INTRODUCTION
The extinction rates over the last century has been estimated to be
higher than the historical background rate (e.g. up to 100 times higher
for vertebrates), with human activity identified as the predominant
driver of this “sixth mass extinction” (Ceballos et al. 2015).
In addition to the positive effects of biodiversity on human wellbeing
and culture (Dereniowska & Meinard 2021), such declines undermine the
stability and resilience of ecological systems on which humanity relies
for food, fresh water, and clean air (Maron et al. 2017). At the
root of human induced extinctions are a suite of stressors - including
habitat loss, pollution, overharvesting, and climatic change (Tilmanet al. 2017) - which can drive declines and erode a population’s
ability to recover in the face of disturbances, increasing the
probability of rapid collapses in the abundance of the populations (van
de Leemput et al. 2018). Indeed, anthropogenic pressure often
creates scenarios where negative biotic and abiotic stressors mutually
reinforce one another and affect, through a domino effect, multiple
facets of populations, driving it precipitously to extinction – the so
called extinction vortex (Fagan & Holmes 2005; Williams et al.2021). Consequently, we are at a critical point for ecosystem management
where, to preserve biodiversity and ecosystem services, we need to
reliably detect not only what systems are being most impacted by
anthropogenic stressors, but which are most at risk of collapse
(Clements & Ozgul 2018).
This need has driven the development of numerous predictive methods that
aim to forecast the risk of population collapse, ranging from classical
Population Viability Analyses (PVA, Shaffer 1991) to more recently
developed Early Warning Signals (EWSs, Clements & Ozgul 2018). However,
the difficultly of surveying wild populations, together with economic
limitations (Gardner et al. 2008), often results in noisy and
short abundance time series data which can detrimentally affect the
accuracy of predictive frameworks such as the EWSs (Clements et al.
2015). Moreover, this approach has neglected other potentially powerful
diagnostic features which theory and evidence suggest should be impacted
by increasing stress, and thus could act as additional indicators of
increasing extinction risk. Indeed, the decline of a population to
extinction is a manifestation of a host of changes to the structure and
dynamics of that population. Such changes occur first at the individual
level (e.g. decline in the body condition of an individual) which, when
a high enough proportion of the population exhibit similar changes of
sufficiently high magnitude, affects the dynamics of that population.
Therefore, additional signatures of approaching collapse could include
changes in the behaviour of individuals (Berger-Tal et al. 2011),
and value of morphological traits (Baruah et al. 2019), alongside
traditional abundance-based measures of extinction risk (Clements &
Ozgul 2018). At the individual level, many behaviours and
morphological traits have a range
of plasticity to maintain fitness in the presence of environmental
variability and stress (Fox et al. 2019); however, if conditions
continue to change, this adaptive plasticity may not be enough to
maintain reproductive capacity and survival of individuals (i.e.
fitness), eventually affecting population abundance.
Thus, the effect of increasing stress on a population propagates from
the individual level to the population level via a successive series of
reactions (or “signals of stress”) through time. Such individual level
responses necessarily take place (and are observable) over smaller time
scales compared to population level signals; an individuals’ behaviour
or morphology can change during their lifespan, while the effect of
stress at the population dynamics (e.g. decreases in abundance) will
happen after one or more generations. Incorporating this
individual-to-population concept offers not only the opportunity to use
individual responses to stress as early indicators of change in
population conditions, but to measure the impacts of this stress on
multiple-dimensions simultaneously. Such an approach expands on recent
work in the field of EWSs, where abundance based EWSs and shifts in the
mean body size of the population are considered concurrently, leading to
an increase of the overall predictive power (Clements & Ozgul 2016;
Clements et al. 2017). These results suggested that integrating
multi-dimensional data to predict population collapse has significant
merit, whereby signals of stress from a range of ecological disciplines
may be combined to increase the reliability of warning signals, decrease
the length of the time series required to generate such signals, and
increase the time prior to collapse in which signals are detectable.
Here we develop a conceptual framework – the “timeline to collapse”
– supported by case studies, where we: i) describe how increasing
environmental stress shapes different features of individuals across
time, ii) state the temporal sequence of observable stress signals from
individual to population level along the path to extinction, iii) show
how this “timeline” of responses provides a signature to corroborate
whether a population is at risk of collapse and iv) outline methods to
gather the data needed to implement such an approach. Additionally, we
will explore how the timeline provide information on individuals and
population stress buffering capacities. The feasibility of this
framework relies on organisms showing measurable behavioural plasticity,
and thus within this review we will primarily consider animals as
examples and case studies, but where applicable we will highlight
concepts that are relevant to non-animal species. Henceforth, we will
refer to “environmental stress” - or more in general to “stress” -
as the presence of biotic and/or abiotic factors (e.g. resource
scarcity, pollution, invasive species etc.) that effects a population in
a negative way. In the following sections we will consider responses to
stress observable over the short term (rapid
changes;<<1 generation), medium term (intermediate
speed changes; ≤1 generation) and long term (slow changes;
>1 generation).
RAPID CHANGES
Behavioural changes are amongst the most rapid changes that individuals
can perform to cope with sub-optimal conditions (Greggor et al.2016). The potential range of behaviours individuals can present in the
face of stress is a result of evolved mechanisms that shape strategies
to maximize fitness, intrinsic plasticity and the past experiences of
the individual (Tuomainen & Candolin 2011). Broadly, such behaviours
comprehend those related to movement and habitat use, foraging
activities, and reproductive and social behaviours (Berger-Tal et
al. 2011). That different categories of behaviour can be modified by
stress is well documented, with many studies showing variation in e.g.
foraging activity and dispersal of individuals in response to declining
resource availability (Couvillon et al. 2014; Fayet et al.2021), climatic change (Hamilton et al. 2015; Holt & Jørgensen
2015; Gauzens et al. 2021), and invasive species (Lenda et
al. 2013). Indeed, changes in the movement patterns, speed, and
position of individuals in their environment can be amongst the first
signals to manifest in response to increasing environmental stress, as
individuals seek to minimise the impacts of, say, declining food
availability by moving to new foraging areas (e.g., increased foraging
effort, Figure 1A) or by reducing activity levels (depressing
metabolism, Trites & Donnelly 2003). Such changes constitute some of
the most easily observed and measurable behavioural signals of
increasing stress, as they can often be captured remotely through e.g.
GPS tracking or remote camera monitoring, techniques which bridge taxa
(vertebrate and invertebrates, Hertel et al. 2019, Tini et
al. 2018) and realms (marine and terrestrial , Shimada et al.2021).
In addition to movement patterns, individuals may react to stress by
altering rates of intra-and -interspecific interactions, with effects
observed also in social and communicative behaviours (Kunc & Schmidt
2021). For instance, resource scarcity may lead individuals to prefer
energy allocation in essential activities (e.g. foraging), decreasing
actions not linked to strict survival such as the engagement in
territorial defense (e.g. in coral fishes,
Keith et al. 2018, Figure
1B). Similarly, acoustically active insects and amphibians may change
the acoustic properties of the mating signals in response to temperature
stress (e.g. crickets call speed increase at high temperature, Singhet al. 2020). Moreover, human disturbance (e.g. presence of
boats) can induce reductions in whistles and echolocation click rates of
social cetaceans (Pellegrini et al. 2021).
Increases or decreases in behavioural metrics (foraging distance,
prevalence of an interaction type, duration in time of given actions
etc.) will vary depending on species environmental tolerance, trophic
level (e.g. prey vs predator) and stressor type. Whilst a lack of
resources may trigger increases in movement, the arrival of an invasive
predator in an ecosystem may induce a prey species to reduce movement
(to reduce encounter rates) or to shift microhabitat use toward a more
shelter-oriented strategy (i.e. less time spent in open areas, McMahan
& Grabowski 2019). In addition to these directional changes,
environmental stress may increase the variance observed while monitoring
behavioural metrics, e.g. poor environmental conditions enhanced the
variability of foraging trip duration in young albatrosses (Patricket al. 2021).
Previous experience may also play a critical role in determining an
individual’s response to stress. Individuals that have previously faced
similar situations may cope better with a novel stress if cues share
similar characteristics to those already experienced. For instance,
compared to naïve individuals, fishes with previous experience of
predation events showed stronger antipredator behaviours (e.g.
decreasing swimming activity) when they were represented with the
chemical cues of the predator (Vilhunen et al. 2005). Likewise,
the evolutionary history of a population can shape an individuals’
capacity to react to environmental pressure. A lizard prey species will
likely recognize a new predatory snake introduced in its habitat as
dangerous and perform antipredatory behaviours if the lizard’s
population have evolved with other snake species, especially if the
predators share similar features (shape, chemical cues etc.) with the
introduced predator (Ortega et al. 2017). On the other hand,
lizards that have never seen predatory snake in their evolutionary past
(e.g. due to geographical isolation, (Durand et al. 2012)) are
less likely to recognize an alien snake as dangerous, and thus may
suffer heavy predation (i.e. lack of antipredatory response). (Sih 2013;
McMahan & Grabowski 2019). Consequently, a population’s ecological and
biogeographical history must be considered when looking for such
behavioural signals of stress.
Such changes in individual’s behaviour can occur over short
(<< 1 generation) timescales, as such shifts are
driven by physiological needs and immediate adaptative reactions which
take place rapidly. For instance, the micro-habitat use shift by an
insect prey can happen overnight after a predator arrival (Pierce 1988);
similarly, an increase in movement of individuals due to food scarcity
can be triggered after months, days, or hours depending on the species
life span and metabolism speed. Regardless, such fast behavioural
changes represent an individuals’ primary stress buffering response, and
consequently will manifest as the first of the suite of detectable
warning signals.
INTERMEDIATE SPEED CHANGES
If rapid behavioural plasticity is not enough to mitigate the effects of
increasing stress, individuals may respond to maximize survival and
reproductive output through changes in morphological traits related to
fitness (Fox et al. 2019). Such changes can include metabolic
adjustments (e.g. reductions in body mass, decreases in growth rate) as
well as antipredatory morphological trait expression, and their
plasticity shapes an individual’s capacity to respond to rapid
environmental change (Fox et al. 2019), thus governing the
vulnerability of populations to extinction (Olden et al. 2007).
Environmental stress substantially affects morphological trait
distributions, both prior to or concurrent with changes in the
demography of the population (Pigeon et al. 2017; Baruah et
al. 2019). For instance, the reduction in body size of populations due
to sub-optimal food consumption is a general response to resources
scarcity (Trites & Donnelly 2003). Reductions in body size are also
directly and indirectly induced by climatic change and habitat
fragmentation, with such shifts being observed across numerous taxa
(Lomolino & Perault 2007; Gardner et al. 2011; Sheridan &
Bickford 2011; Stirling & Derocher 2012; Thoral et al. 2021,
Figure 2A). Indeed, body size is a key trait that directly affects
thermoregulation dynamics and rates of energy and mass intake and
utilization (Gardner et al. 2011), and has recently been
suggested as a possible measure of population stability (Clements &
Ozgul 2016). For example, changes in body size of diatoms algae preceded
a regime shift in a lake ecosystem (Spanbauer et al. 2016), and
experimental populations exhibit the same pattern, showing that – when
resources decrease – declines in average body size precede declines in
population size, and hence could be indicative of a future population
collapse (Baruah et al. 2019).
In situations where measuring body size changes is inappropriate, change
in individual growth rates can be used as an even more accurate stress
signal since growth rate will respond instantaneously to physiological
adjustments made by the individual in response to stress. For example,
Bjorndal et al. 2017 reported a decrease in growth rate of
individuals of three sea turtle species in response to climatic
stressors and anthropogenic degradation of their foraging areas.
Similarly, environmental stress can lead to a decrease in defensive
morphological traits: e.g. light stress in pregnant individuals of a
freshwater cladoceran crustacean induced the reduction of antipredator
spines dimensions in their offspring, with a consequent enhancement of
the predation risk for newborn individuals (Eshun-Wilson et al.2020).
Even if such reductions in the size of morphological traits are the most
likely outcome of stress, particular stressors may result in other
patterns of change. For instance, the novel pressure that an invasive
predator species brings on a native population can trigger the
increasing of body features (predator induced-defenses , Zhang et
al. 2017)) aimed to better escape negative interactions
(attack/predation), if the alien predator is perceived (via visual or
chemical cues) as a threat (Thawley et al. 2019). Moreover,
chemical pollution has been found to increase the occurrence of
fluctuating asymmetry in body traits linked to intraspecific interaction
(i.e. femoral pores, Figure 2B) in lacertids (Simbula et al.2021). Indeed, increase in fluctuating asymmetry has been suggested as
an indicator of the loss of genetic variation possibly occurring prior
to extinction (Leary & Allendorf 1989).
These physiological responses, including (but not limited to) declining
body mass/size, expression of chemical induced antipredatory features,
and asymmetry in meristic features will generally occur over longer time
periods than rapid behavioural changes described above, but may still
occur within the life span of an individual ( i.e. ≤1
generation), or be tracked across multiple sequential generations (e.g.
Clements & Ozgul 2016; Clements et al. 2017). For instance, the
body size reductions induced by food scarcity can be observed both
during an individual’s life and across generations (e.g., seabird annual
breeding season (Fayet et al. 2021)), as the nutrient deficit of
the parents is reflected by loss of condition in the hatchlings.
Likewise, toxic chemicals can accumulate in adult females inhabiting
polluted habitats and be transferred to their eggs, and the induced
traits shift could appear in the offspring over a single reproductive
season (e.g. few months for lizards, Simbula et al. 2021).
Therefore, after behavioural changes, morphological trait shifts
represent the next viable response to stress (i.e., second buffering
level) of individuals, and should thus occur as the second indicator of
increasing stress on a population.
SLOW CHANGES
The signals discussed thus far represent the impacts of stress
observable at the individual level; however, when a high proportion of a
population is similarly stressed, such individual level effects can
propagate to alter the structure and dynamics of a population through
changes in births, deaths, immigration, and emigration. Examples of
fitness related phenotypic changes that shape an individual’s life
history traits (e.g. shift in fecundity (Boggs & Ross 1993)) and thus
population dynamics are numerous in the literature. For example,
climatic change impacted the feeding activity of many polar bear
populations, resulting first in body condition reductions and
subsequently in a decreases of reproductive rates and cubs survival
(Stirling & Derocher 2012). Similarly, behavioural plasticity can
impact population dynamics in the long term: a recent example in
humpback whales has shown that changes in behaviour (shifts in diet and
seasonal movement) driven by environmental change led to a subsequent
decline in calving rates (Kershaw et al. 2021). Such decreases,
which necessarily reduce the lifetime reproductive success of an
individual, represent some of the the last stages of adaptive plasticity
in life history, where resources are reallocated from reproduction to
maintain the survival of the individual whilst allowing for the possible
exploitation of improved future conditions (Fleming et al. 2016).
Although such responses are carried out by the individuals during their
lifetime, i.e. occurring right after to or concurring with the
morphological shifts, the resulting signals become observable over long
(>1 generation) time frames via changes in the abundance
trends of a population. Indeed, such decreases in reproductive success
and increases in mortality will drive fluctuations significantly
different from the preceding stable periods (e.g., increasing variance
EWS (Clements & Ozgul 2018)). However, these changes will not
necessarily trigger/drive continuous declines in the abundance until
extinction (i.e. the population could stabilize at a new carrying
capacity level with lower resources quantity).
These slowly occurring changes represents the ultimate signals (last
stress buffering level) a population may show before the collapse
starts. Indeed, such a state of low recruitment potential may be
critical as the plasticity of behaviours and body traits may have
already been exhausted, and thus a population is more vulnerable to
fluctuations and collapse if stress continues to increase, or through
stochastic factors (e.g., catastrophic events). At this point, if stress
– be that abiotic or biotic – continues to increase then even this
last stress buffering level of the population will be overcome, death
rate will increase, and abundance will start to continuously drop until
the ultimate extinction of the population.
THE TIMELINE TO COLLAPSE
The above changes – ranging from rapid behavioral responses to declines
in the abundance of a population – constitute a predictable succession
of observable signals which we term the “timeline to collapse” (Figure
3). The presence of these signals assumes a continuous increase in
stress – be that biotic or abiotic – such that a population is able to
respond, rather than sudden step-shifts in a stressor which may
eradicate a population in the absence of any indicators (Clements &
Ozgul 2018).
Whilst the time at which behavioral, morphological, and abundances
shifts start (TBs, TMs and
TAs, Figure 3) are expected to be sequential, the time
intervals over which such shifts occur (IB,
IM and IA, Figure 3) may overlap.
Indeed, for an organism, changing a behaviour above a given threshold
may require the use of energy reserves that may trigger a change in
morphological traits. For example, for a seabird population (Figure 3),
increasing foraging distance may be the first response to decreasing
food availability, and in normal conditions the resources found in a
further area may be enough to compensate this additional foraging
effort; but if the food is needed for recruitment (i.e. feeding chicks,
(Fayet et al. 2021)) most of these resources will be transferred
to the offspring, and may not cover the individual’s energy cost of
increasing flight distance. Therefore, an individual will either i) fail
to replenish energy stores (e.g. start to lose weight) or ii) decrease
feeding rate to offspring to ensure they have the energy needed to cope
with the extended foraging distance (Fayet et al. 2021). This
will result in observing flight distance increasing together with
declines in the body weight of adults, offspring, or both. However, in
other scenarios we could observe a clear temporal distinction between
signals of stress time (i.e. no overlap among IB,
IM and IA). For instance, in the
presence of an invasive predator a prey species can go through an
initial fast and discrete behavioural change (e.g., a shift in
microhabitat use (Pierce 1988)), followed by a medium speed response
(e.g., change in body size due to different conditions in the new
microhabitat, (Leibold & Tessier 1991)), without any overlap between
these two signals.
The timeline can act as novel tool to discriminate populations tending
toward extinction from those simply adapting in the face of change. For
instance, focus on a single feature such as behaviour cannot discern a
population where individuals’ behavioural shifts are sufficient to cope
with stress (maintain fitness) from a population where individuals reach
the maximum level of behavioural plasticity and then start to compensate
the fitness loss with changes in morphological traits (e.g. decrease in
body size). In both cases, the monitoring would demonstrate a
significant change in behaviour. Instead, observing the temporal
sequence of changes in all the facets (behaviour, morphological traits,
changes in the variances of abundance, and finally abundance declines)
represent the key indicator that the stress gradually overcomes
individual and population level reactions, and thus collapse is
approaching.
These temporal pattern in signals of stress will necessarily be across
time scales relevant to the study organism, i.e. lifespans and
generations rather than absolute time periods. For small invertebrates,
fast response that may be observable over hours (e.g., Daphniadepth shift, Oram & Spitze 2013) while slow signals will occur over
days. For larger vertebrates, medium speed response may take place over
months (e.g., Steller sea lions weight loss, Trites & Donnelly 2003)
whilst EWSs occurrence and subsequent abundance declines may occur over
years. Regardless of the direction of the shifts and the stress type, we
expect the temporal sequence in the typology of signals (behavioural,
morphological, abundance; Figure 3) to remain broadly consistent.
Data requirements
The conceptual development of the timeline to collapse offers hope that
multiple data streams can be synthesized into a single predictive tool
which incorporates both the timing of changes in signals of stress, and
the order in which such signals occur. To apply such a framework to at
risk populations would require simultaneous monitoring of the behaviour,
morphological and/or life history traits, and abundance of populations.
Whilst such multivariate data may seem challenging to gather in real
world situations, recent technological advancements in data-collection
methods provide the opportunity to generate high throughput information
on these multiple features of populations with a relatively low
cost/benefit ratio (Thompson 2013; Ward et al. 2017). Indeed, GPS
tracking, biologging, acoustic monitoring, and photographic analysis are
now able to extract data on behaviours and morphological traits,
providing invaluable data even from a subset of the population,
(Desjonquères et al. 2020; Williams et al. 2020; Sequeiraet al. 2021; Shimada et al. 2021), and such approaches
have been implemented in vertebrates (both terrestrial and marine) and
invertebrates (Table 1). Biologging sensor are becoming rapidly more
affordable, and research to reduce the relative mass of these devices
ameliorates the ethical implications of weight and invasiveness
(Portugal & White 2018). Current biologger models can already collect,
among other information, data on geographical location, body movement
(e.g. posture, rotation, heading), physiological rates (e.g. heartbeat,
temperature, reproductive periods) and acoustic data (e.g.
vocalizations, external soundscape), widening the possibilities to
observe behavioural stress responses simultaneously in several aspects
of the individuals’ life (Table 1, Williams et al. 2020).
Similarly, for sound-emitting species, passive acoustic monitoring
allows the assessment of individuals’ behaviour, health status,
distribution, and population dynamics (Gibb et al. 2019;
Desjonquères et al. 2020). Acoustic sensors (microphones and
hydrophones) are relatively easy to deploy, can be used in low
visibility environments such as dense forests or deep-water and be leftin situ for long times, and have the advantage of being
non-invasive and able to survey a broad taxonomic range spanning from
vertebrates (e.g., cetaceans Sousa-Lima et al. 2018; bats,
Tuneu-Corral et al. 2020; birds and amphibians, Deichmannet al. 2017; Table 1) to insects (e.g. Orthopterans (Singh et al.
2020)). Moreover, unmanned aircraft systems (e.g. drones) now allow to
perform precise photogrammetric measurements of species that are
challenging to sample: drones photography can take measurements of big
marine mammals like pinnipeds and whales (also good ecosystem health
indicators, Krause et al. 2017; Kershaw et al. 2021)) and
estimate their mass and body condition, thus providing data on possible
shifts in body size (Clements et al. 2018).
Similarly, abundance estimates are being improved through new tools and
statistical models that complement classic approaches like direct
sampling and capture-mark-recapture methods (Seber & Schofield 2019).
Camera traps, and aerial and satellite images can be analyzed with
machine learning techniques to obtain accurate population counts even
for multiple species systems (Linchant et al. 2015; Norouzzadehet al. 2018), and citizen science projects can help to gather and
process such image data (e.g. Penguin Watch (Jones et al.2020)). Moreover, the recent explosion in environmental DNA (eDNA)
analyses can provide a cost-effective way to estimate populations that
is applicable to a large number of systems and taxa (Yates et al.2019). This broad suite of cutting-edge methodologies means that data on
multiple facets of a population will become increasingly available, much
of which has been largely overlooked by predictive ecology but which can
be leveraged under the timeline to collapse framework.
Forecasting
The timeline to collapse provides a conceptual framework to synthesize
multiple types of data to aid predicting the future dynamics of
ecological systems (Clements & Ozgul 2016). However, to apply the
timeline to collapse concept requires identifying appropriate data to
monitor (behaviours, traits), measuring baselines against which change
can be quantified, developing statistical tools to provide robust
detections of increasing stress.
Whilst some behaviours and morphological signals may provide general
indicators of increasing stress (e.g. increased dispersal), selecting
signals which are relevant to the taxa of interest remains key
(McClanahan et al. 2020). Expert knowledge can aid in this
(Reside et al. 2019), identifying which behaviours and traits are
most likely to change given the nature of the stressor, or – in cases
when the identity of the stress is unknown – what can provide general
indicators of an individual’s condition. After choosing what to monitor,
a quantitative and/or qualitative definition of “normal” values for
the identified behavioural, morphological, and abundance indicators is
needed, from which we expect to observe significant deviations when
environmental stress starts to increase (Figure 3). Defining such values
in wild populations ideally requires long term monitoring data (Wauchopeet al. 2021) on the multiple features of a population under
stable conditions. Such data will become progressively more available as
remote sensing and technological advancements continue to automate data
collection at large scales (Krause et al. 2017; Sequeira et
al. 2021). Alternatively, a comparative approach between populations
experiencing different levels of stress can provide baseline values such
as along a stress gradient (Ingram et al. 2021) – a so-called
space-for-time substitution (Keith et al. 2018); Fayet et al.
2021). Such data on non-stressed populations can characterize the range
of variation in the selected behaviours and morphological traits that,
together with the abundance fluctuations, can be analyzed to obtain
means and upper and lower confidence intervals. In the absence of such
long-term monitoring data, methods such as Dynamic Energy Budget Models
(DEMs) could help to set baselines using more general population life
history data. DEMs describe in a single framework how individuals’
energy is distributed for growth, somatic maintenance, development,
maturity, and reproduction (Baas et al. 2018). Standard life-cycle data
that can be obtained over shorter periods of time (e.g. Body length and
weight at birth, growth rate, maximum reproduction rate, lifespan etc.)
feed into the model that derive quantitative parameters describing the
organisms energetics. Trait information can also be incorporated into
the model to provide taxa specific estimates (Baas et al. 2018).
Parameterizing such DEMs with life history data from populations in
stable conditions could represent a viable and generalizable baseline
distribution from which one can compare observed changes (Lika et
al. 2011).
Regardless of how a baseline is defined, comparing these multivariate
estimates to observed changes in behaviours, traits, and abundances is
non-trivial. Recently developed statistical tools provide options to
achieve this; multivariate time series modelling (Wei 2018) may offer a
strong method for analyzing the timeline data (time series of behaviour,
traits, and abundance), whereby the trends of the different variables
can be analyzed through time while taking into account the
inter-dependencies between them (e.g. behaviour and morphology). For
example, Multivariate Autoregressive State Space (MARSS) models (Holmeset al. 2012) can use information on historical trajectories of
multiple variables to forecast future values while accounting for
multiple sources of uncertainty, and thus could represent another
valuable option to predict shifts in, for example, behavioural
indicators (Zhu et al. 2018). Alternatively, deep learning
networks such as recurrent neural networks and temporal convolutional
neural networks (Lai et al. 2018; Bury et al. 2021;
Lara-Benítez et al. 2021) could provide an even more powerful
approach to forecast future trends or state changes in such variables
(Guo et al. 2020), though these tools will require large amounts
of training data. Such approaches could be performed for single
populations but may be stronger at the landscape scale, whereby one
could combine inputs from multiple populations under different
conditions and use the data to train the deep learning algorithms to
then perform generic predictions in new cases. Therefore, despite the
complexity of analyzing multivariate data, these new tools offer the
opportunity to try to implement the timeline forecasting capacity.
Ecological insights
Whilst the temporal order of signals provides information on the
population’s future, the magnitude of the shifts in behavioural,
morphological, and abundance-based metrics may provide measures of a
populations ability to resist stress. For instance, for individuals of a
population suffering from resource loss (Figure 4), increasing their
foraging distance can initially compensate against the increased
stressor levels’ (i.e. buffering a given quantity of stress). However,
above a threshold (Figure 4, Point 1), expanding such behaviour is
insufficient to maintain fitness, and the animals’ body size is
impacted. Body size will also ultimately decrease if stress growth
persists, until a physiological limit is encountered, beyond which the
reproductive ability of a population is impacted (Figure 4, Point 2).
Therefore, the variation from the average value of pre-stress (stable)
conditions, measured in the behaviour and morphological trait prior the
onset of the next signal of stress, represents an intrinsic stress
buffering capacity (C): a measure of the magnitude of stress tolerable
before transitioning to the subsequent stress buffering level along the
timeline. If we define Bs and Ms as the
average values of a monitored behavioural metric and morphological trait
during stable conditions, and Bx and Mxtheir respective values at the onset of the next buffering signal/level
(Figure 4, Point 1 and 2), we may calculate Cb(behaviour C) and Cm (morphological trait C) as follows:
Cb=|Bs –
Bx|; Cm=|Ms – Mx|.
From this framework, average values of Cb and
Cm of individuals can thus be calculated for particular
behaviours or traits that can undergo continuous shifts and compared
among different species and populations. For instance, nematodes and
rotifers show extreme plasticity in morphology (reduction of up to
one-thirds of original body size (Rebecchi et al. 2020)) to cope with
long periods of stress (e.g. exsiccation of habitat), and thus they
would display higher values of Cm compared to e.g.
amphibians species with limited drought resistance. Such high value of
Cm reflects the large amount of stress they can buffer
by changing morphology before the eventual occurrence of abundance EWSs
in the population. Therefore, such buffering capacities may be compared
among different species to indicate which life history traits (group
living vs solitary animals, bigger vs smaller dimensions, specialist vs
generalist etc.) lead to species more resistant to stress. Additionally,
average Cb and Cm may vary among
populations of the same species, due to difference in biogeographic
history and genetic structure (e.g. allelic heterozygosity (Hansson &
Westerberg 2002)), which may provide information on how such factors
shape stress buffering capacity.
Caveats
The timeline to collapse concept necessarily makes assumptions about how
stressors will impact populations. The main assumption is that stressors
will increase over time (Figure 4), allowing populations to respond
gradually to increases in stress. However, as with EWSs and PVA, sudden
and/or catastrophic stress (drought, storms, fires etc.) may lead to
significant changes in the abundance or distribution of a population
without any warning. Moreover, even in cases where stress increases
continuously, the mutable nature of biological systems may create
situations where the sequence of signals may be different (e.g. body
traits shift occurs first, triggering then behavioural shift, or
concomitant abundance and trait change (Burant et al. 2021).
Finally, whilst it is possible to apply parts of the timeline concept
(and indeed doing so has been shown to improve the predictive accuracy
of forecasting tools (Clements & Ozgul 2016, 2018)), applying the
entire framework requires studying species that show quantifiable
behaviours and morphological traits, where gathering data is easier at
the individual perspective, and thus it may not be fully applicable to
animals such as sessile (e.g. Anthozoa), obligate parasite species or to
plants and fungi species. Nevertheless, we believe that in such cases a
partial application of the timeline concept (e.g., monitoring
morphological traits and abundance data) will improve the predictive
horizon of eventual collapses compared to considering only one type of
data.
CONCLUSIONS
Considering how anthropogenic stressors impact populations via changes
in individual-level features provides a key step forward in predicting
populations extinction. Doing so allows us to develop a conceptual
framework, the timeline to collapse, where the temporal aspect of
signals of stress can act as an additional corroborative tool to infer
risk of population collapse. The timeline to collapse approach also
provides a framework for the development of monitoring programs,
highlighting what data might be collected to help enhance biodiversity
monitoring, and how technological innovation might help to increase the
amount of data available (Pimm et al. 2015, Table 1). A holistic view of
how the behaviours, morphological features, and dynamics of populations
change as they become increasingly stressed will improve the
identification of what observable signals precede declines in the
abundance of populations, thus strengthening the tool arsenal for
fighting biodiversity loss.