Resilience of New Zealand indigenous forest fragments to impacts of livestock and pest mammals

Abstract

A number of factors have combined to diminish ecosystem integrity in New Zealand indigenous lowland forest fragments surrounded by intensively grazed pasture. Livestock grazing, mammalian pests, adventive weeds and altered nutrient input regimes are important drivers compounding the changes in fragment structure and function due to historical deforestation and fragmentation. We used qualitative systems modelling and empirical data from Beilschmiedia tawa dominated lowland forest fragments in the Waikato Region to explore the relevance of two common resilience paradigms – engineering resilience and ecological resilience – for addressing the conservation management of forest fragments into the future. Grazing by livestock and foraging/predation by introduced mammalian pests both have direct detrimental impacts on key structural and functional attributes of forest fragments. Release from these perturbations through fencing and pest control leads to partial or full recovery of some key indicators (i.e. increased indigenous plant regeneration and cover, increased invertebrate populations and litter mass, decreased soil fertility and increased nesting success) relative to levels seen in larger forest systems over a range of timescales. These changes indicate that forest fragments do show resilience consistent with adopting an engineering resilience paradigm for conservation management, in the landscape context studied. The relevance of the ecological resilience paradigm in these ecosystems is obscured by limited data. We characterise forest fragment dynamics in terms of changes in indigenous species occupancy and functional dominance, and present a conceptual model for the management of forest fragment ecosystems.

Full text

Dodd et al.: Forest fragment resilience
Resilience of New Zealand indigenous forest fragments to impacts of livestock and
pest mammals
Mike Dodd1*, Gary Barker2, Bruce Burns3, Raphael Didham4,5, John Innes2, Carolyn King6, Mark
Smale2 and Corinne Watts2
1AgResearch Grasslands, Private Bag 11008, Palmerston North 4442, New Zealand
2Landcare Research, Private Bag 3127, Hamilton 3240, New Zealand
3School of Biological Sciences, University of Auckland, Private Bag 92019, Auckland 1142, New Zealand
4School of Biological Sciences, University of Canterbury, Private Bag 4800, Christchurch 8140, New Zealand
5Present Address: School of Animal Biology, The University of Western Australia, 35 Stirling Highway, Crawley WA 6009,
Australia and CSIRO Entomology, Centre for Environment and Life Sciences, Underwood Avenue, Floreat WA 6014, Australia
6School of Biological Sciences, University of Waikato, Private Bag 3105, Hamilton 3240, New Zealand
*Author for correspondence (Email: mike.dodd@agresearch.co.nz)
Published on-line: 19 December 2010
Abstract: A number of factors have combined to diminish ecosystem integrity in New Zealand indigenous
lowland forest fragments surrounded by intensively grazed pasture. Livestock grazing, mammalian pests,
adventive weeds and altered nutrient input regimes are important drivers compounding the changes in fragment
structure and function due to historical deforestation and fragmentation. We used qualitative systems modelling
and empirical data from Beilschmiedia tawa dominated lowland forest fragments in the Waikato Region to
explore the relevance of two common resilience paradigms – engineering resilience and ecological resilience
– for addressing the conservation management of forest fragments into the future. Grazing by livestock and
foraging/predation by introduced mammalian pests both have direct detrimental impacts on key structural and
functional attributes of forest fragments. Release from these perturbations through fencing and pest control
leads to partial or full recovery of some key indicators (i.e. increased indigenous plant regeneration and cover,
increased invertebrate populations and litter mass, decreased soil fertility and increased nesting success) relative
to levels seen in larger forest systems over a range of timescales. These changes indicate that forest fragments
do show resilience consistent with adopting an engineering resilience paradigm for conservation management,
in the landscape context studied. The relevance of the ecological resilience paradigm in these ecosystems is
obscured by limited data. We characterise forest fragment dynamics in terms of changes in indigenous species
occupancy and functional dominance, and present a conceptual model for the management of forest fragment
ecosystems.
Keywords: biodiversity; model; perturbation; system dynamics
Introduction
Lowland native forest fragments are the poorly represented
remnants of one of the most damaged and threatened indigenous
ecosystems in New Zealand (Craig et al. 2000; Ewers et al.
2006). Extensive destruction of the original forests during
two human colonisation events has left a relictual landscape
(sensu McIntyre & Hobbs 1999) in which these vegetation
components are subject to repeated, severe perturbation events
with no comparable historical analogue. Aside from the habitat
loss and fragmentation process itself, these perturbations have
included selective harvesting of certain canopy dominant trees
(Nicholls 1979), intermittent browsing and soil disturbance
by domestic livestock (Jane 1983), the introduction of pest
mammals and plants (Craig et al. 2000), elevated rates of
inorganic nutrient input via agricultural fertiliser drift and/or
animal transfer (Stevenson 2004), altered hydrological regimes
through drainage of the surrounding pastoral land (Whaley
et al. 1997) and exposure to agricultural herbicides. With the
exception of the initial logging, all of these perturbations
can be regarded as ‘press disturbances’ (Bengtsson 2002),
since the agents (in the case of livestock, pests, drainage,
agrichemical and fertiliser transfer) or latent effects (in the
case of fragmentation) continue to operate for at least several
decades and up to a century in some regions.
As a result of the combined effects of multiple
perturbations, the structure and functioning of native forest
fragment ecosystems have been highly modied. Canopy
and subcanopy vegetation cover has been reduced and
regeneration of canopy trees has been inhibited (Smale et al.
2005, 2008; Burns et al. in press), forest soil litter and organic
layers have been reduced or removed, nesting success of all
native bird species has declined or been prevented altogether
(Innes et al. 2004; Boulton et al. 2008) and adult occurrence
reduced to sporadic visits (Stevens 2006), indigenous plant
species diversity has been reduced (Smale et al. 2008) and
some plant and bird species have been extirpated (Whaley
et al. 1997; Miskelly et al. 2008; Innes et al. 2010a). While
the invertebrate fauna has been shown to remain relatively
abundant and diverse in fragments dominated by indigenous
plants (Crisp et al. 1998; Harris & Burns 2000), substantial
dissimilarity in the composition of invertebrate taxa has been
shown relative to more intact reference forests (Didham et al.
2009), with largely unknown effects on ecosystem functioning.
It seems clear that without some countervailing intervention,
native forest fragments will remain in a degraded state and
New Zealand Journal of Ecology (2011) 35(1): 0-0 © New Zealand Ecological Society.
Available on-line at: http://www.newzealandecology.org/nzje/

New Zealand Journal of Ecology, Vol. 35, No. 1, 2011
been suggested that a more appropriate measure of resilience
in this context is the magnitude of the perturbation required to
force the regime shift (Holling 1996). Any measure of resilience
relevant to this scenario must consider the degree of deviation
from the normal domain and how this changes over time.
Orwin and Wardle (2004) have thoughtfully examined such
measures and developed a mathematically and conceptually
robust index.
For whichever concept of resilience one is working with,
it is clear that resilience is a relative concept – there is only
value in considering the relative resilience of either a specic
ecosystem to multiple disturbances, or multiple ecosystems
to the same disturbance. Hence there is a need to be specic
about the system(s) and perturbation(s) of interest (Carpenter
et al. 2001).
may eventually disappear from the lowland landscape, as the
ageing population of indigenous canopy trees gradually dies
out and the historical ‘extinction debt’ (Tilman et al. 1994)
catches up with forest fragments surviving for the present
under the prevailing landscape regime.
In this paper, we discuss whether this grim outlook for
forest fragments is avoidable, given the small size of most
fragments, the high degree of degradation that they have already
sustained and the overwhelmingly adverse surrounding matrix
environment in which they exist. Specically, we question
whether forest fragments can be restored to a self-sustaining
structure and function that is characteristically indigenous,
through management that releases them from the multiple
perturbations they face. This is essentially a question about
resilience (see Box 1), which is a key ecological concept
relevant to understanding the nature and effects of ecosystem
recovery following perturbations. We sought to determine
whether there is good evidence that native forest fragments
Resilience concepts in ecology
Two concepts of resilience are most commonly seen in the
ecological literature – engineering resilience and ecological
resilience (Gunderson 2000; Bengtsson 2002). The rst
incorporates the ability of a system to recover its initial
structure or character following a perturbation (Pimm 1984)
and has been referred to as ‘engineering resilience’ because of
its similarity to attributes measured in that discipline (Holling
1996). This more conventional concept of resilience assumes
there is a global stability domain for a system property (i.e.
the typical or ‘normal’ range in that system property observed
over time or space as a result of environmental variation)
that the system returns to naturally following a signicant
perturbation (see Fig. 1). Following a perturbation that leads
to a deviation from this range, engineering resilience can be
measured as the time taken for the property to return to the
normal range (see Ludwig et al. 1996).
However, some ecosystems have been observed not to
return to an expected global stability domain after perturbation,
but have remained within an alternative stable domain, having
crossed a hysteresis-type threshold (Suding & Hobbs 2009;
Fig. 1). These observations led to the development of another
resilience concept called ‘ecological resilience’ (Holling 1996).
This concept of resilience sets aside the assumption of a single
global equilibrium, in favour of multiple local equilibria
(or domains of attraction). The commonly depicted visual
model for the concept is a ball in a landscape of troughs and
ridges, with the ball representing the system and the troughs
representing alternative domains of attraction. Movement
of the ball between troughs is described as a ‘regime shift’,
since the controls on structure and functions have changed
(Gunderson 2000). The most well-developed cases in ecology
relate to rangeland grazing systems (with the background being
the development of ‘state and transition’ models; Westoby
et al. 1989) and lake systems (Walker et al. 1997; Scheffer &
Carpenter 2003) but many others have been suggested (see
Walker & Meyers 2004). Suding and Hobbs (2009) provide a
useful summary and glossary of these concepts in the context
of restoration ecology.
In the case of a regime shift the notion of a return time
becomes meaningless, and the relative resilience of systems
cannot be quantitatively assessed using this measure. It has
Figure 1. Concepts of engineering and ecological resilience in
ecosystems. A system with greater engineering resilience (A,
long dash line) would have a shorter return-time interval (T1–T0)
compared with a system with lesser engineering resilience (B,
solid line) with a longer return-time interval (T2–T0) following
perturbation of a system property from P0 to Px. A system with
greater ecological resilience (A or B) will return to the ‘normal’
stable domain following a perturbation of the system property
from P0 to Px compared with a system with lesser ecological
resilience (C, short dash line) that enters an alternative stable
domain following a perturbation of the same magnitude, having
crossed a hysteresis-type threshold between P0 and Px.
exhibit the characteristics indicative of either of the types
of resilience – engineering or ecological – described in Box
1. We did not seek to support or refute the validity of the
concepts but rather to explore their relevance to the system
of interest (Suding & Hobbs 2009). In order to develop sound
management approaches for forest fragments, it is useful to
understand how the inherent resilience of these ecosystems
may be utilised, or may represent an obstacle to success, in
achieving management goals.
More specically, our objectives were to (1) explore the
relevance of the two resilience concepts (Box 1) to indigenous
lowland forest fragments; (2) assess the resilience of these
forest fragments to the major perturbations resulting from stock
and pest mammals; and (3) develop a conceptual management
framework that might account for and utilise this resilience,
in order to develop long-term strategies for conservation
management of forest fragments.
46
Figure 1 1
2
Time
Ecosystem property value
Alternative stable
do
m
a
in
Perturbation
T0 T2
‘Normal’ stable
domain
T1
A
B
C
P0
Px
Perturbation
'Normal' stable
domain
Alternative stable
domain
Time
T1T2
T0
P0
Px
Ecosystem property value
BOX 1

Dodd et al.: Forest fragment resilience
Approach
We used two approaches to explore resilience, the rst based
on a consensus model of forest fragment system dynamics
developed by a team of ecosystem scientists and the second
based on empirical data generated in the ‘Forest fragment
resilience’ research project (Innes 2009), supplemented by
other information on forest fragment disturbance dynamics
gleaned from the literature.
Our evidential basis for drawing conclusions about the
relative resilience of forest fragments requires rst that they
show a signicant response of key system properties to the
perturbations of interest (grazing and mammalian pests). This
immediately highlights the issue of how to dene the initial
conditions, in the absence of fragments that have never been
grazed or infested with weeds and pests. Studies to date have
used larger forest tracts, but two difculties are apparent: (1)
the fact that the reference forests are not fragmented, and
thus any fragmentation-specic effects or interactions will
not be expressed, e.g. species–area relationships, dispersal
limitations; and (2) the idiosyncratic management of most
reference forests, in terms of which mammalian pests and
weeds are controlled and over what time period. Despite these
limitations, reference forests represent a major resource for
our ability to infer initial conditions, guided by our growing
understanding of the historical condition and dynamics of
indigenous forest ecosystems. Hence in this study we also
cautiously use reference forests to inform our assessment of
resilience, focusing on high-level indicators to minimise the
associated uncertainty.
Having established the existence of perturbation effects,
relative engineering resilience would be indicated by the
extent to which system properties show a return response
following release from those perturbations, over the timescales
encompassed by the data (one or two decades): high engineering
resilience is indicated by full recovery and low engineering
resilience is indicated by partial recovery (cases A and B in
Fig. 1). By contrast, relative ecological resilience would be
indicated by the occurrence of a return response to release
from the perturbations along with changes in system controls
(as indicated by the feedback loops described below): high
ecological resilience is indicated by a full or partial recovery and
low ecological resilience is indicated by a lack of recovery after
multiple decades accompanied by changes in system controls
(case C in Fig. 1). Such changes in system controls should
distinguish low ecological resilience from low engineering
resilience when observing very slow recovery.
System dynamics model
We developed a system model of forest fragments using the
causal loop diagram approach from the discipline of system
dynamics (Maani & Cavana 2007). The rationale for using this
approach is that an explicit documentation of the dynamics of the
ecosystem should: (a) help integrate our understanding of key
ecosystem processes across different ecosystem components
(plants, invertebrates, soils, mesofauna); (b) highlight the most
important structural and functional parameters of the ecosystem
to guide measurement and monitoring; and (c) reveal a number
of features of system behaviour that are relevant to analysing
resilience, specically the nature of any feedback loops.
Within the system dynamics framework, we identied key
system variables and depicted the nature of the relationships
between those variables with annotated arrows. The direction
of the arrow indicates a cause–effect relationship, while the
annotation indicates the direction of the effect, thus: + represents
a positive effect (i.e. both variables increase together or both
variables decrease together) and − represents a negative effect
(i.e. one variable increases while the other decreases). Within
this system diagram we identied the feedback loops, of which
there are two types:
(1) Reinforcing loops (R), whereby the feedback is positive
overall (when multiplying the signs of the annotations)
and results in enhancement of the initial change in the
key variable, also called a ‘vicious’ or ‘virtuous’ cycle.
(2) Balancing loops (B), whereby the feedback is negative
overall and results in moderation of the initial change in
the key variable.
The system model was initially built for a ‘natural’ system,
excluding the role of the major human-mediated perturbations
of interest (livestock grazing, pest invasion). The relevant
perturbations were then considered in terms of whether they
modied the system state variables, modied the nature of the
feedback loops, or added new feedbacks (see Fig. 2).
Experimental data
Forty-seven low altitude forest fragments in the Waikato
Region, ranging in area from 0.5 to 24 ha, were surveyed during
the summer of 2006/07 (for details see Didham et al. 2009;
Burns et al. in press). The fragments were selected according to
distinct historical management regimes of fencing to exclude
livestock and pest control, arranged in a factorial design. This
arrangement included four categories of fencing: unfenced or
fenced <2 years; fenced 2–10 years; fenced 10–20 years; and
fenced >20 years) and two categories of pest control targeted
at possums (Trichosurus vulpecula) and ship rats (Rattus
rattus): uncontrolled and controlled – the latter dened as
being conducted for a minimum of 2 years and usually > 10
years, with at least annual repeats at a minimum of one trap
or bait station per hectare.
Data on vascular plant structure, plant species composition
and soil characteristics were collected from a subset (41) of tawa
(Beilschmiedia tawa) dominated forest fragments (Burns et al.
in press). Data on forest-oor invertebrate faunal composition
and site characteristics were also collected from a subset (30)
of the fragments (Didham et al. 2009). Bird nesting success
using real and model nests was assessed in fragments with
and without intensive ship rat and possum control, and ship
rat abundance was measured in fragments with and without
grazing (Innes et al. 2010b). Corresponding data were collected
from, or obtained for, three large local forest reserves (Te Miro,
Karakariki and Maungatautari scenic reserves), all of which
were free of livestock but which have had only limited recent
mammalian pest control. Key ecosystem variables incorporated
into the system dynamics model were selected from the plant,
soil, invertebrate and bird data, which were transformed
into semi-quantitative values (i.e. nil–low–medium–high).
These values were compared with estimates of the ‘normal
domain’ from the eld studies and other available literature.
This comparison aimed to examine the system’s responses
to removal of livestock and pest mammal perturbations, and
to determine if the measured responses provided evidence of
engineering and/or ecological resilience.

New Zealand Journal of Ecology, Vol. 35, No. 1, 2011
47
Figure 2 1
2
3
_
_
_
_
_
_
_
+
_
_
_
+
Livestock
densit
y
Mammalian
omnivore density
Mammalian
carnivore density
+
_
+
+
+
+
+
+
+
_
+
+
+
Bpredat
_
+
+
Plant understorey
cover
Plant canopy
cove
r
Mineralisation rate
Litter mass
Plant
biomass
Viable seed
numbers
Sapling
establishment
Bshading
Rdecomp
+
Ground
invertebrate
densit
y
Understorey
light
Bird density
+
Rregen
Bherbiv
Soil fertility
Palatable plant
biomass
Litter quality
+
_
Figure 2. Causal-loop system
dynamics model of a forest
fragment ecosystem showing key
reinforcing loops (regeneration,
Rregen; decomposition, R decomp) and
balancing loops (shading, Bshading;
herbivory, B
herbiv; predation,
Bpredat). New structure created by the
perturbations of interest is shown by
dashed boxes and arrows indicating
major direct effects. Symbols: +
positive effect; − negative effect.
Results and discussion
System dynamics model
The quite complex initial cause-and-effect diagrams were
simplied to include only the key causal loops identied as
major controls (Fig. 2). Two linked clusters of reinforcing
and balancing loops were evident. The rst cluster was
based around plant growth and regeneration. The reinforcing
component of this pair involves the standing vegetation biomass
generating current seed production, which germinates under
favourable conditions, leading to regeneration of saplings in
the understorey (Fig. 2) These saplings contribute initially to
understorey cover but also via a slowly operating process to
canopy cover (for the relevant canopy species, i.e. tawa and
rewarewa – Knightia excelsa). A component of those favourable
conditions is the light environment under the canopy, which
provides the balancing control on sapling regeneration. The
consumption of leaves and owers, predominantly by avian
herbivores, in pre-disturbance New Zealand ecosystems (Clout
& Hay 1989) also provides a balancing loop that checks plant
growth and perhaps regeneration, but native birds are also
crucial pollinators and fruit dispersers (Kelly et al. 2006) and
so are necessary for plant regeneration.
The second cluster of reinforcing and balancing loops
was based around resource–consumer interactions and
decomposition processes (Fig. 2). The rst reinforcing
component of this pair involves the standing vegetation biomass
generating leaf litter, which is decomposed by the invertebrate
community, thereby recycling nutrients for plant growth.
The second reinforcing component consists of the suite of
relatively palatable plants associated with understorey cover,
which increases litter quality and inuences mineralisation
rates (Pastor et al. 1993). The linked balancing control on this
process is the predation of the macroinvertebrate community
by the avifauna. The plant regeneration balancing loop also
provides a long-term control on plant biomass and litter
production.
The direct effects of the two perturbations of interest in the
Viable seed
numbers
Mammalian
carnivore
density
Mammalian
omnivore
density
Bird
density
Plant
biomass
Understorey
light
Plant canopy
cover Litter mass
Ground
invertebrate
density
Sapling
establishment
Plant understorey
cover
Palatable plant
biomass
Litter quality
Mineralisation rate
Soil fertility
Livestock
density

Dodd et al.: Forest fragment resilience
initial model are outlined in the dashed boxes and arrows in
Fig. 2. Livestock browsing can be a pulse or press disturbance,
but in the context of most New Zealand farm systems it involves
rotationally grazed livestock having access for repeated short
periods (e.g. wintering cattle) and thus should be regarded as
a press disturbance. Livestock have several impacts on forest
fragments, including removal of the understorey vegetation
(particularly the most palatable plants) by direct browsing,
suppression of sapling regeneration by direct browsing and
physical damage of seedlings by trampling (Jane 1983; Timmins
2002), introduction of adventive species through propagule
transport from the pastoral matrix (with likely enhancement of
establishment through soil disturbance) and elevated nutrient
supply (via faeces and urine). Overall, these impacts serve
to weaken the plant growth and regeneration reinforcing
loop by damage to seedlings and saplings, while at the same
time strengthening the herbivory balancing loop by addition
of herbivore consumption of foliage and strengthening the
soil fertility loop by addition of nutrients in dung and urine.
This direct effect may serve to offset negative effects on the
decomposition loop (see Wardle et al. 2001). However, it
was considered that the inclusion of livestock in the model
did not create any new reinforcing or balancing loops within
the system.
Livestock impacts
Livestock browsing is not analogous to that of indigenous
herbivore browsing in unperturbed systems. Because livestock
are primarily fed on the pastures surrounding forest fragments,
and their numbers are determined by the farm manager, there
is no balancing feedback loop to control their population and
their rate of consumption of forest fragment vegetation, as is
the case for native herbivores (Fig. 2). This external resource
subsidy effect creates the potential to push the system into a
new domain, through ongoing prevention of the recruitment of
canopy trees. At some point the existing trees will die through
natural attrition and the structure of the forest might change
sufciently such that removal of the livestock will not result
in a return to a pre-grazed state, which would be a case of a
regime shift to an alternative state (Box 1). It is thus valuable
to know for how long a forest fragment can be grazed by
livestock before this shift occurs. This period could be an
empirical measure of the ecological resilience of the forest
fragment to livestock grazing (i.e. the temporal magnitude of
the disturbance required to force the regime shift, sensu Holling
1996). We predict that the time frame of this process would be
related to the longevity of the dominant canopy species, which
for tawa is in the order of 300-plus years (West 1995).
In terms of empirical evidence for this process, Esler
(1978, p. 45) documents the degeneration of indigenous forest
fragments induced by long-term browsing and characterises
it as a three-stage process of understorey destruction, weed
infestation and nally canopy collapse, leaving only scattered
former subcanopy trees in a grassland matrix. On the other hand,
Esler (1978, pp. 73–77) also describes forest regeneration via
bracken/mānuka/gorse shrubland succession, resulting from
decreasing grazing intensity in steep sown grassland. In mesic
environments, vegetation often shows classical secondary
succession through shrubland to indigenous-dominant forest
over timescales of decades (McQueen 1993; Leathwick &
Rogers 1996; Sullivan et al. 2007) to centuries (as modelled
by Meurk & Hall 2006). Thus even in this case of complete
deforestation in mesic regions of New Zealand, an engineering
resilience framework may still be appropriate, since even
forest destruction may not represent a permanent regime shift.
Given the generally poor representation of native plant seed in
non-forest soils (Partridge 1989) and the absence of persistent
seed banks (Sem & Enright 1996; Moles et al. 2000), the local
availability of dispersed propagules from remaining forest and
the continuity of the associated dispersal mechanisms will be
important factors in forest regeneration from grassland (Meurk
& Hall 2006; Standish et al. 2009). Hence the climate and
landscape context of the forest fragment becomes an important
mediating factor in determining resilience.
Mammalian pest impacts
The impact of mammalian pests can also be considered as
a press disturbance in terms of the ubiquitous build-up and
ongoing maintenance of high pest populations in the vicinity of
all fragments (Batcheler & Cowan 1988; King 2005). As with
livestock, individual pest species (e.g. possums, ship rats) can
have multiple impacts on structural and functional components
of the system, which can be exacerbated by the occurrence of
a suite of pest species with multiple functional roles. This is
illustrated in Fig. 2 by the inclusion of two new state variables
representing mammalian omnivores and predators. It is worth
noting that while predation of birds would have been a feature
of natural systems, we did not regard this as a dominant
control on bird populations (Innes et al. 2010a), and hence
indigenous predation was omitted from Fig. 2. Mammalian
pests that are omnivores (possums, ship rats and mice Mus
musculus) have direct negative impacts on ora, though on
different system variables than those identied for livestock
impacts. Thus pest mammal impacts also weaken the growth
and regeneration reinforcing loop by consumption of seeds
and seedlings and strengthen the herbivory balancing loop by
additional herbivore consumption of foliage throughout the
canopy layers. Many omnivorous pests (particularly hedgehogs
Erinaceus europaeus) are also predators of invertebrates,
potentially weakening the decomposition reinforcing loop.
Thus, they are engaged in the herbivory and invertebrate
predation loops in the same manner as indigenous birds (i.e.
as competitors; Nugent et al. 2000; McQueen & Lawrence
2008). However, two of these omnivorous mammalian pests
(possums and ship rats) are also direct predators of indigenous
birds (via nest predation; Innes et al. 2004) and hence operate
as joint competitors and predators. Furthermore, they are also
prey items themselves (along with birds) for the introduced
mammalian predators (e.g. stoats Mustela erminea) and thus
create an important new predation balancing loop, whereby
the mammalian omnivores support a mammalian predator
population that can prey-switch between the mammalian
omnivores and the indigenous bird population (Murphy et al.
2008). Overall, the inclusion of this suite of mammalian pests
represents a new and quite complex dynamic in the forest
fragment ecosystem, which has two major detrimental aspects
for the avifauna, which we term the competitor-predator effect
and the predator-support effect.
The importance of external resource subsidies noted
previously for livestock grazing effects is also relevant for pest
mammals to a large extent, since they typically range widely
and have access to numerous food resources in the wider non-
forest landscape. For example, a substantial component of
possum diet appears to be high quality pasture in agricultural
landscapes (Harvie 1973; Nugent et al. 2000; Dodd et al.
2006). Consequently the populations of both livestock and
pest mammals that inhabit forest fragments are not subject
solely to internal balancing feedback controls within the

New Zealand Journal of Ecology, Vol. 35, No. 1, 2011
fragment, but also to the destabilising inuence of external
resource subsidies.
Resilience in the system model
The question of whether the model has any characteristics
that support the concepts of engineering and/or ecosystem
resilience can be addressed by considering what the model
might qualitatively predict following the release from livestock
and pest mammal perturbations, through management actions
such as fencing and pest control.
In the case of livestock browsing, our model indicates
that this perturbation functions within the context of existing
feedback loops (the reinforcing regeneration loop, and
the balancing shading and herbivory loops). This suggests
that the growth reinforcing loop will continue to function
following removal of the perturbation, leading to recovery
of the vegetative structure of the forest. Therefore, exclusion
of livestock browsing by fencing should lead to fairly rapid
recovery of sapling regeneration and understorey cover, with
slower effects on soil fertility (as elevated nutrients dissipate)
and plant diversity (as adventive herbaceous species become
subject to control by the shading loop). The lack of permanent
changes in system structure and feedback control due to
livestock browsing suggests that the fragment will exhibit
a release response in these variables that is characteristic of
relatively high engineering resilience. However, fencing in the
absence of pest control leaves open the question of whether the
weakening of the regeneration loop due to ower, fruit and seed
herbivory by mammalian omnivores will inhibit understorey
sapling regeneration in the medium term and thus result in a
decline in canopy cover in the longer term.
The effects of pest mammal control will be dependent on
which pests are controlled. Elimination of carnivores (stoats
and feral cats Felis catus) may have little positive impact on
the forest fragment, because of their mainly indirect effects and
low densities (often only 1–2 per 100 ha or fewer). Carnivore
removal may even exacerbate damage by omnivorous rodents
that are no longer subject to this balancing control, thus
ensuring there will effectively be no release from perturbation.
By contrast, the removal of the mammalian omnivores should
result in fairly rapid recovery of plant regeneration and
invertebrate populations, but will not necessarily benet the
avifauna while mammalian carnivores remain in the local
environment. A particular feature of the system is that there
is no balancing feedback from bird populations to either
the carnivore or omnivore mammal guilds (i.e. there is no
reduction in food supply resulting from predation that would
normally make either guild food-limited), since the omnivores
(ship rats, possums, mice) also eat fruit, seeds, leaves and
invertebrates and the carnivores (stoats, cats) also eat other
mammals. Thus, either guild has the potential to drive the
birds to extremely low population levels without any density-
dependent feedbacks. The resultant wholesale replacement of
this important component of the indigenous herbivores with an
introduced fauna could well constitute a ‘regime shift’ in the
language of ecological resilience. It is thus possible to envisage
a release response characteristic of engineering resilience in
components of the system (plants and invertebrates), but no
response in other components (birds) as evidence of a lack of
ecological resilience.
In summary, comparing the system dynamic effects of
livestock and mammalian pests leads to the conclusion that
both have extensive networks of impacts that ramify throughout
the system, which serves to emphasise the devastating effect
they have on lowland forest fragments. However, each operates
largely on different components of the system, which tends
to suggest (1) that the combination of both perturbations
will be far more detrimental to the whole system than either
one operating alone, and (2) that release from one or other
of the perturbations will have differential effects, leading to
differing restoration endpoints and possibly alternative stable
domains.
Empirical data
The system dynamics model indicated that a number of key
ecosystem variables would provide information on dynamic
responses to perturbation release, particularly with respect to
plant regeneration, plant canopy cover, palatable plant biomass,
litter mass, litter decomposition rate, invertebrate density,
soil fertility and bird populations. The semi-quantitative
data derived from the empirical data in the forest fragment
resilience study (Didham et al. 2009; Burns et al. in press),
with quantitative approximations, are shown in Table 1.
The data conrm the substantial effects of perturbation by
livestock grazing and mammalian pests on key indicators of
ecosystem structure in forest fragments relative to ungrazed
reference forest systems. Specic effects include declines in
plant regeneration (low seedling and very low sapling numbers),
palatable plant cover, invertebrate density, litter mass and
decomposition rate and increases in soil fertility (lower soil
C:N). These patterns are also reected in other data from the
limited literature on the ecological condition of forest fragments
in New Zealand. For example Smale et al. (2008) and Dodd and
Power (2007) have also shown inhibited regeneration and low
litter cover in grazed fragments. With regard to soils, Stevenson
(2004) has shown high levels of inorganic phosphorus, but
no differences in C and N, in forest fragments compared with
ungrazed forests. Soil-fertility-related properties appear to be
highly variable between forest fragments, reecting localised
fertiliser and stock management. In mid-elevation Nothofagus
forests in the South Island, Ewers et al. (2007) and Ewers and
Didham (2008) found that beetle community structure was
dramatically altered in small forest fragments and at the edges
of large forests, relative to interior forest sites, with the loss of
some interior forest specialists (Ewers & Didham 2004). Bird
nesting success is low in both large reference forests and small
fragments due to the ubiquitous distribution of pest mammals
(King 2005), although food shortage due to inadequate habitat
area is undoubtedly an additional problem for native birds in
fragmented landscapes (Innes et al. 2010a).
Resilience in the data
All of the indicators in Table 1 show return behaviour with
release of the fragments from either livestock grazing or
mammalian pest impacts. Some show a ‘full’ recovery –
seedling and sapling densities recover with fencing, invertebrate
densities recover with both fencing and pest control. Others
show a ‘partial’ recovery – canopy cover increases with pest
control, litter mass increases with fencing, soil C:N increases
with fencing. Partial recovery may simply be a function of
time, where feedback cycles operate over long periods and
thus rates of change are slow (e.g. soil C:N as a function of
litter return and decomposition). It may also be a function of
the incomplete nature of the perturbation release (e.g. only
possum and rat predation of invertebrates is controlled). Some
variables show no recovery, consistent with a lack of direct
relationship as indicated by the system dynamics model (e.g.

Dodd et al.: Forest fragment resilience
Table 1. Semi-quantitative dynamics of forest fragment ecosystem variables in reference forest, perturbed fragments and
fragments in response to long-term fencing and mammalian pest control, with approximations of the semi-quantitative
variables based on eld data (from Didham et al. 2009; Innes et al. 2010b; Burns et al. in press). DW = dry weight; C:N =
carbon:nitrogen.
__________________________________________________________________________________________________________________________________________________________________
System variable Indicator Units Low Medium High Reference Perturbed With With pest With
variable forest fragment fencing control both
__________________________________________________________________________________________________________________________________________________________________
Plant regeneration Seedling stems m–2 100 1000 10000 Medium– Low– Medium– Medium Medium–
density high medium high high
Plant regeneration Sapling
density stems m–2 10 100 1000 High Nil–low Medium– Nil–low Medium–
high high
Canopy cover Cover of % cover 70 80 90 High Medium Medium Medium- Medium–
canopy high high
species
Palatable plant Palatable % cover 5 10 20 High Nil–low Medium– Low High
biomass plant cover of dened high
spp.1
Decomposition rate Litter bag % mass 40 60 80 Medium Low Medium Low Medium
mass loss loss over
200 days
Litter mass Litter mass t DW ha–1 6 8 12 High Low Medium Low Medium
Invertebrate density Ground number 500 1000 2000 High Low Low– Medium High
invertebrate m–2 medium
density
Soil fertility C:N ratio n/a 10 15 20 Medium– Low– Medium Low– Medium
high2 medium
Bird density Bird % nests 20 40 60 Low3 Low Low High High
nesting that edge
success young
__________________________________________________________________________________________________________________________________________________________________
1Palatable species include Asplenium bulbiferum, Coprosma grandifolia, Cyathea medullaris, Geniostoma rupestre, Schefera digitata.
2See Sparling & Schipper (2002).
3See Innes et al. (2010a).
sapling density is unaffected by pest control, invertebrate
density is minimally affected by fencing).
Other studies in the New Zealand plant ecology literature
conrm this general picture of partial recovery in system
attributes following removal of the agents of perturbation.
Paired browser-exclusion plot studies have shown recovery
of palatable plant species (Smale et al. 1995; Husheer et al.
2005), though the effects have not been consistent, leading
to a number of hypotheses for non-recovery (Coomes et al.
2003). While we know of no published studies of the temporal
sequence of recovery of New Zealand forest fragments
following the alleviation of disturbance, two studies of forest
fragment vegetation have used a space-for-time substitution
approach to study recovery after the exclusion of domestic
livestock. One study was in lowland kahikatea (Dacrycarpus
dacrydioides) fragments (Smale et al. 2005) and the other
in lowland tawa fragments (Dodd & Power 2007). In these
studies, several fragments with differing periods of time since
fencing were assessed for vegetation and soil characteristics.
This approach implicitly follows an engineering resilience
paradigm, i.e. the equilibrium domain is assumed to be a
forest where the browsing disturbance was not present. The
(spatially simulated) changes in structure and composition of
the fragments over time frames of 10–20 years since grazing
exclusion in Smale et al. (2005) and Dodd & Power (2007)
included increases in sapling regeneration / understorey cover
and decreases in soil P fertility, consistent with a view of
the soil and vegetation component of forest fragment plant
communities being resilient to livestock browsing.
In the only nest survival study undertaken in New Zealand
fragments so far, Boulton et al. (2008) found that robin (Petroica
longipes) nest survival ‘marginally decreased’ with fragment
size. While nest survival is generally poor in both small and large
forests, it can be increased methodically in either with predator
control (e.g. Innes et al. 1999, 2004; studies summarised in
Innes et al. 2010a). Rare species may have to be restored to
fragments by translocation. The short-term absence of these
species from fragments despite pest control could be viewed as
an example of a lack of ecological resilience, but other species
have recently been demonstrated to establish new populations
in reserves independent of translocations (Miskelly et al. 2005),
which may then be evidence for engineering resilience on a
longer timescale, and requiring a larger spatial scale of pest
management (e.g. Basse & McLennan 2003) focused on many
rather than single fragments.
The experimental data also revealed unexpected
interactions between components of the model. Ship rats were
signicantly more abundant in fenced than in grazed fragments,
probably due to the higher biomass of vegetation, fruits and
seeds, and litter invertebrates. In this sense, ship rats simply
replace native birds in Fig. 2 in terms of responding to the
abundance of owers, fruit and seed and supplying a balancing
loop that reduces ower, fruit and seed abundance. However,
ship rats cannot replace native birds as effective agents of

New Zealand Journal of Ecology, Vol. 35, No. 1, 2011
ower pollination and seed and fruit dispersal (Williams et al.
2000; Kelly et al. 2006).
Overall, with respect to our objectives, we suggest that the
data present a picture of generally high ecological resilience
(no evidence of a lack of recovery over decadal timescales)
but variable degrees of engineering resilience for different
components of the forest fragment ecosystem (differing
recovery rates). Specically, the vegetation and invertebrate
components show high engineering resilience, while the soil
and bird components show lower engineering resilience. Bird
fauna show the potential for a lack of ecological resilience
(no recovery and changes in system controls), although rm
conclusions are obscured by lack of data. Thus the engineering
resilience paradigm appears to be an adequate model to inform
restoration. However, we emphasise that this assessment is
restricted to fragments in mesic environments, to landscapes
with good opportunities for immigration (i.e. fragmented but
not relictual), and to fragments impacted by livestock and pest
mammals over timescales of multiple decades.
Developing a management framework
One of the key questions that we sought to address from this
work is ‘What are the implications of our systems understanding
for the future management of forest fragments in New Zealand?’
Fundamentally, management works through a process of
developing goals and objectives; developing supporting
indicators and assessing the current state of a system relative
to management goals; applying management actions that have
a reasonable expectation of making progress toward the goals;
and monitoring outcomes using the same set of indicators. Our
results can inform all of these stages to varying degrees:
Goals
A key issue is that of setting appropriate goals for managing
forest fragments, and perhaps the most important message from
the study reported here is that the use of large ungrazed and
pest-controlled forest systems as a benchmark for restoration
is not entirely appropriate, given (a) that for some variables,
long-term fencing and pest control have not led to conditions
similar to ‘reference forests’; (b) known effects of area and
fragmentation on species richness (Hobbs & Saunders 1994;
Lomolino 2000); and (c) tentative evidence of the inability
of some species to recolonise areas after disturbance release.
Given that the prevailing agricultural land-use matrix over
much of New Zealand is unlikely to change in the foreseeable
future, many of the drivers of disturbance will remain present
in the current landscape context indenitely (Norton 2009).
The well-recognised effects of this (i.e. permanent loss of
area-sensitive species, reduction in potential species richness,
and loss of dispersal mechanisms) imply a need for rethinking
of the goals of restoration (Hobbs & Harris 2001). At the very
least we must seek goals that reect reasonable targets for
fragmented ecosystems. Such goals may discard a restoration
paradigm in favour of a reconstruction paradigm that focuses
on ecosystem goods and services rather than biotic history
(Jackson & Hobbs 2009).
Indicators
The development of a system dynamics model and the process
of rening it to depict the key variables (pools and/or processes)
highlights the most useful parameters for assessment and
monitoring of forest fragment condition. Measurement of
canopy cover, plant biomass, plant reproduction (ower/
fruit/seed and saplings), understorey cover, understorey light,
litter mass, invertebrate density, plant diversity, litter quality,
mineralisation rate, soil fertility, and bird populations within a
forest fragment should give a clear indication of the structural
and functional integrity of the ecosystem. However, even this
limited set of indicators is likely to be too onerous for most
land managers to measure or monitor, suggesting that there
will be a strong need for further renement of key indicators.
The system dynamics model would suggest that a minimal
set of indicators comprises the parameters that capture the
operation of the major causal loops, namely: understorey light
(the balancing shading loop); sapling numbers (the reinforcing
regeneration loop); litter mass (the reinforcing decomposition
loop); and bird numbers (the balancing herbivory and predation
loops) (Fig. 2). All have specic methods of visual or aural
assessment that are relatively inexpensive and require minimal
training.
Actions
Assuming that an assessment of the condition of a forest
fragment identies a mismatch between current and desired
state, the manager will probably want to know the extent to
which s/he can rely on natural processes versus the need for
active intervention to achieve the desired result. Our results
indicate that if the manager wishes to restore plant species
diversity and understorey regeneration, we can condently
advise that putting up a livestock-proof fence will achieve
this goal, as the understorey ora will recover without the
need for supplementary planting. This conclusion is based
on our assessment of the relative resilience of the indigenous
understorey ora to livestock browsing, but may not be true
for canopy structure and composition. If the manager wishes
to restore macroinvertebrate fauna, we may advise that stock
fencing or elimination of mammalian omnivores alone will
have limited value, and both are required (Didham et al. 2009).
This is based on our assessment of the relative resilience of
the indigenous invertebrate fauna to the combination of stock
browsing and mammalian predation. If the manager wishes to
restore nesting native bird populations, we may advise that even
complete control of mammalian fauna will not be sufcient
alone, and some form of reintroduction of bird species will
be required. This is based on our assessment of the relative
lack of resilience of the indigenous avifauna to mammalian
predation/competition. The latter example represents a key
dividing line between management by perturbation release
(which relies on the engineering resilience of the ecosystem to
restore structure and function) and management by ecosystem
reconstruction (which accounts for the lack of ecological
resilience and seeks to rebuild the system as well as protect
it from further perturbation).
Monitoring
Given that the manager will also be interested in the likely time
frame over which s/he can expect to observe an improvement
in fragment condition following fencing and pest control, the
results can suggest appropriate monitoring intervals for the
key indicators outlined above. Based on the limited studies
available to date, measurable changes should be apparent after
0–5 years for litter mass, 5–10 years for plant diversity and
bird numbers, 5–15 years for sapling numbers and >20 years
for understorey light levels.

Dodd et al.: Forest fragment resilience
Figure 4. A conceptual state-space for forest fragment ecosystems,
based on indigenous functional dominance and indigenous
community occupancy, showing the major recognisable states
of extant forest (light grey), forest fragments (mid-grey) and
non-forest or forest edge (dark grey); observable trajectories of
degradation (solid lines) and recovery (short-dash lines); and
potential thresholds associated with fragmentation and collapse
(long-dash lines).
A conceptual management model
Finally, we sought to develop a conceptual management model
at the fragment scale that could incorporate the concepts
of resilience discussed in this paper with the more familiar
concepts of forest condition and ecosystem integrity. Our
conclusion about the adequacy of an engineering resilience
paradigm for our system of interest suggested that a linear
framework with few indicator variables would be adequate, so
long as it could incorporate threshold dynamics where evidence
for them emerged. Lee et al. (2005) have suggested three
indices of integrity for cross-scale biodiversity inventory and
monitoring by the New Zealand Department of Conservation:
(1) ‘Indigenous dominance’ (the level of indigenous species’
inuence on the structure and function of ecosystems); (2)
‘Species occupancy’ (the extent to which the indigenous
species capable of living in an ecosystem are present); and (3)
‘Environmental representation’ (the distribution of indigenous
ecosystems across environmental gradients).
We have drawn on the indices of Lee et al. (2005) to
develop a conceptual management model for forest fragments,
by rst creating a semi-quantitative empirical example (Fig. 3)
and then formulating a generalised model (Fig. 4). The model
incorporates a two-dimensional ‘state-space’, akin to the
restoration scenarios depicted in Suding et al. (2004, Fig. 2) to
reect the condition and dynamics of forest fragments in the
context of associated ecosystems. At the scale of individual
forest fragments that we are interested in here, environmental
representation is less relevant to fragment management, so we
have focused on the rst two indices of Lee et al. (2005), with
some modications outlined below.
From the terms described by Lee et al. (2005), we have
modied the terminology of the second index to ‘community
occupancy’ to distinguish it from other denitions of species
occupancy that focus on spatial abundance (e.g. MacKenzie
Figure 3. Semi-quantitative dynamics of plant communities in
forest fragment ecosystems in response to livestock grazing and
mammal pest invasion (solid arrows), and the release from these
press disturbances at two stages of degradation (dashed arrows).
Data based on Smale et al. (2008) and subsequent unpublished
data from that study site.
et al. 2005). In addition, it is likely that the structural component
of an indigenous dominance index (in particular species
composition) will be highly correlated with an occupancy
index, so we have focused the x-axis of our two-dimensional
state-space on the combination of these two components (which
we call indigenous community occupancy), and we focus the
y-axis of the state-space solely on the functional component
of indigenous dominance of ecosystem processes (which we
call indigenous functional dominance).
Semi-quantitative example
Figure 3 populates the state-space using data from the Smale
et al. (2008) study, along with more recent data from the same
Whatawhata site in the western Waikato hill country (MBD,
unpubl. data). These data provide the basis for an analysis using
vascular plant data from forest ecosystems under ve different
management regimes: the interior of a large reference forest,
the interior of a grazed and pest-infested forest fragment, the
edge of a grazed and pest-infested forest fragment, the interior
of a small fragment fenced and pest controlled for 7 years,
and the edge of a grazed and pest-infested forest fragment
fenced for c. 40 years. The use of vegetation data only, rather
than a more complete analysis including faunal data, reects
the relative availability of this information. The compilation
of a list of extant indigenous species from which to estimate
indigenous community occupancy is simple, but determining
the denominator for calculating percent occupancy relative to
the Lee et al. (2005) criteria that ‘indigenous species capable
of living in an ecosystem are present’ is less straightforward.
In this case we have used the mean number of indigenous
species (85) identied in an 800-m2 area of reference forest
from g. 2 of Smale et al. (2008), on the basis that this
represents the approximate spatial scale of the measurements
from the grazed fragments in that study. The general lack of

New Zealand Journal of Ecology, Vol. 35, No. 1, 2011
baseline information on historical species distributions in
New Zealand represents a major limitation to the development
of management goals and indicators. For canopy tree species,
a potential natural vegetation layer has been constructed for
vegetation composition in New Zealand prior to the arrival
of humans (Leathwick 2001), but similar reference points
for oral or faunal composition are not available for the vast
majority of taxa.
In generating the data for the y-axis (indigenous functional
dominance) of Fig. 3, we note that Lee et al (2005, p. 107)
state ‘The cornerstone of continued indigenous dominance is
self-regeneration…’. Therefore we chose to plot functional
dominance in terms of the juvenile abundance of indigenous
species, using a three-stage scoring system (nil = 0; low = 1;
high = 2). We have assumed that all non-woody species with
lifespans less than the period of livestock grazing or mammalian
pest disturbance must have successfully regenerated under these
conditions (i.e. score = 2) and based the score for longer-lived
woody species on sapling abundance. The self-regeneration
index is the sum of the juvenile abundance scores divided
by the number of indigenous species present as adults (×2),
converted to a percentage (×100). This simplistic approach
has some obvious weaknesses (e.g. how to account for the
observation that even when all juveniles are indigenous
species the juvenile population may not encompass the full
range of adult species occupying the site, due to stochastic
reproduction or dispersal failure; and how to account for
juveniles dispersing into plots where no adults are present)
but further development of this index is beyond the scope of
the illustrative discussion in this paper.
Based on this semi-quantitative worked example (Fig. 3),
there is a clear distinction in both indigenous community
occupancy and indigenous functional dominance between the
reference forest and the interior of a degraded forest fragment.
There appear to be two alternative trajectories that the forest
fragment ecosystems at this site have followed, corresponding
to either improvement in indigenous species occupancy
and functional dominance associated with the recovery of
a forest fragment released from livestock grazing and pest
mammal impacts, or further degradation to very low levels of
indigenous community occupancy and functional dominance
that might be reected in current state observed in fragment-
edge environments, which have few remnant canopy trees and
consist primarily of tree ferns and adventive weeds. The data
from the long-term fenced remnant edge indicate substantial
recovery in both occupancy and dominance from even this
highly degraded state.
The semi-quantitative model of plant community dynamics
in Waikato forest fragments (Fig. 3) can be generalised to other
components of the structure and functioning of forest fragment
ecosystems throughout lowland areas of New Zealand (Fig. 4).
In this general conceptual model, we note that the reference
forests to which the condition of forest fragments are often
compared (large tracts of indigenous forests in which some
degree of conservation management has occurred) do not
occupy the extreme top right ‘pristine’ space in the diagram
(Fig. 4), due to species extinctions, associated loss in ecosystem
function, and the inuence of exotic weeds and pests on
ecosystem processes. We also note that non-forest ecosystems
(e.g. pastoral land) do not occupy the extreme bottom left
‘completely degraded’ space in the diagram (Fig. 4), since these
habitats commonly contain some indigenous species that can
have a signicant contribution to ecosystem processes (e.g.
grass grub beetle Costelytra zealandica, and meadow ricegrass
Microlaena stipoides). Nevertheless, it is clear that the reference
forests would be regarded as having relatively high ecological
integrity, while pasture would be regarded as having relatively
low ecological integrity, from the point of view of retaining
the ‘natural character’ of indigenous ecosystems. Between
these two extremes of indigenous occupancy and functional
dominance, there are likely to be a range of degradation and
recovery pathways through which ecosystems might be forced,
but all the degraded forest fragments considered here appear
to occupy the central space in the diagram (Fig. 4).
The process of human-mediated restoration generally
involves two approaches: (1) disturbance release (e.g. fencing,
pest control), which focuses on restoring the indigenous
functional dominance and which by itself implicitly assumes
spontaneous immigration and improvement of community
occupancy; and (2) translocation (e.g. tree planting,
reintroducing birds), which focuses on restoring community
occupancy and which by itself implicitly assumes the presence
of these species will improve indigenous functional dominance.
Thus, recovery of ecosystem integrity is represented by an
upward and/or rightward shift in Fig. 4. The rate or extent to
which this occurs can be considered a measure of engineering
resilience.
Summary
It is of particular interest whether forest fragment recovery
processes are inhibited by the existence of thresholds, since this
would have a bearing on the likelihood of restoration failure
following management intervention. A lack of evidence for
thresholds and associated positive feedback loops operating
within a system would suggest that an engineering resilience
paradigm might be adequate for management purposes. Under
this model, it would be reasonable to expect that mitigation of
adverse drivers would lead to an upward and rightward shift
(i.e. successful recovery) of the forest fragment system. The
results of the studies reported here indicate that this is the
case. We did not nd strong evidence for a lack of ecological
resilience in the empirical data (Table 1, Fig. 3a). However,
the system dynamics model did suggest that low ecological
resilience may be an issue in respect of the avifauna and
mammalian pest effects. In addition, the scope of our study
was conned to already fragmented ecosystems, and further
exploration of the specic effects of the fragmentation process
might provide evidence of a lack of ecosystem resilience in
lowland forest fragments to historical human impacts.
Acknowledgements
We express our thanks to the other members of the ‘Forest
Fragment Resilience’ team – the late Greg Arnold, Bruce
Clarkson, Lisa Denmead and Chris Floyd – for stimulating
discussions. Our ideas and approach benetted from discussions
with Richard Hobbs, Helen Allison and Roger Partt. The
comments of the two journal reviewers helped us improve the
initial manuscript. This study was funded by the New Zealand
Foundation for Research, Science and Technology through
PGSF Contract UOWX0609 to the University of Waikato.

Dodd et al.: Forest fragment resilience
References
Basse B, McLennan JA 2003. Protected areas for kiwi in
mainland forests of New Zealand: how large should they
be? New Zealand Journal of Ecology 27: 95–105.
Batcheler CL, Cowan PE 1988. Review of the status of the
possum in New Zealand. Contract report commissioned
by the Technical Advisory Committee (Animal Pests),
Ministry of Agriculture & Fisheries, Wellington. 129 p.
Bengtsson J 2002. Disturbance and resilience in soil animal
communities. European Journal of Soil Biology 38:
119–125.
Boulton RL, Richard Y, Armstrong DA 2008. Inuence of
food availability, predator density and forest fragmentation
on nest survival of New Zealand robins. Biological
Conservation 141: 580–589.
Burns BR, Floyd CG, Smale MC, Arnold GC 2010. Effects
of forest fragment management on vegetation condition
and maintenance of canopy composition in a New Zealand
pastoral landscape. Austral Ecology: in press. Published
online: DOI: 10.1111/j.1442-9993.2010.02130.x.
Carpenter S, Walker B, Anderies JM, Abel N 2001. From
metaphor to measurement: resilience of what to what?
Ecosystems 4: 765–781.
Clout MN, Hay JR 1989. The importance of birds as browsers,
pollinators and seed dispersers in New Zealand forests.
New Zealand Journal of Ecology (Suppl.) 12: 27–33.
Coomes DA, Allen RB, Forsyth DM, Lee WG 2003. Factors
preventing the recovery of New Zealand forests following
control if invasive deer. Conservation Biology 17:
450–459.
Craig J, Anderson S, Clout M, Creese B, Mitchell N, Ogden
J, Roberts M, Ussher G 2000. Conservation issues in
New Zealand. Annual Review of Ecology and Systematics
31: 61–78.
Crisp PN, Dickinson KJM, Gibbs GW 1998. Does native
invertebrate diversity reect native plant diversity? A case
study from New Zealand and implications for conservation.
Biological Conservation 83: 209–220.
Didham RK, Barker GM, Costall JA, Denmead LH, Floyd
CG, Watts, CH 2009. The interactive effects of livestock
exclusion and mammalian pest control on the restoration
of invertebrate communities in small forest fragments.
New Zealand Journal of Zoology 36: 135–163.
Dodd MB Power IL 2007. Recovery of Tawa-dominated
forest fragments in the Rotorua Basin, New Zealand, after
cessation of livestock grazing. Ecological Management
and Restoration 8: 208–217.
Dodd MB, Power IL, Porcile V, Upsdell M 2006. A measurable
effect of feral grazing on pasture accumulation rate.
Proceedings of the New Zealand Grassland Association
68: 339–342.
Esler AE 1978. Botany of the Manawatu District, New Zealand.
DSIR Information Series 127. Wellington, Government
Printer. 206 p.
Ewers RM, Didham RK 2004. A first assessment of
New Zealand beetles restricted to large forest areas. The
Weta 27: 33–40.
Ewers RM, Didham RK 2008. Pervasive impact of large-scale
edge effects on a beetle community. Proceedings of the
National Academy of Sciences, USA 105: 5426–5429.
Ewers RM, Kliskey AD, Walker S, Rutledge D, Harding JS,
Didham RK 2006. Past and future trajectories of forest
loss in New Zealand. Biological Conservation 133:
312–325.
Ewers RM, Thorpe S, Didham RK 2007. Synergistic
interactions between edge and area effects in a heavily
fragmented landscape. Ecology 88: 96–106.
Gunderson LH 2000. Ecological resilience – in theory and
application. Annual Review of Ecology and Systematics
31: 425–439.
Harris RJ, Burns BR 2000. Beetle assemblages of kahikatea
fragments in a pasture-dominated landscape. New Zealand
Journal of Ecology 24: 57–67.
Harvie AE 1973. Diet of the opossum (Trichosurus vulpecula
Kerr) on farmland northeast of Waverley, New Zealand.
Proceedings of the New Zealand Ecological Society 20:
48–52.
Hobbs RJ, Harris JA 2001. Restoration ecology: repairing the
earth’s ecosystems in the new millennium. Restoration
Ecology 9: 239–246
Hobbs RJ, Saunders DA 1994. Effects of landscape
fragmentation in agricultural areas. In: Moritz C, Kikkawa
J eds Conservation biology in Australia and Oceania.
Chipping Norton, NSW, Surrey Beatty. Pp. 77–95.
Holling CS 1996. Engineering resilience versus ecological
resilience. In: Schulze P ed. Engineering within ecological
constraints. Washington, DC, National Academy Press.
Pp. 31–44.
Husheer SW, Hansen QW, Urlich SC 2005. Effects of red deer on
tree regeneration and growth in Aorangi forest, Wairarapa.
New Zealand Journal of Ecology 29: 271–277.
Innes J 2009. Forest fragments need fencing and pest control.
Open Space 75, March 2009. Wellington, QEII National
Trust.
Innes J, Hay R, Flux I, Bradeld P, Speed H, Jansen P 1999.
Successful recovery of North Island kokako Callaeas
cinerea wilsoni populations, by adaptive management.
Biological Conservation 87: 201–214.
Innes J, Nugent G, Prime K, Spurr EB 2004. Responses of
kukupa (Hemiphaga novaeseelandiae) and other birds to
mammal pest control at Motatau, Northland. New Zealand
Journal of Ecology 28: 73–81.
Innes J, Kelly D, Overton JMcC, Gillies C 2010a. Predation
and other factors currently limiting New Zealand forest
birds. New Zealand Journal of Ecology 34: 86–114.
Innes J, King CM, Bridgman L, Fitzgerald N, Arnold, G, Cox
N 2010b. Effect of grazing on ship rat density in forest
fragments of lowland Waikato, New Zealand. New Zealand
Journal of Ecology 34: 227–232.
Jackson ST, Hobbs RJ 2009. Ecological restoration in the light
of ecological history. Science 325(5940): 567–569.
Jane G 1983. The impact of introduced herbivores on lowland
forests in the North Island. In: Thompson K, Hodder
APH, Edmonds AS eds Lowland forests in New Zealand.
Hamilton, University of Waikato. Pp. 135–152.
Kelly D, Robertson AW, Ladley JJ, Anderson SH, McKenzie
RJ 2006. Relative (un)importance of introduced animals
as pollinators and dispersers of native plants. In: Allen
RB, Lee WG eds Biological invasions in New Zealand.
Ecological Studies 186. Berlin & Heidelberg, Springer-
Verlag. Pp. 227–245.
King CM 2005. The handbook of New Zealand mammals. 2nd
edn. Melbourne, Oxford University Press. 610 p.
Leathwick JR 2001. New Zealand's potential forest pattern as
predicted from current species-environment relationships.
New Zealand Journal of Botany 39: 447–464.
Leathwick JR, Rogers GM 1996. Modelling relationships

New Zealand Journal of Ecology, Vol. 35, No. 1, 2011
between environment and canopy composition in
secondary vegetation in central North Island, New Zealand.
New Zealand Journal of Ecology 20: 147–161.
Lee W, McGlone M, Wright E (comps) 2005. Biodiversity
inventory and monitoring. A review of national and
international systems and a proposed framework for
future biodiversity monitoring by the Department of
Conservation. Contract Report LC0405/122. Lincoln,
Landcare Research. 216 p.
Lomolino MV 2000. Ecology's most general, yet protean
pattern: The species-area relationship. Journal of
Biogeography 27: 17–26.
Ludwig D, Walker B, Holling CS 1996. Sustainability, stability,
and resilience. Conservation Ecology 1: 1– 27.
Maani KE, Cavana RY 2007. Systems thinking, system
dynamics: Managing change and complexity. 2nd edn.
Auckland, Prentice Hall.
MacKenzie DI, Nichols JD, Sutton N, Kawanishi K, Bailey
LL 2005. Improving inferences in population studies of
rare species that are detected imperfectly. Ecology 86:
1101–1113.
McIntyre S, Hobbs R 1999. A framework for conceptualizing
human effects on landscapes and its relevance to
management and research models. Conservation Biology
13: 1282–1292.
McQueen DR 1993. A review of the interaction between
naturalized woody plants and indigenous vegetation in
New Zealand. Tuatara 32: 32–56.
McQueen S, Lawrence B 2008. Diet of ship rats following a
mast event in beech (Nothofagus spp.) forest. New Zealand
Journal of Ecology 32: 214–218.
Meurk CD, Hall GMJ 2006. Options for enhancing forest
biodiversity across New Zealand’s managed landscapes
based on ecosystem modelling and spatial design.
New Zealand Journal of Ecology 30: 131–146.
Miskelly C, Empson R, Wright K 2005. Forest birds
recolonising Wellington. Notornis 52: 21–26.
Miskelly CM, Dowding JE, Elliott GP, Hitchmough RA,
Powlesland RG, Robertson HA, Sagar PM, Scoeld RP,
Taylor GA. 2008. Conservation status of New Zealand
birds, 2008. Notornis 55: 117–135.
Moles AT, Hodson DW, Webb CJ 2000. Seed size and shape
and persistence in the soil in the New Zealand ora. Oikos
89: 541–545.
Murphy E, Maddigan F, Edwards B, Clapperton K 2008. Diet
of stoats at Okarito Kiwi Sanctuary, South Westland,
New Zealand. New Zealand Journal of Ecology 32:
41–45.
Nicholls JL 1979. Ecological survey of New Zealand's
indigenous forests. Forest Service Mapping Series 6.
Sheet 4: Waikato. Forest Research Institute, New Zealand
Forest Service.
Norton DA 2009. Species invasions and the limits to restoration:
Learning from the New Zealand experience. Science
325(5940): 569–571.
Nugent G, Sweetapple P, Coleman J, Suisted P. 2000. Possum
feeding patterns : dietary tactics of a reluctant folivore. In:
Montague TL ed. The brushtail possum: biology, impact
and management of an introduced marsupial. Lincoln,
Manaaki Whenua Press. Pp. 10–19.
Orwin KH, Wardle DA 2004. New indices for quantifying
the resistance and resilience of soil biota to exogenous
disturbances. Soil Biology and Biochemistry 36: 1907–
1912.
Partridge TR 1989. Soil seed banks of secondary vegetation
on the Port Hills and Banks Peninsula, Canterbury,
New Zealand and their role in succession. New Zealand
Journal of Botany 27: 421–436.
Pastor J, Dewey B, Naiman RJ, McInnes PF, Cohen Y 1993.
Moose browsing and soil fertility in the boreal forests of
Isle Royale National Park. Ecology 74: 467–480.
Pimm SL 1984. The complexity and stability of ecosystems.
Nature 307: 321–326.
Scheffer M, Carpenter SR 2003. Catastrophic regime shifts
in ecosystems: linking theory to observation. Trends in
Ecology and Evolution 18: 648–656.
Sem G, Enright NJ 1996. The relationship between seed rain
and the soil seed bank in a temperate rainforest stand
near Auckland, New Zealand. New Zealand Journal of
Botany 34: 215–226.
Smale MC, Hall GMJ, Gardner RO 1995. Dynamics of
kanuka (Kunzea ericoides) forest on South Kaipara spit,
New Zealand, and the impact of fallow deer (Dama dama).
New Zealand Journal of Ecology 19: 131–141.
Smale MC, Ross CW, Arnold GC 2005. Vegetation recovery
in rural kahikatea (Dacrycarpus dacrydioides) forest
fragments in the Waikato region, New Zealand, following
retirement from grazing. New Zealand Journal of Ecology
29: 261–269.
Smale MC, Dodd MB, Burns BR, Power IL 2008. Long-term
impacts of grazing on indigenous forest remnants on North
Island hill country, New Zealand. New Zealand Journal
of Ecology 32: 57–66.
Sparling GP, Schipper LA 2002. Soil quality at a national
scale in New Zealand. Journal of Environmental Quality
31: 1848–1857.
Standish RJ, Sparrow AD, Williams PA, Hobbs RJ 2009. A
state-and-transition model for recovery of abandoned
farmland in New Zealand. In: Hobbs RJ, Suding KN eds
New models for ecosystem dynamics and restoration.
Island Press. Pp. 189–205.
Stevens H 2006. Native birds in forest remnants. Unpublished
PhD thesis, School of Geography and Environmental
Science, University of Auckland, Auckland, New Zealand.
200 p.
Stevenson BA 2004. Changes in phosphorus availability and
nutrient status of indigenous forest fragments in pastoral
New Zealand hill country. Plant and Soil 262: 317–325.
Suding KN, Hobbs RJ 2009. Models for ecosystem dynamics
as frameworks for restoration ecology. In: Hobbs RJ,
Suding KN eds New models for ecosystem dynamics and
restoration. Washington, DC, Island Press. Pp. 3–21.
Suding KN, Gross KL, Houseman GR 2004. Alternative states
and positive feedbacks in restoration ecology. Trends in
Ecology & Evolution 19: 46–53.
Sullivan JJ, Williams PA, Timmins SM 2007. Secondary forest
succession differs through naturalised gorse and native
kānuka near Wellington and Nelson. New Zealand Journal
of Ecology 31: 22–38.
Tilman D, May RM, Lehman CL, Nowak MA 1994. Habitat
destruction and the extinction debt. Nature 371: 65–66.
Timmins SM 2002. Impact of cattle on conservation land
licensed for grazing in South Westland, New Zealand.
New Zealand Journal of Ecology 26:107–120.
Walker BH, Meyers JA 2004. Thresholds in ecological and
social–ecological systems: a developing database. Ecology
and Society 9(2):3.
Walker BH, Langridge JL, McFarlane F 1997. Resilience

Dodd et al.: Forest fragment resilience
of an Australian savanna grassland to selective and
nonselective perturbations. Australian Journal of Ecology
22: 125–135.
Wardle DA, Barker GM, Yeates GW, Bonner KI, Ghani A 2001.
Introduced browsing mammals in New Zealand natural
forest: aboveground and belowground consequences.
Ecological Monographs 71: 587–614.
West CJ 1995. Sustainability of Beilschmiedia tawa-dominated
forest in New Zealand: Population predictions based on
transition matrix model analysis. Australian Journal of
Botany 43: 51–71.
Editorial Board member: Jacqueline Beggs
Received 23 October 2009; accepted 4 August 2010
Westoby M, Walker B, Noy-Meir I 1989. Opportunistic
management of rangelands not at equilibrium. Journal
of Range Management 42: 266–274.
Whaley PT, Clarkson BD, Smale MC 1997. Claudelands
Bush: ecology of an urban kahikatea (Dacrycarpus
dacrydioides) forest remnant in Hamilton, New Zealand.
Tane 36: 131–155.
Williams PA, Karl BJ, Bannister P, Lee WG 2000. Small
mammals as potential seed dispersers in New Zealand.
Austral Ecology 25: 523–532.