Introduction
Phenological shifts, changes the timing of species life history events, have become one of the best-documented signatures of anthropogenic climate change (Forrest & Miller-Rushing, 2010; Parmesan & Yohe, 2003). Many prominent studies of phenology have focused on the timing of reproduction, especially in plants (Parmesan & Yohe, 2003). Some have documented shifts in seasonal migration (Møller et al., 2008). Others have documented changes in emergence from or entry into dormant life stages, including hibernation in vertebrate animals (Inouye, 2000), leaf-out periods for trees (Polgar & Primack, 2011), and diapause in insects (Bale & Howard, 2010). Although the majority of past studies have demonstrated advances in phenology associated with climate warming, there is considerable variation, with some species delaying phenology (Fric et al., 2020; Forister & Shapiro, 2003; Roy & Sparks, 2000).
Despite widespread documentation of changes in phenology, few empirical studies have tested whether phenological changes are associated with long-term trends in population size (Ramula et al., 2015). In terrestrial warming experiments, plants with earlier phenology in warmed plots tended to have higher biomass, growth, and/or reproduction, suggesting that phenological advances represent adaptive responses to environmental change (Cleland et al., 2012). Advanced phenology was also positively associated with long-term population trends of plants in Concord, Massachusetts (Willis et al., 2008) and may be beneficial for insect populations that can avoid competition for resources or feed on a wider range of vegetation (Rathcke & Lacey, 1985). Nevertheless, changing phenology has also been shown to harm species, if changes lead to phenological mismatches with resources or interacting species (Both et al., 2006).
For insects, changes in phenology often arise through changes in the timing of entry into diapause, as well as timing of spring emergence (Bale et al., 2002). Many insect species exhibit geographically variable patterns of voltinism (number of broods within a year) based on location within their range. Under a warming climate, populations which were formerly thermally restricted in parts of their range may be capable of producing an additional generation within the extended growing season (Kozak et al., 2019; Grevstad & Coop, 2015; Mitton & Ferrenberg, 2012; Altermatt, 2010; Tobin et al., 2008) which could be beneficial or detrimental for populations. Species that can successfully add a generation in the extended growing season may benefit from another bout of reproduction, leading to higher overall population growth rates (Kerr et al., 2020; Kerr et al., 2019), and the potential for more rapid evolutionary responses to climate change (Chevin et al., 2010). In spite of its potential benefits, an increase in voltinism can also be detrimental to insects. Mismatched phenological cues could cause a population to start an additional generation that fails to reach its diapausing life stage before frost (Levy et al., 2015; VanDyck et al., 2014). Such developmental traps can potentially cause populations to decline, possibly dramatically, as the flight period increases (VanDyck et al., 2014). Knowing how these different responses affect population dynamics is necessary to understand the consequences of phenological shifts, as well as the longer-term effect of increased growing season length on insect populations.
Butterflies are known to be sensitive to changes in temperature and have been widely used as models to study phenological change (Bale et al., 2002). Phenological studies of butterflies have demonstrated advances in adult emergence, which have been attributed to warming climates (Forister & Shapiro, 2003; Roy & Sparks, 2000). In contrast to numerous examples of phenological advances in the onset of butterfly flight, few studies have investigated potential changes in the end of flight. In the two studies that have investigated empirical patterns of late-season phenology in butterfly communities, Zipf et al. (2017) and Westwood & Blair (2010) both observed later end dates of flight activity, correlated with increasing temperatures. These studies are noteworthy in phenological research because much less is known about late-season phenology than early-season phenology (Gallinat et al., 2015; Karlsson, 2014). Nonetheless, in spite of the large body of work on butterfly phenology, we do not know whether phenological changes are beneficial or deleterious responses to changing environments. One recent study has demonstrated that advanced emergence in British Lepidoptera was associated with significantly higher rates of demographic abundance within multivoltine species (MacGregor et al., 2019). Some studies of birds have also generally demonstrated associations between delayed phenology and decreasing abundance trends (Saino et al., 2011; Møller et al., 2008; Both et al., 2006). Understanding the population dynamics associated with phenology is imperative to translating climate-related phenological changes into their impacts for long-term viability.
One common feature of past research on phenology change is analysis using simple metrics such as first, average, or last observation date. Inferences from such metrics may be limited because first and last observation dates have known biases (Miller-Rushing, 2008) and averages do not capture changes throughout the activity period (Inouye et al., 2019). In this study, we use quantile regression (Cade and Noon, 2003) to evaluate phenological change throughout the activity period. Quantile regression is a statistical modelling technique that enables us to robustly estimate changes in the onset and end of adult flight. Like linear regression, quantile regression generates a slope-intercept line through a specific part of the distribution. Linear regression minimizes squared deviations in the response variable around a trend, whereas quantile regression minimizes the absolute deviations from a trend, subject to the constraint that some proportion of the data be below the line (e.g. the 0.1 quantile, or 10th percentile, fits a line that minimizes absolute deviations with 10% of observations below the line). Quantile regression uses the complete data set to fit this line, and so it differs from fitting a line to the first or last x% of data points, an ad hoc technique that has occasionally been used in ecological studies of phenology (Zipf, et al., 2017; Brooks et al., 2014; Polgar et al., 2013). To date, a handful of studies have used quantile regression to study trends in phenology, mostly in the context of bird migration., These studies have revealed that changes in phenology are not uniform (e.g., Barton & Sandercock, 2018; Gimesi et al., 2012; Gordo et al., 2013). Although under-used, quantile regression represents a formal and robust technique for evaluating changes in phenology throughout the activity period.
Here, we quantify long-term trends in phenology and abundance of butterflies using 27 years of citizen science records from the Massachusetts Butterfly Club (hereafter MBC). Using these observational data for Massachusetts butterflies, we test the relationship between year-to-year changes in flight phenology and abundance. For each species, we estimate abundance trends through time using list length analysis, updating previous analyses of an earlier subset of the same data (Breed et al., 2013), and analyses of counts individuals. We test whether changes in the onset of activity, the end of activity, the average date, and the flight period are associated with increases or declines in abundance. We also evaluate relationships between life history traits, trends in phenology, and trends in abundance using structural equation modelling to elucidate the potential mechanistic pathways. We compare our results to a recent study which documented similar associations between phenology and abundance for British (MacGregor et al., 2019). Finally, we discuss the implications of our findings for insect population viability in their phenological response to global change.