Introduction

Recent climate change has seen the advancement of biological events across many taxonomic groups, including flowering in plants (Fitter and Fitter 2002; Menzel and Fabian 1999), egg laying in birds (Crick et al. 1997; Crick and Sparks 1999) and the migration of a wide range of taxonomic groups (Cherry et al. 2013; Marra et al. 2005). Whilst phenological responses to climate change are common, the rates of change can differ across taxonomic groups (Parmesan 2007; Thackeray et al. 2010). Differences in the rate or direction of change between interacting species can lead to phenological mismatches, with potential consequences for the abundance, distribution and extinction risk of the interacting species (Fabina et al. 2010; Memmott et al. 2007; Parmesan 2007; van Asch and Visser 2007). Interacting species must be temporally, as well as spatially, correlated for interactions to occur. Therefore, it is necessary to consider the effects of changing conditions on inter-dependent species, including pollinators and their food plants (Visser and Both 2005), when predicting the impacts of climate change and planning appropriate conservation measures.

Despite the important roles insects play in ecosystem functioning and crop pollination there has been little research on potential phenological mismatches in this taxa, and the wider ecosystem impacts these could have (though see Bartomeus et al. 2011; Kudo and Ida 2013; Rafferty and Ives 2011). In butterflies, earlier and longer flight periods have been recorded across the United Kingdom (UK) and Europe (Diamond et al. 2011; Roy and Sparks 2000; Stefanescu et al. 2003) with evidence of correlations between temperature and appearance date (Diamond et al. 2011; Roy and Sparks 2000). A warming of 1 °C could advance the first and the peak appearances of most British butterfly species by 2–10 days (Roy and Sparks 2000), whilst warming of 3 °C could advance their appearance by as much as 2 or 3 weeks (Sparks and Yates 1997).

Many adult butterflies are dependent on a restricted number of plant species for nectar (Asher et al. 2001). Whilst the flowering periods of many plants have become earlier under climate change (Calinger et al. 2013; Fitter and Fitter 2002; Primack et al. 2004), there is evidence that butterflies′ flight periods are showing more rapid rates of advancement (Gordo and Sanz 2005; Parmesan 2007). Little is known about the physiological mechanisms controlling the phenology of imago emergence in butterflies; however, there is evidence in some species that the climatic cues used by butterflies differ from those used by plants (Doi et al. 2008), which may increase the chance of phenological mismatches. The phenology of sessile organisms such as plants will be affected by microclimate as well as larger scale climate changes. Plant phenologies, therefore, may be especially affected by topography (Pellerin et al. 2012) and, as a result, topographically heterogeneous sites may have longer site-wide flowering periods than homogeneous sites. This heterogeneity may help to minimise phenological mismatches between butterflies and their nectar sources, especially when the caterpillar food plant and the nectar plants for the imago occur in different microhabitats, as can be the case with Melanargia galathea (Linnaeus 1758) (Asher et al. 2001; Stace 1991; Wilson 1985).

Here we examine the flight period of M. galathea, a generalist butterfly, over the period 2001–2013 at three sites across the species’ UK range. We relate the timing of M. galathea’s flight period to the flowering period of one of its main nectar sources, Centaurea scabiosa (Linnaeus 1758) (Asher et al. 2001). In particular, we explore the possibility that M. galathea’s flight period could be advancing more quickly than C. scabiosa’s flowering period, with the potential for a phenological mismatch to occur. Finally, we consider whether the presence of topographical diversity within a site alters the flowering period of this key nectar source, by comparing flowering phenology across multiple aspects at a site on the species’ northern range margin.

Methods

Study species and sites

M. galathea is found primarily on calcareous grasslands; within the UK it is found predominantly in the South, naturally occurring as far north as Yorkshire (Asher et al. 2001; Fox et al. 2006). It has undergone rapid range expansions at its northern range boundary; a survey in 1995–1999 found it present in 66 % more 10 km grid squares than in 1970–1982 (Asher et al. 2001; Fox et al. 2006). C. scabiosa is one of the species’ main nectar sources although Centaurea nigra (Linneaus 1753) and other purple flowers such as Scabiosa spp. are also used (Asher et al. 2001). C. scabiosa is a perennial and flowers between June and September in the UK. It is common within our study sites and across the natural range of M. galathea in the UK. At our northernmost study site, at the species northern range limit, it is the dominant nectar source during M. galathea’s flight period.

Data were collected at three study sites (Fig. 1): Totternhoe Chalk Quarry Reserve, Bedfordshire (Totternhoe; OS grid reference SP9722), Brockadale Nature Reserve, North Yorkshire (Brockadale; OS grid reference SE5117) and Wingate Quarry Local Nature Reserve, Co. Durham (Wingate; OS grid reference NZ3737). Totternhoe and Wingate are the sites of disused limestone quarries, whilst Brockadale is a narrow valley created by the River Went. All three sites have been classified as Sites of Special Scientific Interest (SSSI) by Natural England. As well as being topographically diverse they contain a variety of habitats, including calcareous grasslands where M. galathea is found. Totternhoe is in the core of M. galathea’s UK range, Brockadale is close to M. galathea’s natural northern range boundary and Wingate is beyond M. galathea’s natural northern range limit. The Wingate study population was introduced in 2000 as part of an assisted colonisation experiment (Willis et al. 2009).

Fig. 1
figure 1

Map showing the location of the three study sites, Totternhoe, Brockadale and Wingate (from South to North)

Full size image

Flight period phenology

Data on the abundance of M. galathea adults were collected using transects, following the standard UK Butterfly Monitoring Scheme (UKBMS) guidelines (Pollard and Yates 1993). The transects at each site run through calcareous grassland with diverse topographies. Transect data were collected at Wingate every 2–3 days from a 1,550 m long and 5 m wide transect, for 9 years between 2001 and 2013. Weekly transect data from Totternhoe and Brockadale were collected by the Yorkshire Wildlife Trust and the UKBMS respectively. 8 and 5 years of data were available from Totternhoe and Brockadale respectively.

Flowering phenology

Data on the flowering phenology of C. scabiosa were collected by counting the number of developing (flower buds), flowering (bright inflorescences and pollen present) and spent (withering flowers, and developing or spent seed capsules) flowers on each plant. A minimum of 30 plants were surveyed on ground of varying topography approximately every 3–4 days during M. galathea’s flight period for nine non-consecutive years between 2001 and 2013 at Wingate and in 2011 at Brockadale and Totternhoe. Plants were selected randomly from across the site.

In 2013, additional data on the effect of topography on the flowering phenology of C. scabiosa were collected at Wingate. Four aspects were studied [north east (NE), south east (SE), south west (SW) and south (S)]. Data were also collected on two types of flat area: flat sheltered (FS) sites were in low-lying quarry bottoms and were sheltered on all sides; whereas, flat exposed (FE) sites were on the quarry tops, though still within the nature reserve, and were not sheltered on any side. 30–40 randomly selected individuals were surveyed on each of the six microhabitat types every 3–4 days.

Statistical analysis of butterfly flight period phenology

The expected number of M. galathea individuals seen on a transect on Julian day t was modelled as

$$\mu (t) = y_{\hbox{max} } \exp \left( { - \frac{1}{2}\left( {\frac{{t - \bar{t}}}{\sigma }} \right)^{2} } \right) ,$$
(1)

where the peak abundance (y max) is reached on day \(\bar{t}\), and σ is positively related to the expected duration of the flight period. When fitting this general model to the data we investigated possible year and site effects on adult abundance by comparing fits for versions of the model that either fixed or varied y max, \(\bar{t}\), or σ with respect to year and site, respectively (Table 1).

Table 1 AIC analysis of the models that describe seasonal changes in adult M. galathea abundance among three sites
Full size table

Observed variation about the expected counts was assumed to have a negative-binomial distribution as there was evidence of overdispersion in the data (i.e. the variance was higher than the mean). The probability of observing i butterflies when µ were expected is:

$$\Pr (i\left| {\mu ,\phi } \right.) = \frac{\Gamma (i + a)}{\Gamma (i + 1)\Gamma (a)}\left( {\frac{b}{1 + b}} \right)^{a} \left( {\frac{1}{1 + b}} \right)^{i} ,$$
(2)

where ϕ is a variance parameter, \(a = \mu /\phi\) and \(b = 1/\phi\). We assumed that ϕ was independent of site and year. We performed model selection using AIC (Burnham and Anderson 2002). Models were considered parsimonious if (1) they were associated with an AIC value within 6 of the minimum calculated from the set, and (2) there were no simpler nested models having a lower AIC value. This pair of selection rules prevents overly complex models from being selected when applying AIC (Richards 2008). The flight period was defined as the number of days during which more than one adult butterfly is expected to be observed alive during a transect survey, and it was calculated using the best AIC model (lowest AIC value).

Statistical analysis of flowering phenology

Next, we sought evidence that the timing and length of the flowering period of C. scabiosa differed across sites and years by modelling the expected cumulative proportion of flowers open by Julian day x as

$${\text{logit}}\,p(x) = \beta (x - \bar{x}) ,$$
(3)

where \(\bar{x}\) is the day by which 50 % of the flowers are expected to be open and β is negatively related to the length of the flowering period. We compared the fit of the model that fixed \(\bar{x}\) and β, with the model that allowed these two parameters to differ among years (2001–2013) and across the three sites (Table 2). We also used this general model formulation to determine whether the timing \((\bar{x})\) and length (β) of the flowering period were affected by topography at Wingate in 2013 (Table 3). In this case we compared model fits where \(\bar{x}\) and β were allowed to differ amongst the four aspects and two types of flat area.

Table 2 AIC analysis of the models that describe flowering phenology of C. scabiosa among three sites
Full size table
Table 3 AIC analysis of the models that describe flowering phenology of C. scabiosa among microhabitats at Wingate in 2013
Full size table

For both of the flowering analyses we assumed that the variation in open flowers about the expected proportion was described by a beta-binomial distribution, which accounts for variation among plants caused by unknown covariates. The probability that n of the N flowers on a plant had opened, given that fraction p were expected to have opened is

$$\Pr (N,n\left| {p,\phi } \right.) = \frac{\Gamma (N + 1)\Gamma (a + b)\Gamma (n + a)\Gamma (N - n + b)}{\Gamma (n + 1)\Gamma (N - n + 1)\Gamma (a)\Gamma (b)\Gamma (N + a + b)} ,$$
(4)

where a = p/ϕ and b = (1 − p)/ϕ (Richards 2008). The data suggested that the degree of variation among plants, quantified by ϕ, was similar between sites, years and microhabitats; as a result, this term was kept constant in all of the models. Model selection was again performed using AIC as above. The most parsimonious model (lowest AIC value) was then used to calculate the flowering period, which was defined as the number of days during which between 5 and 95 % of the flowers had opened.

Results

Flight period phenology

Model selection suggested that the day of peak abundance, the peak abundance, and the length of the flight period differed among years and sites (Table 1). Wingate had the latest predicted day of peak abundance and the shortest predicted flight period (Table 4). At Wingate, the start (Linear regression; R 2 < 0.01, F 1,7 < 0.01, p = 0.96) and the length (Linear regression; R 2 < 0.01, F 1,7 = 0.03, p = 0.86) of the flight period have not changed significantly with year. Whilst the model generally appeared to fit the observed data well, this was not the case for all site-year combinations, particularly in 2009 and 2012 in Totternhoe and in 2008 in Brockadale (Table 4).

Table 4 Estimates of peak abundance (y max ) day of peak abundance \((\bar{t})\) and the length (FPL) and start date (FPS) of flight period for different years and sites, according to the best AIC model (Table 1)
Full size table

Flowering phenology

There was evidence that the timing and the length of the flowering period differed among years and between study sites (Table 2). Where data were available from all three sites (2011), the flowering period was earliest at the southernmost site (Totternhoe) and latest at the most northern site (Wingate; Table 5). At Wingate, where multiple years of data were available, over time the start of the flowering period generally shifted to later in the year (Linear regression; β = 2.72, R 2 = 0.52, F 1,7 = 7.68, p = 0.03). There was also a general decrease in the length of the flowering period over time (Linear regression; β = −3.20, R 2 = 0.59, F 1,7 = 9.95, p = 0.02).

Table 5 Julian day by which 5 and 50 % of the flowers are expected to be open by, and the expected length of the flowering period (FP) for each year and site
Full size table

There was evidence that the period between the start of C. scabiosa flowering and the start of M. galathea imago emergence had decreased over time (Fig. 2; Linear regression; β = −2.73, R 2 = 0.59, F 1,7 = 9.96, p = 0.02). There was no significant relationship between the length of C. scabiosa’s flowering period and M. galathea’s flight period (Linear regression; R 2 = 0.12, F 1,7 = 0.93, p = 0.37).

Fig. 2
figure 2

Difference between the start date of M. galathea’s flight period and C. scabiosa’s flowering period, each estimated from the fitted models. Each point represents 1 year’s data at one of the three study sites. The line shows the linear regression fitted to the data

Full size image

There was also evidence that at Wingate the timing, but not the length of the flowering period, varied among microhabitats (Table 3). Flowering was predicted to be earliest on south and south west facing slopes, where it occurred 14 days earlier than on the flat sheltered areas (Fig. 3). The flowering period in each microhabitat was estimated to be 54 days. Thus, the presence of topographic diversity within the site lengthened the site-wide flowering period by around 25 %, to 68 days across all six microhabitats. M. galathea’s flight period was estimated to start at day 187 in Wingate in 2013, which is after C. scabiosa′s flowering period was estimated to start on three of the six microhabitats studied (Fig. 3). The M. galathea flight period was predicted to start on the same day as the C. scabiosa flowering period on flat exposed areas and before the flowering period began on north-east slopes and in flat sheltered areas.

Fig. 3
figure 3

Mean cumulative proportion of flowers on each plant open in different microhabitats at Wingate each day during the flowering period in 2013. The error bars show standard error of the mean. The lines show the predicted number of open flowers according to the best model fit (M3(\(\bar{x}\)); Table 3). The solid horizontal line above the x-axis shows the estimated M. galathea flight period at Wingate in 2013

Full size image

Discussion

Over recent years, the start of M. galathea’s flight period has become much closer to, or even preceded, the onset of flowering of a key nectar source, at least at its northernmost site in the UK. This has resulted in some butterflies emerging before the start of C. scabiosa’s peak flowering period. Whilst the majority of M. galathea’s flight period still overlaps with C. scabiosa’s flowering period, complete mismatches are not necessary for community effects to occur (Fabina et al. 2010). The observed mismatch, if it continues, could have pronounced effects for M. galathea, given that there appears to be a sexual phenological dimorphism, with males emerging several days before females (personal observation by SGW, 2001–2014), probably to maximise mating opportunities (Bulmer 1983; Wiklund and Solbreck 1982). Even slight phenological mismatch in such a situation could result in substantial population declines, if many males in the population fail to find sufficient nectar following emergence. This effect could be particularly acute at the population margin at Wingate, as other key nectar sources either do not exist (e.g. Origanum majorana), occur only in small numbers (e.g. Knautia arvensis, Scabiosa columbaria), or are not yet flowering at the time of M. galathea butterfly emergence (e.g. C. nigra).

The relatively poor model fit in some year and site combinations in the M. galathea model suggests that site and year are not the only factors affecting the abundance of butterflies and the timing of the flight period. Here, we do not link the phenology of either species directly to climate. Further studies are needed to explain why the phenologies of both species have not advanced as may be expected (Fitter and Fitter 2002; Roy et al. 2001), and such studies could also predict how phenological patterns may change under projected rates of climate change (Murphy et al. 2009). However, as the difference between the start of flowering period and flight period has changed over time, it appears that these two species may use different climatic cues, or possibly that one species may not be able to react as quickly to changes in the environment. This suggests that future predicted climate change could lead to further phenological mismatch between the species, probably to the greater detriment of M. galathea.

The increased length of C. scabiosa’s flowering period seen in the topographically diverse Wingate site could help to reduce the chance of a phenological mismatch occurring between these two species as their phenologies change under climate change. The phenology of individuals may be affected by genetics and biotic or abiotic environmental factors. However, here we suggest that the differences in C. scabiosa’s phenology within the site were most likely to be caused by an abiotic factor, namely the aspect and its effect on microclimate, rather than other factors. There is little chance of genetic differences in C. scabiosa between aspects given their proximity to one another. As microclimate will also be affected by habitat type, this could affect phenology (Altermatt 2012); however, as all of the aspects studied here were calcareous grassland, habitat variability is also an unlikely driver of the observed phenological patterns. Topography is likely to affect the phenology of sessile organisms, through changes to the microclimate. It seems likely that these are not the only species for which topographically diverse sites could help to ameliorate the potential for phenological mismatch under climate change, at least in the short-term.

Mismatches between obligate plant–pollinator systems will obviously have severe impacts. In the same way that pollinator biodiversity could buffer the effects of phenological asynchrony in plants (Bartomeus et al. 2013), diversity in potential nectar sources may reduce the effects of phenological asynchrony on butterfly abundance. Predicting whether generalist species will be affected by phenological mismatches requires detailed studies of interactions with all possible interacting species present, rather than simple paired interactions (Bartomeus et al. 2013). Whilst specialist species are therefore more vulnerable to the effects of phenological mismatches, impacts may still be pronounced for more generalist species that experience decreases in dietary breadth, possibly leading to population declines (Memmott et al. 2007).

Consistent phenological shifts within broad taxonomic groups but that differ among taxonomic groups (Parmesan 2007; Thackeray et al. 2010), e.g. insects versus flowering plants, could result from different cueing systems for e.g. growth/emergence/flowering operating. This could result in substantial changes in ecosystems, even among populations of e.g. generalist butterfly species, if flowering is largely synchronous among nectar provisioning plants. The general trend in flowering plants in the UK has been towards earlier flowering (Amano et al. 2010; Fitter and Fitter 2002), rather than the later flowering seen here in C. scabiosa. However, the current study was over a relatively restricted period, during the latter part of which there were several unusually late and wet summers. The decreasing length of flowering period seen in C. scabiosa may also increase the likelihood of a phenological mismatch occurring with M. galathea, and the effect of this may be exacerbated if shorter flowering periods occur in other nectar sources. Although the negative effect of phenological mismatches are likely to be larger for specialist species, which are unable to substitute resources, the impact of phenological mismatching among generalist species is a field that merits future research. The potential for multiple phenological mismatches in taxa such as Lepidoptera that use different plant resources and species at different life stages must render them especially vulnerable. The ability of butterflies to alter their flight period phenology under climate change may be affected by their larval diet (Altermatt 2010).

The impacts of a plant–pollinator phenological mismatch will depend on whether nectar sources are a limiting factor for that population. Despite one study suggesting that there is little evidence that nectar limitation is causing population declines (Thomas et al. 2011), others have found correlations between nectar abundance and butterfly density (e.g. Wallisdevries et al. 2012); nectar abundance may be more important for determining generalist species abundance, whilst specialist species are more dependent on larval host plant abundance (Kraemer et al. 2012). Specifically, in the case of M. galathea, several studies have shown correlations between the abundance of nectar sources such as C. scabiosa and this butterfly species (Lenda and Skorka 2010; Loertscher et al. 1995).

Climate change is expected to significantly impact many species, communities and ecosystems (Parmesan and Yohe 2003), yet studies often focus on single species, without considering the effects of biotic interactions. Phenological mismatches have the potential to disrupt ecosystem functioning, with consequences not only for the species that have become asynchronous (Fabina et al. 2010; Nakazawa and Doi 2012). Whilst biotic interactions are slowly being included into models of climate impacts on species (Araujo and Luoto 2007; Cormont et al. 2013), and in some cases are considered to exceed the importance of climate in affecting distributions (Mason et al. 2014), there has been a lack of research on how the phenologies of insects, and the species they interact with, will be affected. Phenological change may affect the host plant preferences of generalist species, as growth rates on different host plants may be affected by the time of season (Audusseau et al. 2013).

We have demonstrated here how heterogeneous habitats can serve to ameliorate phenological mismatching, for species that can move between habitats/microclimates. We suggest that sites of topographic heterogeneity might be prioritised for conserving similar species under climate change. The diverse microclimates that such sites provide could also buffer populations against the impacts of the extreme climatic events that are forecast to be more frequent in future (Easterling et al. 2000; van Halder et al. 2011; Weiss et al. 1988). For sites with little topographical diversity, management actions that increase the breadth of microclimatic conditions could be encouraged (Hopkins et al. 2007).

As the distributions of the majority of European butterflies appear to be lagging behind climate change (Devictor et al. 2012; Warren et al. 2001), with specialist species particularly vulnerable, conservation strategies are necessary to prevent widespread extinctions. Management strategies such as increasing connectivity between protected areas may help to increase dispersal, whilst assisted colonisations could be used for those species that are unlikely to be able to remain within climatically suitable habitats (Loss et al. 2011; Vitt et al. 2009). Assisted colonisation remains a controversial strategy, with risks of species potentially becoming invasive, and its cost and feasibility questioned (Ricciardi and Simberloff 2009). However, it may be a necessary conservation strategy if rare species with limited dispersal ability are to be conserved (Hoegh-Guldberg et al. 2008; Thomas 2011) and could also be used to restore ecosystem function (Lunt et al. 2013). The M. galathea population at our main study site is an example of a successful translocation, and provides evidence that the distributions of even relatively mobile generalist species may lag behind changes in suitable climate (Willis et al. 2009). Here we emphasise the importance of considering species interactions before carrying out translocations, including the possibility of temporal mismatch. Paired or multispecies assisted colonisations may be necessary for specialist species that require particular host-plants (McLachlan et al. 2007); however, these approaches may also be necessary for more generalist species. At Wingate, where potential nectar sources from other parts of the range of M. galathea are absent, it may be beneficial to M. galathea to introduce such species that might not otherwise reach the site, to minimise the impact of phenological mismatch.

Here we demonstrate how the phenology of a generalist butterfly may lose synchronicity with the flowering period of a key nectar source. However, this asynchrony may be reduced by the presence of topographical diversity. The role of topography in egg-laying, caterpillar growth and imago emergence in M. galathea has not been studied. Spatial preferences of adult butterflies probably bear little relation to their natal area, apart from during egg-laying, and, as such, could mask the role of habitat heterogeneity on phenology. Microhabitat choice is of particular importance in ectothermic species such as butterflies, especially in the less mobile larval stages, and may affect egg hatching and the survival and growth of the larval stages (Bergman 2001; Bryant et al. 2002; Turlure and Van Dyck 2009; Weiss et al. 1988). Studies on caterpillar topographic preferences could further our understanding of the mechanisms explaining differential advancements in phenology of butterflies versus their nectar sources.