The influence of thermal ecology on the distribution of three nymphalid butterflies


  • Simon R. Bryant,

    1. School of Biosciences, University of Birmingham, Edgbaston, Birmingham B15 2TT, UK; and
    Search for more papers by this author
    • *

      Present address and correspondence: Dr S.R. Bryant, School of Biological and Molecular Sciences, Oxford Brookes University, Gipsy Lane Campus, Headington, Oxford OX3 0BP, UK (fax + 44 1865 483242; e-mail

  • Chris D. Thomas,

    1. Centre for Biodiversity and Conservation, School of Biology, University of Leeds, Leeds LS2 9JT, UK
    Search for more papers by this author
  • Jeffrey S. Bale

    1. Centre for Biodiversity and Conservation, School of Biology, University of Leeds, Leeds LS2 9JT, UK
    Search for more papers by this author


  • 1Studies have shown that many adult and immature insects are able to maintain body temperature well above, and often independently of, ambient temperature in the presence of direct solar radiation. They may do so directly by basking, or indirectly via microhabitat choice. The implications for this are often ignored in development models that serve to predict species’ responses to climate change.
  • 2To investigate the difference in development times attributable to solar input, field development of Aglais urticae, Inachis io and Polygonia c-album larvae (Lepidoptera: Nymphalidae) was followed both in a natural open situation and in an artificially shaded environment. Each species completed development more quickly in the open, equivalent to a 20%, 10% and 15% reduction in development time, respectively.
  • 3Observed development times in the artificially shaded environment were used to compare the techniques of rate summation and degree-day modelling. Rate summation models were found to describe development in the shade best for A. urticae and I. io, although a degree-day model performed best for P. c-album, possible reasons for which are discussed. The best-performing models for each species were modified to include larval thermoregulation data, assuming a linear relationship between body and ambient temperatures during the measured sunshine hours each day, and tested against observed development times in the open situation. Times for 50% adult emergence were predicted exactly for A. urticae, and to an accuracy of 1 day for I. io and 5 days for P. c-album.
  • 4The models were tested further using climate data from 128 UK Meteorological Office weather stations across England, Wales and Scotland. Thermoregulation model predictions matched observed UK distribution and voltinism better than predictions made by the standard unmodified models. It was estimated that larval thermoregulation allows A. urticae and I. io populations to persist approximately 200 km further north than would otherwise be possible, and that the extent of bivoltinism may be shifted northwards by around 300 km.
  • 5These results have significant implications for predicting the effects of global warming on insects’ geographical ranges, the potential distributions of invasive species, and the phenology and voltinism of introduced biocontrol agents as one component of their likely success.


The warming effects of direct radiation from the sun are used by most insects. Both adults and immature stages absorb solar radiation directly by basking, or benefit indirectly via the choice of suitable microhabitats (May 1979). Thermoregulatory behaviours have evolved such that many species are able to achieve and maintain specific body temperatures, which often coincide with the optimum of one or more physiologically important processes, such as feeding rate (Sherman & Watt 1973; Lactin & Johnson 1995), feeding efficiency (Porter 1982), growth rate (Knapp & Casey 1986), development rate (Lactin & Johnson 1996; Bryant, Thomas & Bale 2000) and metabolic efficiency (Kukal, Heinrich & Duman 1988).

Physiological time in developing insects is temperature dependent (Taylor 1981): the higher the temperature, the quicker development proceeds (up to a threshold above which increased temperature soon becomes fatal). This is the basis behind most models for estimating emergence times for pest species (Liu & Meng 1999), understanding species’ phenological patterns (Malcolm, Cockrell & Brower 1987) and predicting shifts in species’ ranges in response to climate change (Porter, Parry & Carter 1991). Most studies use air temperature as the only environmental input to drive the models (Got, Labatte & Piry 1996; Turnock & Boivin 1997; Harari et al. 1998; Liu & Meng 1999), although it is acknowledged that the discrepancy between ambient temperature and the temperature insects actually experience is the biggest source of error in such models (Pruess 1983; Higley, Pedigo & Ostlie 1986). To counter such error in model performance, start dates for models are chosen retrospectively based on observed emergence patterns (Mcbrien & Judd 1998), threshold temperatures are varied to find the best prediction (Davis, Brenes & Allee 1996; Fatzinger & Dixon 1996), or developmental requirements are calculated solely from field observations (Beasley & Adams 1996). Such models may indirectly take microclimate into account, but only on a local scale and without identifying the separate ambient and microclimatic components. These modifications improve the statistical fit, but not necessarily the biological understanding. None the less, in some cases thermoregulation by the species in question has been taken into consideration, either directly (Rawlins & Lederhouse 1981; Carruthers et al. 1992) or indirectly (McDonald & Smith 1988). Microclimate measurements have also been incorporated in some studies (Lyons 1994).

Increased biological understanding is likely to be required if thermally based predictions are to be made for climate scenarios outside our current experience, and there is increasing interest in the response of animal distributions to various climate change scenarios (Porter, Parry & Carter 1991; Drake 1994). In this context, the relevance of macroclimatic measures such as air temperature (often averaged over large areas) to the microclimatic temperatures experienced by insects is questionable (Kennedy 1997). In this study, we tested whether incorporation of more detailed aspects of the thermal biology of three insect species allowed us to predict species’ phenology and distribution accurately. If so, the approach should have general applicability in understanding and predicting the responses of poikilothermic animals to climate change. The study species were the small tortoiseshell Aglais urticae L., peacock Inachis io L. and comma Polygonia c-album L. butterflies (all Lepidoptera: Nymphalidae). These species are medium-sized mobile butterfly species that share the same larval resource, the common stinging nettle Urtica dioica L. (Urticaceae), itself widespread and generally abundant throughout the UK and mainland Europe (Rose 1981). Interestingly, they differ with respect to their thermal ecology. Female P. c-album lay eggs singly while those of A. urticae and I. io lay in large batches, of about 80 and 300–400 eggs, respectively (Dennis 1984). The solitary, cryptically coloured, larvae of P. c-album spend most of their time concealed on the underside of U. dioica leaves, while the gregarious, mainly black, larvae of A. urticae and I. io live within loose silken webs, fully exposed and often near the top of U. dioica plants. The body temperature of P. c-album larvae thus conforms to the surrounding (microhabitat) temperature, but the temperature of A. urticae and I. io larvae is elevated almost independently of ambient in the presence of sunshine, being maintained at around 32·5 °C and 31·5 °C, respectively (Bryant, Thomas & Bale 2000). We explored how these differences affect the butterflies’ phenologies and distributions.

Materials and methods


The field experiment was carried out in spring and summer 1996. Two sites, of approximately 24 m2, were chosen that were about 30 m apart, one in an open situation, the other in a more shaded area beneath several trees. A white tarpaulin was tied between these trees at about 1 m from the ground, to allow air circulation, and held about 2·5 m above the ground in the central area with a series of metal poles to form a tent, completely shading a large enough area to house the required number of cages. Each cage was constructed from fine black netting bags, placed over four garden canes hammered into the ground in a formation 50 × 50 cm, and 100 cm high. The base of the netting was dug into the ground. There was an opening in one side of each cage to allow exchange of old and new food plants, held shut when not in use with clips. There were 15 cages in each ‘treatment’, five for each species: three rows of five cages were arranged under the tent, and two rows of seven and eight cages, each cage 1·5 m apart from its neighbours, were arranged in the open area.

Females of each species were caught around the campus of the University of Birmingham (grid reference SP0483), Birmingham, UK soon after the local populations started emerging from their overwintering sites. Eggs were obtained and kept at 20 °C, 18 light : 6 dark, until they hatched. On hatching, groups of 50 A. urticae and I. io, and 20–40 individual P. c-album, larvae were placed in each cage on potted U. dioica plants. These plants originated from one large nettle bed, and consisted of spring foliage similar to that chosen by ovipositing females in the wild. For the gregarious species, larvae were placed on a terminal shoot near the top of the plant, a position where they most often occur naturally as first instar larvae. Polygonia c-album larvae were distributed over the whole plant one to a leaf, each one being placed on the underside. The start dates for the cohorts were 2 May (P. c-album), 3 May (A. urticae) and 5 May (I. io). The spatial arrangement of each group within each treatment was randomized. All groups were monitored daily, and the stage of development was recorded. New potted plants were introduced when necessary but, as larvae became larger and consumed more, cut nettles placed in water were used to satisfy the increased demand.

Field temperatures in each treatment were recorded at 15-min intervals using a data logger (Tinytalk® IP68, RS components 219–670, internal 10 k NTC thermistor, RS Components Ltd, Corby, Northants, UK) housed in a plastic pipe covered in aluminium foil, held with a clamp stand at a height of 50 cm. This height was deemed to be representative of the starting position of the larvae on the host plant within the cages and therefore the recorded air temperature was assumed to be representative of the microhabitat temperature experienced by the larvae.


The temperature–rate relationship of insect development is usually described by fitting a function to a series of mean rates recorded under controlled temperature regimes in the laboratory. This function may be linear or non-linear, and the two corresponding modelling approaches are compared here. Linear regression of rate on temperature yields a theoretical developmental threshold temperature (TH), where development rate is zero, and a measure of the mean number of thermal units (degree-days, DD) above the threshold required for the population to complete development (Campbell et al. 1974; Lamb 1992). Development is commonly observed to occur below the linear developmental threshold, however, and non-linear functions take into account the curvilinearity observed at low (and high) temperatures, thus providing a more complete description of development and removing the need for a threshold (Logan et al. 1976; Sharpe & DeMichele 1977; Lactin et al. 1995). The time of mean population development under field temperature conditions is calculated by summing fractions of development rate to unity (and is known as rate summation). In this study we used the non-linear function of Lactin et al. (1995), which has the form:

image(eqn 1)

where r(T) is development rate at temperature T, and rho (ρ), Tmax, delta (Δ) and lambda (λ) are fitted parameters.

Variation in development times between individuals within a population is built into the model with the use of cumulative probability distributions. The distribution of development times at each experimental temperature is assumed to be of similar shape, and thus one standard temperature-independent curve describing the pattern of development completion for any stage or stages of the insect can be constructed. We used the Weibull distribution function, which has the form:

image(eqn 2)

where F(x) is the probability (or proportion) of complete development at normalized time x, and parameters gamma (γ), eta (η) and beta (β) are estimated by non-linear regression (Wagner et al. 1984).

The modelling approaches used in this study are similar to those adopted by previous authors, described in detail by Wagner et al. (1984, 1985). Development of a theoretical population is followed through time, driven by the integration of the chosen description of the species’ temperature-dependent development with field temperature data. A temperature-independent description of emergence patterns is used here that takes into account individual variation in development times within the population. Generally, most models attempt to predict the activity of an insect population rather than abundance, and thus usually comprise a fixed number of individuals to allow calculation of percentage emergence. All model populations described in this paper use 100 individuals, each one representing the cumulative percentage development of the population by association. Model parameters were from laboratory experiments detailed in Bryant (1998) and Bryant, Thomas & Bale (2000).

Both degree-day and rate summation models (see Table 1a,b for model parameters) were tested against observed development of larvae in the shaded experiment first, in order to determine the best-performing model for each species in terms of the ambient microclimate temperature component (i.e. other possible temperature effects were excluded). The temperatures used to drive the models were those obtained from the data logger placed amongst the shaded cages.

Table 1.  Summary of (a) degree-day and rate summation development parameters, (b) Weibull distribution estimates and (c) linear larval thermoregulation parameters required for each development model tested. See text for description of parameters. Data from Bryant (1998) and Bryant, Thomas & Bale (2000)
(a)Degree-dayRate summation
Aglais urticaeLarvae205·610·30·00406644·927561·79598−1·03458
 Pupae 96·111·30·00808736·321600·79925−1·08201
Inachis ioLarvae315·2 8·30·00300434·328200·17565−1·02253
Polygonia c-albumLarvae281·2 7·00·00325035·684030·66385−1·02056
 Pupae114·4 9·70·00633733·623050·54448−1·03778
(b)StageWeibull distribution
Aglais urticaeLarvae0·428980·5606612·63091
 Pupae0·757530·21998 2·94989
Inachis ioLarvae0·414890·5782313·58128
Polygonia c-albumLarvae0·432940·5631112·37774
 Pupae0·776540·19083 2·82774
Aglais urticae0·1627·92
Inachis io0·1826·96
Polygonia c-album0·97 2·59

The best-performing model for each species was then modified to include the thermoregulatory component for larvae (Bryant, Bale & Thomas 1998). This required a daily (temporal) measure of sun (sun hours; University of Birmingham Geography Weather Facility, Monthly Records 1996) and a function describing the relationship between larval body temperature (when exposed to full sunshine) and ambient temperature. This relationship was described in the form of a linear equation, body temperature = (slope × ambient temperature) + intercept (Table 1c). The ambient temperature data used were from the data logger placed amongst the cages in the open situation. Details of the timing of sunny periods during each day are not usually available from meteorological stations, so the recorded amount of sun was assumed to occur in one continuous spell, equally distributed either side of the daily maximum temperature, and adjusted to the nearest 15-min temperature record. During this period within the model, predicted body temperature instead of ambient temperature was used to calculate larval development rate (using data from Bryant, Thomas & Bale 2000). At all other times (during the night and in overcast conditions), and for egg and pupa development, it was assumed that air (microhabitat) temperature was a reasonable approximation of body temperature (Bryant, Thomas & Bale 2000). In addition, the relationship between body and ambient temperature for the three species was not measured below 11 °C, so for temperatures lower than this it was assumed that body temperature was the same as ambient.


The development models were extended to predict species’ potential distribution and voltinism patterns, and also to investigate range differences as a consequence of larval thermoregulation. Some additional information was required for this.

UK Meteorological Office climate data were obtained from the British Atmospheric Data Centre, Didcot, UK. Daily maximum air temperature, daily minimum air temperature and hours of sun were used from 128 weather stations across England, Wales and Scotland. These stations were a subset of the total number available, which had no missing data for any of the three required measurements between 1 April and 1 October for at least 3 years between 1994 and 1998 inclusive. The temperature data used to drive the models were calculated for 15-min intervals from a daily temperature pattern described using the triangle method, with one daily maximum and two daily minima (Sevacherian, Stern & Mueller 1977), following Bryant, Bale & Thomas (1998).

To produce a phenological model that spans a whole season, the biology of the species needs to be considered. All three of the study species spend the winter as adults in reproductive diapause, re-emerging on the first warm or sunny days in spring, spending some time replenishing energy reserves before engaging in reproductive activity (Emmet & Heath 1990). The exact environmental conditions required to break the diapause and trigger reproductive development are unknown. The protracted periods of emergence and oviposition in these species in spring makes it difficult to estimate the start date for the model. Some sources state that mating and oviposition occur from March onwards for A. urticae and I. io, and that P. c-album is usually active even earlier than this (Emmet & Heath 1990). However, an analysis of British Butterfly Monitoring Scheme records by Sparks & Yates (1997) suggests that the mean earliest recorded activity dates for A. urticae and I. io fall around late April and early May, respectively. To compromise, a single start date, common to all species, of 1 April was chosen. (Preliminary runs indicated that the model was relatively insensitive to start date because of the longer physiological time scale in cooler spring conditions.) The models were adapted to include degree-day requirements and Weibull distribution functions for egg development, which were not included in the field development models (A. urticae: DD = 69·6, TH = 10·0 °C, γ = 0·79665, η = 0·21516, β = 10·61888; I. io: DD = 96·8, TH = 10·2 °C, γ = 0·82345, η = 0·19964, β = 4·40023; P. c-album: DD = 52·8, TH = 8·75 °C, γ = 0·84112, η = 0·20356, β = 3·47000; S.R. Bryant, unpublished data). The pattern of oviposition (100 ova, each one representing 1% of the total population) was estimated as starting on 1 April and followed a normal distribution for 15 days. This choice was arbitrary but again preliminary runs indicated that the models were fairly insensitive to alternative values due to the longer physiological time scale in spring. ‘Adults’ emerging within the model started producing ‘eggs’ after 7 days, in concordance with observations that adults of each species mate and start oviposition, in captivity, within 7 days after emergence (S.R. Bryant, personal observation). For the estimation of potential distribution, development was continuous: no attempt was made to induce diapause. Another arbitrary choice had to be made, namely the date on which to stop the models. Aglais urticae is said to enter hibernation between mid-September and early November (Emmet & Heath 1990), I. io is rarely observed after mid-September (Baker 1969), while the second generation P. c-album adults emerge during September. Adults require some flying time in order to build up lipid reserves; the less time there is for feeding the lower the chance of surviving the winter (Pullin 1987). The end of September was chosen as a suitable end date.


The start of the experiment at the beginning of May 1996 coincided with a cold, wet spell, with a ground frost occurring on several nights. Most larvae survived, although three groups of I. io under the tent and one group in the open either died or disappeared during the first fortnight. The month of May as a whole was cooler than average for the University of Birmingham site, but with average rainfall and sunshine. June and July were both slightly warmer and sunnier than average (University of Birmingham Geography Weather Facility, Monthly Records 1996).

The position of basking third, fourth and final instar A. urticae and I. io larval masses in the open changed within each cage as the day progressed, presumably to optimize their position with respect to the sun, whereas their shaded counterparts were generally less mobile. The cages in the open were not exposed to the full daily duration of sunshine as they became shaded by trees from about 18:00 onwards. Larvae were rarely observed to stray from the food plant, other than when searching for pupation sites. Development times were shorter for all three species in the open site exposed to solar radiation compared with times achieved in the site shaded by the tent (Table 2). However, the times were not strictly comparable as the temperature regimes recorded in each site differed slightly. The overall mean air temperatures for the period of the experiment were 15·1 °C in the open site and 14·4 °C under the tent, a difference of 0·7 °C.

Table 2.  Summary of development times observed for the open site and the shaded site experiments. n= number of individuals
SpeciesOpen siteShaded site
nDays ± SEnDays ± SE
Aglais urticae12960·7 ± 0·3419475·7 ± 0·15
Inachis io 9872·5 ± 0·12 9580·2 ± 0·12
Polygonia c-album 5465·7 ± 0·60 9077·4 ± 0·28

Observed times for larval and pupal development in the shaded environment, and times predicted using the degree-day and rate summation models, are presented as proportions developing per day to allow direct comparison (Fig. 1). For A. urticae, the degree-day model clearly overestimated both larval and total development time, the prediction error for 50% cumulative population development being +12 days and +6 days, respectively. The rate summation model was more accurate, the prediction error for larval development and adult emergence being −1 day and +1 day, respectively. There was little difference between the models for I. io, although rate summation was slightly more accurate over the whole development period. The degree-day model error for 50% emergence was +3 days for larval development and +2 days for adult emergence, compared to −3 days and −1 day, respectively, for the rate summation model. For P. c-album, however, the degree-day model performed best. The error for 50% emergence was −2 days for larvae and −2 days for adults, compared to −4 days for larvae and −9 days for adults with the rate summation model.

Figure 1.

Observed (shaded bars) and predicted (black squares) development times for larvae (first peak) and adult emergence (second peak) in the shaded experiment. Left-hand graphs show output from degree-day models, right-hand graphs show output from rate summation models.

The rate summation models for A. urticae and I. io and the degree-day model for P. c-album were used to model development in the open site, with and without the solar modification. Excluding the modification, the model predictions overestimated development times, compared with observed times for 50% adult emergence, by +11 days (A. urticae), +4 days (I. io) and +7 days (P. c-album) (Fig. 2a,c,e). These errors represent the total influence that solar radiation had on field development through to adult emergence in this experiment. Inclusion of the modification reduced the error in prediction of 50% adult emergence for all species to 0 days (A. urticae), −1 day (I. io) and +5 days (P. c-album) (Fig. 2b,d,f). However, for A. urticae and I. io, the modified models underestimated larval development time by −4 days and −7 days, respectively, and, as adult emergence was accurately predicted, overestimated pupal development time (Fig. 2b,d). It is possible that the netting cages reduced the levels of sunlight reaching the larvae (although this was not measured) resulting in the underestimation of development time. Observed pupal development time in P. c-album was also slightly shorter than predicted (Fig. 2f). Whereas larvae would normally seek a pupation site amongst foliage, most larvae of all three species pupated on the netting at the top of their cage, and were thus exposed to the sun for long periods. This undoubtedly resulted in some warming effect and thus a reduction in development time.

Figure 2.

Observed (shaded bars) and predicted (black squares) development times for larvae (first peak) and adult emergence (second peak) in the open, sunny site. For each species, the best-performing shade development models, driven using temperatures recorded in the open site, were used unmodified (left-hand graphs) and modified for larval thermoregulation (right-hand graphs).

For each species, models with and without the solar modification were tested using UK climate data. For each of the 128 weather stations, an average cumulative percentage adult emergence was calculated, the number of years used being dependent on the completeness of the climate records for that station. The definition of potential voltinism was based on this average (0–49 = no generation; 50–99 = 1/2 generation; 100–149 = 1 generation; 150–199 = 11/2 generations; 200–249 = 2 generations). The gridlines on the maps showing the model outputs represent an area of 1° longitude by 1° latitude (Fig. 3a). For some of these areas there was no representative weather station. In these cases the average adult emergence was calculated from adjacent areas. Note that the recorded UK distribution maps use a different grid system (Fig. 3c,f,i).

Figure 3.

UK distribution and voltinism patterns for each species, predicted using the standard (unmodified) best-performing shade development models (left-hand column) and using the same models modified for larval thermoregulation (centre column). Recorded UK distribution maps (right-hand column) were obtained from the Joint Nature Conservation Committee, Peterborough, UK.

Differences in distribution due to larval thermoregulation were apparent for A. urticae (Fig. 3a,b) and I. io (Fig. 3d,e). The relevance of the models, however, depends on how well they predicted actual recorded distribution (Fig. 3c,f,i) and voltinism. For A. urticae the overall distribution pattern indicated little, as both models predicted that at least one generation is possible over the whole of the UK. This is undoubtedly a factor in the observed stability of its distributional status (Heath, Pollard & Thomas 1984). The prediction by the thermoregulation model (Fig. 3b) fit observed voltinism better than the corresponding standard model. This species is generally considered to be bivoltine in England and Wales, becoming univoltine in Scotland (Emmet & Heath 1990). However, this varies greatly from year to year (Dennis 1985) and populations in central England sometimes manage only one generation. The prediction of two generations in southern England, 1·5 generations (partial bivoltinism) in central England and Wales and one generation in Scotland and the border regions thus fits observed patterns, particularly considering that the models use temperature data averaged over 5 years.

Inachis io is considered to be univoltine throughout its UK range. At present it occurs in western and southern Scotland, and is currently undergoing range expansion (Asher et al. 2001). Again the thermoregulation model performed better than the standard, at least in terms of distribution prediction (Fig. 3d,e,f). It also predicted the species’ absence from inland areas between 54° and 56° latitude. The standard model suggested an inability of the species to complete one generation over most of Scotland. The prediction of partial bivoltinism in southern England (Fig. 3e) hints at the possibility of a second generation in warmer than average years; occasional records of partial second generations on the south coast of England in 1976 have been reported (Lipscomb 1977; Holmes 1978).

The standard and thermoregulation models for P. c-album produced very similar results. This is not surprising as the body/ambient relationship used in the thermoregulation model only represented a small ambient increase of around 2·5 °C in the presence of sunshine due to microhabitat warming (Bryant, Thomas & Bale 2000). The performance of the models will be considered together. The predicted distribution suggested that P. c-album, like A. urticae, could be univoltine over the whole of the UK (Fig. 3g,h). However, its actual distribution (Fig. 3i) coincides more with its predicted areas of partial bivoltinism. Polygonia c-album has a complicated distributional history (Pratt 1987a,b). It was once fairly common in Scotland, but a decline in its then primary host-plant Humulus lupulus saw the range retract into six southern English counties by 1913. Since then, its range has expanded, probably due to a host-plant switch to U. dioica and an increase in winter and spring temperatures. Since 1982 its expansion has been rapid, and at present it has reached the Scottish border (Asher et al. 2001).The models suggested that it could persist even further north. It is not clear whether the northern populations are largely univoltine as predicted; Swedish populations at 59·5° latitude are univoltine (Nylin 1989).

Until now we have only considered the number of adults produced. As the models were continuous between a common start and end date, it was also possible to look at the total amount of development achieved: a population may be classed as univoltine but that alone does not indicate how much development is possible in the subsequent generation. For this, cumulative percentage egg, larval and pupal development were calculated, i.e. one generation is equivalent to 100 + 100 + 100 = 300 development units. However, the different stages are not comparable in terms of percentage values as each stage has a different thermal requirement for development. To make each stage relative it was multiplied by its own degree-day requirement divided by the degree-day requirements for larvae (as larvae had the largest degree-day requirements for all three species). The thermoregulation and standard models for each species could then be compared by plotting development units against latitude, for each weather station (Fig. 4). The numbers of development units that represent one generation were 180·6 (A. urticae), 165·6 (I. io) and 163·0 (P. c-album). The latitudinal difference in the points at which one generation occurred with and without thermoregulation could be calculated for each species by solving the appropriate regression equations (Table 3a). The large amount of scatter is partly a function of the altitude of the weather stations, as can be seen from the improved fit of the linear regression when an altitudinal correction is employed (R-squared; Table 3). To correct for altitude, air temperatures used to drive the models were increased 0·6 °C 100 m−1 altitude (Critchfield 1974).

Figure 4.

Average number of development units achieved per season against latitude, as predicted by the standard (unmodified) best-performing shade development models (open circles, lower regression lines) and using the same models modified for larval thermoregulation (closed circles, upper regression lines).

Table 3.  Summary of development unit on latitude regressions for models using (a) raw temperature data and (b) temperature data corrected for altitude
(a)Standard modelThermoregulation modelOne generationTwo generations
SpeciesEquationR-squaredEquationR-squared° Latitude differenceDifference in km*° Latitude differenceDifference in km
A. urticaey = −41·0x + 2736·40·648y = −27·6x + 1856·00·6581·7192·62·5274·5
I. ioy = −19·2x + 1254·70·609y = −22·3x + 1460·70·6731·4152·22·6285·6
P. c-albumy = −20·9x + 1406·50·645y = −20·8x + 1408·40·6410·5 52·00·4 46·5
(b)Standard modelThermoregulation modelOne generationTwo generations
SpeciesEquationR-squaredEquationR-squared° Latitude differenceDifference in km*° Latitude differenceDifference in km
  • *

    1 degree of latitude is approximately equivalent to 111·3 km.

A. urticaey = −23·5x + 1585·90·767y = −26·1x + 1791·20·7671·8203·42·6289·6
I. ioy = −19·3x + 1272·90·703y = −21·0x + 1404·00·7961·7185·12·3261·3
P. c-albumy = −19·5x + 1345·80·765y = −19·4x + 1344·50·7640·5 59·00·4 49·9


Development was faster when influenced by the natural input of solar radiation than it was in a completely shaded environment for both a thermal-conforming species, P. c-album, and two behavioural thermoregulators, A. urticae and I. io. The reductions in development period were due to a combination of slightly higher mean air temperatures and a differential between larval body temperature and recorded ambient temperature, and equated to approximately 20%, 10% and 15% for A. urticae, I. io and P. c-album, respectively (from Table 2). While this was expected for the gregarious species, the relative increase in development rate of P. c-album was not. Several possible reasons exist, although none satisfactorily explains the discrepancy in its own right. (i) There was a much greater proportion of reproductive light morphs produced from larvae developing in the open site experiment (44·4% compared with 8·0%), and such individuals have been shown to develop faster than those destined for diapause (Nylin 1992). (ii) Late instar P. c-album reportedly spend some time basking, protected by their resemblance to a bird dropping (Thomas 1986; Emmet & Heath 1990). Individuals that do become exposed to the sun experience raised body temperatures, although not to the same degree as for A. urticae and I. io (Bryant, Thomas & Bale 2000). This effect may have been enhanced as the number of larvae per plant was unnatural, and they were observed to be more mobile than normal due to disturbance increasing their exposure to the sun. (iii) It is possible that the position of the data logger was not representative of the whole open site and that microhabitat temperatures were underestimated. The environment under the tent may also have been altered with respect to other environmental factors that were not measured, such as humidity, wind speed and rainfall.

A field-based comparison of rate and thermal summation techniques has, to our knowledge, never been made using insects that have been allowed to develop in a completely shaded environment, where microhabitat and body temperature effects are controlled. For A. urticae particularly, but also I. io, observed development was faster than predicted by the degree-day approach, but was described very accurately by rate summation. This suggests that development below the theoretical linear threshold did occur. It has also shown that non-linear models based on rearing larvae under constant temperature conditions in the laboratory, do accurately describe development in a field situation. While the underlying temperature-dependence of insect development must form the basis to such models, other results presented here highlight the need for increased awareness of factors that may influence development rates. For P. c-album, it has been shown that directly developing individuals develop faster than those destined for diapause (Nylin 1992). For this species, reproductive state can be judged from adult morphology (Nylin 1989): diapausing individuals (normal form) have much darker undersides than reproductive individuals (hutchinsoni form). Errors in prediction occur when the ratio of diapausing to reproductive individuals produced under field conditions differs from that produced by the laboratory experiments, the results of which are used to produce the development models in the first place. The laboratory experiments that determined the developmental parameters for P. c-album (Bryant, Thomas & Bale 1997) produced 66·7%hutchinsoni form (S.R. Bryant, unpublished data). This is a much higher proportion than that observed for the shaded experiment, for which only 8·0% were hutchinsoni individuals. This may at least partly explain why the rate summation model underestimated development times in this species. Such phenomena are difficult to include within models and for many species the associated error may be small and/or cancelled out by the accumulation of other small errors (Higley, Pedigo & Ostlie 1986).

The thermal modification made to the best-performing model for each species was fairly crude, but effective. Most importantly, however, it was simple to apply and could be for any species with a little additional information about their thermal ecology. Previous studies that have attempted to quantify the effects of solar radiation on development have done so with various types of modification to the development models involved. McDonald & Smith (1988) found that spring development of the Rutherglen bug Nysius vinitor (Hemiptera: Lygaeidae) was more rapid than predicted by the degree-day model used. This was assumed to be due to the insects basking in the sun; the maximum daily temperatures used as model inputs were adjusted in hindsight, to obtain a good fit (see also Kehat & Wyndham 1972). Lyons (1994) compared rate summation models to predict the development of the pine false webworm Acantholyda erythrocephala (Hymenoptera: Pamphiliidae), driven by meteorological station, web and canopy (microhabitat) temperature data. The effect of solar radiation on larval web temperature was accounted for by calculating averages of hourly web and ambient temperature differentials over the entire development period. The resultant function was integrated with air temperature data within the model, for a set period of daylight each day, but this did not provide predictions as good as those using air temperature alone. Rawlins & Lederhouse (1981) related monarch butterfly Danaus plexippus (Lepidoptera: Nymphalidae) larval positioning and larval body temperature when exposed to the sun to time of day, and modified a degree-day model to estimate increases in development accountable to solar radiation, assuming the sun was out for the entire day. The model was not tested. Weiss et al. (1993) modelled the post-diapause development of bay checkerspot butterfly Euphydryas editha bayensis (Lepidoptera: Nymphalidae) larval masses, which were known to raise body temperature significantly in direct sunlight, using a physiological time scale based primarily on levels of insolation, not temperature, as the main growth controlling factor.

Many species are not able to raise their body temperatures by basking to the same extent as A. urticae or I. io, although most will experience microhabitat temperatures that are warmer than surrounding air temperatures. For some this may be relatively insignificant in terms of development modelling, particularly those that live in naturally shaded environments, but for others the effect could be substantial. This may account in part for the observed faster development of P. c-album larvae in the open compared with under the tent. Substrate temperatures at or just above ground level are elevated well above ambient under high levels of insolation, and immature insects that live in this ground layer will undoubtedly benefit, particularly if they are able to move vertically within thermal layers to maintain optimum temperatures (Sherman & Watt 1973). Thus even without using solar radiation directly, the thermal experience of many species may differ greatly from that predicted using macroclimate air temperatures alone. To bridge the gap between macro- and microclimate temperatures for use in development models we also need to classify and quantify the thermal characteristics of ecologically important microhabitats (Kennedy 1997). The methodology outlined here allows for the inclusion of any relationship between macroclimate and larval body temperature, whether the effect is direct through basking (this study), indirect through microhabitat warming, or, preferably, a combination of both.

The UK distribution models that included the modification for thermoregulation indicate that, as a consequence of larval thermoregulation, A. urticae and I. io are able to persist at more northern latitudes than would otherwise be possible, maybe as much as 200 km further north (one generation; Table 3). It follows that voltinism is similarly shifted, and the extent of bivoltinism may be around 300 km further north (two generations; Table 3). This difference in relative developmental benefit between univoltine and bivoltine populations is partly an artefact of the climate in northern England and Scotland, which is oceanic in nature (Dennis 1977). Although day length increases with latitude, average sun hours per day decreased from 6·5 h at 51° latitude to 4·7 h at 57° latitude, thus reducing the potential advantage of thermoregulation in more northern latitudes. However, where two larval generations are possible the advantage of thermoregulation is effectively doubled. This fact alone may explain the unequal voltinism shift. In real terms, a quick spring generation not only allows for more time to complete a second generation, but ensures that the second generation develops during the warmest months of the year (Bryant, Bale & Thomas 1998). Further improvements could be made by addressing some of the assumptions outlined in the Materials and methods, particularly those relating to the start and end dates of the models; it is likely that opportunities for adult activity become reduced in the spring and autumn as latitude increases. Other factors, such as oviposition preferences, may play an important role. Aglais urticae females are thought to favour sunny U. dioica patches with a south-eastern bias (Dennis 1984), potentially increasing the amount of solar radiation received. They may also choose shorter plants compared with I. io (Feber, Smith & Macdonald 1999), and by being closer to the ground experience warmer microclimates during development.

The models suggest that the first noticeable response to global warming would be the increased frequency of (partial) bivoltinism in I. io: there is clearly no constraint in terms of available physiological time for one generation (Fig. 3e). The current range expansion of P. c-album could be construed as a response to global warming, although it is still only regaining territory lost during its late 19th/early 20th century decline. The recovery has been slow, and may point to other factors. There is apparently enough physiological time for at least one generation even in Scotland, but the butterfly's recorded distribution seems to coincide with areas of predicted partial bivoltinism (Fig. 3h,i). There is a correlation between overwintering survival and winter humidity (Pratt 1987b), and populations may suffer with the relatively mild and wet UK winter climate without a boost in numbers from second generation adults.

While the use of climate-based physiological models to map species’ distributions has its limitations, such models remain widely used as they provide an invaluable insight into the potential distribution of a species. Although other factors interact to determine a species’ actual distribution (Davis et al. 1998), understanding potential distribution is key to studying the effects of global warming on insects’ geographical ranges, identifying areas under threat from invasive pest species, and predicting the voltinism of introduced biocontrol agents as one component of their likely success (Barlow, Goldson & McNeill 1994). This study is intended to add to the growing body of work highlighting the importance of thermal ecology to this area of research.


We would like to thank the Biological Records Centre, the Institute of Terrestrial Ecology and the Joint Nature Conservation Committee for access to recorded distribution maps, and the British Atmospheric Data Centre for access to the UK Meteorological Office climate data. Dr T.H. Sparks and an anonymous referee provided helpful comments. This work was supported by a NERC studentship to S.R. Bryant.