SEARCH

SEARCH BY CITATION

Keywords:

  • climate change;
  • light trap;
  • moth;
  • photoperiod;
  • temperature;
  • thermal sensitivity;
  • thermal sum

Abstract

  1. Top of page
  2. Abstract
  3. Introduction
  4. Material and methods
  5. Results
  6. Discussion
  7. Acknowledgements
  8. References
  9. Supporting Information

The magnitude and direction of phenological shifts from climate warming could be predictably variable across the planet depending upon the nature of physiological controls on phenology, the thermal sensitivity of the developmental processes and global patterns in the climate warming. We tested this with respect to the flight phenology of adult nocturnal moths (3.33 million captures of 334 species) that were sampled at sites in southern and northern Finland during 1993–2012 (with years 2005–2012 treated as an independent model validation data set). We compared eight competing models of physiological controls on flight phenology to each species and found strong support for thermal controls of phenology in 66% of the species generations. Among species with strong thermal control of phenology in both the south and north, the average development rate was higher in northern vs. southern populations at 10 °C, but about the same at 15 and 20 °C. With a 3 °C increase in temperature (approximating A2 scenario of IPPC for 2090–2099 relative to 1980–1999) these species were predicted to advance their phenology on average by 17 (SE ± 0.3) days in the south vs. 13 (±0.4) days in the north. The higher development rates at low temperatures of poleward populations makes them less sensitive to climate warming, which opposes the tendency for stronger phenological advances in the north from greater increases in temperature.


Introduction

  1. Top of page
  2. Abstract
  3. Introduction
  4. Material and methods
  5. Results
  6. Discussion
  7. Acknowledgements
  8. References
  9. Supporting Information

Many observations of climate change responses in natural systems come from shifts in phenology (Parmesan, 2006), that is, the seasonal timing of (periodic) life cycle events (Rathcke & Lacey, 1985). Two quantitative, globally comprehensive meta-analyses have both shown stronger phenological shifts towards earlier spring events at higher vs. lower latitudes (Root et al., 2003; Parmesan, 2007). Stronger phenological shifts at higher latitudes could arise because temperatures are increasing more towards the poles (IPCC, 2007), or because an equivalent increase in average temperatures make a larger proportional contribution towards annual thermal sums in high latitude systems where the summers are already short and cool. Also, the phenology of high latitude species could be generally more sensitive to climate because the costs of inappropriate phenology are probably higher than at lower latitude systems with less seasonality (Pau et al., 2011). However, there could be additional, and possibly opposing, effects from latitudinal trends in the physiological sensitivity of organisms to temperature. For example, climate warming could increase the metabolic rates of tropical species more than that of high latitude species (Dillon et al., 2010; see also Deutsch et al., 2008; Tewksbury et al., 2008).

Biological responses to temperature arise from thermodynamic control of biochemical rates (e.g. van der Have & de Jong, 1996). Many biological processes, including development rates of poikilotherms, exhibit similar unimodal thermal responses where rates increase more than linearly from low to moderate temperatures, then approximately linearly through moderate to warm temperatures, until finally decelerating to a maximum beyond which rates decline rapidly (e.g. Logan et al., 1976; Fig. 1a). Thus, there is almost universally a range of cool to at least moderately warm temperatures where increasing temperature yields increasing development rate of poikilotherms. Other things being equal, increasing development rate yields decreased development time, and therefore tends to advance the seasonal timing of life history events (e.g. the time of reproduction in insects or flowering in plants). Therefore, there is a strong theoretical expectation for climate warming to yield advanced phenology in general, and, other things being equal, greater advances in phenology at high latitudes where there is greater warming. However, this general tendency could be modified by latitudinal patterns in physiological responses to temperature.

image

Figure 1. (a) General function of dependence of biological processes on temperature, and (b–d) three forms of genetically based variation in the thermal response. Development rate (1 days−1), development time (days) and predicted shift in in phenology of two theoretical populations with same (e–f) or different (g–h) slope of temperature response, across a range of temperatures.

Download figure to PowerPoint

Developmental responses to temperature vary broadly among species (Honěk, 1996; Trudgill et al., 2005) and populations within species (e.g. Lonsdale & Levinton, 1985; Ayres & Scriber, 1994). For example, poleward populations of Papilio canadensis in Alaska develop faster, especially at low temperatures, than midlatitude populations in Michigan (Ayres & Scriber, 1994). Genetically based variation in the form of thermal response can include vertical shifts (‘faster-slower’), horizontal shifts (‘hotter-cooler’), or ‘generalist-specialist trade-offs’, that is, variation in the width of the thermal response function (sensu Izem & Kingsolver, 2005; Fig. 1b-d). Such genetic variation can enable populations to adapt to different temperature regimes (Miller & Castenholz, 2000). If as a result there are predictable latitudinal patterns in physiological responses to temperature, this could amplify, weaken, or reverse (depending on the details) the tendency from global temperature trends alone to advance phenology more at high latitudes than at low latitudes. Three simple expectations can be derived from considering the effects of changes in temperature and thermal response functions within the range of temperatures where response functions are approximately linear (Fig. 1e–h): (i) because development time is the inverse of development rate, equivalent warming has larger effect on development time when the baseline temperatures are cooler (warming scenario 12–>15 °C) vs. warmer (15–>18 °C); (ii) if poleward populations (HL) tend to have more rapid development rates than midlatitude populations (ML), especially at cool temperatures (Fig. 1e–f), then their development times would be less sensitive to warming, especially at cooler temperatures (smaller change in HL development time vs. ML in both warming scenarios); (iii) if poleward populations have relatively higher development rates at cooler temperatures and concomitant relative decreases in development rates at warmer temperatures (Fig. 1g–h), as predicted by ‘generalist-specialist trade-offs’ in thermal responses, this also has the effect of decreasing the effects of warming on development time.

In addition to temperature, photoperiod is the other prominent environmental modifier of developmental schedules and therefore phenology in poikilotherms (Bradshaw & Holzapfel, 2010). It can cue the termination of diapause (Tauber et al., 1986; Danks, 1987), and influence development rates during active life stages (Nylin et al., 1989). In general, photoperiodic controls attenuate phenological plasticity. Interannual phenological variation would be zero in a population with complete photoperiodic control of life history schedules. Thus, latitudinal patterns in the strength of photoperiodic controls would influence latitudinal patterns in phenological responses to climatic variation: for example, if poleward populations tended to have weaker photoperiodic controls than temperate populations, then poleward populations would be more responsive than otherwise to climatic warming, and vice versa.

Studies addressing intra and interspecific variation in phenological controls (Hodgson et al., 2011; Valtonen et al., 2011) have been rare because long-term data sets covering wide latitudinal ranges and a high number of species are difficult to collect. Here, we compare the latitudinal differences in the phenological sensitivity to temperature of annual flight times in Finnish nocturnal moths. The spatially replicated time series included 362 species-generation combinations, 12 years, and >1010 km of latitude (from 60 to 68°N). The models fits were cross-validated with additional 8 years of data from the same sites. We examined whether moth communities in northern vs. southern Finland differed in the strength and form of controls on phenology from temperature and photoperiod. With the best-fit models of empirical responses of temperature, we compared the responses of northern and southern moth communities to the same forcing from scenarios of climate warming.

Material and methods

  1. Top of page
  2. Abstract
  3. Introduction
  4. Material and methods
  5. Results
  6. Discussion
  7. Acknowledgements
  8. References
  9. Supporting Information

Moth data

Our analyses were based on moth captures from a network of light traps (in total 208 traps) located in forested areas across Finland during 1993–2012 (Finnish Moth Monitoring Scheme Nocturna, coordinated by the Finnish environmental administration). Only trap data covering the entire flight season of macrolepidoptera were selected, and in other cases species flying in missing months were excluded from the data (details in Appendix S1). Data from traps located less than 10 km apart were combined and treated as the same sites, resulting in 94 sites and 620 site–year combinations. For modelling purposes, the sites had to be divided into two groups. Because thermal conditions in Finland are affected both by latitude and the maritime/continental effect, we decided to divide the sites approximating the +3 °C isocline of yearly average temperatures in Finland (Finnish Meteorological Institute, 2013b). Altogether 50 sites were in southern Finland (59.8–63.1°N) and 44 in northern Finland (62.0–68.9°N; Fig. 2).

image

Figure 2. Map of locations of the 50 southern study sites, 44 northern study sites and 50 weather stations.

Download figure to PowerPoint

For the model fitting phase, data of years 1993–2004 were used, including 701 species of macro moths. We excluded one species that has two geographically overlapping subspecies with differing phenologies, one species that can have either 2 or 1-year development time with different phenologies, and one species that partly lives inside houses, as well as 14 migratory species (Huldén et al., 2000). Of the remaining 684 species, 453 were univoltine with a nonadult overwintering stage, 20 species were univoltine with adult overwintering stage, 210 species were bivoltine, and one species produces three generations per year. For all species that overwinter as adults and for 89 bivoltine species with distinct generations, we were able to assign a cut-off date based on literature (Huldén et al., 2000), dividing the individuals into two generations (cf. Pöyry et al., 2011). This left us with 562 species and 660 species-generation combinations. Finally, only species generations with more than five individuals per site year and more than nine site years per geographical zone (south or north) during the period 1993–2004 were selected. These arbitrary thresholds were selected to ensure informative data while not excluding all rare species from the analyses. In the south, this left 322 species, 348 species generations and 1 576 473 individuals, and in the north, 195 species, 211 species generations, and 1 076 056 individuals (197 species generations occurred in both south and north). Species belonged to the superfamilies Lasiocampoidea, Bombycoidea, Geometroidea, Noctuoidea and Hepialoidea. Systematics and nomenclature follow Kullberg et al. (2008).

For the cross-validation phase, data from the same species generations and sites from the years 2005–2012 were used. Only species generations with more than five individuals per site year and more than two site years were selected, leaving us 279 species generations and 417 670 individuals in the south, and 172 species generations, and 262 684 individuals in the north.

For analyses, we scored the day of capture as the middle date of the weekly trapping period and converted this to days from the last winter solstice (details in Appendix S1).

Temperature data and thermal sum calculations

We obtained daily minimum and maximum temperatures from 50 stations of the Finnish Meteorological Institute (Fig. 2). We then estimated the daily minimum and maximum temperatures at each study site by fitting a trend surface over a grid (with least squares method) using the daily weather station observations. The trend surface was fitted with package ‘spatial’ (Ripley, 2010) in R version 2.12.0 (R Development Core Team, 2008) using a 4th order polynomial (after evaluating 1st to 6th order polynomials). For each trap site, we then downscaled daily maximum and minimum temperatures to hourly temperatures for use in model fitting (details in Appendix S1). The hourly temperatures, instead of daily average temperatures, were used in the analyses, because in many spring days the daily average temperature can be below the developmental threshold, although some hourly temperatures rise above the base (Ruel & Ayres, 1999).

Competing models for environmental controls of phenology

We compared the ability of eight alternative theoretical models to predict the peak flight dates for species generations across all years (1993–2004) and sites of each geographical zone. Peak flight dates were calculated as the median date of capture for each species generation in each year and site.

  1. The null model against which all other models were compared was the solar day (Sday) model, under which the peak flight date across years and sites was predicted to be the average day of peak flight across years and sites (Fig. 3a). If all sites experience the same photoperiodic regime (e.g. for species present only in the southernmost part of the country), Sday model describes photoperiodic control of phenology. When sites span different photoperiodic regimes, the Sday model represents the theoretical possibility that individuals are adapted to fly on about the same calendar date even when distributed across a range of thermal and photoperiodic environments.
  2. For the photoperiod (Photo) model, we predicted the peak flight date across years and sites from a threshold day length, representing minimum or maximum required day length for species generations with peak flight before or after midsummer respectively (Fig. 3a–b). In this model, the predicted day length of peak flight was the average of day lengths observed at peak flight day of each year and site (details in Appendix S2).
  3. For the model based on thermal sum (Tsum), we solved for the best-fit model to predict the peak flight dates across years and sites from a threshold thermal sum, starting to accumulate at previous winter solstice (Fig. 3c–d). The thermal sum for each hour (th) was calculated as the cumulative sum of rates of forcing:
    • display math(1)
    where t1 = first h after winter solstice and xt = the estimated hourly temperature. The rate of forcing in this model is given as
    • display math(2)
    where Tb is the parameter value for lower developmental threshold (i.e. base temperature). This involved optimizing the parameter value for Tb using function ‘fmincon’ in MATLAB (R2010a-R2012a, The MathWorks Inc., Natick, MA, 2000). The details of the optimization process are given in Appendix S2.
  4. For the model based on thermal sum and solar day (Tsum∩Sday), we solved for the best-fit model to predict the peak flight date across years and sites from a threshold thermal sum, starting to accumulate after a threshold number of hours had elapsed from the previous winter solstice (t1 of formula 1). This involved optimizing the parameter values for Tb and start hour.
  5. In insects, there can be multiple pathways to diapause completion in the spring and sometimes it can be stimulated by high temperatures (Hodek & Hodková, 1988). For the model based on thermal sum and high temperature threshold terminating the diapause (Tsum∩HiT), we solved for the best-fit model to predict the peak flight date across years and sites from a threshold thermal sum, starting to accumulate after a threshold high temperature had been achieved (t1 of formula 1). This involved optimizing the parameter values for Tb and high temperature threshold (Thigh).
  6. For the model based on nonlinear effect of temperature on poikilotherm development rate (Tsum.nl; e.g. Logan et al., 1976), we solved for the best-fit model to predict the peak flight across years and sites from a threshold thermal time, where the rate of forcing followed a logistic sigmoid function (Hänninen, 1990):
    • display math(3)
    where parameter b < 0 and c > 0. This involved optimizing the parameter values for b and c.
  7. For the model based on nonlinear effect of temperature and solar day (Tsum.nl∩Sday), we solved for the best-fit model to predict the peak flight across years and sites from a threshold thermal time in the same way as in Tsum.nl model, but the thermal time started to accumulate after 3000 h (125 days, ˜ April 25) had elapsed from the previous winter solstice (t1 of formula 1), which approximates the median of optimized start values in Tsum∩Sday models across species generations. We could not optimize the value for start hour, because the method used allowed us to optimize only two parameters (here b and c).
  8. For some insect species, the adult flight could also be cued by onset of low temperatures in the fall (Masaki, 1980). For the low temperature model (LoT), fitted only to species generations with peak flight in September or later, we predicted the peak flight date across years and sites from threshold cumulative minimum temperature, starting to accumulate after midsummer.
image

Figure 3. Graph showing how the model predicted peak flight days were determined for Sday, Photo and Tsum models of Abraxas grossulariatus (southern population). (a) The average solar day (horizontal line) of peak flight days across years and sites (circles) is the predicted day of Sday model, (b) the solar day (vertical arrow) when the average length-of-day of peak flights across years and sites (vertical line in a) is reached is the predicted day of Photo model, (c–d) the solar day (vertical arrow in d) when the average thermal sum of peak flights across years and sites (vertical line in c) is reached is the predicted day of Tsum model.

Download figure to PowerPoint

Statistical analyses

The overall responsiveness of moth communities to thermal conditions of the season was tested with ancova, where the annual averages of peak flight days (calculated across all species generations in a community) at each site (censused during the entire season) was the response variable, site a fixed factor, and thermal sum (base +5 °C) accumulating until midsummer, or until the end of season, a covariate.

We compared the ability of the alternative theoretical models to predict peak flight time using the corrected Akaike Information Criteria (AICc; Anderson, 2007). If the difference in AICc between the highest and the second highest ranked model (=ΔAICc) was ≥2, the highest ranked model was assigned as the most likely (top) model for the species generation. Species generations were further classified into four phenological classes: (i) Photoperiodic control of phenology if Sday or Photo model was selected as the top model or if together they were the two most likely models (with ΔAICc < 2, but to the other models ≥2); (ii) Thermal∩Photoperiodic control of phenology if Tsum∩Sday or Tsum.nl∩Sday model was selected as the top model or they together were the two most likely models; (iii) Thermal control of phenology if Tsum or Tsum.nl or Tsum∩HiT was selected as the top model or some combination of thermal models (Tsum, Tsum∩Sday, Tsum∩HiT, Tsum.nl or Tsum.nl∩Sday) were the most likely models and (4) Other (in all other cases). For all models we calculated root mean square errors (RMSE), which describes the accuracy of the estimate in days. Averages in RMSEs of the four phenological classes were compared with anovas and Tukey's post hoc multiple comparisons. For evaluation of thermal models, we also estimated the proportion of variance explained (R2adj) relative to the solar day -model. All formulae are given in Appendix S2.

We tested for differences in the frequency of the four phenological classes between northern and southern Finland. We also conducted randomization tests to ask whether the four phenological classes were randomly distributed among the species (details in Appendix S3). To evaluate the patterns in thermal sensitivity, we selected the highest ranked models of species generations classified to Thermal∩Photoperiodic or Thermal control of phenology, but excluded those having start hour less than 1 week before the earliest observed peak of flight across years and sites (because these models could describe the effect of high temperatures enhancing flight activity of adults (e.g. Battisti et al., 2006) rather than increasing the development rate of earlier life stages). We then calculated the development rate (proportion of development/day) at 10, 15 and 20 °C. For Tsum, Tsum∩Sday, and Tsum∩HiT models the development rate at temperature t was calculated as (t - Tb)/model predicted thermal sum at peak flight. For Tsum.nl and Tsum.nl∩Sday models, the development rate at temperature t was (28.4/(1 + exp(* (t - c))))/model predicted thermal sum at peak flight.

Finally, to compare northern and southern Finland with respect to the expected phenological shift in moth flight times from climate warming, we used the historical hourly temperatures and the highest ranked Thermal or Thermal∩Photoperiodic models to calculate the average peak flight dates before and after incrementing hourly temperatures by 3 °C (approximates A2 scenario for 2090–2099 relative to 1980–1999; IPCC, 2007).

Cross-validation of model fits

We also tested how well the highest ranked models were able to predict the peak flight days in an independent, temporally nonoverlapping data set, covering the years 2005–2012. For each species generation, year and site, the predicted peak flight day was calculated based on observed temperatures and the highest ranked model. For each species generation, the model accuracy was then estimated by RMSE. All analyses were conducted using program R 2.12.0 (R Development Core Team, 2008).

Results

  1. Top of page
  2. Abstract
  3. Introduction
  4. Material and methods
  5. Results
  6. Discussion
  7. Acknowledgements
  8. References
  9. Supporting Information

Variation in peak flight days

The observed variation in timing of flight (years 1993–2004) was generally high, which provided a basis for comparing the alternative models. The root mean square error (RMSE) of the null model (Sday) averaged 9.1 (range from 4.0 to 20.9) days in southern Finland (n = 348 species generations) and 8.7 (2.4–21.9) days in northern Finland (n = 211). The phenological range from early years and sites to late years and sites was similarly high throughout Finland: averaged across species, the difference between earliest and latest peak flight day in southern vs. northern Finland was 41.9 days (range 14–105) vs. 40.3 (8–96) respectively.

At the community level, moth phenology was responsive to the thermal conditions of the season (Fig. 4). The annual averages of peak flight days, at sites censused the entire season, were negatively associated with the thermal sums at midsummer, at the same sites (years 1993–2012; ancova; effect of thermal sum in south F1, 104 = 101.9, P < 0.0001; in north F1, 350 = 264.5, P < 0.0001). The moth phenology was significantly, but more weakly, associated with thermal sums accumulating over the entire season (effect of thermal sum in south F1, 104 = 11.2, P = 0.001; in north: F1, 350 = 184.7, P < 0.0001).

image

Figure 4. Relationship between the annual average of peak flight days (calculated across all species generations in a community) and thermal sum accumulating until midsummer (at sites censused the entire season) in (a) north and (b) south (data of years 1993–2012 combined).

Download figure to PowerPoint

Competing models of controls on phenology

There was a relatively good success in identifying top models for phenological control. In 330 (59%) of the 559 species generations modelled, one competing model emerged as a better fit than the alternatives (Table 1). Models (or combination of models producing coarsely equal fit) including some form of thermal control of phenology were by far the most common: 367 (66%) of the species generations studied. The low-temperature threshold (LoT) was the only model that did not emerge as a top model to any of the species generations studied.

Table 1. Frequency (and % of total) of species-generation combinations with different top models for control of flight phenology
Top models for control of flight phenologySouthNorth
  1. a

    Model including thermal control of phenology.

Sday61 (18%)39 (18%)
Photo7 (2%)7 (3%)
Tsuma34 (10%)11 (5%)
Tsum∩Sdaya52 (15%)54 (26%)
Tsum∩HiTa9 (3%)11 (5%)
Tsum.nla13 (4%)8 (4%)
Tsum.nl∩Sdaya17 (5%)7 (3%)
LoT0 (0%)0 (0%)
Sday or Photo20 (6%)13 (6%)
Tsum∩Sday or Tsum.nl∩Sday8 (2%)0 (0%)
Thermal (=other combination of a)99 (28%)44 (21%)
Other combinations of models28 (8%)17 (8%)
Total348211

The highest ranked models fitted generally well to the observed data, the average RMSE being 7.3 days (range from 3.0 to 20.9) in southern data and 7.1 days (2.4–21.9) in northern data. The averages of RMSEs differed among the four phenological classes both in southern (one-way anova F3, 344 = 18.4, P < 0.001) and in northern data (F3, 207 = 8.6, P < 0.001; Fig. 5a–b). Both in the south and the north, the models including thermal control of phenology fitted significantly better to data compared to Photoperiodic models (see Tukey's post hoc multiple comparisons of means in Fig. 5a–b). Further details of model comparison results and summaries of optimized parameter values of the top models are in Appendix S4. For example, the lower developmental threshold was significantly lower in top Tsum∩Sday models (average = −2.9) than in Tsum models (5.0) in southern data (Welch two sample t-test; t = 8.2, df = 68.0, P < 0.001) but not in northern data (t = 2.2, df = 11.2, P = 0.054).

image

Figure 5. Boxplots showing the variation in model fits (RMSE) among four phenological classes (Photoperiodic/Thermal∩Photoperiodic/Thermal/Other) for (a–b) years 1993–2004 for which the models were fitted, and (c–d) for the cross-validation years 2005–2012. Pairs of models with significant differences in mean RMSE are shown with different letter symbol (Tukey's post hoc multiple comparisons).

Download figure to PowerPoint

Phenological classes

There were no statistical differences in the frequency of the four phenological classes (Photoperiodic/Thermal∩Photoperiodic/Thermal/Other) between south and north (χ2 = 5.7, df = 3, P = 0.13). Altogether 107 (54%) species generations, out of 197, were classified to the same phenological class in the two geographical zones (Table 2). However, in both the south and the north, there was evidence of phylogenetic structure to phenological classes from randomization test with phylogenetic trees (Appendix S3).

Table 2. Frequency of the four phenological classes for the 197 species generations with data from both south and north
SouthNorth
PhotoperiodicThermal∩PhotoperiodicThermalOther
Photoperiodic35641
Thermal∩Photoperiodic731148
Thermal1120417
Other3270

Differences in development rates between southern vs. northern populations

The average development rate (proportion of development/day) was significantly higher in northern compared to southern populations of the same species at 10 °C (n = 105; paired samples t-test; t = 4.3, df = 104, P < 0.0001), but not at 15 °C (t = 0.4, df = 104, P = 0.66), or at 20 °C (t = 0.5, df = 104, P = 0.62; Fig. 6).

image

Figure 6. (a) Average (±SE) development rate (proportion of development/day) and (b) average development time (1/development rate) in three temperatures for the 105 species classified to Thermal∩Photoperiodic or Thermal control of phenology both in southern and northern data.

Download figure to PowerPoint

When poorly fitting models (RMSE > 10 days) were removed, and the analyses were repeated (n = 99), the direction or significance of the results did not change: the average development rate was significantly higher in northern populations at 10 °C (t = 3.9, df = 98, P = 0.0002), but not at 15 °C (t = 0.4, df = 98, P = 0.67), or at 20 °C (t = 0.4, df = 98, P = 0.67).

Predicted phenological response to increase of 3 °C

The average estimated phenological shift of the 232 southern species generations classified to either Thermal∩Photoperiodic or Thermal control of phenology was 16.7 ± 0.3 days (range from 3 to 27) and of the 135 species generations in north 13.2 ± 0.4 days (3–26; Fig. 7). Of the 105 species generations with Thermal or Thermal∩Photoperiodic control both in the south and the north, the average estimated shift was 2.8 ± 0.5 days higher in southern vs. northern populations (t = 5.1, df = 104, P < 0.0001). When poorly fitting models (RMSE > 10 days) were removed, and the analyses were repeated (n = 99), the direction or significance of the results did not change: the average estimated shift was still significantly higher in southern populations (t = 4.9, df = 98, P < 0.0001).

image

Figure 7. Predicted phenological shift after an increase of 3 °C in (a) south (n = 232) and (b) in north (n = 135) among species generations with strong thermal control of phenology. (c) Difference in predicted shift between south and north among species with strong thermal control of phenology in both zones (n = 105).

Download figure to PowerPoint

Cross-validation of model fits

The highest ranked models fitted relatively well to the cross-validation data (2005–2012), the average RMSEC-V being 8.4 days (range from 2.3 to 26.5) in southern data (n = 279) and 8.5 days (2.8–25.1) in northern data (n = 172; Fig. 5). The RMSEC-V differed among the four phenological classes both in southern (one-way anova F3, 275 = 2.7, P = 0.048) and in northern data (F3, 168 = 2.8, P = 0.042; but no pairwise differences were identified in Tukey's post hoc multiple comparisons in either geographical zone; Fig. 5c–d).

When models fitting poorly to cross-validation data (RMSEC-V > 10 days) were excluded, and the analyses were repeated (n = 62), the average development rate remained to be significantly higher in northern populations at 10 °C (t = 3.2, df = 61, P = 0.002), but not at 15 °C (t = 0.5, df = 61, P = 0.60), or at 20 °C (t = 0.2, df = 61, P = 0.83), and the estimated shift remained significantly higher in southern populations (t = 5.0, df = 61, P < 0.0001).

Discussion

  1. Top of page
  2. Abstract
  3. Introduction
  4. Material and methods
  5. Results
  6. Discussion
  7. Acknowledgements
  8. References
  9. Supporting Information

Interspecific variation in phenological controls

One fundamental challenge of phenological studies is to understand which environmental controls or cues are important in dictating the phenology of different species. Based on our results, approximately two-thirds of the 183 Lepidoptera species studied here expressed some kind of thermal control of phenology. Similarly, Valtonen et al. (2011) found 51% in analyses of a subset of the same Finnish data, and 60% of the 15 butterfly species in Britain displayed phenology that varied with thermal conditions across years and sites (Hodgson et al., 2011). A literature review, covering mostly spring phenophases of plant species from tropics to arctic, revealed that thermal control of phenology was important in 86% of the >300 species (Pau et al., 2011).

In addition to distinguishing between thermal and photoperiodic controls of phenology, our results revealed a range in the forms of thermal control, both linear and nonlinear. The wide interspecific variation in phenological control mechanisms and thermal sensitivity suggest that the future climate warming is likely to shift the phenology of most, but not all, species and those that do shift their phenology will display variable responses. This will ensure that the composition of ‘phenological communities’ will change with warming climate. The capacity for adaptive phenotypic adjustments of phenology can influence population sizes, population persistence and community structure (e.g. Møller et al., 2008). If interacting species respond differently to climate change this can affect predation and plant–herbivore interactions (e.g. Visser & Holleman, 2001), competition (Gange et al., 2011), parasitism (Møller, 2010), pollination (Memmott et al., 2007) and seed dispersal (Warren et al., 2011). If species at higher trophic levels tend to have higher thermal sensitivity than species at lower trophic levels, as hypothesized by Berggren et al. (2009), we could predict that plants will advance their phenology less than herbivores, and herbivores less than their (poikilotherm) predators. Consistent with this, a meta-analysis indicated that butterflies and birds have shown more than three times larger phenological shifts compared to plants (Parmesan, 2007).

Intraspecific variation in phenological controls

A second fundamental challenge of phenological studies is to understand the intraspecific variation in phenological controls and how these could contribute to the degree of phenological shifts in the future. About half of the southern and northern populations of the same species differed in the form of their phenological controls. Northern populations with strong thermal control of phenology expressed higher development rates in low (10 °C) but not in higher temperatures. This explains the smaller phenological shifts predicted in north vs. south (our theoretical predictions 2 and 3). If no differences in development rates had been found, the equivalent warming (+3 °C in both south and north, in our case) would have been predicted to lead to larger changes in northern compared to southern populations (prediction 1).

Therefore, our initial predictions of how differences in development rates can shape differences in phenological shifts (Fig. 1) were consistent with our process-based model projections for phenological shifts across latitudes. This provides a basis for general predictions regarding phenological plasticity when there is knowledge of patterns in thermal sensitivity among species, clades, assemblages, or trophic levels. For example, if it is general that lower latitude species have higher thermal sensitivity (see also Amarasekare & Savage, 2012), we could anticipate larger phenological shifts than in analogous high latitude species given the same climatic warming. This opposes the effect of a general pattern in greater warming at higher latitudes. The opposition of these two effects could be why the meta-analyses of Parmesan (2007) found surprisingly weak empirical trends for an effect of latitude on advancing phenology (<4% of the variation in phenological shifts).

Differences in development rates can also shed light on the evolutionary pressures faced by populations at different latitudes. Higher development rates of cold source populations at low, but not at high temperatures, as found here and in some other poikilotherm species (Lonsdale & Levinton, 1985; Ayres & Scriber, 1994), matches most closely the outcome of the ‘generalist–specialist trade-off’ in thermal performance curves (sensu Izem & Kingsolver, 2005). Differences in development rates are important also in another fundamental way: thermal sensitivity of development, fecundity and mortality (all with different thermal performance relationships) together determine the thermal sensitivity of fitness and population growth rates (Amarasekare & Savage, 2012).

Predicted phenological response to increase of 3 °C

A third fundamental challenge of phenological studies is to predict the degree of phenological shifts in the future. Based on our results, among the two-thirds of the species generations with strong thermal control of phenology, the average phenological shift after a 3 °C increase in temperatures is predicted to be 17 days in south and 13 days in north, but the rest of the species (one-third) are not predicted to shift their phenology at all, or the degree or direction of phenological shifts in these species is uncertain.

The predicted phenological shifts are well within the historical variation in peak flight dates across years and sites (average range = 40–42 days in north and south respectively). The length of the growing season (when the mean temperatures exceed 5 °C) in Finland varies between 100 and 180 days (Finnish Meteorological Institute, 2013a) and is projected to lengthen by >40 days by the end of the century (A2 scenario; IPCC, 2007; Ruosteenoja et al., 2011). If an equal lengthening of growing season both in spring and fall is assumed, the 20 days earlier start in the spring would match well with our predicted shifts in moth phenology. Our estimates are, however, larger than what could be predicted by extrapolating from the phenological changes oberved so far in Europe. Over the past decades, spring or summer phenology of plants and animals across Europe have, on average, advanced 2.5 days per 1 °C increase in temperatures (Menzel et al., 2006). However, in North America the change in spring leaf phenology has been, on average 4.8 days per 1 °C increase in temperatures (Groffman et al., 2012). Our predicted shifts are also opposite to the phenological changes observed so far, which have been (marginally) stronger towards higher latitudes (Parmesan, 2007), but they are in line with the observation by Pöyry et al. (2011) of the highest increase in moth multivoltinism in southern Finland compared to more northern areas in response to the recent warming.

Statistical models predicting shifts in insect phenology have produced results very similar to ours. Memmott et al. (2007) estimated the future phenological shift of plants and pollinators by extrapolating from phenological shifts observed so far, predicting 1–3 weeks earlier phenology by the end of the 21st century. Similarly, the spring phenology of British butterflies was estimated to shift by 10–30 days after a 3 °C increase in temperatures (Sparks & Yates, 1997).

However, process-based models taking into account winter cold temperatures have produced very different estimates of future phenological shifts in North American tree species, for which the lack of sufficient chilling in the winter delays the break of bud dormancy (Morin et al., 2009). These models predict that some, but not all, species will consistently advance their phenology more towards higher latitudes (Morin et al., 2009). Low temperatures in winter can sometimes be influential for termination of insect diapause (Gray et al., 2001), but the chilling effect was not included into our models because in the present climate the demand of low temperatures is almost surely fulfilled across Finland where mean temperatures are below 0 °C 100–200 days annually (Finnish Meteorological Institute, 2013a).

To predict the future phenological shifts by using process-based phenological models, it is important to evaluate, whether the temperature variability in the study years was representative of temperature variability that is expected to occur over the long term. During the 12 years (1993–2004) used to parameterize our models, the variation in daily temperatures was high: the estimated daily minimum and maximum temperatures of our study sites ranged between −52 and +33 °C, which represents well the temperature variation recorded at all weather stations across Finland between 1960s and 2013 (from −52 to 37 °C; Finnish Meteorological Institute, 2013a). For simplicity, possible long-term changes in temperature variability were not included in our analyses predicting phenological response to increase of 3 °C, although slight increases in interannual summer temperature variability are possible in the future (Räisänen, 2002).

Strengths and weaknesses of our modelling approach

In this study, we used the high interannual and spatial variation in species' phenologies to fit process-based phenological models. This allowed us to compare the differences in phenological controls among species and populations, and to predict the future phenological shifts under climate change, for a large number of species. Therefore, our results are not affected by the possible trends in temperatures during the study period (or lack of it).

In this study, we used a novel way of optimizing the parameter values for phenological process-based models. Our approach allowed us to inspect visually each 2D or 3D landscape showing the difference between observed and model predicted days of peak flight with different combinations of parameter values (examples in Appendix S2). This allowed us to ensure that the global instead of local minimum was found and that an approximate estimate for parameter values was found also when the surface of observed values was very rugged. The disadvantage of this method is that we were able to optimize only two parameters at a time. In addition, the sites had to be divided into two groups (southern vs. northern), to produce a sufficient number of site–year combinations for modelling purposes (with 12 years of data it was not feasible to fit the models separately for each site to model the thermal sensitivity as a function of latitude as a continuous predictor).

Species-specific experimental studies are needed, for example, to investigate the variation in phenological controls between different life stages and to study what cue (or alternative cues) is used to terminate the diapause in the spring. We can only assume that the start hours in our models reflect a photoperiodically cued termination of diapause, but there could also be alternative physiological mechanisms producing the same outcome (Ti et al., 2004). Experimental studies are also needed to understand in detail the form of thermal control under low temperatures. Our analyses produced some top models in which the estimated base for the lower developmental threshold was <−5 °C (Appendix S4), even though meaningful development at such low temperatures seems improbable (Pritchard et al., 1996). The most obvious explanation is the nonlinearity of thermal responses, which even our Tsum.nl models were not able to capture. Alternatively, in early spring, the snow or soil cover protecting the individuals could provide warmer conditions than air temperatures suggest, and therefore some development could take place even on coldest hours.

Anticipating how climate change will affect phenology across latitudes requires that we can compare the relative strengths of latitudinal trends in (i) temperature increases and (ii) thermal sensitivities of the species that are present. This task will be aided by increasingly sophisticated regional projections of increase in temperatures, rate of climate change and shifts in seasonal timing of temperatures (Burrows et al., 2011). Analyses such as reported here also contribute via the development and parameterization of process-based models that can capture the diversity of physiological controls on phenological responses within and among biological communities.

Acknowledgements

  1. Top of page
  2. Abstract
  3. Introduction
  4. Material and methods
  5. Results
  6. Discussion
  7. Acknowledgements
  8. References
  9. Supporting Information

We thank Liisa Tuominen-Roto and Guy Söderman (Finnish Environment Institute) for their help with Nocturna database, Matti Rousi and Hanni Sikanen for allowing us to use the hourly temperature data from Punkaharju, Tommi Nyman, Sanna Leppänen, Seppo Neuvonen and the anonymous reviewers for help and comments to this manuscript. We are especially grateful to the voluntary Finnish lepidopterists for maintaining the traps and identifying the moth samples. This study was funded by Joensuun Yliopiston Tukisäätiö.

References

  1. Top of page
  2. Abstract
  3. Introduction
  4. Material and methods
  5. Results
  6. Discussion
  7. Acknowledgements
  8. References
  9. Supporting Information

Supporting Information

  1. Top of page
  2. Abstract
  3. Introduction
  4. Material and methods
  5. Results
  6. Discussion
  7. Acknowledgements
  8. References
  9. Supporting Information
FilenameFormatSizeDescription
gcb12372-sup-0001-AppendixS1-S4.docxWord document2460K

Appendix S1. Details of source data.

Appendix S2. Details of competing models.

Appendix S3. Effect of phylogeny on phenological controls.

Appendix S4. Model comparison results.

Please note: Wiley Blackwell is not responsible for the content or functionality of any supporting information supplied by the authors. Any queries (other than missing content) should be directed to the corresponding author for the article.