen shrub > 0.019 m3) by time steps under a 1-yr fire frequency arranged by gradient position (columns) and vegetation initialization scheme (rows). The proportion of cells with non-negligible volume was the mean across lattices (n = 100) by cell gradient position. Cells in the Field 2012 schemes were initiated with our observed (2012) vegetation structure values. 3 Cells in the High Grass schemes were initiated with a high value for C4 grass (0.25 m ) and a low value for evergreen shrub (0.025 m3), and vice versa for the lattice cells of the High Shrub schemes. This plot depicts the results for our model instances that were simulated with field-based and enhanced-feedback rules for updating flammability and vegetation response to burn status.

165

Fig. B.6 The mean proportion of plots by vegetation type (C4 grass [solid brown line], evergreen shrub [dashed green line]) with non-negligible volume (i.e. C4 grass > 0.004 m3, evergreen shrub > 0.019 m3) by time steps under a 3-yr fire frequency arranged by gradient position (columns) and vegetation initialization scheme (rows). The proportion of cells with non-negligible volume was the mean across lattices (n = 100) by cell gradient position. Cells in the Field 2012 schemes were initiated with our observed (2012) vegetation structure values. 3 Cells in the High Grass schemes were initiated with a high value for C4 grass (0.25 m ) and a low value for evergreen shrub (0.025 m3), and vice versa for the lattice cells of the High Shrub schemes. This plot depicts the results for our model instances that were simulated with field-based and enhanced-feedback rules for updating flammability and vegetation response to burn status.

166

Fig. B.7 The mean proportion of plots by vegetation type (C4 grass [solid brown line], evergreen shrub [dashed green line]) with non-negligible volume (i.e. C4 grass > 0.004 m3, evergreen shrub > 0.019 m3) by time steps under a 9-yr fire frequency arranged by gradient position (columns) and vegetation initialization scheme (rows). The proportion of cells with non-negligible volume was the mean across lattices (n = 100) by cell gradient position. Cells in the Field 2012 schemes were initiated with our observed (2012) vegetation structure values. 3 Cells in the High Grass schemes were initiated with a high value for C4 grass (0.25 m ) and a low value for evergreen shrub (0.025 m3), and vice versa for the lattice cells of the High Shrub schemes. This plot depicts the results for our model instances that were simulated with field-based and enhanced-feedback rules for updating flammability and vegetation response to burn status.

167

Fig. B.8 The mean proportion of plots by vegetation type (C4 grass [solid brown line], evergreen shrub [dashed green line]) with non-negligible volume (i.e. C4 grass > 0.004 m3, evergreen shrub > 0.019 m3) by time steps under fire suppression arranged by gradient position (columns) and vegetation initialization scheme (rows). The proportion of cells with non-negligible volume was the mean across lattices (n = 100) by cell gradient position. Cells in the Field 2012 schemes were initiated with our observed (2012) vegetation structure values. Cells in the 3 High Grass schemes were initiated with a high value for C4 grass (0.25 m ) and a low value for evergreen shrub (0.025 m3), and vice versa for the lattice cells of the High Shrub schemes. This plot depicts the results for our model instances that were simulated with field-based and enhanced-feedback rules for updating flammability and vegetation response to burn status.

168

Appendix C: Chapter 5 Supplementary Tables

Table C.1 Results (loadings and summary) for a principal-component analysis on soil properties Soil property PC1 PC2 Cation exchange capacity (meq 100 cm-3) 0.4554 -0.0980 Base saturation (%) -0.0855 0.4427 Exchangeable acidity (meq 100 cm-3) 0.4348 -0.1673 pH -0.2683 0.4518 P (mg dm-3) 0.1732 0.0460 K (mg dm-3) 0.3070 0.1203 Ca (mg dm-3) 0.3215 0.3098 Mg (mg dm-3) 0.3945 0.1741 S (mg dm-3) -0.0290 0.2726 Mn (mg dm-3) 0.1698 0.4046 Zn (mg dm-3) 0.1584 0.1846 Cu (mg dm-3) -0.1563 0.3641 Na (mg dm-3) 0.2483 0.1216

Summary Standard deviation 2.09 1.65 Proportion of variance 33.71% 21.03% Cumulative proportion 33.71% 54.74%

169

Table C.2 Results of linear mixed-effect models evaluating May 2013 seedling volume (cm3) by environmental variables. Fixed effect: variable, Random effects: uncorrelated random intercept (site). Marginal Species Variable Est. SE df z P AIC R2 All species Soil bulk density (g cm-3) 0.27 0.11 103.38 2.35 0.020 0.02 20.25 Soil moisture (%) -0.89 0.48 159.10 -1.87 0.063 0.01 22.35 PC2 0.02 0.01 139.28 1.72 0.087 0.01 22.55 Years since fire -0.05 0.03 13.54 -1.79 0.095 0.02 21.79 Canopy closure (%) 0.00 0.00 55.23 -1.38 0.174 0.01 23.16 Humic matter (%) 0.02 0.02 158.31 0.83 0.407 0.00 24.51 PC1 0.00 0.01 352.76 -0.39 0.694 0.00 24.96

E. umbellata Years since fire -0.02 0.02 4.00 -1.07 0.344 0.19 -19.29 PC1 -0.04 0.04 4.00 -0.97 0.385 0.16 -19.05 Canopy closure (%) 0.00 0.00 4.00 -0.92 0.408 0.15 -18.93 PC2 0.00 0.01 4.00 -0.62 0.571 0.07 -18.32 Soil moisture (%) 0.56 1.03 1.26 0.54 0.667 0.06 -18.23 Soil bulk density (g cm-3) -0.06 0.27 1.94 -0.23 0.843 0.01 -17.90 Humic matter (%) -0.01 0.09 2.08 -0.16 0.890 0.01 -17.84

M. azedarach Soil moisture (%) 1.06 0.47 56.00 2.26 0.028 0.08 -172.41 Soil bulk density (g cm-3) -0.10 0.05 56.00 -2.03 0.047 0.07 -171.46 Years since fire -0.01 0.01 56.00 -1.79 0.078 0.05 -170.57 Humic matter (%) 0.01 0.01 56.00 1.08 0.285 0.02 -168.53 PC2 0.00 0.00 20.36 -0.78 0.446 0.01 -167.76 Canopy closure (%) 0.00 0.00 5.36 0.26 0.808 0.00 -167.44 PC1 0.00 0.01 12.11 0.24 0.812 0.00 -167.44

N. domestica PC2 0.01 0.00 54.49 2.58 0.013 0.06 -519.99 Humic matter (%) -0.02 0.01 51.88 -1.36 0.180 0.02 -515.24 Years since fire 0.01 0.01 8.59 1.14 0.285 0.04 -514.79 Soil moisture (%) 0.21 0.24 38.32 0.86 0.398 0.01 -514.07 Canopy closure (%) 0.00 0.00 33.16 0.79 0.433 0.01 -513.93 Soil bulk density (g cm-3) 0.02 0.05 82.48 0.32 0.753 0.00 -513.41 PC1 0.00 0.00 164.59 -0.09 0.926 0.00 -513.32

170

Table C.2 (continued). Results of linear mixed-effect models evaluating May 2013 seedling volume (cm3) by environmental variables. Fixed effect: variable, Random effects: uncorrelated random intercept (site). Marginal Species Variable Est. SE df z P AIC R2 P. calleryana Humic matter (%) 0.14 0.03 57.71 4.78 0.000 0.19 -102.21 Canopy closure (%) -0.01 0.00 21.62 -2.92 0.008 0.08 -90.17 Soil moisture (%) -0.37 0.52 26.86 -0.71 0.481 0.00 -84.12 PC1 0.01 0.01 64.77 0.60 0.549 0.00 -84.42 Soil bulk density (g cm-3) 0.06 0.13 23.59 0.47 0.643 0.00 -84.22 PC2 0.01 0.01 36.03 0.39 0.699 0.00 -84.13 Years since fire 0.00 0.03 16.78 -0.01 0.989 0.00 -84.08

T.sebifera PC1 -0.03 0.02 109.86 -1.44 0.154 0.02 61.33 Soil moisture (%) 1.15 1.12 82.64 1.03 0.304 0.02 62.35 Canopy closure (%) 0.00 0.00 33.05 -0.53 0.602 0.01 63.05 Humic matter (%) -0.04 0.08 65.87 -0.47 0.637 0.00 63.13 Years since fire -0.03 0.09 14.82 -0.29 0.772 0.00 63.27 PC2 0.01 0.03 40.54 0.28 0.783 0.00 63.23 Soil bulk density (g cm-3) -0.03 0.30 66.85 -0.10 0.921 0.00 63.33

171

Table C.3 Results of generalized linear mixed-effect models evaluating 1-year (May 2013 – 2014) survival rate (proportion of seedlings that survived from the beginning until the end of the 1-yr period) by environmental variables for unburned sites. Fixed effect: variable, Random effects: uncorrelated random intercept (site) Marginal Species Variable Est. SE z P AIC R2 All Species PC2 0.28 0.17 1.59 0.111 0.03 260.37 Soil moisture (%) 12.99 9.99 1.30 0.194 0.02 261.23 Years since fire -0.35 0.32 -1.09 0.277 0.01 261.78 Canopy closure (%) -0.02 0.02 -0.92 0.359 0.01 262.13 Soil bulk density (g cm-3) -0.89 1.95 -0.45 0.649 0.00 262.77 PC1 -0.01 0.17 -0.04 0.971 0.00 262.98 Humic matter (%) 0.00 0.45 0.01 0.996 0.00 262.98 M. azedarach PC2 0.35 0.16 2.16 0.031 0.14 39.94 Humic matter (%) -1.34 0.83 -1.61 0.107 0.08 45.11 Soil moisture (%) -129.74 84.78 -1.53 0.126 0.41 41.60 PC1 -0.52 0.42 -1.25 0.213 0.10 46.32 Soil bulk density (g cm-3) -1.75 1.96 -0.90 0.370 0.01 46.47 Years since fire -0.38 0.53 -0.71 0.476 0.03 40.09 Canopy closure (%) 0.00 0.02 0.17 0.864 0.00 47.13 N. domestica PC2 0.56 0.26 2.13 0.033 0.15 81.48 Canopy closure (%) -0.03 0.03 -0.96 0.339 0.04 85.22 Soil moisture (%) -9.57 12.12 -0.79 0.429 0.02 85.65 Years since fire -0.09 0.44 -0.20 0.842 0.00 86.19 PC1 -0.04 0.23 -0.17 0.864 0.00 86.20 Humic matter (%) -0.08 0.65 -0.13 0.900 0.00 86.21 Soil bulk density (g cm-3) 0.25 2.94 0.09 0.932 0.00 86.22 P. calleryana Soil moisture (%) 21.63 16.07 1.35 0.178 0.10 53.38 Years since fire -0.73 0.57 -1.28 0.202 0.05 53.49 PC2 0.25 0.40 0.63 0.531 0.02 54.54 PC1 0.10 0.28 0.34 0.731 0.01 54.85 Humic matter (%) -0.12 0.79 -0.15 0.882 0.00 54.94 Soil bulk density (g cm-3) 0.27 3.58 0.08 0.940 0.00 54.95 Canopy closure (%) 0.00 0.04 0.06 0.954 0.00 54.96 T. sebifera Soil moisture (%) 41.85 16.60 2.52 0.012 0.21 73.25 Canopy closure (%) -0.05 0.03 -1.58 0.113 0.10 78.23 Soil bulk density (g cm-3) -1.41 3.48 -0.40 0.686 0.01 80.91 Years since fire 0.24 0.69 0.35 0.726 0.01 80.96 PC1 -0.10 0.32 -0.29 0.770 0.00 81.00 PC2 -0.06 0.31 -0.20 0.845 0.00 81.05 Humic matter (%) 0.05 0.84 0.06 0.953 0.00 81.08

172

Table C.4 Results of generalized linear mixed-effect models evaluating 1-year (May 2013 – 2014) survival rate (proportion of seedlings that survived from the beginning until the end of the 1-yr period) by environmental variables for burned sites. Fixed effect: variable, Random effects: uncorrelated random intercept (site) Marginal Species Variable Est. SE z P AIC R2 P. calleryana Humic matter (%) 2.34 0.71 3.32 0.001 0.31 32.81 Canopy closure (%) -0.26 0.09 -2.72 0.006 0.22 40.58 PC1 0.60 0.22 2.68 0.007 0.28 44.14 PC2 1.83 0.73 2.49 0.013 0.21 45.67 Soil bulk density (g cm-3) 17.25 8.68 1.99 0.047 0.18 48.49 Years since fire -0.92 0.71 -1.30 0.071 0.25 46.27 Soil moisture (%) -21.09 22.43 -0.94 0.347 0.06 51.72

173

Table C.5 Results of generalized linear mixed-effect models evaluating 1-year (May 2013 – 2014) survival rate (proportion of seedlings that survived from the beginning until the end of the 1-yr period) by seedling size (cm3). Fixed effect: variable, Random effects: uncorrelated random intercept (site). Marginal Burn status Species Est. SE z P AIC R2 Unburned All species 3.91 0.73 5.33 0.000 0.47 551.86 N. domestica 9.61 2.45 3.92 0.000 0.16 224.75 M. azedarach 6.44 4.92 1.31 0.191 0.04 78.87 P. calleryana 26.37 8.42 3.13 0.002 0.89 69.57 T. sebifera 5.79 1.30 4.46 0.000 0.79 92.18 Burned P. calleryana 6.74 2.83 2.38 0.017 0.64 59.10

174