CONTAGIOUS DISTURBANCE AND ECOLOGICAL RESILIENCE
GARRY D. PETERSON
A DISSERTATION PRESENTED TO THE GRADUATE SCHOOL
OF THE UNIVERSITY OF FLORIDA IN PARTIAL FULFILLMENT
OF THE REQUIREMENTS FOR THE DEGREE OF
DOCTOR OF PHILOSOPHY
UNIVERSITY OF FLORIDA
Many people helped sustain, enrich and strengthen this dissertation. Without them
this document would not exist. Buzz Holling encouraged me to avoid the deceptively safe
traps of stable, small-scale science and pursue research that was difficult and dangerous,
but ultimately general and practical. He greatly assisted my development as a scientist by
continually confronting me with new ideas, and introducing me to exciting scientists. He
also ensured that his lab was an excellent place to work.
The collegial atmosphere and fellowship of the Arthur R. Marshall Jr. Ecological
Sciences Laboratory encouraged discussions that developed this research as well as
providing the friendship necessary to support my work, and for this I would like to thank
my fellow graduate students Craig Allen, Jan Sendzimir, Rusty Pritchard, Carla Restrepo,
Joe Fragoso, Paul Marples, and Mark Hostetler. Frequent discussions with Lance
Gunderson helped shape my dissertation and clarified my developing ecological ideas.
Toni Carter guided and protected me on my navigation through the ever shifting twists and
turns of the University of Florida's bureaucracy, as well as making the Arthur R. Marshall
Lab an excellent place to work.
My dissertation committee along with a number of other professors provided me
with a wide variety of assistance. Steve Sanderson introduced me to the human
dimensions of land-use/land-cover change, and greatly enriched my thinking on human
dimensions of resilience and disturbance. Colin Chapman encouraged an interest in the
scaling of seed-dispersal and provided clear, helpful critiques of many manuscripts. Jack
Putz encouraged me to learn more about tropical and sub-tropical ecology, and was always
willing to engage in ecological conversation. Lauren Chapman thoughtfully helped me
clarify many proposed portions of this document. Doria Gordon greatly enhanced my
understanding of north Florida ecosystems. I enjoyed working with them all.
The work on longleaf pine dynamics, described in Chapter 4, was conducted with
the collaboration and support from the Florida Nature Conservancy in partnership with
Eglin Air Force Base. Many people within both institutions helped build and test this
model. I would particularly like to thank Jeff Hardesty and Doria Gordon for their help in
developing this model, and the fire managers of Jackson Guard at Eglin Air Force Base for
testing and refining it. The work on fire in the boreal forest, in Chapter 5, was greatly
improved by discussions and data sharing with Paul Marples. Many discussion and grant
proposals written with Craig Allen produced the ideas that lead to Chapter 7.
Finally, I would like to thank my friends in Gainesville, Lance Gunderson, Jan
Sendzimir, Rusty and Joanna Pritchard, Kathleen 'Netjes' Ragsdale, Jessica Anders, Hank
Green, Joe Fragoso, Kirsten Silvius, and Ravic Nijbroek, whose friendship and support
helped me research and then write this dissertation. Most of all I thank Marieke Heemskerk
for her love and encouragement.
Finally, I would like to acknowledge that this work was supported by a Post-
Graduate Fellowship from the National Science and Engineering Research Council of
Canada and an Earth System Science Fellowship from NASA.
TABLE OF CONTENTS
A C K N O W L E D G M E N T S .................................................................... ii
ABSTRACT ............................................................................... viii
IN TROD U CTION ................................................................. ...... 1
Ecological Reorganization .................................................... ...... 1
Resilience and Disturbance ................................................... ...... 1
E cological Scale .............................................................. ....... 2
A D issertatio n O u tlin e ............................................................ ....... 3
GLOBAL CHANGE AND ECOLOGICAL RESILENCE .............................. 5
G lobal C change ................................................................. ...... 5
E arth Sy stem s ............................................................ ....... 7
E arth and the Sun ............................................ ............ ....... 7
Earth's Internal Dynamics ..................................................... 8
Origin of Earth and the Biosphere............................................. 9
Early Life the Archean .............................................. ......... 9
Transition to an Oxygen Atmosphere......................................... 10
Land Plants ....................................... .................. ............... 11
Expansion of Carbon Cycling ............................................... 12
D evelopm ent of the Biosphere................. .................... ......... 12
Multiple Climate States ......................................................... 13
Expansion of Humanity..................................... 14
Malthus & Darwin: Limits & Adaptation..................................... 15
H um an R egim es ............................................ ................. ........ 16
H um an D om ination......................... ........................................ 19
Land-use/Land Cover Change.............................................. 19
A tm ospheric C change .................................................. ......... 20
E cological A ppropriation............................................. ......... 21
Modification of Biogeochemical Cycles...................................... 21
Loss of Biological Diversity ...................................... ............... 22
Comparing Human and Historic Global Change................................. 22
Sum m ary ...................................................................... ........ 2 5
SCALING ECOLOGICAL DYNAMICS: SELF-ORGANIZATION,
HIERARCHICAL STRUCTURE & RESILIENCE ................................ 32
In tro d u c tio n ...................................................................... ......... 3 2
Scaling Problems .......................................................... ........ 32
Non-Linearity and Heterogeneity.............................................. 33
Different Processes Dominate at Different Scales ............................. 34
Cross-scale Connections ....................................................... 34
Em ergent Processes ............................................................. 35
Coping with Scaling Problems ............................... .............. 37
Cross-Scale Interactions ............................................................. 37
Hierarchy Theory ...................................................... ......... 38
H ierarchical O organization ............................................... .... 38
Dynamic Hierarchy ............................................................. 41
C ross-Scale C change ................................................... ......... 42
Cross-Scale Resilience ............................................................... 44
Avian Predation and Spruce Budworm....................................... 45
Scaling and Ecological Reorganization ....................................... 48
C o n c lu sio n s ...................................................................... ........ 4 8
FIRE, SEED DISPERAL AND MULTIPLE STABLE STATES IN
LONGLEAF PINE FOREST .............................................. 56
Introduction ............................................................................... 56
Ecological Management.............................. ... ............. 57
Longleaf Pine Forest on Eglin Air Force Base................................... 58
Ecological Caricatures and Management Alternatives ............................. 60
Caricatures of Ecological D ynam ics ............................................... 61
Modeling Approach................................................... ............ 62
Succession and Fire Temporal Dynamics....................................... 64
Modeling Vegetation Dynamics................................................ 64
Likelihood of States...................................................... ......... 73
Succession and Fire Spatial Dynamics.......................................... 75
M odel Organization .............................................................. 76
M odel Spatial Processes ......................................................... 79
Landscape Resilience ........................................................... 82
M odel T testing ............................................ ..... ..................... 84
Spatial Dynamics Management Options......................................... 87
Management Vegetation Classification........................................ 88
M managing Prescribed B urns....................... .................. ......... 88
Burning Strategy ................................................................ 89
Management Results ........................................................... 92
D is c u s s io n ............................................................................ 9 5
FIRE IN BOREAL FOREST: EMERGENT LANDSCAPE DYNAMICS ......... 138
In tro du action ................................................................. . . ..... 13 8
Fire in the Boreal Forest ........................................... ...... .......... 140
Topography and Dominant Tree Species..................................... 141
C lim ate .............................................................................. 14 1
Fire in the Forest of Southeastern Manitoba .................................. 142
A approach .............................................................. .... ............. 142
Modeling Approach ................................ ................. ........... 143
M odel Organization ................................................... ......... 144
Fire & F orest M odels ........................................ ............ .......... 145
N ull M odel ............................................................. ......... 146
In teractio n M o d el ......................................................... ........... 14 6
Fuel M odel ............................................................. ......... 146
Alternative Models ...................................... ......................... 147
M odel B behavior ............................................................ ......... 147
S elf-O rg an iz atio n ......................................................... ........... 14 7
F ire P attern s ............................................................ ... ......... 14 9
Summary of Model Behavior.................................................. 150
E m pirical C om prison ....................................... ............ .......... 151
M ethods .......................................................................... .... 151
R e su lts ....................................... .................. . . ............ 1 5 1
Summary of Comparison between Models and Manitoban Data ............ 153
Climate Change and The Self-Organizing Forest ................................ 154
C constant Clim ate Sim ulations....................................... ........... 155
Climate Change Experiments ................................ ........... ..... 155
Clim ate M odel .......................................................... ......... 155
E xperim ents ............................................................ ......... 156
R e su lts ................................................................................ 1 5 6
Summary of Climate Change Modeling Experiments........................... 158
C o n c lu sio n s .............................................................................. 1 5 9
A GENERAL MODEL OF CONTAGIOUS DISTURBANCE ........................ 177
Contagious Disturbance........................ .. ............. 178
A General Model of Contagious Disturbance......................................... 179
Model States ........................................................... .......... 179
M odel Processes........................ ......................................... 180
Types of Contagious Disturbance.................................................. 182
Disturbance Agents ............................................ ................. 183
Stress Accumulation and Release.............................................. 183
Chain Reaction .................................................................... 185
Evolutionary Epidem ic ........................................ ....... ............ 186
Comparisons of Contagious Disturbances........................................ 189
Disturbance Spread................ .... ................. 190
Disturbance Behavior: Constant vs. Adaptive ............................... 190
Disturbance and Ecological Dynamics........................................ 192
Disturbance Patterns..................................................................... 193
Stress Accumulation and Release.............................................. 193
Chain Reaction ................................................... .... ............... 195
Response to Environmental Change .......................................... 196
Summary .................................................................... .......... 197
ECOLOGICAL RESILIENCE, BIODIVERSITY AND SCALE ..................... 209
In tro d u c tio n ............................................................................... 2 0 9
Models of Ecological Organization .............................................. 210
Sp ecies-D iv ersity ..................................................... .......... 2 11
Idiosyncratic ........................................................... .......... 211
R iv ets ............................................................ .. ....... . . ...... 2 12
Drivers and passengers ............................. ................ ........... 213
Model Synthesis ...................................................... .......... 213
Resilience .................................................................. ........... 216
Multiple Stable States ................................... ............................. 218
Irreversible State Transitions............................................... 221
Slow Dynamics ....................................................... .......... 222
Scale ........ ..... ............................... ... ......... 223
Species, Scale and Ecological Function............................... ............ 225
Avian Predation of Insect Defoliators............................. ............ 226
Mammalian Seed Dispersal in an African Tropical Forest.................... 228
Potential Tests of Cross-scale Resilience ............................... .......... 229
C conclusions ................................................................. ......... 2 30
SUM M ARY .................................................................... .......... 244
Translating Across Scales ............................. .................. .......... 244
Floridan Fire M anagem ent ........ ........................... .......... ... 245
Boreal Forest Fires and Forest Resilience .............................. ........... 245
Contagious Disturbance ................................. ................ ........... 246
C ross-scale R esilience ................................................... .......... 247
BIBLIOGRAPHY ............................................................. .......... 248
BIO GRAPH ICAL SKETCH ...................................................... 261
Abstract of Dissertation Presented to the Graduate School
of the University of Florida in Partial Fulfillment of the
Requirements for the Degree of Doctor of Philosophy
CONTAGIOUS DISTURBANCE AND ECOLOGICAL RESILIENCE
Chairman: C.S. Holling
Major Department: Zoology
Managing the consequences of anthropogenic transformation of the biosphere
requires understanding the processes that create, maintain, destroy, and restore ecological
organization. In this dissertation I examine the ability of contagious disturbance processes,
such as fire and insect outbreaks, to create and destroy ecological resilience.
Using computer models I examine the consequences of changes in fire regime on
the organization of two specific ecosystems. I demonstrate that in north Florida forests
both wildfire and the current management practice of periodic controlled burning are
insufficient to maintain the landscape, but a policy that applies controlled burns based upon
the vegetative condition of an area could be effective.
I compare the responses of climate and fuel-accumulation driven models of fire in
the boreal forest. I demonstrate that climate models are more sensitive to climate change
than fuel accumulation models, but are more vulnerable to abrupt reorganization. I use
these specific explorations to construct two general conceptual models: the first of
contagious disturbance, and the second of ecological resilience.
My model of contagious disturbance defines disturbance agents, stress
accumulation and release, chain reactions and evolutionary epidemics as four key sub-types
of contagious disturbance. I identified initiation, reproduction, contagion, virulence, and
recovery as the five key properties of a contagious disturbance regime, and show how a
change in the relationship of these variables alters the behavior of contagious disturbances.
These models provide general insights into the consequences of specific ecological
organizations, allowing qualitative predictions to be made about the behavior of a poorly
understood disturbance process.
My model of cross-scale resilience proposes that ecological resilience is generated
by diverse, but overlapping, functions within a scale and by apparently redundant species
that operate at different scales. The distribution of functional diversity within and across
scales allows regeneration and renewal to occur following ecological disruption over a wide
range of scales. The consequences of species or ecological process loss may not be
immediately visible, but it decreases ecological resilience to disturbance or disruption. It
produces ecosystems that are more vulnerable to ecological collapse, and reduces the
variety of possible alternative ecological organizations.
The Earth is currently undergoing a biospheric reorganization, as the expansion of
human constructed or dominated ecosystems at the expense of natural ecosystems is
altering climate, mineral cycles, land cover, and biotic communities (Vitousek et al. 1997).
Human ecological expansion is mainly occurring at the meso-scale, between the global and
the local. The meso-scale is the scale at which people live and transform ecosystems
through activities such as agriculture, human settlement, and forestry (Harris 1984). It is
these activities that drive much of global change and it is through changes to these activities
that people will experience much of the impacts global change.
Anticipating and coping with the surprises that these anthropogenic transformations
will bring requires an understanding of how the interaction of ecological dynamics across
scales shapes ecological organization (Kates and Clark 1996). It requires understanding
the processes that create, maintain, and restore resilient ecosystems, as well as
understanding the processes that disrupt and destroy them. In this dissertation, I examine
the cross-scale relationship between ecological resilience and disturbance.
Resilience and Disturbance
Ecological resilience is the ability of a system to persist despite disruption (Holling
1973). Ecological resilience assumes that an ecosystem can exist in alternative self-
organized or 'stable' states. It is a measure of the amount of change or disruption that is
required to transform a ecosystem from being maintained by one set of mutually reinforcing
processes and structures to a different set of processes and structures.
A disturbance is an event that disrupts the functioning of an ecosystem. A
disturbance process produces disturbances at a frequency slower than the time it takes the
disturbed system to recover. I differentiate between contagious and non-contagious
disturbances. Non-contagious disturbances have a fixed temporal frequency and spatial
impact that is externally imposed upon the disturbed system. Contagious disturbances are
disturbances that propagate through a system, such as fire or a flood. Consequently, their
size and behavior emerges out of the interaction of the disturbance process and the system
being disturbed. In this dissertation I disregard non-contagious disturbances and focus
upon contagious disturbances, because contagious disturbance processes connect
ecosystems across scales.
Disturbance and resilience are intimately intertwined with one another. Disturbance
disrupts ecosystems. This disruption can overwhelm an ecosystem's resilience causing it
to reorganize. However, resilience is often maintained by the diversity and heterogeneity
produced by disturbance. Disturbance both produces and destroys resilience. The key to
this apparent paradox is the scaling relationship of disturbance and resilience.
I define a scale as a range of spatial and temporal frequencies. This range of
frequencies is defined by resolution below which faster and smaller frequencies are noise,
and the extent above which slower and larger frequencies are background. Different
ecological processes occur at different scales. For example, forest fires occur at different
scales in space and time from a deer mouse's foraging behavior. Processes that operate at
the same scale interact strongly with each other, but the organization and context of these
interactions are determined by the cross-scale organization of an ecosystem. It is this type
of cross-scale interaction that typically occurs between disturbance and an ecosystem.
Small, frequent disturbance often helps an ecosystem, survive, and recover from large,
severe disturbance. An example of this dynamic between disturbance and resilience
recently occurred in the South Florida. During Hurricane Andrew in 1992, mature
mangrove trees were killed by wind damage, however many young mangroves survived.
Many of these young mangroves were located in gaps caused by lightning. Local lighting
disturbances provided the mangrove population with increased resilience to the large-scale
disturbance produced by Hurricane Andrew (Smith et al. 1994). It is these cross-scale
interactions between resilience and disturbance that I explore in this dissertation.
A Dissertation Outline
This dissertation investigates the dynamics of resilience and disturbance from a
combination of theoretical and empirical perspectives. I begin the dissertation by providing
a motivation for this investigation of disturbance and resilience.
In the second chapter, I discuss the ecological dynamics of global change. I
distinguish between natural and anthropogenic global change, provide a brief history of
natural and anthropogenic global change, and place current anthropogenic global change in
a historical context. This chapter illustrates how a variety of cross-scale processes have
provided the resilience that has allowed the biosphere to recover from past global changes,
and that humanity's present combination of global changes is reducing the adaptive capacity
of the biosphere. Along with the general concern for the health of the biosphere, this also
identifies the importance of understanding the resilience of ecosystems and the dynamics of
In the third chapter, I discuss problems with translating data and understanding
across scales. Assessing the impact of change on ecosystems requires the use of scaling
methods, but the resilience of ecological organization to changes in key processes
determines the situations in which scaling methods apply, require adjustment, or break
down. I propose that ecological resilience can be used to develop scaling methods that
incorporate ecological reorganization.
In chapter four, I move from the theoretical considerations of chapter three to
detailed cross-scale investigation of forest dynamics in northwest Florida. Resilience and
scaling are used to analyze the consequences of contagious fire and seed dispersal across a
meso-scale landscape. These concepts are used to compare alternative models of forest
dynamics and assess the consequences of alternative land management strategies.
Chapter five, addresses another investigation of forest dynamics, this time in the
boreal forest. The organization of the landscape under several alternative models is tested
against empirical data, and the sensitivity of the alternative models to climate change is
Chapter six compresses and generalizes the disturbance models developed in
chapters four and five, to produce a general model of contagious disturbance. I identify the
key features of contagious disturbance, the variables that define those features, and outline
how different relationships among these variables produce different disturbance dynamics.
I use these attributes to define four sub-models of my general model, which represent a
variety of abstract types of contagious disturbance. These models provide general insights
into the consequences of specific ecological organizations, allowing qualitative predictions
to be made about the behavior of a poorly understood disturbance process by fitting it into
the classification of rates and their relationship.
Chapter seven returns to investigate resilience and its linkages to disturbance.
In this chapter, I present a cross-scale model of resilience that integrates disturbance and
ecological processes. The model proposes that ecological resilience is generated by
diverse, but overlapping, function within a scale and by apparently redundant processes
that operate at different scales, thereby reinforcing function across scales. The distribution
of functional diversity within and across scales allows regeneration and renewal to occur
following ecological disruption over a wide range of scales.
GLOBAL CHANGE AND ECOLOGICAL RESILENCE
This chapter illustrates how a variety of cross-scale processes have provided the
resilience that has allowed the biosphere to recover from past global changes, and that the
diverse nature of present day anthropogenic global changes is reducing the adaptive
capacity of the biosphere. This decline in the general adaptive capacity of the biosphere
will likely manifest itself far more seriously in specific regions. This loss of adaptive
capacity supports my argument that understanding the dynamics of ecological resilience and
disturbance should lie at the center of applied ecology.
In this chapter I discuss the history and dynamics of natural and anthropogenic
global change, before comparing modern anthropogenic global change to past global
changes, and presenting a simple model of how global change is eroding adaptive capacity.
Global change is different from ecological change at smaller scales. The dynamics
of a local or regional ecosystem may be strongly influenced by fluxes and migrations from
neighboring systems, but there is little interaction of the global biosphere with its external
environment (Barlow and Volk 1992). The globe receives energy from the sun, and emits
heat to space. It receives heat and minerals from beneath the crust, and returns reorganized
materials through the ocean floor. However, these fluxes are tiny compared to the fluxes
of material and energy in Earth's biogeochemical cycles (Schlesinger 1991).
The closed nature of the Earth means an accumulation or decline of material at a
global scale will not be counteracted by external fluxes. This is not true at a local scale.
For example, as a plant removes CO2 from the air near the plant that CO2 is rapidly replaced
by CO2 from the surrounding air. At the global scale, CO2 emissions and fixation have to
match, or the stocks of CO2 begin to change.
At the global scale, change that occurs is largely due to endogenous dynamics.
Evidence suggests that an asteroid may have been precipitated the Cretaceous mass-
extinctions, but the other mass-extinctions appear not to have had an external trigger
(National Research Council 1995). Similarly, recent glacial cycles are probably driven by
changes in the Earth's orbit around the sun. However, the magnitude of the global change
caused by relatively minor changes in solar input is due to the amplifying power of Earth's
endogenous dynamics (Bartlein and Prentice 1989). These internal dynamics are
dominated by the biosphere.
Over the history of the Earth, the expansion and diversification of the biosphere
have caused massive changes in Earth's physical, chemical, and biological organization
altering the Earth's response to external variation (Schlesinger 1991). Today, the
expansion of the Earth's human population and the enormous amplification in the power of
human technology that has accompanied it is driving a transformation of global flows of
energy, materials and biota that overshadows natural global change (Vitousek et al. 1997).
This anthropogenic change is combined and intertwined with the natural dynamics of the
Earth system. In one sense this combination is trivial, in that for those impacted by global
change the cause of that change does not matter. But, understanding how the Earth works,
and designing effective policies to mitigate and adapt to global change requires
distinguishing the natural and anthropogenic drivers of global change.
Understanding consequences of the human transformation of the Earth system
requires understanding the natural dynamics of the Earth system itself. The study of the
dynamics of the Earth system is a prerequisite for disentangling the natural and
anthropogenic components of global change. In the following sections I outline these two
interacting components of global change. I begin with a discussion of the evolutionary
dynamics of the Earth system, before describing the expansion of humanity within the
The evolution of Earth's biosphere has driven periods of revolutionary global
change. The study of history provides a rich set of past world organizations that represent
alternative configurations of the Earth system. The exploration of how the Earth system
responded to past global change helps provide clues to how the Earth will respond to
present day global change. In this section I outline Earth's past global changes to place
present day ecosystems and ecological change in perspective. I discuss the Earth's
relationship with the sun, its internal geophysical dynamics, and the evolution of the
Earth and the Sun
The sun has changed over the history of the Earth. It has gradually become
brighter, about 25% brighter over the past 4 billion years, and this change has increased the
amount of solar radiation reaching the Earth (Lovelock 1988). Over shorter periods, Earth
experiences significant variations in the distribution of solar energy upon its surface and
timing of that energy. Along with this gradual increase in luminosity, the Earth's orbit
around the sun introduces cyclical variation into the amount of sunlight the Earth receives.
Variation in Earth's orbit and a wobble in the tilt of its axis vary the sunlight the
Earth receives (Bartlein and Prentice 1989). There are three cycles occurring at different
time scales. The slowest change is in the eccentricity of the Earth's orbit. Over 100,000
years its varies between being near circular and being more elliptical. This change alters the
difference between Earth's seasons. Over a period of 41,000 years the tilt of the Earth's
axis varies. The degree of tilt alters the Earth's seasonality, and therefore temperature
differences between the northern and southern hemispheres. The fast cycle, with a period
of 23,000 years, corresponds to change in the seasonality in Earth's orbit of Earth's closest
approach to the sun. This change either mitigates or enhances seasonality and differences
between the Northern and Southern Hemisphere. It appears that these orbital variations
interact in some way with one another and Earth's biosphere, but the interaction is not
clear. Variations in polar ice volume are synchronized with these cycles. Until 800,000
years ago, ice ages corresponded strongly with the frequency of axis wobbles, but after
that time, ice ages correspond better with the variation in the shape of Earth's orbit
(National Research Council 1995). The reasons for this shift in the process driving ice
ages are unclear, but either changes in the Earth's biosphere or global oceanic or
atmospheric circulation patterns could alter the relative importance of different forms of
Earth's Internal Dynamics
The long term cooling of the Earth drives continental drift (Westbroek 1992). The
difference between the hot molten core of the Earth and its surface produces convective
currents within the Earth. These currents gradually move the pieces of Earth's crust apart
and against one another, generating continental drift. These forces continually rework
Earth's surface, splitting continents apart and crashing continents together (Schlesinger
1991). Continental drift has had a major impact on the history of life, bringing together
biota that are novel to one another, as well as encouraging speciation by dividing
populations and ecologies (Briggs 1995). Also continental drift has driven climate change,
by changing the configuration of ocean basins and the location of mountain ranges has
altered global oceanic and atmospheric circulation patterns (Ruddiman 1990). Over long
periods, continental drift has altered atmospheric composition by producing gas-emitting
volcanoes. This effect was strongest early in the history of the Earth.
The development of the Earth's biosphere has diversified Earth's biogeochemical
cycles. Mutual reinforcement among these processes tends to self-organize sets of
processes that are relatively stable, because interactions among processes that are unstable
tend to extinguish themselves (Lovelock 1988). However, the biosphere has periodically
been reorganized as new biogeochemical pathways have been created by interaction of
biotic evolution with changes in Earth's geochemical environment. These revolutions have
produced fundamentally different biospheres.
One of the central aims of global change research is to assess to what degree human
alteration of the Earth is eroding the ability of the current Earth system to maintain itself
despite disruption. By examining past reorganizations of the biosphere, current changes
and their future consequences can be placed in context. In the following sections, I
summarize the history of different Earth system regimes, and the transitions between them,
focusing on the development of the biosphere and evolution of novel biogeochemical
Origin of Earth and the Biosphere
The Earth was formed about 4.5 billion years ago, as planets condensed into orbit
around the sun. During it initial development it was bombarded with cosmic debris, which
increased its size, and heated the planet to such an extent that it was liquefied, and denser
substances migrated towards the core of the Earth. Over time, collisions reduced the
amount of the solar debris and the rate at which the Earth was pummeled by meteorites. At
this time, about 3.8 billion years ago, the first traces of life are found (National Research
The composition of the Earth and its position in the solar system define internal and
external bounds on Earth's biota. The Earth receives the bulk of its energy from the sun.
The Earth itself provides heat from radioactive decay, but its main contribution to life is the
materials that comprise the Earth, and the chemical energy that they contain. Between the
rock of Earth and the space of the solar system the biosphere developed. In the following
sections I sketch a brief history of the biosphere, and then discuss how variations in Earth
and Solar dynamics have altered the dynamics of the biosphere.
Early Life the Archean
The biosphere covers the surface of the Earth, and extends at least several
kilometers in the Earth's crust and up into the Earth's atmosphere is supported by energy
from the sun, mineral energy from the Earth (Pedersen 1993). The earliest life was almost
certainly dependent upon chemical energy, but photosynthetic bacteria soon evolved
directly linking life to incoming solar energy (Volk 1998).
While the early history of life on Earth is difficult to reconstruct, one of the earliest
consequences of life was a reduction in atmospheric CO2, as bacteria sequestered carbon.
This decrease in CO2 liberated oxygen. Rather than increasing atmospheric oxygen levels,
this oxygen reacted with oxidizable material in the Earth's oceans, atmosphere, and crust.
The early Earth was probably a reducing environment, and certainly there were many
oxidizable materials on the early Earth. The oxidization of iron dissolved in seawater,
produced iron deposits that provide the earliest evidence for photosynthesis (Schlesinger
Transition to an Oxygen Atmosphere
Between about 2.2 and 1.9 billion years ago (bya) the oxygen increased from being
just a trace gas to composing as much as 3% of the atmosphere (Volk 1998). The cause of
this rise in oxygen is unclear. It could have been due to increased rock weathering burning
more carbon, a decrease in emissions of reducing gases and CO2 from volcanoes as the
Earth cooled, or an increase in the biomass of aerobic bacteria.
It is suggestive that this period corresponds with an expansion of the biosphere.
During this time eukaryotes evolved from prokaryotes. Eukaryotes are more efficient
photosnythesizers than prokaryotes. Whether an increase in oxygen facilitated the
evolution of eukaryotes, or eukaryotes facilitated the rise in oxygen levels in unclear.
However, the proliferation of eukaryotes established a positive feedback, as aerobic
bacteria encouraged the growth of more aerobic bacteria.
A combination of these processes may have produced the rise in oxygen, because
processes such as rock weathering and biotic oxygen production are not mutually
exclusive, rather they are mutually reinforcing. As an increase in bacteria populations,
particularly if they colonized land and rock fractures, would have accelerated rock
weathering (Schwartzman and Volk 1989).
There was another pulse in atmospheric oxygen between 1 0.6 bya. This oxygen
pulse also occurred synchronously with a diversification of life. In this case, the rise of
oxygen accompanied the development of metazoan life (Canfield and Teske 1996).
Another suggestive feature common to both these revolutionary reorganizations of the
biosphere is that both rises of oxygen were accompanied by declines in CO2 and severe
glaciations (Hoffman et al. 1998).
The paleoecological record suggests that as the biosphere became more diverse, the
gradual accumulation of oxygen triggered global ecological crises. However, these crises
were likely overcome by the negative feedback effect of biospheric cycles. For example, if
an increase in oxygen lowered global atmospheric CO2, reducing the warming produced by
the greenhouse effect, this cooling would reduce the rate of plant growth, allowing CO2 to
increase again established a climate more favorable to life (Lenton 1998).
Following the origin of metazoan life, the next major reorganization of the Earth
system was the colonization of land by plants. Plants accelerated weathering by a variety
of means. Plants increased the surface area of rock weathered through their root networks.
By breaking apart rocks they facilitated the creation of organic soil, which increased the
ability of water to weather minerals in the soil. These increases in weathering probably
removed atmospheric CO2 (Berner 1997).
Furthermore, this period coincides with the evolution of lignin by plants. It appears
that plants used lignin for several million years before fungi developed the means to break
apart lignin (Robinson 1990). During this lag period, large amounts of lignin were buried,
removing large amounts of carbon from the atmosphere until fungi evolved the ability to
decompose lignin. This process may have allowed oxygen levels to climb to their highest
levels ever, perhaps even up to 30% of the atmosphere (Lovelock 1988). Beyond this
oxygen concentration, organic matter becomes extremely combustible, and fire would act to
decrease oxygen concentrations.
Expansion of Carbon Cycling
Another key ecological transformation was the evolution of calcium carbonate
precipitating plankton. These plankton fundamentally altered Earth's carbon cycle, and
perhaps even plate tectonics when they evolved 200 million years ago (mya). By
producing calcium carbonate (limestone) from CO2 and calcium they removed CO2 from the
atmosphere. By moving carbon deposition into the deep sea these plankton, surprisingly,
made CO2 more available to the atmosphere. Large volumes of limestone precipitated by
plankton were pushed underneath continental plates. Some of the carbon in this limestone
was returned to the surface through volcanoes, recycling carbon through a slow cycle
driven by continental drift. Limestone deposition may even accelerate continental drift, and
consequently carbon cycling (Lovelock 1988, Westbroek 1992).
Development of the Biosphere
The functional elaboration and diversification of life on Earth over geological time
has increased the number of biogeochemical pathways in the biosphere. These pathways
often perform similar functions, but do so at different scales. For example, carbon is
cycled between the ocean and the atmosphere over minutes, while carbon moves from the
seabed to the atmosphere over hundreds of millions of years.
A general pattern in the development of the biosphere is that the gradual
accumulation of low entropy, high-embodied energy waste products, such as oxygen or
lignin, accumulated to the extent that they provided a utilizable resource. In turn the
evolution of new functional groups to take advantage of this concentrated energy lead to the
accumulation of new waste products, containing ever more embodied energy. This ratchet
of complexity has allowed the elaboration and articulation of Earth's biosphere.
The existence of a diverse set of ecological functions, such as the various steps of
biogeochemical cycles, that are replicated at many temporal scales, such as the many
separate pathways in the carbon cycle, suggests that the development of the biosphere may
have produced an increasing ability to respond to variation or disturbance across wide
variety of scales. Whether this is indeed the case, or not, is unclear. Life has been able to
maintain itself and the Earth's temperature, in the face of a 25% change in the sun's output
(Lenton 1998, Lovelock 1988). The response of the present globe to more subtle changes
in the in Earth's distance from and orbit around the sun, and continental drift is more
Multiple Climate States
Biospheric reorganizations appear to have interacted with the Earth's geophysical
dynamics and solar forcing to produce alternative stable states in the Earth system. At the
scale of hundreds of millions of years the Earth appears to have alternated between periods
in which glaciation was possible, what have been termed 'icehouse' worlds, and those in
which it is not, 'greenhouse' worlds.
The most recent transition between these organizations of the world was 34 mya.
At that time continental drift caused a change in the world ocean circulation triggering a
planetary cooling, shifting the Earth from a 'greenhouse' to an 'icehouse' state. This
occurred when continental drift caused Australia to separate from Antarctica, allowing
water to circulate uninterrupted around Antarctica at the South Pole. Prior to this event, the
Earth's oceans were broken into a series of basins, in which cooler and warmer water
mixed. The circulation of cool water around the south pole, cut off polar water from
mixing with warmer temperate water, and established a steep temperature gradient between
pole and equator which lead to the deep water circulation of cold Antarctic water across the
ocean floor (Prothero 1994).
The cooling of Antarctica may have established a positive feedback cycle, because
as Antarctica cooled it began to accumulate ice, which in turn increased the albedo of
Antarctica, causing it to absorb less heat from the sun, cooling the continent further,
allowing glaciers to develop and spread (Briggs 1995). Another possible cause of global
cooling was the uplift of the Tibetan plateau, which probably lead to increased weathering,
removing CO2 from the atmosphere (Ruddiman 1990). Regardless of the exact cause or
combination of causes, these changes moved the world from a 'greenhouse' to an
During this current icehouse regime, the world has been repeatedly glaciated. It has
alternated between, short warmer inter-glacial periods and longer, cooler glacial periods.
During the course of this recent 'icehouse' regime, oceans have shrunk, novel plant
metabolic pathways formed, and new animal communities emerged (Cerling et al. 1997).
It was in this low CO2, cool, and dry world that humanity's ancestors evolved, and
humanity came into being.
Expansion of Humanity
One hundred thousand years ago the ancestors of modern humanity were a
population of several hundred thousand large hunter-gathers living upon the African
savanna. Today, a population of almost 6 billion people live in all over the world, in
highly organized aggregations, in ecosystems that have been largely shaped by human
action (Cohen 1995). During this time, an ice age ended, global sea levels have changed -
altering the configuration of continents, a wide variety of large mammals, birds and reptiles
became extinct, and entire ecosystems have reorganized.
The greater part of humanity's expansion occurred during the past few centuries
(Figure 2-1), whether measured in absolute population size, population growth rate, or
energy and material use per capital. In the two hundred years from 1600 to 1800 the
world's population increased about 1.7 times, while in the two centuries from 1800 to
1995 the world's population increased 6.3 times (Figure 2-2). Similarly, the amount of
non-animal or plant energy used per person/year increased about 28 times from 1800 to
1900, and another four times from 1900 to 1987 (Cohen 1995).
Malthus & Darwin: Limits & Adaptation
All living things are subject to limits to their growth. Malthus observed that
populations often grow faster than their ability to acquire new resources, leading to
resource shortages (Malthus 1798). These resource limitations bound the increase of a
population. This Malthusian relationship suggests that populations should grow until they
reach their limits, and then remain or oscillate around that steady state. Darwin's theory of
natural selection extended this Malthusian dynamic (Darwin 1859). He noted that not all
individuals in a population are identical. If some individuals have traits that enable them to
obtain resources better than other individuals, then those individuals are more likely to
reproduce than others. Darwin's theory of natural selection, showed how a population
could adapt to a set of environmental limits and possibly find ways to escape from that set
Human societies being able to learn and reorganize their means of production are
more able than most species to adapt and innovate to escape from Malthusian pressures.
Indeed, the demographic expansion of humanity has progressed hand-in-hand with the
expansion of humanities ability to manipulate and construct ecosystems. Locally increased
population densities, put more pressure on local ecosystems. This pressure causes
ecological degradation and, in some cases, inspires innovative forms of ecological
engineering. While many societies become caught in a Malthusian trap some societies
managed to innovate to temporarily escape from Malthusian pressures, by reorganizing
their local ecologies to increasingly support humans rather than other non-human ecologies,
and these innovations usually spread to other groups (Goldstone 1991, Spicer 1996).
While these reorganizations have enabled phenomenal growth in humanity, by increasing
the amount of ecological production aver by people, they have removed resources and
support from non-human ecosystems.
Many innovations have expanded the resources available to people. To organize
my discussion I consider four expansions of anthropogenic ecological domination: the
domestication of fire, agriculture, long distance trade, and public health. These ecological
innovations were made possible by combinations of technological and social innovations
that did not occur simultaneously around the globe, but rather occurred unevenly as
innovations spread out from various centers. However, the spread of these ideas by
conquest and trade makes these transitions meaningful at the coarse global scale (Spicer
1996), and despite some regions of poverty and disorder, the major proportion of the
world's population has passed or is passing through these ecological transformations.
There is a suggestion of these transitions in the population data of Figure 2-1 and
Figure 2-2. When the same data is displayed on a log-log graph, the ever accelerating
nature of these transitions can be seen (Figure 2-3). On a log-log graph, exponential graph
produces a curve that initially climbs steeply, before flattening off. Therefore, any portion
of a log-log graph with an increasing or constant slope indicates an accelerating growth
rate. On Figure 2-3, there is an acceleration at about 10,000 years ago that coincides with
the introduction of agriculture. Similarly, the expansion of global trade 500 years ago, and
global public health measures of the past 50 years both boost growth until the fertility
declines of the past decade, which is more clearly shown in Figure 2-4. While the quality
of the data plotted in Figure 2-3 is low, the figure does suggest that humanity as a whole
has been quite successful at evading Malthusian limits. I discuss each of the expansions of
human dominated ecosystems in more detail below.
Fire has large ecological effects, strongly influencing vegetation growth and
succession, and consequently animal community dynamics (Bond and van Wilgen 1996).
Humans probably domesticated fire about half a million years ago (Goudsblom 1992).
Fire may not have been the first, but it was certainly one of the most powerful early
ecological engineering tools that humanity mastered. The effect of this event on human
population growth is hidden in the distant past, but it is reasonable to assume that it was
People have used fire for hunting, war, and agriculture all over the world. Indeed,
on of the most common paleoecological records of human settlement is an increase in
charcoal in lake sediments (Goudsblom 1992). Records from Australia (Kershaw 1988),
New Zealand (Flannery 1994), and North America (Cronon 1983) all show evidence of
increases in fire frequency with the arrival of people. The domestication of fire also aided
in food preparation, allowing people to expand the range of food items that they could
Agriculture probably was invented about 10,000 to 12,000 years ago, following the
last ice age (Diamond 1998). Agricultural domestication of plants and animals allowed
people to develop novel variations of natural ecosystems that channeled their production
into resources for people.
Agriculture is unremarkable in a fashion; after all ants have domesticated aphids and
fungi. However, people have relentlessly innovated expanding the range of plants and
animals domesticated, and physically restructuring places to channel energy and resources
away from non-human ecosystems towards human-dominated ecosystems.
Agricultural innovation was expanded through trade. The local diffusion and
spread of technologies and domesticated plants and animals has occurred for a long time, as
is evidenced by the spread of the banana from New Guinea to Africa (Diamond 1998).
However, ecological engineering received a large leap forward following the European
discovery of the Americas, which integrated the world's population into a single world
system (Wallerstein 1974). Between 1650 and 1850s, this system became tightly
integrated through trade, and migration. The transport of domesticated plants and animals
among disparate regions greatly increased agricultural yields (Cohen 1995). The impact of
the potato and maize upon Europe and the sweet potato and Manioc upon the tropics
provide two examples of the greatly increased agricultural yields that followed these
agricultural modifications (Crosby 1972, Diamond 1998).
Humanity's novel ecosystems produced novel problems. By requiring denser
human settlement agriculture changed conditions for human ecosystem parasites and
diseases (McNeill 1976). Global trade and transport connections also encouraged the
development of epidemic diseases. While novel human ecosystems provided support for
humanity, they also created new ecological niches for parasites, such as rats and
cockroaches, and pathogens, such as tuberculosis and smallpox (Crosby 1986). While
medicine improved during the 19th century, it was only in the 20th century that public
health programs managed to massively decrease mortality from epidemic diseases such as
smallpox, malaria, and tuberculosis. At the same time, sanitation measures re-engineered
urban ecosystems reducing mortality from infection. Following the Second World War
massive worldwide immunization and sanitation programs public began to have great
success (McNeill 1976). These programs can be regarded as ecological engineering. By
eliminating disease populations, by immunization, and reorganizing human ecosystems
through sanitation and other public health measures public health programs greatly reduced
or even eliminated many of humanities parasites. This reduced human mortality, allowing
population growth rates to increase to their highest levels (Figure 2-4).
Recently, a fifth transition appears to be occurring. In the past decade fertility rates
have declined (Figure 2-4). The phenomena is diverse and not well understood, however it
may represent an important advance in the ability of people to regulate their own
reproduction, or it may be a temporary side-effect of an increasing average age (Cohen
1995, Lutz et al. 1997).
Despite this recent decline in the fertility rates population growth is expected to
continue until the world's population is near 10 billion people, six times the world's
population at the beginning of the 20th century. This growth in absolute population size,
the human ability to innovate, and the desire for an improved material standard of living is
expected to increase the human domination of Earth above its already high levels.
Humans currently dominate the structure and functioning of Earth's ecosystems.
The human impact on the Earth has increased at an accelerating rate due to the increase in
human population and concomitant increases in per capital use and consumption of natural
resources, expanded technological capacity to access resources, and the expansion of
human settlement over the globe (Turner et al. 1993). These increases mean that humanity
has, and will probably continue, to expropriate an ever-increasing share of the Earth's
renewable and nonrenewable resources and increasingly rearrange and reengineer the
Earth's physical, chemical, and biological systems.
Anthropogenic global change is usual divided into a number of separate classes that
are usually considered in isolation, despite the well recognized links that exist among them.
Below I provide a quick outline of the dimensions of several key types of anthropogenic
global change, specifically land-use/land-cover change, atmospheric change, ecological
appropriation, biodiversity loss, and the modification of biogoechemical cycles. I follow
this discussion with my approach to integration.
Land-use/Land Cover Change
Humanity currently uses 2/3 of vegetated land and an estimated 40% of net primary
productivity (NPP) of this land (McNeely et al. 1995). The ability of humans to transform
landscape is ancient, perhaps beginning with the domestication of fire thousands of years
ago. The invention of agriculture, and irrigation amplified the role of humans as ecological
Land use change involves conversions from one type of land cover to another, as
well as changes in management practices. For example, the area of forest converted to
agriculture or grassland since pre-agricultural times is estimated to be about 8 million km2,
with 3/4 of this land cleared after 1680 (Turner et al. 1993). Similarly, about 7 million km2
of grassland have been converted to agriculture, and 0.5 million km2 has been submerged
beneath artificial lakes (Turner et al. 1993).
Land-use/land cover change directly alters the amount and distribution of various
ecosystems, and biota. It also alters ecological dynamics such as the dispersal of biota
(Harris 1984), the provision of ecological services to neighboring locations (Forman
1995), and the interactions of between vegetation and the atmosphere (Pielke et al. 1993).
Anthropogenic emissions are altering the composition of the atmosphere. Land use
change has resulted in net addition of large amounts of CO2 and methane, along with other
gases to the atmosphere (Walker and Steffen 1997). More importantly, the combustion of
large volumes of fossil fuel has also increased atmospheric concentrations of CO2 and
The atmosphere's current CO2 concentration is about 360 ppm, and increasing
(Walker and Steffen 1997). In the centuries before industrialization, the atmosphere
maintained a concentration near 280 ppm (Schlesinger 1991). Global average temperature
is closely correlated to atmospheric CO2 levels. The Intergovernmental Panel on Climate
Change 1995 assessment (Intergovernmental Panel on Climate Change 1995) concludes
that increased concentrations of CO2, and other greenhouse gases, have probably already
altered the world's climate, and that this alteration is likely to become more extreme in
Furthermore, the production and emission of CFCs has caused decreases in
stratospheric ozone, as well as being a strong greenhouse gas. These decreases in ozone
have increased the amount of ultraviolet radiation reaching the Earth's surface, particularly
near the North and South poles.
Humanity uses a significant proportion of the Earth's total ecological production.
Recent estimates put humanity's use of terrestrial ecological production at between 39 %
and 50 % and our use of oceanic production at about 8% (Vitousek et al. 1997). This
ecological production is unavailable to non-human ecosystems, and is suggestive of both
the large stress that humanity is placing on the global biosphere, and of the limits to further
growth in the amount of ecological production that humanity appropriates.
Modification of Biogeochemical cycles
Human action has for most of history altered biogeochemical cycles through the
modification of the environment, for example by burning a grassland. However, the ability
of modem industrial societies to mine and synthesize minerals and chemicals has caused
humans to directly alter global chemical cycles.
The synthesis of phosphate and nitrogen fertilizer has doubled the global cycle of
these key ecological nutrients (Vitousek 1994), driving a global eutrophication of terrestrial
and aquatic ecosystems.
Mining and fossil fuel use has greatly magnified the amount of poisonous heavy
metals in the globe's biogeochemical cycles. For example, the amount of lead has
increased twenty five times, cadmium seven times and mercury ten times (Turner et al.
The success of synthetic chemistry has unleashed a massive wave of novel
synthetic substances into the global environment that have produced novel biogeochemical
processes. These introductions have often had, unexpected negative impacts on people and
animals. The most recent danger is the potential consequences of ubiquitous endocrine
disrupting chemicals (Colbom et al. 1996).
Loss of Biological Diversity
Humanity's spread across the world over the past tens of thousand years appears to
have initiated a mass extinction episode. The spread of people to the New World (Martin
1984), Australia, and New Zealand (Flannery 1994), the Pacific archipelago (Steadman
1995), and other islands (Crosby 1986, Diamond 1992) all instigated regional mass
extinctions. Estimates of current extinction rates suggest that the current rate of species
loss is 100 to 1000 times greater than the average pre-human rate of species loss (Lawton
and May 1995).
Currently several different forces appear to be driving biodiversity loss: habitat
loss, harvesting, and biological invasions. Habitat loss is due to land-use/land cover
change, which was mentioned above, as well as ecological appropriation of the resources
that animals and plants require to survive and reproduce. Secondly, harvesting and hunting
of animals is due to ecological appropriation of the animal and plant populations
themselves. Thirdly, biota has been purposely and accidentally transported within human
transport networks. Transported biota can compete with, prey upon, or modify the
physical environment of native biota, frequently producing large ecological impacts (Lodge
1993). Furthermore, biodiversity loss may occur due to previous extinctions. A species
creates habitat or provides food for other species, its extinction can lead to the extinction of
its dependent species.
Comparing Human and Historic Global Change
The magnitude of modem anthropogenic change appears to be roughly comparable
to past revolutionary global changes. The biosphere has been able to adapt to these
changes. For example, about 6 to7 million years ago, the continuing decline in
atmospheric CO2 lead to the evolution of C4 metabolism plants, which are able to survive
warm temperatures and moisture stress better than C3 plants at current atmospheric CO2
concentrations. However, these adaptations have all occurred on an Earth in which the
biosphere has been unconstrained in its response to these changes.
Present day anthropogenic global change is disrupting a disrupted biosphere. The
current biosphere is not as free to adapt as in the past. Humanity demands a variety of
ecological services, such as food production, from human ecosystems thereby constraining
the degree to which these ecosystems can be allowed to adapt to global change. Humanity
and our constructed ecosystems have greatly reduced the extent of non-human dominated
ecosystems. This reduction may have been as much as 50% (Vitousek et al. 1997), but it
appears to be at least 10% (Gorshkov 1995). This elimination of much of the Earth's
adaptive capacity will likely reduced the ability of the biosphere to respond to
anthropogenic global change.
Past innovations in biogeochemical cycling appear to have often resulted in crises in
global biogeochemical regulation. As discussed above, the development of an oxidizing
atmosphere and the evolution of land plants, lead to periods in which global regulation of
climate broke down, leading to global glaciations.
Present day anthropogenic global change is reorganizing global cycles to a degree
comparable to past ecological revolutions. Human alteration and acceleration of
biogeochemical cycles, the burning of fossil fuel, and the synthesis of novel chemicals has
produced a wide variety of new biogeochemical pathways, many of which are significant
proportions of global fluxes (e.g. N, P, Hg) or are new materials with significant effects
(e.g. CFCs). The synchrony and extent of these changes increases the possibility that
another collapse of Earth's homeostatic processes may occur. The biosphere has managed
to survive past collapses at the cost of mass extinctions and ecological reorganization.
Probably, humanity could also survive an ecological collapse, but such a collapse would
have immense human, economic, and spiritual costs.
A simple conceptual model of the impacts of global change illustrates the risk of the
simultaneous degradation and alteration of global ecological processes. The integrated
impact of various global change processes depends upon the degree of interaction among
these separate changes. If these processes are independent of one another, then their
combined impacts can be obtained by adding the individual impacts together. If global
change processes are not independent of one another, but rather positively interact with one
another, then their integrated impacts can be obtained by multiplying the impacts together.
I illustrate this model by considering the linkages among the global change
processes. Usually, the response or adaptation to a global change process depends upon
the flexibility of other components of an ecosystem. For example, more reliable ecosystem
services appear to be produced by diverse ecosystems (Holling et al. 1995). Reductions in
the diversity of species inhabiting a watershed may reduce the ability of that watershed to
reliably provide an ecological service, such as clean water. Similarly, the adaptation of
ecosystems to climate change, depends upon the presence of a diverse pool of species that
can flourish under different sets of conditions. If biodiversity loss and land-use/land-cover
change, eliminate species or their ability to migrate across a landscape, the historical ability
of ecosystems to adjust to climate variation will be greatly reduced.
The different risks associated with the independent and interaction models can be
clearly illustrated by representing the using a Gaussian probability distribution to
characterize the impact of a global change process. For example, consider the interaction of
four different processes (Figure 2-5). A comparison of the independent and interactive
models of integrative global change reveals two key differences between them (Figure 2-6).
First, the uncertainty surrounding the integrated impacts of global change is much less if
processes are independent than if they interact with one another. Secondly, severe impacts
are much more likely if processes interact with one another. This suggests that the
framework used to integrate global change processes strongly influences the prediction of
risk. While this model of system responses to change is simple, it illustrates the
importance of interaction between processes, and the importance of taking an integrated
Anthropogenic global change appears to be of the same magnitude as Earth's
previous periods of revolutionary ecological change. However, the biosphere and
humanity are probably in a more brittle state during this transition than during past
transitions, due to the increasing proportion of the Earth's biosphere that is used by
humanity. The likelihood that humanity's domination of the biosphere is reducing the
resilience of the Earth to anthropogenic global changes underscores the importance of
analyzing the resilience of Earth's biosphere.
The meso-scale is also the scale at which people live and structure the landscape.
These meso-scale activities drive much of global change, and it is through changes to these
activities that people will experience much of the impact of global change. In the following
chapters I analyze how changes in larger scale processes impact the dynamics of the meso-
scale, and the interactions that meso-scale processes have with larger and smaller scale
World 5.00 -
Population 4.00 -
(billions) 3.00 -
0.00 , i .--..-..----
-1000000 -800000 -600000 -400000 -200000 0
Years Before Present (1998)
Figure 2-1. The size of the global human population over the past million years. The
early data is estimated with much less accuracy than the most recent data, but on the scale
of this graph even large relative errors in estimating the past and dwarfed by the magnitude
of current human population. Data are from Cohen (1995).
World o.uu -
Population 4.00 -
(billions) 3.00 -
-2000 -1500 -1000 -500 0 500 1000 1500 2000
Figure 2-2. The size of the global human population over the last four thousand years.
Examining the recent past more closely indicates the most of human population growth has
occurred quite recently. Data are from Cohen (1995).
1000000 100000 10000 1000 100 10 1
Years Before Present (1998)
Figure 2-3. Population of the Earth over the one past million years. The data is plotted on
a log-log scale, diminishing relatively recent large absolute changes. This presentation
shows that the population growth rate has varied greatly over time. On this graph simple
exponential growth follows a curve with a decreasing slope. Data are from Cohen (1995).
0 .0 0 I i -I I I i I I I I i I I I I i i I
1500 1550 1600 1650 1700 1750 1800 1850 1900 1950 2000
Years Before (1998)
Figure 2-4. Estimated yearly growth rate of humanity over the last five hundred years.
The global population growth rate has clearly increased over the past few centuries, and
especially over the last fifty years. While recent growth rates are lower than the 1950's or
1960's they are still higher than any time before then. Data are from Cohen (1995).
- Change A
- Change B
-- Change C
-2 -1 0 1 2
3 4 5 6
Figure 2-5. Example estimated probability distributions of impacts of four imaginary
global change processes. The impact of A is low and quite certain. The impact of B is less
certain, but expected to be larger. C is also expected to produce a large impact, but less
certainly than B. It could even have a positive impact. Similarly, the impact of D is
expected to be large, but it is highly uncertain. Its impact may quite substantial, or it may
Probability 0.6 -
-20 -10 0 10 20 30 40 50 60
Figure 2-6. The integrated effect of the global change processes A, B, C, and D is quite
different depending upon whether they interact following the Interaction or the Independent
model. The cumulative probability distribution for the integrated impacts, shows that
minimal or extreme impacts are much more likely with process interaction case, than with
process independence. If the processes are independent, they have a 95% chance of having
a cumulative impact less than 10. However, if the processes are interactive they have an
only a 70% chance of having a cumulative impact less than 10, and a 95% chance of having
a cumulative impact of less than 24.
SCALING ECOLOGICAL DYNAMICS: SELF-ORGANIZATION, HIERARCHICAL
STRUCTURE & RESILIENCE
Global change research centers much of its efforts on understanding the global
consequences of local actions, and how global changes impact specific, local sites. These
efforts require techniques that are able to translate understanding developed at one scale to
other scales. Unfortunately, it is usually quite difficult to 'upscale' results from small
scales to large scales and 'downscale' the results of global models to local sites (Ahl and
Allen 1996, Levin 1992, Wiens 1989).
Finding useful ways of translating across scales requires understanding the
problems that scaling presents, which I outline in the next section. I then discuss cross-
scale interactions, focussing on the analysis of hierarchical structure, the identification of
hierarchical reorganization, and the search for the potential alternative organizations.
Finally, I discuss how ecological resilience and scaling are intertwined using an example
from the boreal forest. But first, I discuss the problems of scaling.
Global change research continually confronts upscaling and downscaling problems.
Upscaling involves applying the results of a local study to a wide region, such as
extrapolating the effects of an experiment on CO2 fertilization of trees in a forest plot to the
response of the entire boreal forest to an increase in global CO2 levels (Korner 1996).
Downscaling involves moving in the opposite direction. Downscaling attempts to
determine the local consequences of a large scale change on a local site, such as the impact
of global climate change on agricultural production in a field (Parry et al. 1996). Most
global change research has used simple scaling methods, such as averaging or
interpolation, to translate data across scales.
Unfortunately, these methods often fail to adequately map processes from one scale
to another. For example, variation of ecosystem productivity to CO2 fertilization is not the
sum of the variation in the response of individual species, but due to compensation and
complementarity among species the response is far less (Korner 1996). These methods fail
for four related reasons: non-linearity and environmental heterogeneity, dominance of
different processes at different scales, cross-scale connections, and emergent processes.
Non-Linearity and Heterogeneity
First, and perhaps fundamentally, ecological processes are usually non-
linear.Local non-linearity combined with the spatial heterogeneity of the world means that
the aggregate large-scale behavior of local processes is difficult to predict (Levin 1976).
For example, the process of fire interacts with the ecological structure provided by tree
species in North Florida sandhill communities. Either oak (Quercus spp.) or longleaf pine
(Pinuspalustris) tends to dominate these communities. The interaction of fire with fire-
suppressing oaks and fire-encouraging pines produces a non-linear relationship between
the time since a site has burned and it combustibility. A lack of fuel tends to suppress fire
in oak stands, encouraging the growth of more oaks, while fuel accumulation in stands of
pine tend to encourage fire, suppressing oaks and encouraging the growth of pine
(Abrahamson and Hartnett 1990, Rebertus et al. 1993).
The ability of fire to spread, and consequently the stability of patches of oak or
pine, is determined by the distribution of oaks and pine across the landscape. The
combination of spatial heterogeneity and positive feedbacks make the dynamics of forest
difficult to predict from the study of a local site, since processes that define a site are
determined by the properties of its neighbors.
Different Processes Dominate at Different Scales
Different processes dominate at different scales. This observation can be illustrated
by examining vegetative pattern in the North American boreal forest (Figure 3-1). The
growth of an individual tree is determined by vegetative processes influenced by factors
such as light availability, soil moisture, nutrient availability, and temperature. Studies of
tree growth, in small 1 m2 plots over several years, have produced a great deal of
knowledge of tree growth, and these studies have been synthesized in simulation models.
These models predict that shade-intolerant tree species should gradually be replaced by
shade-tolerant trees (Bormann and Likens 1981). However, this replacement does not
occur in the boreal forest, because large-scale spruce budworm (Choristoneurafumiferana)
outbreaks periodically defoliate and kill large areas of shade-tolerant balsam fire (Blais
1983). These outbreaks require large areas of susceptible trees to begin, but the presence
or absence of susceptible areas cannot be detected by small-scale studies. Therefore, up-
scaling local observations requires the accounting for the presence of budworm dynamics at
The presence of different dominant processes at different scales means that as a
scaling method attempts to span a wider ranger of scales, it needs to incorporate the effects
of an increasing number of processes. Scaling methods can either ignore or incorporate
shifts in dominant processes across scales, but both approaches have problems. Ignoring
changes in structuring processes can produce an inaccurate scaling method, however more
complex scaling methods that incorporate multiple processes may be too complicated or
costly to use effectively.
Processes at different scales do not function independently of one another.
Processes operating at small and fast scales are constrained by processes that operate at
slow and large scales, but these large and slow processes are constructed and organized by
the interactions of many small fast processes (Ahl and Allen 1996). These cross-scale
connections make scaling difficult, because scaling relationships only hold if organization
at other scales remains constant. However, change occurs at all scales.
The dynamics of spruce budworm illustrate the importance of cross-scale
connections. Large scale budworm outbreaks are connected to local forest dynamics via
tree mortality. At the local scale, an individual tree is connected to the large scale by
providing a rich food source and refuge from predation that are required for outbreaks to
occur (Ludwig et al. 1978). In this case, these cross-scale connections allow the forest to
be organized into either a cycle of budworm outbreaks or local low-level budworm
infestation (Clark et al. 1979). In the forests of New Brunswick, logging and spraying to
suppress the budworm changed the existing state of the forest from large even-aged stands
with periodic large outbreaks, to smaller and denser even-aged stands with chronic shifting
budworm infestations (Baskerville 1995).
Alternative behaviors, emerging from the same processes, demonstrate how
understanding developed at one scale may only be relevant or useful as long as the
structures and processes at other scales that are maintaining system organization persist.
These cross-scale connections suggest that scaling can only apply over specific ranges of
scale and in specific situations. Forest pattern would scale differently in a forest that has
periodic budworm outbreaks, than it would in a forest experiencing chronic budworm
infestation. By recognizing alternative ecological organizations and assessing what drives
the shifts among them, analysts could improve scaling methods. Separate scaling methods
could be developed for each alternative organization, and the rules for transitions among
alternatives could be used to synthesize these separate scaling methods into an aggregate
scaling method that incorporates the behavior of the ensemble of states.
Occasionally the interaction of process and pattern at one scale produce emergent
organization at a larger, slower scale. For example, the interaction of air and water
circulation over tropical oceans can produce a self-reinforcing vortex, which strengthens
itself as it draws in increasing volumes of warm air. The continued growth of this vortex
produces a hurricane (Barry and Chorley 1992). Emergent processes form due to non-
linearity across heterogeneous space (Nicolis and Prigogine 1977). Many small fast
processes repeatedly interact to produce a larger slower structure that constrains the
behavior of the small processes in such a way that they mutually reinforce one another.
Such emergent processes are self-organized. They are not created by some outside force,
but are created from the mutual reinforcement of their component processes. Emergent
processes present the strongest challenge to scaling theories, because they demonstrate that
in some situations systemic change is not only non-linear, but also structural.
Emergent processes are common in ecology (Perry 1994). They exist in all of the
previous examples. For example, the interaction of tree growth, avian predation, and
budworm population dynamics can create either budworm outbreaks or chronic budworm
infestations. Slight changes in avian predation can shift budworm dynamics from
infestation to outbreak behavior (Figure 3-2). The ecological consequences of outbreaks
are qualitatively different from dispersed chronic budworm infestation. Budworm
outbreaks increase budworm densities over a thousandfold above non-outbreak budworm
densities. At these high densities budworm defoliation kills trees over large areas. The
death of a large number of trees during a short period of time over large area alters nutrient
cycling, tree growth, forest combustibility, and create large even-aged stands that have the
potential to trigger future budworm outbreaks (Clark et al. 1979).
The production of emergent pattern and process complicates scaling, for it suggests
that the organization that is being analyzed can abruptly reorganize, rendering previously
developed scaling relationships invalid as the structures and processes that they incorporate
cease to exist. This issue is particularly important in global change research, as global
change will change ecological organization and processes in novel ways leading to the
production of unexpected ecological processes and structures (Kates and Clark 1996).
Coping with Scaling Problems
Simple scaling, using addition or integration fails when the scaling problems
discussed above apply. Attempting to scale processes by adding new processes and
variables as the scale range of a model increases will work across small ranges of scale, but
will become unmanageably complex across broader ranges of scale. Both of these
approaches are static. They are based upon the assumption that the current organization of
structure and processes will persist over time. The existence of emergent processes
demonstrates that cross-scale organization is dynamic. Translating across these scales in a
dynamic manner requires understanding existing cross-scale ecological organization,
possible alternative organizations, and pathways by which systems can reorganize (Figure
Self-organizing systems appear to be structured by a few key variables (Holling et
al. 1996), which suggests that identifying these variables, and constraints under which they
interact will allow a dynamic, parsimonious method of translating across scales. In the
following sections of this paper, I present a framework for the analysis of cross-scale
interactions based upon self-organization, and then apply this framework to the analysis of
Ecological organization emerges from the interaction of structures and processes
operating at different scales. Traditionally, the effects of scale on the organization of
ecological systems have been analyzed using hierarchy theory. Hierarchy theory focuses
upon the consequences of hierarchical organization, but translating information across
spatial and temporal scales also requires understanding how hierarchical organization is
constructed and how it falls apart.
Hierarchy theory is an extension of systems theory that attempts to analyze the
effects of scale on the organization of complex systems (Simon 1974). Hierarchy theory
does not assume that a system is necessarily hierarchically constructed. Rather, it divides
the world into hierarchical levels to simplify the analysis of cross-scale interactions (Ahl
and Allen 1996, Allen and Starr 1982, O'Neill et al. 1986).
Hierarchy theory regards the world as systems that are comprised of components.
It views a system as a set of interacting components (i.e., lower-level entities) that is itself a
component of a larger system (i.e., a higher level entity). A hierarchical system is
composed of a set of coupled subsystems (O'Neill et al. 1986). Hierarchical organization
removes or attenuates interactions between system components. In a non-hierarchical
system any components may interact with any other; however in a hierarchical system
components are grouped into sub-systems that interact strongly internally, but only weakly
with other sub-systems (Simon 1974).
Hierarchy theory analyzes ecological patterns by separating the large from the
small, and the fast from the slow. Hierarchy theory has been used to organize analysis at
different scales, both spatially and temporally. Ecologists have analyzed how vegetation
patterns and the processes organizing pattern change with spatial scale (Krummel et al.
1987, O'Neill et al. 1991). Similarly, the constraints of slower temporal dynamics on
faster processes have been analyzed at a variety of scales. For example, fire frequencies in
the southwest United States are constrained by El Nifio fluctuations (Swetnam and
Betancourt 1990). This approach is useful for analyzing existing ecological configurations,
but it is not as useful for understanding the formation, dynamics, and reorganization of
The concept of a hierarchy is a human construct, but natural hierarchical
organizations may have arisen because hierarchy offers life a more stable alternative to
'flat', non-hierarchical organization. Because hierarchical construction increases stability,
systems with hierarchical organizations should be able to form and persist (Holling 1992a,
Simon (1974) uses a parable of watchmakers to illustrate the advantages of
hierarchy. He describes one watchmaker who uses a hierarchical watch construction,
while another does not. When the work of the two watchmakers is periodically disrupted
each watchmaker loses his work, but the watchmaker using hierarchical construction, loses
only that component. Therefore, even if the costs of hierarchical construction are high and
the disturbance rate low, the hierarchical watchmaker is able to produce far more watches
than the non-hierarchical watchmaker. This parable illustrates the virtue of modularity as a
defense against disturbance. It argues that the ideal complexity of a module depends upon
the disturbance rate, and the cost of isolating function within a module. A low disturbance
rate encourages complex modules, while a high disturbance rate encourages simple
Ecological organization occurs as processes interact with structure across space and
time. For example, the mutual reinforcement between the process of fire and structure
provided by longleaf pine will only occur if the fires are started frequently and fires are able
to spread across a large area. Otherwise, sites will bum infrequently and fire-encouraging
vegetation will be replaced by fire-suppressing vegetation. The mutual reinforcement of
fire and vegetation pattern produce spatial patterns of fire, longleaf pine and hardwoods, at
a scale larger and longer lasting than any individual fire or mix of vegetation.
Simon's argument can be inverted to understand why and how self-organized
hierarchies form in ecosystems. Ecosystems are hierarchically organized, because they are
composed of individual organisms and abiotic processes that operate at a variety of scales.
The persistence of a particular organization depends upon the degree to which the
interactions are mutually reinforcing relative to the degree to which the organization is
disrupted. If conditions are favorable, spatial-temporal interaction of structures and
processes produces an emergent pattern at a larger and slower scale than the scale of the
processes and structures themselves. These self-organized patterns and processes can,
through interaction with other larger and slower processes, organize still larger and slower
sets of pattern and process. This type of interaction produces nested hierarchies in nature.
I claim that a nested hierarchy exists in the boreal forest. The forest is composed of
stands, which are composed of trees. Patterns a higher scales emerge from interactions at
the smaller levels. At the scale of a tree, interactions over tens of meters between trees
determine ecological structure. The cumulative impact of a tree and its neighbors, shapes
fuel accumulation over a larger area. The larger scale process of fire interacts with this fuel
accumulation, to produce stand creating, tree-destroying forest fires. The pattern of
patches across the landscape influences larger scale processes, such as moose foraging
(Pastor et al. 1993) and weather (Knowles 1993), shape the forest at larger scales. The
emergence of new structuring processes at larger scales means that it is inappropriate to
consider larger scales simply as aggregations of a large number of small-scale entities.
Hierarchy theory suggests that the rules that organize a hierarchy are the ones that
should be used to translate across scales. For example, if higher hierarchical levels average
changes in the lower levels, then simple scaling methods, such as interpolation or
averaging, are likely to be successful. While, if non-linear processes produce the higher
levels then a scaling rule that is appropriate to those specific non-linear interactions can be
discovered. For example, the frequency of forest fires of different sizes can be appears to
be a function of fire size raised to an exponent (Malamud et al. 1998). This type of power
law relationship can be used to scale forest or fire dynamics.
While hierarchy theory can explain why hierarchical organization offers advantages
over flat organization, it does not provide much insight into how hierarchical structure
forms, changes, or dissolves. Understanding these processes requires examining process
The conventional view of hierarchies is static and structural (Pickett et al. 1989, Wu
and Loucks 1995), however this approach ignores the processes that build and destroy
hierarchical structure. In the past decade, complexity theory has expanded and enriched
systems theory, producing a dynamic conception of hierarchy. Holling (1986) has
proposed a general model of ecological change that I use to organize my discussion of
dynamics in hierarchies.
Holling's adaptive cycle model proposes that the internal dynamics of systems cycle
through four phases: rapid growth, conservation, collapse, and re-organization. As
unorganized processes interact, some processes reinforce one another, rapidly building
structure, or organization. This organization channels and constrains interactions within
the system. However, the system becomes dependent upon structure and constraint for its
persistence, leaving it vulnerable to either internal fluctuations or external disruption.
Eventually, the system collapses leaving disorganizing structures and processes to
reorganize a new set of structures. This dynamic can be seen in a forest stand. As a young
stand grows it gradually becomes denser, accumulates fuel, and becomes increasingly
susceptible to fire. Following a fire, the stand is reorganized as plants resprout from roots
or seeds, producing a new forest stand.
The key point of this model is that during the development of a system, its
organization changes, and due to these changes there are times when it is either more or
less vulnerable to internal and external fluctuations. This point of view acknowledges the
importance of top-down constraint on processes, but it also recognizes that during some
periods large-scale processes are unstable and are vulnerable to change from below. For
example, in a dense old forest damage to an individual tree may allow bark beetles
(Coleoptera: Scolytidae) to reach densities at which they can attack and kill nearby healthy
trees, initiating an outbreak that can alter forest pattern over a large area (Berryman et al.
1984). Similarly, after clear-cut logging, quick small-scale processes, such as seed
dispersal and competition between seedlings, shape the subsequent development of that site
for decades (Bormann and Likens 1981). During other times, slower and larger scale
processes are stable and resilient, constraining the lower levels and integrating noisy
variation of small, faster processes. It is during the periods of destruction and
reorganization that a system is most vulnerable to small, fast processes (Gunderson et al.
I propose that there are three ways that change propagates through dynamic
hierarchies (Figure 3-4). First, change at a higher level alters a lower level due to the
constraints that it places upon it. Second, small-scale disturbance can trigger larger scale
collapse if the larger system is at a brittle stage in its adaptive cycle. Third, following the
collapse of a system, small-scale organization drives the formation of large-scale structure
during periods of reorganization. These changes occur within a dynamic hierarchy that is
embedded in a larger environment, that is itself composed of other dynamic hierarchies.
By considering these hierarchical dynamics a richer set of scaling methods can be
developed. For example, a forest that is vulnerable to a bark beetle outbreak will be better
represented by different scaling methods than a forest that is not. Identifying the
reorganization events that cause an ecosystems scaling to change allows an integrated
scaling method to be developed that identifies the conditions in which different scaling
relationships are appropriate.
Ecological organization across scales is the result of the interaction of processes
operating at different scales. For organization to persist these interactions must not
immediately disrupt one another, rather they should be independent of one another or
reinforce one another. But as these processes alter ecological structures, the interactions
that are possible change. For example, the accumulation of oak leaves in a forest can
suppress the growth of understory vegetation, reducing the possibility of fire. Changes in
ecological organization can: alter the scale at which a process functions, eliminate a process
by disrupting the interactions that maintain it, or create new interactions that enable a new
process to occur.
The spatial and temporal scale of a fire regime is defined by the cross-scale
interaction of fire with relatively slower vegetative processes and faster atmospheric
processes (Figure 3-1). The key processes defining a fire regime are the rate of fire
ignition, the rate of forest recovery following a fire, and the rate of fire spread. If fire
spreads faster than a forest recovers from fire, and forest recovery is faster than the ignition
of another fire, then the average fire size is proportional to the rate of recovery divided by
the rate of ignition (Drossel 1997). These rates emerge from the interaction of vegetative
and atmospheric processes. Changes in these processes that alter these key variables will
alter the fire regime. For example, a change in climate that increases the frequency of
thunderstorms could increase the rate of fire initiations, which would produce smaller and
more frequent fires. Alternatively, a region invaded by pyrogenic grass can recover from
fires far quicker, and this would increase the frequency and spatial extent of fire
(D'Antonio and Vitousek 1992).
Changes in the scale of fire regime will alter the scale range over which a scaling
method works. Incorporating the variables that drive these changes into a scaling method
allows scaling methods to become dynamic, because it incorporate changes in scaling
relationships into a scaling method. For example, given a scaling law that describes forest
fire frequency as a power law of forest fire size (Malamud et al. 1998), simulation
modeling could be used to assess the sensitivity of the scaling exponent to changes in fire
ignition, forest recovery, and fire spread. This type of analysis could potentially identify
when the scaling behavior of an ecosystem would break down.
The emergence, maintenance, and destruction of ecological organization are due to
cross-scale interactions. The stability of a given organization over time and space depends
upon the nature of these interactions. The strength with which they reinforce one another,
their sensitivity to disruption, and the presence of alternate configurations of interactions.
These attributes of ecological organization can be integrated and analyzed from the
perspective of cross-scale resilience.
Ecological resilience is a measure of the amount of change or disruption that is
required to cause a ecosystem to switch from being maintained by one set of mutually
reinforcing processes and structures to a different set of processes and structures (Holling
1973). As long as an ecological organization maintains its resilience, its organization
persists. If the scaling behavior of a given ecological organization can be represented by a
scaling method, maintaining resilience also means maintaining the ability to use a scaling
method. The limits of an ecosystem's resilience are also the limits to the applicability of a
given scaling method. A loss of resilience causes an ecosystem to reorganize, and the
scaling behavior of a new ecological organization should be represented by a new scaling
method. Resilience emerges from both cross-scale and within-scale interactions (Peterson
et al. 1998).
Cross-scale resilience is produced by the replication of process at different scales.
The apparent redundancy of similar function replicated at different scales adds resilience to
an ecosystem, because disturbances are limited to specific scales, functions that operate at
other scales are able to persist. For example, Fragoso (1997) has shown that in Brazil's
Maraca Island Ecological Reserve palm seeds are dispersed across a range of scales by a
variety of species. Seed dispensers range in size from small rodents, which typically
disperse seeds within 5m of parent trees, to tapirs (Tayassu tajacu), which disperse seeds
up to 2 km. Seed dispersal at multiple scales allows the palm population to persist despite a
variety of disturbance processes occurring at different scales, because the trees are
distributed across the landscape at different scales. Cross-scale resilience is complemented
by within-scale resilience.
Within-scale resilience is produced by compensating overlap of ecological function
between similar processes that occur at the same scales. In the previous example a variety
of species, including deer, pecaries, primates, and rodents, disperse palm seed short
distances (Fragoso 1997). Population fluctuations of a species do not have a large effect
on the rate of short distance seed dispersal, because a decline in seed dispersal by one
species leaves more seeds available for other species. An increase in seed availability
allows other species to increase their seed dispersal compensating for initial decline in
dispersal. This compensating complementarity among seed dispersers reduces the impact
of population fluctuations on ecosystem function (Frost et al. 1995), increasing its
By focussing on the cross-scale and within scale sources of an ecological
organization's resilience one can assess the vulnerability of a particular scaling relationship
to within-scale and cross-scale changes. In the following section I discuss how such an
approach has been applied to boreal forest dynamics.
Avian Predation and Spruce Budworm
In the forests of the Canadian province of New Brunswick, outbreaks of spruce
budworm periodically defoliate and kill large areas of mature boreal fir forest (Clark et al.
1979, Morris 1963). The specific form that the budworm/forest system organizes into is
controlled by a variety of forces including climate, landscape pattern, and avian predation
on budworm. Changes in any of these factors can alter the organization and scaling pattern
of the forest, but I will focus upon the effects of variation in avian predation.
Avian predation on budworm slows budworm population growth, altering the
frequency and extent of budworm outbreaks (Holling 1988). Avian predation can prevent
budworm population growth over a wide range of budworm densities. However, at high
densities avian predation has a minimal impact on budworm population dynamics, because
the per budworm effect of avian predation declines. This decline allows the budworm
population as a whole to grow at a faster rate, escaping from avian predation. The range of
budworm densities over which avian predation can control budworm population growth
controls the frequency and intensity of budworm outbreaks, and the effectiveness of avian
predation is determined by within-scale foraging diversity and cross-scale reinforcement.
Within-scale foraging diversity means that similar sized bird species prefer different
prey items. These differences in preferences increase the robustness of predation, because
if a particular prey item increases in abundance, it becomes easier to capture, and species
that did not previously prey upon it will begin to consume it (Murdoch 1969). These
predators switch from their preferred prey because the increasing relative abundance of a
resource makes its use less costly. This switching behavior increases the effectiveness of
avian budworm predation, because the group of species provides more resilient predation
than any single species.
Cross-scale functional reinforcement occurs because different species of birds
forage for budworm at different scales. A bird's body size has a strong influence on how it
forages, determining the amount of food a bird can consume and the scale at which it
searches for food (Peters 1983). At high population densities, budworms are spatially
aggregated. The concentration of food that is available in these aggregations allows larger
birds to consume budworm. These birds do not eat budworm when it is dispersed,
because searching for individual budworm is not worth the effort. Additionally, larger
birds forage over wider areas than smaller birds. Consequently, the size of birds that are
able to effectively consume budworm and the distance from which they are attracted will
increase as the size of local aggregations of budworm increase.
While within-scale and cross-scale diversity do produce effective predation on
budworm, budworm populations can escape from control by birds. Local concentrations
of budworm can be controlled by bird migration, but budworm increases over a wide area
require a huge amount of predation to control. The density of budworm at which avian
predation is unable to control budworm populations determines the frequency of budworm
outbreaks and their spatial scale. Holling (1988) used Ludwig et al.'s (1978) mathematical
model to calculate that large changes in bird predation, an approximately 50% decline in
effectiveness, would be necessary to significantly to shift the scale of budworm outbreaks.
I used the same model to illustrate how different levels of avian predation alter the temporal
scaling behavior of forest vegetation.
I used Ludwig et al.'s (1978) simple analytical model of budworm outbreaks to
simulate budworm-forest dynamics. This is the same model that I used to generate Figure
3-2. This model simulates budworm-forest dynamics through three coupled differential
equations that represent slow changes in forest volume, medium term changes in tree
foliage, and fast changes in budworm densities.
I used the parameter values of Ludwig et al (1978) as a base case, I altered the
maximum amount of predation that was possible in the model. This alteration is equivalent
to simulating increases or decreases in avian populations. I generated a millennium of
forest dynamics, and then analyzed the variance in forest age using a range of temporal
windows. I normalized these variance relationships to be proportional to the maximum
variance in each simulation run. Changes in the normalized variance reveal the relative
amount of variance present at a scale (Figure 3-5).
The greater the rate of decrease in relative variance at a scale, the greater the
proportion of variance present at that scale. Vertical arrows indicate the time scale at which
the greatest amount of variance has been lost. These scales correspond to the periodicity of
budworm outbreaks at different levels of predation. Reducing predation decreases the scale
over which variation occurs. Increasing the amount of predation increases the scale of
budworm outbreaks. When predation is increased more than 17% above the base level
predation becomes strong enough to prevent outbreaks from occurring, causing forest
dynamics to follow a fundamentally different scaling relationship.
The analysis of the sensitivity of scaling relationships to changes in the processes
that support a specific ecological is useful in showing how changes in underlying processes
shift scaling relationships, and cause them to fall apart. Figure 3-5, shows that the
difference between no predation and 50 % of the base predation is very slight, but that at a
20% increase over the base predation would cause the existing scaling relationship to
collapse. This collapse in the scaling relationship corresponds to the difference between the
chronic budworm infestation and budworm outbreak that are shown in Figure 3-2.
Such a change occurred in the forests of New Brunswick when pesticide spraying,
which by killing budworm is similar to an increase in predation, eliminated budworm
outbreaks causing a chronic budworm infestation (Baskerville 1995).
Scaling and Ecological Reorganization
Scaling relationships are essentially models that are used to predict the behavior of a
system over space and time, given a few measurements. These models are constructed
based upon the analysis of the existing cross-scale dynamics of a system. However, a
given set of scaling relationships is unlikely to persist in a world experiencing global
change. Global changes in climate, land-use/land-cover, and biodiversity will all alter
ecological organization and scaling.
Analyzing the cross-scale dynamics of existing ecological organization allows the
assessment of the degree to which change alters scaling relationships, or causes them to
collapse entirely. As was illustrated in budworm predation example, models can be used to
assess the resilience of ecological organizations. Additionally, if alternative ecological
organizations are known, and the causes of switches between these configurations can be
determined then multiple relatively simple scaling rules can be used in large models,
without explicitly incorporating the dynamics that drive changes in scaling into a scaling
The dynamics of ecological organization are an essential component of global
change research. Global change is the convolution of a huge variety anthropogenic
changes in Earth's biosphere, geosphere and atmosphere that all, to some degree, alter
ecosystem functioning. Assessing the impact of change on ecosystems requires the use of
scaling methods, but these scaling methods may not apply in an altered world. The
resilience of ecological organization to changes in key processes determines the situations
in which scaling methods apply, require adjustment, and break down. These analyses can
be used to developing scaling methods that dynamically compensate for shifts in ecological
Translating either data or understanding across scales is often difficult, because the
interaction of ecological processes self-organize dynamic hierarchical structure. Ecological
change can alter the resilience of an ecosystem, cause ecological reorganization, and
eliminate alternative ecological organizations.
Ecological change that alters the resilience of an ecosystem may shift its scaling
relationship, reducing or expanding the scale range over which a relationship applies, as
shown by the analysis of changes in avian predation on budworm. Reduction in predation
leads to more frequent outbreaks, reducing the scale of budworm outbreaks. This changes
the scaling of forest pattern. Similarly, the loss of large-scale seed dispersal would reduce
that scale of forest patterning.
Change may push an ecosystem beyond the limits of its resilience, causing it to
reorganize. Ecosystems as diverse as coral reefs (Done 1992), shallow lakes (Scheffer
1998), and woodlands (Dublin et al. 1990) have been clearly shown to switch between
alternative ecological organizations. Such ecological reorganization alters an ecosystem's
scaling, as new ecological processes operate over different scales.
Furthermore, the addition or subtraction of processes from an ecosystem may
eliminate alternative ecological organizations. This appears to have occurred during the
Pleistocene when the extinction of mega-herbivores eliminated the disturbance processes
that maintained specific ecological organizations (Owen-Smith 1989, Zimov et al. 1995).
The loss of a form of ecological organization is extremely difficult to reverse, particularly if
that organization was generated by a species that has become extinct (Flannery 1994).
I have argued that by focusing upon the cross-scale processes that create, maintain,
and destroy ecological organization scaling relationships can include important ecological
dynamics. By analyzing the cross-scale organization of ecosystems, methods for
translating across scales can be developed. These scaling methods can be improved by
considering the dynamics of cross-scale organization. By perturbing dynamic ecological
models, the effects of change on scaling methods can be calculated, and the limits of a
system's resilience predicted.
Ecological change that alters the resilience of an ecosystem may shift its scaling
relationship; reducing or expanding the scale range over which a relationship applies.
Humanity lives and manages the landscape at the meso-scale. As well as changing
the distribution and organization of ecosystems, these activities have also altered the scaling
of ecological processes. For example, fire suppression and landscape fragmentation have
altered the scaling behavior of fire. In the following chapters I analyze how these types of
larger scale processes impact the dynamics of meso-scale processes, and how, in turn,
these transformed meso-scale processes influence larger scales.
100 1 000 the
km km Earth
-5 -4 -3 -2 -1 0 1 2 3 4
Log Space (Kilometers)
O Vegetative structure Disturbance processes
0 Atmospheric processes
Figure 3-1. Time and space scales of boreal forest structures (Holling 1992a) are compared
with disturbance and atmospheric processes which structure the forest. These processes
include spruce budworm (Choristoneurafumiferana) outbreaks, fire, atmospheric
processes, El Niho, and the rapid CO2 increase in modem times (Clark 1985).
9.0 Budworm 1.8
Log 7.0 1.4
Budworm 6.0 1.2
(larvae/acre) 5.0 1.0 Surface
0 50 100 150 200 250 300
A) Time (years)
9.0 rBudworm -1.8
Log 7.0 1.4
Budworm 6.0 1.2 Forest
(larvae/acre) 5.0- 1.0 Surface
0 50 100 150 200 250 300
B) Time (years)
Figure 3-2. Two plots illustrating alternate relationships between spruce budworm and
boreal forest volume, generated from Ludwig et al.'s (1978) simple analytical model of
budworm outbreaks. Graphs A and B use Ludwig et al's base parameters, except the
maximum avian predation rate is 25% higher than the base in graph A. A) Non-outbreak
spruce budworm, infecting a forest at relatively high densities. Budworm populations are
controlled by higher than observed levels of avian predation. The forest suffers some loss
of productivity from budworm, but little mortality. B) When avian predation is reduced to
observed levels, budworm outbreaks occur. These outbreaks increase budworm densities
a thousandfold over the non-outbreak budworm densities. At these high densities
budworm defoliation kills trees. The outbreaks entrain forest structure, producing a dis-
equilbrium forest that has a periodicity of about seventy years.
Oc m 10-13
0 10 20 30 40 50 60 70 80 90 100
A) Time (years)
0 10 20 30 40 50 60 70 80 90 100
B) Time (years)
Figure 3-3. Simulation of two longleaf pine forest with the same initial composition of
forest types, but different spatial distributions of those types. The model used is a one
dimensional spatial model of fire and forest dynamics (Peterson In Press). A) Two old
sites are adjacent to one another. The sites remain old, while the other sites remain young.
B) Two old sites are separated by several young sites. Fire is prevented from burning
these sites, and they become old. The other young sites remain young. After 100 years of
simulation, both the mix of site types and their spatial pattern are quite different. If the
behavior of these models had been extrapolated from either the initial conditions of sites, or
the initial mix of sites, those extrapolations would have been inadequate, unless the spatial
relationships among sites are also considered.
larger constrain i constrain constrain
slower V I noise V noise noise
scale constraint constraint
smaller noise V I 'sparks' 'seeds'
faster a: (c b) c: = i >
Figure 3-4. An idealized map of the relationships between a hierarchical level and the
levels directly above and below it, as the nature of the central level is dominated by
different types of processes. a) A system during 'normal' times. Each level is strongly
constrained by the level above it. 'Noise' from the faster levels below each level has a
minimal effect upon higher levels. The system is resilient. b) A 'brittle' system strongly
constrains the level beneath it, but also extremely vulnerable to any change in the level
beneath it. This brittleness allows a small disturbance to rapidly propagate through the
entire system. c) A reorganizing system only weakly constraints its subsystems, making it
prone to organize around any 'seeds' of order that emerge from change in the lower levels.
(V/Vmax) 0.1 _- -50%
:- .- - >+17%
S V V Predation
1 10 100
Scale (window size years)
Figure 3-5. Changes in the intensity of predation alter the scaling of forest pattern in a
simple model of budworm outbreaks. Ludwig et al.'s (1978) simple analytical model of
budworm outbreaks was used to calculate variance in forest age over a range of temporal
window size. These variances were normalized based upon the maximum variance at each
level of predation. Using the parameter values of Ludwig et al (1978) as a base case, I
altered the amount of avian predation that was possible in the model. Reducing the
predation rate decreases the scale over which variation occurs, by decreasing the frequency
of budworm outbreaks. Increasing the maximum amount of predation, increases the scale
of budworm outbreaks, until outbreaks can no longer occur. Predation levels greater than
15% above the base case are sufficient to prevent outbreaks, causing the forest pattern
follows a fundamentally different scaling relationship. The vertical arrows indicate the time
scale over which almost all the variance has been lost for the 'Base Predation' case and
reduced predation cases. These time scales correspond to the periodicity of outbreaks,
which are respectively: 17 years, 24 years, 35 years, 65 years, 89 years, and no outbreak.
FIRE, SEED DISPERAL AND MULTIPLE STABLE STATES IN LONGLEAF PINE
This chapter focuses upon understanding the cross-scale dynamics of human
dominated longleaf pine (Pinuspalustris) ecosystem in northwest Florida. It applies the
general discussion of resilience, ecological scaling, and large-scale global dynamics to a
specific, complicated managed ecosystem.
I use the concept of ecological resilience to construct a set of alternative models of
longleaf pine sandhill forests that are used to explore the ecological behavior of the
managed forest of Eglin Air Force Base in northwest Florida. In this introduction, I
discuss my approach towards ecosystem management and then describe the longleaf pine
forests of Eglin Air Force Base. The remainder of this chapter is divided into four
sections. First, I present the set of simple alternative models, or caricatures, that I use to
explore the structure, dynamics, and management of northwest Florida longleaf pine forest.
Second, I describe and use a non-spatial model to explore the temporal dynamics of
longleaf pine forest. Third, I describe and use a spatially explicit model to explore the
spatial-temporal dynamics of longleaf pine under a number of management regimes. I
conclude by discussing the ecological and management lessons learned from this
exploration. This includes identifying the important areas of uncertainty that remain and
suggesting appropriate management actions.
Human transformation of the earth has largely occurred at the meso-scale. These
transformations have altered the abundance and distribution of biota. More subtly, human
action has altered the functioning of ecological processes, encouraging some, suppressing
others, and creating others. These changes in ecological processes often have gradual, but
cumulative effects that present a significant challenge to humanity's ability to effectively
manage ecosystems. These changes produce ecosystems that are organized in novel ways.
Learning how to structure human relationships with poorly understood, changing
ecosystems is one of the central challenges of modem ecology.
Applied ecologists are caught in a trap. Ecological change requires that they act, but
ecological change means that they have limited knowledge of how they should act. Applied
ecology, or ecosystem management, is based upon an abstraction of nature. From the vast
complexity of nature, managers abstract a set of system attributes that they feel capture the
structure and dynamics of the system of interest, and then use that set of relationships to
guide their decision making. Usually this decision process is based upon some form of
ecological theory, even if it is unarticulated. Often scientists work with managers with the
aim of improving the ecological model used by managers. However, often there are a
variety of competing ecological models that can not be rejected based upon existing
knowledge of the ecosystem being managed. One approach to this dilemma, which is
advocated by adaptive management (Holling 1978, Walters 1986), is to base management
upon a set of alternative ecological models rather than attempt to discover which model
performs the best according to a set of criteria.
Exploring a broad set of alternative models helps researchers ask better questions
about how ecosystems work. This allows managers and ecologists to engage in an iterative
process of synthesis, reflection, and experimentation that helps them move towards a richer
understanding of ecosystem dynamics. Perhaps more importantly, the continual use and
evaluation of a diverse set of models reminds ecological managers to attend to the
uncertainty that exists over how the system operates, encouraging management that tests
rather than follows models.
Developing such models is difficult, but the general ecological concepts of multiple
stable states, resilience and scaling, as discussed above, can provide a framework for
model construction. Useful models of complex situations compress a great deal of
understanding into simple, yet rich models that exhibit the minimum complexity to describe
important system dynamics. Such models cannot be developed immediately. Often they
can only be developed after many different complicated interactions are explored and tested.
Such a process usually narrows the set of interactions that have significant effects on
ecological dynamics. This process allows complex models to be compressed into minimal
models. I used these concepts to develop multiple ecological models to explore
management alternatives for forest restoration in Northwest Florida.
Longleaf Pine Forest on Eglin Air Force Base
From the end of the Pleistocene until recently, the majority of the coastal plain of
the southeastern U.S.A. was covered by longleaf pine (Pinuspalustris) forest. During the
past two centuries, human activities such as logging, agriculture and fire suppression have
reduced the area covered by longleaf pine from about 250,000 km2 to less than 12 500 km2
(Schwartz 1994). The largest remaining area of contiguous old-growth longleaf pine forest
is located in the northwest Florida, within the 1 870 km2 Eglin Air Force Base.
Either longleaf pine or various oak species (Quercus spp.) can dominate forest sites
located on the sandy soils of northern Florida (Abrahamson and Hartnett 1990). The
frequency of ground fires usually determines which vegetation type dominates at a specific
location (Heyward 1939). Mature stands of longleaf pine often produce open savanna
ecosystems that possess an understory rich in grasses, and herbs. Longleaf pine shed
needles that, in combination with the understory vegetation, provide a combustible medium
for fire. While longleaf pine stands encourage fire spread, oak stands tend to inhibit fire
spread. Mature oaks form dense stands that reduce the amount of light that reaches the
ground, producing a sparse understory. Oaks shed leaves that form a compressed, low
oxygen litter layer that suppresses the accumulation of understory vegetation and other
potentially combustible detritus.
Fire plays a key role in determining what types of tree seedlings survive to become
trees. Longleaf pine, in all but its young sapling state, survives fire. It also drops many
fine needles which ignite. Young oaks, however, are killed by fire. The absence of fire
from oak stands or longleaf pine stands allows the growth of young oaks. However,
regular fires suppress oak growth allowing longleaf pine to thrive (Rebertus et al. 1989).
The difference between fire's relationship with longleaf pine and oaks drives forest
dynamics in northwest Florida. Longleaf pine through the combination of fire resistant and
the production of fine fuels drive a forest site towards a frequently burned longleaf pine
savanna. While oaks through the combination of fire inhibition and rapid growth in the
absence of fire drive forest sites towards infrequently burned oak scrub-forest.
While the tension between oak and longleaf pine dominance occurs over much of
the southern coastal plane, a third tree species, sand pine (Pinus clausa), further
complicates ecological dynamics on Eglin Air Force Base. Sand pine historically occurred
along the coast, but in past decades has spread into areas formerly occupied by longleaf
pine (Provencher et al. 1998). The Choctawhatchee variety of sand pine that inhabits
northwest Florida is ecologically distinct from other sand pine. While interior Florida sand
pine are serotinous, requiring fires to open their cones before seeds can regenerate, the
Choctawhatchee variety is not (Parker and Hamrick 1996). Sand pine is not as fire tolerant
as longleaf pine, but sand pine reaches sexual maturity much faster than longleaf pine.
These differences mean that sites that are not regularly burned can be invaded by
neighboring sand pine within several years.
On Eglin Air Force Base, longleaf pine savanna has been replaced by pine
plantations and invaded by oaks and the Choctawhatchee variety of sand pine. Presently,
the land managers at Eglin Air Force Base are attempted to restore the now rare longleaf
pine savanna ecosystem (Jackson Guard: Eglin Air Force Base 1993).
Restoring the forest requires understanding its ecology and its history. Ecological
understanding is needed to develop useful management plans, while an understanding of
the past is required to understand what processes and states should be encouraged or
discouraged. While there is a large body of knowledge concentrated on longleaf pine forest
(Glitzenstein et al. 1995, Platt et al. 1988), this knowledge has often been developed at the
scale of forest stands over several years. However, ecosystem management operates at the
scale of thousands of square kilometers and decades. This scale mismatch requires
translating understanding and uncertainty across scales. The development of simulation
models provides a means of integrating ecological and management knowledge. By
constructing alternative models, different models of cross-scale dynamics can be
formulated, explored, and tested. It was this approach that I applied to Eglin Air Force
Ecological Caricatures and Management Alternatives
Management alternatives are developed based upon models of how the forest
functions. From meetings and discussions with land managers and ecologists I developed
a set of simple caricatures of Eglin's forest types and dynamics. These caricatures were
developed to articulate the concepts that implicitly underlie most actual and proposed
management strategies. These caricatures assume specific types of landscape change.
Testing these caricatures may eliminate a several of them, along with the management
strategies that they imply. Furthermore, testing the caricatures reveals the key differences
among them, and the situations in which these differences matter to management.
Identifying these points allows a more refined set of alternative hypotheses to be posed and
tested through management actions.
Caricatures of Ecological Dynamics
I identified five caricatures of the changes that have occurred in Eglin's forests.
Two focus on the temporal dynamics of the forest, while three present alternative models of
spatial and temporal interactions. Each caricature and a management strategy associated
with it are defined below.
Out of Tune Forest: Fire frequency determines whether vegetation is maintained
or changed. The frequency of fire in Eglin has decreased, leading to an increased
presence of hardwood tree species and sand pine in the forest.
Suggested Management: Restoring longleaf pine requires returning the forest's fire
frequency to its historical level. This task requires discovering the 'right fire
frequency.' Apply this frequency of fires to the landscape and the forest will be
Transformed Forest: The invasion of hardwoods and sand pine has
fundamentally changed the forest. Fire behaves differently and has different
impacts in this transformed forest.
Suggested Management: Restoring longleaf pine savanna requires intensive fires to
transform the forest. Once the forest is restored it can be maintained with less
frequent fires. This requires policy that varies the fire frequency applied to an area
based upon the vegetation of that area.
Edaphic Forest: Soils and topography limit vegetation to specific locations on the
landscape. Changes in vegetation occur only within a limited set of alternatives
depending upon local site characteristics.
Suggested Management: Some sites will be difficult or impossible to change.
Managers should focus their efforts on the areas that can be improved should areas
and ignore the areas cannot be altered.
Divided Forest: Development has broken the landscape into many pieces,
altering the spread of fire across the landscape, which has decreased the frequency
of fire experienced by sites. This has lead to the loss of longleaf pine and the
invasion of hardwoods and sand pine
Suggested Management: The scale of fire must be expanded from beyond fragments
to cover the entire landscape.
Invaded Forest: Sand pine has invaded the forest, altering its dynamics.
Suggested Management: Fire frequency must be sufficient to prevent sand pine
spread. The areas that are occupied by sand pine will expand over time unless the
spread of sand pine is controlled. Managers should focus upon rapidly removing
areas of sand pine that have the potential to invade substantial areas.
The first two caricatures focus upon the interaction of fire and vegetative processes
through time. The next three caricatures focus upon how changes in landscape pattern
shape fire and seed dispersal, which are the processes that strongly influence the pattern of
the landscape itself. These caricatures are not mutually exclusive. Changes in the forest
are likely due to a combination of these processes, but management actions will be quite
different based upon which of these caricatures best captures the forces shaping Eglin's
landscape. A modeling process was used to test and explore the differences among these
Ecological modeling serves of variety of goals. Primarily, a model is a synthetic
hypothesis. It combines many diverse and heterogeneous results, observations, and
theories into a unified framework that proposes a model of how an ecological system
The process of developing a model can be useful in and of itself. By assembling
the concepts surrounding an issue in a system, and attempting to synthesize them is an
opportunity for learning about the system. It provides participants to learn about areas
outside their area of expertise, can reveal hidden connections, and establish important gaps
in existing understanding (Walters 1986).
Once a model is constructed, it is a tool for reflection. When its builders examine it
they may not like what they see, but what they see is what they chose to put there, because
a model embodies their prejudices, and ignorance as well as their knowledge. Models
allows individuals and groups to see if they believe what they thought they believed, by
clearly and explicitly laying out the things and processes that people consider to be
important to the function of a system.
The process of reflection often results in the development of alternative models as
processes that were thought to be important are shown to be peripheral, and marginalized
processes are shown to be important. Changes in the processes considered by a model
usually requires changes in the variables that are included within a model. Often a model
evolves over several years, from discussions that alternate with attempts at synthesis, that
are followed by periods of reflection, criticism, speculation and model reformulation.
Frequently, this process can produces a set of models, or a modeling framework that is
acceptable to the people involved in the modeling processes.
For Eglin Air Force Base, I produced a set of models that embodied the ecological
caricatures above. These models represented what workshop participants felt were key
ecological structures, processes, and management actions. At this point the model's basic
structure, but not the rules describing how vegetation changes, were accepted as being
reasonable by the various workshop participants. The model provides a formal document
of consensus opinion of how, in broad terms, of how and why sandhill forest in north
Florida changes. I used this modeling framework to conduct two sets of simulation
experiments to identify what aspects of forest dynamics have important consequences for
landscape management. The first set of experiments used a non-spatial model to explore
forest dynamics under a wide variety of environmental regimes. The second set of
experiments used a spatially explicit model to explore spatial-temporal forest dynamics
under a narrower set of forest management scenarios.
Succession and Fire Temporal Dynamics
Modeling Vegetation Dynamics
The central abstraction in the model is its representation of forest in Eglin as a mix
of the three dominant tree species. I assumed that an area of forest could be represented as
a mix of longleaf pine (LL), sand pine (SP) and hardwood species (HW). This allows the
character of a forest patch to be characterized by its relative proportion of these three tree
types. This mix of tree types can be plotted as a position within a triangle, because of the
assumption that the proportions of the three types comprise the total canopy (i.e. LL + SP
+ HW = 1.0). Changes at a site in the forest can therefore be thought of as tracing a
trajectory through the triangular state space representing the various potential canopy types
The various possible combinations of longleaf pine, sand pine and oaks that are
possible in Figure 4-1, were reduced to a set of discrete states. These states were based
upon translating existing management classifications into the triangular state-space
representation of forest composition. The infinite number of possible mixes of tree types
was reduced to eight distinct vegetation states, each representing a discrete type of forest.
These classes are briefly described below, and illustrated in Figure 4-2.
Longleaf pine (LL) : Longleaf pine with recruitment, hardwood understory,
and continuous herb cover.
Longleaf pine and hardwoods (LLHW) : Longleaf pine canopy without
recruitment, hardwood midstory, and discontinuous herb cover.
Longleaf pine and immature Sand pine (LLSP) : Longleaf pine canopy
without recruitment, sexually immature sand pine present, and some hardwoods.
Hardwood/Longleaf pine (HWLL): Hardwood and longleaf pine canopy
with no longleaf recruitment, sparse herb cover.
Sand pine/Longleaf pine (SPLL) : Sand pine and longleaf pine canopy with
no longleaf recruitment, hardwood mid-story, sparse herb cover.
Hardwood (HW) : Hardwood canopy, with sand pine possibly present.
Sand pine and Hardwoods (SPHW): Mixed co-dominant sand pine and
Sand pine (SP): Sand pine canopy, with hardwood midstory.
These vegetative states can be visualized as representing areas within the triangle
defined by the combination of longleaf pine, sand pine and hardwoods (Figure 4-3).
Fire and vegetation succession determine the transitions among these states. The
model represents fifteen different transitions among the eight states (Table 4-1). These
rules were derived from a series of workshops and discussions with people from: The
Nature Conservancy, the land managers at Jackson Guard on Eglin Air Force Base, other
longleaf pine managers, and ecologists. The general structure of these transition rules, for
example that unburned longleaf pine sites will be invaded by sand pine and hardwoods, is
well understood and has been reviewed in many publications (Heyward 1939, Platt et al.
1988, Rebertus et al. 1993). However, the specific fire frequencies that are necessary to
shift a site from one state to another are much less well known. What is certain is that these
frequencies vary with soil, topography and climate conditions. Therefore, the rules that
have been proposed to model Eglin Air Force Base will likely be subtly different from the
transition rules that would describe even neighboring stands of longleaf pine in different
soil conditions. For example, Blackwater River State Park located just north of Eglin Air
Force Base has less sandy soils than Eglin making it more difficult for fires to kill
hardwoods, among other differences.
Table 4-1. Vegetation transition rules used in the Eglin Longleaf pine model. Transitions
among vegetation states are based upon a sites time since fire (TSF) and the presence of
seeds from either or both pine species.
From LL LLHW LLSP HWLL SPLL HW HWSP SP
LL TSF>12 TSF>5
years years &
LLHW TSF<5 TSF>35
LLSP TSF< TSF>12
HWLL TSF<5 TSF>30 TSF
years years & >250
for last sand years
SPLL TSF<5 TSF
for last years
HW TSF<2 TSF<2
years & years &
S P TSF<2
The transitions among LL, LLHW, LLSP and the processes of sand pine and
hardwood invasion are relatively well understood. The transition rules describing the
conversion of HW, SP, HWSP, SPLL, and HWLL to sites with greater amounts of LL are
the least certain parts of the model. These transitions need to be further explored by field
experimentation. The transitions in the model are based upon the experience of land
managers at Eglin. There burning has been more successful at eliminating hardwoods than
in many other areas of Florida, however the amount of burning to decrease the hardwoods
in a HWLL site to the extent that it can be considered a LLHW is uncertain. In view of this
uncertainty, the model uses the same transition rule for both sand pine and hardwood based
upon the idea that it in the absence of evidence to the contrary these transitions will be
assumed to be equally likely. Below, I explain the derivation of these rules.
Longleaf pine (LL) will remain longleaf pine if it is frequently burnt. Without
frequent fire the longleaf pine will be invaded by hardwoods or sand pine, if sand pine
seeds reach the site. Hardwoods are assumed to be always present at low densities at
Eglin, however after twelve years without fire they reach a size that is sufficient to survive
future fires (Rebertus et al. 1993). Managers in Eglin have observed that sand pine can
quickly invade neighboring sites. If these sites are not burned within 5 years of sand pine
establishment, the sand pine reach sexual maturity and begin producing cones (Parker and
Hamrick 1996). However, if these sand pines are burned within twelve years of
establishment they can be killed, or at least have their growth slowed.
I, ( ,f Pine and Hardwoods
Forest in the longleaf-hardwood (LLHW) state requires three fires within five years
to be restored to a longleaf pine savanna state (LL). One or two fires, within a 3-5 year
interval, are not sufficient to eliminate hardwood trees from longleaf stands (Robbins and
Myers 1992). More fires are required, but how many and at what frequency was uncertain
and debated. For this model, three fires within less than 5 years of one another was settled
upon as a reasonable threshold.
Ii rfPine and Sand Pine
The mixed longleaf-sand pine (LLSP) state is composed of sexually mature, but
young sand pine trees. If this site is burned before the sand pine reach sufficient size the
sand pine will be killed and the site will return to being a longleaf forest (LL). However, if
the sand pine remain unburned for ten years, they will reach sufficient size to be more
difficult to kill with fire, and the site converts to the sand pine-longleaf (SPLL) state
(Provencher et al. 1998).
Hardwoods and I, .'l fPine
The hardwood-longleaf (HWLL) state can only be burned with difficulty. The
model assumes that some hardwoods can be killed with repeated fires. This assumption is
supported by field experience at Eglin Air Force Base (Provencher et al. 1998), however
the periodicity of fires that is necessary is not at all certain. Given this uncertainty, I
decided to require three fires within less than 5 years of one another to convert the site to a
mixed longleaf-hardwood state (LLHW). This transition rule is extrapolated from the
transition rule from LLHW to LL states. It is assume that since sizeable hardwoods are
being killed in the LLHW to LL transition, a similar fire frequency is required to kill some
of the hardwood in a HWLL state. This transition is reasonable, and consistent with field
experience at Eglin Air Force Base, but unsupported by any experimental studies.
However, such studies are on-going at Eglin Air Force Base.
Within an unburned hardwood-longleaf (HWLL) site, longleaf pine cannot
successfully regenerate. The longleaf pine in these sites will eventually die, shifting the site
to the hardwood (HW) state. Longleaf pine can live over four centuries, but most trees do
not (Platt et al. 1988). Over time, older trees are killed by lightning and other events. The
model uses the estimates that 250 years without fire are required to eliminate longleaf from
a hardwood-longleaf site.
Forest in the hardwood-longleaf state can be invaded by sand pine. If sand pine
seeds are available, and the site remains unburned for more than thirty years, these sand
pine accumulate and transform the forest to a hardwood-sand pine (HWSP) state. This rate
is based on observed changes in the Eglin landscape over the past fifty years (Provencher et
Sand Pine and In ..'lfPine
Forest in the sand pine longleaf (SPLL) state is difficult to bum. Fire is assumed
to be able to kill sand pine and convert a SPLL site to LLSP, but little is known about the
specifics of such a process (Provencher et al. 1998). I assumed that sand pine can be killed
by the same fire frequency that the model requires to reduce hardwood densities, which is
at least three fires with a frequency of less than five years. It is possible that even more
frequent fires, over a longer time are needed. Without fire, longleaf pine will persist along
with the sand pine, but will not regenerate. Without longleaf regeneration a site will
convert to pure sand pine. Longleaf mortality is assumed to be have completed the
transformation from a SPLL to a SP state after 250 years without fire.
Forest dominated by hardwoods (HW) can be invaded by longleaf or sand pine, but
only during specific short periods. Immediately after a fire, longleaf or sand pine can
invade a site if their seeds are present. Longleaf pine is assumed to be only able to invade
if the longleaf pine is producing a mast crop of seeds that year (Platt et al. 1988).
Hardwood Sand Pine
Hardwood-sand pine sites are stable, but without fire sand pine will not be able to
survive in the understory, and as with the hardwood-longleaf sites, after 250 years without
fire a hardwood sand pine site will return to a pure hardwood stand. This transition is
unimportant to the model, because it will almost never have the time to occur during a
Sand pine sites can be invaded by longleaf pine immediately after a fire, if longleaf
seeds are available. However unless fires continue to frequently occur, the site will remain
dominated by sand pine with a few longleaf
Fire and vegetation
The transitions rules described above are driven by the frequency of fire. Fire
frequency is itself influence by the time since last fire and the vegetation that is being
burned. This section describes how fire dynamics were represented in the model.
Ground fires in the forest of north Florida consume fine fuels, and kill the above
ground portion of understory and midstory vegetation, but they usually do not burn the
canopy trees of the forest. The top-killed understory vegetation is usually able to
regenerate very quickly. Within a year, or less in the case of spring fires, understory
vegetation is able regrow to such an extent that the accumulated fuel can once again
propagate fire. After a site has gone several years without fire, its combustibility begins to
decreases (Platt et al. 1988, Robbins and Myers 1992).
There are significant differences in the combustibility of different forest states,
however almost all forest states are most combustible in the years immediately following a
fire. Sites containing a greater proportion of longleaf pine are more combustible than sites
that contain less longleaf (Rebertus et al. 1989, Robbins and Myers 1992). Sand pine and
oak dominated sites also differ in their combustibility. Assuming a constant amount of
longleaf, sites with a significant sand pine component are less combustible than sites that
have significant hardwood species present (Provencher et al. 1998).
Vegetation State Combustibility
To quantify the relative combustibility of different vegetation types at different times
since fire, Eglin's fire managers were asked for their assessment of the relative
combustibility of different types of vegetation. While there are eight vegetation types,
several of these states were thought to have similar functional relationships between the
time since fire and vegetation combustibility (Figure 4-4).
Two sets of vegetation states were assumed to have probabilities of combustion that
were independent of the time since last fire. Sites with a significant longleaf component,
LL, LLHW, and LLSP, were assumed to have a constant high probability of combustion:
Pcomb,LL = 0.25.
While hardwood sites were assumed to have a constant low probability of
Two other sets of vegetation states were assumed to have probabilities of
combustion that decreased with the time since last fire. Hardwood-longleaf sites were
assumed to bum well if recently burned, but as time since fire increase their combustibility
decays. This decline in the probability of combustion of a site was represented as a
negative exponential function of time since fire, which eventually reaches the same low
probability of burning as a pure hardwood stand. Specifically, the equation used is:
Pcomb,HWLL(TSF) = P e(-r*TSF) Pcomb,HW
Where P0=0.25 and rHW =0.035.
SPLL, HWSP, and SP, are assumed to be easier to burn following a fire than
HWLL vegetation, but their combustibility decrease, at an increasing rate following after
several years. This relationship was represented by a negative logistic relationship.
Specifically, the equation used is:
Pcomb,sp(TSF) = (1 -1/ ( 1/PO+((K+PO)/(K*PO)) *e(-r*TSF)))/3
Where K=0.9, P0=0.25, and rsp=.125
These probabilities of fire combustion change at the time scale of years. There is a
substantial body of fire management literature on the seasonal effects on burning (Robbins
and Myers 1992). Seasonality of fire is not included in the Eglin longleaf model. There
are two reasons for this omission. The first is that there is a minimal effect of burning
season on longleaf pine (Glitzenstein et al. 1995). Secondly, the model is focussing upon
decadal-scale landscape dynamics rather than short-term consequences of a single fire.
This focus reduces the importance of seasonal variation in bum, since over the long-term
variation should approach a long-term average.
An integrated model of vegetation dynamics was be constructed (Figure 4-5). This
model represents the forest with two sets of variables. The first represents the vegetation
of the forest as one of the eight states (Figure 4-3). The second represents the fire history
of the forest by storing the time since last fire, and the frequency of previous fires. These
two variables describe the state of the forest. Every year, the vegetation transition rules are
applied to the forest (Table 4-4), and the forest vegetation may change state. This forest is
subjected to a fire regime that periodically attempts to ignite the forest. The success of
these ignition events is determined by the probability of combustion functions (Figure 4-4).
This non-spatial model was used to test the Out of Tune Forest, Transformed Forest, and
Invaded Forest caricatures by exploring the dynamics of vegetation under a broad set of fire
Succession and fire regime
Fire frequency determines what changes occur in the landscapes of north Florida.
Alternative visions of this relationship are embodied in the Out of Tune Forest and
Transformed Forest caricatures. The simulation model can be used to distinguish between
these caricatures. The key difference between these two caricatures is in the homogeneity
of a forest's response to fire. The Out of Tune Forest caricature proposes that specific
forest types correspond to specific fire frequencies, while the Transformed Forest
caricature proposes that fire suppression causes fundamental changes in the organization
and combustibility of a forest. These changes produce a hysteretic effect, because shifting
a forest from an unburned forest to a burned forest requires more fire than maintaining a
forest as a burned forest.
These caricatures were tested by experimentally assessing the impact of a wide
variety of fire frequencies on a variety of vegetation types. Each vegetation type was
burned at a range of fire frequencies, and its behavior observed over 100 years. Fire
frequency was controlled using the yearly probability of a site's ignition. The inverse of
the probability of ignition is a sites expected fire frequency. For example, the probabilities
0.01 and 1 correspond to average fire frequencies of 100 years and 1 year respectively.
Using a probability of ignition allows the time between fires at a site to vary, but still
correspond to an average fire frequency. These probabilities were varied from 0 to 1, in
0.01 increments. The state of a site after 100 years of burning was recorded. Five
hundred replicates of each vegetation type were exposed to each fire regime. From these
replicates the likelihood of a vegetation class of maintaining its existing vegetation, or
shifting to another vegetation type can be calculated.
These calculations were conducted for the vegetation types LL, LLHW, LLSP,
HWLL, SPLL, HW, and SP. The vegetation type HWSP was ignored, because it is not
present in the existing landscape, it occurs infrequently in the simulation model, and it does
not change in periods less than 250 years rendering its dynamics uninteresting over the
period of 100 years. Each of these sites was simulated with and without the dispersal of
longleaf or sand pine seeds into the site. These cases explore the impact of neighboring
sites on a site's dynamics. These results clearly show that for all fire frequencies a few
vegetation states are much more likely than others (Figures 4-6, 4-7, and 4-8).
These results reveal that some states are more likely than others across all fire
regimes, that the likelihood of vegetation states is strongly influenced by the presence of
external seed sources, and that the initial vegetation state influences model dynamics.
These results invalidate the Out of Tune Forest caricature, and support the Transformed
Likelihood of States
The model output shows that LL is the most likely state when fires are frequent, but
that any of HWLL, SPLL, SP or HW is likely to occur when fire is infrequent. Which of
these states occurs depends upon what a site's initial conditions are, and whether external
seed sources are present. If a site is initially in LL, LLHW, or HWLL and there is no
dispersal of sand pine the site is most likely to end up in either LL, if fires are frequent, or
HWLL, if fires are infrequent. There is a much lower likelihood of the site being in the
intermediate state of LLHW. Similarly, if the site is initially in LLSP or SPLL, or if there
is sand pine seed dispersal into LL, LLHW, or HWLL sites, then the site will likely end up
in LL, at high fire frequencies and SPLL, at low fire frequencies. Although, the situation
is more complex at intermediate fire frequencies on LLHW and HWLL sites that are being
invaded by sand pine.
Without seed dispersal HW and SP are completely stable states, regardless of the
fire frequency they will maintain themselves. However, their behavior becomes more
complicated when there is seed dispersal. Longleaf dispersal without sand pine is not
particularly important for any states except for SP and HW, because all the other states
already contain longleaf pine. However, in the case of HW it can be invaded by LL if a
longleaf seed source, but no sand pine seeds are present. However, longleaf is only
weakly able to invade HW, due to the low probability of a mast year coinciding with a burn
of a hardwood site. Sand pine can also be invaded by longleaf at high fire frequencies.
When both sand pine and longleaf seed sources are present, then longleaf is not able to
invade hardwood sites, being pre-empted by sand pine, but it still manages to invade sand
There are several ecological consequences of these transitions. Firstly, the
important transitions are from the initial states of LL, LLHW, LLSP, HWLL, and SPLL as
the other states (HW, SP, and HWSP) occur over minor areas of the landscape. In these
five states, three states are the most likely to occur. With frequent fires, LL dominates,
while infrequent fires cause HWLL or SPLL to dominate, with SPLL being dominant
when sand pine seeds are available.
Despite the fact that the same small set of states dominates the model's ecological
dynamics there are quite different dynamics depending upon the initial state of the model.
That is, different states respond differently to the same fire frequency. These differences
are clearly shown in Figure 4-9 and Figure 4-10. These figures plot the fire frequencies in
which a state has a greater than 50% chance of maintaining the same proportion of longleaf
pine. For example, if a site containing LL without seed dispersal is burned at a fire
frequency less than one fire every four and a half years, it will not be maintained as LL.
Similarly, if a HWLL site is not exposed to fires more frequently than once every six years
it will remain HWLL. Figure 4-10, reveals how adding longleaf and sand pine seed
dispersal alters the sites ecological dynamics, for example increasing the frequency of fire
needed to maintain LL and decreasing the frequency of fire needed to maintain SP and
HWLL. These graphs show that a similar fire regime has quite different effects upon
different vegetation types, and that there is no fire frequency that will maintain all
vegetation types. These results support the Transformed Forest caricature of Eglin Air
Force Base's ecological dynamics.
Comparing these seed dispersal and non-seed dispersal cases (Figures 4-6 to 4-10)
reveals that the strong influence that the presence of seed sources has on the response of
vegetation to fire. These results indicate that applying a specific fire frequency across a
landscape will produce variable results even in the same vegetation type, based upon the
input of other site's seeds into that site. This finding supports the Invaded Forest
caricature, by illustrating that the location of other forest types has a strong impact on
ecological dynamics. This support suggests that the spatial dynamics across the landscape
influence temporal dynamics, and need to be incorporated into management planning. To
do this, a spatially explicit version of the Eglin succession model was constructed.
Succession and Fire Spatial Dynamics
Spatial processes fundamentally influence the modeled ecological dynamics of Eglin
Air Force Base. Understanding the interactions of these processes over time required the
construction of a spatially explicit model that allows the spatial heterogeneity of the
landscape to influence fire and seed dispersal.
As discussed above, there are three spatial caricatures of Eglin Air Force Base that
this model had to address the Edaphic Forest, Divided Forest, and Invaded Forest
caricatures. To explore these caricatures the model had to include topographic variability,
fragmentation and seed dispersal along with the central processes of fire and fire
A forest landscape is represented as a matrix of sites. Each of these sites undergoes
the same successional dynamics as represented in the temporal model, but fire and seed
dispersal are explicitly modeled. Additionally, considering the landscape requires the
addition of some new landscape types to the model. I will now address each of these
components of the spatial model in more detail.
The spatial model divides the landscape into a matrix of sites, each of which is 60 m
on edge (Figure 4-11). This resolution was chosen for two reasons. Firstly, it is a
reasonable scale to develop management alternatives, displaying heterogeneity that is
important to management while not providing overwhelming detail (Jackson Guard: Eglin
Air Force Base 1993). Secondly, tree size appears have most of its variation at scales less
than 28 m (Platt and Rathburn 1993), making 60m an ecologically reasonable scale to
represent a portion of forest.
The forest is modeled using a set of overlying matrices that interact with one
another (Figure 4-12). The model stores information of vegetation, and fire history as does
the succession model, but spatial explicitness also requires a map of seed availability and
topography. Seed availability is generated by seed dispersal from existing sites, a process
that will be explained below. The topography layer is simply used to provide a template of
ecologically active and ecologically inert sites. This layer allows the presence of roads or
developed areas to block the spread of seeds and fires.
New model states
Considering the entire landscape requires a consideration of states other than those
that occur in Florida's sandhill communities, because if other states are intermixed with the
sandhill vegetation they can alter that vegetation indirectly by influencing the spread of
seeds and fire.
Two topographically limited states were added to the model to represent wet forest
sites. The sites are both wet forest, but one contains sand pine (WETSP) and the other
does not (WET). These forests are wetter than usual sandhill habitat occurring in creek
beds. Sand pine is assumed to be able to invade wet forests in the same fashion that it
invades hardwoods. The ability of fire to remove sand pine from this forest type is
uncertain, but Eglin's land managers think that it lies somewhere between the difficulty of
removing sand pine from LLSP and SPLL sites. The rules used in the model are shown in
Table 4-2. Vegetation transition rules used for the wet forest sites in the .
From WET WETSP
WETSP Two fires
Both WETSP and WET sites impede the spread of fire. WET sites are slightly
easier to bum than WETSP sites. WET sites are assumed to combust similarly to HWLL
sites, and WETSP sites are assumed to combust similarly to SPLL sites (Figure 4-4).
WET sites do not disperse any seeds, while WETSP sites produce sand pine seeds.
Controlled bums of hardwood-longleaf pine sites (HWLL) are difficult to manage,
because the forest is difficult to ignite. Conditions that allow burning can result in locally
intense fires that kill existing old longleaf (Robbins and Myers 1992). To explore the
effects of this process and help assess the degree of longleaf pine mortality due to particular
control burning strategies, a new state consisting of a relatively open stand of hardwoods
that includes a significant herb layer was added to the model. Without fire the hardwoods
on the site will grow larger and come to dominate the site, shifting into the HW state.
Longleaf seeds combined with frequent fires will allow longleaf to reestablish and convert
the site back to HWLL. At an intermediate fire frequency or in the absence of longleaf
seeds the site will remain HHW. These transition rules are shown in Table 4-3.
Table 4-3. Modifications of vegetation transition rules to include the herbaceous-hardwood
vegetation state in the Eglin Longleaf pine model. These additions were added to the
transition rules illustrated in Table 4-1
From LL LLHW LLSP HWLL SPLL HW HWSP SP HHW
HWLL TSF<5 TSF> TSF>2 1% chance
years 30 50 of
for last years years transition
three & sand during a
fires pine prescribed
HW TSF<2 TSF<2
years & years &
HHW TSF<6 TSF>1
years for 5 years
Herbaceous hardwood sites do not burn as well as longleaf sites, but they also are
not as difficult to burn as pure hardwood sites. The open structure of these stands allows a
combustible understory to develop. In the model, HHW sites are assumed to combust
similarly to HWLL sites (Figure 4-4).
The temporal vegetation model (Figure 4-3), is expanded by the addition of HHW,
and the wetland states, WET and WETSP. These states and the transitions among them
produce a spatial vegetation model (Figure 4-13).
The topographic layer added to the model was extremely simple. It simply defined
areas where vegetation dynamics could occur and areas that were outside the model. This
layer was used to define the boundaries of the simulation area, which in this case were the
boundaries of Eglin Air Force Base, and of the central watersheds of Eglin. It was also
used to define ecological boundaries within Eglin. Roads and developed areas block the
flow of fire and seeds across the landscape. The topographic layer maps these boundaries
among all the disparate layers of the model. The more subtle effects of regional topography
were implicitly included in the model through the wet forest states (WET, and WETSP) that
were discussed above.
Model Spatial Processes
The spatial dynamics of fire spread and seed dispersal are explicitly incorporated in
the Eglin Landscape Model. I will describe how each of these processes was represented
Seed dispersal is fairly simply represented with the landscape. While sand pine,
longleaf pine and hardwoods all disperse seeds only the dispersal of longleaf and sand pine
are explicitly modeled. Hardwood is assumed to be always present within a site, due to
resprouting as well as prolific seed production. Both sand pine and longleaf pine disperse
their seeds over short distances. The resolution of the model is 60m, and since most cones
disperse under 60m 70% under 20m (Boyer and Peterson 1983, Burns 1978), the model
only includes seed dispersal to the four nearest cells neighboring a site (Figure 4-14).
Longleaf pine seed production varies from year to year. Infrequently, about once a
decade, longleaf pine produces a large number of cones (Platt et al. 1988). The model,
somewhat conservatively, assumes that these mast events occur once a decade. In other
years no longleaf seed dispersal occurs. Sand pine does experience such fluctuations in
seed production. Unlike southern sand pine, the cones of the Choctawhatchee variety of
sand pine that are native to northwest Florida, do not require fire to open (Provencher et al.
1998), and therefore sand pine seed dispersal occurs continuously.
Wildfires and prescribed fire were both modeled in this landscape model, but not
necessarily at the same time. Wildfires produce their own spatial pattern through
interaction with vegetation, while land managers impose the pattern of prescribed bums.
Wildfires are ignited by lightning across the landscape, and then spread. The area a fire
bums is the product of chance and the existing landscape. Prescribed fire, or prescribed
bums, are different, because land managers chose an area of the landscape to bum, and
then attempt to burn it. However in prescribed bums fire may not spread into all parts of
the designated burn area, especially in areas that are difficult to ignite in large bums.
Wildfire was modeled as a process that propagates itself across a landscape. Every
year a number of sites are struck by lightning, and ignited. The lightning based fire
initiation rate is about one fire started/ 10 km2 (Henkel 1995, Platt et al. 1988), which is
equivalent to about 17 fire initiations in the central three watersheds of Eglin Air Force Base
every year. A burning site can spread fire to its neighbors. This process is probabilistic
and is dependent upon the vegetation type and time since last fire of the fire's neighboring
sites (Figure 4-15). Sites that fire spreads into can then ignite their neighbors. This
process continues until the existing burning sites fail to ignite any new sites. An example
of this process is provided in Figure 4-16. As fire spreads across the landscape it does not
burn homogeneously. Remenant patches of unbumt vegetation remain with a fire, and fire
spread is impeded by older, less combustible vegetation. The spread of fire can be halted
by roads, or diffciult to bum vegetation. An animation of wildfires in a simulation is
shown in Figure 4-17.
Prescribed burns, rather than wildfires, dominate the Eglin landscape. Prescribed
bums occur when land mangers go out and burn an area of the base. Weather, ignition
pattern, the skill of the bum team and other factors affect the results of any prescribed fire,
however these variations are outside the focus of this model.
Prescribed burns can be ignited both from the air and the ground. Aerial ignitions,
in which fire is dropped from helicopters, allow larger areas to be burned. Depending
upon the spacing of ignition sources and the distribution of fule fire may bums these large
areas unevenly. Ground ignitions are started by using drip torches to establish a fire front
that is allowed to bum over a plot and into another fire. Ground ignitions can provide more
control over fire spread, and may bum more evenly than aerial ignition. However, it is
difficult to start and control large fires from the ground.
The simulated large aerial bums and smaller ground bums. The aerial
bums are set to cover 24,000 acres (9 km2), while the ground burns cover 6,000 acres (2
1/4 km2). The probability of a site burning depends upon its vegetation type and time
since last fire as it does in a wildfire. However, in bum blocks it is assumed that each site
has multiple chances to be ignited. Aerial bums are assumed to provide three chances and
ground bums are assumed to provide four chances. These rates correspond to a fire front,
and an encircling fire respectively. These probabilities produce bum blocks that bum with
completeness similar to bum blocks in actual prescribed bums. When burning a hardwood
invaded longleaf site a ground bum will actually ignite about 75% of the area, while an
aerial bum will ignite about 60% of the area. Choosing which blocks to bum in which year
requires a management strategy, and this strategy can be varied within the simulation.
The ability of a state to persist despite disruption is termed resilience (Holling
1973). Ecological resilience is a measure of the amount of change or disruption that is
required to transform an ecosystem from being maintained by one set of mutually
reinforcing processes and structures to a different set of processes and structures.
The concept of resilience can be illustrated by modeling an ecological 'state' as the
position of a ball on a landscape (Figure 4-18). Gravity pulls the ball downwards towards
pits in the surface. These pits are stable states. The deeper a pit the more stable it is,
because increasingly strong disturbances are required to move an ecological state away
from the bottom of the pit. The steepness of the sides of a stability pit corresponds to the
strength of negative feedback processes maintaining an ecosystem near its stable state.
These are the forces that transform the ecosystem, back towards a stable state following a
disturbance. The width of a pit is the range of change that a state can experience before
transforming to another state. The width of the stability pit corresponds to the ecological
resilience of a state (Figure 4-18).
The concept of resilience can be adapted from the analysis of a system of
continuously varying variables to a system that experiences transitions among discrete
states. In the ball and landscape illustration above, the state of the system was represented
by a single variable. The use of multiple variables is possible, but the analysis and
visualization of the results becomes increasingly complicated. In the discrete case the use
of more than two states is similarly complicated, unless the states are arranged along an
axis and transitions do not jump across intervening states. By dividing the multiple
possible states in two states, the resilience of discrete states can be simply and clearly
assessed by analyzing the likelihood of transitions between states.
In the spatial Eglin model these states are defined on basis of their longleaf density
as 'poor' (P) and 'good' (G). Two states produce four likelihood ratios, because each state
can either remain in its state or switch to the other state. Because a state can either remain
in that state or shift to the other state, these likelihood can be written as:
PtoG + PtoP = 1.0 and
GtoG+ GtoP = 1.0.
The probability of a state remaining within that state (i.e., Ptop and GtoG)
corresponds to the resilience of that state, as this probability represents the chance that the
state will persist given a particular environmental regime. It is important to note that the
probabilities of changing from poor to good, and from good to poor are not necessarily
This model can be applied to data by gathering multiple observations of a system
and calculating the likelihood of the four types of transition. For example, the transition
likelihood given in a table can be converted into a resilience ball and landscape model by
assigning the likelihood of remaining in a state to the width of a state, and the likelihood of
leaving the state as its potential energy. The states are then connected by barrier to state
change that is equal in height to the probability of state change:
Prob. Of State Change = 1.0 the likelihood of remaining within the state.
An example table and diagram are shown in Figure 4-19.
This approach of assessing resilience is observationally based. It does not attempt
to analyze the underlying dynamics of a system. Rather it takes observed state-transitions
and applies the concept of resilience to them. If this process is inappropriate this method
will reveal that the states) have no resilience by producing a flat line. This observational
approach does not easily allow results that are collected from a system under one
environmental regime to be applied to the system under another environmental regime.
However, by focusing on transitions between defined states, this method does provide a
method of analyzing the integrated response of a system to a given environmental regime
that is based upon management criteria.
The state-transition resilience approach can be applied to landscapes. The spatial
distribution of the likelihood of state transitions can be assessed by classifying an initial
landscape into two states, and then repeatedly applying a environmental regime to that
landscape and recording the post-treatment landscape states.
This approached was applied to the Eglin Air Force Base landscape by classifying
Eglin's vegetation-based land management goals. At Eglin managers wish to expand the
area in longleaf pine savanna. I used this goal to classify states as either being 'poor' or
'good' longleaf sites. I classified LL, LLSP, LLHW, and WET as good longleaf sites. All
the other sites were considered poor LL. By running the simulation model many times, the
likelihood of a given site on the landscape changing states can be assessed and plotted on
the landscape. This measure not only allows one to assess the stability of a given
landscape formation, but also measure the type of change. This approach was used to
compare the response of the Eglin landscape to multiple fire management strategies.
The Eglin Landscape Model was tested in a number of fashions. Quantitative tests
of the model were impossible, due to a lack of comparable data. However, because this
purpose of this model is not to predict the future landscape dynamics of Eglin, but rather to
explore the consequences of different sets of assumptions, this lack of quantitative
comparison is not a serious failing. The model's dynamics were tested against its ability to
reproduce historical landscape dynamics, and its behavior was tested against the
expectations of Eglin's land managers. There were two main tests of the model's
credibility: its ability to reproduce and maintain the historical Eglin landscape, and its ability
to reproduce hardwood and sand pine invasion into longleaf sites. These tests were
conducted using the concept of landscape resilience, which I will briefly outline, before
discussing the tests themselves.
The landscape of northwest Florida was in the recent past dominated by longleaf
pine savanna (Schwartz 1994). The Nature Conservancy used a variety of historical and
ecological sources to attempt to reconstruct the landscape of Eglin Air Force Base (Kindell
et al. 1997, Provencher et al. 1998). This reconstruction did not have sufficient detail for
the Eglin Landscape Model, so a number of further assumptions on the distribution of sand
pine and hardwoods dominance had to be made before a historical landscape could be
produced for the model. This historical landscape, which I extrapolated from The Nature
Conservancy's work, is shown in Figure 4-20.
Historical Proposition: If the model was accurately mimicking the natural processes
of north Florida, wildfire should be able to maintain this landscape in a shifting
steady state fairly close to this reconstructed historical condition.
This proposition was tested by running the model for one hundred years starting
from the initial conditions of the estimated historical landscape (Figure 4-20). This model
used the estimated historical fire initiation rate, and the vegetation transition rules described
above. Ten model runs were used to assess the resilience of the landscape to the historical
The landscape resilience of the historical landscape to wildfire is shown in Figure 4-
21. Most of the landscape area is maintained as either good longleaf or poor longleaf. The
areas of degradation, by sand pine spread, are relatively minimal and since the initial
distribution of sand pine was itself largely unknown, some variation is to be expected.
Therefore, the model successfully maintained the estimated historical landscape.
Fire suppression test
A second test of the model's functioning is its ability to recreate the invasion of
longleaf by hardwoods and sand pine following landscape fragmentation and fire
suppression. We used the following proposition:
Suppression Proposition: Fragmenting the landscape and suppressing fire for a
similar time in the simulation model should produce a situation qualitatively similar
to the existing landscape.
Fire suppression began about fifty years ago. Therefore, the current road network
fragmented the historical landscape, and fire was suppressed to one fifth of its original
level. The model was allowed to run for fifty years, at which time the simulated landscape
was compared to the Eglin Air Force Base's present landscape. An animation of vegetation
changes during 100 years of fire suppression is shown in the Figure 4-22.
The model landscape successfully reproduces the invasion of sand pine out of
creeks, and the limited amount of longleaf pine on the landscape. The vegetation of Eglin
Air Force Base has been fragmented by an undocumented, complex set of logging, tree
planting, and road cutting that makes quantitative comparison between the simulation and
the existing Eglin landscape unreasonable, because these processes are not included in the
model. To test the ability of the model to reproduce larger scale changes in the Eglin
landscape, the landscape dynamics predicted by the model were shown to and discussed
with Eglin's land managers. They felt that the landscape produced by this method
accurately captured the speed of hardwood and sand pine invasion of longleaf pine. While
this test is not a strong one, it is still a test, because the model could have failed to produce
reasonable dynamics. Passing this test does not validate the model, rather it improves the
model's credibility. However, the vegetation pattern produced after fifty years of fire
suppression was considered to be enough a reasonable representation of the current
landscape, that this simulated landscape could be used to represent the current landscape for
the comparison of alternative management scenarios. To distinguish this simulated
landscape from the actual landscape, I refer to it as the 'current' landscape (Figure 4-23).
This model was developed not to predict the future of Eglin Air Force Base, but
rather to serve as a tool to integrate and test existing ecological knowledge. Part of its
utility has been to identify important gaps in that understanding. The Eglin Landscape
Model embodies the consensus ecological understanding of Eglin's managers. By
observing the dynamics of the model and comparing these dynamics to their tacit feeling for
how the landscape functions, managers are presented with an opportunity for learning.
Conflict between expectations and model dynamics means that either the model needs to be
revised, or that the managers understanding of the landscape needs to be changed. In
either case, learning occurs.
The Eglin Landscape Model passed through a series of workshops, in which it was
used, discussed and revised. This process lead to a model in which the land managers had
some confidence, due to their feeling that the modeling accurately captured their
understanding of the forest, and it produced realistic forest dynamics. This confidence
allow the model to be used as a tool to investigate alternative land management strategies.
Spatial Dynamics Management Options
Ecosystem management has been ongoing since the early 1990's. Burning the
forest using prescribed fire has been one of the main components of this activity. This
burning is motivated by both ecological and safety concerns. Ecologically, prescribed
bums are conducted to attempt to increase the area of longleaf pine forest. Additionally,
they inhibit fuel accumulation reducing the risk of uncontrollable wildfire.
The main purpose of modeling the Eglin landscape was to explore and challenge the
ideas underlying management. In this sprit, the model was used to determine the potential
consequences of continuing the current of wildfire suppression without also conducting
prescribed bums. This test was conducted by applying the policy of fire suppression to the
'current' landscape (Figure 4-23). Under the policy of fire suppression the 'current'
landscape is quite unstable. The landscape continues to lose areas of longleaf pine, and
experience sand pine invasion.
Calculating the landscape resilience of fire suppression on the 'current' landscape
reveals that a continued policy of fire suppression would lead to the degradation of large
areas of the base, and the maintenance of longleaf in only a few small areas (Figure 4-24).
Given the management goals of expanding the area in longleaf, this response is not
acceptable. For this reason land managers have been applying prescribed fire to the
landscape and logging areas of sand pine and hardwood that they wish to restore.
Management Vegetation Classification
Alternate management polices were evaluated, and prescribed bums implemented,
based upon a simplification of vegetation into three classes. These classes divided the
vegetation types into three states. Because, an increase in the area of longleaf pine is a
major management goal at Eglin Air Force Base, these three classes rank vegetation as
either being 'best', 'good' or 'poor' based upon the amount of longleaf pine a vegetation
state contains (Figure 4-25).
Together the 'good' and the 'poor' class encompase all the land that can be
potentially occupied by longleaf pine. The 'best' class is a subset of the 'good' class. The
'best' sites are longleaf pine savanna (LL). All the sites in which longleaf pine is a
dominant component (LL, LLHW, and LLSP) are 'good' sites. All the other sites that
could include longleaf are considered 'poor sites' (HWLL, SPLL, HW, HWP, SP and
HHW). The vegetation types WET and WETSP are excluded from these divisions as they
do not contain and cannot change to contain longleaf pine.
Land managers would like to convert the 'poor' sites into 'good' sites, and improve
the 'good' sites into 'best' sites. However, in meeting these goals they also would like to
ensure that they at least maintain the existing area of 'good' sites.
Managing Prescribed Bums
Prescribed bums are difficult to organize. They require trained fire managers, and
prescriptions require specific weather conditions. Eglin managers estimate that ecological
and social conditions are appropriate for burning about 80 days a year. Optimistically, they
suggest that they could conduct two bums per day during about half of these burn days.
These estimates produce a high estimate of 120 prescribed bums a year.
As discussed above, prescribed burns can be ignited from either the ground or the
air. Managers estimate that prescribed bums would be evenly split between aerial and
ground ignition each year. These estimates provide an upper bound of 60 ground and 60
aerial fires per year. The landscape area being used in the simulation model corresponds to
a third of Eglin, which means that the maximum number of prescribed bums in the
modelled area is 20 aerial bums and 20 ground burns per year.
These estimates of the number of prescribed bums that could be applied a year are
optimistic and do not account for extreme weather, military missions or crises in
management such as budget problems or understaffing. Therefore it is useful to observe
what effect lower rates of prescribed burning would have on the Eglin landscape.
Additional, applying prescribed bums is expensive. If fewer fires could restore the
landscape effectively, spending money on over-burning the landscape would not be
sensible. To explore a range of prescribed bum levels, I ran the model with 20, 15, 10,
and 5 pairs of aerial and ground fires a year.
Using fire to manage a landscape requires a strategy for deciding when and where
to bum. Part of the purpose of developing a simulation model is help develop such
strategies. Using the model to evaluate land management strategies offers a simple way to
test ideas about management ideas that combine both social and ecological models and
to locate important uncertainties and unintended consequences.
Based upon discussions with land mangers I developed four sets of land
management strategies. As with most of this model, these strategies were not meant to be
realistic, but rather simple charicatures of different management approaches. Three of these
approaches use prescribed fire to manage the landscape. These approaches are named
'maintenance', 'conversion' and 'rotation'. The fourth approach uses wildfire, and is
The 'fire' management strategy is used as a base case, against which to compare the
other strategies. The 'fire' strategies do not use prescribed fire, therefore they cannot be
applied at the same frequencies as the prescribed fire strategies. Rather, wildfire is
conducted at two levels the historical rate of wildfire initiation and the current suppressed
rate of forest fires.
The simplest management strategy that uses prescribed fire is 'rotation.' This
strategy divides the landscape into bum blocks. Each block is burned with the same
frequency. This frequency is determined by the number of blocks burned per year divided
by the total number of burn blocks.
Both the 'maintenance' and 'conversion' strategies alter the fire frequency of a bum
block in response to how the block's vegetation compares to the remainder of the
landscape. Both these strategies vary the fire frequency of bum blocks depending upon the
vegetation within a block. These strategies counts the amount of 'best', 'good', and 'poor'
sites in a bum block and then use a formula to rank the blocks. The 'conversion' and
'maintenance' use different forumulas. Each year bum blocks are burned in the order their
ranking, until the available burns for the year are complete. Sites that have been burned
within the last three years are not returned. The 'converison' strategy aims to convert poor
sites to good sites, while also maintaining good sites. The maintenance strategy trys to
maintain the good sites.
The rule used for the 'conversion' strategy is:
Score = Good + Poor 2/3 + Best 3/5
The rule used for the 'maintenance' strategy is:
Score = Good Best/3
These alternative management strategies were used to probe the dynamics of the
managed Eglin landscape. The purpose of these probes is to test the caricatures discussed
at the beginning of this chapter. While these simulation exercises cannot be expected to
predict the details of specific management regime, they can identify which caricatures
appear to better represent the functioning of the sandhill communities of north Florida, and
where there are important gaps in ecological understanding.
Four different management strategies were tested: 'maintenance', 'conversion',
'rotation' and 'wildfire.' These strategies were each tested at several different fire
frequencies. Prescribed bums were tested at levels of 20, 15, 10, and 5 pairs of large and
small bums per year. While the 'wildfire' strategy was tested at 'historical' and
'suppressed' levels. In combination a total of fourteen different management regimes were
tested. The combinations of fire frequency and management strategies tested are shown in
Table 4-5. Alternative management regimes combining yearly bum amount and fire
Fire Types Fire Strategy
Aerial and Wildfire
Ground Initiations Rotation Maintenance Conversion Wildfire
20,20 X X X
15,15 X X X
10,10 X X X
5,5 X X X
S Historical X
The success of these management alternatives was evaluated based upon the
changes they produced in the landscape over fifty years. The pattern of change produced
by each management strategy has a complex temporal dynamic (Figure 4-26). These
experiments, however, focus upon the spatial aspect of northwest Florida forest. To
simplify the analysis of changes in the spatial distribution of vegetation a fixed time period
of fifty years was used for analysis. The 'future' landscape at the end of these fifty years
was compared with the 'present' landscape, and changes calculated.