Spatio-Temporal Connectivity in Dynamic Tropical ... · Spatio-Temporal Connectivity in Dynamic...
Transcript of Spatio-Temporal Connectivity in Dynamic Tropical ... · Spatio-Temporal Connectivity in Dynamic...
Spatio-Temporal Connectivity in Dynamic Tropical
Fragmented Landscapes
by
Alexandre Camargo Martensen
A thesis submitted in conformity with the requirements
for the degree of Doctor of Philosophy
Department of Ecology and Evolutionary Biology
University of Toronto
© Copyright by Alexandre Camargo Martensen 2017
ii
Spatio-Temporal Connectivity in Dynamic Tropical Fragmented
Landscapes
Alexandre Camargo Martensen
Doctor of Philosophy
Department of Ecology and Evolutionary Biology
University of Toronto
2017
Abstract
Expanding human occupation on the planet reduces and fragments native habitats, threatening
biodiversity, ecosystem functioning, and services. Tropical regions have recently been
experiencing unprecedented amounts of forest conversion and fragmentation. As tropical forests
harbor a large fraction of the world’s biodiversity, their loss and fragmentation has spearheaded
the sixth mass extinction. Nevertheless, the tropical regions experience unique low intensity
land-use and bioclimatic characteristics that result in highly dynamic forested landscapes. These
dynamical landscapes, when subjected to the current scenario of intense global change, poses
particular challenges for biodiversity conservation in human-modified landscapes. This thesis
provides insights towards (i) the development of new metrics to quantify landscape dynamics;
(ii) the assessment of the effects of land-use intensification on spatio-temporal dynamics and
connectivity; and (iii) the quantification of potential drivers of these changes in the spatial
dynamics.
In the first part of my thesis, I developed a graph-theoretical method that incorporates the
spatial dynamics of the landscape in the evaluation of landscape connectivity. I tested this
method using a large set of Atlantic Forest landscapes of Brazil. In the second part of the thesis, I
evaluated the effects of different drivers of landscape spatial dynamics, particularly focusing on
land-use intensification alongside its economic and social drivers. My results pointed to the fact
that land-use intensification reduces spatio-temporal dynamics of landscapes, as a large fraction
iii
of the land is locked up into one or a few intensified land cover types and the proportion of land
abandoned to native habitat regeneration is low.
Taken together, my findings have two broad impacts: (i) the new spatio-temporal indices
reveal insights about landscape connectivity missed by purely spatial connectivity indices; and
(ii) land-use intensification is happening across the globe, independent of the agricultural
commodity that is being produced, reducing spatial dynamicity, which will lead to a decline in
connectivity. Therefore, more land should be spared for biodiversity conservation in more highly
intensified landscapes. Both findings have direct implications for spatial planning for
conservation.
iv
Acknowledgments
This work would not have been possible without a supportive group of collaborators, friends and
family. I want to thank, first and foremost, three people who were particularly vital for the
conclusion of my PhD. Without the unconditional love, support, patience, dedication and
partnership of Kika and Cauê, none of this would have been possible. For their sacrifices, I will
be eternally grateful. Also to my mentor Dr. Marie-Josée Fortin, who has taught me so much
well beyond academic knowledge. Her example as an advisor has gone far beyond what I
expected, which has set the bar really high for mentorship. No words can express my gratitude
for how comprehensive she was with aspects of my personal life and the way that she welcomed
me, Kika and Cauê. I am also thankful to my committee members, Dr. Don Jackson and Dr. Ben
Gilbert, for their kind support, guidance and encouragement. I am very grateful for the help of
Dr. Santiago Saura, who embraced my research, illuminated my ideas and helped me to format
my results as papers. I am thankful for the long standing partnership with Dr. Milton Cezar
Ribeiro who has been influential since the initial stages of this work, including the generation of
the Atlantic Forest dataset used in the second and third chapters. Also many thanks to Kate Kirby
for helping me in different phases of this study. The Connaught International Scholarships for
Doctoral Students provided financial support for my work and a Discovery Grant and CRC Tier
1 awarded to my advisor, Dr. Fortin. Additional funding was provided by the Department of
Ecology and Evolutionary Biology at the University of Toronto. I was privileged to have worked
with a great group of friends in the LeLab, all of whom contributed in some way to this work.
Huge thanks to Colin, Amanda, Andrew, Chris Edge, Chris Blackford, Stephanie, Stephen,
Lanna, Kate, and many others who helped me translate what I have written to English. Enormous
thanks to Amanda, Colin, Paul, Kate, Ilona, Lanna, Aaron, Andrew, Alex, Cassidy, Carina,
Korryn, Flávia, Claudia, Iñaki, Jennifer, Anna, Henrique, Fernando, Jonathan, Emily and many
v
others who made my life in Toronto much happier and easier. Finally, I want to thank my Mom,
Dad and brother for their constant support and encouragement.
vi
Chapter Acknowledgements
This thesis is composed of five chapters divided as follow: three analytical chapters (Chapters 2
to 4), that are either in review or in preparation for submission; one introductory one (Chapter 1)
that provides the basis for the dissertation; and a final closing chapter (Chapter 5) that presents
some conclusion remarks, as well as future research avenues that could be a starting point for
forthcoming inquires. The experimental design, analyses, and manuscript preparation were all
carried out by myself. Co-authors of the chapters contributed on conceptual discussions (Saura
and Fortin), maps datasets (Ribeiro), expertise in computer programming (Saura), and editing of
written materials (Saura and Fortin).
1. Martensen, A.C., Saura, S. and Fortin, M.-J. (In review) Spatio-temporal connectivity:
Assessing the amount of reachable habitat in dynamic landscapes (Chapter 2)
2. Martensen, A.C., Saura, S., Ribeiro, M.C. and Fortin M.-J (In prep.) Land-use intensification
constraints spatio-temporal connectivity of fragmented tropical forest landscapes (Chapter 3)
3. Martensen, A.C., Saura, S., and Fortin, M.-J. (In prep.) Forest and land-use dynamics in the
Amazon: Convergent effects of human activities (Chapter 4)
vii
Table of Contents
Abstract ........................................................................................................................................... ii
Acknowledgments.......................................................................................................................... iv
Chapter Acknowledgements .......................................................................................................... vi
Table of Contents .......................................................................................................................... vii
List of Tables ...................................................................................................................................x
List of Figures ............................................................................................................................... xii
- Spatio-temporal dynamics of tropical forest .............................................................................1
General introduction ............................................................................................................1
Research questions and thesis roadmap ...............................................................................4
- Spatio-temporal connectivity: Assessing the amount of reachable habitat in dynamic
landscapes ...................................................................................................................................8
Abstract ................................................................................................................................8
Introduction ..........................................................................................................................9
Methods..............................................................................................................................10
2.3.1 Spatio-temporal landscape networks .....................................................................10
2.3.2 Modelling movement in dynamic landscapes ........................................................11
2.3.3 Metrics of spatio-temporal habitat reachability .....................................................13
2.3.4 Case study in the Atlantic Forest ...........................................................................15
2.3.4.1 Model parametrization .............................................................................16
Results ................................................................................................................................17
Discussion ..........................................................................................................................19
Appendix ............................................................................................................................34
- Land-use intensification constraints spatio-temporal connectivity of fragmented tropical
forest landscapes .......................................................................................................................45
Abstract ..............................................................................................................................45
viii
Introduction ........................................................................................................................47
Methods..............................................................................................................................48
3.3.1 Spatio-temporal model ...........................................................................................48
3.3.2 Metrics of habitat reachability in the spatio-temporal network .............................49
3.3.3 Model parametrization ...........................................................................................50
3.3.4 Study region ...........................................................................................................51
3.3.5 Sampled landscapes ...............................................................................................52
3.3.6 Statistical analysis ..................................................................................................52
Results ................................................................................................................................53
3.4.1 Spatio-temporal connectivity variation from 1990 to 2007 ...................................53
3.4.2 Proportion of native habitats and land-use intensity on spatio-temporal
connectivity ............................................................................................................54
Discussion ..........................................................................................................................61
Appendix ............................................................................................................................64
- Forest and land-use dynamics in the Amazon: Convergent effects of human activities ........69
Abstract ..............................................................................................................................69
Introduction ........................................................................................................................70
Methods..............................................................................................................................73
4.3.1 The study region ....................................................................................................73
4.3.2 Drivers of land-use change ....................................................................................74
4.3.3 Spatial patterns of forest and pastures losses, gains and stability ..........................76
4.3.4 Spatio-temporal dynamics evaluation: losses, gains and stability .........................76
4.3.5 Socio-economic drivers of landscape dynamics ....................................................77
Results ................................................................................................................................79
4.4.1 Patterns of forest loss, regeneration/gain and stability ..........................................79
4.4.1.1 30% of class (pasture or forest) cover .....................................................79
ix
4.4.1.2 70% of class (pasture or forest) cover .....................................................80
4.4.2 Patterns of landscape dynamics .............................................................................85
4.4.3 Drivers of landscape dynamics ..............................................................................87
Discussion ..........................................................................................................................93
Appendix ............................................................................................................................97
- Conclusions ...........................................................................................................................100
Thesis summary ...............................................................................................................100
Future research directions ................................................................................................107
Bibliography ................................................................................................................................110
x
List of Tables
Table 2.1: List of variables and keywords. ................................................................................... 30
Table 2.2: Movement possibilities along temporal connections between source (t1) and
destination (t2) nodes, not considering the spatial constraints. A value of 1 indicates that such
movement is possible at some moment within the analyzed period. A value of 0 indicates that it
is not possible from t1 to t2. Values of 0.5 indicate that the movement is possible given some
assumptions on the co-occurrence of nodes in time. Temporal movement possibilities are
directional (asymmetric) from t1 to t2 (source to destination). ...................................................... 31
Table 2.3: Metrics description and equations. All metrics can be calculated for a spatial-only
model (denoted with the suffix s) or for the proposed spatio-temporal model (denoted with the
suffix st). ....................................................................................................................................... 32
Table 2.4: variation of the percentages of habitat amount in the 200 landscapes at the three
landscape sizes. ............................................................................................................................. 35
Table 2.5: Absolute values of habitat loss and gain in hectares, and percentages of habitat loss
and gain as a function of habitat amount in t1. .............................................................................. 36
Table 3.1: Medians and standard deviations among treatments of each of the spatio-temporal
connectivity metrics (ECAst: Equivalent connected area and the PCst fractions. ....................... 54
Table 3.2: For the runs with habitat regeneration with the same quality of the regenerated habitat
the PCst were better explained by the following models. ............................................................. 56
Table 3.3: Results for the models when considering all native habitats and only the initial and
intermediate native habitats (* represents cases where a negative influence was observed, but
largely influenced by only one landscape).................................................................................... 58
Table 3.4: Land-use contribution for each of the two first PCA axes, the proportion of variance
and the cumulative variance.......................................................................................................... 66
Table 4.1: Results of the class metrics summarized by the median obtained across the 11 counties
for each category and period. ........................................................................................................ 82
xi
Table 4.2: Results of the PCfractions (PCintrast, PCdirectst and PCstepst) for each county per
period. ........................................................................................................................................... 86
Table 4.3: Selected models (Δ AICc < 2) explaining the variation of PCnumst, PCintrast,
PCintrast %, PCdirectst, PCdirectst %, PCstepst, PCstepst %, for each explanatory variable
importance for each model selection processes, and if it has a positive or negative relationship. 90
xii
List of Figures
Figure 2.1: Spatial (solid lines) and spatio-temporal connectivity (dashed arrows). The grey solid
lines in t2 represent patch locations at t1. In (a) the letters represent the isolated populations of a
given species with a particular dispersal capacity. Population A is connected at t1, since both
patches are within the species dispersal capacity. The same happens for population D at t2.
However, A and D are considered isolated when t1 and t2 are analysed separately (i.e. without
accounting for temporal connections). In (b), although the patches have different sizes and
species compositions (different dark grey geometric shapes) at t1, the spatial aspects of t1 do not
affect their biological composition at t2. When accounting for both spatial and temporal
connections, in (c) a given individual, represented by the star, could be in the left fragment at t1,
and in the right fragment at t2, but not the other way around (from right to left, temporal
directional connection). Additionally, population A, present in the left and central fragments in
t1, became isolated in the left patch at t2, but is mixed with population B in the central and right
patches at t2, as represented by AB. In (d), the large patch in t1 could provide to the small patch
in t2 more species than an already small patch in t1 can do, as represented by the different width
of the dashed arrow, and by the different dark grey geometric shapes. ........................................ 23
Figure 2.2: Spatial and spatio-temporal connectivity. (a) Purely spatial connections, (b) Spatio-
temporal direct movements, (c) Spatio-temporal stepping-stones movements, and (d) entire
connectivity pattern including both direct and indirect movements. The hollow polygons at t2
represents the polygons that were lost. ......................................................................................... 25
Figure 2.3: Contribution of the spatio-temporal connectivity ECAst compared to the purely
spatial connectivity ECAs at t2 (100(ECAst / ECAst2)-100); (a) density functions of the
contribution of the ECAst as a function of ECAs at t2. Positive values represent a positive
influence of the spatio-temporal connectivity over the purely spatial connectivity in t2.
Negative/zero values represent cases where either there was no influence of spatio-temporal
connectivity, or the increase in the purely spatial connectivity in t2 was so huge, that any increase
in connectivity caused by the spatio-temporal metrics was surpassed by the purely spatial
connectivity at t2. (b) The linear models of the percentage of the increment given by ECAst
compared to ECAs at t2 for all dispersal capacities. ..................................................................... 26
xiii
Figure 2.4: Contribution of ECAst compared to ECAs t2 (100(ECAst / ECAst2)-100) as a
function of the amount of habitat in t2. ......................................................................................... 27
Figure 2.5: PC fractions independent of dispersal capacities. (a) PCdirects in t1, t2 and PCdirectst
in the spatio-temporal model; (b) PCintras in t1, t2 and PCintrast; and (c) PCsteps in t1, t2 and
PCstepst......................................................................................................................................... 27
Figure 2.6: PC fractions contributions according to dispersal capacity (50 and 1000 m) in t1, t2
(for the spatial-only PCs) and in the spatio-temporal model (PCst). ............................................ 28
Figure 2.7: PCst fractions (PCdirectst, PCintrast and PCstepst) contribution as a function of the
percentage of habitat loss for 50, 200 and 1000 m dispersal distances. ....................................... 29
Figure 2.8: Contribution of the spatio-temporal connectivity ECAst compared to the purely
spatial connectivity ECAs at t2 (100(ECAst / ECAst2)-100). Density functions of the contribution
of the ECAst as a function of ECAs at t2 according to species dispersal capacity (50, 100, 200,
500 and 1000 m) and landscape size (25000, 50000, 100000 ha). Positive values represent a
positive influence of the spatio-temporal connectivity over the purely spatial connectivity in t2.
Negative/zero values represent cases where either there was no influence of spatio-temporal
connectivity, or the increase in the purely spatial connectivity in t2 was so huge, that any increase
in connectivity caused by the spatio-temporal metrics was surpassed by the purely spatial
connectivity at t2. Medians are shown. ......................................................................................... 37
Figure 2.9: The linear models of the percentage of the increment given by ECAst compared to
ECAs at t2 based on the differences in habitat amount for all landscape sizes (25000, 50000 and
100000 ha) and dispersal capacities (50, 100, 200, 500 and 1000 m). The betas are shown. ...... 39
Figure 2.10: PC fractions for the 25000 and 50000 ha landscapes. For the 25000 ha landscapes
(upper panels): PC fractions in t1 (a), t2 (b) and spatio-temporal (st) (c); (d) PCdirect in t1, t2 and
spatio-temporal (st); (e) PCintra in t1, t2 and spatio-temporal (st); and (f) PCstep in t1, t2 and
spatio-temporal (st); For the 50000 ha landscapes (lower panels): PC fractions in t1 (a), t2 (b) and
spatio-temporal (st) (c); (d) PCdirect in t1, t2 and spatio-temporal (st); (e) PCintra in t1, t2 and
spatio-temporal (st); and (f) PCstep in t1, t2 and spatio-temporal (st). .......................................... 40
xiv
Figure 2.11: PC fractions contributions for different dispersal capacities in t1, t2 and spatio-
temporal for the landscapes with 25000 ha................................................................................... 41
Figure 2.12: PC fractions contributions for different dispersal capacities in t1, t2 and spatio-
temporal for the landscapes with 50000 ha................................................................................... 42
Figure 2.13: PCst fractions contribution (PCdirectst, PCintrast and PCstepst) as a function of the
percentage of habitat loss for three dispersal distances (50, 500 and 1000 m) for landscape sizes
of 25000 and 50000 ha. ................................................................................................................. 43
Figure 2.14: Contribution of the PC fractions for species with larger dispersal capacities in the
100000 ha landscape. .................................................................................................................... 44
Figure 3.1: The responses for all dispersal distances together for the variation in the quality of
the regenerated habitat. ................................................................................................................. 64
Figure 3.2: Changes in land-use through time of the studied landscapes. The top panel shows the
59 landscapes in the three times evaluated, the black lines unit the landscapes in t1 (1990) and t2
(2000), whereas the red line between t2 (2000) and t3 (2007). The bottom panel showed the
proportion of change in terms of proportion in the PCA axis 1 (black) and PCA axis 2 (red)
between t1-t2(circles) and t2-t3 (crosses). ....................................................................................... 67
Figure 4.1: Histograms of the influence of the spatio-temporal connectivities over the purely
spatial ones [((spatiotemporal/purely spatial)-1)*100], for forests and pastures per period. The
dashed lines represent the comparisons between the spatio-temporal and the first year of purely
spatial metrics, whereas the solid line represents the comparisons among the spatio-temporal and
the second year purely spatial metrics. ......................................................................................... 89
Figure 4.2: Barplot of the number of patches, mean patch area (ha), maximum patch area and %
of the counties occupied by each class. Dark grey: 30% of forest in the landscape; and light grey:
70% of forest in the landscape. ..................................................................................................... 99
1
- Spatio-temporal dynamics of tropical forest
General introduction
Species are not distributed evenly across the globe (Pianka 1966), but rather present a clear
pattern with a greater number in the tropical regions, both in terrestrial and in aquatic realms
(Lomolino et al. 2010). Tropical forests cover less than 10% of the terrestrial surface, but harbor
around 50% of the terrestrial species (Wilson 1988; Mayaux et al. 2005), and even more extreme
percentages are observed for tropical marine ecosystems (Reaka-Kudla 1997). Although this
pattern has been known for more than three centuries, the driving processes are still unknown,
and a growing number of competing hypotheses are still being discussed (Brown 2014). While
tropical regions were less modified through human history (FAO 2010; Goldewijk et al. 2011),
tropical forests became the main source of new agricultural land in recent decades (Laurance &
Bierregaard Jr. 1997; Mayaux et al. 2005; Gibbs et al. 2010), and 50% of global forest losses
occurred in the tropics in the latter years (Hansen et al. 2013). This is obviously of concern for
biodiversity conservation, given the disproportional biodiversity that occurs in the tropical
regions (Laurance, Sayer & Cassman 2014). Indeed, native habitat destruction, fragmentation
and degradation are the main causes of the global biodiversity crises (Rands et al. 2010; Pereira
et al. 2010), and today over 50% of tropical forests are already lost, and the remaining forests are
in most cases severely fragmented (Haddad et al. 2015).
Astonishingly, there is little evidence of species extinctions in the tropics due to the
reduction and fragmentation of the native habitats (Brown & Brown 1992; Heywood & Stuart
1992; Sodhi et al. 2010). This has motivated a large body of research that have studied some
tropical areas that have somewhat longer histories of land-use and degradation, such as the
Atlantic Forest of Brazil (Brown & Brown 1992; Dean 1996). These tropical regions have
therefore, become important laboratories to understand the potential long-term effects of land
change in the tropics, mainly the ones related to species extinctions (Brown & Brown 1992). One
usual approach to investigate species loss, is to relate the amount of habitat loss with the rates of
species extinctions by reversing the species-area curve (Simberloff 1992). This practice to
predicts species extinction (e.g., Pimm & Askins, 1995; Pimm et al., 1995; Brooks et al., 1999a,
2002; Pimm & Raven, 2000; Hanski et al., 2013; Rybicki & Hanski, 2013), almost invariably
overestimates the actual observed species extinctions (Heywood et al. 1994; He & Hubbell 2011,
2
2013). The detection of these overestimations generated an intense debate in the literature, where
different arguments were opposed. Some have argued that there are mathematical reasons for
these differences (He & Hubbell, 2011, but see Pereira et al., 2012). Others suggested that
species traits, for example, pre-adaptation to patchy environments given topographic
characteristics or even extinctions that have occurred unnoticed might also have a role on these
divergences, although are less supported by empirical evidence (Brown & Brown 1992). One of
the most supported hypothesis is that there is a time-delay of species becoming extinct following
deforestation (Tilman et al. 1994; Hanski & Ovaskainen 2002). Therefore, given the tropical
degradation in recent history, landscapes could be in a non-equilibrium period before extinctions,
called “relaxation time” (Diamond 1972). In these cases, species are not extinct yet, although
they are “committed” to extinction, as shown by the many species described in the IUCN red list
(Heywood et al. 1994).
This time-lag to extinction has been studied in different ways (e.g.: Tilman et al., 1994;
Brooks & Balmford, 1996; Brooks et al., 1999; Hanski & Ovaskainen, 2002; Ferraz et al., 2003;
Metzger et al., 2009; Korfanta et al., 2012; Wearn et al., 2012; Uezu & Metzger, 2016). The
general trend is that after an initial great loss of species, due to many factors associated with the
habitat loss, such as proximity to edges, and disturbance frequency, the remaining number of
species would follow a steady reduction over the years (Krauss et al. 2010). For instance, many
species or taxa that are long-lived, or can survive in resistant life-cycle stages, such as long-lived
trees, would take longer to “pay the debt”, than short-lived species, such as small mammals and
frogs (Metzger et al. 2009; Hylander & Ehrlén 2013). Dispersal is also a key aspect in species
maintenance in fragmented landscapes, since current (Martensen, Pimentel & Metzger 2008) and
past landscape connectivity patterns (Lindborg & Eriksson 2004) have both shown to be
influential in species distribution. Therefore, the path to relaxation, which is the pattern dynamics
in which a system changes from one state to another, is still largely unknown (Malanson 2002,
2008; Wearn, Reuman & Ewers 2012). Nevertheless, the relaxation path can be initially
described based on two factors, the differences between before and after equilibria in terms of
species numbers, and the dynamics of perturbation and resistance forces (Malanson 2008;
Hylander & Ehrlén 2013). Yet, the magnitude of the extinction debt is also influenced by the
dynamics of perturbation and resistance forces (Malanson 2008). For instance, high levels of
connectivity in fragmented landscapes can act as a resistance force preventing extinctions
(Hanski & Ovaskainen 2000, 2002).
3
Moreover, tropical fragmented forested landscapes are particularly hyperdynamic
(Laurance 2002). Forest loss and regeneration occur rapidly and at intense rates (Hansen et al.
2013). All these alterations result in a scenario where, while irreplaceable for biodiversity
conservation, mature forests continue to shrink (reviewed in Gibbs et al., 2010). However,
second-growth forests have expanded in many places (Chazdon 2003, 2008; Aide & Grau 2004;
Mayaux et al. 2005), notably Puerto Rico (Rudel, Perez-Lugo & Zichal 2000; Grau et al. 2003),
Costa Rica (Meyfroidt, Rudel & Lambin 2010), and large regions in other tropical countries
(Rudel, Bates & Machinguiashi 2002; Grau et al. 2003; Aide & Grau 2004). While habitat loss
can result in decreasing landscape connectivity, habitat regeneration could reconnect habitat
patches, and these processes happen simultaneously in landscapes (Hansen et al. 2013).
Therefore, landscapes have been transformed worldwide into dynamic anthropogenic mosaics
with patches of forests scattered in different successional states, which has profound impacts on
several ecological processes and species persistence (IPCC 2001; Hanski 2011).
One way to mitigate landscape fragmentation is to restore landscape connectivity.
Landscape connectivity is the degree in which landscapes facilitate or reduce organismal
movements (Taylor et al. 1993), and it affects gene flow (Coulon et al. 2004), population and
metapopulation dynamics (Wiegand et al. 1999, Baggio et al. 2011), community structure
(Martensen et al. 2008), ecosystem functioning (Fischer et al. 2006), and ecosystem services
(Mitchell et al. 2013). Connectivity is the integration of the physical structural connectivity of a
landscape with each species own functional response to the spatial layout of habitat, known as
functional connectivity (Tischendorf & Fahrig 2000; Bélisle 2005; Urban et al. 2009). A broad
array of methods and approaches are used to measure connectivity (Rayfield, Fortin & Fall
2011), such as those that account for presence or absence of corridors, corridor configuration,
presence and density of stepping-stones, distance between patches, contagion, percolation,
matrix permeability and probability of moving between patches (Kindlmann & Burel 2008). All
these metrics consider a vast array of potential connectivity aspects in static landscapes
(Moilanen & Hanski 2001; Kindlmann & Burel 2008).
Among the many potential frameworks to analyse landscape connectivity,
graph/network-theoretical approaches are of increasing interest (Urban et al. 2009; Dale & Fortin
2010; Blonder et al. 2012) and use (Borrett, Moody & Edelmann 2014). Network theory
provides a framework to analyze network topology and flow under different circumstances (Fall
et al. 2007, Dale & Fortin 2010) such as: in dendritic networks, like riverine and cave systems
4
(Peterson et al. 2013); in 2-D landscapes, in ponds (Ribeiro et al. 2011) and in forest fragments
(Minor & Urban 2007); or generating hypothesis for testing genetic relationships (Dyer & Nason
2004). However, network dynamics, i.e., changes in spatial characteristics over time is
considered a complex mathematical process, which is pointed as a reason for not being widely
considered (Blonder et al. 2012). Thus, a general gap in all connectivity studies is that they
usually consider landscapes as static entities ignoring the effects of landscape dynamics that is
particularly crucial in the highly dynamic tropical landscapes.
The high spatial and temporal heterogeneity in the tropics is facilitated by fast forest
regeneration (Aide et al. 2000), given both climatic characteristics (Chazdon 2014) and low
intensity in land uses (Jakovac et al. 2015). In fact, land-use in the tropics is historically of low
intensity (IPCC 2001), which confers an overall high resilience (Chazdon 2014). However, in the
last few decades, market pressures are driving agriculture towards intensification in an attempt to
increase crop yield (Phalan et al. 2011). In addition, land-use intensification has been claimed as
the best available option to meet the growing human demands for food, at the same time
avoiding expansions into native habitats (Green et al. 2005; Phalan et al. 2011, 2016; Foley
2011). Moreover, the increasing demands for biofuels and fibers productions add pressure for
agriculture intensification (Assessment 2005; Ragauskas et al. 2006; Tscharntke et al. 2012).
Land-use intensification is associated with increasing amounts of fertilizers, pesticides, irrigation
(Foley et al. 2011; Mueller et al. 2012), and reductions in spatial heterogeneity and temporal
dynamics, causing direct effects on population dynamics (North & Ovaskainen 2007; Fahrig et
al. 2011), threatening biodiversity (Sodhi et al. 2004), and ecosystem services (Tscharntke et al.
2012).
All these aspects place tropical regions in a pressing and fascinating time for studying
spatio-temporal dynamics effects on biodiversity. The current trends in global changes,
following climatic alterations and land-use dynamics will alter spatial patterns directly
influencing biodiversity patterns and processes with potential impacts in species conservation.
Research questions and thesis roadmap
In this context, the main objective of my thesis is to develop a model framework that allows me
to consider spatial and temporal relationships between landscape features in highly dynamic
5
tropical landscapes. It has been suggested that landscape legacies may account for a large
fraction of contemporary biological patterns (Brooks & Balmford, 1996; Brooks et al., 1999a;
Metzger et al., 2009; Ewers et al., 2013; Uezu & Metzger, 2016). However, the temporal
interactions among landscape features and the rates of spatial changes through time are largely
ignored. This research will contribute to the basic understanding of ecological dynamics in
highly dynamic tropical fragmented forest landscapes and can support management decisions in
fragmented forested landscapes. To address these objectives, I ask several questions in three
analytical chapters.
Chapter 2 presents a novel model framework for evaluating spatio-temporal connectivity
dynamics and to calculate metrics to evaluate the spatio-temporal influences of landscape
dynamics that could be compared to purely static ones, in order to evaluate the influences of
landscape dynamics. Moreover, this chapter also tests these metrics in real-world situations, and
evaluate influences of landscape extent and species dispersal capacity in their behaviour. Finally,
this chapter compares spatial metrics to the new spatio-temporal metrics developed here by
evaluating which circumstances spatio-temporal metrics could greatly contribute to the
understanding of the spatial patterns.
Based on the model framework and metrics developed in the first chapter, in the
subsequent chapters, I evaluate drivers of landscape dynamics, such as land-use intensification,
economy, and time since land-use establishment. In order to do that, I used two different study
systems. In chapter 3, my study system encompasses over 2 million hectares in the south of
Bahia and a small portion of the east of Minas Gerais, in Brazil. The forest in the region is one of
the most diverse and richly endemic places on Earth (Thomas et al. 1998; Martini et al. 2007). It
is also an emblematic region of the Atlantic Forest, as it encompasses the first spot that the
Europeans arrived and settled in Brazil. Therefore, as a unique aspect compared to other tropical
regions, this region has a long history of degradation, that started with the selective logging of
Pau-Brasil (Caesalpinia echinata, Dean, 1996) in the 1500s. In the beginning of the 18th
century, cacao (Theobroma cacao) was introduced into the region, and the plantations expanded
quickly (CEPLAC - http://www.ceplac.gov.br/). However, cacao is mainly planted in agroforest
schemes associated with the native forests. Later, after the middle of the twentieth century, the
forests and the shaded cocoa plantations were largely converted to pastures and to agricultural
fields (Thomas et al. 1998). Since 1990, a fast expansion of highly intensified Eucalyptus
6
plantations occurred, which today, covers more than 10% of the area (Ribeiro et al. 2012).
Therefore, the region is unique for studying tropical systems fragmentation, since it has a longer
history of human occupation than most tropical systems. Additionally, it has been experiencing a
process of land-use intensification for the last three decades, which is a trend that many tropical
areas are now experiencing. Hence, using three different maps from the same region in different
years, which covers the period before land-use intensification (t1=1990), during the most intense
period of land-use intensification (t2=2000/2001), and after land-use intensification (t3=2007), I
evaluate how land-use intensification alters spatio-temporal patterns of native forested habitats.
Additionally, I assess how these changes influence animal species, as a function of their dispersal
capacity and habitat requirements. Finally, I evaluate the trade-offs of forest dynamics, and
therefore opposing habitat-aging effects to the spatio-temporal connectivity that a highly
dynamic landscape can experience.
In chapter 4, I evaluate the spatio-temporal connectivity dynamics in different
Amazonian counties. As an opposite scenario compared to chapter 3, this study system is
composed of newly occupied areas. However, these counties represent diverse histories in terms
of land-use and economic pressures, and therefore, an important study system to evaluate these
driving forces over landscape dynamics. The main goal in this chapter is to assess how much
influence the long establishment of land-use influences landscape dynamics, as well as the
relevance of the global economic crisis of 2008. Long-standing occupied areas are expected to be
less dynamic than newly occupied areas. Therefore, counties that are occupied for a long time
and are severely deforested long ago would be less dynamic than counties that have the same
amount of forest, but where, this reduction happened more recently in time. Finally, landscape
dynamics in counties that are long established would be less variable as a function of the global
crisis of 2008 compared to those that are more recent occupied. With the contrasting study
systems from chapters 3 and 4, I intend to gain knowledge about some of the most important
driving forces of landscape dynamics, and how they can influence spatio-temporal connectivity,
and therefore, different ecological processes and patterns.
Taken together, these three chapters address the relevance of the temporal directional
interactions among landscape features, associated with their spatial interactions. There is a
pressing gap of knowledge that needs to be filled, as tropical regions are immense reservoir of
species, and a large number of extinctions are forthcoming. The knowledge obtained in my thesis
7
could help to manage tropical forest fragmented landscapes in order to prevent at least some of
them from happening. A conclusion (chapter 5) with my main findings closes the thesis, also
pointing for future directions in the study of spatio-temporal connectivity.
8
- Spatio-temporal connectivity: Assessing the amount of reachable habitat in dynamic landscapes
Abstract
1. Landscape heterogeneity and connectivity affect species movements, and therefore, they
play an important role in determining the likelihood of species persistence, as well as diversity
patterns. However, landscape connectivity is usually evaluated using static snap-shots, which do
not account for the sequential interactions among habitat patches through time.
2. We developed a network-based model of landscape dynamics and corresponding
connectivity metrics to account for the reachable habitat across space and time. We illustrate the
behaviour of these metrics using fragmented forested landscapes in the Atlantic Forest of Brazil,
parametrizing the models based on the dispersal capacities of selected bird and small mammal
species.
3. We found that, by considering spatio-temporal links, connectivity is estimated to be on
average 30% higher (up to 150% higher) than what is estimated from purely spatial models. This
higher degree of spatio-temporal connectivity arises due to connections through temporal
stepping-stone patches that appear (habitat gain) and disappear (habitat loss) over time. Species
with short dispersal distances (< 1000 m) particularly benefited from the spatio-temporal
connections. The contribution of spatio-temporal connectivity to habitat reachability increases
with higher habitat loss rates. Moreover, it depends on the amount of habitat in the landscape,
being higher at intermediate habitat amount (~30%).
4. We showed that accounting for spatio-temporal connectivity is critical for understanding
ecological patterns and processes in dynamic landscapes, and that a series of purely spatial
connectivity metrics underestimates actual connectivity patterns across time. Changes in climate
and land-use are altering landscape dynamics, potentially causing additional threats to
biodiversity conservation in fragmented landscapes. The proposed spatio-temporal connectivity
approach and metrics can be applied to evaluate the effective connectivity patterns and trends in
a variety of dynamic landscapes, avoiding the potential overestimates of population isolation and
extinction probabilities that may result from widely used spatial-only connectivity models.
9
Introduction
Landscape connectivity, i.e., the degree to which landscape heterogeneity affects organism’s
dispersal, directly influences species movement, and therefore modulates gene flow (Coulon et
al. 2004), affecting populations, communities and ecosystems (Mitchell, Bennett & Gonzalez
2013). Connectivity measurements have received great scientific attention, and a broad array of
methods and approaches have been used to support its evaluation (Rayfield, Fortin & Fall 2011).
However, the influences of landscape connectivity on ecological processes and subsequent
patterns are generally evaluated using only static snap-shots (Moilanen & Hanski 2001;
Kindlmann & Burel 2008; Claudino, Gomes & Campos 2015), which do not capture temporal
interactions among habitat patches that occur in many rapidly changing landscapes (Hanski
2011).
One of the most promising and integrative approaches for evaluating landscape connectivity
is the development and application of methods based on network (graph) theory (Urban et al.
2009; Dale & Fortin 2010; Blonder et al. 2012). Network-theory has been suggested as a good
practical tool to asses connectivity, because it is more informative than simple landscape metrics,
yet less demanding in terms of biological data than individual-based or metapopulation models
that require movement and/or demographic data (Bodin & Norberg 2007; Fall et al. 2007).
However, using network dynamics to capture changes in spatial characteristics over time is a
mathematically complex process, and as a result, these methods are poorly developed (Blonder et
al. 2012). To date, the ecological impacts of changes in landscape connectivity have been
determined by comparing spatial connectivity analyses performed independently at multiple
points in time (e.g., Metzger et al. 2009; Saura et al. 2011; Bommarco et al. 2014) ignoring the
effects of temporal interactions among habitat patches and the rates of spatial changes through
time (Figure 2.1).
Here, we propose a novel spatio-temporal network approach and corresponding metrics for
quantifying both spatial and temporal connectivity in an integrated fashion. The proposed
approach calculates the amount of habitat that can be reached through both spatial and temporal
connections, and provides new metrics that are directly comparable to purely static metrics that
have been widely used in previous studies (e.g., Saura & Rubio 2010; Saura et al. 2014 and
10
citations therein). We illustrate the behaviour of these metrics by evaluating a large number of
fragmented forested landscapes in the Brazilian Atlantic Forest. Specifically, we assess their
behaviour as a function of habitat amount, rate of habitat change (loss and gain), size of the
analyzed landscapes, and species dispersal capacities. Finally, we highlight the differences in the
amount of estimated connectivity based on spatial versus spatio-temporal connectivity models.
We discuss the potential implications of this novel spatio-temporal connectivity approach
towards improving our understanding of the ecological patterns and processes that occur in
dynamic landscapes.
Methods
2.3.1 Spatio-temporal landscape networks
Using a network approach, habitat patches are represented as nodes, and their potential direct
connections as links or edges (Urban & Keitt 2001); landscape dynamics can then be
characterized through changes in both nodes and their links. Changes in nodes may include
losses of entire patches or of parts of patches (shrinkage), patch enlargement, creation of new
patches, or changes in their habitat quality. These changes in nodes can also translate, depending
on species dispersal abilities, into connectivity gains or losses because of the changes in links
between patches, including changes in the distances between patches and in the availability of
intermediate stepping-stones that facilitate movement between patches. Even without changes in
habitat patches (nodes), links can vary, for example, because of changes in land-use between
patches, which can facilitate or impede movement across the matrix, or as a function of seasonal
changes, such as floods and droughts, which can temporally connect and disconnect water
bodies.
Spatio-temporal connectivity is composed of two features: (1) the spatio-temporal paths
i.e., a sequence of links that can be used to move between two nodes in a network and (2) the
spatio-temporal legacy (Figure 2.1). As spatio-temporal legacy has been widely discussed (see
reviews in Kuussaari et al. 2009; Hylander & Ehrlén 2013), we here only consider the spatio-
temporal paths. Spatio-temporal paths require the consideration of both spatial and temporal
11
links among patches. A spatio-temporal link across a network represents the possibility of an
individual moving from a given habitat location at time t1 to a different habitat location at a later
time t2 (see acronyms and definitions in Table 2.1). From a biological perspective, spatio-
temporal paths can be used to calculate the probability that an individual will survive from t1 to t2
in a dynamic landscape, particularly when the patch that holds the individual in t1 does not exist
in t2 (habitat loss). In addition, spatio-temporal paths can be used to calculate the probability of a
particular individual reaching a given location in t2 from t1, and thus, allowing for individual and
gene flow, and hence, connectivity between populations.
2.3.2 Modelling movement in dynamic landscapes
Given two dates (t1 and t2) all patches in a landscape can be classified into one of the following
types:
Stable: habitat in t1 and in t2.
Loss: habitat in t1 but not in t2.
Gain: not habitat in t1 but habitat in t2.
We assume that no more than one type of habitat change occurs between t1 and t2. In other
words, the data are measured frequently enough to avoid back-and-forth changes (e.g., habitat
loss and gain) for a location within the same timestep, yet at a temporal resolution long enough
to capture changes between timesteps.
Based on the above classification of patches, we create a network model in which nodes
corresponding to each of the patches are assigned a type of Stable, Loss or Gain. For two patches
to be considered connected habitat over space and time, a path across the network must exist
between the patches where the starting node of the path in t1 should be of type Stable or Loss and
the final node of the path in t2 should be of type Stable or Gain.
The spatio-temporal links among patches can be of two forms (Table 2.2, Figure 2.2):
Direct movements, which correspond to a movement using a single link from a patch
with habitat at t1 (node of type Loss or Stable) to another patch with habitat at t2 (node of type
Stable or Gain), without passing through other intermediate stepping-stone patches.
12
Indirect or Stepping-stones movements, which correspond to movements from a patch
with habitat at t1 to another patch with habitat at t2 that involves multiple links (steps) through
one or more intermediate stepping-stone patches. There are two types of movements that are
partial in the sense that, for being successful, they would need to be combined with another
previous or subsequent movement step. In the first type, an individual moves to a location with
habitat at t1 and at tx (t1<tx<t2), but with no habitat at t2 (hence of type Loss). In this case, the
individual will need to move somewhere else in an additional movement step to be made before
t2. An example of this type would be a movement from a node i of type Loss to a different node j
of type Loss, and from node j then to a node k of type Gain, where k can be reached from j but
not from i because i is too far from k based on the dispersal abilities of the focal species (here j
acts as a stepping stone allowing final arrival to k as a result of two movement steps). In the
second type of partial movement, an individual makes a movement from a location of type Gain
in time tx where there is habitat at tx and t2 but not at t1. In this case, the original starting point for
the individual movement at t1 cannot be a patch of type Gain, as habitat did not exist there in t1.
Therefore, the individual had to be in some location other than Gain at t1, and hence a previous
movement (direct or stepping-stone) from some other initial location must have occurred
between t1 and tx (e.g., from Loss or Stable directly to Gain, or from Loss or Stable to Gain
through another intermediate stepping-stone). By considering these two types of stepping-stone
movements, we can account for all possibilities of spatio-temporal connectivity between patches
through one or more stepping-stones, while also acknowledging that not all combinations of
stepping-stone movements will lead to a successful movement that allows for individual
survival.
A movement from node i to node j is considered possible from a temporal perspective
(value of 1 in Table 2.2) when i and j simultaneously exist in the landscape at some time tx; this
is the case for movement from Loss to Stable, or from Gain to Stable. A movement from Loss to
Gain or from Gain to Loss may or may not be possible, depending on when the losses and gains
occur for the different patches, i.e. they may exist or not simultaneously at some time tx. For this
reason, these movements are given a likelihood value of 0.5, although any value between 0 and 1
may be given according to particular cases (Table 2.2).
13
In addition, the possibility of movement in a temporal perspective (e.g. from Loss to
Stable) does not mean that such movement is possible from a spatial perspective. If two nodes
are too far apart from each other based on the dispersal abilities of the focal species, or are
separated by a hostile land-use that acts as a barrier, the movement between the nodes will not be
possible even if the patches exist at the same time. Hence, both the spatial and temporal
constraints (possibilities) for movement are considered and integrated in the connectivity model.
The spatial probabilities of connectivity are obtained from the combination of the species
dispersal abilities with the distance between patches. This distance between patches could be of
any kind, such as the Euclidean distance, or a more complex cost-weighted distance accounting
for matrix heterogeneity and resistance. Note that these spatial probabilities for movement, as
given by dispersal kernels (e.g. negative exponential functions), will decrease when larger
distances need to be traversed (either as a result of a long direct movement or of the combination
of several stepping-stone movements) to reach a node j from a node i. In the model, the spatial
probabilities for movement are multiplied by the temporal probability for movement (0, 0.5 or 1,
Table 2.2) to obtain the final spatio-temporal links and their associated movement probabilities.
Finally, we need to consider the possibility of movements from nodes Gain or Stable at tx
to some other different node Gain or Stable at t2. These movements are not strictly necessary for
survival given that the individual at tx is already in a location with habitat that will remain so at
t2. However, the model accounts for the entire set of nodes that can be reached by moving
through the network (directly or indirectly). In other words, even if an individual can move to a
particular Gain or Stable node, this does not exclude the possibility that it also may be able to
reach, with some probability (even if lower given a longer distance), other node of types Gain or
Stable in the network.
These rules form a spatio-temporal model that corresponds to a directed network with
asymmetric links, since even if movement from i to j may be possible, it does not imply that
movement from j to i is also possible.
2.3.3 Metrics of spatio-temporal habitat reachability
Given our network model of spatio-temporal connectivity, we now generalize and adapt two
existing habitat availability (habitat reachability) metrics, namely the Probability of Connectivity
14
(PCs) and Equivalent Connectivity (ECs) (Saura & Rubio 2010; Saura, Bodin & Fortin 2014), to
account for both the spatial and temporal dimensions. Henceforth, we use PCst and ECst to refer
to our new spatio-temporal metrics, and PCs and ECs for the standard spatial-only metrics; note
that the values of the spatio-temporal metrics here proposed are directly comparable with those
obtained from the purely static analyses (Table 2.3).
These metrics express connectivity as the amount of reachable habitat resources in a
landscape. They account for both the habitat resources (e.g., habitat area) that can be reached
within the patches (intrapatch connectivity) and for the habitat resources that can be reached by
moving to other patches through the links in the network (interpatch connectivity). Intrapatch
spatio-temporal connectivity occurs when individuals can survive by staying in the same Stable
patch from t1 to t2. Interpatch spatio-temporal connectivity occurs when an individual moves
from a Stable or Loss node in t1 to a different Stable or Gain node in t2.
Given two point locations (source i and destination j) randomly selected within the
landscape (i in t1 and j in t2), PCst is defined as the probability that i and j fall into habitat areas
that are spatio-temporally connected so that it is possible for an individual located in i at t1 to
move to j at t2. PCst is hence the sum of the probability corresponding to the intrapatch
connectivity (i and j falling within the same habitat patch, and that patch being of type Stable)
and the interpatch connectivity (i and j falling into different but connected patches).
ECst is defined as the amount of resources (e.g., habitat area, nesting spots) of a single
Stable habitat patch (existing throughout t1 to t2) that would provide the same probability of
spatio-temporal connectivity (PCst) as the network composed by the multiple Loss, Gain and
Stable habitat nodes of the landscape. EC is denoted as Equivalent Connected Area (ECA) if
habitat area is used as the attribute of the nodes in the network, as we will do hereafter. ECAst
gives the effective area of habitat that individuals would be able to reach in the spatio-temporal
network, and is calculated as the square root of the numerator of PCst (see Table 2.3 for details).
In addition, the spatio-temporal connectivity PCst can be divided into three fractions:
PCintrast, PCdirectst and PCstepst, which are here expressed as percentages of PCst (Table 2.3).
Each of these fractions quantifies a different contribution to the spatio-temporal connectivity of
the landscape. PCintrast corresponds to the intrapatch connectivity (amount of reachable habitat
15
within stable patches), PCdirectst corresponds to the interpatch connectivity provided by
complete direct spatio-temporal connections between patches (without using intermediate
stepping-stone patches), and PCstepst corresponds to the interpatch connectivity provided by
indirect connections made possible by stepping-stone nodes between the source and destination
nodes (see Saura & Rubio 2010; Saura et al. 2014 for details of the fractions for the purely
spatial metrics).
To calculate PCst, the three PCst fractions, and ECAst we: (1) account for the types of
movements (i.e. direct and indirect) and connections among patches (Table 2.1); (2) define two
‘duplicated’ nodes in the network for Gain and Loss habitat patches, because in some cases, such
patch types can be an initial starting point (at t1) or a final destination (at t2) for a direct
movement, and in other cases they may only act as stepping-stone at time tx in a multi-step
movement among other patches; (3) calculate the intrapatch connectivity for the stable patches,
as it is possible that an individual that is in a stable habitat patch at t1 remains in the same patch
until t2; and finally (4) combine the potential movements in the temporal dimension with the
spatial constraints for movement. Spatial constraints are determined, for example, by the
combination of the Euclidean or effective (cost-weighted) distance between patches and the
species dispersal ability. Calculations are performed by combining an R script with a command
line version of the Conefor software package adapted to spatio-temporal directed networks (both
available as supplementary material).
2.3.4 Case study in the Atlantic Forest
To demonstrate the proposed network approach and metrics, and to explore their behavior in
dynamic landscapes with different amounts of habitat and rates of land-use change, we used a
spatial dataset from a 2 million hectares region in the northeast of Brazil. The area was originally
covered by Atlantic Forest, one of the most fragmented and species rich biomes of the world
(Myers et al. 2000). We analyze forest habitat changes between 1990 and 2001 in 200
landscapes of 25,000, 50,000 and 100,000 ha (see Supplementary Material for additional details).
16
2.3.4.1 Model parametrization
In order to keep our example simple, we used patch size in hectares as the node attribute for
calculating PCst, PCs and related metrics. However, users can select any other attribute, such as
population size, habitat quality, or patch area weighted by habitat quality. Here, we selected five
median species dispersal distances: 50, 100, 200, 500 and 1,000 m. A negative exponential
function of interpatch distance was used to obtain the probability of direct movement between
any pair of patches (although other dispersal kernels could be also used in the model). The
function was parameterized so that it gave a 0.5 probability of movement (gap crossing) between
patches when the patches were separated by an edge-to-edge Euclidean distance equal to the
considered median dispersal distance. For the tropical biomes (Moore et al. 2008) and the
Atlantic Forest in particularly, few studies are available for birds and small mammals, but they
suggest that the bulk of species have dispersal capacities below 200 meters across gaps, with a
few species able to cross gaps of many hundreds of meters (Crouzeilles et al. 2010). Therefore,
with the selected dispersal distances, we are covering a large range of potential dispersal abilities
for species in the region. Nevertheless, to further explore the behaviour of the different metrics
for larger dispersers, we also used medians of 2500, 5000, 7500 and 10000 m for dispersal
distances for a subset of the landscapes (n=5). We used Euclidean distances and hence treated the
non-forest matrix as homogeneous in order to keep the illustrative case study simple, but
resistance surfaces and cost-weighted (effective) distances between patches could also be used.
We also built nine linear univariate models to evaluate the effect of the following
landscape characteristics on the PCst fractions: (1) habitat amount in t1, (2) habitat amount in t2,
(3) amount of stable habitat, (4) the net difference in habitat amount, (5) the amount of habitat
that was lost, (6) the amount of habitat that was gained, and (7) the proportion of area that was
lost, as well as (8) gained and (9) the proportion of the differences between total habitat amount
in t1 and t2, based on the amount of habitat in t1. Finally, we used the model that was best
supported among those outlined above to evaluate how PCst fractions respond to these changes.
17
Results
Spatio-temporal connectivity (ECAst) is approximately 30% higher than spatial-only
connectivity in t2 (ECAs), reaching close to 150% in some cases (Figure 2.3), with a slight
influence of species dispersal capacity (Figure 2.3b). The additional contribution of spatio-
temporal connectivity is not much affected by landscape size, but is slightly lower in smaller
landscapes (Figure 2.8). The increase in ECAst compared to ECAs t2 is higher with larger
amounts of habitat change (Figure 2.3). Longer-dispersing species are less influenced by
reductions in habitat amount than shorter dispersers (Figure 2.3b). Even landscapes with a stable
net habitat (similar gains and losses) show increases in ECAst over ECAs t2 of around 10%. The
higher value of ECAst over ECAs t2 holds even for landscapes that have a net habitat gain of
around 5% (Figure 2.3b). How much greater ECAst is when compared to ECAs t2 depends on
habitat amount, with a peak at around 30% of habitat, independent of species dispersal capacity
(Figure 2.4).
The contribution of PCstep and PCintra to total connectivity varies as a function of the
three analyzed scenarios (purely spatial ones for t1 and t2 and spatio-temporal one), whereas
PCdirect represents around 20% of the total PC in all the three scenarios (Figure 2.5). PCintra is
lower and PCstep higher for t1 and st, whereas for t2 is the opposite (Figure 2.5). When
considering species dispersal capacities, there is a general trend of increasing PCstep fraction
associated with increased dispersal capacities for all scenarios (t1, t2 and st, Figures 2.6 and 2.7).
However, this PCstep increment happens for shorter-dispersing species in t1 and st, whereas just
for longer dispersers in t2 (Figure 2.6). These results are largely independent of landscape size
(Figures 2.9 to 2.14).
Considering all landscape sizes and dispersal capacities, the best supported models (total
models = 45) that were able to explain the variations in the fractions of PCst were the ones that
included the proportion of habitat loss (60%, 27 times) and also the amount of habitat lost (31%,
14 times). The PC fractions behavior, according to the proportion of habitat loss, varies as a
function of species dispersal capacity (Figure 2.7). For short-dispersal species, with low amounts
of habitat loss, PCintrast is by far the most important fraction, accounting for around 80% of the
PCst, whereas low values are obtained for PCdirectst and PCstepst although slightly higher for
PCdirectst (Figure 2.8). However, when the amount of habitat loss increases, the importance of
PCintrast drops, and the importance of PCstepst increases, whereas PCdirectst remains stable.
18
For short-dispersal species (50 m), at around 20% of habitat loss, PCstepst started to be more
relevant than PCintrast. For species with longer-dispersal capacities (1000 m for example), the
fractions are somewhat similarly important for low amounts of habitat loss, but PCstepst
increases (and PCdirectst and PCintrast decrease) for larger amounts of habitat loss. For these
longer-dispersing species, PCstepst is larger than PCintrast at already 10% of habitat loss. For
species that are able to disperser for even longer distances, such as 2500, 5000 or 10000m, the
importance of PCstepst is reduced, and PCdirectst increases (Figure 2.14).
19
Discussion
Spatio-temporal connectivity has a positive effect on landscape connectivity for all considered
landscape sizes and species dispersal capacities. Increases in connectivity occur primarily
through additional spatio-temporal pathways that appear or disappear between timesteps (i.e.,
temporal stepping-stone patches). We showed that measuring connectivity based on purely static
spatial metrics substantially underestimates connectivity levels, usually by 30%, but in some
cases by nearly 150%. Therefore, not accounting for spatial dynamics could severely
overestimate population isolation and extinction probabilities in changing landscapes.
We also demonstrated that accounting for spatio-temporal connectivity is particularly
important in landscapes with high levels of habitat change, and with net habitat loss, which is
common in the tropical regions (Hansen et al. 2013). In the tropics, both afforestation, given
passive forest regeneration, and deforestation, given demands for agricultural expansion, are
occurring concomitantly and at high rates (Lambin, Geist & Lepers 2003). These factors
ultimately generate a scenario where spatio-temporal patterns are complex, but particularly
relevant drivers of ecological processes and patterns. Whereas the spatio-temporal legacy is
commonly investigated in extinction debt studies (Hylander & Ehrlén 2013; Essl et al. 2015), the
spatio-temporal path approach, as presented here, and its influences on the spatio-temporal
legacy were, until now, not considered in landscape connectivity models. Studies that investigate
the spatial-temporal legacy suggested that the relaxation time and its trajectory are affected by
different resistances forces (Malanson 2002; James et al. 2007), and landscape connectivity is
one of the most influential ones (Jackson & Sax 2009). We suggest that the spatio-temporal
pathways, given its overarching positive influence in landscape connectivity, could significantly
contribute to inform on how to prevent species extinctions, especially in highly dynamic
landscapes, such as most of the tropical ones. Moreover, they might have a strong influence on
the duration, as well as in the trajectory of the relaxation time. Therefore, spatio-temporal
connectivity could help to avert species extinctions, and can partially be the responsible for the
over-estimations of species extinctions due to habitat loss, usually attributed to time-lagged
effects (Tilman et al. 1994). Such findings are opposite of what was previously suggested based
on metapopulation studies in simulated static and dynamic landscapes. In metapopulation
studies, dynamic landscapes experience more rapid declines in patch occupancy associated with
20
habitat reductions, and extinctions occur at higher levels of habitat (Wimberly 2006).
Nevertheless, our study involves real landscape dynamics that could depart from simulated ones,
particularly in terms of the spatial arrangements of stable and dynamic habitats (Matlack &
Monde 2004).
We observed a positive curvilinear effect of the amount of habitat on spatio-temporal
connectivity, with a peak in connectivity at around 30% of habitat (Figure 2.4). For low amounts
of habitat landscapes connectivity plays a minor role, as habitat patches are so far apart that
individuals rarely cross these large gaps; therefore, community composition and species
abundance are more related to patch size than to connectivity (Martensen et al. 2012). In
landscapes with reduced habitat coverage and connectivity, the spatial dynamics could be
detrimental, since they might reduce overall habitat quality (younger habitat), whereas not being
able to enhance connectivity (Wimberly 2006). At intermediate habitat amounts, the spatial
dynamics could enhance landscape connectivity between more stable or to newly created
patches, therefore increasing metapopulation survival probabilities (Matlack & Monde 2004;
Wimberly 2006), as seen by the larger relevance of the PCstepst fraction. In contrast, for larger
amounts of habitat, where purely spatial connectivity is already high, and therefore habitats are
already well connected, habitat dynamics could again reduce overall habitat quality, whereas not
significantly increasing an already high landscape connectivity (Figure 2.4).
The positive effect of spatio-temporal connectivity varies as a function of the percentage
of habitat change, at least between 30% of habitat loss and 5% of habitat gain (Figure 2.3a). With
additional habitat gain, purely spatial connectivity surpasses the influence of spatio-temporal
connectivity, and temporal aspects end up having no relevant influence. With net habitat loss
there is an increase in the importance of spatio-temporal connectivity, at least until 30% of
habitat loss, which is the investigated range. Even in landscapes with a stable net amount of
habitat, spatio-temporal connectivity has a positive influence on purely spatial connectivity,
since the patchwork of losses and gains promotes an increase in effective connectivity of around
10%. However, this relationship might not be linear, and additional studies should investigate
this further.
21
In our case study, we did not consider differences in habitat quality. Dynamic landscapes
are known to drive forest habitats to early successional stages, since new habitat is constantly
created, and more mature ones are lost (Teixeira et al. 2009). In dynamic landscapes, local
extinction could happen due to habitat destruction and overall reduction in habitat quality,
whereas connectivity can provides access to newly created patches, partially compensating these
extinctions (Matlack & Monde 2004). Agricultural intensification is expanding over tropical
regions, which has promoted large-scale spatial homogenization and reductions in landscape
dynamics (Fahrig et al. 2011), which can reduce spatio-temporal connectivity, whereas
generating conditions for habitats to age. Habitat regeneration speed, and species habitat
requirements and life span all play important roles in these dynamics. We believe that the model
and metrics presented here could be extremely helpful in the understanding of this balance, and
future efforts in this direction should incorporate information about habitat quality into node
attributes.
Additionally, we observed that species dispersal capacity is directly related to how
species are affected by the spatio-temporal dynamics, as larger dispersal capacities reduce the
dependence on landscape connectivity to sustain populations in fragmented landscapes (Hanski
1999). Nevertheless, we show that for species that could disperse up to 1000 m between habitats,
which is greater than the dispersal capability of most tropical forest birds and small mammals
species (Moore et al. 2008), the spatio-temporal dynamics of common tropical landscapes could
significantly enhance landscape connectivity beyond what is predicted by purely spatial models.
Therefore, we expect that for a large portion of tropical forest species, accounting for spatio-
temporal effects is key for the understanding of their dynamics in fragmented landscapes, and the
methods presented here could help in this endeavour.
Finally, we showed that most of the spatio-temporal connectivity happens through
habitats that are lost or gained between timesteps, which are used as stepping-stones (PCstepst)
to move between stable or to newly created patches (Matlack & Monde 2004; Wimberly 2006).
These dynamics are fundamental to linking habitat in a spatio-temporal context, by including
connectivity that does not exist in a purely spatial perspective. This generates temporal
directional connections, i.e., situations where patch A is connected to patch B, but B is not
connected to A (Figure 2.1c), differing from bidirectional spatial connections. This combination
22
of directional and bidirectional connections serves to mix populations and gene pools in a highly
heterogeneous manner.
The study of landscape dynamics is particularly pressing in the current changing world
(Auffret, Plue & Cousins 2015). Climate change is altering the speed of habitat regeneration in
some regions (Whitmore 1998) and increasing disturbance (Dale et al. 2001). Increases in
agricultural intensification are expected to reduce spatial and temporal heterogeneity, and large-
scale spatio-temporal homogenization is already happening in many regions (Fahrig et al. 2011).
In summary, the type, rate and intensity of disturbances are changing, and therefore landscape
dynamics are also changing (Turner 2010). To understand the effects of spatial dynamics and of
these changes in dynamics is vital for fine-tuning the understanding of ecological processes and
guiding landscape management. Changes in these dynamics could shorten relaxation time
periods, or accelerate extinction debt effects. We believe that these are pressing questions, and
that the model and metrics outlined here are an important contribution for future applications and
developments both in scientific and in management applications.
23
Figure 2.1: Spatial (solid lines) and spatio-temporal connectivity (dashed arrows). The grey solid
lines in t2 represent patch locations at t1. In (a) the letters represent the isolated populations of a
given species with a particular dispersal capacity. Population A is connected at t1, since both
patches are within the species dispersal capacity. The same happens for population D at t2.
24
However, A and D are considered isolated when t1 and t2 are analysed separately (i.e. without
accounting for temporal connections). In (b), although the patches have different sizes and
species compositions (different dark grey geometric shapes) at t1, the spatial aspects of t1 do not
affect their biological composition at t2. When accounting for both spatial and temporal
connections, in (c) a given individual, represented by the star, could be in the left fragment at t1,
and in the right fragment at t2, but not the other way around (from right to left, temporal
directional connection). Additionally, population A, present in the left and central fragments in
t1, became isolated in the left patch at t2, but is mixed with population B in the central and right
patches at t2, as represented by AB. In (d), the large patch in t1 could provide to the small patch
in t2 more species than an already small patch in t1 can do, as represented by the different width
of the dashed arrow, and by the different dark grey geometric shapes.
25
Figure 2.2: Spatial and spatio-temporal connectivity. (a) Purely spatial connections, (b) Spatio-
temporal direct movements, (c) Spatio-temporal stepping-stones movements, and (d) entire
connectivity pattern including both direct and indirect movements. The hollow polygons at t2
represents the polygons that were lost.
26
Figure 2.3: Contribution of the spatio-temporal connectivity ECAst compared to the purely
spatial connectivity ECAs at t2 (100(ECAst / ECAst2)-100); (a) density functions of the
contribution of the ECAst as a function of ECAs at t2. Positive values represent a positive
influence of the spatio-temporal connectivity over the purely spatial connectivity in t2.
Negative/zero values represent cases where either there was no influence of spatio-temporal
connectivity, or the increase in the purely spatial connectivity in t2 was so huge, that any increase
in connectivity caused by the spatio-temporal metrics was surpassed by the purely spatial
connectivity at t2. (b) The linear models of the percentage of the increment given by ECAst
compared to ECAs at t2 for all dispersal capacities.
27
Figure 2.4: Contribution of ECAst compared to ECAs t2 (100(ECAst / ECAst2)-100) as a
function of the amount of habitat in t2.
Figure 2.5: PC fractions independent of dispersal capacities. (a) PCdirects in t1, t2 and PCdirectst
in the spatio-temporal model; (b) PCintras in t1, t2 and PCintrast; and (c) PCsteps in t1, t2 and
PCstepst.
28
Figure 2.6: PC fractions contributions according to dispersal capacity (50 and 1000 m) in t1, t2
(for the spatial-only PCs) and in the spatio-temporal model (PCst).
29
Figure 2.7: PCst fractions (PCdirectst, PCintrast and PCstepst) contribution as a function of the
percentage of habitat loss for 50, 200 and 1000 m dispersal distances.
30
Table 2.1: List of variables and keywords.
Variables / keywords Type Description
t1, t2, tx, ty Points in time Initial, final and intermediate points of the time step (t1<tx<ty<t2)
Direct movement Movement type Direct movement consisting of a single link from a patch with habitat at t1 (node of
type Loss or Stable) to another patch with habitat at t2 (node of type Stable or Gain),
without using intermediate stepping-stone patches.
Stepping-stone
movement (indirect
movement)
Movement type Movement comprising multiple links in an indirect path from a patch with habitat at t1
to another patch with habitat at t2 going through one or several intermediate
stepping-stone patches.
i, j, k Nodes Any given node
31
Table 2.2: Movement possibilities along temporal connections between source (t1) and
destination (t2) nodes, not considering the spatial constraints. A value of 1 indicates that such
movement is possible at some moment within the analyzed period. A value of 0 indicates that it
is not possible from t1 to t2. Values of 0.5 indicate that the movement is possible given some
assumptions on the co-occurrence of nodes in time. Temporal movement possibilities are
directional (asymmetric) from t1 to t2 (source to destination).
Type of source node:
individual location at
t1 for the direct
movements or at tx
(t1<tx<t2) for the
indirect movements
Type of destination node: individual location after t1
Direct movements (individual
location in t2)
Indirect movements (individual
location in ty, tx<ty<t2)
Stable Loss Gain Stable Loss Gain
Stable 1 0 1 N/A 1 N/A
Loss 1 0 0.5 N/A 1 N/A
Gain 0 0 0 1 0.5 1
N/A: Not applicable
32
Table 2.3: Metrics description and equations. All metrics can be calculated for a spatial-only model (denoted with the suffix s) or for
the proposed spatio-temporal model (denoted with the suffix st).
Metrics Description Equation
PCs
PCst
Given two locations (source and destination) randomly selected within the landscape, the Probability of
Connectivity (PC) is the probability that these two locations fall into habitat areas that are connected, so that
it is possible for an individual to move from source to destination. PC can be partitioned in three fractions
that are described below (PC=PCintra+PCdirect+PCstep).
∑ ∑ 𝑎𝑖𝑎𝑗𝑝𝑖𝑗∗𝑛
𝑗=1𝑛𝑖=1
𝐴𝐿2
ECAs
ECAst
The Equivalent Connected Area (ECA) is the size of a single patch that provides the same value of the
Probability of Connectivity (PC) as the observed habitat pattern in the landscape. √∑∑𝑎𝑖𝑎𝑗𝑝𝑖𝑗
∗
𝑛
𝑗=1
𝑛
𝑖=1
PCintras
PCintrast
The fraction of PC that corresponds to the intrapatch connectivity – i.e. the amount of reachable habitat
within stable patches.
∑ 𝑎𝑖2𝑛
𝑖=1
𝐴𝐿2
PCdirects
PCdirectst
The fraction of PC that corresponds to the interpatch connectivity provided by complete direct connections
between patches (without using intermediate stepping-stones)
∑ ∑ 𝑎𝑖𝑎𝑗𝑝𝑖𝑗𝑛𝑗=1,𝑖≠𝑗
𝑛𝑖=1
𝐴𝐿2
PCsteps
PCstepst
The fraction of PC that corresponds to the interpatch connectivity provided by indirect connections made
possible by intermediate stepping-stone patches between source and destination patches.
∑ ∑ 𝑎𝑖𝑎𝑗(𝑝𝑖𝑗∗ − 𝑝𝑖𝑗)
𝑛𝑗=1,𝑖≠𝑗
𝑛𝑖=1
𝐴𝐿2
n is the number of habitat patches, ai and aj are the attributes of the patches (here habitat area), p*ij is the maximum product
probability of the paths between patches i and j (accounting for both direct and indirect stepping-stone movements), pij is the direct
33
dispersal probability between patches i and j (without using any intermediate stepping-stone patch), and AL is the maximum landscape
attribute (here total landscape area). See Saura & Rubio (2010) and Saura, Bodin & Fortin (2014) for further details.
34
Appendix
Dataset
The dataset used in our analysis consists of two different maps of land-use and cover from the
same region in different years (t1=1990 and t2=2001). The timestep of 10 years was chosen
because of its relevance to tropical forest regeneration dynamics in the region (Piotto 2011),
land-use characteristics (Ribeiro et al. 2012), as well as for many species life span (e.g.
understory birds). The maps were produced based on visual interpretation of images at a
resolution of 1:50 000 (see details in Ribeiro et al. 2012), with the different forest successional
classes merged, given that our purpose was to demonstrate the application of our network model
rather than addressing site-specific questions. Nonetheless, this decision is also biologically
supported, as a large fraction of the forested species will also use early stages of second-growth
forests, whereas a much smaller fraction will be restricted to mature forests (Chazdon et al.
2009).
Sampled landscapes
We randomly sampled 200 points that were at least 17841 m within the study region boundary,
therefore ensuring that all the sampled landscapes were completely inside the study region. In
order to test for the potential effect of landscape size on the spatio-temporal network assessment,
we used three landscape sizes of 25000, 50000 and 100000 ha centered at each point.
The proportion of area covered by habitat varied greatly among sampled landscapes,
from 5% to around 80%, with a median of about 25% (see details in Table 2.5). This large range
35
of variation in habitat amount represents almost the full spectrum of potential variation of
amounts of habitat in fragmented landscapes. Landscapes with more than 80% of habitat could
be considered as a continuous habitat landscape, whereas those with less than 5% can be
considered largely altered, where landscape connectivity could be of less ecological relevance
(Martensen, Pimentel & Metzger 2008; Martensen et al. 2012). The changes of habitat amount in
the analysed period also varied greatly among the sampled landscapes. There are cases where
habitat net gains were close to 2.5% of the total landscape area. However, the median amount of
net habitat change (net gains and losses) was of about 2.5% of habitat loss, with some cases
where more than 13% of the habitat was lost. In terms of proportion of habitat loss, the set of
landscapes that we analyzed encompass a large range of variation in changes in habitat amount,
from a loss of around 40% of habitat to a gain of a similar percentage (Table 2.4). It is expected
that a larger change in habitat amount would generate a greater effect on biodiversity (Wearn,
Reuman & Ewers 2012); therefore, by accounting for a large range of changes, both in terms of
habitat loss and gain, our sampled landscapes could represent the behaviour of the proposed
metrics over a broad set of real landscapes.
Table 2.4: variation of the percentages of habitat amount in the 200 landscapes at the three
landscape sizes.
Landscape sizes
Minimum Median Maximum
t1
(%)
t2
(%)
Stable
(%)
t1
(%)
t2
(%)
Stable
(%)
t1
(%)
t2
(%)
Stable
(%)
25000 5.72 5.02 4.55 26.65 23.8 19.65 80.72 76.00 72.34
50000 7.26 5.87 5.36 27.93 24.35 20.82 65.67 58.72 54.21
36
100000 7.59 7.97 7.07 26.65 23.03 20.43 57.49 53.04 47.94
Table 2.5: Absolute values of habitat loss and gain in hectares, and percentages of habitat loss
and gain as a function of habitat amount in t1.
Sizes
Loss Gain
Min median max min median max
(ha) (%) (ha) (%) (ha) (%) (ha) (%) (ha) (%) (ha) (%)
25000 24.05 0.72 1430.65 20.99 4440.51 46.47 19.55 0.42 697.21 9.56 2482 45.34
50000 44.96 0.74 3090.26 22.39 7982.05 41.03 139.4 1.77 1398.06 10.33 4891.5 34.78
100000 331.11 3.07 6224.1 22.84 14741.7 39.96 447.29 3.33 2678.29 10.5 7923.47 28.34
37
Figure 2.8: Contribution of the spatio-temporal connectivity ECAst compared to the purely
spatial connectivity ECAs at t2 (100(ECAst / ECAst2)-100). Density functions of the contribution
38
of the ECAst as a function of ECAs at t2 according to species dispersal capacity (50, 100, 200,
500 and 1000 m) and landscape size (25000, 50000, 100000 ha). Positive values represent a
positive influence of the spatio-temporal connectivity over the purely spatial connectivity in t2.
Negative/zero values represent cases where either there was no influence of spatio-temporal
connectivity, or the increase in the purely spatial connectivity in t2 was so huge, that any increase
in connectivity caused by the spatio-temporal metrics was surpassed by the purely spatial
connectivity at t2. Medians are shown.
39
Figure 2.9: The linear models of the percentage of the increment given by ECAst compared to
ECAs at t2 based on the differences in habitat amount for all landscape sizes (25000, 50000 and
100000 ha) and dispersal capacities (50, 100, 200, 500 and 1000 m). The betas are shown.
40
Figure 2.10: PC fractions for the 25000 and 50000 ha landscapes. For the 25000 ha landscapes
(upper panels): PC fractions in t1 (a), t2 (b) and spatio-temporal (st) (c); (d) PCdirect in t1, t2 and
spatio-temporal (st); (e) PCintra in t1, t2 and spatio-temporal (st); and (f) PCstep in t1, t2 and
spatio-temporal (st); For the 50000 ha landscapes (lower panels): PC fractions in t1 (a), t2 (b) and
spatio-temporal (st) (c); (d) PCdirect in t1, t2 and spatio-temporal (st); (e) PCintra in t1, t2 and
spatio-temporal (st); and (f) PCstep in t1, t2 and spatio-temporal (st).
41
Figure 2.11: PC fractions contributions for different dispersal capacities in t1, t2 and spatio-
temporal for the landscapes with 25000 ha.
42
Figure 2.12: PC fractions contributions for different dispersal capacities in t1, t2 and spatio-
temporal for the landscapes with 50000 ha.
43
25000 ha landscapes
50000 ha landscapes
Figure 2.13: PCst fractions contribution (PCdirectst, PCintrast and PCstepst) as a function of the
percentage of habitat loss for three dispersal distances (50, 500 and 1000 m) for landscape sizes
of 25000 and 50000 ha.
44
Figure 2.14: Contribution of the PC fractions for species with larger dispersal capacities in the
100000 ha landscape.
45
- Land-use intensification constraints spatio-temporal connectivity of fragmented tropical forest landscapes
Abstract
1. Land-use intensification is expanding over large regions in the tropics, altering landscape
heterogeneity, composition and dynamics, affecting the likelihood of species movements and
the diversity patterns at different scales.
2. Here we investigate how land-use intensification (landscape dynamics and habitat-turnover
rates), influences landscape spatio-temporal connectivity (i.e., the reachable habitat through
spatial and temporal connections). Moreover, we evaluate if this connectivity influence varies
as a function of species dispersal capacity, land uses and habitat quality of the newly
regenerated forests. Specially we evaluate the trade-off between (i) a rapid turnover of forest
habitat, which results in lower habitat quality but a more dynamic landscape, providing
potential opportunities for species movements through the landscape, and (ii) a scenario in
which forest turnover is reduced, leading to more mature forests stands but to lower spatial
and temporal heterogeneity.
3. We used a network-based model of landscape dynamics and calculated spatio-temporal
connectivity metrics to account for the reachable habitat. We evaluated 59 circular landscapes
with a radius of 10000 m in the Atlantic Forest of Brazil that are experiencing different levels
of land-use intensification, from the expansion of Eucalyptus plantations, to the retention of
low intensified shaded-cocoa agroforests and pastures.
4. The amount of forest was a key driver of spatio-temporal connectivity, but land-use was
highly influential as well. High-intensity land-use (Eucalytpus plantation) had a strong overall
effect reducing spatio-temporal connectivity, whereas the low-intensity land-use (shaded
cacao, low intensified pastures and agricultural fields) had positive connectivity effects,
mainly because it allows for the existence of ephemeral stepping-stones patches that connect
fragments that would otherwise remain isolated through time. Additionally, the decrease in
overall forest quality following vegetation turnover in low-intensity land-use landscapes had a
minor impact on species habitat availability compared to the increase in spatio-temporal
connectivity that turnover provided.
46
5. We show that land-use intensification reduces spatio-temporal connectivity by slowing down
landscape spatial dynamics, holding many habitat patches into static ‘isolation traps’ in which
there are not temporal windows through which colonisations or migrations can happen. To
maintain connectivity levels, it is necessary to spare more land for biodiversity conservation
in landscapes that are experiencing land-use intensification than in those in which low-
intensity land-use predominates. The spatio-temporal connectivity metrics that we applied
could help in the evaluation of the amount and location of land that needs to be spared to
maintain or increase the connectivity levels in landscapes under different land-use
intensification situations. Finally, we argue that traditional low-intensity land-use in the
tropics is able to sustain high levels of spatio-temporal connectivity, and therefore, may
importantly contribute to preventing landscape-scale extinctions in dynamic, fragmented
landscapes.
47
Introduction
The rising demands for food, biofuels, and fibers are important drivers of landscape composition,
heterogeneity, and dynamics (Assessment 2005) causing biodiversity losses and threats to the
provision of ecosystem services (Cardinale et al. 2012). As market pressures continue to grow,
locations where land-use intensification are technologically and economically possible will be
increased (Foley et al. 2005). Land-use intensification reduces landscape functional
heterogeneity (Fahrig et al. 2011) by removing small forest fragments and reallocating land to
increase field size (Tscharntke et al. 2005). The elimination of ephemeral forested patches
immersed in the agricultural fields could result in reductions of spatio-temporal connectivity,
given that these patches could be used as stepping-stones among more stable patches (Chapter
2). Native habitats are often constrained into sites with low aptitude for crops (Latawiec et al.
2015), and a reduced number of intensified uses dominate the higher productivity areas.
Increasing land-use intensity is also known to lead to higher matrix harshness, resulting in lower
permeability to movement among native habitats, and reducing its use as complementary habitats
(Perfecto & Vandermeer 2008). Land-use intensification also reduces landscape spatial
dynamics, because a large part of the land is locked up into one land-cover type and land
abandoned to native habitat regeneration is low (Angelsen 2010). These changes on spatial
patterns ultimately decrease landscape connectivity (Tscharntke et al. 2005). Connectivity could
prevent species extinctions by allowing the use of multiple fragments in daily or occasional
movements (Martensen, Pimentel & Metzger 2008), rescue effect (Brown & Kodric-Brown
1977), or remediate local losses with recolonizations (Kuussaari et al. 2009). In summary, higher
levels of connectivity in fragmented landscapes could counter-act extinctions and potentially
influence the extinction debt by altering the pattern of the relaxation time (Malanson 2008).
Low-intensified dynamic tropical fragmented forested landscapes show a progressively
younger secondary forest composition (Teixeira et al. 2009), which could potentially reduce their
capacity of hosting sensitive forest species. However, studies that compare species composition
in secondary and pristine forests are usually similar, suggesting that secondary forests can sustain
a large fraction of the species found in pristine ones (see review in Chazdon et al., 2009),
although some species are found exclusively in mature forests (Gibson et al. 2011).
Nevertheless, the speed and trajectories of forest recovery, is better known for plants (Liebsch,
48
Marques & Goldenberg 2008; Piotto et al. 2009), and biomass/carbon (Poorter et al. 2016). Yet,
they are consistent in suggesting that after a fast and robust initial recovery, additional gains,
particularly in species composition, are very slow (Martin, Bullock & Newton 2013). Therefore,
reducing habitat turnover could allow habitat to mature, and therefore, to sustain a larger number
of species sensitive to forest disturbance. The effects and the trade-offs of the spatio-temporal
dynamics and habitat quality on biodiversity conservation are still open fundamental questions,
particularly in the current situation of global changes. A better understanding of these aspects
could shed light into how species are behaving in dynamic landscapes and how they will be
affected by changes in spatio-temporal dynamics (Melo et al. 2013).
Here we quantify the amount of reachable forests through both spatial and temporal
connections in landscapes that are experiencing different trends of land-use intensification using
a spatio-temporal network model (Chapter 2). Specifically, we assess how land-use
intensification alters spatio-temporal patterns of native forests, and how these changes influence
habitat availability and connectivity for animal species, as a function of their dispersal capacity
and habitat requirements. Furthermore, we evaluate the trade-offs of forest dynamics, by
accounting for habitat-aging effects. We hypothesized that, for a similar net amount of habitat
through time, a more dynamic landscape would have higher spatio-temporal connectivity,
provided by forest fragments that appear and disappear over time, therefore providing enhanced
overall spatial connectivity, while more stable landscapes would have less spatial and spatio-
temporal connectivity. Yet, landscapes that are more dynamic would have lower overall habitat
quality, compared to more stable ones, given the constant appearance and disappearance of
habitat, which resulted in reduced population viability of sensitive species on these landscapes.
Methods
3.3.1 Spatio-temporal model
To evaluate forest habitat reachability through space and time, we developed and used a novel
spatio-temporal network approach (Chapter 2). As in common landscape network models,
habitat patches (here forest patches) are represented as nodes, and their potential connections via
species dispersal abilities as links or edges (Fall et al. 2007). Our approach however integrates
49
both spatial and temporal links in the assessment of habitat connectivity in dynamic landscapes.
In our approach, a spatio-temporal link is the possibility of an individual moving from a given
patch location at time 1 (t1) to a different patch location later in time 2 (t2). The spatio-temporal
paths (combination of one or multiple spatio-temporal links in the movement through the
network) provide a proxy for gene flow and connectivity between populations across time.
For each time-step (pair of years), forest patches were classified into three states: Stable
(patch exists at t1 and t2); Loss (patch lost from t1 to t2); and Gain (patch appears from t1 to t2).
We assume that time-steps are short enough, so that no more than one of the three patch states
could occur in the period between t1 and t2 for any given location (e.g., habitat loss and gain not
happening in the same location during the considered period).
The spatio-temporal links among patches can be of two forms. A direct path, i.e. a single
link movement from a patch at t1 to another patch at t2. Alternatively, it could be a stepping-stone
path, i.e. a multiple link (step) movement from a patch at t1 to another patch at t2, which involves
one or more intermediate stepping-stone patches.
3.3.2 Metrics of habitat reachability in the spatio-temporal network
We calculated the spatio-temporal Probability of Connectivity (PCst) and the Equivalent
Connected Area (ECAst, Chapter 2). Given two randomly selected locations (source location at
t1 and destination location at t2); PCst is defined as the probability that the source and the
destination locations are spatio-temporally connected such that an individual in the source
location at t1 can move to the destination location at t2. ECAst is the amount of area reachable
given the spatio-temporal connectivity (PCst). Therefore, these metrics account for: (a) the area
of each patch (intra-patch connectivity) and (b) the amount of patch area reachable by moving
through the links in the network (inter-patch connectivity).
The spatio-temporal connectivity PCst can be decomposed into three fractions: PCintrast
(intra-patch spatio-temporal connectivity), PCdirectst (inter-patch spatio-temporal connectivity
provided by direct spatio-temporal connections between patches), and PCstepst (inter-patch
spatio-temporal connectivity provided by stepping-stone connections between patches); see
Chapter 2 for additional details. Calculations were performed by combining an R script with a
50
command-line version of the freely available spatio-temporal ConeforST software package
(Chapter 2).
3.3.3 Model parametrization
We used a negative exponential function of inter-patch distance to obtain the probability of direct
movement between any pair of patches. We considered five values of the median dispersal
distance: 50, 100, 200, 500 and 1000 m. Gap-crossing capacities for tropical birds are usually
below 200 meters (e.g. Pyriglena leucoptera and Sclerurus scansor ~ 60 m - Uezu et al., 2005,
Hansbauer et al., 2008, Xiphorhynchus fuscus < 100 m - Boscolo et al., 2008), also for small
mammals (Marmosops incanus, Micoreus paraguayanus ~ 100 m - Forero-Medina and Vieira,
2009), with a few species able to cross gaps of many hundreds of meters (Moore et al., 2008,
Crouzeilles et al. 2010). Therefore, with the selected dispersal distances, we cover a large range
of potential dispersal abilities for species in the region. As our intention was to evaluate the
changes in forest dynamics, we considered matrix permeability to be uniform independent of the
land-use class, i.e., we do not consider possible differences between Eucalyptus and shaded-
cocoa plantations matrix permeability. Hence, we used Euclidian distance among patches, and
the kernel dispersal function for each species was parameterized so that it gave a 0.5 probability
of movement (gap crossing) between patches, when the patches were separated by an edge-to-
edge distance equal to the considered median dispersal distance.
Additionally, in order to evaluate the trade-offs between the reduced quality of the
regenerated habitat and their role in spatio-temporal connectivity, we simulated different
qualities of regenerated forests (i.e. gain patches). We used a random uniform distribution to
assign the quality of each regenerated patch: 0-25; 0-50; 0-75; and 0-100% of the habitat quality
encountered in a patch that was already present in t1. Therefore, a Gain patch with 0% habitat
quality is exclusively used for connectivity purpose, i.e., individuals would not stay in this patch,
therefore, its area is not considered as an available resource. From another side, a Gain patch
with 100% of habitat quality would allow species to fully use it as habitat, i.e., not only with
connectivity purpose, but also as habitat resources. The results among different regenerated-
patch qualities were not statistically significant most of the time, as well as when considering the
51
different types of habitat (Supplementary Figure 1). Therefore, we show only the extreme cases
of regenerated patch quality; 0 and 100%.
3.3.4 Study region
Our study region covers over 2 million hectares of the Atlantic Forest, south of Bahia and a small
portion of the east of Minas Gerais, in Brazil. The Atlantic Forest is one of the most diverse and
has one of the highest levels of endemism on Earth (Thomas et al. 1998; Martini et al. 2007).
Selective logging of Pau-Brasil (Caesalpinia echinata) began in the 20th century (Dean 1996),
and cacao (Theobroma cacao) plantations began at the start of the 18th century and quickly
expanded (CEPLAC - http://www.ceplac.gov.br/). Cacao is mainly planted in agroforest
schemes, in association with the native forests; hence, the region remained well preserved until
the middle of the 20th century. In the 1980s the forests and the shaded cacao plantations were
largely converted to low intensified pastures and agricultural fields (Thomas et al. 1998). Since
1990 a fast expansion of highly intensified Eucalyptus plantations has occurred, which today,
covers more than 10% of the area (Ribeiro et al. 2012). Overall, forest regeneration is very fast,
and is associated with former land-use intensity, time since forest conversion, and distance to
seed sources (Piotto 2011).
The database is composed by three different maps from the same region in different years
(t1 = 1990, t2 = 2000 and t3 = 2007), which covers the period since the beginning of land-use
intensification in the region. The 1990 and 2000 maps were produced based on visual
interpretation of images with a resolution of 1:50 000 covering an area of ~2.3 millions of
hectares, whereas the 2007 map was produced by visual interpretations of images of resolution
of 1:20 000 (see details in Ribeiro et al. 2012), and latter degraded to make maps compatible.
Although the original maps differentiate many land uses and cover classes, we opted to combine
the non-native land uses into four classes: pastures/agricultural fields, Eucalyptus plantations,
shaded cacao and urban areas, to reduce classification errors. Urban areas were always present in
low amounts; therefore, it was used in the Principal Component Analyses (PCA), but not in the
regression models (see below).
Two different species profiles in terms of habitat requirements were evaluated. The first
included species that are able to use any kind of forest successional stage (initial, intermediate
52
and mature forests), and the second, with species that use only the most dynamic successional
stages (initial and intermediate stages of succession). Given map characteristics, it was not
possible to subset map classes in order to define a class that included intermediate and mature
forests, but not the initial ones. However, a large fraction of the forested species will also use
second-growth forests, particularly in intermediate stages of succession, whereas a much smaller
fraction will be restricted to mature forests (Chazdon et al. 2009).
3.3.5 Sampled landscapes
We systematically sampled 59 points that were at least 10000 m from the borders of the study
region, ensuring that all landscapes fall completely inside the study region. Around these points
we create a buffer of 10000 m, therefore, generating landscapes that were 31416 ha. The
proportion of forest area varied greatly among sampled landscapes (from 6.49 to 84.99%, median
= 28.02%), as well as the land covers (shaded-cacao from 0 to 47.03%, median = 0.12%; pasture
from 14.8 to 93.42%, median = 60.06%; Eucalyptus from 0 to 37.8%, median = 0.32%; urban
from 0 to 5.59%, median = 0.06%). The large range in variation for forest amounts, land uses,
landscape intensification, and transitional pathways among landscape features (Supplementary
Figure 2) generate a comprehensive picture of the potential spatio-temporal influences occurring
in landscapes, and its relationship with land-use intensity.
3.3.6 Statistical analysis
We used factorial ANOVA to evaluate the main effects of the dispersal capacities (50, 100, 200,
500 and 1000 m), species habitat requirements (mature, intermediate and initial, or just initial
and intermediate), regenerated habitat quality (0 and 100% of habitat quality) and first and
second time-steps on the spatio-temporal connectivity metrics. Then we used linear models to
evaluate the influence of the amount of native habitat and of the different land uses in the spatio-
temporal connectivity. In order to obtain the effects of the land uses independently of the amount
of forest, we used the residuals of the regressions of each land-use variable by the amount of
native habitat (based on the different habitat requirements, only initial and intermediate, or all
forest types) in the second time-step for each analyzed period. We applied a model selection
protocol using AICc and all combinations of the four variables (habitat amount, and the residuals
53
of the three land cover types: amount of pastures, shaded cacao, and Eucalyptus) and the null
model, for each of the two habitat requirements species profiles (mature, intermediate and initial
habitats together, or only intermediate and initial). We used a change in AICc > 2 to flag the
models with better support. Finally, we sum all the AICc weights for any model that each of the
variable were present, obtaining the importance of each variable in a multi-model inference
framework (Burnham & Anderson 2002).
Results
3.4.1 Spatio-temporal connectivity variation from 1990 to 2007
There is a 19% reduction of the spatio-temporal equivalent connected area (ECAst) from t1990-
t2000 to t2000-t2007 (p < 7 e-13, Table 3.1). However, this difference of ECAst between periods is
similar to the reduction in time between the time-steps (i.e. 10 years from 1990-2000 and 7 years
2000-2007). ECAst increases with increments in the dispersal capacities considered (p < 2e-16,
Table 3.1), and by the different types of habitat considered, since we considered all types of
habitat, when there is more habitat in the landscapes there is more connectivity (p < 2e-16, Table
1). ECAst was not influenced by the different patch qualities considered for the regenerated
habitat (p = 0.727, Table 3.1), although a slightly greater ECAst was observed for higher quality
habitats (Table 3.1). The interactions between variables (periods analyzed 1990-2000 and 2000-
2007, dispersal capacities, habitat types and regenerated patch quality) were not influential.
The different periods (1990-2000 and 2000-2007), the dispersal capacities (50, 100, 200,
500 and 1000m), and the composition of the native habitat (initial, intermediate and mature, or
only initial and intermediate) were influential in all fractions of the PCst (p < 1 e-7, for all cases).
The regenerated habitat quality was also influential for PCst intra (p = 0.005), and marginally
significant for PCst direct (p = 0.057). For PCst step and PCst intra the interaction between the
periods of time (1990-2000 and 2000-2007) and the dispersal capacities (50, 100, 200, 500 and
1000m) were also significantly influential (respectively p = 0.001 and p = 5.6 e-7), suggesting
that the influence of the dispersal capacities changes between the first (1990-2000) and second
(2000-2007) periods analyzed.
54
Table 3.1: Medians and standard deviations among treatments of each of the spatio-temporal
connectivity metrics (ECAst: Equivalent connected area and the PCst fractions).
ECAst PCstepst PCdirectst PCintrast
Treatment Median
Standard
deviation Median
Standard
deviation Median
Standard
deviation Median
Standard
deviation
time
1990-2000 2561.25 4318.54 46.78 24.16 19.15 7.61 22.96 23.61
2000-2007 2067.31 3744.30 42.49 23.35 20.59 7.59 28.75 23.62
Dispersal
distances
50m 1555.12 3662.97 26.24 22.26 16.32 5.56 47.78 23.64
200m 2215.17 3959.98 40.87 22.81 20.14 6.28 28.44 21.34
1000m 3492.12 4334.08 60.55 19.77 23.82 8.59 10.77 15.12
Habitat
composition
Initial, intermediate
and mature 3050.90 4791.55 40.47 24.17 19.05 7.86 31.15 25.01
Initial, intermediate 1778.42 2806.37 47.47 23.28 20.54 7.34 22.35 21.46
Regenerated
habitat quality
0% 2268.65 3976.19 44.60 23.65 19.61 7.83 27.69 24.07
100% 2370.59 4139.78 44.14 24.03 19.89 7.42 24.61 23.29
3.4.2 Proportion of native habitats and land-use intensity on spatio-
temporal connectivity
The amount of native habitats was the main driver of the spatio-temporal connectivity (Table
3.2). However, land-use types were also relevant for every dispersal distance and for the
different classes of habitat considered (Table 3.2). The Eucalyptus plantation was the most
influential land-use and had a strong negative influence in the ECAst. Pastures were the second,
followed by shaded-cacao, both with generally positive influences in the ECAst. However,
55
shaded-cacao had a low negative influence in some of the models (marked with a * in Table 3.2).
These results were largely influenced by one of the landscapes (landscape 54) that presented a
large percent of shaded-cacao (~50%). When we excluded landscape 54 from the analysis, the
influence of the shaded-cacao variable was positive. When considering only the initial and the
intermediate habitats, the amount of native habitat presented higher importance compared with
the rest of the land covers; nevertheless, land-use influence was still relevant and the results were
similar when the analysis considered all native habitat, i.e. including mature forests (Table 3.2).
Results were also similar when we considered the effect of the variable quality of the regenerated
habitat (Supplementary Material Figure 3.1).
The amount of native habitat always had a positive influence in all the PCst fractions.
The pastures were the most relevant land-use in 14 models, whereas the Eucalyptus plantations
were the most important land-use in 10 models (Table 3.3). The intensified land-use
(Eucalytpus) has predominantly a negative influence in all PCst fractions, with exception of the
initial and intermediate habitat type for PCstepst. From another side, the low intensified land
uses (pastures and shaded-cacao) have a predominantly positive influence in the PCst fractions,
with only three exceptions for pastures and a few more for the shaded-cacao, again largely
influenced by the landscape 54 (see cases with the * in the Table 3.3). The shaded-cacao was the
least influential land-use. For the PCintrast the Eucalyptus plantations were particularly relevant
when considering all native habitats, whereas the pastures were more significant when
considering only the initial and the intermediate habitats, both independent of the dispersal
capacities. For PCdirectst, pasture was the land-use with the strongest influence, independent of
the types of habitat considered, however, for the short dispersers the Eucalyptus were more
relevant between t1999-t2000 (Table 3.3). For the PCstepst both Eucalyptus (5 times) with a
predominantly negative influence and pastures (3 times) with a predominantly positive influence
were relevant.
56
Table 3.2: For the runs with habitat regeneration with the same quality of the regenerated habitat
the PCst were better explained by the following models.
All Native Habitats
Initial and Intermediate Native Habitats
50
m
Native
habitat Eucalyptus Pastures Cacao*
Native
habitat Eucalyptus Pastures Cacao*
t1990-
2000
Importance: 1 0.95 0.43 0.25
1 0.52 0.32 0.24
# of models 8 8 8 8
8 8 8 8
+ - + +
+ - + +
Native
habitat Eucalyptus Pastures Cacao
Native
habitat Pastures Eucalyptus Cacao *
t2000-
2007
Importance: 1 0.84 0.53 0.33
1 0.6 0.37 0.28
# of models 8 8 8 8 8 8 8 8
+ - + + + - + +
200
m
Native
habitat Eucalyptus Pastures Cacao *
Native
habitat Eucalyptus Pastures Cacao *
t1990-
2000
Importance: 1 0.96 0.38 0.25
1 0.65 0.26 0.24
# of models 8 8 8 8
8 8 8 8
+ - + +
+ - + +
Native
habitat Eucalyptus Pastures Cacao
Native
habitat Pastures Eucalyptus Cacao *
Importance: 1 0.84 0.53 0.33
1 0.47 0.41 0.26
57
t2000-
2007 # of models 8 8 8 8 8 8 8 8
+ - + + + - + +
10
00
m
Native
habitat Eucalyptus Pastures Cacao *
Native
habitat Eucalyptus Cacao * Pastures
t1990-
2000
Importance: 1 0.97 0.35 0.26
1 0.69 0.25 0.24
# of models 8 8 8 8
8 8 8 8
+ - + +
+ - + +
Native
habitat Eucalyptus Pastures Cacao
Native
habitat Eucalyptus Pastures Cacao *
t2000-
2007
Importance: 1 0.77 0.56 0.32
1 0.35 0.31 0.24
# of models 8 8 8 8 8 8 8 8
+ - + + + - + +
58
Table 3.3: Results for the models when considering all native habitats and only the initial and intermediate native habitats (*
represents cases where a negative influence was observed, but largely influenced by only one landscape).
All Native Habitats Initial and Intermediate Native Habitats
PC
step
st
t1990-
2000
50
m
Native habitat Eucalyptus Pastures Cacao
Native habitat Eucalyptus Pastures Cacao
Importance: 1 0.25 0.24 0.24
1 0.46 0.28 0.24
+ - + +
+ - + +
t2000-
2007
Native habitat Pastures Eucalyptus Cacao
Native habitat Pastures Eucalyptus Cacao
Importance: 1 0.41 0.25 0.25
1 0.55 0.36 0.27
+ + - +
+ + - +
t1990-
2000
100
0m
Native habitat Pastures Eucalyptus Cacao
Native habitat Eucalyptus Pastures Cacao
Importance: 1 0.7 0.39 0.35
1 0.64 0.32 0.25
+ - + -
+ - - +
t2000-
2007
Native habitat Eucalyptus Pastures Cacao
Native habitat Eucalyptus Pastures Cacao *
Importance: 1 0.88 0.27 0.25
1 0.31 0.26 0.24
59
+ + - -
+ - + +
PC
dir
ects
t
t1990-
2000
50
m
Native habitat Eucalyptus Pastures Cacao
Native habitat Pastures Eucalyptus Cacao *
Importance: 1 0.9 0.43 0.26
1 0.48 0.41 0.25
+ - + +
+ + - +
t2000-
2007
Native habitat Pastures Cacao Eucalyptus Native habitat Pastures Cacao * Eucalyptus
Importance: 1 0.92 0.67 0.39
1 0.92 0.39 0.31
+ + + -
+ + + -
t1990-
2000
100
0m
Native habitat Pastures Eucalyptus Cacao *
Native habitat Pastures Eucalyptus Cacao *
Importance: 1 0.81 0.74 0.32
1 0.43 0.41 0.25
+ + - +
+ + - +
t2000-
2007
Native habitat Pastures Cacao Eucalyptus Native habitat Pastures Eucalyptus Cacao *
Importance: 1 0.98 0.73 0.36
1 0.56 0.38 0.27
+ + + -
+ + - +
PC
in
tras
t
50m
Native habitat Eucalyptus Pastures Cacao
Native habitat Pastures Eucalyptus Cacao *
60
t1990-
2000
Importance: 1 0.92 0.44 0.25
1 0.6 0.52 0.26
+ - + -
+ + - +
t2000-
2007
Native habitat Eucalyptus Pastures Cacao
Native habitat Pastures Eucalyptus Cacao
Importance: 1 0.93 0.31 0.25
1 0.36 0.34 0.26
+ - + +
+ + - +
t1990-
2000
10
00
m
Native habitat Eucalyptus Pastures Cacao *
Native habitat Pastures Eucalyptus Cacao *
Importance: 1 0.92 0.44 0.25
1 0.6 0.52 0.26
+ - + +
+ + - +
t2000-
2007
Native habitat Eucalyptus Pastures Cacao
Native habitat Pastures Eucalyptus Cacao
Importance: 1 0.93 0.31 0.25
1 0.36 0.34 0.26
+ - + + + + - +
61
Discussion
The amount of native habitat has a preeminent positive effect on spatio-temporal connectivity;
however, land-use is also important. Land-use intensification reduces spatio-temporal
connectivity independently of the amount of native forests, species dispersal capacity and types
of native forest considered. Agricultural intensification has been claimed as the best available
option to enhance harvest of food, fibres and biofuels without additional natural habitat
reductions (Green et al. 2005; Phalan et al. 2011; Foley et al. 2011). However, land-use
intensification is associated with lower resilience of the native habitats (Karp et al. 2012;
Jakovac et al. 2015), with lower forest regeneration and reduced temporal dynamics in
fragmented landscapes (Fahrig et al. 2011). We demonstrated that land-use intensification causes
a decrease in spatio-temporal connectivity on these landscapes, given by the reduction in both
spatial heterogeneity and temporal dynamics, even if the same amount of habitat is retained
(Table 3.2). Therefore, to maintain connectivity levels in landscapes that are experiencing land-
use intensification it is necessary to spare more land for native habitats, than when under low-
intensified land uses.
Today, more than half of the tropical forest is gone, and what remains are in most cases
severely fragmented (Haddad et al. 2015). However, there is little evidence of species extinctions
that can be directly attributed to the reduction and fragmentation of the native habitats in the
tropics (Heywood & Stuart 1992; Sodhi et al. 2010). This lag of extinctions is usually attributed
to the recent degradation of the tropical biomes (Assessment 2005; FAO 2010), and an expected
delay in species extinctions (Tilman et al. 1994; Hanski & Ovaskainen 2000). Additionally,
maintaining high levels of connectivity in fragmented landscapes can prevent extinctions
(Hanski & Ovaskainen 2000; Malanson 2002). We argue that the traditional low-intensified land
uses in the tropics sustain high levels of spatio-temporal landscape connectivity, and therefore,
contribute to prevent landscape scale extinctions in dynamic fragmented landscapes. With the
mechanization associated with the process of land-use intensification, low aptitude sites (for
example, steep portions of the landscape) are usually abandoned (Lambin & Geist 2008). Yet, in
many of our sampled landscapes, the simple abandonment of these sites was not enough to
compensate for the reductions in spatio-temporal connectivity (Table 3.3). Therefore, to
62
guarantee the maintenance of the previous connectivity levels, land-use intensification needs to
be accompanied by a considerable planned set-aside of areas to overcome losses in spatio-
temporal connectivity. When land-use intensification is high, the spatio-temporal Equivalent
Connected Area metric (ECAst) could be applied to determine the amounts of habitat that should
be restored and the best spatial locations to overcome the reductions.
Land-use intensification is also associated with larger field sizes and a reduced number of
land-cover types, which allied with the intensified production, usually culminates in a low matrix
permeability (Goulart, Salles & Machado 2013). That being said, tropical low-intensified
fragmented landscapes are shown to be particularly hyperdynamic (Laurance 2002), which could
differentiate these landscapes from more static intensified ones (Karp et al. 2012). Therefore,
there could be synergistic effects between the reduction in structural spatio-temporal
connectivity and the increment of matrix harshness caused by land-use intensification. Such
synergies would increase functional spatio-temporal isolation of populations in intensified
landscapes. In our study, we did not consider matrix permeability; however, the available
evidence shows that corridors and other structural connectivity features such as stepping-stones,
which were all considered in our study, are even more important under the intensification
scenario that tropical regions are experiencing (Uezu, Beyer & Metzger 2008). Contrasting,
stepping-stones habitats, including temporal ones (i.e., that appear and disappear over time), are
more common and therefore, more relevant for spatio-temporal connectivity in lower intensified
landscapes (Table 3.1). Additionally, different types of low-intensity land uses could provide
habitat with different qualities (Green et al. 2005), which could be an important trade-off
between the amount and quality of the available habitat for different species. For example, the
shaded-cacao plantations in our study region are used by many species of mammals, birds,
butterflies, frogs and lizards, as well as ferns and trees (Faria et al. 2006; Pardini et al. 2009).
However, land-use intensification could create situations where habitat stability allows
native habitat to age. Mature forests have been identified as irreplaceable for biodiversity
conservation (Gibson et al. 2011), however, second-growth forests (< 40 years) quickly recover
biodiversity up to a certain level (Liebsch, Marques & Goldenberg 2008; Piotto et al. 2009;
Martin, Bullock & Newton 2013), hosting many taxonomic groups, including some sensitive
species (Dunn 2004; Chazdon et al. 2009). Our simulations suggest that only small amounts of
63
forests are gained and lost among the evaluated time-steps (7 to 10 years), and therefore do not
significantly change overall landscape forest suitability between the scenarios with different
qualities of the regenerated habitat. However, the regenerated forests play an important role in
maintaining connectivity (Laps 2006), by positively influencing landscape spatio-temporal
connectivity. Clearly, this will be directly influenced by regeneration characteristics (Poorter et
al. 2016), and by the speed of forest conversion. Field studies in the region have demonstrated
that vegetation structure (Faria et al. 2009) and fauna composition (Faria et al. 2006) in early
secondary forests is different than in mature stands. Actually, many places across the tropics are
experiencing an overall reduction in forest quality due to younger secondary forest composition,
what reduces its ability to sustain sensitive species (Teixeira et al. 2009). Additionally, corridors
composed by early stages of successional forests, particularly the ones that present a large
length/width ratio, are used less frequently than more mature forests (Lees & Peres 2008).
Therefore, although our results support the notion that the reduction in overall landscape forest
quality presented less influence compared to the increment in connectivity in dynamic
landscapes, species use of these newly created habitats should be better studied, and the potential
detrimental influences of larger amounts of early successional forests on species conservation
assessed. Additionally, developing management techniques that focus in speeding up
regeneration of these areas, especially in order to improve habitat use, could generate win-win
situations, where connectivity as well as habitat use could be improved, without additional set
aside of areas for conservation.
The current rapid pace of land-use intensification that is occurring in the tropics has
major effects on spatio-temporal connectivity, mainly by reducing spatio-temporal stepping-
stones patches. Moreover, land-use intensification will reduce matrix permeability further
reducing connectivity (Perfecto & Vandermeer 2010). With these different sources of declines in
landscape connectivity, either by reducing spatio-temporal connectivity, or by decreasing matrix
permeability, it is clear that biodiversity patterns and processes in fragmented landscapes will be
directly affected. We forecast that the relaxation time will be reduced, extinction debts will be
paid more promptly, and if management strategies in order to spare additional land for
conservation are not urgently adopted, we are going to start observing many extinctions across
the tropics very soon.
64
Appendix
Figure 3.1: The responses for all dispersal distances together for the variation in the quality of
the regenerated habitat.
65
Landscape changes pathways
In order to investigate land-use intensification pathways in the sampled landscapes, we ran a
Principal Component Analysis (PCA, Supplementary Table 3.4 and Supplementary Figure 3.2)
with the land uses (pastures/agricultural fields, Eucalyptus plantations, shaded cacao and urban
areas) from the three different years (1990, 2000 and 2007) based on a covariance matrix. We
then link sampled landscapes loadings in the different years analyzed, and used the differences in
axis 1 and in axis 2 as metrics of land-use change.
Among our evaluated landscapes, there are four temporal dynamics that we observed; (i)
low-intensified landscapes that stayed as low intensified landscapes thru all the evaluated period
(e.g. landscape 54 to 59); (ii) low-intensified landscapes that became highly intensified in the
first period of time evaluated (e.g. landscape 54 to 59); (iii) low-intensified landscapes that
became highly intensified ones in the second time-step evaluated (e.g. landscapes 8, 16, 26, 39,
46 and 47); (iv) and the ones that changed in both periods of time (e.g. landscapes 11 to 13, 28,
29 and 35) (Supplementary Figure 3.2). The first PCA axis explains 83.4 % of the variation, and
is strongly negatively correlated with the amount of pastures (Supplementary Table 3.4). The
second PCA axis explains an addition of 11.4%, and is strongly positively correlated with the
amount of Eucalyptus plantations and negatively correlated to the shaded cacao plantations
(Supplementary Table 3.4).
66
Table 3.4: Land-use contribution for each of the two first PCA axes, the proportion of variance
and the cumulative variance.
PC1 PC2
Shaded cacao 0.14 -0.54
Pastures -0.98 0.06
Eucalytpus 0.16 0.84
Urban 0 0.01
Proportion of Variance 0.83 0.11
Cumulative Proportion 0.83 0.95
67
Figure 3.2: Changes in land-use through time of the studied landscapes. The top panel shows the
59 landscapes in the three times evaluated, the black lines unit the landscapes in t1 (1990) and t2
(2000), whereas the red line between t2 (2000) and t3 (2007). The bottom panel showed the
68
proportion of change in terms of proportion in the PCA axis 1 (black) and PCA axis 2 (red)
between t1-t2(circles) and t2-t3 (crosses).
69
- Forest and land-use dynamics in the Amazon: Convergent effects of human activities
Abstract
1. Tropical forest loss and fragmentation could be transient, since forest regeneration is very fast
unless prevented by continued disturbance. Hence, a large fraction of the altered area by
human activities can quickly revert to secondary forests. In this context, fragmented
secondary forests are particularly relevant for biodiversity conservation.
2. Here, we investigate the spatio-temporal dynamics of fragmented landscapes in the Brazilian
Amazon, particularly how forest and land-use respond to different histories of management,
amounts of pristine forests, human population sizes, and economy (global crises of 2008,
county GDP and per capita GDP).
3. We evaluated 11 counties that are located in two different regions in the Amazon having
similar biophysical characteristics but different management histories and amounts of pristine
forests in 2012 (7 with ~ 30% and 4 with ~70%). We compared maps from 2004, 2008 and
2012 to evaluate the spatial attributes (e.g., size, shape, number of patches and fragmentation)
of patches of forest and pastures that are lost, gained or remained stable through time. We
then, evaluate the spatio-temporal dynamics of forests and pastures, particularly in terms of
landscape connectivity that can sustain individual movements in fragmented landscapes, thus
facilitating the maintenance of animal species in such environments. Finally, we tested how
forest and pasture dynamics are affected by cattle numbers, per capita GDP, and human
population sizes for each county, for each evaluated period (2004-2008 and 2008-2012).
4. Our results pointed to a reduction of spatio-temporal dynamics across time, particularly in the
30% pristine forests counties, where the numbers of forest patches that are gained and lost
decreased, and the patches that were gained were of reduced size, generating landscapes that
are more homogeneous through space and time. Overall, pasture gains are more clumped than
forest gains resulting in more fragmented forests, whereas pasture gains increase in size over
the 8-year of the study. Regression models with cattle numbers and human population sizes
better explain pasture dynamics. However, the spatio-temporal connectivity influences differ
according to the per capita GDP.
70
5. We showed that (i) landscapes in the Amazon are progressively becoming more homogeneous
through space and time; (ii) this is caused by a reduction in the forest dynamics, associated
with smaller patches of forest gains, and a reduction in numbers of gains and losses of pasture
patches (though pasture gains were of a larger size); and (iii) pasture expansion results in
more fragmented landscapes that are less prone to biodiversity conservation. In closing, we
suggest strategies to sustain spatial and temporal dynamics in key selected areas, such as
among protected areas, or surrounding first nation lands.
Introduction
Tropical forests originally covered roughly 7% of the global terrestrial surface yet harboured
around 50% of terrestrial species. Currently, half of global forest losses occur in the Tropics
(Hansen et al. 2013), due to tropical forests being converted into agricultural areas at a fast rate
since the 1980s (Gibbs et al. 2010). From 1950 to 2000, it is estimated that at least 350 million
hectares of tropical forests were clear-cut (ITTO 2002), resulting in around 50% of the original
tropical forest converted to other land uses (Wright 2005). This forest loss affects global carbon
and hydrological cycles, as well as biodiversity conservation and the livelihoods of millions of
people across the region (Wright 2010).
In many tropical habitat fragmentation studies it is assumed that converted areas stay
locked up as farmlands for a long time, while the enduring habitat remains fragmented. This
ignores that tropical forests regeneration is particularly fast (Aide et al. 2000), and unless
prevented by continued disturbance, a large fraction of the converted area can quickly revert
back to secondary forests (Corlett 1995). Nowadays, roughly 60% of the total area classified as
forest in the tropics is expected to be second-growth forest or degraded primary forest (ITTO
2002). As such, most of the tropical forest countries currently have more secondary forests than
pristine ones (Brown & Lugo 1990). Although the Amazon still harbors a large fraction of
pristine forest, the accumulated pristine Amazonian forest loss in Brazil alone is around 1 million
km2 (INPE/PRODES 2015). Nevertheless, in the Amazon, second-growth forests already cover
almost one fourth of this 1 million km2 and are steadily expanding (INPE/TerraClass 2016).
Therefore, it is clear that fragmented landscapes in the tropics need to be considered under a
71
dynamic framework (Chapter 2), departing from the more static traditional view of forest
fragmentation assessments.
The Amazon has long been viewed as one of the top deforestation regions on the planet
(Kim, Sexton & Townshend 2015), however, bioclimatic factors, as well as low intensity and
ephemeral land-use, also make the region a place of intense forest regeneration. The landscape
spatial patterns of forest loss and its anthropogenic drivers are relatively well studied (e.g.,
Nepstad et al. 2001, 2009, Laurance et al. 2001, 2002; Soares-filho et al. 2006; Fearnside et al.
2012), however, the spatial patterns of forest recoveries and their driving processes are less
understood. Yet, land-use dynamics in the Amazon are known to follow an economic boom–bust
cycle (Homma & Furlan Jr. 2001), with widespread land abandonment and intense forest
regeneration during economic recessions, sometimes, surpassing the amount of forest losses
(Perz & Skole 2003). Additionally, given its vast extent and settlement history, different parts of
the Amazon experience the multi-scale processes that drive landscape dynamics in different
ways, directly influencing landscape spatial patterns (Perz & Skole 2003). Furthermore, at any
single time, spatial landscape patterns are the result of combined effects of forest losses and
gains resulting in intricate complex spatial arrangements. These dynamic anthropogenic mosaics
contain uneven amounts of forest, scattered into dynamic patches of different sizes, connectivity
levels, and successional states, that are immersed in variable agricultural matrices having
different permeabilities for animal movement and native forest regeneration likelihoods (Hanski
2011; Haddad et al. 2015).
For instance, forest loss and/or regeneration can change forest patch size, shape and
habitat quality, altering population and community structures (Bregman et al., 2014). Forest
regeneration can occur either as part of a forest fragment, or as an isolated patch, sometimes
elongated along riverine systems (Teixeira et al. 2009), all potentially acting as a stepping-stone
(Uezu, Beyer & Metzger 2008) or corridor (Martensen, Pimentel & Metzger 2008) favoring
animal movement between more established forest patches. Hence, these regenerated patches
could facilitate organismal movements between fragments, resulting in enhanced connectivity
when considering the temporal perspective (Chapter 2). Yet, additional reductions of forest
amounts could act in an opposite direction, reducing the available forest habitat and the
connectivity among patches.
72
These changes in forest habitat and their effects on species movements are not always
immediate, and might be influenced by the dynamics of spatial changes, potentially causing
extinction debts or immigration credits where biological systems have not yet adjusted to the
new spatial conditions (Tilman et al. 1994; Kuussaari et al. 2009; Essl et al. 2015). Such spatial
changes, as well as delays in species extinctions and colonisations, ends up influencing overall
species distributions and turnovers (Jackson & Sax 2010). Hence, to comprehend biodiversity
patterns in changing fragmented landscapes, it is important to understand the spatio-temporal
dynamics that generates the spatial and ecological patterns, as well as the drivers of these
dynamics.
The goal of this study is to evaluate how forest and pasture spatio-temporal dynamics
respond to different types and intensities of anthropogenic drivers in the Brazilian Amazon.
Specifically, we selected 11 counties, having different percentages of pristine forest remaining:
seven counties with around 30% and four counties with 70%. These counties were spread among
two regions of the Brazilian Amazon (see Perz & Skole, 2003 for additional detail); one that is
considered a “frontier” region (i.e., a region with rapid land settlement along new roads since the
1960s, and particularly after 1980), and a “remote” region (i.e., limited settlement until 1990s,
mainly in traditional riverine communities and few cities, with a recent expansion in non-
traditional settlement). We evaluated the spatial dynamics of these counties across two different
periods in time, from 2004 to 2008, and from 2008 to 2012. These two time-steps reflect very
distinct periods of Amazon deforestation history. Until 2007, the values of agricultural and
ranching production in the Amazon were highly correlated with deforestation rates (Barreto &
Silva 2012), and fluctuated as a function of agricultural commodity prices (Silva 2009).
However, from 2008 onwards, the value of agriculture production kept growing faster, whereas
deforestation rates decreased (Barreto & Silva 2012). This was due to different combined
important factors, such as increases in agricultural commodities prices after 2008 (OECD/FAO
2011), and enhancements in production (Macedo et al. 2012), both by applying new
technological developments, and given the utilization of vast newly open and still highly
productive areas (Barreto & Silva 2012). On a global scale, 2008 was marked by the start of an
intense global economic crises that has multiple effects on different levels of the economy (Kotz
2009). Little is known about the effects of this economic crisis over landscape dynamics in the
Amazon, other than that 2008 was the last year where pristine forests conversions where above
73
10000 km2, and was followed by an intense drop in pristine forests conversions in 2009, which
were followed by constant smaller decreases until 2012 (PRODES/INPE 2016). Actually, in
2012 the smallest amount of pristine forest deforestation was observed (4571 km2) in the
Brazilian Amazon since 1988, when the measures started to be conducted annually (INPE 2016).
After 2012, there was some variation from year to year, but the amount of pristine forests
deforestation remained fairly stable from 2013 to 2015 (5891, 5012 and 6207 km2), therefore, it
appears to have reached a new basal limit, that will demand additional and different strategies for
further reductions. Hence, understanding the spatial dynamics during these two periods could
help us to better comprehend the current drivers influencing these dynamics. This could support
new strategies for additional reductions in forest losses, and particularly biodiversity
conservation in fragmented landscapes in the region.
Methods
4.3.1 The study region
We applied a subdivision proposed by Perz & Skole (2003), where they split the Amazon into 3
different regions: settled, frontier and remote. We then randomly selected 11 counties in the
frontier and remote regions that have proportions of remaining pristine forests varying between
25 and 35 (hereafter 30%) and between 65 and 75% (hereafter 70%), based on the PRODES
dataset from 2012 (INPE/PRODES 2015). The counties varied in size from 1988 to 12295 km2
(mean = 5465 km2). In addition, we carefully selected counties based on their spatial location;
therefore, they were from a similar latitude in order to standardize some of the biophysical
characteristics that could influence forest regeneration. These counties belong to two different
states, Acre and Mato Grosso, seven with 30% (Vila Rica, Castanheira, Sinop, Vera, Carlinda,
Senador Guiomard and Plácido de Castro), and four with 70% (Marcelândia, Nova Bandeirantes,
Rio Branco and Brasiléia). In the remote area, two counties have 30% of forests, whereas the
other two have 70% of forests. In the frontier area, five counties have 30% of forest, whereas two
have 70%. Using these two different sets of amount of forest landscape (30 and 70%), we
compared the dynamics of forests and pastures when they are present in similar proportions. In
other words, we compared the dynamics of forest where pristine forests cover 30% of the
74
landscape with the dynamics of pastures where pastures occupy 30% of the landscape (i.e. 70%
of pristine forests).
We started by first simplifying the TerraClass maps for these counties, by reducing to
only two variables: pastures and forests. The pasture class was composed by the original “clean
pastures” and “pastures with scattered trees” classes (respectively, “pasto limpo” and “pasto
sujo”), and the forest class is composed of “forest”, “secondary forests” and “pasture with
regeneration” classes (respectively, “floresta”, “vegetação secundária” and “regeneração com
pasto”). The remaining classes, such as agriculture, urban or water, were considered as other
matrix, as well as the areas covered by clouds.
Then, we overlaid these reclassified maps for each county from 2004 and 2008, and the
maps from 2008 and 2012. These series of composite maps allowed us to assess the amount and
spatial characteristics of forest and pastures losses, gains, and stability among periods (2004-
2008 and 2008-2012). For example, a forest fragment could lose two different portions of the
fragment; therefore, we call each of these parts as a patch of forest loss. The same happens for
forest/pasture gains, and the concomitant process of forest/pasture losses and gains could split or
unite stable patches. The spatial patterns of forest/pasture patch losses, gains and stability are
evaluated in the first part of the analysis (e.g., size, shape, edge amounts). In the second part of
the analysis, we evaluate the spatio-temporal dynamics of these changes and their influence on
connectivity dynamics. Finally, in the last part, we evaluate some of the potential socioeconomic
drivers of these dynamics.
4.3.2 Drivers of land-use change
We selected four variables as surrogates for human drivers of land-use change dynamics on these
11 counties: (1) number of beef cattle; (2) human population size; (3) gross domestic product
(GDP), and (4) GDP per capita.
The beef cattle numbers in the 11 counties grew from approximately 2.51 million in
2004, to 2.82 million in 2008, to 3.15 million in 2012 (IBGE). The number of cows per county
varied greatly, from 17537 in 2004 in the county of Vera to 709879 in 2012 in the county of Vila
Rica. Moreover, the cattle dynamics differ among counties. Some counties, such as Vila Rica
75
and Nova Bandeirantes, presented an almost steady and strong increase in cattle numbers. But
other counties presented a more complex pattern, such as Sinop and Senador Guiomard, that
increased and then decreased in the second period the cattle herd numbers. In 2012, in the county
of Senador Guiomard, the cattle herd numbers dropped lower than their 2004 value. Rio Branco
and Carlinda showed a decrease in their cattle numbers between 2004 and 2008, and then an
increased between 2008 and 2012. Even with this increase, Carlinda’s 2012 cattle numbers were
lower than in 2004.
Human population also increased in the counties from just over half a million to over 600
thousand from 2004 to 2012 (IBGE), an increment of around 20% in just 8 years. Nine counties
have human populations lower than 20,000 inhabitants, whereas one has almost 100,000
inhabitants. The largest county, Rio Branco, the capital of the State of Acre, had 286082
inhabitants in 2004. Human population patterns were fairly constant. In 2012, 6 counties had less
than 20000 inhabitats, whereas three of them had between 20000 and 23000, one had almost
120000 and the larger one almost 350000 inhabitants. Most of the counties show a steady
increase in population over the evaluated periods, however, two counties (Marcelândia and Vera)
showed a reduction in human population from 2004 to 2012. Some counties presented a strong
increase in population size, such as, Nova Bandeirantes (~38%) and Brasiléia (31.5%).
All counties increased in GDP, some of them were 4 times higher in 2012 than in 2004,
like Plácido de Castro and Nova Bandeirantes. The mean increase of GDP was 2.8 times more in
2012, compared to 2004. The county that showed the smallest increase in GDP was Marcelândia,
with 50% increase from 2004 to 2012, followed by Castanheira and Carlinda. The increase in per
capita GDP was also huge, but lower than the increase in GDP itself. Per capita GDP was 2.48
times higher in 2012 than in 2004, with the largest per capita GDP increase in Plácido de Castro,
and the smallest in Castanheira, Carlinda and Vila Rica. The smallest GDP was only the fifth
smallest per capita GDP, with 2.28 times increases from 7878.00R$ in 2004 to almost
18000.00R$ in 2012.
Therefore, we opted to not include GDP as it is highly correlated with the population size
and thus, we used three explanatory groups of variables for each county: cattle numbers in each
year, population size in each of the years, and per capita GDP for each of the years. The selected
76
counties presented ample variation amongst the drivers that were chosen as potentially
influencing landscape dynamics in the region, which allowed us to evaluate the independent
effects of theses variables on landscape features dynamics across the frontier region and the
remote region, which encompasses the largest fraction of the Brazilian Amazon.
4.3.3 Spatial patterns of forest and pastures losses, gains and stability
In order to evaluate the spatial patterns of forest/pasture losses, gains, and stability, we calculate
the following metrics for each of the overlaid maps obtained from each pair of years (2004-2008
and 2008-2012), for each county: number of patches, mean patch area, proportion of the
landscape, landscape shape index, largest patch index, maximum patch area, aggregation,
splitting index and effective mesh size. All of these metrics were then compared considering the
period, the class (pasture or forest), based on forest proportion and on region (remote and
frontier), and then, compared among classes (pasture and forest).
4.3.4 Spatio-temporal dynamics evaluation: losses, gains and stability
In order to evaluate the dynamics of forest/pasture loss, gain and stability, and the effects of
spatio-temporal connectivity on these dynamics, we evaluate the spatial and temporal
connections based on a spatio-temporal network approach (Chapter 2). In this approach, the
patches are represented as nodes, and their potential connections as links or edges (Fall et al.
2007). Landscape dynamics are characterized by changes in both nodes (e.g., patch – expansion
or shrinking) and links (e.g., multiple species dispersal abilities). A spatio-temporal link is the
possibility of flow from a given patch location at time 1 (t1) to a different patch location later (t2).
For each time-step (i.e. pair of years), patches were classified into three states: Loss (patch lost
from t1 to t2); Gain (patch appears between t1 and t2); and Stable (patch exists at t1 and t2). We
assume that time-steps are short enough such that no more than one type of patch change is
possible between t1 and t2 (e.g. consecutive loss and gain) for any given location. The spatio-
temporal link among patches can be of two forms: (i) direct path, i.e. a single link movement
from a patch at t1 to another patch at t2, or (ii) partial stepping-stone path, i.e. multiple links
(steps) allowing the flow from a patch at t1 to another patch at t2, using one or more intermediate
stepping-stone patches at time tx (t1 ≤ tx ≤ t2).
77
Based on this model framework, we calculated the spatio-temporal probability of
connectivity (PCst) and the Equivalent Connected Area by using patch area as attribute, ECAst
(sensus Chapter 2). These metrics account for: (a) the area of each patch (intra-patch
connectivity) and (b) the amount of patch area reached by moving to other patches through the
links in the network (inter-patch connectivity). Given two patches (source patch at t1 and
destination patch at t2) randomly selected, PCst is defined as the probability that the source and
the destination patches are spatio-temporally connected such that an individual located in the
source patch at t1 can move to destination patch at t2. The spatio-temporal connectivity PCst can
be decomposed into three fractions: PCintrast (intra-patch connectivity through time), PCdirectst
(inter-patch connectivity provided by direct spatio-temporal connections between patches), and
PCstepst (interpatch connectivity accounting for stepping-stones connections between the source
and destination patches, see Chapter 2 for additional details).
We tested different values of spatial constraints by using a negative exponential function
of inter-patch distance to obtain the probability of direct movement between any pair of patches.
We considered inter-patch habitat as uniform, therefore, we used Euclidian distance between the
patches (edge to edge). The kernel dispersal function was parameterized so that it gave a 0.5
probability of movement (gap crossing) between patches, when the patches were separated by an
edge-to-edge distance equal to the considered median dispersal distance. After a careful
evaluation of the spatial characteristics of the maps and a literature search of the dispersal
capacity of species in the region, we opted to use a dispersal distance of 100 m. Calculations
were performed by combining an R script with a command line version of the ConeforST
(available at http://www.conefor.org/files/usuarios/conefor_directed.zip).
For this part of the analysis we did not considered Nova Bandeirantes, therefore, only 10
counties were considered. We compared the spatio-temporal metrics with the purely spatial
metrics obtained for the first and second time-step for each evaluated period.
4.3.5 Socio-economic drivers of landscape dynamics
In order to evaluate the effects of county economic dynamics (gross product), total human
population size, cattle herd size per county, national cattle exports, price of agricultural
commodities (soy and beef), and the effects of the global economic crises of 2008 on the patterns
78
and in the dynamics of forest/pasture loss, gain and stability, we evaluate how well these
socioeconomic drivers can predict the changes in spatial patterns and on spatio-temporal
dynamics in our study counties.
79
Results
4.4.1 Patterns of forest loss, regeneration/gain and stability
4.4.1.1 30% of class (pasture or forest) cover
In the first evaluated period (2004-2008), the most common patch type is pasture gain (5614
patches), followed by forest loss (4082 patches) and forest gain (2063 patches). In the second
time period, there were much less patch loss and gain of any kind, and pasture loss dominated
(2594.5 patches), followed by pasture gain (2230 patches) and then forest loss (1776 patches).
Forest loss and gain were of a similar size (6.39 and 5.36 ha) during the first period, whereas
pasture loss and gain were much larger (9.83 and 17.95 ha). Forest-patch gain decreased in size
(4.07 ha), whereas forest-patch loss increased (7.86 ha) in the second period, whereas an
opposite trend happened for pasture gain (13.38 ha) and loss (9.7 ha). In the first period, forest-
patch gain accounted for a median of 3.9% of the area of the counties, and the median maximum
area of the largest patch gain was 964.75 ha (Table 4.1, Supplementary Figure 4.2). In the first
period, the effective mesh size for pasture gain is almost three times higher than for forest gain,
and it reduced for forest gain, and increased for pasture gain in the second (Table 4.1). The
effective mesh size for pasture and forest loss were both higher in the first period, than in the
second, and also higher than the gain (Table 4.1).
In the second period (2008-2012), forest-patch gain accounted for only 1.7% of the
county areas, with a maximum patch gain of about one third of the first period (Table 4.1). Forest
losses accounted for a much larger proportion of the counties (7%) in the first period, however,
this value declined to less than half that value in the second period (3.2%). The median largest
patch of forest loss was just over 1000 ha in the first time and was also reduced to 728.5 ha in the
second time period (Table 4.1). Pasture gains accounted for 4.6% of the counties, whereas losses
were 2.8 % of counties in the first period, whereas for 3.1 and 3 % of counties respectively in the
second. In the first period, the median largest patch area of pasture gains was larger than the
forest ones, as well as the losses, which were more than three times higher than the pasture losses
(Table 4.1). In the second period, the median of the maximum size of pasture gains were more
than seven times higher than forest gains, whereas forest (728.5 ha) and pasture (785.25 ha)
80
median maximum patch losses areas were similar (Table 4.1, Supplementary Figure 4.2). Pasture
gains were more aggregated than forest gains, with pasture gains presenting a higher landscape
shape index, splitting index and aggregation (Table 4.1).
4.4.1.2 70% of class (pasture or forest) cover
Forest loss dominated the first period (6566.5 patches), followed by pasture gain (4272 patches),
and forest gain (2952.5 patches), whereas in the second period, forest gain where the most
common patch type (4778 patches), followed by forest loss (2685 patches). The mean forest
patch gain (6.05 and 7.41 ha) and loss (7.46 and 8.63 ha) were similar in both periods, however,
forest loss was somewhat higher than gain and both were slightly higher in the second period
(Table 4.1, Supplementary Figure 4.2). The median mean patch of pasture gain was smaller in
the first period (5.84 ha), compared to the second one, which was almost three times bigger
(15.84 ha). The pasture loss behaved in an opposite direction, with the sizes in the second period
half of the ones in the first (Table 4.1). The median of the larger patches of forest that were
gained in each county was almost two times bigger in the first period compared to the second,
whereas the biggest patches of forest loss, also reduced but not as much as the gain (Table 4.1).
In the case of the pasture, the largest patches of pasture gain increased from the first to the
second periods (respectively, 1050.75 and 1319.5 ha), whereas the larger patches of pasture that
were loss decreased (respectively, 1816.5 and 695.75 ha). Forest gain comprised 2.2% of the
counties, whereas forests loss comprised 4.3% in the first period. Forest gain covered a larger
area than forest losses in the second period (respectively 3.2 and 2.2%). Pasture gain covered
6.7% of the counties and loss 5.4% in the first period, whereas the gain covered 6.1% and loss
2% in the second period (Table 4.1). Forest loss is more aggregated (80.2%) and presented a
larger effective mesh size (12.9 ha), than forests gain (respectively 73.2% and 2.5 ha), which also
reflected in a disproportional larger splitting index for forest gain (Table 4.1). In the second
period, both forest loss and gain presented a similar aggregation, effective mesh size and splitting
index (Table 4.1, Supplementary Figure 4.2). Pasture loss and gain presented a similar
aggregation index in both periods, however, they were slightly higher in the second period.
Pasture gain effective mesh size is 11.3 ha and loss 15.1 ha in the first period, however pasture
81
loss mesh size reduced to almost one fourth in the second period (4.5 ha), whereas increased for
gain to 15.4 ha. Comparing with forest loss and gain, forests presented a considerably higher
splitting index than pastures for all cases (Table 4.1).
82
Table 4.1: Results of the class metrics summarized by the median obtained across the 11 counties for each category and period.
% Class Years Patch
type
Number of
patches
Mean patch
area (ha)
Max.
patch area
(ha)1
% of
the
county2
Landscape
Shape Index
Splitting
index
Effective
mesh size
(ha)
Aggregation
%
30 %
of
each
cla
ss
Fore
st
2004-2008
gain 2063 6.39 964.75 3.9 57.4 101240.07 4.8 74.4
loss 4082 5.36 1050.75 7 63.5 31276.88 12.7 78.9
stable 1291 94.33 77954.25 46 39.2 30.51 20811.1 95.9
2008-2012
gain 775 4.07 326.25 1.7 27.2 361362.98 1.3 74.3
loss 1776 7.86 728.5 3.2 55.4 92795.34 3.7 77
stable 1584 79.86 74637.25 45.3 45.6 30.37 16262.7 95.1
Pas
ture
2004-2008
gain 5614.5 9.83 1656.5 4.6 78.2 122976.74 13.5 82.8
loss 1319.5 17.95 3400 2.8 54.9 69183.48 18.7 83.9
stable 764 189.32 43134.38 18.1 40.3 373.05 2517 95
2008-2012
gain 2230 13.38 2374.12 3.1 59.3 111191.3 14.1 84.1
loss 2594.5 9.7 785.25 3 57.2 246042.84 4.4 80
83
stable 874 164.85 48356.5 22 41.6 311.69 3182.1 94.6
70 %
of
each
cla
ss
Fore
st
2004-2008
gain 2952.5 6.05 1199.12 2.2 77.6 527957.8 2.5 73.2
loss 6566.5 7.46 1369.12 4.3 88.3 189581.08 12.9 80.2
stable 2047 415.5 776391.5 77.3 26.3 2.15 529211.3 98.3
2008-2012
gain 4778 7.41 693.62 3.2 77.4 305750.73 3.8 78.1
loss 2685 8.63 1067.88 2.2 67.1 371465.20 3.2 78.5
stable 1984 372.39 788358.88 79.4 28.1 2.04 545669.6 98.3
Pas
ture
2004-2008
gain 4272 5.84 1050.75 6.7 64.7 26619 11.3 78.4
loss 1325 12.91 1816.5 5.4 52.5 29017.31 15.1 78.8
stable 431 390.03 82352 45.7 29 12.43 26366.3 95.7
2008-2012
gain 1658 15.84 1319.5 6.1 48.5 36558.05 15.4 82.6
loss 776 6.29 695.75 2 28.1 57508.46 4.5 82.3
stable 320 444.42 102762 51.3 28 9.07 46830.2 96.7
84
%: The percentage of pristine forest present in the county in 2012; Class: land-use/cover class evaluated; Years: Period evaluated, first
period 2004- 2008, and second period 2008-2012; Patch type: if it is a patch of forest/pasture gain, loss or stable. 1: The median of the
maximum patch area in the given category; 2: The percentage of the county cover by this category and 3: Proportion like adjacencies.
85
4.4.2 Patterns of landscape dynamics
As expected, the counties with high percentage of pristine forest (~70%) have a larger portion of
the connectivity composed by intra-patch connectivity through time (PCintrast), however, when
pasture cover around a similar proportion, the fraction composed by intra-patch connectivity is
much smaller, around half what was observed for forest (Table 4.2). The PCdirectst and PCstepst
fractions were much smaller for forest in 70% pristine forested landscapes, and the connections
by stepping-stones were more important (Table 4.2). When pasture covers larger portions of the
counties, PCstepst was the most important fraction, followed by PCintrast and by PCdirectst
(Table 4.2). In the first period 2004-2008, for the counties with less amount of forest (~30%), the
PCstepst was by far the most important fraction. However, for the second period 2008-2012, the
PCintrast presented a somehow higher importance, in some cases, similar with the PCstepst,
although PCstepst was still the most important fraction (Table 4.2). For pasture, the PCstepst was
more relevant in the first period, however, PCintrast was also important in the second period,
with the exceptions of a few counties such as Sinop, Vera and Castanheira (Table 4.2).
86
Table 4.2: Results of the PCfractions (PCintrast, PCdirectst and PCstepst) for each county per period.
Counties
Forest Pasture
% of
Forest
Region 2004-2008 2008-2012 2004-2008 2008-2012
intra direct step intra direct step intra direct step Intra direct step
Brasiléia 81.38 7.04 11.58 77.04 8.10 14.86 26.53 15.04 58.43 20.11 10.68 69.21 H R
Carlinda 10.50 3.29 86.22 15.15 5.58 79.27 40.73 11.66 47.61 79.90 17.11 2.99 L F
Castanheira 14.15 14.71 71.15 23.53 20.06 56.41 20.57 16.53 62.90 32.56 16.68 50.75 L F
Marcelândia 70.73 12.99 16.28 85.02 5.94 9.04 18.76 13.83 67.41 20.14 19.83 60.02 H F
Plácido de Castro 14.69 10.18 75.14 18.93 22.16 58.91 34.20 15.25 50.55 41.03 25.85 33.12 L R
Rio Branco 84.65 3.60 11.76 83.73 4.13 12.14 13.80 15.86 70.34 36.34 10.14 53.52 H R
Sen. Guiomard 28.45 10.49 61.06 43.75 11.72 44.53 36.17 11.95 51.88 48.50 29.06 22.44 L R
Sinop 18.86 18.82 62.32 28.30 28.84 42.87 24.36 20.89 54.75 6.69 13.30 80.01 L F
Vera 20.90 18.26 60.84 36.91 15.50 47.59 6.85 12.46 80.69 4.03 62.08 33.89 L F
Vila Rica 21.79 21.91 56.30 44.69 5.00 50.31 46.94 18.31 34.75 70.32 16.41 13.27 L F
87
4.4.3 Drivers of landscape dynamics
The cattle numbers in 2004, 2008 and 2012 were highly significantly linearly correlated with
each other (r > 0.9), and presented a low positive correlation with population size (cattle for 2004
r ~ 0.3, cattle 2008 r ~ 0.13 and cattle 2012 r ~ 0.22) and with GDP (cattle 2004 r ~ 0.26, cattle
2008 r ~ 0.07 and cattle 2012 r ~ 0.13). However, cattle numbers were negatively low correlated
with per capita GDP (2004 r ~ -0.18, 2008 r ~0.44, 2012 r ~ -0.44). Human population size was
highly correlated between years (r ~0.99) and also with GDP (r > 0.98). However, population
size presented no significant correlation with per capita GDP (2004 r ~- 0.01, 2008 r ~ 0.08 and
2012 r ~ -0.10). GDP was highly correlated among years (r ~ 0.99), however, not significantly
correlated with the per capita GDP (2004 r ~ 0.05, 2008 r ~ 0.15, 2012 r ~ - 0.01).
The forest and pasture dynamics of the region are high, and these dynamics influence
spatio-temporal connectivity (Figure 4.1). The spatio-temporal connectivity of pastures had an
ample variation of influence, varying between extremely negative, which means a much stronger
influence of the purely spatial connectivity, to a strong (almost 300% addition in connectivity)
influence of the spatio-temporal dynamics on connectivity patterns. The median influence of the
spatio-temporal connectivity over the beginning of each time period was 34 and 25%, whereas
the second half of each period was much smaller, and negative for the second period (Figure
4.1). There is a general tendency of a reduction of influence of spatio-temporal connectivity of
pastures across time, both between 2004-2008 and 2008-2012, and among the comparisons of
the spatio-temporal connectivity between the first and last years of each time-step (Figure 4.1).
Overall, spatio-temporal connectivity of forests was less variable and predominantly positive,
suggesting that spatio-temporal dynamics usually positively influence spatio-temporal
connectivity (Figure 4.1). The median of the spatio-temporal connectivity in comparisons with
the first years were smaller (seven and 4%), than when compared with the second year (37 and
29%). In general, the influence of spatio-temporal dynamics on connectivity were slightly
smaller between 2008-2012, compared with the 2004-2008 period, and higher when compared
with the first few years of these time periods, than to subsequent years (Figure 4.1).
None of the drivers were significantly influential in spatio-temporal forest dynamics
(PCnumst) or in the PCfractions (intra, direct and step). From another side, cattle numbers had a
strong positive influence over the spatio-temporal dynamics of pastures. In 2004, population size
88
positively impacted spatio-temporal dynamics, and 2008 and 2012 population size were both
negatively influencing spatio-temporal dynamics (Table 4.3). Per capita GDP from all years were
of minor effects on PCnumst. PCintrast fractions were also strongly positively associated with
cattle numbers, whereas less negatively influenced by population sizes, and again, per capita
GDP was of no influence (Table 4.3). From another side, when considering the percentages of
PCintrast, per capita GDP was the most influential metric, followed by cattle and population
size. However, the null model was among the best model selected for all years, suggesting that
none of the models strongly explained the data. For the PCdirectst fraction, cattle numbers,
followed by population size were the best supported models. In these cases, cattle had a positive
influence, whereas population size mostly had a negative one. For the PCdirectst percentage, per
capita GDP was the most influential variable for 2008 and 2012, however, not for 2004. For
PCstepst, the variable cattle numbers was the most significant explanatory one for PCstepst
variation; however, for 2008 and 2012, the null models were among the selected best models (Δ
AICc < 2, Table 4.3). Finally, for PCstepst percentages, the null model was among the best
supported models for all years, however, in 2004 per capita GDP and cattle numbers were also
selected among the best models.
89
2004-2008 2008-2012 F
ore
st
Pas
ture
Figure 4.1: Histograms of the influence of the spatio-temporal connectivities over the purely
spatial ones [((spatiotemporal/purely spatial)-1)*100], for forests and pastures per period. The
dashed lines represent the comparisons between the spatio-temporal and the first year of purely
spatial metrics, whereas the solid line represents the comparisons among the spatio-temporal and
the second year purely spatial metrics.
90
Table 4.3: Selected models (Δ AICc < 2) explaining the variation of PCnumst, PCintrast,
PCintrast %, PCdirectst, PCdirectst %, PCstepst, PCstepst %, for each explanatory variable
importance for each model selection processes, and if it has a positive or negative relationship.
Variable Years Selected models (Δ AICc <2),
AIC weights and Δ
Cattle Population Per capita
Impor
t
signal impor
t
signal impor
t
signal
PCnumst
2004 Cattle + Population (w=0.723) 1 + 0.74 + 0.04 -
2008
Cattle + Population (w=0.63)
Cattle (w=0.34, Δ = 1.23)
1 + 0.64 - 0.03 -
2012 Cattle + Population (w=0.805) 1 + 0.82 - 0.02 -
PCintrast
2004
Cattle + Population (w=0.512)
Cattle (w=0.42, Δ = 0.32)
0.96 + 0.52 - 0.03 -
2008
Cattle (w=0.609)
Cattle + Population (w=0.342,
Δ = 1.15)
0.99 + 0.35 - 0.04 -
2012
Cattle + Population (w=0.564)
Cattle (w=0.393, Δ = 0.72)
0.99 + 0.57 - 0.03 -
PCintrast
%
2004
Per capita (w=0.382)
Intercept (w=0.285 , Δ = 0.58)
0.19 + 0.19 - 0.52 -
2008
Cattle (w=0.328)
Per capita (w=0.309, Δ =0.12)
0.44 + 0.07 - 0.41 -
91
Intercept (w=0.207, Δ = 0.92)
2012
Per capita (w=0.424)
Cattle (w=0.237, Δ = 1.16)
Intercept (w=0.205, Δ = 1.46)
0.32 + 0.08 - 0.51 -
PCdirect
st
2004
Cattle (w=0.607)
Cattle + Population (w=0.325,
Δ = 1.25)
1 + 0.34 + 0.07 -
2008
Cattle + Population (w=0.658)
Cattle (w=0.308, Δ = 1.52)
0.99 + 0.67 - 0.02 -
2012
Cattle + Population (w=0.688)
Cattle (w=0.254, Δ = 2)
0.96 + 0.70 - 0.03 -
PCdirect
st %
2004 Intercept (w=0.681) 0.108 + 0.135 + 0.094 +
2008
Per capita (w=0.567)
Per capita + Population
(w=0.305, Δ = 1.24)
0.07 0.32 0.92
2012 Per capita (w=0.775) 0.05 - 0.15 - 0.96 +
PCstepst
2004 Cattle (w=0.88) 1 + 0.05 + 0.07 -
2008
Cattle (w=0.485)
Intercept (w=0.352, Δ = 0.64)
0.53 + 0.07 + 0.09 -
2012 Cattle (w=0.469) 0.52 + 0.07 + 0.10 -
92
Intercept (w=0.355, Δ = 0.56)
PCstepst
%
2004
Intercept (w=0.359)
Per capita (w=0.328, Δ = 0.18)
Cattle (w=0.136, Δ = 1.95)
0.21 - 0.13 + 0.41 +
2008 Intercept (w=0.611) 0.169 - 0.157 + 0.097 +
2012 Intercept (w=0.625) 0.155 - 0.156 + 0.099 +
93
Discussion
The spatial heterogeneity of the 11 counties is highly dynamic; however, we observed a general
tendency of reduced landscape heterogeneity and dynamics from the first period (2004-2008) to
the next one (2008-2012), resulting in an overall decrease of spatial and spatio-temporal
connectivity across time. The multiple events of forest and pasture losses and gains across time
resulted in less amount of forests in the counties than in 2004, and more isolation of forest
patches by 2012. Counties that had 30 or 70% of pristine forests were at different stages of
historical changes and therefore showed different spatial heterogeneity and dynamic patterns. For
instance, in counties that had 30% of pristine forests, there was a general tendency of a reduction
in forest spatial dynamics, seen as a function of the reduced number of the forest patch gain and
loss, and an increase in patches that were stable (Table 4.1). These reductions in numbers were
also accompanied by a reduction in the sizes of patches that are gained and an increase in the size
of the patches that were lost. This resulted in an overall homogenization of the landscape in these
counties (Table 4.1). From another side, in counties with 70% of pristine forests, the number of
patches that are gained increased, and fewer patches were lost. The sizes of gains and losses
increased in these counties suggesting that forest regeneration remained highly dynamic
throughout the analysed 8-year period. Although there was a trend of reducing spatial
heterogeneity through time due to the larger sizes of the patches (Table 4.1). This is particularly
relevant due to the fact that cattle ranching dominates the deforested areas of the Amazon and in
most regions it is only marginal profitable (Nepstad et al. 2009). Therefore, ranchers rely on
infrastructure and market expansions to enhance the price of their ranch, which are then sold for
more lucrative agricultural commodity production (Bowman et al. 2012). Therefore, for a large
portion of the region, cattle ranching is seen as a way of securing land tenure and clearing for
land speculation (Bowman et al. 2012). Yet at least one-fourth of the ranch land commonly
bounces back to second-growth forests (INPE/TerraClass 2016). However, second-growth
forests do not have the same conservation or legislation status when compared to pristine forest
(Metzger 2010).
To avoid deforestation in the Amazon, land intensification by increasing agricultural
productivity per hectare has been proposed (Martinelli et al. 2010; Sparovek et al. 2010). These
efforts might be misguided in many regions, however, because the primary causes of forest
94
clearing is granting land tenure towards land speculation, and not agriculture production. Yet, per
hectare productivity is known to be influenced by market pressures and related to it, and
associated with higher human population densities (Vale 2014). Also, land-use intensification
depends on infrastructure expansions, whereas infrastructure developments are known to cause
severe forest losses (Laurance et al. 2001). Additionally, technical assistance necessary for land-
use intensification is almost exclusively available for large and capitalized land owners (IBGE
2006). This, associated with increased land prices, caused a “Pervasive transition of land-use
system” in Brazil, reinforcing traditional inequality in land ownership (Lapola et al. 2014). This
inequality caused forced migration towards cities, or towards more remote areas, which starts the
process of clear-cutting forests all over again in a new region. Hence, land-use intensification in
the Amazon will only reduce deforestation if controlled under robust governance and
accompanied by a focus in solving enduring land-tenure problems, thereby granting property
rights not only for powerful sectors (Lapola et al. 2014).
Land-use intensification reduces spatio-temporal dynamics and connectivity (Chapter 3),
amplifying the threats to biodiversity conservation in fragmented landscapes. Also, pasture
dynamics were better modeled by the selected anthropogenic variables than forest dynamics
(Table 4.3). Moreover, the numbers of patches, amounts of changes and sizes of the loss and
gains of pastures and forests flawed match each other, different from what is expected if it was a
perfect “black and white” dynamics. This suggests that although landscape changes across the
Amazon are modeled as an irreversible, one-way process, where forests are turned into pastures,
landscape dynamics are more complicated than this simplified picture. Complex spatio-temporal
forest dynamics occur even in remote and frontier regions, and additional attributes than are
typically included in studies are needed to fully understand forest dynamics. For example,
although the number of patches of forest loss and pasture gain matched reasonably well in the
period 2004-2008 for both counties with 30 and 70% of pristine forests, then in the period 2008-
2012 patches of forests gain governs the dynamics in landscapes where pristine forests dominate
(Table 4.1). This happened even with the increased agricultural commodities prices that began
after 2008 (OECD/FAO 2011). Although agricultural prices raised, the fluctuation of these prices
also increased (OECD/FAO 2011). These fluctuations in price that were associated with the
global economic crisis of 2008 might had brought additional uncertainty to farming operations in
these newly settled counties. Nevertheless, this period corresponded to the start of the
95
disassociation between production and deforestation observed in the region, given harvest
increases (Macedo et al. 2012). It is important to note however, that this relationship between
production and deforestation could have been different across the Amazon, for instance, in
counties with different proportions of forests and with different infrastructure facilities.
Therefore, different characteristics and history of occupation drive the landscape spatial patterns,
ultimately influencing ecological processes across the Amazon in different ways (Perz & Skole
2003).
The Amazon has experienced an extreme reduction in deforestation in the recent years
due to: the precise application of traditional strategies; named market regulation (Rudorff et al.
2011); creation of vast and strategically placed protected areas (Walker et al. 2009); credit
obstacles’ to properties which did not comply with the environmental legislation; and command
and control policies regarding illegal deforestation (Nepstad et al. 2009). This has happened even
in light of a growing demand for Brazilian beef (Bowman et al. 2012) and increases in prices of
agricultural commodities (OECD/FAO 2011). Brazil has been able to increase agricultural
production without expanding into pristine forests, decoupling agriculture production from
deforestation (Macedo et al. 2012). Nevertheless, increases in productivity were associated with
a reduction in landscape heterogeneity across space and time, and increases in the sizes and
aggregation of pastures. These changes were constantly promoted by scattered and reduced size
of patches of forest regeneration and aggregated and enlarging patches of pasture growth. These
changes highlight the economic path that Brazil’s agriculture is taking: gradually moving
towards an intensified and exported oriented large-scale commodity farming. Additionally, these
traditional conservation strategies might had reach their limits, as seen by the stabilization of the
pristine-forest conversion amounts of each year since 2012 (PRODES/INPE 2016).
Under the continuous reductions of pristine forest, the importance of secondary-growth
forests increases especially for biodiversity conservation (Chazdon et al. 2009), carbon
sequestration (Poorter et al. 2016), climate regulation (Pütz et al. 2014), ecosystem functioning,
and ecosystem services in general (Edwards et al. 2014). The Amazon is therefore a vast
storehouse of biodiversity, estimated to hold more than 10% of the world’s species (Jenkins et al.
2010, 2015), and its importance in regulating climate goes well beyond its borders, affecting
rainfall patterns throughout much of South America (Nobre, Sellers & Shukla 1991). It is also a
large carbon pool, playing a major role on the global carbon stocks (Pütz et al. 2014). As pristine
96
forest losses continue to happen, the spatial dynamics of fragmented landscapes should gain
additional attention in order to reconcile biodiversity conservation and agricultural production in
the vast and growing portion of the Amazon that is fragmented. Important institutional and
political complexities, as well as strong economic interests are driving the spatial dynamics in
the Amazon. As human population growth and infrastructure expands over the region, it is
expected that additional reductions in spatial heterogeneity and forest dynamics will occur, and
therefore biodiversity conservation in fragmented landscapes will struggle. In this context, the
network of protected areas and indigenous lands, which cover around 45% of the biome or 54%
of the remaining forest (Soares-Filho et al. 2010), will be even more crucial for biodiversity
conservation. As the dynamics of the region vary across spatial and temporal scales, so too
should the strategies to reconcile biodiversity conservation and agricultural production. Brazil
has a pioneering experience in setting aside tracks of land for sustainable use management,
including by local communities, which has great potential as a complementary tool to the strictly
protected conservation units (Peres 2011). This alternative could play a relevant spatial role in
maintaining landscape heterogeneity in space and time in key places to sustain landscape scale
connectivity and should regain attention in upcoming years. In these situations REDD+ strategies
could be adopted and successful cases have appeared (Sills et al. 2014). Moreover, Brazil holds
the title of the most unequal land distribution in the planet (Oliveira 2001), where less than 1% of
the properties cover around 50% of the agricultural land of the country (IBGE 2006). Therefore,
the claims for land reform are fair, and recently small-land owners and land settlers have in many
times allied forces towards sustainable agricultural development, frequently in opposition to
large non-sustainable powerful agricultural sectors (Braga & Martensen in press). Land-reform
settlements could be important areas to develop environmental friendly agricultural productive
strategies, such as the ones based on agroecology, which have multiplied across the Amazon
(Fearnside 1992; Smith, Dubois & Current 1998). In summary, the future of the region depends
on devising strategies to protect the environment while also allowing sustainable use of its vast
resources. The Amazon needs to generate a societal return, which should not be privatized by a
few large landowners, in order to generate incentive for people to maintain such a large forested
environment.
97
Appendix
Nu
mb
er o
f p
atch
es
98
Mea
n p
atch
are
a (h
a)
Max
imu
m p
atch
are
a (h
a)
99
% o
f th
e co
un
ty o
ccu
pie
d b
y ea
ch c
lass
Figure 4.2: Barplot of the number of patches, mean patch area (ha), maximum patch area and %
of the counties occupied by each class. Dark grey: 30% of forest in the landscape; and light grey:
70% of forest in the landscape.
100
- Conclusions
Thesis summary
In recent years, global environmental change and its detrimental effects have steadily increased
in headlines across global media outlets. For instance, the current staggering loss of forests
worldwide has driven awareness in climate and land-use change (Fleming et al. 2011). In
particular, tropical forests are extremely vulnerable to climate and land-use change, given the
elevated number of species and endemism rates found in these bioregions (Laurance, Sayer &
Cassman 2014). The recent and intensive degradation of tropical forests for anthropogenic use
(FAO 2010; Goldewijk et al. 2011) has led to the loss of over 50% of the tropical forests, with
much of the remaining forest habitat severely fragmented (Haddad et al. 2015). The devastating
loss to tropical forest habitat, its fragmentation and consequent degradation renewed the threats
of a mass global extinction (Rands et al. 2010; Pereira et al. 2010), where species over-
exploitation, native habitat loss and fragmentation continue to be the main drivers of species
extinctions (Maxwell et al. 2016).
The understanding of the effects of habitat loss and fragmentation tends to ignore the
unique characteristics of tropical forests. For instance, many studies assume that habitat loss is
constant or immediate, where, following habitat loss, the enduring habitat is fragmented and
spatial patterns are directly related to the amount of the remaining habitat (see reviews in
Debinski & Holt 2000; Fahrig 2003; Haddad et al., 2015; Wilson et al., 2016). However, forest
regeneration is particularly fast in the tropics, where unless prevented by continued disturbance,
large portions of the deforested area quickly revert to second-growth forests (Corlett 1995).
Moreover, tropical landscapes are complex assortments of native habitats in different
successional stages, with different spatial characteristics (e.g., shape, size). These habitats are
immersed in a dynamic and complex mosaic of matrix types (Melo et al. 2013), where historical
processes also imprint upon the habitats’ present spatial features (Ewers et al. 2013). Therefore,
traditional forest-loss and fragmentation models, which assume a one-way route of habitat
destruction, are overly simplistic and fail to capture the complex and dynamic nature of tropical
fragmented forest landscapes, even if only one land-use occurs in the area.
In my thesis, I developed, in collaboration with my co-authors, a new method to
incorporate landscape dynamics into connectivity metrics. My approach significantly departs
101
from traditional methods, which use purely spatial metrics, in application to dynamic fast-growth
tropical landscapes. I demonstrated the importance of accounting for the complex spatio-
temporal interactions of tropical forest habitats, as it is vital to understand biological patterns and
processes in tropical fragmented landscapes. Additionally, I showed the dire potential impacts on
biodiversity conservation from land-use intensification, where market pressures across the
tropics have changed the spatial dynamics of fragmented landscape.
Recent failed attempts to incorporate spatial dynamics into the landscape ecology
understanding of fragmented landscapes, has resulted in inadequate predictions of land-use
changes impacts on biodiversity (Ewers et al. 2013). For instance, one approach compares
multiple static temporal snap-shots of distinct years of landscapes (e.g., Metzger et al., 2009;
Lira et al., 2012). While this approach has found empirical evidence for time-lagged effects of
habitat changes on species distribution (Tilman et al. 1994; Hylander & Ehrlén 2013), it however
fails to incorporate the temporal interactions among years. Another approach accounts for the
temporal dependencies, such as the ‘terrageny’, where phylogenies represent the historical
spatial separation of habitat fragments (Ewers et al. 2013). This approach, however, models
landscapes as a simplistic one-way route and does not account for multiple connections and
separations of habitat fragments. In this context, network (graph) theory is one of the most
promising and integrative approaches for evaluating landscape connectivity (Urban et al. 2009;
Dale & Fortin 2010; Blonder et al. 2012)—it is more informative than simple landscape metrics,
yet less demanding in terms of biological data than individual-based or metapopulation models
(Bodin & Norberg 2007; Fall et al. 2007). However, the employment of network theory to
evaluate spatial dynamics are still poorly developed (Blonder et al. 2012), where there is a clear
knowledge gap of the influences of spatio-temporal dynamics on habitat connectivity in
fragmented landscapes.
In Chapter 2, I proposed a novel spatio-temporal network approach and corresponding
metrics for quantifying the amount of habitat that can be reached through both spatial and
temporal connections. The new metrics are directly comparable to purely static metrics that have
been widely used in previous studies (e.g., Saura & Rubio 2010; Saura et al. 2014 and citations
therein). I demonstrated that spatio-temporal connectivity in dynamic fragmented landscapes of
the Atlantic Forest in the northeast of Brazil is on average 30% higher than purely spatial static
metrics, and is sometimes 150% higher. This influence of spatio-temporal connectivity arises due
102
to connections through temporal stepping-stone patches that appear (habitat gain) and disappear
(habitat loss) over time. Landscape connectivity affects organisms’ dispersal, directly influencing
populations, communities and ecosystems dynamics (Mitchell, Bennett & Gonzalez 2013).
Landscape connectivity plays therefore an important role countering relaxation time and
extinctions debt (Malanson, 2002; James et al., 2007, Jackson & Sax, 2009). Hence, the hyper-
dynamism of tropical landscapes can sustain high levels of spatio-temporal connectivity, and as
such, could significantly prevent species extinctions. This could explain, in part, the lack of
massive amount of extinctions in the tropics due to habitat loss and fragmentation.
My study region for my second and third chapters encompasses the first location where
Europeans arrived and settled in Brazil. This regions long history of degradation dates back from
the 1500s when the selective logging of the pau-brasil (Caesalpinia echinata) started (Dean
1996). Given its long degradation history in comparison to other tropical forests around the
world, the Atlantic Forest is an ideal case study for time-lagged effects of habitat loss and
fragmentation in the tropics (Brooks & Balmford 1996; Brooks et al. 1999; Metzger et al. 2009;
Lira et al., 2012; Uezu & Metzger 2016). The Atlantic Forest originally covered 150 million
hectares of the coastal area of Brazil, but also portions of inland Argentina and Paraguay.
However today, it covers only 12% of its original expanse and is fragmented into more than
247,000 patches, where 83% of the fragments are now less than 50 ha (Ribeiro et al. 2009).
Despite the extensive damage to the native habitat, only one documented Atlantic Forest species
is considered to be extinct in the wild: the Alagoas Curassow (Mitu mitu) has not seen since the
early 1980s. This species was endemic to a small area of less than 2,500 km2 in the north-east of
the Atlantic Forest, that is today extremely reduced and almost exclusively composed of second
growth forests (Silveira, Olmos & Long 2004). This singular documented extinction as a result
of habitat loss and fragmentation is however biased and does not fully reflect species loss in the
area—curassows were an important game-species, with extensive records of the intense hunting
pressure for this species, even in the last decades of its existence in nature (Silveira, Olmos &
Long 2004).
Another important result from the second chapter is that species with short dispersal
distances (< 1000 m) are benefit from temporal connections, and the spatio-temporal dynamics
are predominantly influential when the reductions in habitat amount were greater between time-
steps. This is crucial given the fast forest conversion in many tropical regions (Hansen et al.
103
2013). Purely spatial connectivity evaluations on these areas might be seriously misleading by
overstating the estimation of populations’ isolation. Predicting when species extinctions will
occur is difficult yet it is vital for proper management towards conservation (Wearn et al., 2012).
I believe that some of the recent projections are overstated when the spatio-temporal connectivity
paths of habitats are not considered. Considering spatio-temporal path connectivity would offer a
broader window of opportunity for restoration and conservation actions to be implemented.
Additionally, spatio-temporal influences vary as a function of habitat amount, and is
greater at intermediate habitat amounts (~30%). This finding is similar of what was previously
observed for purely spatial connections (Andrén 1994). This is again highly relevant for
management purpose, since sustaining ~30% of habitat in landscapes is suggested to be
beneficial for sustaining high levels of connectivity in spatial context (Andrén 1994; Martensen,
Pimentel & Metzger 2008; Martensen et al. 2012), and this could be a pattern reinforced in a
temporal perspective.
Environmental legislation in Brazil defines two different conservation requirements in
rural properties: the permanent protected areas, which are mostly along the rivers and in
extremely steep areas; and the legal reserves, which are percentages of the properties that need to
be set aside for forest management and conservation purpose (Boscolo & Martensen 2011).
These percentages vary as a function of the region of the country, and used to be 20% in the
Atlantic Forest (Silva, Nobre & Manzatto 2011). This environmental legislation has, however,
been recently relaxed, and more changes are to come (Tollefson 2016; Editorial Nature 10
November 2016). Among the relaxations of the environmental legislations are reducing
permanently protected areas and the percentages of the legal reserves (Metzger 2010).
Furthermore, the legislation may account for exotic species in plantations, such as Eucalyptus
and Pinnus, as legal reserves, and expand the possibilities for consider the legal reserve outside
the farm, even in other regions of the country (Metzger 2010). My results from Chapters 2 and 3
suggest that these aforementioned changes erroneously tried to complement biodiversity
conservation and agriculture production, by considering plantations as reserves, and reducing the
percentages for conservation.
Marketing pressures influenced by the rising demands for food, biofuels, and fibers have
moreover intensified land-use in tropical regions (Foley et al. 2005). Land-use intensification
104
reallocates land to increase field sizes, and eliminates small ephemeral forested patches
immersed in agricultural fields (Tscharntke et al. 2005). Such land intensification is a huge
financial investment (Angelsen 2010) with high external input, such as fertilizers, pesticides and
irrigation (Green et al. 2005), such that land abandoned back to native habitat regeneration is
low. Native forested habitats are constrained into sites with low aptitude for crops (Latawiec et
al. 2015) and a reduced number of land uses dominate the landscape. Land-use intensification is
also known to promote higher matrix harshness, which leads in lower permeability of movement
among native habitats (Perfecto & Vandermeer 2008); moreover, it reduces the likelihood of
agricultural fields to be used as complementary habitats to the native ones (Dunning, Danielson
& Pulliam 1992). This large-scale intensification of land may however decelerate habitat loss, as
it already facilitates the tracking of habitat changes by remotely sensed data. Moreover,
associated infrastructure improvements in roads and cities provide access to areas and promotes
control and law enforcement (Arima et al. 2014). Therefore, land-use intensification may
decelerate habitat loss, where habitats can mature into overall higher quality and structured
habitats (Arima et al. 2014).
A large portion of the study region of Chapters 2 and 3 is of Veracel property. Veracel
formerly belonged to both the Swedish-Finnish company Stora Enso and to a Norwegian-
Brazilian company Aracruz Celulose, which was recently incorporated by Fibria, a full Brazilian
company. In 2008, Veracel—a company that produces 1.1 million tons of cellulose pulp and is
FSC certified1--was found guilty of illegal deforestation of large areas of pristine and mature
second-growth Atlantic Forests. Veracel was fined millions of dollars and responsible for
restoring 47,000 ha of forests in the region2. However Veracel was not solely responsible for the
deforestation in the region (Ribeiro et al. 2012). Actually, in low-intensified areas the
deforestation was even more pronounced, however with similar amounts of low-intensified fields
bouncing back to regeneration forests, a pattern that was not accompanied in the Eucalyptus
plantations (Ribeiro et al. 2012). My results from Chapter 3 pointed exactly to the differences of
1 http://www.veracel.com.br/en/about-veracel/
2 http://www1.folha.uol.com.br/poder/2008/07/421375-justica-federal-condena-veracel-celulose-a-pagar-multa-de-r-
20-milhoes.shtml?mobile
105
these two dynamics observed in low- and high-intensified fields, particularly in terms of changes
in spatio-temporal connectivity. Therefore, in highly intensified fields, the low agricultural
aptitude lands are abandoned and potentially restored (Latawiec et al. 2015), which is an
opportunity to increase connectivity among stable fragments. Yet if there are not enough of these
low aptitude sites, or if these sites are not spatially placed in a way that it could overcome the
reductions in connectivity given the reductions in spatio-temporal dynamics, then highly
intensified landscapes will present an overall reduction in connectivity, compared to low-
intensified ones with similar amounts of native habitats. In these cases, additional land should be
spared for biodiversity conservation. This should be implemented as a key management strategy
in landscapes experiencing land-use intensification, and my proposed metrics from Chapter 2
should be used to define the amounts and spatial locations of land that should be set aside to
overcome reductions in spatio-temporal connectivity.
Finally, land-use intensification is generally associated with intense land grabbing and
changes in agricultural crops employed in the area, i.e., adoption of agricultural commodities
such as soy, corn, palm oil or Eucalyptus. In most of the Amazon, the intensification occurs
within the same land-use, where pastures have been slowly intensified. This could be considered
as a first step into land-use intensification, as cattle ranching is in a large portion of the Amazon
but only marginal profitable (Bowman et al. 2012). Additionally, in places where infrastructure
is already in place such as the county of Sinop (study area in Chapter 4) soy, corn, cotton and
other agricultural commodities are already the main land-use (TerraClass 2016). This is also the
case along the main highways that dissect the Amazon region (Perz et al. 2008), such as the
BR010-Belém-Brasília highway, BR-163-Cuiabá-Santarém highway (where Sinop is located),
particularly in the portion inside the state of Mato Grosso, which also has the BR0158 highway,
as an important vector of land-use intensification. The former governor of the state of Mato
Grosso, current a senator, personally made an extreme political and personal effort to have
appointed one of his people to head the DNIT, the federal department that deals with transport
infrastructure3. The large infrastructure plans that were conducted across Mato Grosso, confers
extreme competitive advantage to the agriculture of the state of Mato Grosso, helping the state to
3 http://blogs.oglobo.globo.com/blog-do-moreno/post/governador-vem-pessoalmente-articular-indicacao-para-dnit-
68670.html
106
become a major global agricultural producer (Macedo et al. 2012). This change in land-use had
direct feedbacks into the deforestation dynamics because agricultural fields are larger than
pastures (Morton et al. 2006). My analyses also find these trends in the intensification process
across the evaluated counties, which are greatest in the state of Mato Grosso.
In my thesis, I explore ways in which our conceptual understanding of landscape ecology
and its applications towards biodiversity conservation benefit from considering spatio-temporal
dynamics. The methods and software I present in this thesis could facilitate our understanding of
the ecological processes in dynamic fragmented landscapes, and reconcile agriculture production
and biodiversity conservation in the tropics. My finding—that low-intensified landscapes are
more dynamic than highly intensified ones, potentially affecting populations by being less
isolated in these sorts of landscapes—could shed some light on “Why we haven’t seen many
extinctions given habitat loss and fragmentation in the tropics…”. This is currently solely
attributed to time-lagged effects, and I argue that the spatio-temporal connectivity could play a
larger, but until now ignored role on this manner. I highlight that in landscapes that are
experiencing land-use intensification, additional land should be set aside for biodiversity
conservation, and active policies to implement these practices should be prioritized. I also stress
that land-use intensification is occurring across the tropics, and sometimes, within the same land-
use type. Therefore, one should not expect the full range of changes, such as increased the
amounts of fertilizers and pesticides, to observe the impacts on spatio-temporal dynamics. In
summary, I present strong evidence that the ecological patterns and processes in tropical
fragmented landscapes cannot be fully understood under a spatially static landscape
framework—not even by comparing multiple spatial static snapshots. Tropical fragmented
landscapes should be considered as complex interdependent spatio-temporal dynamic systems,
with intricate spatial and temporal relationships among its landscape features, which directly
affect different ecological processes. The conceptual model and the spatio-temporal metrics
developed during this thesis, along with the experiments carried out in two different and
important tropical regions, provide a foundation to evaluate spatio-temporal connectivity, as well
as an initial understanding of the impacts that the changes on these dynamics could promote in
tropical fragmented landscapes.
107
Future research directions
While many important questions arose throughout the course of my thesis, the scope of my thesis
is however constrained to address the drivers of spatio-temporal dynamics in fragmented
landscapes, as well as the potential influences of the spatio-temporal paths (see Figure 2.1) on
these dynamics. However, it is imperative to incorporate all aspects of spatio-temporal dynamics,
namely, the time-lagged decay of individuals/species numbers within fragments in the algorithms
that calculates the network and metrics. While this aspect is present in the conceptual model of
Chapter 2, it however was not included in the experiments because the time-lagged effects were
already explored elsewhere (see for example Wearn et al., 2012; Gilbert & Levine, 2013;
Hylander & Ehrlén, 2013), and the inclusion of this other aspect could obfuscate the temporal-
path analysis we presented. Nevertheless, based on the current understanding of time-lagged
effects, I expect that the temporal-paths, as I proposed here, will mediate the relaxation time by
slowing extinction debts, as well as enabling potential immigration credits. Therefore, a holistic
understanding of the spatio-temporal relationships will demand the inclusion of the time-lagged
effects in the algorithm which calculates the spatio-temporal dynamics. Nevertheless, it does not
invalidate our conclusions, contrary, exacerbate the effects observed in my thesis.
Also, the relevance that I found for the spatio-temporal dynamics on the isolation of
species with various dispersal abilities calls for a more robust empirical evidence of these
influences, including in relation to habitat use, and dispersal behaviour. For example, to evaluate
into what extent the species could use the forests in the initial stages of succession, for instance if
they can use these areas as habitat, or exclusively as a permeable matrix enhancing connectivity,
or if they cannot use at all. Professor Eduardo Mariano, from the Federal University of Bahia,
coordinated a project that sampled 16 sites within the study region of Chapters 2 and 3, for the
abundance of different taxonomic groups, including trees and birds. Among the 16 sites, a
diverse array of forests successional stages was sampled, including four that regenerated within
our study period (i.e., after 1990): three as fragment expansions, and another one as a complete
new fragment isolated from others. This opportunistic dataset could provide an important initial
evaluation of habitat use for different species. In initial inspections of the tree composition
dataset, we found forests that regenerated within our study period to have different species
composition compared to older ones. These newly regenerated fragments are dominated by
108
pioneer/wind-dispersed species, independently if they were physically linked to a fragment, or
regenerated isolated from any existing patch of forest.
Tropical animal species inter-habitat dispersal capacities still largely unknown
(Crouzeilles, Lorini & Grelle 2010). The best available information are for birds, which
suggested that understory bird species individuals avoid any open gaps, sometimes not even
using edge habitats (Hansbauer et al. 2008b). However, the same species are able to, in very
unfrequently occasions, make some longer dispersions and reach other fragments (Hansbauer et
al. 2008a). The available data on species dispersal are still restricted to a dozen or so species. To
proper parameterize the spatio-temporal models as here proposed, a better understanding of the
dispersal capacities, as well as of life span are key.
There are also some refinements to the model, which could be promptly applied.
However, again the lack of biological information makes it challenging to do so. For example,
least-cost path connectivity could bring additional reality to the connectivity metrics. Information
about matrix permeability is largely absent, and very few experiments evaluate species dispersal
behaviors among different matrix in the tropics (e.g. Gascon et al. 1999; Antongiovanni &
Metzger 2005; Umetsu & Pardini 2007). Matrix composed of tree plantations have similar
structure compared to native forests, and therefore are expected to provide shelter against
predators, shade and microclimatic protections, when compared to open fields, and therefore, are
expected to be more permeable (Taylor et al. 1993). This is a vision that was supported by
studies conducted either long ago, when Eucalyptus plantations were of longer rotations and with
an intense understory growth (e.g. Stallings 1991), or by forest plantations managed in an
ecological-sustainable fashion (Fonseca et al. 2009). Studies conducted in modern intensified
agriculture have shown that only very few generalist, open-habitat species could use the
plantations (Umetsu & Pardini 2007). Additionally, in most of the cases, these species are able to
use the Eucalytpus plantations in the initial of the rotation cycle (Rosalino et al. 2014), when the
plantations are similar to open fields. In a recent collaboration, I compare Eucalyptus, with
highly intensified sugar-cane plantations, a matrix that is expected to be highly impermeable
(Giubbina et al. in review). Nevertheless, both matrices presented similar high levels of
resistance, even for low sensitive edge understory bird species, including one species that is
frequently found in open habitats close to forest patches, as well as in initial stages of forest
succession (Giubbina et al. in review). Therefore, the use of similar matrix permeability values in
109
my thesis, i.e., Euclidian distance among forest patches, could be considered a very conservative
interpretation of the connectivity patterns in the light of land-use intensification. Therefore,
including matrix harshness would almost certainly reinforce my results about the reduction in
habitat connectivity in intensified land uses landscapes.
In summary, there are many different avenues in which one can apply the conceptual
framework and models that I developed during my thesis. I believe that my findings could
improve our ability to understand ecological process and patterns in dynamic landscapes; but
moreover, could help us better manage fragmented landscapes and improve landscape
connectivity and therefore, long-term population viability. My thesis provided therefore new
insights and important findings to the understanding of the dynamic network structure of
fragmented landscapes, which could also have profound impact on biodiversity conservation.
110
Bibliography
Aide, T.M. & Grau, H.R. (2004) Globalization, migration, and Latin American ecosystems.
Science (New York, N.Y.), 305, 1915–1916.
Aide, T.M., Zimmerman, J.K., Pascarella, J.B., Rivera, L. & Marcano-Vega, H. (2000) Forest
regeneration in a chronosequence of tropical abandoned pastures: Implications for
restoration ecology. Restoration Ecology, 8, 328–338.
Andrén, H. (1994) Effects of habitat fragmentation on birds and mammals in landscapes with
different proportions of suitable habitat - a review. Oikos, 71, 355–366.
Angelsen, A. (2010) Policies for reduced deforestation and their impact on agricultural
production. Proceedings of the National Academy of Sciences of the United States of
America, 107, 19639–44.
Antongiovanni, M. & Metzger, J. (2005) Influence of matrix habitats on the occurrence of
insectivorous bird species in Amazonian forest fragments. Biological Conservation, 122,
441–451.
Arima, E.Y., Barreto, P., Araújo, E. & Soares-Filho, B. (2014) Public policies can reduce
tropical deforestation: Lessons and challenges from Brazil. Land Use Policy, 41, 465–473.
Assessment, M.E. (2005) Millennium Ecosystem Assessment Synthesis Report. Millennium
Ecosystem Assessment.
Auffret, A.G., Plue, J. & Cousins, S.A.O. (2015) The spatial and temporal components of
functional connectivity in fragmented landscapes. AMBIO, 44, 51–59.
Barreto, P. & Silva, D.S. da. (2012) How Can One Develop the Rural Economy without
Deforesting the Amazon?, 1st ed (ed Imazon). IMAZON - Amazon Institute of People and
Environment, Belém.
Bélisle, M. (2005) Measuring landscape connectivity: The challenge of behavioral landscape
ecology. Ecology, 86, 1988–1995.
Blonder, B., Wey, T.W., Dornhaus, A., James, R. & Sih, A. (2012) Temporal dynamics and
network analysis. Methods in Ecology and Evolution, 3, 958–972.
Bodin, Ö. & Norberg, J. (2007) A network approach for analyzing spatially structured
populations in fragmented landscape. Landscape Ecology, 22, 31–44.
Bommarco, R., Lindborg, R., Marini, L. & Öckinger, E. (2014) Extinction debt for plants and
flower-visiting insects in landscapes with contrasting land use history. Diversity and
Distributions, 20, 591–599.
Borrett, S.R., Moody, J. & Edelmann, A. (2014) The rise of Network Ecology: Maps of the topic
diversity and scientific collaboration. Ecological Modelling, 293, 111–127.
111
Boscolo, D., Candia-Gallardo, C., Awade, M. & Metzger, J.P. (2008) Importance of Interhabitat
Gaps and Stepping-Stones for Lesser Woodcreepers (Xiphorhynchus fuscus) in the Atlantic
Forest , Brazil. Biotropica, 40, 273–276.
Boscolo, D. & Martensen, A.C. (2011) Alterações no Código Florestal afetam todos os
brasileiros. Ciência Hoje, 48, 74–75.
Bowman, M.S., Soares-Filho, B.S., Merry, F.D., Nepstad, D.C., Rodrigues, H. & Almeida, O.T.
(2012) Persistence of cattle ranching in the Brazilian Amazon: A spatial analysis of the
rationale for beef production. Land Use Policy, 29, 558–568.
Braga, A.C.R. & Martensen, A.C. Smallholders: Drivers or targets of Amazonian deforestation?
Human Geography, In Press.
Brooks, T.M. & Balmford, A. (1996) Atlantic forest extinctions. Nature, 380, 115–115.
Brooks, T.M., Mittermeier, R.A., Mittermeier, C.G., da Fonseca, G.A.B., Rylands, A.B.,
Konstant, W.R., Flick, P., Pilgrim, J., Oldfield, S., Magin, G. & Hilton-Taylor, C. (2002)
Habitat loss and extinction in the hotspots of biodiversity. Conservation Biology, 16, 909–
923.
Brooks, T.M., Pimm, S.L. & Oyugi, J.O. (1999) Time lag between deforestation and bird
extinction in tropical forest fragments. Conservation Biology, 13, 1140–1150.
Brooks, T., Tobias, J. & Balmford, A. (1999) Deforestation and bird extinctions in the Atlantic
forest. Animal Conservation, 2, 211–222.
Brown, J.H. (2014) Why are there so many species in the tropics? Journal of Biogeography, 41,
8–22.
Brown, K.S.J. & Brown, G.G. (1992) Habitat alteration and species loss in Brazilian forests.
Tropical deforestation and species extinction (eds T.C. Whitmore & J.A. Sayer), pp. 119–
142. Chapman & Hall, London.
Brown, J.H. & Kodric-Brown, A. (1977) Turnover rates in insular biogeography: Effect of
immigration on extinction. Ecology, 58, 445–449.
Brown, S. & Lugo, A.E. (1990) Tropical secondary forests. Journal of Tropical Ecology, 6, 1–
32.
Burnham, K.P. & Anderson, D.R. (2002) Model Selection and Multimodel Inference: A Practical
Information-Theoretic Approach (2nd Ed).
Cardinale, B.J., Duffy, J.E., Gonzalez, A., Hooper, D.U., Perrings, C., Venail, P., Narwani, A.,
Mace, G.M., Tilman, D., Wardle, D. a, Kinzig, A.P., Daily, G.C., Loreau, M., Grace, J.B.,
Larigauderie, A., Srivastava, D.S. & Naeem, S. (2012) Biodiversity loss and its impact on
humanity. Nature, 486, 59–67.
Chazdon, R.L. (2003) Tropical forest recovery: Legacies of human impact and natural
112
disturbances. Perspectives in Plant Ecology, Evolution and Systematics, 6, 51–71.
Chazdon, R.L. (2008) Beyond deforestation: Restoring forests and ecosystem services on
degraded lands. Science (New York, N.Y.), 320, 1458–1460.
Chazdon, R.L. (2014) Second Growth: The Promise of Tropical Forest Regeneration in an Age
of Deforestation (ed C.U. Press). Chicago.
Chazdon, R.L., Peres, C.A., Dent, D., Sheil, D., Lugo, A.E., Lamb, D., Stork, N.E. & Miller,
S.E. (2009) The potential for species conservation in tropical secondary forests.
Conservation Biology, 23, 1406–1417.
Claudino, E.S., Gomes, M.A.F. & Campos, P.R.A. (2015) Extinction debt and the role of static
and dynamical fragmentation on biodiversity. Ecological Complexity, 21, 150–155.
Corlett, R.T. (1995) Tropical secondary forests. Progress in Physical Geography, 19, 159–172.
Coulon, A., Cosson, J.F., Angibault, J.M., Cargnelutti, B., Galan, M., Morellet, N., Petit, E.,
Aulagnier, S. & Hewison, A.J.M. (2004) Landscape connectivity influences gene flow in a
roe deer population inhabiting a fragmented landscape: an individual-based approach.
Molecular Ecology, 13, 2841–2850.
Crouzeilles, R., Lorini, M.L. & Grelle, C.E.V. (2010) Deslocamento na matriz para espécies da
Mata Atlântica e a dificuldade da construção de perfis ecológicos. Oecologia Australis, 14,
875–903.
Dale, M.R.T. & Fortin, M.J. (2010) From graphs to spatial graphs. Annual Review of Ecology,
Evolution, and Systematics, 41, 21–38.
Dale, V.H., Joyce, L.A., McNulty, S., Neilson, R.P., Ayres, M.P., Flannigan, M.D., Hanson, P.J.,
Irland, L.C., Lugo, A.E., Peterson, C.J., Simberloff, D., Swanson, F.J., Stocks, B.J. &
Michael W.B. (2001) Climate change and forest disturbances. BioScience, 51, 723.
Dean, W. (1996) A Ferro E Fogo: A História E a Devastação Da Mata Atlântica Brasileira.
Companhia das Letras, São Paulo.
Debinski, D.M. & Holt, R.D. (2000) A survey and overview of habitat fragmentation
experiments. Conservation Biology, 14, 342–355.
Diamond, J.M. (1972) Biogeographic kinetics: Estimation of relaxation times for avifaunas of
Southwest Pacific Islands. Proceedings of the National Academy of Sciences of the United
States of America, 69, 3199–3203.
Dunn, R.R. (2004) Recovery of faunal communities during tropical forest regeneration.
Conservation Biology, 18, 302–309.
Dunning, J.B., Danielson, B.J. & Pulliam, H.R. (1992) Ecological processes that affect
populations in complex landscapes. Oikos, 65, 169–175.
113
Dyer, R.J. & Nason, J.D. (2004) Population graphs: The graph theoretic shape of genetic
structure. Molecular Ecology, 13, 1713–1727.
Editorial Nature 10 November. (2016) Don’t bypass Brazil’s environmental protections. Nature,
539, 139–139.
Edwards, D.P., Tobias, J.A., Sheil, D., Meijaard, E. & Laurance, W.F. (2014) Maintaining
ecosystem function and services in logged tropical forests. Trends in Ecology and
Evolution, 29, 511–520.
Essl, F., Dullinger, S., Rabitsch, W., Hulme, P.E., Pyšek, P., Wilson, J.R.U. & Richardson, D.M.
(2015) Delayed biodiversity change: no time to waste. Trends in Ecology & Evolution, 30,
6–9.
Ewers, R.M., Didham, R.K., Pearse, W.D., Lefebvre, V., Rosa, I.M.D., Carreiras, J.M.B., Lucas,
R.M. & Reuman, D.C. (2013) Using landscape history to predict biodiversity patterns in
fragmented landscapes ed M. Bonsall. Ecology Letters, 16, 1221–1233.
Fahrig, L. (2003) Effects of habitat fragmentation on biodiversity. Annual Review of Ecology,
Evolution, and Systematics, 34, 487–515.
Fahrig, L., Baudry, J., Brotons, L., Burel, F.G., Crist, T.O., Fuller, R.J., Sirami, C., Siriwardena,
G.M. & Martin, J.-L. (2011) Functional landscape heterogeneity and animal biodiversity in
agricultural landscapes. Ecology Letters, 14, 101–112.
Fall, A., Fortin, M.J., Manseau, M. & O’Brien, D. (2007) Spatial graphs: Principles and
applications for habitat connectivity. Ecosystems, 10, 448–461.
FAO. (2010) Global Forest Resources Assessment 2010. Rome.
Faria, D., Laps, R.R., Baumgarten, J. & Cetra, M. (2006) Bat and bird assemblages from forests
and shade cacao plantations in two contrasting landscapes in the Atlantic Forest of Southern
Bahia, Brazil. Biodiversity and Conservation, 15, 587–612.
Faria, D., Mariano-Neto, E., Martini, A.M.Z., Ortiz, J.V., Montingelli, R., Rosso, S., Paciencia,
M.L.B. & Baumgarten, J. (2009) Forest structure in a mosaic of rainforest sites: The effect
of fragmentation and recovery after clear cut. Forest Ecology and Management, 257, 2226–
2234.
Fearnside, P. (1992) Agroforestry in Brazil’s Amazonian development policy: The role and
limits of a potential use for degraded lands. Brazilian Perspectives on Sustainable
Development of the Amazon Region (ed M. Clüsener-Godt & I. Sachs), pp. 125–148. France
and Parthenon Publishing Group, Carnforth, UK.
Fearnside, P.M., Laurance, W.F., Cochrane, M. a., Bergen, S., Delamônica, P., Barber, C.,
D’Angelo, S. & Fernandes, T. (2012) The future of Amazonia: models to predict the
consequences of future infrastructure in brazil’s multi-annual plans. Novos Cadernos NAEA,
15, 25–52.
114
Ferraz, G., Russell, G., Stouffer, P., Bierregaard, R., Pimm, S. & Lovejoy, T. (2003) Rates of
species loss from Amazonian forest fragments. Proceedings of the National Academy of
Sciences of the United States of America, 100, 14069–73.
Fleming, R., Kanowski, P., Brown, N., Jenik, J., Kahumbu, P. & Plesnik, J. (2011) Emerging
perspectives on forest biodiversity. UNEP Year Book: Emerging Issues in Our Global
Environment, 46–59.
Foley, J.A. (2011) Can we feed the world & sustain the planet? Scientific American, 305, 60–5.
Foley, J.A., Defries, R., Asner, G.P., Barford, C., Bonan, G., Carpenter, S.R., Chapin, F.S., Coe,
M.T., Daily, G.C., Gibbs, H.K., Helkowski, J.H., Holloway, T., Howard, E. a, Kucharik,
C.J., Monfreda, C., Patz, J. a, Prentice, I.C., Ramankutty, N. & Snyder, P.K. (2005) Global
consequences of land use. Science (New York, N.Y.), 309, 570–4.
Foley, J.A., Ramankutty, N., Brauman, K.A., Cassidy, E.S., Gerber, J.S., Johnston, M., Mueller,
N.D., O’Connell, C., Ray, D.K., West, P.C., Balzer, C., Bennett, E.M., Carpenter, S.R.,
Hill, J., Monfreda, C., Polasky, S., Rockström, J., Sheehan, J., Siebert, S., Tilman, D. &
Zaks, D.P.M. (2011) Solutions for a cultivated planet. Nature, 478, 337–42.
Fonseca, C.R., Ganade, G., Baldissera, R., Becker, C.G., Boelter, C.R., Brescovit, A.D.,
Campos, L.M., Fleck, T., Fonseca, V.S., Hartz, S.M., Joner, F., Käffer, M.I., Leal-Zanchet,
A.M., Marcelli, M.P., Mesquita, A.S., Mondin, C.A., Paz, C.P., Petry, M. V., Piovensan,
F.N., Putzke, J., Stranz, A., Vergara, M. & Vieira, E.M. (2009) Towards an ecologically-
sustainable forestry in the Atlantic Forest. Biological Conservation, 142, 1209–1219.
Forero-Medina, G. & Vieira, M.V. (2009) Perception of a fragmented landscape by neotropical
marsupials: effects of body mass and environmental variables. Journal of Tropical Ecology,
25, 53.
Gascon, C., Lovejoy, T.E., Bierregaard Jr., R.O., Malcolm, J.R., Stouffer, P.C., Vasconcelos,
H.L., Laurance, W.F., Zimmerman, B., Tocher, M. & Borges, S. (1999) Matrix habitat and
species richness in tropical forest remnants. Biological Conservation, 91, 223–229.
Gibbs, H.K., Ruesch, A.S., Achard, F., Clayton, M.K., Holmgren, P., Ramankutty, N. & Foley, J.
a. (2010) Tropical forests were the primary sources of new agricultural land in the 1980s
and 1990s. Proceedings of the National Academy of Sciences, 107, 16732–16737.
Gibson, L., Lee, T.M., Koh, L.P., Brook, B.W., Gardner, T. a, Barlow, J., Peres, C. a, Bradshaw,
C.J. a, Laurance, W.F., Lovejoy, T.E. & Sodhi, N.S. (2011) Primary forests are
irreplaceable for sustaining tropical biodiversity. Nature, 478, 378–381.
Gilbert, B. & Levine, J.M. (2013) Plant invasions and extinction debts. Proceedings of the
National Academy of Sciences, 110, 1744–1749.
Goldewijk, K. k., Beusen, A., Van Drecht, G. & De Vos, M. (2011) The HYDE 3.1 spatially
explicit database of human-induced global land-use change over the past 12,000 years.
Global Ecology and Biogeography, 20, 73–86.
115
Goulart, F.F., Salles, P. & Machado, R.B. (2013) How may agricultural matrix intensification
affect understory birds in an Atlantic Forest landscape? A qualitative model on stochasticity
and immigration. Ecological Informatics, 18, 93–106.
Grau, H.R., Aide, T.M., Zimmerman, J.K., Thomlinson, J.R., Helmer, E. & Zou, X. (2003) The
ecological consequences of socioeconomic and land-use changes in postagriculture Puerto
Rico. BioScience, 53, 1159.
Green, R.E., Cornell, S.J., Scharlemann, J.P.W. & Balmford, A. (2005) Farming and the fate of
wild nature. Science (New York, N.Y.), 307, 550–5.
Haddad, N.M., Brudvig, L.A., Clobert, J., Davies, K.F., Gonzalez, A., Holt, R.D., Lovejoy, T.E.,
Sexton, J.O., Austin, M.P., Collins, C.D., Cook, W.M., Damschen, E.I., Ewers, R.M.,
Foster, B.L., Jenkins, C.N., King, A.J., Laurance, W.F., Levey, D.J., Margules, C.R.,
Melbourne, B.A., Nicholls, A.O., Orrock, J.L., Song, D.-X. & Townshend, J.R. (2015)
Habitat fragmentation and its lasting impact on Earth’s ecosystems. Science Advances, 1,
e1500052–e1500052.
Hansbauer, M.M., Storch, I., Leu, S., Nieto-Holguin, J.-P., Pimentel, R.G., Knauer, F. &
Metzger, J.P.W. (2008a) Movements of neotropical understory passerines affected by
anthropogenic forest edges in the Brazilian Atlantic rainforest. Biological Conservation,
141, 782–791.
Hansbauer, M.M., Storch, I., Pimentel, R.G. & Metzger, J.P. (2008b) Comparative range use by
three Atlantic Forest understorey bird species in relation to forest fragmentation. Journal of
Tropical Ecology, 24, 291–299.
Hansen, M.C., Potapov, P. V, Moore, R., Hancher, M., Turubanova, S.A., Tyukavina, A., Thau,
D., Stehman, S. V, Goetz, S.J., Loveland, T.R., Kommareddy, A., Egorov, A., Chini, L.,
Justice, C.O. & Townshend, J.R.G. (2013) High-resolution global maps of 21st-century
forest cover change. Science, 342, 850–853.
Hanski, I. (1999) Metapopulation Ecology. Oxford University Press, Oxford.
Hanski, I. (2011) Habitat loss, the dynamics of biodiversity, and a perspective on conservation.
Ambio, 40, 248–255.
Hanski, I. & Ovaskainen, O. (2000) The metapopulation capacity of a fragmented landscapes.
Nature, 404, 755–758.
Hanski, I. & Ovaskainen, O. (2002) Extinction debt at extinction threshold. Conservation
Biology, 16, 666–673.
Hanski, I., Zurita, G. a, Bellocq, M.I. & Rybicki, J. (2013) Species-fragmented area relationship.
Proceedings of the National Academy of Sciences of the United States of America, 110,
12715–20.
He, F. & Hubbell, S.P. (2011) Species-area relationships always overestimate extinction rates
from habitat loss. Supplementary information. Nature, 473, 368–371.
116
He, F. & Hubbell, S. (2013) Estimating extinction from species-area relationships: Why the
numbers do not add up. Ecology, 94, 1905–1912.
Heywood, V.H., Mace, G.M., May, R.M. & Stuart, S.N. (1994) Uncertainties in extinction rates.
Nature, 368, 105–105.
Heywood, V.H. & Stuart, S.N. (1992) Species extinctions in tropical forests. Tropical
deforestation and species extinction (eds T.C. Whitmore & J.A. Sayer), pp. 91–117.
Chapman & Hall, London.
Homma, A.K.O. & Furlan Jr., J. (2001) Desenvolvimento da deindeicultura na Amazônia:
cronologia. Agronegócio do dendê: uma alternativa social, econômica e ambiental para o
desenvolvimento sustentável da Amazônia. (eds A.A. Müller & J. Furlan, Jr.), pp. 193–207.
Embrapa Amazônia Ocidental, Belem.
Hylander, K. & Ehrlén, J. (2013) The mechanisms causing extinction debts. Trends in Ecology &
Evolution, 28, 341–346.
IBGE. (2006) Censo Agropecuário 2006. Rio de Janeiro.
IPCC. (2001) Climate Change 2001: Mitigation.
ITTO. (2002) ITTO Guidelines for the Restoration , Management and Rehabilitation of
Degraded and Secondary Tropical Forests.
Jackson, S.T. & Sax, D.F. (2009) Balancing biodiversity in a changing environment: extinction
debt, immigration credit and species turnover. Trends in Ecology & Evolution, 25, 153–160.
Jackson, S.T. & Sax, D.F. (2010) Balancing biodiversity in a changing environment: extinction
debt, immigration credit and species turnover. Trends in Ecology & Evolution, 25, 153–160.
Jakovac, C.C., Peña-Claros, M., Kuyper, T.W. & Bongers, F. (2015) Loss of secondary-forest
resilience by land-use intensification in the Amazon. Journal of Ecology, 103, 67–77.
James, P.M.A., Fortin, M.J., Fall, A., Kneeshaw, D. & Messier, C. (2007) The effects of spatial
legacies following shifting management practices and fire on boreal forest age structure.
Ecosystems, 10, 1261–1277.
Jenkins, C.N., Alves, M.A.S. & Pimm, S.L. (2010) Avian conservation priorities in a top-ranked
biodiversity hotspot. Biological Conservation, 143, 992–998.
Jenkins, C.N., Alves, M.A.S., Uezu, A. & Vale, M.M. (2015) Patterns of Vertebrate Diversity
and Protection in Brazil. PloS one, 10, e0145064.
Karp, D.S., Rominger, A.J., Zook, J., Ranganathan, J., Ehrlich, P.R. & Daily, G.C. (2012)
Intensive agriculture erodes β-diversity at large scales. Ecology Letters, 15, 963–970.
Kim, D., Sexton, J.O. & Townshend, J.R. (2015) Accelerated deforestation in the humid tropics
from the 1990s to the 2000s. Geophysical Research Letters, 42, 3495–3501.
117
Kindlmann, P. & Burel, F. (2008) Connectivity measures: a review. Landscape Ecology, 23,
879–890.
Korfanta, N.M., Newmark, W.D. & Kauffman, M.J. (2012) Long-term demographic
consequences of habitat fragmentation to a tropical understory bird community. Ecology,
93, 2548–2559.
Kotz, D.M. (2009) The financial and economic crisis of 2008: A systemic crisis of neoliberal
capitalism. Review of Radical Political Economics, 41, 305–317.
Krauss, J., Bommarco, R., Guardiola, M., Heikkinen, R.K., Helm, A., Kuussaari, M., Lindborg,
R., Ockinger, E., Pärtel, M., Pino, J., Pöyry, J., Raatikainen, K.M., Sang, A., Stefanescu, C.,
Teder, T., Zobel, M. & Steffan-Dewenter, I. (2010) Habitat fragmentation causes immediate
and time-delayed biodiversity loss at different trophic levels. Ecology Letters, 13, 597–605.
Kuussaari, M., Bommarco, R., Heikkinen, R.K., Helm, A., Krauss, J., Lindborg, R., Öckinger,
E., Pärtel, M., Pino, J., Rodà, F., Stefanescu, C., Teder, T., Zobel, M. & Steffan-Dewenter,
I. (2009) Extinction debt: A challenge for biodiversity conservation. Trends in Ecology &
Evolution, 24, 564–571.
Lambin, E.F. & Geist, H.J. (2008) Land-Use and Land-Cover Change: Local Processes and
Global Impacts.
Lambin, E.F., Geist, H.J. & Lepers, E. (2003) Dynamics of land-use and land-cover change in
tropical regions. Annual Review of Environment and Resources, 28, 205–241.
Lapola, D.M., Martinelli, L.A., Peres, C.A., Ometto, J.P.H.B., Ferreira, M.E., Nobre, C.A.,
Aguiar, A.P.D., Bustamante, M.M.C., Cardoso, M.F., Costa, M.H., Joly, C.A., Leite, C.C.,
Moutinho, P., Sampaio, G., Strassburg, B.B.N. & Vieira, I.C.G. (2014) Pervasive transition
of the Brazilian land-use system. Nature Climate Change, 4, 27–35.
Laps, R.R. (2006) Efeito Da Fragmentação E Alteração Do Hábitat Na Avifauna Da Região Da
Reserva Biológica de Una, Bahia. Universidade Estadual de Campinas - Unicamp.
Latawiec, A.E., Strassburg, B.B.N., Brancalion, P.H.S., Rodrigues, R.R. & Gardner, T. (2015)
Creating space for large-scale restoration in tropical agricultural landscapes. Frontiers in
Ecology and the Environment, 13, 211–218.
Laurance, W.F. (2002) Hyperdynamism in fragmented habitats. Journal of Vegetation Science,
13, 595.
Laurance, W.F., Albernaz, A.K.M., Schroth, G., Fearnside, P.M., Bergen, S., Venticinque, E.M.
& Da Costa, C. (2002) Predictors of deforestation in the Brazilian Amazon. Journal of
Biogeography, 29, 737–748.
Laurance, W.F. & Bierregaard Jr., R.O. (1997) Tropical forest remnants: Ecology, management,
and conservation of fragmented communities. Tropical forest remnants: Ecology,
management, and conservation of fragmented communities, xv+616p.
118
Laurance, W.F., Cochrane, M.A., Bergen, S., Fearnside, P.M., Delamonica, P., Barber, C.,
D’Angelo, S. & Fernandes, T. (2001) The Future of the Brazilian Amazon. Science, 291,
438–439.
Laurance, W.F., Sayer, J. & Cassman, K.G. (2014) Agricultural expansion and its impacts on
tropical nature. Trends in Ecology and Evolution, 29, 107–116.
Lees, A.C. & Peres, C.A. (2008) Conservation value of remnant riparian forest corridors of
varying quality for amazonian birds and mammals. Conservation Biology, 22, 439–49.
Liebsch, D., Marques, M.C.M. & Goldenberg, R. (2008) How long does the Atlantic Rain Forest
take to recover after a disturbance? Changes in species composition and ecological features
during secondary succession. Biological Conservation, 141, 1717–1725.
Lindborg, R. & Eriksson, O. (2004) Historical landscape connectivity affects present plant
species diversity. Ecology, 85, 1840–1845.
Lira, P.K., Ewers, R.M., Banks-Leite, C., Pardini, R. & Metzger, J.P. (2012) Evaluating the
legacy of landscape history: Extinction debt and species credit in bird and small mammal
assemblages in the Brazilian Atlantic Forest. Journal of Applied Ecology, 49, 1325–1333.
Lomolino, M. V., Riddle, B.R., Whittaker, R.J. & Brown, J.H. (2010) Biogeography, 4th ed.
Sinauer Associates, Inc.
Macedo, M.N., DeFries, R.S., Morton, D.C., Stickler, C.M., Galford, G.L. & Shimabukuro, Y.E.
(2012) Decoupling of deforestation and soy production in the southern Amazon during the
late 2000s. Proceedings of the National Academy of Sciences, 109, 1341–1346.
Malanson, G.P. (2002) Extinction-debt trajectories and spatial patterns of habitat destruction.
Annals of the Association of American Geographers, 92, 177–188.
Malanson, G.P. (2008) Extinction debt: Origins, developments, and applications of a
biogeographical trope. Progress in Physical Geography, 32, 277–291.
Martensen, A.C., Pimentel, R.G. & Metzger, J.P. (2008) Relative effects of fragment size and
connectivity on bird community in the Atlantic Rain Forest: Implications for conservation.
Biological Conservation, 141, 2184–2192.
Martensen, A.C., Ribeiro, M.C., Banks-Leite, C., Prado, P.I. & Metzger, J.P. (2012)
Associations of forest cover, fragment area, and connectivity with neotropical understory
bird species richness and abundance. Conservation Biology, 26, 1100–1111.
Martin, P., Bullock, J. & Newton, A. (2013) Carbon pools recover more rapidly than plant
biodiversity in secondary tropical forests. Philosophical Transactions of the Royal Society
B, 280, 20132236.
Martinelli, L.A., Naylor, R., Vitousek, P.M. & Moutinho, P. (2010) Agriculture in Brazil:
Impacts, costs, and opportunities for a sustainable future. Current Opinion in Environmental
Sustainability, 2, 431–438.
119
Martini, A.M.Z., Fiaschi, P., Amorim, A.M. & Paixão, J.L. da. (2007) A hot-point within a hot-
spot: A high diversity site in Brazil’s Atlantic Forest. Biodiversity and Conservation, 16,
3111–3128.
Matlack, G.R. & Monde, J. (2004) Consequences of low mobility in spatially and temporally
heterogeneous ecosystems. Journal of Ecology, 92, 1025–1035.
Maxwell, S.L., Fuller, R.A., Brooks, T.M. & Watson, J.E.M. (2016) The ravages of guns, nets
and bulldozers. Nature, 536, 146–145.
Mayaux, P., Holmgren, P., Achard, F., Eva, H., Stibig, H.-J. & Branthomme, A. (2005) Tropical
forest cover change in the 1990s and options for future monitoring. Philosophical
Transactions of the Royal Society B: Biological Sciences, 360, 373–384.
Melo, F.P.L., Arroyo-Rodríguez, V., Fahrig, L., Martínez-Ramos, M. & Tabarelli, M. (2013) On
the hope for biodiversity-friendly tropical landscapes. Trends in Ecology & Evolution, 28,
462–468.
Metzger, J.P. (2010) O Código Florestal tem base científica? Natureza & Conservação, 8, 92–
99.
Metzger, J.P., Martensen, A.C., Dixo, M., Bernacci, L.C., Ribeiro, M.C., Teixeira, A.M.G. &
Pardini, R. (2009) Time-lag in biological responses to landscape changes in a highly
dynamic Atlantic forest region. Biological Conservation, 142, 1166–1177.
Meyfroidt, P., Rudel, T.K. & Lambin, E.F. (2010) Forest transitions, trade, and the global
displacement of land use. Proceedings of the National Academy of Sciences, 107, 21300–
21305.
Minor, E.S.E.S. & Urban, D.L. (2007) Graph theory as a proxy for spatially explicit population
models in conservation planning. Ecological Applications, 17, 1771–1782.
Mitchell, M.E., Bennett, E. & Gonzalez, A. (2013) Linking landscape connectivity and
ecosystem service provision: Current knowledge and research gaps. Ecosystems, 16, 894–
908.
Moilanen, A. & Hanski, I. (2001) On the use of connectivity measures in spatial ecology. Oikos,
95, 147–151.
Moore, R.P., Robinson, W.D., Lovette, I.J. & Robinson, T.R. (2008) Experimental evidence for
extreme dispersal limitation in tropical forest birds. Ecology Letters, 11, 960–968.
Morton, D.C., DeFries, R.S., Shimabukuro, Y.E., Anderson, L.O., Arai, E., del Bon Espirito-
Santo, F., Freitas, R. & Morisette, J. (2006) Cropland expansion changes deforestation
dynamics in the southern Brazilian Amazon. Proceedings of the National Academy of
Sciences of the United States of America, 103, 14637–41.
Mueller, N.D., Gerber, J.S., Johnston, M., Ray, D.K., Ramankutty, N. & Foley, J.A. (2012)
Closing yield gaps through nutrient and water management. Nature, 490, 254–7.
120
Myers, N., Mittermeier, R.A., Mittermeier, C.G., da Fonseca, G.A.B. & Kent, J. (2000)
Biodiversity hotspots for conservation priorities. Nature, 403, 853–858.
Nepstad, D., Carvalho, G., Barros, A.C., Alencar, A., Capobianco, J.P., Bishop, J., Moutinho, P.,
Lefebvre, P., Silva, U.L. & Prins, E. (2001) Road paving, fire regime feedbacks, and the
future of Amazon forests. Forest Ecology and Management, 154, 395–407.
Nepstad, D., Soares-Filho, B.S., Merry, F., Lima, A., Moutinho, P., Carter, J., Bowman, M.,
Cattaneo, A., Rodrigues, H., Schwartzman, S., McGrath, D.G., Stickler, C.M., Lubowski,
R., Piris-Cabezas, P., Rivero, S., Alencar, A., Almeida, O. & Stella, O. (2009) The end of
deforestation in the Brazilian Amazon. Science, 326, 1350–1351.
Nobre, C.A., Sellers, P.J. & Shukla, J. (1991) Amazonian Deforestation and Regional Climate
Change. Journal of Climate, 4, 957–988.
North, A. & Ovaskainen, O. (2007) Interactions between dispersal, competition, and landscape
heterogeneity. Oikos, 116, 1106–1119.
OECD/FAO. (2011) OECD-FAO Agricultural Outlook 2011. OECD Publishing.
Oliveira, A.U. De. (2001) A longa marcha do campesinato brasileiro: movimentos sociais,
conflitos e Reforma Agrária. Estudos Avançados, 15, 185–206.
Pardini, R., Faria, D., Accacio, G.M., Laps, R.R., Mariano-Neto, E., Paciencia, M.L.B., Dixo, M.
& Baumgarten, J. (2009) The challenge of maintaining Atlantic forest biodiversity : A
multi-taxa conservation assessment of specialist and generalist species in an agro-forestry
mosaic in southern Bahia. Biological Conservation, 142, 1178–1190.
Pereira, H.M., Borda-de-Água, L. & Martins, I.S. (2012) Geometry and scale in species–area
relationships. Nature, 482, E3–E4.
Pereira, H.M., Leadley, P.W., Proenca, V., Alkemade, R., Scharlemann, J.P.W., Fernandez-
Manjarres, J.F., Araujo, M.B., Balvanera, P., Biggs, R., Cheung, W.W.L., Chini, L.,
Cooper, H.D., Gilman, E.L., Guenette, S., Hurtt, G.C., Huntington, H.P., Mace, G.M.,
Oberdorff, T., Revenga, C., Rodrigues, P., Scholes, R.J., Sumaila, U.R. & Walpole, M.
(2010) Scenarios for global biodiversity in the 21st century. Science, 330, 1496–1501.
Peres, C.A. (2011) Conservation in sustainable-use tropical forest reserves. Conservation
Biology, 25, 1124–1129.
Perfecto, I. & Vandermeer, J. (2008) Biodiversity conservation in tropical agroecosystems: A
new conservation paradigm. Annals of the New York Academy of Sciences, 1134, 173–200.
Perfecto, I. & Vandermeer, J. (2010) The agroecological matrix as alternative to the land-
sparing/agriculture intensification model. Proceedings of the National Academy of Sciences
of the United States of America, 107, 5786–5791.
Perz, S., Brilhante, S., Brown, F., Caldas, M., Ikeda, S., Mendoza, E., Overdevest, C., Reis, V.,
Reyes, J.F., Rojas, D., Schmink, M., Souza, C. & Walker, R. (2008) Road building, land use
121
and climate change: prospects for environmental governance in the Amazon. Philosophical
transactions of the Royal Society of London. Series B, Biological sciences, 363, 1889–95.
Perz, S. & Skole, D. (2003) Secondary forest expansion in the Brazilian Amazon and the
refinement of forest transition theory. Society & Natural Resources, 16, 277–294.
Phalan, B., Green, R.E., Dicks, L. V, Dotta, G., Feniuk, C., Lamb, A., Strassburg, B.B.N.,
Williams, D.R., Ermgassen, E.K.H.J. z. & Balmford, A. (2016) How can higher-yield
farming help to spare nature? Science, 351, 450–451.
Phalan, B., Onial, M., Balmford, A. & Green, R.E. (2011) Reconciling food production and
biodiversity conservation: Land sharing and land sparing compared. Science, 333, 1289–
1291.
Pianka, E.R. (1966) Latitudinal gradients in species diversity: A review of concepts. The
American Naturalist, 100, 33–46.
Pimm, S.L. & Askins, R.A. (1995) Forest losses predict bird extinctions in eastern North
America. Proceedings of the National Academy of Sciences, 92, 9343–9347.
Pimm, S.L. & Raven, P. (2000) Extinction by numbers. Nature, 403, 843–845.
Pimm, S.L., Russell, G.J., Gittleman, J.L. & Brooks, T.M. (1995) The future of biodiversity.
Science, 269, 347–350.
Piotto, D. (2011) Spatial Dynamics of Forest Recovery after Swidden Cultivations in the Atlantic
Forest of Southern Bahia, Brazil. Yale University.
Piotto, D., Montagnini, F., Thomas, W., Ashton, M. & Oliver, C. (2009) Forest recovery after
swidden cultivation across a 40-year chronosequence in the Atlantic forest of southern
Bahia, Brazil. Plant Ecology, 205, 261–272.
Poorter, L., Bongers, F., Aide, T.M., Almeyda Zambrano, A.M., Balvanera, P., Becknell, J.M.,
Boukili, V., Brancalion, P.H.S., Broadbent, E.N., Chazdon, R.L., Craven, D., de Almeida-
Cortez, J.S., Cabral, G.A.L., de Jong, B.H.J., Denslow, J.S., Dent, D.H., DeWalt, S.J.,
Dupuy, J.M., Durán, S.M., Espírito-Santo, M.M., Fandino, M.C., César, R.G., Hall, J.S.,
Hernandez-Stefanoni, J.L., Jakovac, C.C., Junqueira, A.B., Kennard, D., Letcher, S.G.,
Licona, J.-C., Lohbeck, M., Marín-Spiotta, E., Martínez-Ramos, M., Massoca, P., Meave,
J.A., Mesquita, R., Mora, F., Muñoz, R., Muscarella, R., Nunes, Y.R.F., Ochoa-Gaona, S.,
de Oliveira, A.A., Orihuela-Belmonte, E., Peña-Claros, M., Pérez-García, E.A., Piotto, D.,
Powers, J.S., Rodríguez-Velázquez, J., Romero-Pérez, I.E., Ruíz, J., Saldarriaga, J.G.,
Sanchez-Azofeifa, A., Schwartz, N.B., Steininger, M.K., Swenson, N.G., Toledo, M.,
Uriarte, M., van Breugel, M., van der Wal, H., Veloso, M.D.M., Vester, H.F.M., Vicentini,
A., Vieira, I.C.G., Bentos, T.V., Williamson, G.B. & Rozendaal, D.M.A. (2016) Biomass
resilience of Neotropical secondary forests. Nature, 530, 211–214.
Pütz, S., Groeneveld, J., Henle, K., Knogge, C., Martensen, A.C., Metz, M., Metzger, J.P.,
Ribeiro, M.C., de Paula, M.D. & Huth, A. (2014) Long-term carbon loss in fragmented
Neotropical forests. Nature Communications, 5, 5037.
122
Ragauskas, A.J., Williams, C.K., Davison, B.H., Britovsek, G., Cairney, J., Eckert, C.A.,
Frederick Jr., W.J., Hallett, J.P., Leak, D.J., Liotta, C.L., Mielenz, J.R., Murphy, R.,
Templer, R. & Tschaplinski, T. (2006) The path forward for biofuels and biomaterials.
Science, 311, 484–489.
Rands, M.R.W., Adams, W.M., Bennun, L., Butchart, S.H.M., Clements, A., Coomes, D.,
Entwistle, A., Hodge, I., Kapos, V., Scharlemann, J.P.W., Sutherland, W.J. & Vira, B.
(2010) Biodiversity conservation: Challenges beyond 2010. Science, 329, 1298–303.
Rayfield, B., Fortin, M.-J. & Fall, A. (2011) Connectivity for conservation: A framework to
classify network measures. Ecology, 92, 847–858.
Reaka-Kudla, M.L. (1997) The global biodiversity of coral reefs: A comparison with rainforests.
Biodiversity II: Understanding and Protecting our NAtural Resources (eds M.L. Reaka-
Kudla, D.E. Wilson, & E.. Wilson), p. Joseph Henry/National Academy Press, Washington
D.C.
Ribeiro, R., Carretero, M.A., Sillero, N., Alarcos, G., Ortiz-Santaliestra, M., Lizana, M. &
Llorente, G.A. (2011) The pond network: Can structural connectivity reflect on (amphibian)
biodiversity patterns? Landscape Ecology, 26, 673–682.
Ribeiro, M.C., Holvorcem, C.G.D., Marques, A., Martensen, A.C., Metzger, J.P. & Tambosi,
L.R. (2012) Monitoramento Independente Da Cobertura Florestal Das Bacias Setentrionais
Do Extremo Sul Da Bahia.
Ribeiro, M.C., Metzger, J.P., Martensen, A.C., Ponzoni, F.J. & Hirota, M.M. (2009) The
Brazilian Atlantic Forest : How much is left , and how is the remaining forest distributed ?
Implications for conservation. Biological Conservation, 142, 1141–1153.
Rosalino, L.M., Martin, P.S., Gheler-Costa, C., Lopes, P.C. & Verdade, L.M. (2014) Neotropical
small mammals’ diversity in the early cycle of commercial Eucalyptus plantations.
Agroforestry Systems, 88, 427–436.
Rudel, T.K., Bates, D. & Machinguiashi, R. (2002) A Tropical forest transition? Agricultural
change, out-migration, and secondary forests in the Ecuadorian Amazon. Annals of the
Association of American Geographers, 92, 87–102.
Rudel, T.K., Perez-Lugo, M. & Zichal, H. (2000) When fields revert to forest: Development and
spontaneous reforestation in post-war Puerto Rico. The Professional Geographer, 52, 386–
397.
Rudorff, B.F.T., Adami, M., Aguiar, D.A., Moreira, M.A., Mello, M.P., Fabiani, L., Amaral,
D.F. & Pires, B.M. (2011) The soy moratorium in the Amazon biome monitored by remote
sensing images. Remote Sensing, 3, 185–202.
Rybicki, J. & Hanski, I. (2013) Species-area relationships and extinctions caused by habitat loss
and fragmentation. Ecology Letters, 16, 27–38.
Saura, S., Bodin, Ö. & Fortin, M.-J. (2014) Stepping stones are crucial for species’ long-distance
123
dispersal and range expansion through habitat networks. Journal of Applied Ecology, 51,
171–182.
Saura, S., Estreguil, C., Mouton, C. & Rodríguez-Freire, M. (2011) Network analysis to assess
landscape connectivity trends: Application to European forests (1990–2000). Ecological
Indicators, 11, 407–416.
Saura, S. & Rubio, L. (2010) A common currency for the different ways in which patches and
links can contribute to habitat availability and connectivity in the landscape. Ecography, 33,
523–537.
Sills, E.O., Atmadja, S.S., Sassi, C. de, Duchelle, A.E., Kweka, D.L., Resosudarmo, I.A.P. &
Sunderlin, W.D. (2014) REDD+ on the Ground: A Case Book of Subnational Initiatives
across the Globe (eds E.O. Sills, S.S. Atmadja, C. de Sassi, A.E. Duchelle, D.L. Kweka,
I.A.P. Resosudarmo, & W.D. Sunderlin). CIFOR, Bogor, Indonesia.
Silva, J.H.G. (2009) Economic causes of deforestation in the Brazilian Amazon: an empirical
analysis of the 2000s. Xxxvii Encontro Nacional De Economia, 1–26.
Silva, J., Nobre, A. & Manzatto, C. (2011) O Código Florestal E a Ciência: Contribuições Para
O Diálogo.
Silveira, L.F., Olmos, F. & Long, A.J. (2004) Taxonomy, history, and status of Alagoas
Curassow Mitu mitu (Linnaeus, 1766), the world’s most threatened cracid. Ararajuba, 12,
125–132.
Simberloff, D. (1992) Do species–area curves predict extinction in fragmented forest? Tropical
deforestation and species extinction (eds T.C. Whitmore & J.A. Sayer), pp. 75–90.
Chapman & Hall, London.
Smith, N., Dubois, J. & Current, D. (1998) Agroforestry Experiences in the Brazilian Amazon:
Constraints and Opportunities. Brasília.
Soares-Filho, B.S., Nepstad, D.C., Curran, L.M., Cerqueira, G.C., Garcia, R.A., Ramos, C.A.,
Voll, E., McDonald, A., Lefebvre, P. & Schlesinger, P. (2006) Modelling conservation in
the Amazon basin. Nature, 440, 520–523.
Sodhi, N.S., Koh, L.P., Brook, B.W. & Ng, P.K.L. (2004) Southeast Asian biodiversity: An
impending disaster. Trends in Ecology & Evolution, 19, 654–660.
Sodhi, N.S., Wilcove, D.S., Lee, T.M., Sekercioglu, C.H., Subaraj, R., Bernard, H., Yong, D.L.,
Lim, S.L.H., Prawiradilaga, D.M. & Brook, B.W. (2010) Deforestation and avian extinction
on tropical landbridge islands. Conservation Biology, 24, 1290–1298.
Sparovek, G., Berndes, G., Klug, I.L.F. & Barretto, A.G.O.P. (2010) Brazilian agriculture and
environmental legislation: Status and future challenges. Environmental Science &
Technology, 44, 6046–6053.
Stallings, J.R. (1991) The importance of understorey on wildlife in a brazilian eucalypt
124
plantation. Revista Brasileira de Zoologia, 7, 267–276.
Taylor, P., Fahrig, L., Henein, K. & Merriam, G. (1993) Connectivity is a vital element of
landscape structure. Oikos, 68, 571–573.
Teixeira, A.M.G., Soares-Filho, B.S., Freitas, S.R. & Metzger, J.P. (2009) Modeling landscape
dynamics in an Atlantic Rainforest region: Implications for conservation. Forest Ecology
and Management, 257, 1219–1230.
Thomas, W., Carvalho, A., Amorim, A., Garrison, J. & Arbeláez, A. (1998) Plant endemism in
two forests in southern Bahia, Brazil. Biodiversity and Conservation, 7, 311–322.
Tilman, D., May, R.M., Lehman, C.L. & Nowak, M.A. (1994) Habitat destruction and the
extinction debt. Nature, 371, 65–66.
Tischendorf, L. & Fahrig, L. (2000) On the usage and measurement of landscape connectivity.
Oikos, 1, 7–19.
Tollefson, J. (2016) Brazil debates loosening environmental protections. Nature, 539, 147–148.
Tscharntke, T., Clough, Y., Wanger, T.C., Jackson, L., Motzke, I., Perfecto, I., Vandermeer, J. &
Whitbread, A. (2012) Global food security, biodiversity conservation and the future of
agricultural intensification. Biological Conservation, 151, 53–59.
Tscharntke, T., Klein, A.M., Kruess, A., Steffan-Dewenter, I. & Thies, C. (2005) Landscape
perspectives on agricultural intensification and biodiversity - Ecosystem service
management. Ecology Letters, 8, 857–874.
Turner, M.G. (2010) Disturbance and landscape dynamics in a changing world. Ecology, 91,
2833–2849.
Uezu, A., Beyer, D.D. & Metzger, J.P. (2008) Can agroforest woodlots work as stepping stones
for birds in the Atlantic forest region? Biodiversity and Conservation, 17, 1907–1922.
Uezu, A. & Metzger, J.P. (2016) Time-lag in responses of birds to Atlantic forest fragmentation:
Restoration opportunity and urgency. PLoS ONE, 11, 1–16.
Uezu, A., Metzger, J.P. & Vielliard, J.M.E. (2005) Effects of structural and functional
connectivity and patch size on the abundance of seven Atlantic Forest bird species.
Biological Conservation, 123, 507–519.
Umetsu, F. & Pardini, R. (2007) Small mammals in a mosaic of forest remnants and
anthropogenic habitats - Evaluating matrix quality in an Atlantic forest landscape.
Landscape Ecology, 22, 517–530.
Urban, D. & Keitt, T. (2001) Landscape connectivity: A graph-theoretic perspective. Ecology,
82, 1205–1218.
Urban, D.L., Minor, E.S., Treml, E. a & Schick, R.S. (2009) Graph models of habitat mosaics.
125
Ecology Letters, 12, 260–273.
Vale, P.M. (2014) The conservation versus production trade-off : does livestock intensification
increase deforestation ? Evidence from the Brazilian Amazon. Centre for Climate Change
Economics and Policy the Environment Working paper.
Walker, R., Moore, N.J., Arima, E., Perz, S., Simmons, C., Caldas, M., Vergara, D. & Bohrer, C.
(2009) Protecting the Amazon with protected areas. Proceedings of the National Academy
of Sciences of the United States of America, 106, 10582–6.
Wearn, O.R., Reuman, D.C. & Ewers, R.M. (2012) Extinction debt and windows of conservation
opportunity in the Brazilian Amazon. Science, 337, 228–232.
Whitmore, T.C. (1998) Potential Impact of Climatic Change on Tropical Rain Forest Seedlings
and Forest Regeneration. Climatic Change, 39, 429–438.
Wilson, E.O. (1988) Biodiversity. National Academy of Sciences, Washington D.C.
Wilson, M.C., Chen, X.-Y., Corlett, R.T., Didham, R.K., Ding, P., Holt, R.D., Holyoak, M., Hu,
G., Hughes, A.C., Jiang, L., Laurance, W.F., Liu, J., Pimm, S.L., Robinson, S.K., Russo,
S.E., Si, X., Wilcove, D.S., Wu, J. & Yu, M. (2016) Habitat fragmentation and biodiversity
conservation: Key findings and future challenges. Landscape Ecology, 31, 219–227.
Wimberly, M.C. (2006) Species dynamics in disturbed landscapes: When does a shifting habitat
mosaic enhance connectivity? Landscape Ecology, 21, 35–46.
Wright, S.J. (2005) Tropical forests in a changing environment. Trends in Ecology and
Evolution, 20, 553–560.
Wright, S.J. (2010) The future of tropical forests. Annals of the New York Academy of Sciences,
1195, 1–27.
126