Each life stage matters: the importance of assessing the response to climate change over the complete life cycle in butterflies



  1. As ectothermic organisms, butterflies have widely been used as models to explore the predicted impacts of climate change. However, most studies explore only one life stage; to our best knowledge, none have integrated the impact of temperature on the vital rates of all life stages for a species of conservation concern.
  2. Besides, most population viability analysis models for butterflies are based on yearly population growth rate, precluding the implementation and assessment of important climate change scenarios, where climate change occurs mainly, or differently, during some seasons.
  3. Here, we used a combination of laboratory and field experiments to quantify the impact of temperature on all life stages of a vulnerable glacial relict butterfly. Next, we integrated these impacts into an overall population response using a deterministic periodic matrix model and explored the impact of several climate change scenarios.
  4. Temperature positively affected egg, pre-diapause larva and pupal survival, and the number of eggs laid by a female; only the survival of overwintering larva was negatively affected by an increase in temperature. Despite the positive impact of warming on many life stages, population viability was reduced under all scenarios, with predictions of much shorter times to extinction than under the baseline (current temperature situation) scenario. Indeed, model predictions were the most sensitive to changes in survival of overwintering larva, the only stage negatively affected by warming.
  5. A proper consideration of every stage of the life cycle is important when designing conservation guidelines in the light of climate change. This is in line with the resource-based habitat view, which explicitly refers to the habitat as a collection of resources needed for all life stages of the species. We, therefore, encourage adopting a resource-based habitat view for population viability analysis and development of conservation guidelines for butterflies, and more generally, other organisms. Life stages that are cryptic or difficult to study should not be forsaken as they may be key determinants in the overall response to climate change, as we found with overwintering Boloria eunomia larvae.


Climate change-related conditions are among the biggest threats to biodiversity in the near future (e.g. Thomas et al. 2004), and among frequently studied topics in butterfly ecology (Hellmann 2002; Roy, Brodeur & Cloutier 2003; Wilson et al. 2007). As ectothermic organisms, butterflies are especially sensitive to changes in temperature, one of the leading abiotic factors of climate change (Bale et al. 2002). Different aspects of butterfly life have received attention in the light of climate change: for example, behaviour (Kemp & Krockenberger 2002; Cormont et al. 2010), phenotypic plasticity (Jong et al. 2010), survival (Mercader & Scriber 2008), phenological responses (Altermatt 2010; Singer & Parmesan 2010), range shifts (Wilson et al. 2007) and population viability (McLaughlin et al. 2002; Schtickzelle & Baguette 2004).

Most of the studies that focus on species response to temperature were clearly restricted to a single life stage (e.g. Petersen, Woods & Kingsolver 2000; Nice & Fordyce 2006; Mercader & Scriber 2008; Bjorkman et al. 2011); only a couple of studies concerned more than a single life stage (Hirose et al. 1980; Feeny, Blau & Kareiva 1985; Koda & Nakamura 2010). However, the response to changing environment can differ largely for each life stage; moreover, complex interactions often come into play, because of existing trade-offs in the allocation of resources to different processes at different life stages (Petersen, Woods & Kingsolver 2000; Boggs & Freeman 2005). Our ability to predict how climate change will affect populations in the future, therefore, depends on our understanding of the key effects of temperature at the level of each stage of the life cycle as they will all combine to determine local population demography, and hence population dynamics and viability.

In this context, syntheses of the effects of temperature on the life cycle as a whole are highly desirable for butterflies. Such a holistic approach is in agreement with the resource-based habitat view (Dennis 2010), defining habitat as a collection of resources needed at each stage of the species life cycle. All the resources used by an organism can be split into consumables, which are directly used by an organism (e.g. nectar and host plants for butterflies) and utilities, representing the physical conditions necessary for the organism existence (e.g. microclimatic conditions, roosting and egg-laying sites for butterflies). Despite the concept being widely accepted nowadays, most of the research focused on the description of consumables (Turlure et al. 2009; Merckx & Berwaerts 2010; Sarin & Bergman 2010), ignoring the impact of utilities (but see: Turlure et al. 2010; Choutt et al. 2011). As an integral part of the resource-based habitat, the knowledge on microclimatic requirements of ectothermic species would improve the design of conservation guidelines and enhance management programmes (Suggitt et al. 2011).

The rare studies that consider the impact of temperature on each life stage of insect species belong almost exclusively to life table studies (Bellows, Vandriesche & Elkinton 1992). Pest organisms lead the studies of life table analyses and most research has focused on the sphere of biological control. Thus, of the 73 insect life tables published from 1954 to 2004 reviewed by Peterson et al.(2009), 37% were on forest insects, while the rest concerned species living in agroecosystems. In a similar way, according to Cornell & Hawkings (1995), only seven life tables were constructed with an aim of conservation, while 117 concerned pests and biocontrol agents (e.g. Barron, Barlow & Wratten 2003; Wang et al. 2009; Haye et al. 2010).

In this article, we focus on a vulnerable butterfly species, the bog fritillary butterfly Boloria eunomia (Esper 1799). This glacial relict species, whose current habitat zones in Western Europe (peat bogs and wet meadows) are separated by thousands of kilometres of hostile matrix from the suitable habitats northwards (Fichefet et al. 2008), is likely to show a very limited ability to withstand an increase in temperature expected with global change. B. eunomia's biology and ecology have been widely studied during the last 20 years, and extensive knowledge was accumulated on adult behaviour (Baguette, Convié & Neve 1996; Baguette et al. 1998), dispersal (Schtickzelle, Mennechez & Baguette 2006; Schtickzelle et al. 2007), metapopulation genetics (Vandewoestijne & Baguette 2004) and metapopulation viability (Schtickzelle & Baguette 2004). Nevertheless, immature stages have received far less attention (but see Turlure et al. 2009; Choutt et al. 2011; Turlure et al. 2011).

Here, we investigate the impact of temperature on the survival of every life stage of B. eunomia through controlled laboratory experiments complemented by field studies to measure the effect of temperature in conditions as close as possible to nature, while minimizing the possible negative impacts of manipulation on this vulnerable and legally protected species and its fragile habitat. Then, we combined the effect of temperature on vital rates (survival and fecundity) for each life stage in deterministic periodic life tables (Caswell 2002) to forecast the expected future of the species under various scenarios of climate change.

Materials and methods

Study species and area

Boloria eunomia is a glacial relict species, listed as ‘vulnerable’ in the European Red List (van Swaay, Warren & Lois 2006). It is highly specialized, with both adults and larvae feeding on the same host plant, Persicaria bistorta. The species is univoltine with non-overlapping generations (Fig. 1). In June, females lay clutches of c. 2 to 20 eggs on the underside of host plant leaves or on surrounding plants. Eggs hatch after 12 days on average (Schtickzelle 2003) and caterpillars feed during approximately 3 weeks before entering the diapause as the fourth instar. Around mid-April, caterpillars terminate diapause and feed for a further 4 weeks prior to pupation. The adult flight season lasts on average from end of May to end of June in the study area, the Fange de Pisserotte peat bog nature reserve, located in South-Eastern Belgium (50°13′N 5°4′E).

Figure 1.

(a) Annual life cycle of B. eunomia. Individuals pass through egg, pre-diapause larva, diapausing (overwintering) larva, post-diapause larva, pupa and adult stage within a year. The time-scale (in months) and the timing of experiments are indicated. At any point in time, almost all individuals occur in the same stage. (b) A periodic matrix for B. eunomia baseline scenario. The value in each cell shows the probability to switch from the stage specified in column to the one specified in row; the subdiagonal then contains survival rates, whereas the uppermost row contains fecundity.

Experiments in semi-natural and controlled conditions

Experiment 1. Influence of temperature on the number of eggs laid by a female

To estimate the impact of temperature on the number of eggs laid by females daily, 16 recently emerged B. eunomia females were captured in the field in May 2009. Females were kept individually in flight cages (28 × 28 × 30 cm) in a green house. Every day we supplied a fresh bunch of the host plant with flowers (both for ovipositing and feeding) and a fresh cotton with 10% honey–water solution as an additional food supply (Schtickzelle 2003). Oviposited eggs were counted and collected daily. The temperature in the green house was automatically recorded every half an hour with an electronic temperature recorder (Onset HOBO H08-004-02). Impact of average daily temperature and female's age on the number of eggs laid by a female daily was tested with a generalized linear mixed model (Poisson error distribution, log link, Gauss–Hermite quadrature estimation method with PROC GLIMMIX in sas 9.3 software: http://www.sas.com/) (Bolker et al. 2009). Female identity was included as a random effect to take into account intrinsic differences of fecundity between females because of, for example, genetic or maternal effects.

Experiment 2. Influence of temperature on egg survival (hatching probability)

Thousand two hundred and five eggs (laid by females in experiment 1) were randomly assigned to 3 days temperature conditions: 15 °C (398 eggs), 20 °C (405 eggs) and 27 °C (402 eggs), controlling for the ‘maternal effect’, that is, splitting the eggs laid by a female evenly between the three temperature treatments. Photoperiod was 12L:12D and night temperature was kept around 10 °C for all three conditions, with progressive increase in the morning (from 09:00) and decrease in the evening (from 18:00), mimicking natural conditions (the 25th and 75th percentiles of the night temperature recorded in the field in June and July 2001–2004 + 2009 at 30 cm height, the average height of P. bistorta leaves, were 8·5 °C and 13·1 °C, respectively). Petri dishes with eggs were supplied with a humid piece of cotton, changed every day. Egg hatching was recorded daily. The effect of temperature on hatching probability was tested with a generalized linear mixed model (binomial error distribution, logit link, Gauss–Hermite quadrature estimation method with PROC GLIMMIX). Female identity was added as a random effect to account for variation in egg quality, and hence in survival probability, according to female body condition.

Experiment 3. Influence of temperature on pre-diapause larva survival

Four hundred and fifteen larvae (obtained from the eggs in experiment 2) were split into three groups and assigned to the same three temperature regimes already used in experiment 2 (15 °C: 159 larvae, 20 °C: 96 larvae and 27 °C: 160 larvae), controlling for the temperature during egg development, to create nine combinations of egg*larvae temperatures. Every 3 days, we cleaned the Petri dishes, replaced the P. bistorta leaves and the humid cotton and recorded dead and alive larvae. The effect of temperature during both egg and larval development and their interaction on larval survival was tested with a generalized linear model (binomial error distribution, logit link with PROC GENMOD).

Experiment 4. Influence of tussock microclimate on the survival over larvae (end of the pre-diapause and overwintering) and pupa stages

Tussocks were shown to create a special microclimate impacting the behaviour and thermoregulation of B. eunomia larvae (Turlure et al. 2011). Therefore, we hypothesized that tussocks of different height would offer different microclimates (in terms of temperature but not exclusively: Turlure et al. 2011) and hence impact the survival of larvae. To check this hypothesis, we installed eight round cages (S = 0·36 m2) into the natural B. eunomia habitat, at four places with a big (>20 cm high) tussock and four places with a small (<20 cm high) tussock. 436 third-instar (on average 23 days old) larvae (from experiment 2) were gently deposited on bistort leaves in the cages (57 larvae were put in seven cages and 37 larvae in the last cage, as it contained lower number of P. bistorta leaves), controlling for the temperature they experienced during development (experiment 2). The effect of tussock height on the survival rate over larvae and pupa stages was tested with a generalized linear model (binomial error distribution, logit link with PROC GENMOD).

Experiment 5. Influence of tussock microclimate on pupal survival in the field

In order to estimate the impact of different microclimatic conditions on pupal survival and disentangle pupal survival from the larvae-pupa survival (estimated in experiment 4), the same cages described in experiment 4 were used for a pupal survival experiment. In the end of April and beginning of May 2009, last instars larvae were collected in the field and brought to the laboratory, where they were kept singly in Petri dishes with bistort leaves and humid cotton (changed every 2 days) until pupation (20 °C during the day, 10 °C at night, 12L:12D photoperiod). Nine pupae (3 days old on average) were put into each cage in the field. The cages were kept open for 10 days to allow for natural mortality from predators and/or parasitoids and closed with a mesh on the 11th day. After 12 days, the cages were checked every 2 days, and emerged adults marked with a unique number and released. The effect of tussock height on pupa survival rate was tested with a generalized linear model (binomial error distribution, logit link with PROC GENMOD).

Experiment 6. Influence of temperature on pupa survival in laboratory conditions

According to experiment 5, survival of pupa was not significantly different under the two microclimates created by tussocks of different sizes (see 'Results'). To verify whether there is indeed no temperature impact on pupa survival, we conducted a controlled laboratory experiment with three fixed temperature treatments in spring 2010. Twenty-six last instars larvae were collected in the field in April–May 2010 and kept in the laboratory until pupation (20 °C during the day, 10 °C at night, 12L:12D photoperiod), with bistort leaves and humid cotton renewed every 2 days. Nine pupae were placed in the 14 °C and 22 °C temperature regimes, and eight pupae in the 18 °C temperature regime (all regimes with temperature stable during the day and decreasing to around 10 °C during the night, as in experiments 2 and 3, with 12L:12D photoperiod). Every 2 days pupae were supplied with fresh humid cotton and checked for adult emergence. Survival of pupae as a function of temperature was tested using a general linearized model (binomial error distribution, logit link with PROC GENMOD).

Influence of temperature on overall population growth rate

To assess how the overall B. eunomia population may be impacted by temperatures that are currently (or will likely be in the future) experienced by the individuals in different life stages in the field, we used a periodic deterministic matrix model (Caswell 2002). A periodic matrix model is a special modification of population matrix models used for annual life cycle organisms with non-overlapping stages occurring within a year, such as univoltine butterflies like B. eunomia (Fig. 1). Each within-year stage is called a phase and the change in the population size (n) after a year is calculated as following:

display math(eqn 1)

with m the number of phases distinguished within the yearly life cycle, B1Bm a set of matrices with Bm matrix projecting the population from phase m−1 to m, nt the population size at time t, and Am the product of matrices BmB1 projecting the population through a complete life cycle. Projection matrices can be directly analysed using matrix algebra to derive the finite rate of population increase (the dominant right eigenvalue of the matrix: Caswell 2002).

A baseline and three alternative scenarios were compared (see Table 1 for details on vital rate estimates for each scenario):

  1. The baseline scenario, corresponding to a matrix of vital rates as measured under the temperature regime the species is currently experiencing in the field (medium temperature treatment for all stages).
  2. The temperature increase scenario, mimicking simple and direct climate warming; vital rates were those obtained from the high-temperature treatment for all stages.
  3. The European climate change scenario, where winter and spring temperatures are more likely to increase than summer ones (Christensen & Newitson 2007); vital rates were those obtained under high temperature for pupa and overwintering larva stages, and medium temperature for other stages.
  4. The increased temperature variability scenario, corresponding to a combination of high and low temperatures for different stages, as a proxy for the increased variation in temperature and occurrence of extreme events that is often predicted to accompany climate change (Schar et al. 2004). Four sub-scenarios were designed: 4a. Low temperature during May (pupa survival) and June (egg laying and egg survival), and high temperature during July (pre-diapause larva survival) and October–March (overwintering larva survival) months; 4b. High temperature during May and June and low temperature during July and October–March; 4c. High temperature during May and July and low temperature during June and October–March; 4d. Low temperature during May and July and high temperature during June and October–March (Table 1). Every year we randomly choose one of the four sub-scenarios describing possible temperature combinations within a year and used its corresponding finite rate of increase to project the population size into the next year.
Table 1. The impact on population dynamics of four realistic climatic scenarios, as estimated with a deterministic periodic matrix model (see text for details). For each scenario are shown: the value of the five vital rates (cf matrix in Fig. 1), the resulting finite rate of population increase λ and its value when rescaled to obtain a match between the baseline scenario and the population growth rate over 8 years (see text for details), and the time to extinction. Shading corresponds to the level of temperature treatment: light for low, semi-dark for medium and dark for high temperature. The ‘increased temperature variability’ scenario is based on the four sub-scenarios with temperature varying among different life stages; each year the finite rate of increase is chosen randomly among the four finite rates of increase for these sub-scenarios to predict the fate of the populationThumbnail image of

Deterministic matrix models for all scenarios were initialized with the size of the B. eunomia population in Pisserotte study site (1436 individuals), as estimated from capture–mark–recapture (CMR) data in 2004 according to the method described in Schtickzelle, Le Boulengé & Baguette (2002). We considered a population extinct when the population size dropped to two individuals. Finite rate of increase obtained with the baseline scenario (0·608) was lower than the geometric mean of population growth rates (0·97) observed in the study system during 8 years (2004–2011: N. Schtickzelle & C. Turlure, unpublished data), meaning some of the survival rates were likely underestimated because of manipulations (see 'Discussion'). To avoid incorporating this experimental error into the model, we rescaled the finite rates of increase for all scenarios to obtain a match between the baseline scenario and the observed time series; this was carried out by multiplying them with a 1·595 rescaling factor, result of the division of the finite rate of increase from the baseline scenario by the geometric mean of growth rates from the field data.

We report two population viability measures: (i) the finite rate of increase and (ii) the time to extinction, which is used to quantify the future population size and is more simple to conceptualize than the finite rate of increase.

Additionally, we assessed the sensitivity of the finite rate of increase to the changes in every vital rate, using the analytical formula for periodic matrix models (Caswell 2002):

display math(eqn 2)

with math formula the sensitivity matrix of the elements of matrix Bm projecting the population from phase m−1 to m, DT = (Bm-1Bm-2Bm+1)T a transpose of the product of phase-specific matrices excluding math formula the sensitivity matrix of the matrix Am (a product of all phase-specific matrices), calculated as math formula , with v and w, respectively, the left and the right eigenvectors of Am.


Estimation of temperature effect on vital rates

Experiment 1. Number of eggs laid by a female

The daily number of eggs laid by a female was impacted by both average daily temperature and female age (Table 2) with existing optima. The daily number of eggs laid by an average female (Neggs) according to the best selected model was given by:

display math(eqn 3)
Table 2. Testing for the effect of average daily temperature, female age and female identity on the number of eggs laid (experiment 1) using generalized linear mixed models (Poisson error distribution, log link) with AICc model selection (Burnham & Anderson 2002). Female identity was used as a covariate in all models and was included as a random effect. Temperature and age variables were included as fixed effects. Both linear and quadratic effects of temperature and age variables were tested. To take into account the overdispersion math formula present in the data, we used QAICc (Burnham & Anderson 2002); this non-negligible overdispersion is likely due to daily variation in egg laying that could not be explained by female age and daily temperature. For each model, we present the list of variables in the model, the number of parameters, the QAICc value and the QAICc difference (Δ) with the lowest QAICc model. Supported models (ΔQAICc < 2) are shown in bold
ModelNumber of parametersQAICcΔQAICc
age + age2 + T + TFemale 6 138·2 0·0
age + age2 + T + Female 5 138·6 0·4
T + Female3150·412·2
age + T + T2 + Female5150·412·3
age + T + Female4151·813·7
age + Female3151·913·7

With age the female's age in days, and T the daily average temperature. The standard deviation (SD) between females was 0·83. The maximum number of eggs was laid in the 25–31 °C temperature range by females 3–7 days old (Fig. 2).

Figure 2.

The number of eggs laid daily by an average female as a function of its age and daily temperature.

Experiment 2. Egg survival

Overall, 874 larvae hatched from 1205 eggs, corresponding to an average (±SEM) hatching rate of 0·750 ± 0·012 for an average female. The SD between females was 0·69. Egg survival (hatching probability) increased with temperature and differed according to the female identity (Table 3; Fig. 3a).

Figure 3.

Vital rates (mean and 95% confidence interval) measured for B. eunomia life stages. Survival of: (a) eggs (experiment 2), (b) pre-diapause larvae (experiment 3; temperature treatment during the egg stage is also shown), (c) larvae and pupae (experiment 4), (d) pupae in field (experiment 5) and (e) pupae in laboratory (experiment 6). (f) Number of eggs laid under 5th (low), medium, and 95th (high) percentiles of daily maximum temperature in June, as predicted using eqns 5 and 6.

Table 3. Modelling the survival of immature stages of B. eunomia using AICc model selection. Generalized linear mixed models (binomial error distribution, logit link) were applied to test the egg survival; generalized linear models (binomial error distribution, logit link) were applied to test the survival of other stages. Temperature was used as a categorical variable (low, medium, high). Female identity was used as a covariate and was included as a random effect in the egg survival experiment only. For the pre-diapause stage, the effects of temperatures during the larva (Tlarva) and egg (Tegg) stages and their interaction were tested. Tussock size was used as a categorical variable in the field experiments on overwintering larva and pupa stages. For each model are shown the list of variables in the model, its number of parameters, AICc value and AICc difference (Δ) with the lowest AICc model. Supported models (ΔAICc < 2) are shown in bold
StageModelNumber parametersAICcΔAICc
Egg (experiment 2) T + Female 4 1346·4 0·0
Pre-diapause larva (experiment 3) Tlarva + Tegg + Tlarva*Tegg 9 506·4 0·0
Tlarva + Tegg5514·68·3
Intercept only1535·328·9
Overwintering larva and pupa (experiment 4) Tussock size 2 64·8 0·0
Intercept only 1 65·4 0·6
Pupa (field, experiment 5) Tussock size 2 102·3 0·0
Intercept only 1 102·6 0·3
Pupa (laboratory, experiment 6) Intercept only 1 34·3 0·0
Tpupa 3 34·7 0·5

Experiment 3. Pre-diapause larva survival

Of 415 first-instar larvae, 273 survived until the third instar, corresponding to an average (±SEM) survival rate of 0·658 ± 0·023 (Fig. 3b). Survival of pre-diapause larva was impacted by temperatures both during the egg and larva development, in an interactive way (Table 3; Fig. 3b): survival was generally higher at higher temperature during larva stage; however, larvae that experienced a high temperature at the egg stage showed a high survival at low temperature during the larval stage.

Experiment 4. Survival over the larva (end of pre-diapause and overwintering) and pupa stages

There was a great variability in the number of adults emerging from the cages, with an average survival rate over both overwintering larval and pupa stages of 0·014 (range among cages: 0–0·035). The mean survival rate over overwintering larva stage only was 0·025 (obtained by dividing the aforementioned value by the average pupa survival rate: 0·549). The overwintering larvae had a five times higher probability to survive until the adult stage in the big tussocks (0·022) than in the small tussocks (0·0048) (Fig. 3c). Because of the large variability between replicate cages, the models with and without an effect of tussock size received very similar support (Table 3).

Experiments 5 and 6. Pupa survival

The mean (±SEM) survival rate of pupa in the field (experiment 5) was 0·549 ± 0·057. It was higher in the cages with big tussocks (0·639 ± 0·078) than in the cages with small tussocks 0·459 ± 0·079 (Fig. 3d). However, the models with and without the effect of tussock size received very similar support because of the large variability (0·33–0·99) between replicate cages (Table 3).

The mean (±SEM) survival rate of pupa in the laboratory (experiment 6) was 0·692 ± 0·088 (Fig. 3e). It showed some increase with temperature, but the models with and without an effect of temperature received once again very similar support (Table 3).

Prediction of population growth rate

The baseline scenario was characterized by the highest finite rate of population increase (Table 1). If the temperature regime is changed, viability of the species is predicted to decline; this is true whatever the change but is clearly more pronounced for some scenarios than for others (Table 1): 88% viability reduction (in terms of shorter time to extinction compared to the baseline scenario) for the increased temperature variability scenario; 94% viability reduction for the temperature increase; and 97% viability reduction for the European climate change scenario.

The finite rate of increase was the most sensitive to changes in the survival of overwintering larva (from the end of pre-diapause and on) and virtually insensitive to changes in adult fecundity (Table 4).

Table 4. Sensitivity of the finite rate of increase to the change in each vital rate as calculated using the analytical solution for matrix models (see text for details). Finite rate of increase was the most sensitive to the survival of larvae from the end of pre-diapause and on (highlighted in bold)
Vital rateSensitivity
Egg survival0·817
Pre-diapause larva survival1·242
Overwintering larva survival 24·32
Pupa survival0·811
Adult fecundity0·007


The life stages of the endangered B. eunomia butterfly were all impacted by temperature but positively for egg survival, pupa survival and fecundity, and negatively for overwintering larva survival. Altogether, population viability was predicted to be highest under temperature conditions currently experienced (baseline scenario) and drastically reduced (>88%) under three climate change scenarios implying some global warming, especially under the European climate change scenario, which is forecasted as very likely for Europe according to the IPCC report (Christensen & Newitson 2007). This is in line with the expectations for such a glacial relict species, restricted to relatively cool and humid biotopes (peat bogs and wet meadows: Turlure et al. 2009), and already declining in Belgium (Fichefet et al. 2008).

Even if most of the vital rates increased in warmer conditions, the overall population response was negative because of the high sensitivity of population viability to survival during winter diapause. Warmer winter temperatures lead to an elevated mortality of overwintering larvae (Schtickzelle & Baguette 2004; N. Schtickzelle & C. Turlure, unpubl. data), probably due to increased incidence of diseases and/or fungal infections, both favoured by mild winter (Harcourt 1966; Dempster 1984). This illustrates the necessity to integrate the response over the whole life cycle to make meaningful predictions of the climate change effects. Predictions based on one stage only would likely and mistakenly indicate that the species is not facing any direct risk from global warming. This might be the case for other studies. For example, Bjorkman et al. (2011) use the relation between fecundity and temperature derived from laboratory experiment to parameterize a heritability population dynamics model for a leaf beetle. They conclude that an increase in frequency of warm summers will lead to higher probability of population outbreaks. However, these model predictions were based on female fecundity only, ignoring the other vital rates.

Overwintering larva was the stage to which model predictions were the most sensitive. Unfortunately, it is also the least studied one because of technical difficulties. On the one hand, it is virtually impossible to study overwintering larvae in the field because they bury themselves in the litter/upper soil layer for diapause. On the other hand, all attempts to establish laboratory rearing of the species have so far failed, particularly because of the high incidence of fungal infection (A. Joiris and J. Choutt, personal communication); this difficulty also concerns some North American fritillary butterflies (A. Shapiro, personal communication). In the framework of this study, we once more undertook a laboratory rearing experiment; however, none of the 225 larvae from experiment 3 put for diapause in late summer 2009 survived. This emphasizes the necessary complementarity of laboratory studies and field work, for only the long-term demographic monitoring of the species in the field (CMR over 20 generations: 1992–2011, Schtickzelle, Turlure & Baguette 2012) allowed an estimation of the impact of temperature on overwintering larvae, which, in turn, made possible the viability predictions on the overall population.

As the survival of overwintering larva was the most sensitive of all vital rates, this makes it a primary target for conservation management. The management efforts therefore should favour development of tussock structures, as they were previously demonstrated to be used by caterpillars for behavioural thermoregulation (Turlure et al. 2011). In the Pisserotte study site, this would require removing the trees in close vicinity of the peat bog to allow increasing the level of water table, thereby promoting the development of Deschampsia cespitosa tussocks. Other conservation studies also demonstrated the necessity to preserve larval habitat in addition to adult's (Euphydryas aurinia: Anthes et al. 2003; Konvicka, Hula & Fric 2003; Maculinea arion: Thomas, Simcox & Clarke 2009; Icaricia icarioides fenderi: Schultz 2001), underlining the usefulness of adopting a resource-based habitat approach for applied species management.

We chose deterministic matrix model as a straightforward way to integrate the impact of temperature over the whole life cycle. It yielded predictions about increased extinction risk owing to climate change that matched our expectations for this glacial species and underlined the importance of survival of overwintering larvae. However, this simple modelling approach (Beissinger & Westphal 1998) ignores any source of stochasticity encountered in real systems. Moreover, we did not consider the indirect effects of climate change, which will likely also affect other biotic (e.g. B. eunomia parasitoids) or abiotic factors (e.g. B. eunomia habitat). Additionally, the ability of the species to adapt to warmer conditions is unknown. All these will in turn affect population dynamics of B. eunomia in a way that is currently difficult to predict, because synergetic and/or buffer interactions are likely.

The estimated finite rate of increase in the baseline scenario (0·608) was lower than the geometric mean of growth rates observed in the studied system during the last eight years (0·97), which could signal underestimated survival rates in the laboratory compared to the wild. Plausible reasons for this could be usage of fixed temperature treatments that limit the ability of larvae to move and escape unfavourable conditions by behavioural thermoregulation (Casey 1976; Nice & Fordyce 2006; Turlure et al. 2011), or manipulation of individuals. This brings another insight: laboratory experiments are often likely to estimate the relative impact of a factor on vital rates more correctly than absolute values of these vital rates. Independent estimates of the absolute value of vital rates are needed to rescale estimates obtained in the laboratory if one wants to use them to make predictions. However, the rather simplistic rescaling approach we used here assumed, in the absence of further information, that the error in estimation of all vital rates was the same, despite the survival estimates of some life stages might be more affected by the manipulation than others.

Given the nature of the model, the rescaling approach, and more generally the inevitable uncertainties in model parameters, predictions shown here must therefore not be taken as reliable predictions of time to extinction. Translating absolute predictions to the field situation, for example, for direct conservation management, must generally be cautious, as it is always the case for PVA results; however, the predictions can be compared between the scenarios (Beissinger & Westphal 1998; Reed et al. 2002).

Scenarios incorporating increase in temperature variability pose rather high threat to population viability (Table 1). The conditions simulated in this increased temperature variability scenario, implying increased weather variability and higher incidence of extreme events, are predicted to occur in the future (Schar et al. 2004; Scherrer et al. 2005). Nevertheless, PVA models constructed for butterflies so far mostly tested for changes in mean temperature (Schtickzelle & Baguette 2004: B. eunomia, Schtickzelle, WallisDeVries & Baguette 2005: B. aquilonaris, but see Saltz, Rubenstein & White 2006 on Asiatic wild ass, Hunter et al. 2010 on polar bears, Nicolè et al. 2011 on Dracocephalum austriacum). Here we demonstrated how even such a simplistic matrix model can be used to explore the plausible effects of increased within- and between-yearly temperature variability on butterfly population dynamics.

To conclude, when predicting the population's future and designing conservation measures in a resource-based habitat framework, we stress the need of estimating and integrating the response of all life stages to the driving forces of climate change, such as temperature. Cryptic or difficult to study life stages, in particular, should not be forsaken as they may be key determinants in the overall reaction to climate change. So far, most of the PVAs on butterflies use a yearly population growth rate to project the population in the future (Schtickzelle & Baguette 2009 and references therein, but see Cormont 2011; Harrison, Hanski & Ovaskainen 2011). They do neither permit predictions for scenarios implying temperature change during specific months, such as the European climate change scenario (Christensen & Newitson 2007), nor target conservation measures towards the most sensitive life stage.


We would like to thank Eleonore Danhier for her help with laboratory experiments and Kate Mitchell for language improvements of the manuscript. V.R. was supported by a Ph.D. grant from the FRIA-fund. C.T. is Postdoctoral Researcher and N.S. is Research Associate from the Fund for Scientific Research-FNRS and they both acknowledge its financial support. This study is contribution BRC275 of the Biodiversity Research centre at UCL.