1201
CONCEPTS & SYNTHESIS
EMPHASIZING NEW IDEAS TO STIMULATE RESEARCH IN ECOLOGY
Seasonality and predictability shape temporal species diversity
JONATHAN D. TONKIN ,
1,5
MICHAEL T. BOGAN,
2
NÚRIA BONADA,
3
BLANCA RIOS-TOUMA,
4
AND DAVID A. LYTLE
1
1
Department of Integrative Biology, Oregon State University, Corvallis, Oregon 97331 USA
2
School of Natural Resources and the Environment, University of Arizona, Tucson, Arizona 85721 USA
3
Grup de Recerca Freshwater Ecology and Management (FEM), Departament d’Ecologia, Facultat de Biologia, Institut de Recerca
de la Biodiversitat (IRBio), Universitat de Barcelona, Diagonal 643, 08028, Barcelona, Catalonia, Spain
4
Facultad de Ingenierías y Ciencias Agropecuarias, Ingeniería Ambiental, Unidad de Biotecnología y
Medio Ambiente (BIOMA), Campus Queri, Calle José Queri y Av, Granados, Edificio #8 PB,
Universidad de las Américas, Quito, Ecuador
Abstract. Temporal environmental fluctuations, such as seasonality, exert strong controls
on biodiversity. While the effects of seasonality are well known, the predictability of fluctuations
across years may influence seasonality in ways that are less well understood. The ability of a
habitat to support unique, non- nested assemblages of species at different times of the year
should depend on both seasonality (occurrence of events at specific periods of the year) and
predictability (the reliability of event recurrence) of characteristic ecological conditions. Drawing
on tools from wavelet analysis and information theory, we developed a framework for quanti-
fying both seasonality and predictability of habitats, and applied this using global long- term
rainfall data. Our analysis predicted that temporal beta diversity should be maximized in highly
predictable and highly seasonal climates, and that low degrees of seasonality, predictability, or
both would lower diversity in characteristic ways. Using stream invertebrate communities as a
case study, we demonstrated that temporal species diversity, as exhibited by community turno-
ver, was determined by a balance between temporal environmental variability (seasonality) and
the reliability of this variability (predictability). Communities in highly seasonal mediterranean
environments exhibited strong oscillations in community structure, with turnover from one
unique community type to another across seasons, whereas communities in aseasonal New
Zealand environments fluctuated randomly. Understanding the influence of seasonal and other
temporal scales of environmental oscillations on diversity is not complete without a clear under-
standing of their predictability, and our framework provides tools for examining these trends at
a variety of temporal scales, seasonal and beyond. Given the uncertainty of future climates,
seasonality and predictability are critical considerations for both basic science and management
of ecosystems (e.g., dam operations, bioassessment) spanning gradients of climatic variability.
Key words: climate; communities; desert annuals; migratory waterfowl; periodicity; seasons; stream
invertebrates; temporal beta diversity; turnover; wavelets.
To everything there is a season (Ecclesiastes 3)
INTRODUCTION
Ecologists have long understood that environmental
heterogeneity is intimately connected to species diversity.
In theory, predictable oscillations in an environment
should allow the coexistence of a great number species
over a given timeframe, with each species experiencing
optimum conditions at a different time and none
experiencing poor conditions for too long a time
(Hutchinson 1961). Indeed, temporal periodicity in envi-
ronmental conditions is a central component of eco-
systems worldwide, and within- year seasonality is among
the strongest and most well- known forms of such perio-
dicity (Fig. 1). More recently, seasonality has been
invoked to explain general phenomena such as life history
adaptations (McNamara and Houston 2008), latitudinal
diversity gradients (Hurlbert and Haskell 2003, Dalby
et al. 2014), and community structure (Chesson 2000,
Chase 2011), as well as specific phenomena such as
migratory dynamics of birds (Somveille et al. 2015), estu-
arine fish diversity (Shimadzu et al. 2013), and stream
invertebrate diversity (Bogan and Lytle 2007, Bonada
Ecology, 98(5), 2017, pp. 1201–1216
© 2017 by the Ecological Society of America
Manuscript received 26 September 2016; revised 23 January
2017; accepted 24 January 2017. Corresponding Editor: Helmut
Hillebrand.
5
E-mail: jdtonkin@gmail.com
1202 Ecology, Vol. 98, No. 5JONATHAN D. TONKIN ET AL.
ConCepts & synthesis
and Resh 2013). However, efforts to generalize how both
temporal variability in environmental conditions (sea-
sonality) and the reliable recurrence of these different
environments (predictability) determine species diversity
within a single location require a common currency to
measure predictability and seasonality.
Drawing on tools from modern wavelet analysis and
information theory, as well as ecological theory, we
provide a simple framework for understanding how the
seasonality and predictability of an environment interact
to shape temporal patterns of local diversity. We
emphasize that, while seasonality is a well- known struc-
turing force on biodiversity, our understanding of its reg-
ulatory influence on local communities is not complete
without understanding its predictability. Following a dis-
cussion on the ways in which seasonality of environmental
conditions can regulate diversity, we use this framework
to generate predictions about within- latitude temporal
diversity patterns, emphasizing the mechanisms that
promote overall temporal diversity (i.e., temporal
turnover), particularly for short- lived organisms. We then
demonstrate the framework using stream community case
studies from different regions at similar latitudes that
span a seasonality–predictability gradient in rainfall
patterns (mediterranean- climate western United States,
arid southwestern United States, maritime New Zealand).
We focus on the types of systems and dynamics that this
framework can directly address, but also demonstrate
that the methods we outline allow the identification of
various ecological trends across a variety of temporal
scales. By providing a framework for quantitatively con-
sidering seasonality and predictability of environmental
fluctuations, we hope to (1) spark renewed interest in the
role of seasonality and (2) stimulate new research on the
less well studied role of environmental predictability in
governing diversity of natural systems. We believe that
considering both concepts together can shed more light on
temporal patterns of local diversity in a variety of systems
allowing for better prediction and management of biodi-
versity under the ever- increasing threat of global change.
INCORPORATING SEASONALITY GIVES US
A RICHER VIEW OF ECOSYSTEMS
Most environmental phenomena occur with seasonal
oscillations, particularly temperature and precipitation,
but even regular oscillations can vary in biologically
important ways. For example, while the total annual
FIG. 1. Examples of seasonal changes in two ecosystems: Sonoran Desert sand dunes at El Pinacate y Gran Desierto de Altar
Biosphere Reserve (Sonora, Mexico) (A) before and (B) after winter rains and Chalone Creek at Pinnacles National Park (California,
USA) during the (C) dry and (D) wet season.
A
B
CD
May 2017 1203SEASONALITY–PREDICTABILITY FRAMEWORK
ConCepts & synthesis
duration of daylight is invariant across the globe, the
seasonal distribution of daylight varies dramatically with
latitude. As a result of these differences in distribution,
many ecosystems experience distinct seasonal conditions
that can favor entirely different communities and food
webs during parts of each year (McMeans et al. 2015)
(Fig. 1). Seasonality tends to increase in importance with
increasing distance from the equator. However, even
tropical regions that are not subject to extreme temper-
ature variations can experience seasonal fluctuations in
key environmental characteristics, such as precipitation
on land and upwelling in the ocean. These seasonal
abiotic oscillations lead to seasonal pulses of resources,
which in turn open up temporal niches for a wide variety
of species to reside in local habitats.
Climatic variability is at the heart of the distribution of
species globally and comprises many different compo-
nents, including seasonality, harshness, predictability,
and length of the favorable period for occupation
(Jocque et al. 2010). Species life histories are finely tuned
to capitalize on specialized temporal niches associated
with this variability (Chesson 2000, Chase 2011; Fig. 2).
Consequently, temporal diversity can be promoted
through a variety of channels, such as migration to exploit
resources and escape competition (Somveille et al. 2015),
highly synchronous seasonal reproduction, or seasonal
fluctuations in abundance (Shimadzu et al. 2013; Fig. 2).
In fact, oscillations in environmental conditions, such as
seasonal shifts in productivity, can explain discrepancies
in latitudinal diversity gradients (Hurlbert and Haskell
2003, Dalby et al. 2014). Nevertheless, this very season-
ality can reduce diversity outside of tropical regions by
acting as an environmental filter for organisms (Gouveia
et al. 2013). Thus, while seasonality is clearly important,
it appears to interact with other forces to produce observed
patterns of biodiversity.
Several studies have suggested mechanisms by which
temporal fluctuations affect pairwise or community- wide
species interactions, and thus patterns of biodiversity.
Although limited resources and interspecific competition
can lead to the exclusion of species (Connell 1978), sea-
sonal variation in environmental conditions can facilitate
the persistence of similar species (Tilman and Pacala
1993). Such seasonal variation (temporal environmental
FIG. 2. Examples of taxa with life cycles that are synchronized to take advantage of predictable, seasonal changes in
environmental conditions. Citations: 1, Hynes (1976); 2, Jacobi and Cary (1996); 3, López- Rodríguez et al. (2009); 4, Mulroy and
Rundel (1977); 5, Guo and Brown (1997); 6, Mathias and Chesson (2013); 7, Craig et al. (2004); 8, Jeffres et al. (2008); 9, Arthington
and Balcombe (2011); 10, Dalby et al. (2014); 11, Keeley and Zedler (1998); 12, Collinge and Ray (2009); 13, Kneitel (2014).
Examples
Aquatic insects
in temporary
streams
Annual plants
in deserts
Fish
in floodplains
Waterbirds
in lakes and
wetlands
Plants and
crustaceans
in vernal pools
Seasonal factor
Stream flow
Precipitation
River flooding
Vegetation growth
Hydroperiod
Mechanism/process
Dormant egg and larval stages
to survive dry seasons,
reactivate during wet seasons;
terrestrial insects occupy
streambed during dry season.
Rapid development and growth
with seeds that lay dormant until
following wet season; some
species grow during winter-
spring rains, others during
summer rains.
Disperse into floodplains when
rivers top their banks, consume
abundant resources in
floodplains, return to primary
river channels as flow recedes.
Migratory waterfowl occupy
productive mid-latitude lakes and
wetlands when vegetative
growth is highest (summer), then
migrate to other habitats during
unproductive winter season.
Crustaceans (e.g., fairy shrimp)
active during wet season, enter
dormant stage when pools dry;
plants grow during drying phase,
dormant seeds persist through
wet phase.
Citations
Canada,
Spain, USA
(1,2,3)
USA
(4,5,6)
Australia,
Bangladesh,
USA
(7,8,9)
Global
(10)
Australia,
Chile, USA
(11, 12, 13)
1204 Ecology, Vol. 98, No. 5JONATHAN D. TONKIN ET AL.
ConCepts & synthesis
variation) operates in a conceptually similar way to
physical heterogeneity (spatial environmental variation)
through preventing competitive exclusion and creating
niches for species in different seasons. For instance, the
storage effect allows multiple species to occupy similar
habitats through populations “storing” the gains made in
good years to buffer against losses in bad years (Chesson
2000). Thus, theory suggests that, at least in principle,
organisms can capitalize on seasonal dynamics in ways
that enhance the overall diversity of a single habitat.
LIFE- HISTORY EVOLUTION IN SEASONAL ENVIRONMENTS
The life histories of organisms are intimately connected
to seasonality. Fundamental vital rates such as growth,
mortality, and reproduction seldom remain static over
time, but vary in response to environmental conditions
that change seasonally. Thus, many important life-
history decisions are likely governed by seasonality,
including age and size at maturity, timing of migration or
breeding, and allocation to growth vs. reproduction
(McNamara and Houston 2008). Life- history models
that explicitly incorporate seasonality predict that when
disturbances such as floods, droughts, or fires recur with
sufficient seasonality, selection will favor strategies that
produce synchrony with this disturbance regime (Cohen
1966, Rowe et al. 1994, Lytle 2001). Seasonality can thus
be viewed as an adaptive force that entrains the life his-
tories of organisms into specific temporal strategies.
From a community perspective, this should have a direct
effect on the composition of the local species pool. In
environments with strong, recurrent seasonality, species
should possess specific adaptations or abilities for coping
with seasonal environments (due to selection for seasonal
life histories) and community composition should differ
from one part of the season to another (due to life- history
trade- offs that favor specialization on a particular
season). The latter prediction is directly testable by exam-
ining patterns of community structure across seasons,
and is the focus of this study.
Life history evolution is not confined to annual time
scales, however. Processes operating at other temporal
frequencies, such as diel fluctuations in light levels,
monthly changes in ocean tidal cycles, and supra- annual
changes in oceanic and atmospheric conditions can all
potentially drive the evolution of life histories. The rel-
evant factor is the temporal scale of the environmental
phenomenon with respect to the lifespan of the organisms.
For example, models examining the timing of maturation
in seasonal environments predict a strong evolutionary
response when the lifespan of the organism corresponds
roughly to the frequency of environmental fluctuations
(Iwasa and Levin 1995, Lytle 2001). On the other hand,
organisms with too- short or too- long lifespans may fail
to evolve synchronous life history strategies, even though
this may reduce fitness and even result in ecological
exclusion from the system (Lytle and Poff 2004). For
these reasons, we need analytical methods that visualize
ecological processes across a range of time scales in order
to identify the most important frequencies, with respect
to the organisms of interest.
SEASONALITY AND PREDICTABILITY DEFINED
Seasonality can be defined in many ways, depending on
the application. The astronomical definition of the four
seasons relates to the timing of the summer and winter
solstices and vernal and autumnal equinoxes, which
differs slightly from the meteorological definition based
on calendar dates (Timm et al. 2008). From an economic
perspective, Hylleberg (1992:4) defined seasonality as
“the systematic, although not necessarily regular,
intra- year movement caused by the changes of the
weather, the calendar, and timing of decisions, directly or
indirectly through the production and consumption deci-
sions made by the agents of the economy.” We used a
definition of seasonality of environmental phenomena
based on Lieth (1974:5): “Seasonality is the occurrence of
certain obvious biotic and abiotic events or groups of
events within a definite limited period or periods of the
astronomic (solar, calendar) year.” Essentially, this rep-
resents the degree to which within- year conditions are
distinct. Thus, a mediterranean- zone climate creates a
highly seasonal environment because summer conditions
are dry and warm while winter conditions are cool and
wet. Colwell (1974) used information theory to formalize
this notion: Colwell’s M, or “contingency,” measures the
degree to which biological events such as flowering, or
physical events such as monthly rainfall totals, are tied to
specific times of the year (Box 1). Colwell’s M is useful as
a measure of seasonality, in that high values indicate
strong differences across seasons.
While environments can be seasonal, however, they are
not necessarily predictable (and vice versa). Predictability
can be variously defined depending on the phenomenon
of interest and has been the topic of much discussion
(Resh et al. 1988, Poff 1992, Lytle and Poff 2004). For
our purposes, we define predictability as the regularity of
recurrence of the within cycle (e.g., annual) distribution
of events across multiple cycles. For instance, if the
annual distribution of rainfall recurs each year, regardless
of what that annual distribution is, it is considered pre-
dictable. Thus, in the case of our mediterranean- zone
example, the dry season is highly predictable if it recurs
reliably during the same summer months from year to
year. By contrast, some temperate oceanic regions (e.g.,
New Zealand, as demonstrated in the following case
study) can experience wet months nearly any time of the
year, varying from year to year, so predictability is low.
In many biologically important cases, events recur on an
annual time scale, but this need not be the case: phe-
nomena such as lunar tidal cycles, El Niño – Southern
Oscillation, and the Pacific Decadal Oscillation all recur
at non- annual time scale, and thus have a characteristic
and measurable predictability associated with them.
Wavelet analysis (Box 1) can be employed to quantify the
May 2017 1205SEASONALITY–PREDICTABILITY FRAMEWORK
ConCepts & synthesis
predictability of periodic phenomena such as rainy and
dry seasons (Daubechies 1990, Torrence and Compo
1998). A benefit of wavelet analysis is that it allows us to
quantify the strength of predictability at any time scale.
Although we are often interested in phenomena that
recur on an annual basis, we can also explore data for
patterns that recur at other time scales.
Wavelet analysis has been used for a variety of appli-
cations in ecology and other fields (Daubechies 1990,
Cazelles et al. 2008), such as comparing compensatory
Box 1. A framework for measuring seasonality and predictability
Wavelets
Spectral analysis partitions the variability within a time series into different components characterized by
different frequencies. The contribution of each frequency (period) to the variability (power) within a time series
can be revealed by plotting the power spectrum (power as a function of frequency). This power spectrum then
enables examination of key temporal scales of variability within the time series.
Some have recommended the use of the Fourier transform to decompose variation in ecological or environmen-
tal phenomena, such as streamflow (Sabo and Post 2008). However, wavelets have the advantage over Fourier
transforms in their scale independence and ability to examine multiple scales simultaneously (Torrence and Compo
1998). Rather than simply detecting the dominant frequencies averaged over an entire time series, wavelets can
preserve the location of an event in space or time, enabling tracking of periodic phenomena over the time series.
Essentially, wavelet transforms decompose a time series into three- dimensional space: time, scale/frequency, and
power, where power represents the magnitude of variance at a given wavelet scale and time. Thus, they can help
to reveal more subtle structures that would otherwise be missed in multi- scaled, non- stationary, time series data
(Smith et al. 1998).
Depending on the data and objectives, there are a variety of wavelet functions that can be used, including Morlet,
Mexican hat, and Paul. The Morlet wavelet is well suited for hydrological time- series data, being a nonorthogonal,
complex wavelet transform (Torrence and Compo 1998). Nonorthogonal wavelets tend to be more robust to noise
and to variations in data length than other decompositions (Cazelles et al. 2008). Moreover, “complex” wavelets
are better at capturing oscillatory behavior than “real” wavelets, which are better used for isolating individual peaks
or discontinuities (Torrence and Compo 1998). The significance of wavelet power spectra can be tested against
background (noise) spectra. In this case, we used the default white noise spectrum (constant variance across all
scales). Using this approach allows direct comparison between time series, as after detrending the time series, it is
standardized to obtain a measure of wavelet power relative to unit- variance white noise. When the wavelet power
exceeds the background, it is deemed significant (here at 95% confidence level; Torrence and Compo 1998).
Colwell’s indices
Colwell (1974) devised three interrelated metrics based on information theory to quantify the general character-
istics of periodic phenomena: predictability (P), constancy (C), and contingency (M). P represents the relative
certainty of knowing a state at a given time, and is the sum of constancy and contingency. C represents the degree
to which a state stays the same throughout all seasons. M describes how closely different states correspond to
different time periods within a year. Thus, M contains information about the degree of seasonality experienced by
an environment. As P is the converse of uncertainty, it stands to reason that its calculation is based on the math-
ematics of information theory (Colwell 1974). P is maximized when the environmental phenomenon is constant
throughout the year, if the seasonal fluctuation is consistent across all years, or a combination of both. It is
important to note that Colwell’s P is a fundamentally different metric from our wavelet- derived estimate of pre-
dictability. When analyzing annual data, Colwell’s P is linked to within- season dynamics, while our wavelet-
derived measure can be applied at any temporal scale.
Combined
In our framework, we combine the simplicity of Colwell’s information- theoretic metrics and the power and
graphical quality of wavelets to define seasonality and predictability, respectively. We opted to combine these two
approaches as Colwell’s metrics are inherently linked (i.e., predictability is the sum of contingency and consist-
ency). Specifically, our measure of seasonality is M/P, which is Colwell’s measure of contingency standardized by
Colwell’s within- season predictability. This measures the degree to which the environment varies during the course
of a single year. Our measure of predictability uses the proportion of wavelet power that is significant at the
12- month interval across the entire time series. That said, we could have also compared the standardized wavelet
power at the 12- month interval to achieve a similar result (as per Fig. 4D). Furthermore, the chosen period of
interest can be adapted to any given recurrence interval and wavelets can be employed as an exploratory tool to
identify dominant frequencies in the data. This flexible measure of predictability allows us to examine the impor-
tance of phenomena that recur at intervals less than or greater than the typical annual cycle.