Seasonality shapes the amplitude of vole population dynamics rather than generalist predators

1 –––––––––––––––––––––––––––––––––––––––– © 2019 The Authors Oikos published by John Wiley & Sons Ltd on behalf of Nordic Society Oikos This is an open access article under the terms of the Creative Commons Attribution License, which permits use, distribution and reproduction in any medium, provided the original work is properly cited. Subject Editor and Editor-in-Chief: Dries Bonte Accepted 5 September 2019 00: 1–7, 2019 doi: 10.1111/oik.06351 doi: 10.1111/oik.06351 00 1–8


Introduction
Periodic fluctuations in the size of small mammal populations (i.e. population cycles) are common in northern ecosystems (Finerty 1980, Krebs 1996, Myers 2018. These fluctuations tend to have a greater amplitude with increasing latitude. Such a relation is even found in other organisms such as grouse, hares and forest insects (Ims et al. 2008, Myers 2018). In small mammals, increasing amplitude of population fluctuations are associated with a longer winter season (Hanson andHenttonen 1985, 1988) and stronger delayed density-dependence related to winter predation by specialist predators (Stenseth et al. 2003). At the other end of the latitudinal gradient, stronger direct density-dependence occurs due to the presence of a larger number of generalist predators further south (Hanson and Henttonen 1985, Bjørnstad et al. 1995, Ims et al. 2008, Henden et al. 2009).
2 Empirically-based modelling studies have confirmed that seasonality is a driver of the dynamics of cyclic populations, with density-dependent regulation occurring in winter (Stenseth et al. 2003). Generally, these results show that longer winters are associated with longer cycles and greater amplitudes (Stenseth et al. 2003, Lambin et al. 2006. These studies describe stronger interspecific competition during winter and greater intraspecific competition in both summer and winter at higher latitudes. While seasonal effects have primarily been used to explain the latitudinal gradient in the dynamics of small rodents, Strann et al. (2002) suggested similar effects for variations in the dynamics of the cycles along a coastal-continental gradient.
The mechanisms underlying the seasonality hypothesis remain unclear, but a high density-dependence during winter is expected if resources are limited (Hansen et al. 1999a). Another obvious mechanism is the connection between seasonality and community processes such as predation along the north-south gradient. Similar mechanisms may also be expected along an elevational gradient as generalist predator densities (especially the red fox Vulpes vulpes) tend to be higher in valleys close to human activities (Salek et al. 2014, Walton et al. 2017. Snow cover and condition also vary with elevation, with less stable winters in the valleys causing periodic melting and icing on the ground which limits access to resources such as nests and conspecifics for thermoregulation, due to reduced mobility under the snow. Predation from generalist predators like the red fox is also expected to decrease with increasing snow cover Henttonen 1988, Lindström andHörnfeldt 1994).
In a recent review, Myers (2018) concluded that one of the remaining mysteries of population cycles is the variation in amplitudes. Here we used an elevational gradient as a proxy for variations in the winter season to study how the amplitude of bank vole Myodes glareolus populations correlate with seasonality. We monitored 30 local vole populations during two peaks and one low phase along an elevational gradient ranging from 260 m to 801 m a.s.l. Populations at the highest elevation attained snow cover earlier, and permanent snow cover lasted longer than at low elevations. The difference was approximately two months longer snow cover at 800 m than at 260 m.
We tested the seasonality hypothesis empirically, expecting greater amplitudes in the population cycles at high than at low elevation. We also tested whether seasonality, described by the elevation proxy, or the small predator community best described the amplitude variations.

Study area
The study was conducted in the boreal forest of Stor-Elvdal municipality, southeast Norway (61°N, 11°E, Fig. 1). The area has a relatively continental winter climate (Boonstra et al. 2016) with vegetation dominated by Norway spruce Picea abies and Scots pine Pinus sylvestris at low and medium elevations, and by mountain birch Betula pubescens and Norway spruce at higher elevations. The forest ground vegetation layer is dominated by bilberry Vaccinium myrtillus, mosses e.g. Pleurozium schreberi and lichens, e.g. Cladonia rangiferina. The average winter length (number of days with mean temperature below 0°C) during the study period was at the nearest weather stations 2.5 times longer at 930 m a.s.l.

Trapping plots
The bank vole is a small arvicoline rodent, found primarily in the mature forests of Europe (Myllymäki 1977). We haphazardly selected 30 bilberry-spruce forest sites suitable for bank voles along forest roads to establish an elevation gradient in presumed good habitat for bank voles. The mean distance 3 between plots was 1037 m, and minimum 500 m. At each of the 30 sites one trapping plots was established at 5-10 m from the forest roads (see Johnsen et al. 2017 for more details about field procedures). For logistical reasons, all plots could not be trapped every year and the number of trapped plots ranged from 24 to 6 (Table 1).
Trapping plots ranged from 250 to 801 m a.s.l. in a total area of ca 200 km 2 . However, due to lower trapping effort in 2017 and 2018 (Table 1), the highest elevation was then 650 m a.s.l. Traps were allocated within the plots in a cross-shaped design of 60 × 60 m ( Fig. 1; Ehrich et al. 2009). Each plot consisted of 16 Ugglan Special live traps. The cross design was chosen to maximise the area covered with a limited number of traps. If the cross design did not fit within the suitable habitat in a plot, we used trapping lines with a total 9, 10 or 12 traps, each 15 m apart. This adjusted trapping design covered approximately the same total trapping area (ca 0.6 ha) as the cross design. Traps were placed close to vole runways, dead trees or potential holes to increase the probability of catching voles. We avoided anthills, ant paths or areas with potential exposure to sun or water, to increase the survival probability of trapped individuals.

Field procedures
We monitored the plots every June and August (primary session; see Table 1 for details about trapping frequencies) during the years 2013-2018. Trap checking took place over three days with two checks per day, morning and evening (secondary trapping occasions). Traps were baited with carrots and oats, and activated 12 h before the first trap check. Traps were checked four to six times per session (Table 1). On the final check, we removed any remaining bait to avoid the supplementation of food and we left the traps open, i.e. deactivated, until the next trapping session.
We marked each new individual weighing more than 10 g by injecting a small passive integrated transponder (PIT) tag (7 mm length) into the subcutis. For each capture, we recorded identity, trap location, sex, sexual maturity and we weighed the trapped animals to the nearest gram.

Density estimates
Each plot was assumed to have an independent local population. Population size was estimated as the minimum number of animals known to be alive (MNA; Krebs 1966) from trapping data. An individual was defined as present in the study area at a primary trapping session if it was caught at least once during the secondary occasions. In addition, an individual was assumed to be alive and present in the trapping plot at primary session t if it had been caught before during primary session t − 1, and subsequently during primary session t + 1. MNA of the local populations was calculated over time from the estimated individual capture histories. When an animal was found dead in the trap it was counted as present in the trapping station, but the life history ended at that point, with no possibility of surviving and affecting the number of individuals in the next trapping session.
The trapping plots covered ca 0.6 ha. We obtained an approximate estimate of local density by multiplying the population-specific MNA by 1/(0.6 × 0.6) ha. We did not correct population size estimated by the MNA for differences in capture probabilities as this was impossible for populations with very low densities. Mean capture probability, estimated from the number of individuals observed at a primary trapping session t divided by the MNA at session t, is normally high during the summer season (0.94 ± 0.01 SE) (Aars and Ims 2002).
Season was defined as summer from June to August trapping, and winter from August to June. Trapping twice a year (spring and fall), is a common way to create time series of voles and lemmings. Hence, the population density was estimated at the end of the winter in June, and at the end of summer in August. Amplitude was estimated per plot by using the maximum difference between maximum and minimum monthly estimate of MNA.
We applied generalised linear mixed models with population ID as a random factor in all models and carried out all statistical analyses using the package glmmTMBR in R ver. 3.5.2 software (Brooks et al. 2017, < www.r-project.org >).

Small predator community
We estimated track frequencies of generalist predators red fox and pine marten, Martes martes, by snow tracking along 2.95 km (SD = 0.5) transect lines in January for the period from 2003 to 2014. The transect lines were part of a nationwide monitoring program for Eurasian lynx Lynx lynx and were based on voluntarily work from members of the Hedmark Chapter of the Norwegian Association of Hunters and Anglers (Tovmo and Brøseth 2011). The transect line density was three (SE = 4) lines per 100 km 2 . Of a total of 621 different lines, 281-484 lines were surveyed annually during favourable snow conditions, i.e. 2 ± 5 days after snowfall (see Breisjøberget et al. 2018 for more details). We used estimates of track frequencies, i.e. the number of tracks per km divided by the number of days since last snowfall, from 19 transect lines allocated between 300 m and 800 m a.s.l. and within 20 km south and north of the vole trapping plots.

Results
The population trajectories showed that yearly and seasonal variations in the minimum number of animals alive in each population correlated with elevation as a continuous variable (Table 2). We categorised plots into two: 15 low elevation plots < 550 m and 15 high elevation plots > 350 m (Fig. 2). The highest peak population densities occurred in fall 2014, with up to 125 animals per ha in some of the high elevation populations. Low elevation populations had less than 100 animals per ha at the peak of 2014. The minimum densities in both low and high elevation populations were <2.2 animals per ha and did not differ between high and low elevations (Table 3). Peak population densities were 1.5 times greater in high elevation than low elevation populations (Table 3).
There was a strong positive correlation between the amplitude and elevation (r 2 = 0.28, slope = 0.09 ± 0.03 (SE), F 1,28 = 10.76, p = 0.003; Fig. 3). The indices of fox and generalist predators were highly correlated (r > 0.99), and both predator indices were also correlated negatively with elevation (r = −0.98). Hence, elevation and generalist predators were highly confounded. According to AIC weights, the univariate model with elevation as the predictor of amplitude was twice as likely to be the best model than the models including generalist predators (Table 4, Fig. 4).

Discussion
Our results confirmed our expectations that winter length was positively associated with the amplitude of vole population cycles. A two-month longer winter season resulted in    Figure 3. The regression between amplitude (no. of animals ha −1 ) and elevation. Each symbol is one population. a 1.5-fold greater cycle amplitude of the local population fluctuations over the six-year study period. This was mainly due to the higher maximum densities at high elevations in the first peak (year 2014) in the time series. Profound temporal variation in cycle amplitude is a common characteristic in rodent populations (Kleiven et al. 2018, Soininen et al. 2018. Also, the fact that fewer plots with a lower altitude range were trapped in 2018 may have contributed to the lower amplitude of the second peak of the time series. Hence, the seasonality effect is elevational as well as latitudinal. As far as we know, we have for the first time tested directly whether the community of small generalist predators is part of the seasonality effect. Obviously, the small mammal community is completely confounded with elevation and season and it may be impossible to falsify the predation hypothesis completely. Furthermore, we cannot exclude the possibility of an elevational variation in predation impact due to a functional response rather than a numerical response of the predators. However, our results demonstrate that seasonality alone as a predictor of amplitude is twice as likely to be the best model, compared to models including generalist predators. Seasonality is important for population cycles to occur (Stenseth et al. 2003), although this assumption is not supported by cyclic vole populations in agricultural areas in southwest France (Lambin et al. 2006), it seems robust at least for more natural boreal and arctic ecosystems. Empirically-based modelling studies based on time series from Fennoscandia have shown the significance of seasonality as a driver of the dynamics of cyclic populations showing strong density-dependence in winter (Bjørnstad et al. 1995, Hansen et al. 1999a, b, Kleiven et al. 2018. These studies describe strong interspecific competition during winter and high intraspecific competition in both summer and winter. High density-dependence during winter is expected if resources are limited (Hansen et al. 1999a). Indeed, several studies of the winter ecology of cyclic vole populations confirm the significance of food resources during winter for winter survival (Ylönen and Viitala 1991, Schweiger and Boutin 1995, Eccard and Ylönen 2001, Huitu et al. 2003, 2007, Boonstra and Krebs 2006, Johnsen et al. 2017, Soininen et al. 2018, but see also Yoccoz et al. 2001). Food availability influences the bank vole social system (Ostfeld 1990), and overwintering groups have been reported to operate on highquality patches with potentially high survival Viitala 1991, Sundell et al. 2012).
Other resources than food may be limited during winter. For instance, Korslund and Steen (2006) found that survival of tundra voles Microtus oeconomus increased with increasing space availability in the subnivean area. Similar results show that snow depth could increase the summer density of brown lemmings Lemmus trimucronatus, affecting the amplitude and possibly the periodicity of the cycle (Bilodeau et al. 2013). Reid and Krebs (1996) found that shallow winter snow appeared to be a strong limiting factor in the population growth of collared lemmings Dicrostonyx kilangmiutak, and that the density of winter nests in voles and lemmings increased with snow depth (Reid et al. 2012). Finally, Ylönen and Viitala (1985) found that bank voles aggregated in areas with brush-vegetation  6 before winter, which were also the areas with thickest snow cover during winter (see also Hambäck et al. 2002). Winter aggregations benefit from a high level of social interactions (Ylönen and Viitala 1991) which allow high reproduction at the onset of the breeding season in spring (Andreassen et al. 2013, Hansen et al. 2013, Rémy 2013, Radchuk et al. 2016. This may give rise to high population growth rates in summer and consequently greater cycle amplitudes. It is to be noted that the described resources could be interacting with other factors and be used to minimise the direct causes of mortality such as predation or diseases during winter. In contrast, populations at low elevation may be more vulnerable to unstable mild winter weather that reduces habitat availability and quality, with variable snow cover, due to repeated thawing and freezing phases at ground level (Aars and Ims 2002, Stien et al. 2012, Bilodeau et al. 2013, Hansen et al. 2013. This may limit access to food resources and shelter while increasing exposure to mammalian and avian predators and to lower critical temperatures (Hansson and Henttonen 1985, Aars and Ims 2002, Hoset et al. 2009, Haapakoski and Ylönen 2013. Breeding during winter is a characteristic of increasing vole populations (Krebs and Myers 1974). This may be restricted by weather conditions and some studies suggest that the timing of ice and snow melt in spring might be the most critical factor, as it is the spring population that initiates the peak phase of the population cycles (Aars and Ims 2002, Korslund and Steen 2006, Kausrud et al. 2008, Hoset et al. 2009, Ims et al. 2011, Cornulier et al. 2013, Haapakoski and Ylönen 2013, Fauteaux et al. 2015. Still, more knowledge is needed on small rodent winter ecology, and winter breeding is poorly understood (Krebs 1993, Aars andIms 2002).
In conclusion, seasonality is an important factor in shaping the dynamics of cyclic vole and lemming populations. The effect may be seen as an intrinsic characteristic of cyclic vole populations. However, more probable the effect is connected to winter resource availability and inter-and/or intraspecific density-dependence. It may be empirically impossible to untangle the many confounding factors of the small mammal community, elevation and season. In a theoretical modelling approach Tyson and Lutscher (2016) found that even a small change in season length would have large effects on a simple predator-prey system. In this study the modelled predator behaved as specialist predator during the winter and a generalist during the summer. Thus, to elucidate to what extent predators act as a decisive mechanism connected to seasonality may thus require more detailed data than were available in the present study. Still we have been able to show that the abundance of generalist predators were not the best predictor of variation in cycle amplitude of a rodent species in boreal forest.