AbstractPhenology refers to the seasonal timing patterns commonly exhibited by life on Earth, from blooming flowers to breeding birds to human agriculture. Climate change is altering abiotic seasonality (e.g. longer summers) and in turn, the phenological patterns contained within. However, how phenology evolves is still an unsolved problem. This problem lies at the crux of predicting future phenological changes that will likely have substantial ecosystem consequences, and more fundamentally, of understanding an undoubtedly global phenomenon. Most studies have associated proximate environmental variables with phenological responses in ways that are case-specific, making it difficult to contextualize observed changes within a general evolutionary framework. We advocate for general theory of phenological evolution centered around constructing null hypotheses to explain the disparate cases of phenological change in a systematic manner, and to distinguish when cases are surprising, and why. We outline the necessarily complex but universal ways in which timing within seasonal windows map onto evolutionary fitness. Throughout, we borrow lessons from life history theory and evolutionary demography that have benefited from a more first principles-based theoretical scaffold. Lastly, we identify key questions for theorists and empiricists to help synthesize and advance our general understanding of phenology.Introduction Phenology—the seasonal timing of biological events on scales ranging from individual life cycles to global cycles—is a universal feature across plants and animals, from ecosystems (e.g. flowering, emergence, migration) to human systems (e.g. agriculture) [1–3]. Phenology’s ubiquity is perhaps unsurprising: the revolution of the Earth around the Sun preceded the origin of life itself, and underlay the course of evolution ever since. Thus, phenology is arguably one of the deepest themes in ecology. The rapidly growing interest in phenology over the last few decades has focused on consequences of climate change [2,4]. But explanations of recent phenological changes are typically system-specific and focused on empirical cues and responses. This focus has not been matched by developments of a higher-order organization of the principles of phenological selection despite phenology’s global existence [5,6]. Distillation of the first principles of phenological evolution is urgently needed to synthesize and contextualize the large body of disparate reports and explanations of phenological divergence unfolding under climate change. Moving forward, such a theoretical organization will 1) make phenological research more streamlined as new knowledge gets compared against and added to a common conceptual framework, 2) enable baseline predictions of future phenological change even where data to parameterize models are yet insufficient for a system of interest.Phenology—regardless of scale or system—describes cyclical patterns in the dimension of time (Fig. 1). Historically, spatial pattern-thinking has arguably biased many fundamental frameworks in ecology and evolution theory from island biogeography [7] to regional-local community hierarchies [8] to species ranges [9], perhaps due to the immediate obviousness of spatial patterns. However, decades of phenological observations show that there are repeatable and predictable biological patterns in the dimension of time as well. The Earth’s physical environment is structured by temporal cycles, even in comparatively less seasonal environments such as the Tropics in lower latitudes [10]. Such physical cycles bound the time windows for predictable biological dynamics such as seasonal life history events of individual organisms, oscillations in the numbers of individuals in a population expressing such seasonal traits, or in the number of species expressing them. The evolution of cyclical timing patterns presents an interesting quantitative puzzle, not least because evolution is itself a temporal process.Climate change influences cyclical timing patterns in two main ways. The first is via overall warming [11], e.g. increases in mean annual temperature, which influences rates of biological processes such as development. Studies typically analyze the timing of measurable state transitions such as bud-burst or flowering for plants [3,11–13], and breeding or migration for animals [14–17]. Research on phenological shifts (Box 1) has disproportionately focused on the warming aspect [18]. The second is via entire ‘climatic’ growing seasons (e.g. the continuous frost-free period of the year) being extended by earlier springs and later autumns [18–23], as well as potentially becoming more variable [19]. The climatic growing season is a period when biological activity is favorable [24], or possible at all especially in high latitude or altitude systems. Therefore, changes to the length and predictability of the climatic growing seasons represent an alteration to the arena needed for the unfolding of life cycles, population dynamics, and larger scale ecosystem processes. There is mounting evidence that the warping of the seasonal time window dramatically drives rapid evolution of individual phenological traits [5,24–28], and whole phenophases (Box 1; [29]). That the very temporal arena containing temporal phenological patterns is itself morphing makes the evolutionary process of phenology an ever more complex and intriguing puzzle.Perhaps the most unresolved conundrum is that the same change in climatic growing seasons often induces very different phenological shifts between organisms occupying the same habitat in both direction (when) and magnitude (by how much). Discrepancies in shifts are observed among individuals [30], between traits in a single species (e.g. early-life traits shifting more than late-life traits [25,31,32]), as well as among species in ecological communities [33]. Longer climatic growing seasons are not necessarily beneficial, nor do they have the same consequence for different populations and species. For instance, longer growing seasons have benefited some species (e.g. orchids in Norway [34] or yellow-bellied marmots in the US [35]), but markedly decreased growth rates of others (e.g. mustard white butterflies in the US [36]). Interestingly, these discrepancies might illuminate an interaction between phenology and demography that makes a wide array of responses more tractable, which is gaining attention [35–41].In summary: the rapidly growing body of top-down (observations first) studies of phenological change is brimming with contrasting effects and specific explanations, which makes it difficult to generalize the eco-evolutionary links between seasonality change and phenological change across systems [16,42,43]. What is comparatively missing is a bottom-up approach to phenological evolution that first defines null expectations and testable hypotheses. Such a focus can help us study which, when, how, and why theoretical assumptions might be wrong, and how they should be updated to build general theoretical explanations of a phenomenon that is undoubtedly ubiquitous. Further, a focus on first principles causality might enlighten commonalities between seemingly disparate cases of phenological shifts.In our first section, we argue that a first principles view of phenological evolution starts with a recognition of a simple truth in any system: the fitness of phenological traits varies over time within a year. In other words, if any phenological manifestation confers an equivalent consequence for evolutionary fitness, there would be no discernible phenological pattern around the planet. Drawing analogies from spatially-oriented theory, we highlight that treating the dimension of time in bounded units enables quantitative conceptualization of how phenological variation maps to variation in fitness. Then, we outline how this variation in fitness produces selection pressures on phenological variation, first at the scale of individuals and populations, and then at the scale of multi-species communities. Our overarching goal is to introduce theorists to the unsolved puzzle of general selective forces acting on phenology around the world, and invite empiricists to test those general hypotheses to advance cross-system understanding.Phenology: cyclical patterns in the dimension of time.Patterns, in any dimension, can only be quantified and systematically compared in the context of defined bounds and scales. Drawing analogies from the more familiar spatial dimension helps crystallize this point (Fig. 1). Ecologists commonly discretize infinite space into appropriately sized frames for the question at hand even if the chosen scale is imperfect and arbitrary [44]. Consider how we bound nature with transects or grids. We use statistical tools to translate observations within the bounds to an understanding of how and why entities are distributed in space, even though a vector crossing the surface of a spherical planet is in reality infinite. Similarly, while time is in reality boundless, delineation (e.g. the climatic growing season) allows systematic quantification and comparison of phenological timing patterns contained within (Fig. 2). In both space and time, some delineations are non-arbitrary and important for biological dynamics, such as islands or habitat boundaries in space, and daily or seasonal cycles in time.Focusing on the temporal bounds that encompass annual patterns might be an important step for first principles theory development because bounds are one of few parameters that are universal. In other words, any system that exhibits cyclical phenological patterns has a beginning and an end to the seasons that constrain the sequence. The relationship between the size of the bounded domains in natural systems—whether in space (area) or time (duration)—and the biological patterns they contain is often complex, and thus inevitably requires mathematical modelling. For example, in space, the size of islands or habitat patches nonlinearly determines biodiversity, distribution, and coexistence patterns contained within that space [7,9,45–48]. Analogously, expansions of the climatic growing season (the ‘size’ of the bounded domain in the time dimension) are associated with complex and often unintuitive phenological pattern changes among species within the seasonal window [11,23,49–51]. In contrast to the space dimension analogue, theoretical understanding of how changes in the size of the seasonal time window drive phenological change is much more unresolved. Recent theoretical work, however, showed that simple contractions or protractions of the cyclical time window alone can drive diverse and dramatic changes in life-history strategies that underpin phenology [38,52].In theorizing the causality behind any change in temporal patterns, it is also important to keep in mind that cyclical phenological patterns are distinct from emergent ‘phenomenological cycles’ that arise from internal systems dynamics (e.g. predator-prey cycles) or Markovian transition processes (e.g. ecological succession). In contrast, phenological patterns are evolutionarily adaptable strategies that are repeatedly expressed within periods of geophysical environmental cycles (Fig. 1; [27,28,53]). As an example for the adaptive nature of phenology, studies using model systems such asArabidopsis show that phenological traits like flowering time are expressed in the laboratory even in the absence of climatic cues characteristic of the natural populations’ localities, and can even be mapped to specific genes under selection [54].Lastly, two caveats should be seriously considered when inferring evolutionary causality behind cyclical phenological patterns:temporal contingency in abiotic and biotic dynamics, andscale relativity between life cycles and seasonal cycles.Temporal contingency . Phenological patterns in a given seasonal time window are deeply contingent on past windows, importantly with respect to both the abiotic as well as biological dynamics. Here, the analogy between spatial pattern formation and temporal pattern formation breaks: in space, causation can be transferred bidirectionally in three dimensions, but causation is unidirectional (“anisotropic”) in time, from past to future. The anisotropic nature of temporal patterns makes causal influences stronger in time than in space since effects from multiple directions can be counteracted or obfuscated in space [55]. In other words, some or all environmental factors as well as surviving individuals in a biological system in a given time period necessarily had to arise from past time periods. Abiotically speaking, future environments are dependent on past windows, often in an autocorrelative manner with a few dominant time lags. The consequences of temporal autocorrelation in environmental variables such as temperature or food availability have been extensively studied in the contexts of population dynamics [56,57], and life-history evolution [58,59]. However, the effect of autocorrelation and temporal contingency on the natural selection of cyclical phenological patterns is much less well understood (but see [60,61]). Biologically speaking, individuals’ future phenological timings are inherently dependent on the individuals’ past investments and trait expressions (e.g. energy expenditure in early life phenological traits influences the amount of resources individuals need to accrue for subsequent growth, survival, or reproduction, and thus the timing of those transitions, e.g. [50,62]; Fig. 3). The population-level distribution of phenotypes is constrained by those whose phenological timing in past windows was compatible with their survival. One fruitful avenue might be to adopt modeling methods developed in evolutionary demography and life history theory that set up the environmental cycles and biological dynamics interactively; models such as adaptive dynamics [50,63] allow the interplay between the two types of temporal dependencies to analyze the effect of eco-evolutionary feedback dynamics.Scale relativity . A collection of species that exhibits a repeatable phenological pattern year to year in the same space may consist of strikingly different generation times or activity schedules; hence, those species’ population dynamics actually operate on very different scales of time (e.g. the community of phytoplankton and zooplankton in Lake Washington, USA shows predictably synchronized seasonal temporal patterns but the two trophic levels have very different generation times [64]). Annual organisms fit one generation within one period of an annual cycle, whereas perennial organisms experience multiple periods per generation, and shorter-lived organisms fit multiple generations within the same annual cycle [65]. In space, too, local patterns are influenced by processes larger than the scope of study, which are invisible to the local observer [44]. Analogously, longer processes are invisible to the ‘brief’ observer of natural systems. The key point is that the delineation of time into bounded units is necessary for standardized measurement of the distribution of biological events within time units, and development of explanatory theory. The goal is to develop theories that generally explain the widespread phenomenon of seasonal biological rhythms in nature, despite the fact that the scale of seasons means very different things to species with vastly different generation times.Towards that goal, we ask a general guiding question: how do organisms that live in environments with periodic time windows evolve to utilize non-random portions of the windows? We break the question down to two key hierarchies: single species phenological evolution, and community interactions that influence multiple coexisting phenologies.How can life-history and demographic theory help establish first principles of phenological selection?Phenological timing is typically studied as a variable responding to seasonal transitions in the abiotic environment (e.g.temperature, snow melt, photoperiod, precipitation). Responses to seasonal environmental variables, often involving plastic expressions of traits [2,66], constitute proximate phenological causality (Box 1). Environmental cues often have tractable effects on the timing of trait expression, and will continue to be important targets of research as cues will likely continue to shift and become more unpredictable under climate change [2,67,68]. However, proximate investigations often cannot fully explain or predict phenological shifts in many cases; species in the same space experiencing the same change in seasonal cycles often exhibit unexplained variability in phenological shifts [33,43,69–72]. These formerly surprising discrepancies appear to be commonplace, and confirm three notions: 1) there is of course no single optimal phenological timing, or shift, for all species, 2) there are unexplained evolvability differences between species with respect to their phenologies in response to environmental change, and 3) investigating the correlative trait responses to environmental variables might not be sufficient for understanding the general selective pressures acting on phenological change.A sense of what constitutes ‘correct’ timing, or the baseline null expectation of how phenological timing should change given some change in the environment, is currently not theoretically generalized. Expectations are often set by intuitions that can arise form system-specific knowledge, e.g. food availability for birds [15]. However, given the geometric nature of population growth and fitness, it is at least theoretically conceivable that a seemingly imperfect matching of phenological timing with respect to some relevant target such as seasonal food peak is actually optimal due to longer-term payoffs [6,73,74]. Post-hoc statistical analyses of phenological change with candidate environmental variables cannot easily integrate responses across the lifespan to reveal impacts on lifetime reproductive success, and selection over multiple generations, to offer explanations of ultimate causality (Box 1). Most importantly, a generalized evolutionary framework can allow one to quantify how unexpected an observed phenological shift really was (e.g.statistically unlikely) against null expectations. For instance, [38] theoretically showed that with small differences in the combinations or magnitudes of life-history trade-offs, populations can have dramatic—and directionally opposite—shifts in life histories even when given the same change in environmental seasonality.Life history theory and evolutionary demography (Box 1) have consistently provided biologists with remarkable causal explanatory power based on simple, species-agnostic frameworks [75,76]. While life history theory has certainly entered the field of phenology [2,24], the likes of the bottom-up theoretical structure that exists in the former discipline has not been established in the latter. Life history and evolutionary demographic theoretical frameworks consider fundamental processes that are universal across organisms such as birth, growth, reproduction, and death. The classic models are free from species-specific assumptions (e.g.[77–80]), and draw broad conclusions about the direction in which life-history evolution should proceed if, for example, certain age classes experience selective mortality. The classic models then extend the calculation to the population level by conceptualizing the relative fitness differences among individuals along some phenotypic or external (e.g. environmental or food type) gradient, which provides the basis for natural selection [81]. These calculations are then said to provide null, testable hypotheses. Such a species-agnostic, general theoretical backbone has motivated decades of life history research across vastly different systems in a systematic manner [82,75,83]. As a famous example, reduced adult survival was predicted to drive evolution towards earlier maturation and increased reproductive effort, which was repeatedly supported in Trinidadian guppies [82,84]. The philosophy of such fields is not to precisely explain every system with one model but to provide a common-language framework to be flexibly parameterized and tested by researchers to study their specific systems.Similar to the phenotype-to-fitness mapping considered by life history and demography theorists, timing of occurrence or trait expression within seasonal windows is an axis of fitness variation upon which natural selection can act [4,85,86]. The key practical benefits for phenology that these neighboring disciplines offer might be quantitative tools to deal with temporal contingencies within and across seasonal time windows. Namely, life history models specify temporal contingency in two main forms: 1) an individual organism’s current investments into biological functions influence its own future investments, and 2) current biological investments have rippling consequences for future generations [74,75]. Thus, selection on phenological timing within one seasonal window depends on selection in past and future seasonal windows (Box 2; Fig. 3; [87,88]). Demography integrates temporal contingencies in the dynamics of stage-/age-/size-/sex- structures of populations into selection dynamics. For example, fluctuating age- or stage-structures of populations, as opposed to simply population size, influence population growth trajectories, as well as calculations of optimal phenotypes [76]. Calculations of selection on life-histories when such real structural complications are considered can be very different from when they are not considered [89,90]. Another real complication of natural populations that demographic theory is suited to deal with is that individuals in populations exhibit variations in phenological schedules. For example, sexes of the same species often have different courses of seasonal developmental sequences and are affected differently by change in seasonality [91]. Seasonal synchrony of sexes is important for mating or even predator swamping [92]. For Scottish red deer, climate change has induced unequal advancements of phenological traits between males and females, leading to a contraction of their seasonal breeding window [93]. Further, different life stages of a single species can be differentially shifted by climate change. For example, in yellow-bellied marmots, advancements in dates of emergence from hibernation and weaning, but not of the onset of hibernation, led to the lengthening of their growing season, and to increases in body mass, reproduction, and population size [35].Above examples of studies that incorporated life history interdependencies and demographic structure into phenological analysis demonstrate that phenology is a highly eco-evolutionary process (Box 1), and would benefit from being modelled as such. For instance, phenological selection shapes the individual variation of life cycle schedules within a seasonal window. The life cycle decisions made in that seasonal window have consequences on the survival and life cycles of genotypes that make it to future seasons due to intergenerational trade-offs [74]. These genotypes then shape the standing variation of traits and population structure that comprise the raw material available for selection in future windows, completing the eco-evolutionary loop. Such a demographically explicit conceptualization of phenological evolution may be one of the most promising targets of theoretical progress [32,40].In testing phenological evolution theory using the common types of phenological data, a nuanced conceptual gap that needs to be bridged is one between how ‘rate’ (i.e.speed of processes or number of events in a time interval; e.g.oscillation frequency) evolves and how ‘timing’ (i.e. the occurrence of events in reference to a clock; e.g. oscillation phase) evolves. Rates are the parameters typically manipulated in demographic and life history models due to the time differential nature of dynamical systems modeling. Such models ask what happens over a fixed time step, whether that be a large step (e.g. a month), or an infinitesimally small one (e.g. \(\operatorname{}\frac{x}{t}\)). Conceptions of rate, such as growth, force the theorist to confront the fact that all phenology-related processes require time to complete such as size growth and physiological development. For example, when a flowering event is detected, it represents the culmination of a series of upstream biological steps leading up to that point; these can be aggregated to express a rate to reach that point. Therefore, one needs to consider the correlated and sometimes antagonistic selection pressures involved prior to the detectable timing of an event. However, the actual timingof an event is often what affects intra- and interspecies interactions such as mating or predator avoidance, and determines the set of environmental conditions experienced by an individual. Timing is a measurable point event that affects survival, and thus is potentially more ‘visible’ to selection [2]. Further, events like flowering reflect actual categorical change with a binomial property and is thus more easily measurable than rates. Likely for these reasons, data on timing dominate phenological studies (e.g. [94]). As a starting point, rate and timing are analogous in simple cases such as annual organisms that start and end their lives in a year (i.e.fast -growers mature earlier in a season). For species with more complex life histories, this conversion does not necessarily hold true. Marrying rate-based theoretical foundations with decades of existing timing data will unlock important advances in our general understanding of phenological evolution.How do species interactions produce and maintain diverse phenologies in the same space?Phenological evolution occurs in the context of ecological communities. The challenge is to understand how periodic interactions between coexisting species influence each species’ adaptive occupation of different portions of seasonal windows. Empirical evidence shows that different types of ecological interactions such as competition, invasion, or consumer-resource dynamics can alter the occurrence or trait expression timing of species in a community. Broadly, periodic interactions can favor overlap (Fig. 4A) or segregation (Fig. 4B) of phenologies between two species within seasonal time windows. Mechanisms depend on context and history. For example, experimental reduction of plant species diversity in a serpentine grassland community in California, USA advanced the phenology of remaining species, suggesting an infilling of newly available temporal niches [95]. This suggests that competition may limit co-occurrence. Analogously, exotic plant species may invade a new community by exploiting early-season phenological niches in which competition by co-occurrence with native species is lower [96] (but see [97]). A similar pattern can be achieved through a consumer-resource dynamic: introduction of large vertebrate herbivores may have selected for advanced flowering time in forage species in the US Southwest because earlier flowering reduces herbivory-induced loss of reproductive structures [98]. Mismatches in phenological shifts across trophic levels can have adverse effects on reproduction, survival and fitness of coexisting species, and cause rapid increases in extinction probability of populations [87] or health of whole ecosystems [5]; some trophic links such as plant-pollinator pairs appear capable of advancing constituent phenologies fairly synchronously [99,100], possibly suggesting that at least in some cases the selective forces on phenology imposed by species interactions are dominant over those imposed by single-species life history optimization. One fruitful avenue of theoretical advancement will be to incorporate the various modes of periodic phenological interaction into models of single-species phenological evolution. Interactions can be treated as dynamic time-dependent parameters that modify fitness landscapes of each involved species. Viewing phenological communities as dynamical systems in this way might help explain many of the incongruous cases of phenological shifts that appear unintuitive when studied out of the context of the community.The key ecological consequence of the differential expansions, contractions, and shifts among species’ phenologies is that the interaction potential between combinations of species can change within seasonal time windows [33,101,102]. Thus, novel ‘no-analog’communities (sensu [103]) can form through the season. For example, a recent 12-year observational study of 14 co-existing vascular plant species at a low-Arctic study site in Greenland revealed that differential advancement of spring emergence among the species [104] increased temporal segregation of the early- and late-phenology species from other species [25]. Among species of coexisting plants in a subalpine meadow in Colorado, USA, differential rates of advance of first, peak, and last flowering time have altered the phenological sequence and co-flowering patterns through the season [33]. Similar phenomena are now documented across a broad range of biological systems including butterflies [105], anurans [106,107], vascular plants [33,108,109], and vertebrate herbivores [25]. These cases of temporal shuffling of phenological communities highlight the issue that co-existence in the same space does not necessarily mean co-occurrence. Interaction potentials are as periodic as the occurrence of each species in seasonal systems, and are being perturbed under climate change. One important question that emerges—connected to the broader disciplines of species coexistence and biodiversity research—is how perturbations to multi-phenological systems influence interaction dynamics among species within seasonal time windows, and thus long-term ecological community stability and maintenance of phenological diversity (Box 3).
