Hot topics in butterfly research: Current knowledge and gaps in understanding of the impacts of temperature on butterflies

As small poikilotherms, insects are largely dependent on their environment for thermoregulation and so are particularly vulnerable to changing temperatures. Butterflies are a well‐studied group often used as models to investigate insect responses to temperature. However, little has been done to synthesise and present this large volume of literature in an accessible format, particularly with reference to knowledge gaps and areas rich in information. Using a systematic mapping method, we synthesised the last 40 years of research on the topic of butterfly responses to temperature. We identified and coded 451 research papers, in which butterfly species were studied 3198 times. We identified taxonomic groups, regions and experimental designs that were well or poorly represented. We found that there was a relatively good balance of representation across butterfly families in relation to the number of species within each family. The tropics were less frequently studied than temperate regions, and there were more studies reporting outcomes on adults than at any other life stage. Finally, in situ studies were more common than ex situ studies. Taken together, the higher representation of certain regions, life stages and approaches could lead to an incomplete understanding of the impacts of temperature on butterflies, potentially resulting in ill‐informed decisions. We make suggestions for how to resolve these discrepancies in representation, including calling for an increased focus on the tropics, the establishment of butterfly monitoring schemes in the global south, a greater focus on the effects of temperature on non‐adult life stages, an increase in experiments investigating fluctuating thermal regimes and the incorporation of more behavioural responses to temperature in future research. Only by addressing these disparities can we gain a complete understanding of how butterflies will respond to climate change.


INTRODUCTION
Under climate change, mean temperatures, the frequency of extreme weather events and temperature variability are all projected to increase (IPCC, 2014).The effects of changing temperatures have been detected in many taxa, particularly insects (Andrew et al., 2013;Elias, 1991;Kingsolver et al., 2011).As small poikilotherms, insects are largely dependent on their environment for thermoregulation, so changing global temperatures will likely influence insect populations and the ecosystems they inhabit in many ways.In particular, temperature has been found to impact insect growth (Maino et al., 2016), distribution (McCain & Garfinkel, 2021), behaviour and survival (Kingsolver et al., 2013), and the synchronicity of ecological interactions (Cornelissen, 2011).
Butterflies make a good model group to investigate the effects of temperature on insects, as they are well-studied (Sparks & Yates, 1997;Warren et al., 2001), fecund, and have generally high dispersal as adults, relatively resolved systematics, and short generation times.Therefore, changes in abundance and distribution can be detected over short time periods (Brown, 1997;Parmesan, 1996;Thomas, 2005).Temperature impacts butterfly emergence (Gezon et al., 2018), development and survival (Bauerfeind & Fischer, 2014), reproduction (Jones et al., 1982) and phenotype (Stevens, 2004).A broad body of research has investigated the effects of temperature on butterflies, covering a variety of life stages, species, response variables, research focuses and approaches, and geographic areas.For example, research can cover the whole life cycle of butterflies (Klockmann et al., 2017) or focus on a specific life stage (Scriber et al., 2012).It can cover single species (Zalucki, 1981), or entire communities consisting of thousands of species (García-Barros, 2000).
Response variables can range from critical thermal maximum studies assessing thermal tolerance in laboratories (York & Oberhauser, 2002) to long-term field studies of changes in abundance in relation to climate variables (Brooks et al., 2017).The purpose of these studies ranges from conservation (Koda & Nakamura, 2010) and pest control (Pandey et al., 2015) to maximising production for commercial butterfly gardens (Kim et al., 2012).
However, despite the large volume of diverse literature investigating the effects of temperature on butterflies, little has been done to synthesise and present this information in an accessible format, particularly with reference to knowledge gaps and areas rich in information.Only through a comprehensive understanding of the complex effects of changing temperatures on butterflies can conservation, policy and coordination of research funding continue to be informed and effective.Here, we use a systematic mapping approach to provide an overview of patterns in existing research on the effects of temperature on butterflies.We flag areas where further data synthesis and meta-analysis would be profitable, as well as under-represented areas that should be prioritised for future research.Though some trends in global research are well known (e.g., Stroud & Feeley, 2017), it is important to test whether these research gaps are being addressed and reduced over time.Specifically, we answer the following key questions: 1. What is the distribution of research across taxonomic groups (at both the family and species level), how does this compare to the number of described species in each family and has this changed over time?
2. What is the distribution of research on butterfly life stages, including when interventions are applied and outcomes are measured, does this differ between families and has this changed over time?
3. What is the distribution of research geographically for both in situ and ex situ studies and has this changed over time? 4. Are observational or experimental approaches more common and has this changed over time?
5. What are the most commonly recorded outcomes (e.g., physiology and ecology), does this differ between families and regions, and has this changed over time?

Search strategy
We extracted globally published peer-reviewed literature from 10 commonly used online databases available in English (Web of Science Core Collection, SCOPUS, BIOSIS Citation Index, Current Contents Connect, Data Citation Index, Derwent Innovations Index, KCI-Korean Journal Database, MEDLINE, Russian Science Citation Index and SciELO Citation Index), between 1980 and 2020 (search conducted on 14 May 2020).This date range was chosen to reduce selection bias, due to limited pre-1980 literature being digitised in the databases searched.We also searched Google Scholar, which provides a wider coverage of literature than many conventional databases, with the first 50 hits evaluated for relevance.A limit of 50 records was used due to the high prevalence of duplicates and irrelevant literature that we identified in this search engine.

Search string and search comprehensiveness
To generate an appropriate search string, we selected a set of 20 relevant studies as benchmark papers to ensure search comprehensiveness (Table S1).Benchmark papers were selected in discussion with experts in the field, as good examples of the literature that we wanted to include, covering a variety of species and topics within the subject of temperature and butterflies, and published over a wide time span.
All benchmark papers were found using Google Scholar or were already known to the authors.We modified the search string by adding relevant synonyms and keywords, and using wildcard search terms, until all benchmark papers were returned across all databases.
The final search string that extracted all 20 benchmark papers was as follows:

Supplementary searches
We also searched references cited in the 20 benchmark papers in a 'snowball design' to identify additional relevant publications not extracted by our search string (Naderifar et al., 2017).Due to the high frequency of duplicate records from the benchmark paper reference lists with the initial extracted database, this process was only carried out for the benchmark reference lists.The results from all databases and searches were combined and further checked for duplicates before proceeding to the article screening stage.

Inclusion/exclusion criteria
For an article to be included, it had to fulfil all of the inclusion criteria, while not fulfilling any exclusion criteria (Table S2).Our criteria were developed to capture all relevant literature that provided novel data on the effects of temperature on butterflies in the English language while aiming to minimise bias.Inclusion and exclusion criteria were defined following the PICOS framework: Population (research focus or study species: butterflies), Intervention (action or process being applied to the study species: temperature), Comparison (comparisons across groups to identify responses: within or between butterfly species or thermal regimes), Outcomes (the recorded impact of the intervention: all), and Study (design or type of study: presenting novel data and peer-reviewed) (Methley et al., 2014).This provided a structure to quantitatively and qualitatively assess literature while limiting the number of irrelevant articles.
The inclusion/exclusion criteria were applied to the extracted datasets in three rounds (screening titles, abstracts and full texts).The first two rounds were each undertaken by two reviewers and the final round by eight reviewers, with each reviewer receiving a subset of articles to screen.To ensure repeatability and consistency, decisions about relevance were tested between all reviewers.Each reviewer was presented with the same random subset of papers for that stage (10% of the total), and their decisions were compared using a Kappa coefficient between all pairs of reviewers (Cohen, 1960).Reviewers did not proceed to screen the dataset for that round until a score of >0.7 was achieved (following Collaboration for Environmental Evidence, 2018 guidelines) (Table S3).All disagreements were resolved in discussion, after which there was never any ambiguity as to an article's relevance.All reasonable effort was then made to secure the full texts of relevant articles, but any articles without accessible full texts were excluded (n = 197) (Figure S1).At the fulltext screening stage, the reason for exclusion was recorded.

Data coding strategy
A standardised questionnaire was used for each article to extract meta-data into Microsoft Excel (version 16.30), using a mixture of free-form and closed-entry fields.Repeatability of this process was tested between all three reviewers using a 10% subset of articles, until 70% agreement was reached, using a similar procedure to that outlined above for article relevance screening (Table S4).
Each article was coded across three categories: population (species, life stage and country of origin), intervention (in/ex situ, thermal regime temperature ranges and durations), and outcome (categorised within six broad categories: behaviour, development, ecology, evolution, physiology and population).For more detail on how these were coded, see Supporting Information S2, Methods 1.

Data analyses
All analyses took place in R (version 4.2.0,R Core Team, http://www.r-project.org).
What is the distribution of research across taxonomic groups (at both family and species level), how does this compare to the number of described species in each family and has this changed over time?
To describe research coverage within and between the six butterfly families, two values were calculated; the number of unique species studied within each family, and the total number of times any species within a family was studied (allowing double-counting of species between studies).These two values were used to describe the amount of research interest a family has received (number of unique species studied) and the current level of knowledge per family (the number of times species of each family have been studied).The number of times species were studied per family was compared to the total recorded number of species within each family with a Pearson's chi-square test, to examine differences in representation between families in relation to the size of the family.The total number of species within each family was taken from Shields (1989).The most commonly represented species were identified by a count of occurrence across all studies.To investigate change over time, a Poisson regression was fitted with the count of publications as the dependent variable, and year and family as explanatory variables.An interaction term was included between year and family to test whether the rate of publication differs between families.Data dispersion was checked with the 'AER' package (Kleiber & Zeileis, 2008), and data were not overdispersed.The publication rate was then extracted per family.
What is the distribution of research on butterfly life stages, including when interventions were applied and outcomes measured, does this differ between families and has this changed over time?
The number of times species were studied at each life stage (ova, larva, pupa and adult) was counted.One study included over 1000 species at multiple life stages; this study was excluded from changeover-time analyses and removed from plots, so the rest of the data could be seen (see Supporting Information S2 for plots including this study).To identify the life stage that most commonly experienced the temperature intervention and the life stage at which outcomes were most commonly recorded, the number of times species were studied was counted for each life stage for interventions and outcomes separately, and tested with individual Pearson's chi-square tests, within and across families.To test whether these changed over time, the first two Poisson regressions were fitted (as data were not overdispersed), with the count of publications as the dependent variable, and year and life stage that either experienced the temperature intervention or at which the outcomes were recorded as explanatory variables.An interaction term was included between year and life stage to test whether the rate of publication increase differed between life stages.
The publication rate was then extracted per life stage.
What is the distribution of research geographically for both in situ and ex situ studies and has this changed over time?
The number of times species were studied per continent and country was counted for all studies, and then split into in situ and ex situ studies.For ex situ studies, the country of origin of the butterflies was used, rather than the country of laboratory location.The number of times species were studied in tropical or temperate regions was also counted.To test whether temperate and tropical publication rates have changed over time, first, a Poisson regression was fitted (as data were not overdispersed), with the count of times species were studied as the dependent variable, and year and region as explanatory variables.An interaction term was included between year and region to test whether the rate of change differed between tropical and temperate regions.The publication rate was then extracted per region.
Are observational or experimental approaches more common and has this changed over time?
The number of times species were studied where the temperature regime was manipulated (experimental) or observed (observational) was counted.The experimental approaches were described and counted, such as the temperature range used, the average number of thermal regimes, the minimum, maximum and mean temperatures used, and the duration of the thermal regimes.To test whether thermal regimes have changed over time, the minimum, maximum and mean temperatures within thermal regimes were tested in simple linear regressions against year.To test whether the duration of thermal regimes has changed over time, the minimum, maximum and mean durations within thermal regimes were tested in simple linear regressions against year.
What are the most commonly recorded outcomes (e.g., physiology, ecology), does this differ between families and regions and has this changed over time?
The recorded outcomes (according to the six broad categories, Table S6) were counted each time a species was studied.Where individual studies reported multiple outcomes, all outcomes were counted individually.To test whether the categories of outcomes have changed over time, first, a Poisson regression was fitted (as data were not overdispersed), with the count of times species were studied as the dependent variable, and year and outcome category as explanatory variables.An interaction term was included to determine whether the rate of change differed between outcome categories.The publication rate was then extracted per outcome category.To test whether the outcomes recorded differed between families and regions (tropical or temperate), two Pearson's chi-square tests were used between outcome categories across families and regions.

RESULTS
A total of 451 relevant papers were included in the systematic map database, covering 1973 unique species across all butterfly families.There were 3198 instances of butterfly species having their responses to temperature investigated.
What is the distribution of research across taxonomic groups (at both family and species level), how does this compare to the number of described species in each family and has this changed over time?
There was no significant difference in the number of publications relative to the number of species in each family (χ 2 = 30.0,d.f.= 25, p = 0.224) (Figure S2).See Supporting Information S2, Results 1, for publication counts per family.
For individual species, the most commonly represented in the literature was Bicyclus anynana, followed by Pieris rapae, Pararge aegeria and Danaus plexippus (Figure 1).
The number of butterfly species across all families whose responses to temperature have been published has increased over the last 40 years (at a rate of 4.1 species studied per year, χ 2 = 3145.2,d.f.= 40, p < 0.001) (Figure 2).However, publications have increased at different rates between the six families (F = 18.5, d.f.= 1,4, p < 0.001), with Nymphalidae having the greatest rate of publication increase (1.7 species studied per year, χ 2 = 1312.9,d.f.= 40, p = 0.011), followed by Lycaenidae (1.0 species studied per year,

What is the distribution of research on butterfly life stages, including when interventions are applied and outcomes are measured, does this differ between families and has this changed over time?
There was no significant difference in the frequency at which each life There was a significant increase in the number of times each life stage was exposed to a temperature intervention over time (egg: χ 2 = 3255.6,d.f.= 40, p < 0.001; larva: χ 2 = 3128.9,d.f.= 40, p < 0.001; pupa: χ 2 = 3132.7,d.f.= 40, p < 0.001 and adult: χ 2 = 3351.9,d.f.= 40, p < 0.001).There was no significant difference in the rate of increase over time between life stages (χ 2 = 72.9,d.f.= 120, p = 0.999) (ranging from 3.8 to 4.0 species studied per year from 1980 to 2020).
There was a significant increase in the number of recorded outcomes at specific life stages over time (egg: χ 2 = 96.1,d.f.= 40, p < 0.001; larva: 223.5, d.f.= 40, p < 0.001; pupa: 159.0, d.f.= 40, p < 0.001; and adult: χ 2 = 3253.8,d.f.= 40, p < 0.001) (Figure 3).The number of publications recording outcomes on adults (produced at a The cumulative number of times butterfly species had their response to temperature investigated from 1980 to 2022, split by butterfly family.For visualisation purposes, this plot excludes a single paper with over 1000 recorded species studied (the equivalent plot including this paper is given in Figure S3).rate of four species studied per year from 1980 to 2020) has increased at the greatest rate over time compared with the other life stages (egg: 0.06 species studied per year, larva: 0.29 per year, pupa: 0.18 per year, χ 2 = 600.2,d.f.= 120, p < 0.001), therefore increasing the disparity between studies on adults versus other life stages over time.
What is the geographic distribution of research for both in situ and ex situ studies and has this changed over time?
We recorded studies from 45 countries across six continents.The majority of butterfly species studied were sourced from Europe Of the 3198 times species studied, 1833 were sourced from temperate study sites, compared with 106 from tropical study sites, and 1257 where the location was not stated (Table S7).
The rate of increase in the number of times species have been studied differed between temperate and tropical regions (χ 2 = 226.1,d.f.= 40, p < 0.001).The number of times species from temperate regions have been studied has increased over time (3.9 species studied per year, χ 2 = 2140.1,d.f.= 1, p < 0.001).The number of times species from tropical regions have been studied has also increased, but at a slower rate (0.07 species studied per year, χ 2 = 11.6,d.f.= 1, p < 0.001).
Over the last 40 years, the rate of increase of species from temperate regions being studied is 52 times greater than the rate of increase from tropical regions (Figure 5), meaning that the disparity in the number of studies in tropical versus temperate regions has increased over time.
The cumulative number of times butterfly species had the outcome of their response to temperature investigated at specific life stages from 1980 to 2020.For visualisation purposes, this plot excludes a single paper with over 1000 recorded outcomes (the equivalent plot including this paper is given in Figure S5).Although the majority of outcomes were recorded at the adult life stage, this was not reflected consistently across outcome categories.

Adults were best represented in the categories Ecology and
Population (with outcomes on the adult life stages representing 95% and 90% of the total for these categories, respectively).Development was the only category where the adult life stage did not make up the majority of studies, with larvae (32%) and pupae (31%) being more commonly recorded (Figure 8).
The distribution of outcome categories recorded differed between temperate and tropical regions (χ 2 = 281.4,d.f.= 10, p < 0.001).The majority of temperate studies recorded outcomes on ecology, whereas the majority of tropical studies recorded outcomes on physiology (Figure S8).

F I G U R E 7
The cumulative number of times butterfly species have been studied where outcomes in each of six categories were recorded in response to temperature from 1980 to 2020.For visualisation purposes, this plot excludes a single paper with over 1000 recorded outcomes (the equivalent plot including this paper is given in Figure S7).

Overview of results
We found substantial taxonomic, geographic and experimental variation in the depth of research on different aspects of butterfly responses to temperature.Of the six butterfly families, Nymphalidae was most represented in the last 40 years of research, whereas Riodinidae was least represented, although this was consistent with the relative number of species described in each family.The most commonly studied species was B. anynana (Nymphalidae), followed by P. rapae (Pieridae) and P. aegeria (Nymphalidae).The differences between life stages exposed to a temperature intervention were minimal, within and across families, with all stages being similarly represented in the literature.In contrast, outcomes were largely recorded on the adult life stage, with more outcomes recorded on adults than all other life stages combined.There were also differences between families and the life stage at which outcomes were recorded, with Hesperiidae, Lycaenidae and Pieridae having a particularly high proportion of outcomes recorded at the adult life stage.The majority of species studied were from Europe and North America, with the fewest from South America.The majority of studies were from temperate regions, with very few from the tropics.Most studies were conducted in situ rather than ex situ, and the majority of species were studied observationally, rather than experimentally.For experimentally manipulated temperature regimes, the majority used constant rather than fluctuating temperatures.Experimentally manipulated temperatures have generally increased over time, whereas the duration of temperature regimes applied has decreased.For observed temperature regimes, very few stated the temperatures experienced, but for those that did, there was a decrease in temperatures used, whereas the duration of these studies has increased over time.The most frequently recorded outcome was ecology, whereas behavioural outcomes were recorded the least.The outcomes recorded differed between regions, with temperate studies tending to record more outcomes on ecology, but tropical studies tending to record more outcomes on physiology.The outcomes recorded differed between families, with Pieridae having a higher representation in behavioural outcomes, and Hesperiidae having more population outcomes.Outcomes also differed between life stages, where outcomes recorded on adults were generally ecological or at the population level, whereas the non-adult life stages tended to have more developmental and physiological outcomes recorded.

Taxonomic groups and geographic distributions
The relative frequency of studies did not differ from the number of described species per family, suggesting that the pattern of high representation of Nymphalidae reflects the high number of species within this family.The high number of studies on Nymphalidae and the rapidly growing body of literature on this family implies a good level of understanding for this large and diverse group of butterflies.As this family represents a substantial portion of global butterfly diversity, this suggests that consensus findings may be representative of a large proportion of species.However, a large number of studies are on single nymphalid species, often captive bred (B.anynana).This could be problematic for global understanding of temperature impacts.In addition, it is not clear how results may be affected by long-term captive breeding and a relatively small founder population.In contrast to Nymphalidae, we found far fewer studies on Riodinidae, though not disproportionately compared with the size of this family.Riodinidae contains a relatively low number of species, which tend to be small, inconspicuous and exhibit elusive behaviour, which may contribute to their lack of representation.
The lack of research from the tropics is not a surprise; the neglect of the tropics is a widespread issue throughout ecology and evolutionary research (Clarke et al., 2017;Stroud & Feeley, 2017).Research effort is often linked to availability of resources, funds, and accessibility of an established research institute, and can be restricted by logistical hurdles, as well as policies, politics and local sociocultural factors (Stroud & Feeley, 2017).Despite how well documented the lack of research effort in the tropics has been over the last few decades, we found very little evidence that this gap is being addressed, with no major change in publication rate over the last 40 years.However, it is important to note that it is possible that not all relevant literature from the tropics was included in this map.For example, only English language publications were included in our search, potentially leading to a bias in our sampling towards temperate studies.Despite this, it is worth noting that when translated into Spanish, our search string only pulled a further 31 studies (unscreened for relevance), so it is unlikely that the literature missed would have changed the trends we detected.
In contrast to the tropics, we found a far higher number of studies in the temperate north, with the rate exponentially increasing over time.These tend to be regions where established research institutes are located, and where policies and active conservation are often already in place and are funded to protect declining species.As climate change is predicted to have amplified effects at higher latitudes, this could mean that research is targeted to the taxa expected to experience the greatest change, facilitating more effective conservation efforts in these regions.However, given that the majority of butterfly diversity is concentrated in the tropics (May, 1992), there is evidence to suggest that tropical species may be more vulnerable to even minor changes in ambient temperature conditions (Ghalambor et al., 2006;Grinder & Wiens, 2023;Janzen, 1967), and this finding also highlights a lack of research in the most diverse areas and a lack of understanding on the impacts of temperature change in the tropics.
It is also possible that species in different regions or biomes may respond to temperature change in different ways, so studies concentrated within a few countries may lack global transferability.
At the species level, the number of studies seemed to be related to commonness or pest status, as well as the use of species as model organisms.The most commonly studied species was B. anynana (Nymphalidae), which is used as a model for developmental biology (particularly in relation to eyespot formation) and has been maintained in a laboratory stock in Europe since 80 gravid females were collected in Malawi in 1988.The frequent study of this species implies a good understanding of butterfly wing spot formation under various temperature regimes, assuming that this species is comparable to others and that long-term maintenance in laboratory settings has not altered its biology.The second most commonly studied species was P. rapae, a Pieridae native to Europe that has become naturalised around the world and is a common agricultural pest of wild and domesticated Brassica and other plants in the mustard family.

Research interventions and experimental approaches
We found an equal representation across the four life stages being exposed to temperature interventions.This may be due to the large number of in situ studies, where multiple life stages are often exposed to ambient temperature conditions in the same study, or where ambient conditions are not restricted to particular life stages, often with the purpose of recording adult responses to temperature intervention.
In situ studies are conducted within the ecological context of the species and can provide insights into complex interactions and behaviours not possible within a laboratory or other captive setting.However, confounding factors, such as habitat, may influence results, which can be accounted for in ex situ studies.In addition, time and space can be limiting factors in the sample sizes and number of species studied in situ, particularly for rare or univoltine species.However, accessible long-term datasets, often conducted using citizen science approaches, can remedy this limitation, all of which can be found in the most frequently studied countries (North American Butterfly Association counts, the Catalan Butterfly Monitoring Scheme, Butterflies of India/ IFoundButterflies).The strong representation of in situ studies implies a growing understanding of the long-term effects of ecologically relevant temperatures on butterflies in the natural world.However, this dominance also indicates that there is a lack of understanding of future or abnormal temperatures on butterflies, which are more easily studied ex situ.
The increased use of long-term butterfly monitoring datasets also explains the high proportion of observed thermal regimes (in contrast to experimentally manipulated thermal regimes).However, studies with observed temperature regimes rarely stated the temperatures the butterflies experienced during the study.This lack of accessible information means that the outcomes recorded lack context when compared with other studies.In the few studies that did state the temperatures observed, temperatures were decreasing over time.This may be the result of an increased focus on in situ studies in high latitude or altitude locations, as these areas are of increasing conservation concern (e.g., Ashton et al., 2009).We also recorded an increase in the duration of observed temperature regimes over time, which may be testament to the availability of ongoing long-term in situ butterfly datasets and climate variables, which are often freely available.
In contrast, there were fewer ex situ studies in the literature, with the majority of the butterflies used in these studies coming from the USA, Japan, Australia and Malawi.This reflects the distribution of butterfly sources for established laboratories used to conduct experiments, which require significant investment and knowledge to establish but also implies a relatively global distribution of species studied with experimental temperature interventions.A limitation of ex situ studies is that species need to be accessible, policies must be in place to allow for the collection of individuals, and methods for the successful rearing or maintenance of species in captivity must be understood, which can take a substantial amount of time and resources.This may restrict ex situ studies to existing well-studied species.Ex situ studies also take place outside of the ecological and environmental context of the species, and so may yield relatively little information about how species respond to real-world challenges.
However, a benefit of ex situ studies is that they allow for species to be exposed to temperatures outside of their natural range; the frequent use of B. anynana, sourced from Malawi and maintained in a laboratory stock in Europe, is a testament to this.
Ex situ studies tended to rely on experimentally manipulated thermal regimes using constant rather than fluctuating temperature, despite the latter being more biologically relevant.For example, studies in other insect groups have shown that constant and fluctuating temperature regimes can alter responses to temperature (Clarkson et al., 2013;Paaijmans et al., 2013).Studies on butterflies show responses can be similar in direction but differ in strength between constant and fluctuating temperature regimes (Fischer et al., 2011), so caution is needed when interpreting outcomes from constant temperature regimes (Bauerfeind & Fischer, 2014).Constant thermal regimes may be favoured by researchers due to their relative simplicity in setting-up observations compared with fluctuating regimes, and the reduced technological investment needed.The temperatures used in experimentally manipulated experiments have increased over time.
This may reflect an increased focus on climate change and the consequences of rising global temperatures, with experimental temperatures increasing to reflect biologically relevant temperatures butterflies may experience in the future.The length of experimentally manipulated temperature regimes has decreased over time, implying a shift in focus from long-term exposure to short-term exposure experiments.This too may reflect an increased focus on the effects of extreme temperature events on butterflies, with an eye to predicting the consequences of climate change.These focuses will be valuable for understanding the response of butterflies to short-term temperature changes outside of what is currently being naturally experienced, but not longer-term changes.

Research outcomes
We found a high proportion of outcomes recorded at the adult life stage, particularly for ecological and population-level outcomes.The higher representation of outcomes recorded at the adult life stage implies that more is known about the impacts of temperature change on adult butterflies than other life stages.As life stages can be differentially impacted by temperature (Radchuk et al., 2013), and lessmobile larvae may be more sensitive to temperature change than adults, this could mean that findings do not accurately reflect species' responses to temperature across their entire life cycle (e.g., Ashe-Jepson et al. 2023b).Adults tend to be more conspicuous than the other life stages, with response traits that are easy to record (e.g., emergence time and changes to wing patterns), particularly as part of citizen science projects and long-term monitoring datasets.
However, to our knowledge, all these schemes collect data on adult butterflies only.The increased use of these datasets and the investigation of changes in response to widely available climate variables may be responsible for the uneven recording of outcomes.These datasets may also be responsible for the high representation of ecological outcomes, such as changes to phenology, particularly in the last 20 years.Butterfly monitoring schemes often record variables relevant to phenology, such as first emergence or flight period, and are often recorded at scales applicable to population-level outcomes.The strong representation of ecological and population-level outcomes is likely to result in a greater understanding of the effects of temperature on temperate adult butterfly ecology, particularly phenology, and may have informed a range of effective conservation interventions that have been applied to protect butterflies from the effects of climate change in temperate regions in recent years (e.g., Bonoan et al., 2021).
In contrast, there were few studies on the effects of temperature on butterfly behaviour.This may be due to behaviour being one of the few outcomes restricted to particular life stages (larvae and adults), but may also be the result of the complexity involved in quantifying behavioural responses to temperature.Behaviour can be environment-specific, species-specific or sex-specific, and it can be challenging to capture natural behaviours in unnatural settings, such as in ex situ studies.Behaviour is also rarely captured in citizen science projects, making the data less readily available than ecological or population-level responses.Often studies reporting behavioural responses to temperature depend on human observations over short time periods, which require abundant species with conspicuous behaviours, such as male territoriality in P. aegeria (e.g., Vande Velde et al., 2011).These challenges associated with recording and quantifying behavioural responses to temperature may explain its lack of representation in the literature.However, behaviour can mediate the impacts of changing temperatures (Stevenson, 1985), for example, by shifts in habitat use (Ashton et al., 2009), behavioural thermoregulation (Kemp & Krockenberger, 2002), or oviposition preference (Eilers et al., 2013).Behavioural thermoregulation can help species persist outside of their climate niche, but recent evidence also suggests that it may hinder adaptation to thermal tolerance by reducing the selective pressures that individuals are exposed to (Ashe-Jepson, Cobo, et al., 2023;Buckley et al., 2015).As a result, behaviour may play a critical role in population persistence under future climate change.
The lack of understanding of butterfly behavioural responses to temperature may provide inaccurate predictions of species responses to climate change.
For the non-adult life stages, the majority of outcomes recorded were developmental or physiological.Developmental responses to temperature are generally well-studied in insects (Ratte, 1984), and can be important for understanding how phenology and the synchronicity of ecological interactions may change under climate change.Developmental and physiological responses tend to be conspicuous (such as time to reach specific life stages or changes in morphology) and can be restricted to early life stages, particularly developmental responses.Developmental responses to temperature can explain both environmental requirements to complete the life cycle (such as degree-day requirements), and the distributions of species (e.g., Bryant et al., 2003).This is particularly relevant when predicting species' responses to climate change, as range shifts are a particularly common response (e.g., Macgregor et al., 2019;Wilson et al., 2015).
The families differed in the outcomes recorded, with Pieridae having a higher proportion of behavioural outcomes, and Hesperiidae having a higher proportion of population outcomes.The focus on behavioural responses to temperature in Pieridae may be the result of attempts to predict how pest species may behave either during the invasion of a new environment or under future climate change (e.g., Pandey et al., 2015).Hesperiidae may be the focus of population-level outcomes due to the relative complexity in data collection for this group; they tend to be small and fast fliers (Betts & Wootton, 1988), making them challenging to accurately record or identify during surveys, particularly by the public during citizen science projects.These traits may make it challenging to collect data on wild individual butterflies during in situ studies and explain the focus on population-level outcomes reported.Hesperiids are often grassfeeders as larvae (Sahoo et al., 2017), which may make them particularly challenging to locate at non-adult life stages or captive breed for ex situ studies.

Suggestions for future research
To address the gaps highlighted, we propose the following areas for research to focus on.A shift in focus is required to address the research gap in the tropics.Long-term monitoring schemes have become a staple within this area of research, and so increasing the establishment of butterfly monitoring schemes would be a valuable asset to many regions around the world, particularly the tropics and global south.However, monitoring schemes (both new and existing) should consider including non-adult life stages in their surveys to address the gap in understanding non-adult responses to temperature.
We recognise that surveying for non-adult life stages is generally more challenging than surveying for adults; however, long-term monitoring of these life stages has been successful even in complex tropical systems (Janzen et al., 2009;Janzen & Hallwachs, 2021).To address the gap in outcomes recorded, there should be an increase in recording behavioural outcomes in response to temperature.This can be achieved through monitoring schemes including the recording of behaviour, such as activity, during surveys.Searching for eggs during surveys would also provide implications as to oviposition preferences, without the need to follow single butterflies for extended periods of 2. We found limited evidence of taxonomic gaps in our current understanding.Nymphalidae were well represented but also represent the majority of butterfly diversity.In contrast, Riodinidae were less represented.This family is concentrated in the Neotropics and includes species with complex ecological interactions with other organisms that may be at risk under changing global temperatures.To address this, future research should include more species of Riodinidae.
3. We found gaps in current understanding of butterfly responses to temperature across the life cycle, whereby more outcomes were recorded at the adult life stage than all other life stages combined.
This may be the result of the increased use of butterfly monitoring scheme datasets, which are largely focussed on the adult life stage.

(
butterfl* OR lepidopter*) AND (temperature OR therm*) AND (adult* OR imago OR pupa* OR chrysalis OR larva* OR caterpillar OR egg OR ovum OR ova).

F
I G U R E 1 The 22 most commonly researched species appearing in peer-reviewed publications from 1980 to 2020 on the effects of temperature on butterflies, ranked by frequency.Only the top 10 ranked positions are shown.Bars are coloured by family.

F
I G U R E 4 Map showing the global distribution of the number of times butterfly species sourced from each country had their responses to temperature investigated (with continent totals as text).Colour is on a scale from red (most species) to blue (fewest species) split by deciles.All 440 studies that state their study sites are shown, those which did not (11) are excluded.Map produced on mapchart.net.Are observational or experimental approaches more common and has this changed over time?Of the 3198 times species were studied, thermal regimes were manipulated 377 times and observed 2760 times.A combination of both manipulated and observed thermal regimes was used 61 times.Both manipulated and observed experimental designs have increased in frequency over time, with observed studies increasing at a greater rate (manipulated: 0.4 species studied per year, F = 17.8, d.f.= 1,39, p < 0.001; and observed: 3.7 species studied per year, F = 25.2, d.f.= 1,39, p < 0.001).Of the 377 times the thermal regime was manipulated, the majority used constant thermal regimes (n = 297) compared with fluctuating regimes (n = 45), with 37 using a combination of the two.Manipulated thermal regimes ranged from À35 to 50 C, with an average of 3.6 different thermal regimes used per study, an average difference of 5.3 C between thermal regimes, and an average range of 16.1 C between the minimum and maximum temperatures across regimes (Figure6a).There was a significant increase in maximumF I G U R E 5The cumulative number of times butterfly species that had their response to temperature investigated were sourced from tropical or temperate study sites from 1980 to 2020.The plot includes all 440 studies that stated butterfly source location.F I G U R E 6 The (a) mean temperatures and (b) durations of manipulated thermal regimes, and the (c) mean temperatures and (d) durations of observed thermal regimes for each study that investigated the response of butterflies to temperature from 1980 to 2020.Points represent individual studies.Lines show the minimum and maximum temperatures (a and b) and minimum and maximum duration (c and d) used in each study with a range.Years with multiple studies have all ranges shown.(F = 52.5, d.f.= 1379, p < 0.001), minimum (F = 10.7,d.f.= 1375, p = 0.001) and mean temperatures (F = 32.0,d.f.= 1381, p < 0.001) used in thermal regimes over time.Of the 377 manipulated thermal regimes, duration ranged from 7.2 s to 245 days (Figure6b).The thermal regime was limited to a life stage rather than to a specific time period 210 times.There was a significant decrease in maximum (F = 48.4,d.f.= 1132, p < 0.001), minimum (F = 28.4,d.f.= 1122, p < 0.001) and mean (F = 51.9,d.f.= 1122, p < 0.001) duration of manipulated thermal regimes over time.Of the 2760 times species were studied under an observed thermal regime, very few stated the minimum or maximum temperature range experienced during the study (n = 50).For the small number of studies that did state it, temperatures ranged from 0 to 35 C, with an average range of 8.0 C (Figure6c).There was a significant decrease in maximum (F = 21.8,d.f.= 1.50, p < 0.001), and mean (F = 10.4,d.f.= 1.50, p = 0.002) temperatures in observed thermal regimes over time, but no change in minimum temperatures (F = 0.60, d.f.= 1.50, p = 0.458).The duration of observed thermal regimes was greater than that of manipulated temperature regimes, ranging from 10.2 min to 173 years (Figure6d).There was a significant increase in maximum (F = 26.4,d.f.= 1.1330, p < 0.001), minimum (F = 26.0,d.f.= 1.1327, p < 0.001) and mean (F = 26.1,d.f.= 1.1327, p < 0.001) length of temperature regimes over time.What are the most commonly recorded outcomes (e.g., physiology, ecology), does this differ between families and regions and has this changed over time?The most commonly recorded outcome across all species studied was the effect of temperature on butterfly ecology (n = 2287), followed by physiology (n = 1623), development (n = 1407), evolution (n = 1318), population (n = 769) and finally behaviour (n = 208) (Figure 7).The number of times species were studied and a broad outcome category was recorded has significantly increased over time for all categories, but at different rates (χ 2 = 2049.1,d.f.= 200, p < 0.001); the number of times ecological outcomes have been recorded has increased at the greatest rate (2.9 species studied per year, χ 2 = 2972.5,d.f.= 40, p < 0.001), followed by population (1.6 species studied per year, χ 2 = 2111.8,d.f.= 40, p < 0.001), physiology (0.67 species studied per year, χ 2 = 886.4,d.f.= 40, p < 0.001), behaviour (0.31 species studied per year, χ 2 = 289.3,d.f.= 40, p < 0.001), development (0.15 species studied per year, χ 2 = 112.7,d.f.= 40, p < 0.001) and evolution (0.12 species studied per year, χ 2 = 143.9,d.f.= 40, p < 0.001) (Figure 7).The frequency of outcome categories differed significantly across the six butterfly families (χ 2 = 172.6,d.f.= 25, p < 0.001).This was driven by behavioural outcomes being recorded more frequently in Pieridae, and population outcomes being more frequently recorded in Hesperiidae.

F
I G U R E 8 Percentage of times butterfly species were studied and an outcome was recorded in response to temperature at specific life stages from 1980 to 2020, split by outcome and stacked by life stage.Includes all studies.
time.Experimentally manipulated thermal regimes should consider increasing the use of ecologically relevant fluctuating temperatures outside of the normal range butterflies experience in their natural environment, to replicate those which might be experienced under future climate change.Experimentally manipulated thermal regimes should also consider running experiments over longer periods of time, as we detected a decrease in regime duration over time, to improve understanding of butterfly responses to long-term changes in temperature.Finally, a wider variety of species should be studied ex situ to prevent understanding of what could be species-specific responses being generalised across all butterfly species.CONCLUSION 1.We have described the distribution of the last 40 years of research on butterfly responses to temperature, and identified taxonomic, geographic and experimental gaps in current understanding.
To address this, future research should either develop new or expand existing butterfly monitoring datasets to include non-adult life stages, or new studies should consider including non-adult life stages in their data collection.4. We found a large gap in current understanding of tropical butterfly responses to temperature.The dominance of temperate butterflies in the literature could be the result of the increased use of butterfly monitoring schemes, which are largely concentrated in temperate regions.The consequence of this is a lack of understanding of species' responses to climate change in the tropics, where the majority of global biodiversity is concentrated.To address this, future research should focus on tropical systems, where long-term monitoring schemes could be established.5.There was a large disparity in experimental approaches, whereby in situ studies, with observed thermal regimes, were far more common than ex situ studies, with generally constant experimentally manipulated thermal regimes.The dominance of in situ studies implies a good understanding of the impacts of real-world temperature change on butterflies within their ecological context.The scarcity of ex situ studies and the decreasing durations of studies over time implies a lack of understanding of the long-term impacts of temperatures outside of those currently experienced, and the dominance of constant thermal regimes raises concern as to the biological relevance of results.To address this, future research should consider including fluctuating thermal regimes for longer time periods.6.The majority of outcomes recorded were ecological, with behavioural outcomes being rarely recorded.This may be the result of butterfly monitoring schemes recording data that is relevant to ecology, and also the relative difficulty of recording behavioural outcomes.The result is possible inaccuracies as to how species will respond to climate change.To address this, future research should consider including behavioural responses to temperature.stages from 1980 to 2020.Plot includes all studies, including a single paper with over 1000 species studied.

Figure S6 .
Figure S6.Map showing the global distribution of the number of times butterfly species sourced from each country had their responses to temperature investigated (A) in an ex situ setting and (B) in an in situ setting from 1980 to 2020.Colour is on a scale from red (most species) to blue (fewest species), split by deciles for each map separately.All 472 studies that state their study sites are shown, those which did not (11) are excluded.Maps produced on mapchart.net.

Figure S7 .
Figure S7.The cumulative number of times butterfly species have been studied where outcomes in each of six categories were recorded in response to temperature from 1980 to 2020.Plot includes all studies, including a single paper with over 1000 species studied.

Figure S8 .
Figure S8.Percentage of times butterfly species were studied and an outcome was recorded in response to temperature from butterflies sourced from temperate or tropical study sites, stacked by outcome category and split by region.Includes all studies.