Active around the year: Butterflies and moths adapt their life cycles to a warming world

Living in a warming world requires adaptations to altered annual temperature regimes. In Europe, spring is starting earlier, and the vegetation period is ending later in the year. These climatic changes are leading not only to shifts in distribution ranges of flora and fauna, but also to phenological shifts. Using long‐term observation data of butterflies and moths collected during the past decades across northern Austria, we test for phenological shifts over time and changes in the number of generations. On average, Lepidoptera adults emerged earlier in the year and tended to extend their flight periods in autumn. Many species increased the annual number of generations. These changes were more pronounced at lower altitudes than at higher altitudes, leading to an altered phenological zonation. Our findings indicate that climate change does not only affect community composition but also the life history of insects. Increased activity and reproductive periods might alter Lepidoptera–host plant associations and food webs.

showed that the resulting environmental stress causes increased extinction rates of local butterfly populations in regions that became climatically suboptimal.Particularly species restricted to mountain areas were found to respond sensitively to rising environmental temperatures (Maharjan et al., 2023;Theurillat & Guisan, 2001).
In contrast to the well-understood climate driven changes in spatial distributions, empirical work addressing the impact of climate change on annual activity periods and voltinism is less studied and mainly restricted to temporal shifts in emergence (cf. Dunn et al., 2023;Halsch et al., 2021).In Europe, leaves now unfold earlier in spring (Wang et al., 2022), and insect adults emerge several weeks earlier than some decades ago (Vitali et al., 2023).Faltýnek Fric et al. (2020) and Teder (2020) studied the geographic variation in voltinism of European Lepidoptera to identify species traits that might predispose species to change their phenology and become multivoltine.As a consequence of altered activity times, species interactions and food chains might get broken (Saunders et al., 2023;Schweiger et al., 2012), with possible repercussions (or effects/impacts) on successful organismal development and survival (Renner & Zohner, 2018).However, local trends in phenology might strongly vary across regions, making generalization from single case studies challenging (Dunn et al., 2023;Forrest, 2016).
Prolonged vegetation periods and increased temperature might also prolong the annual activity of arthropods (Altermatt, 2010;Freimuth et al., 2022;Hällfors et al., 2021; Figure 1).For instance, Bentz et al. (2010) linked the delayed diapause and extended reproductive capacity of selected spruce beetles with warmer autumn temperatures.Tick activity is prolonged in warmer autumns (Hancock et al., 2011) and the delayed occurrence of first frost nights might explain the survival of adult Lepidoptera until late in autumn or early winter (Merckx et al., 2021).However, apart from a few case studies on single species, it is largely unknown to what degree prolonged vegetation periods affect the phenology of insects and other arthropods.For their development, insects often depend on specific host plants as larval food.Their hormonal cycles are not only temperature but also day length driven (Mukherjee et al., 2020).
Given the rather species-specific reactions of plants to warmer autumns, we might not expect uniform trends of adult insects toward prolonged autumn activity but rather species-specific patterns and diffuse trends (Forrest, 2016;Li, Wang, et al., 2023).Given the scarcity of comprehensive studies addressing the question of changes in activity periods, it is largely unknown how many insect species prolong their annual activity period and have additional generations.
In mountain regions like the Alps, climate change has differential impacts.In the western Alps, increasing annual temperatures can be seen across a wide altitudinal gradient (Beniston et al., 1994;Dumas, 2013).However, this trend is not continuous since comparatively higher than average annual temperature in the 1940s to 1960s and since 2000 were interrupted by two decades of cooler weather conditions (Beniston, 2006;Dumas, 2013).Such long-term variability has to be accounted for when inferring insect life history trends.Additionally, rising regional temperatures tend to modulate the altitudinal gradient making lower and higher altitudes increasingly less different in annual average temperatures (Beniston, 2006;Ohmura, 2012).This fact is backed by the observed uphill shift of many lowland butterfly species (Bonelli et al., 2021;Habel et al., 2023).However, long-term observations in mountain regions also indicate increased variability of weather conditions at higher altitudes that might counteract phenological variation (Barry, 2008).
Therefore, analyses of insect life history and phenology need the inclusion of altitudinal gradients.We are not aware of studies that simultaneously trace insect phenology in time and across different altitudes.
In this study, we therefore take advantage of a large long-term data set on Lepidoptera (butterflies and moths) from northern Austria.Prior work on part of these data (i.e., the butterflies) revealed community-wide reactions of the insects to increasing annual temperatures.These include an uphill shift of occurrences of mountain butterfly species (Rödder et al., 2021) but also lowland taxa (Habel et al., 2023), predicable changes in community composition toward habitat generalists and dispersive species (Habel et al., 2023), and synchronized population fluctuations at regional scales (Ulrich, Schmitt, et al., 2023).Here, we use all available data that comprises observations of adult butterflies and moths with precise information about the geographical location and time (day) of all records.Based on these data, we aim to answer the following questions: 1. Do species shift their activity periods towards earlier spring and later autumn occurrences of adults in response to the warming climate?
2. Do such prolonged adult activity periods lead to additional generations per year? 3. Do these trends differ across the altitudinal gradient?F I G U R E 1 Temperature increase might force spring species or mobile (dispersive) species to appear earlier in the year (red phenology) with respect to the original state (green), and to prolong their time of appearance.These species might have additional generations (broken line).Autumn species (grey) might appear later in the year with prolonged phenology.Such species might have flatter temporal distributions with less pronounced peaks.We assessed the shifts in the phenological distribution from the change in the positions (in days) of the upper (CL early ) and lower (CL late ) onesided confidence limits of this distribution (the day where already 5% of individuals were recorded in spring or await recording in autumn).

| Dataset
We used data on Lepidoptera (butterflies and moths) recorded in the federal state of Salzburg (northern Austria) and stored in the entomological collection of the "Haus der Natur", Museum of Natural Sciences in Salzburg (https:// www.hausd ernat ur.at/ en/ ).Data were collected by different entomologists and kept in the form of collections and observation lists.The dataset was completed by recent Lepidoptera assessments, from diverse literature sources, and various mobile apps (e.g., Obser vation.org).All this information is stored in a single online data base.
In total, the data contain 250,108 single records of 2275 lepidopteran species, collected in the time frame of 1900-2022 across a large altitudinal gradient (380-3105 m a.s.l.).Exact dates (day, year) and places of capture (converted to a GPS coordinate) were available for 247,371 records of 2264 species.The coordinates were recalculated to altitude.As the precision of these coordinates is limited to about 15 m, we used the average altitude of a square of 100 m 2 around each coordinate to assess the record's altitude.We only considered observations of adults, but not records of other developmental stages (e.g., eggs, larvae, pupae).The complete set of raw data together with extracted phenologies and kernel density analyses are available in the figshare repository (Ulrich, Habel, et al., 2023).
For the present study, we constructed three types of time windows containing species time series (phenologies) from the cumulated annual records.We used the data since 1960 to ensure sufficient annual sample sizes >500 records (cf.Habel et al., 2022).
Each time series was based on daily records (a vector with 366 rows).
We excluded records from the months of January, February, and December as these records were possibly either erroneous notations or hibernating adults.Within each time window, we categorized the records into three altitudinal bands (<800 m a.s.l., 800-1500 m, and >1500 m).We first cumulated all records between 1960 and 1990 into the time window T pre , with which we compared two other more recent types of time windows.For these, we second constructed seven shifting windows (1985-1995, … 2005-2015, 2010-2020, 2015-2022) each containing cumulated records.The shifting window grouping resulted for each species in 7 time series × 3 altitudinal bands (total of 13,584 time series).Since each of these seven time windows contained a comparably small number of phenologies that did not allow for a comparison of altitudinal bands, we third also cu-  60 (1985-1995) to 222 (2005-2015) single time series (total of 1016 series).This restriction to occurrence in both time series ensured that we only compared intraspecific phenologies and not changes in phenology at the community level due to differential species composition.Appendix S1 contains all time series of the 486 T pre and T post series with more than 50 individuals and an overlay figure for visual inspection.

| Statistics
To assess the numbers of generations and the days of maximum appearance (the generation peak), we first smoothed the time series by kernel density estimation with Gaussian kernel.As larval development during the vegetation period (between the first and the following generations) generally lasts from 2 to 4 weeks, we used a fixed band width of 15 days.Lower band width caused a false detection of generations in a few cases while a larger bandwidth failed to detect additional generations in cases of low record numbers.The raw data of all 13,584 time-series together with the fitted kernel density estimates, and the Fortran source code for calculating the time series are contained in Ulrich, Habel, et al. (2023).
For each of the 486 kernel density series with at least 50 records, we determined the position (in days) of the single peak (univoltine species) or peaks (bi-or multivoltine species).We then assessed the numbers of species that shifted towards earlier and later peak positions, the difference (in days) of peak positions, and the difference in the number of generations.This analysis was done separately for each time period × altitudinal band combination.
In each analysis (shifting windows and cumulated data), we compared the phenologies of both time windows separately for each species and altitudinal band which occurred in both time windows.
The assessment of earliest and latest appearances was hampered by unequal sample sizes of the time windows.Lower sample sizes reduce the chance to record specimen at very early or late days.
To adjust sample sizes, we first drew random samples of single records from the time series with more records (most often the series with data from 2000 to 2022) until the record number of the smaller series was reached.We repeated this sampling 1000 times.This provided a random distribution of the rarefied time series with which we compared the smaller one.To assess changes in the early and the late appearances, we used standardized effect sizes SES = d obs − d exp ∕ exp ; where d obs and d exp denote the observed and expected appearance days (the latter being the arithmetic mean of the random distribution) of the two time windows, and σ exp is the standard deviation of the random distribution.To estimate the days of earliest and latest activity, we define two threshold days of early and late appearances before which the 5% earliest (CL early ) and after which the 5% latest (CL late ) individuals had been recorded (5% single-sided confidence limits of a time series, Figure 1).SES late was calculated only for species being at least bivoltine.We did not use observed earliest and latest records because single outlying records might distort the results.We also calculated SES kurtosis values for excess kurtosis [ = E(X− ) 4 4 − 3, where E(X − ) 4 refers to the fourth central moment of the distribution and μ and σ denote the respective mean and standard deviation], measuring the importance of outliers of a distribution, in the present case of very early and very late appearances.Here, we focused on SES values above 1.65 and below −1.65 that indicate significant shifts in phenology at the 5% single-sided error level assuming a normal distribution of the random samples.Finally, we calculated for all bi-or multivoltine species the differences between the later (T post ) and the earlier (T pre ) time windows for the average days of the first annual occurrence peak [i.e., ΔP early = P early (T post ) − P early (T pre )] and the late annual occurrences peak [i.e., ΔP late = P late (T post ) - P late (T pre )].Identically, we calculated the differences ΔCL early and ΔCL late in the position of the early (CL early ) and late (CL late ) threshold days of the time series, number of species that increased or decreased their number of generations ΔS peak , and the respective difference in SES kurtosis (ΔS kurtosis ).The Fortran software code for these calculations is available together with the raw data in Ulrich, Habel, et al. (2023).Statistical inference was done with two-way ANOVA, contingency table analysis with Monte Carlo permutation χ 2 -testing, and ordinary linear least squares regression as implemented in spss 29.0.

| RE SULTS
Comparison of the period 1960-1990 (i.e., T pre ) with the shifting 10years windows (starting with 1985-1995) revealed major temporal changes in lepidopteran phenologies (Figure 2).We found significant evidence [p(F) < .001]that flight peaks in spring appeared increasingly earlier than in the period 1960-1990 (Figure 2a).The last peaks of lowland bi-or multivoltine species appeared significantly [p(F) < .001]later in the year (Figure 2b).The random sample model indicated that the adults of 11.9% of these 82 species are occurring later in the year in the time window 2000-2010 in comparison to 1960-1990 (Figure 2b).Since 2000, these shifts to earlier dates in first generation flight periods were 6.5 ± 4.0 (mean ± two standard errors) to 10.6 ± 1.9 days at lower and 12.8 ± 9.3 to 24.1 ± 14.5 days at intermediate altitudes (Figure 2a).At lower altitudes, last gen-  dates since 2005 (Figure 2b).Similarly, between 14 and 22 species annually increased the number of generations within each shifting time window since 2000 (Figure 2c).However, the data for intermediate altitudes was based on a small number of cases (Figure 2c).
The results based on two fixed time windows (i.e., 1960-1990: T pre ; since 2000: T post ) were consistent with the ones obtained for the analyses based on shifting windows.Thus, more species had significantly earlier than later annual adult appearance since 2000 compared to 1960-1990, with the difference being significant at lower altitudes, existing but not being significant due to small numbers at intermediate altitudes, and not existing at higher altitudes (Figure 3b).The first appearance peak of adults shifted on average by 13 and 11 days to earlier dates at lower and intermediate altitudes, respectively; the difference was negligible at higher altitudes (Table 1).Such a shift was visible for 146 species (76.0%out of 192) at lower, 26 species (72.2% out of 36) at intermediate, and 9 (60.0%out of 15) species at higher altitudes, while opposite trends appeared in only 44 (22.9%), 6 (16.7%), and 5 (33.3%) species, respectively (Table 2).
Late in the years, we found evidence for a later temporal activity of Lepidoptera adults since 2000.At lower altitudes, 14 bi-or multivoltine species (10.9% of species in this group) significantly shifted their annual flight periods towards later dates (Figure 3c).The peak average of last generations shifted by 5 days (lower) and 21 days (intermediate altitudes), but the latter value is based on a small number of species, hence just depicting a trend (Table 1).At lower altitudes, but not at intermediate and higher altitudes, kurtosis quantifying the proportion of outliers was significantly smaller in T post in comparison to T pre (Figure 2d).The shifting windows indicated that lowland species had comparatively lower numbers of outliers (negative ΔS kurtosis ) in the earlier time windows (1990)(1991)(1992)(1993)(1994)(1995)(1996)(1997)(1998)(1999)(2000) in comparison with the more recent ones (Figure 3d).We also found a prevalence of negative ΔS kurtosis values in the lowland and positive ΔS kurtosis at intermediate altitudes (Figure 2d).
We observed 48 species (25.0% of all 192 species) at lower altitudes to have increased the numbers of generations since 2000 compared to 1960-1990, while the opposite occurred only in 11 species (5.7%) (Figure 3a).At intermediate altitudes, more species decreased (four species) than increased (two) their number of generations; however, the difference was low and not statistically significant (Figure 3a).Nevertheless, the average duration of annual adult appearance over all species was mostly stable over time at lower altitudes, but strongly prolonged by on average 22 days at intermediate altitudes (Table 1).We also found significant correlations between ΔN peak and the duration of the adult activity period for lower and intermediate altitudes, with more generations, shifting the onset of their flight period to earlier and its end to later dates (Figure 4).

| DISCUSS ION
Answering to our first starting question, we corroborate prior work showing earlier annual appearance of insect adults and birds (Li, Zhang, et al., 2023;Vitali et al., 2023, with references therein).For example, long term data on western and Central European longhorn beetles (Vitali et al., 2023) and lepidopterans (Asher, 2001;Richert, 2018) revealed shifts of about 2-4 weeks within the past decades.In comparison to these lowland data, we reported an average shift of mountain Lepidoptera by 13 days at lower and 11 days at intermediate altitudes (Table 1), implying a phenological shift of about 3 days per decade since 1990.These results match current estimates on climate induced phenological changes (Kharouba et al., 2018;Menzel et al., 2006).This observed trend includes species such as Vanessa atalanta, Macroglossum stellatarum, and Agrotis ipsilon, which previously migrated mostly to Central Europe from southern overwintering retreats in spring, but for several years have been observed to more and more overwinter successfully in Central Europe due to increasingly mild winters (Burton & Sparks, 2003).Likewise, typical spring species which overwinter as pupae (such as Anthocharis cardamines, Pieris rapae, Callophrys rubi, Celastrina argiolus, Pararge aegeria) have also been shown to fly earlier, as these species eclose above a certain temperature threshold independent of day length (Burton & Sparks, 2003).Earlier appearance was also observed in species, which hibernate as adults and can take direct advantage of the mild climate in spring (e.g., Gonepteryx rhamni, Aglais io, A. urticae).Consequently, in our study region, typical spring butterflies like A. io (17 days earlier), Papilio machaon (19), Leptidea sinapis agg.( 22), or Gonepteryx rhamni (26), but also early flying moth species like Orthosia incerta (5) and O. gothica (7) shifted their flight periods considerably.
Despite the overall trend towards earlier annual flight periods (Table 1), we also found species now having their first generation later in the year, although these shifts were at most moderate.
For instance, the adults of Boloria euphrosyne and Parnassius mnemosyne, both typical light forest species (Reinhardt et al., 2020), appeared 9 days and 4 days later, respectively, since 2000, while Pieris bryoniae, a high-altitude species (Reinhardt et al., 2020), even was 17 days later.However, these species with later adult appearance of the first generation are not typical early spring species, and other factors, for instance the interplay between precipitation and temperature, might have influenced these changes.

Of the typical spring species, only Celastrina argiolus and
Pyrausta purpuralis delayed the flight period of their first generation by more than 1 week.In this respect, Ellwood et al. (2012) already reported mixed responses of Japanese insects to global warming depending on annual drought periods.Furthermore, earlier appearing Lepidoptera also need their host plants to develop in parallel.Earlier leaf unfolding and flowering, as well as altered seed production cycles, have already been reported in many plants (e.g., Hacket-Pain & Bogdziewicz, 2021;Menzel et al., 2006;Parmesan & Yohe, 2003), but increased periods of drought in late spring and summer might delay or even cease the development of some plants and consequently of the insects feeding on them (Li, Zhang, et al., 2023).Climatic change does not only lead to an earlier onset of plant growth and activity of their insect hosts.The vegetative phase as a whole should be prolonged in association with delayed leaf fall; however, this pattern might be much less pronounced than the rather consistent spring shift (Forrest, 2016;Li, Wang, et al., 2023).We expected to see highly species-specific responses to the interplay of temperature, precipitation, food plant availability, and predator abundances at the end of the flight season.Nevertheless, our longterm data indicate a clear signal of extended butterfly and moth flight periods towards the end of the year (Figures 2 and 3; Table 1).
In 29 multivoltine species, we obtained prolonged activity at the end of the flight season for the last generation, on average by 6 days.
These include butterflies commonly active in autumn like A. io (31 days later), V. atalanta (39), and Lycaena phlaeas (28).At lower altitudes, these species are now found regularly until late October, with single records even in December and January.Of the typical autumn moths, Operophtera brumata (7 days) and Erannis defoliaria (3 days) TA B L E 1 Summary table providing sample sizes (N 1st Gen , N last Gen ), mean days for the peaks of the first and last generation (the latter excluding univoltine species), the earliest (CL early ) and latest (CL late ) 5% of records, and the annual duration (D in days) of appearance (D = CL late − CL early ).In time series, kurtosis quantifies the proportion of outliers and the tendency towards narrow temporal distributions.With respect to phenological series, it is a measure of the scatter of occurrence.

Altitude
Therefore, under the assumption of prolonged annual activity periods of adults, we expected to see an increase in kurtosis in recent decades.This was not the case (Figures 2d and 3d).Particularly at lower altitudes, we found significantly decreased kurtosis in T post .
Apparently, the shifts in activity period caused very early or very late appearances which are no more isolated outliers and that annual activity today is more continuously distributed between March and October in comparison to the narrower activity periods in T pre .
Additionally, it is well known that insect longevity decreases with increasing environmental temperatures (Karlsson & Wiklund, 2005).
This dependence might reduce the number of very late records despite of prolonged activity periods.
A prolonged vegetation period might also allow for additional generations, particularly in bi-and multivoltine species.Indeed, and answering to our second starting question, we found an increase of generations per year in 21.9% of species across all altitudes and a close correlation between this increase and the prolongation of their flight period (Figure 4).As predicted, obligatorily univoltine species did not prolong their annual flight periods.Both trends are best explained as a reaction to increasing temperatures during the past four decades (Altermatt, 2010;Pöyry et al., 2011).
Indeed, Freimuth et al. (2022)  from higher altitudes above 1500 m a.s.l.Probably, the rather short higher altitude vegetation periods still do not allow for additional generations, especially as many species at these altitudes, in particular high mountain taxa as many Erebia species, often need 2 years for their development (Stettmer et al., 2022).If such biannual species shift to annual, univoltine cycles, they need to adjust their entire flight periods to later dates and the whole development from egg to adult has to be "squeezed" into a single year (cf.Altshuler & Dudley, 2006).The fact that as much as 33% of species at higher altitudes prolongated their flight periods might be a first indication for such an adaptation.
In addition, climate is much more than just temperature, and climate change is much more complicated than just rising temperatures.This is particularly important in high mountain areas where snow cover and its duration are the most important influencing factors on these ecosystems (Löffler, 2007;Wang et al., 2018;Xu et al., 2022).Importantly, Morán-Tejeda et al. ( 2013) revealed for the Swiss Alps a turning point at 1400 m a.s.l.(±200), below which snow cover is mostly correlated with temperature, but above which it is mostly triggered by the amount of precipitation.While increasing temperature is a fact generally agreed on, precipitation models for the Alps predict constant annual amounts, but with increasing values in winter and declines in summer (Beniston, 2006;Gobiet & Kotlarski, 2020;Smiatek et al., 2009).Indeed, Beniston (2006) and Hüsler et al. (2014) reported constant or even increasing winter length (the period of snow cover) in the Alps during the second half of the 20th century.The average snow cover at high altitudes was strongly variable with obvious temporal trends (Hüsler et al., 2014;Rohrer et al., 1994).These climate facts explain our results on unchanged flight periods at higher altitudes because the disappearance of snow cover is a prerequisite for an earlier onset of flight.However, future models predict shrinking winter length for the next decades (Gobiet et al., 2014) and higher altitude lepidopteran populations might also adjust their phenology (Konvicka et al., 2021).Whether these predicted and the observed phenological changes at lower altitudes will be detrimental, neutral or even positive for the affected populations is still heavily debated (e.g., Konvicka et al., 2016;Roland & Matter, 2016).The fact that the butterfly communities in the present study region on one hand face compositional homogenization with increasing dominance of a few widespread habitat generalist species (Habel et al., 2022(Habel et al., , 2023) ) and on the other hand still uphold their species diversity (Gros, 2023;Habel et al., 2022)  writing -review and editing.
mulated data from 2000 to 2022 (the T post time window) and compared the phenologies of the time windows T pre and T post .The latter comparison allowed for a more precise assessment of recent temporal shifts in phenology.Sufficient sample sizes are necessary to reliably assess the important parameters of phenology.We only use time series containing at least 50 records, leaving for T pre and T post comparisons 486 annual time series of 223 species (384 time series from 192 species at low, 72 series from 36 species at intermediate, and 30 series from 15 species at high altitude).The shifting windows contained from erations' peaks shifted from 14.8 ± 10.5 to 22.8 ± 12.1 days to later F I G U R E 2 Summary results of shifting window analyses of Lepidoptera phenologies (seven windows in the time frame 1985-2022) in comparison with the T pre window (1960-1990).Year denote the midpoints of the 10-year time windows.Differences in the days of the peak of the first [(a) ΔP early ] and the last generation [(b) ΔP late ].(c) Numbers of species which increased their number of generations (ΔS G ).(d) Numbers of species which increased or decreased kurtosis (ΔS kurtosis ).Data shown separately for lowland (brown bars) and intermediate altitudes (green bars).Error bars in (a, b) refer to two standard errors (approximately 95% two-sided confidence limits).Linear regression lines (given are the numbers of data points N and the coefficients of determination r 2 ) are parametrically significant at *p(F) < .05 and ***p(F) < .001.In (b) intermediates altitudes are not shown due to the low number of data points in each window (cf.

F
I G U R E 3 (a) Numbers of species (S) that decreased (yellow bars) or increased (blue) the number of generations between the time windows T pre and T post (≥2000) across three altitudinal bands.(b, c) Numbers of species with standardized effect sizes (SES) of SES early (b) and SES late (c) being smaller than −1.65 (green, earlier appearance in T post ) or larger than +1.65 (brown, earlier appearance in window T pre ).In (c), only species of at least bivoltine phenology were included.(d) Numbers of species with SES kurtosis being larger than +1.65 (green, lower proportions of outlying recording days in T post ) or smaller than −1.65 (brown, higher proportions of outlying recording days in T post ).Sample sizes at lower (<800 m a.s.l.), intermediate (800-1500 m), and higher altitude (>1500 m): 170, 62, and 17 species, respectively.Included are only species, which jointly occurred in both time windows and which had at least 50 records per time series.Permutation based contingency table significances: *p(χ 2 ) < .05,**p(χ 2 ) < .01,***p(χ 2 ) < .001.
prolonged their annual flight activity period.However, 31 species were found to shorten their flight activity period in late summer and autumn, 10 days on average.In five moth species, this shortening was even linked to a reduction in the number of generations.These shifts in both directions demonstrate the highly species-specific reactions to climate change.Further studies need to infer which species traits and which climatic parameters are particularly linked to changes in phenology.
might corroborate both sides.In conclusion, our results show how fast alpine Lepidoptera adjust their phenology to changing environmental conditions.In less than three decades, a modified phenological zonation along the altitudinal gradient appeared, where earlier activity in spring, prolonged flight activity in autumn, and a tendency towards multivoltinism prevail at lower altitude.Higher in the mountains, the effects of rising temperatures are still counterbalanced by winter length and snow cover.Future studies have to reveal how and how fast these changes affect food web structures and pave the way to new ecological adaptations.Conceptualization; project administration; validation; writing -review and editing.Thomas Schmitt: Data curation; validation; writing -review and editing.Patrick Gros: Data curation; investigation; resources; writing -review and editing.Werner Ulrich: Conceptualization; formal analysis; funding acquisition; methodology; software; visualization; writing -original draft;

Table 1
, lack of data in two time windows).
For the latest 5% of records, only at least bivoltine species were included.Data from 486 time series with at least 50 records per time series.Comparisons of temporal shift of Lepidoptera phenologies between the time windows 1960-1990 and since 2000 across three altitudinal bands.Note: Given are numbers of species for which the peak days for the first and last generations (excluding univoltine species in the latter case) was earlier or later in the later time window (i.e., since 2000).Included were 486 time series with at least 50 records per time series.
Box and whiskers plots between ΔN peak and ΔP early (a, b) and ΔP late (c, d) for lower altitudes (a, c) and intermediate altitudes (b, d).For higher altitudes, the number of data points was too low.Number in (a, b) denote the sample sizes.The changes in ΔP early and ΔP late with respect to ΔN peak were significant at p(F) < .001 in both altitudinal bands.
specific shifts in phenology (Figures2 and 3; Table1).Unfortunately, the differences in sample sizes among the altitudinal bands used here do not allow for straightforward interpretation as the density of data reduces with altitude.Nevertheless, there is strong indication that phenology shifts were particularly evident at lower and intermediate altitudes, but mostly or even completely absentF I G U R E 4