Population dynamics in a cyclic environment: consequences of cyclic food abundance on tawny owl reproduction and survival


Correspondence author. E-mail: patrik.karell@helsinki.fi


1. Understanding which factors regulate population dynamics may help us to understand how a population would respond to environmental change, and why some populations are declining.

2. In southern Finland, vole abundance shows a three-phased cycle of low, increase and decrease phases, but these have been fading out in recent years. During five such cycles (1981–1995), all tawny owls Strix aluco were censused in a 250-km2 study area, and their reproduction and survival were monitored.

3. Males and females showed similar dynamics, but experienced breeders recruited more offspring and had higher survival than first breeders. Offspring recruitment, but not survival of breeding individuals varied in accordance with vole abundance.

4. The population’s numerical response to prey abundance was primarily due to first-breeding individuals entering the population in the increase phase when immigration was the highest. First-breeding birds were younger, but experienced breeders were older in more favourable vole years.

5. A stage-specific matrix population model integrating survival and fecundity showed that, despite obvious variation in fecundity between vole cycle phases, this variation had limited importance for overall tawny owl population dynamics, but that the survival of experienced breeders during the low phase is most important for population growth.

6. Model and data agreed that the vole cycle drives the dynamics of this avian predator by limiting the recruitment of new breeders during the low phase. Population dynamics hence differ not only from the classic example of the species in a more temperate region in the UK where the number of territories is stable across years, but also from the dynamics of other avian vole predators in Fennoscandia where the recurring crash in vole abundance drastically lowers adult survival thereby creating vacancies.


Since the seminal work of Charles Elton (1927), population cycles of predators and their prey have been a focal topic of animal ecology and population dynamics (Southern 1970; Hanski, Hansson & Henttonen 1991; Hanski & Korpimäki 1995; Krebs et al. 1995; Lambin, Petty & MacKinnon 2000; Lindström et al. 2001; Gilg, Hanski & Sittler 2003; Sundell et al. 2004; Korpimäki et al. 2005a, b). Cyclic fluctuations in the abundance of herbivores are commonly found in populations on high latitudes and/or high altitudes (Lindström et al. 2001). Because these herbivores typically are basal to the ecosystem, the effect of the cycles in their abundance reverberates across the food web (Ims, Henden & Killengreen 2008). The consequences of herbivore cycles are thus apparent also on higher trophic levels, even when these predators do not directly drive the cycle. This perspective is in contrast to the classic view of predator–prey dynamics as a Lotka–Volterra type of dynamics, where the predator drives the fluctuations in the abundance of the prey and shows similar cycles as its prey but lagging in time. A classic example of predator-prey population dynamics where the predator’s population size does not track the fluctuations in the abundance of its main prey, voles, is provided by Southern’s (1970) study of tawny owls Strix aluco Lin. in southern England.

In Northern boreal environments in Fennoscandia, tawny owls and other birds of prey occur in such low densities that they do not have the potential to impose sufficient predation pressure to make a serious impact on the vole dynamics and hence are, themselves, not driving the cyclic fluctuations in their main prey (Korpimäki et al. 2002; Norrdahl et al. 2004). Resident owl species (Ural owl Strix uralensis Pall. and tawny owl) respond to fluctuations in food abundance by adjusting their reproduction, but – once they have occupied a territory – do not disperse to breed where there are plenty of voles as other (semi-) nomadic species do (Andersson 1980).

By refraining from breeding when food is scarce, the proportion of breeding site-tenacious owls can increase rapidly with increasing numbers of voles, without any delay (Southern 1970; Brommer, Pietiäinen & Kolunen 2002; see also Korpimäki & Norrdahl 1989, 1991; Rohner 1996). On the other hand, the mortality of territorials (and their offspring) is drastically increased when the voles crash in abundance every third year (Brommer et al. 2002). This recurring ‘bottleneck’ creates opportunities for prebreeding individuals (floaters; Rohner 1996) to start breeding when food abundance increases again. As a consequence, fluctuations in food abundance generate changes in the population’s age distribution, as the proportion of young, first-breeding individuals in the population increases when food abundance becomes more favourable (Brommer, Pietiäinen & Kolunen 1998).

For a variety of reasons, young and/or inexperienced individuals may respond differently to environmental fluctuations than older and experienced ones (Metcalf & Pavard 2007). The change in age structure over a cycle therefore potentially creates marked variation in the population’s reproductive output and survival. One powerful way to incorporate such individual differences in performance is to group the individuals in relevant stages. In general, such grouping has important consequences for the understanding of population growth and dynamics (Caswell 2001). In case of a population living in a cyclic environment, changes in population structure across the cycle need to be incorporated and the consequences of a variable population structure for reproduction and survival need to be understood when modelling cyclic population dynamics.

In many places, and particularly in Fennoscandia, herbivore cycles are fading out (Ims et al. 2008), which is expected to present a major change in the environment for many other species that are (partly) dependent on these herbivores. Avian predators of voles are prime candidates for species likely to be negatively affected by changes in the vole dynamics (Hörnfeldt, Hipkiss & Eklund 2005). Although the tawny owl is a generalist predator in Northerly populations, it almost non-exclusively uses voles as a prey when vole abundance is high (Petty 1999) and it is highly dependent on voles for reproduction (Kekkonen et al. 2008). In this study, we determine how reproduction and survival, which together define population growth, of different stages of tawny owls respond to variation in food supply during 15 years of persistent cycles in vole abundance. Our aim was to provide a benchmark for understanding the cyclic tawny owl – vole system to evaluate changes in this system when vole cycles fade out. In particular, we aim to assess the relative importance of variation in reproduction and survival for the dynamics of a predator population subsisting on prey that shows periodic fluctuations in its abundance. Previous studies have shown that reproduction and survival vary in such an environment (e.g. Brommer et al. 1998), but no study has – to our knowledge – quantitatively compared the importance of such variation for the population dynamics. It is not obvious how variation in reproduction can be compared with variation in survival without formal consideration in a population dynamical model. We therefore construct a matrix population model based on our study of fecundity and survival to perform an elasticity analysis, in which we quantitatively resolve how a change in survival and fecundity rate of different stages across the vole cycle would alter population growth rate.

Materials and methods

Study area

Tawny owls were studied in a study area of c. 250 km2 in southern Finland (60°15′ N, 24°15′ E). Between 1975 and 1980, the study area was established by setting up boxes. Pairs nested almost exclusively in nest boxes, which were provided in high abundance (c. 125 were available from 1980 onwards). Tawny owls were also ringed and controlled by ornithologists in regions surrounding the study area. We here consider data collected from the study area during 1981–1995, when vole dynamics were cyclic.

Field protocol

The census and handling of all tawny owls were carried out by KA and TK. Territorial tawny owls are rather vocal, especially prior to the breeding season. In early spring (March–April), recordings of hooting tawny owls were played at regular locations along roads transecting the study area. The location where tawny owls responded was marked on a map. Starting at the end of April, all boxes and other possible breeding sites were checked. Considerable effort was put into finding the nests of all tawny owls by searching for natural nest sites and using new boxes set up by private individuals in the approximate area where a hooting owl was recorded and where a breeding thus was expected. We here distinguish between the number of territories and the number of pairs that are breeding, where the former refers to the number of sites where tawny owls responded to the playback, and the latter to the number of pairs observed breeding.

Practically all females and males were trapped when the offspring were 1–2 weeks old. Brooding females were taken from their nest boxes in the evening by netting them at the opening of the nest box. After handling, the female was put back into the nest box and a swing-door trap for the male was mounted in front of it and left over night. In the following, morning traps were checked and the males were handled. When the oldest chick was c. 25 days old, all offspring in the brood were ringed with a unique aluminium ring to allow lifelong individual identification. Adult birds were categorized into three age classes at ringing −1, 2 and +2 years old – using characteristic patterns on juvenile and adult primary and secondary feathers (Ahola & Niiranen 1986). Individuals were considered as first breeders if they were unringed when first caught, or if they had been ringed as pulli and did not have a previous breeding record. Because considerable effort has been put into finding all nests and ringing all nestlings, we considered all individuals that were unringed when first caught as immigrants.

Prey abundance

Spring and autumn snap trapping of small mammals was carried out by KA and TK within the study area from 1981 onwards to estimate the abundance of prey. Snap trapping was conducted in two localities on each trapping event: one in the eastern part of the study area and one in the western part of the study area. Each trapping locality consists of open (field/clear-cut) habitat and forest habitat. Traps were set as a transect of 16 trapping spots (15 m between) with three traps each, giving a total of 48 traps per habitat (96 traps per replicate). All traps were triggered for two consecutive nights (192 traps for two nights = 384 trap nights in total per trapping). Not only field voles Microtus agrestis and bank voles Clethrionomys glareolus were caught in the traps, but also, to some extent, wood mice Apodemus flavicollis and shrews Sorex araneus. We include all species in the analysis of prey abundance (number of individuals caught per 100 trap nights). Based on these trap indices, we categorized years into low, increase and decrease phases of the vole cycle (see, e.g. Brommer et al. 2002).

Life-history stages

We here distinguish owls according to their sex and whether they are first breeders or have bred before (termed experienced breeders). Differences in reproductive performance between the stages of first and experienced breeders are expected (Metcalf & Pavard 2007). Our main motivation for classifying individuals into first vs. experienced breeders is that such a classification is possible in a non-arbitrary way for any individual in a population, whereas ageing all individuals is not always possible. For example, age classes in tawny owls older than 2 years cannot be distinguished (Ahola & Niiranen 1986), such that the exact ages of only a subset of individuals is known, and similar constraints apply to most avian populations. We here consider the sex of an individual as a second-stage variable. An individual’s sex may be particularly important in birds of prey, where males are smaller and mainly provide food (almost exclusively so prior and during incubation), whereas the larger females incubate and defend the brood. Such sex differences may conceivably lead to differences in their sensitivity to changes in the environmental conditions that occur when food supply fluctuates. Sex differences in reproduction and survival in birds of prey have rarely been examined (but see Altwegg, Schaub & Roulin 2007), mainly due to the difficulty of capturing males.

Statistical analyses

Statistical analyses were carried out using r 2.5.0 (R Development Core Team 2007). To investigate associations between vole and owl numbers, we assumed a saturating nonlinear response and used linear regression with square root of the prey trapping indices in the preceding autumn to explain the number of breeding individual owls. A saturating response is expected and biologically motivated as follows: (i) voles are the only available prey of tawny owls in Finland prior to breeding (most birds migrate); and (ii) a previous study of Finnish tawny owls has shown that breeding parameters correlate with voles but not with other prey items (Kekkonen et al. 2008). In the other analyses (on reproduction and survival), vole phase was entered as a three-level categorical variable (low, increase and decrease phases). We grouped years according to vole cycle phases rather than describing variation across years due to vole abundance (i.e. as a continuous function of prey abundance), because the phase-based approach captures not only the present, but also the future dynamics of the vole cycle. For example, in March–April when the breeding season for tawny owls starts in Finland, the vole abundance is reasonably high both in vole increase and in vole decrease phases, but the autumn density of voles is high in the former whereas voles are almost absent in the latter (Karell et al. 2009). Because we study five vole cycles, the phase-based approach constitutes a replicated design (each phase occurs five times). Hence, to explicitly explore whether such categorization indeed captures an intrinsic part of the dynamics, we specifically modelled the effect of years (nested in phase) in all analyses [denoted ‘year(phase)’ in the results]. In case phases are irrelevant to the dynamics and variation across years is in fact the main driver, analyses should show that phase is insignificant.

The effects of breeding experience on breeding activity and recruit production were analysed separately for males and females, and we thus only made a qualitative comparison of the sex-related differences in breeding performance. Data on both sexes were not combined in the same analyses because this would require analysing the data per pair.

Survival of first and experienced breeders in relation to the vole cycle was estimated using capture–mark–recapture (CMR) methodology on live encounters data (Cormack–Jolly–Seber model, CJS) using the program MARK (White & Burnham 1999). With the CJS model, one can separate survival probability (Φ) from recapture probability (P) using a maximum likelihood approach. We used data on individual encounters from 1981 to 1996 in order to also get an estimate of survival for the last year of the study (1995). Individuals were categorized into males and females to test for sex-specific differences in survival. We built a full model coding for sex and including two experience categories (first breeder denoted by ‘first’ and experienced breeder denoted by ‘exp’) and full-time dependence in survival and recapture [Φ(sex × timefirst/timeexp)p(sex × timefirst/timeexp)]. To correct for overdispersion, we calculated the parameter ĉ as the ratio of the observed deviance of the full model over the mean deviance achieved from 500 bootstrap simulations of the same model. Time dependence was replaced by vole cycle phase to test if Φ and p of the different age and sex categories were affected by the vole cycle. Models were built by entering the vole cycle phases directly into the Parameter Index Matrix in program MARK as a three-class variable coding for low, increase and decrease phase. Therefore, Φ and p could be either time dependent (time), constant (c) or vole phase dependent (phase) for each experience class (first or experienced breeder) and sex. Models were ranked on the basis of their quasi-likelihood Akaike information criterion (QAIC) calculated as −2 ln(L)/ĉ + 2K + (2K(K + 1)/(n – K – 1), where L is the likelihood of the model, K the number of parameters and n the effective sample size. We used the following approach in model selection: for ‘Φ’ we first tested all candidate models nested under the full model (n = 22) while keeping ‘p’ constant at ‘sex × time/time’. Models including ‘sex’ as a factor did not receive high support (QAIC), and therefore two-way interactions of ‘sex × experience’ and ‘sex × phase’ were not tested. The models with the ighest QAIC among these 22 models (QAIC weight more than 5%, n = 5) were then tested with all the different ‘p’ possibilities (n = 22) giving a total of 127 models [n = 22 + (5 × 21)]. The five models with the highest QAIC were ranked in an equal order for all the different recapture possibilities. Because we did not find strong support for a single model (see Results), we obtained estimates of survival and recapture probabilities through model averaging. Model averaging calculates an average value over all models in the candidate model set with common elements in the parameter structure, weighted by normalized QAIC model weights [exp(−ΔQAIC/2)/∑(exp(−ΔQAIC/2))]. For more information on the modelling approach, see Burnham & Anderson (1998) and Cooch & White (2001).

Proportional data on population composition were analysed using generalized linear models (GLM) with binomial errors and logit links. We estimated reproductive success in the population as the proportion of recruits per number of eggs produced in a given year. To allow qualitative comparison between sexes recruitment probability was analysed separately for males and females in (nested) GLMs with binomial errors emphasizing vole cycle and experience-dependent effects. We constructed the minimal adequate models by stepwise backwards modelling in which we compared the models using likelihood ratio tests (LRT, χ2; Crawley 2002). We also compared the models using AIC. Both the AIC and LRT produced the same minimal adequate models for the data. Although our data on reproduction is individual-based, we chose to analyse it on an annual basis. Our aim with the analysis of recruitment was to evaluate the population-level productivity of first and experienced breeders, and hence, to deliver a general picture of reproduction in relation to the vole cycle as a baseline for the population dynamical model. Although generalized linear mixed models (LMMs) can account for individual characteristics, we chose not to use such a model because the focus is on the population-level contribution of first and experienced breeders. We thus accepted some level of pseudo-replication (120/258 individuals are included both as a first breeder and as an experienced breeder) as the preference of a straightforward and robust method of analysis. We report in the results the minimal adequate models achieved through the stepwise backwards procedure.

Age composition was based on individuals that were ringed as nestlings and whose age was therefore known exactly. The data were analysed with LMMs with ‘individual identity’ as a random effect, because certain individuals occurred repeatedly in the analysis. We constructed the minimal adequate model in the same manner as for GLMs by stepwise backwards modelling in which we compared LMMs solved under maximum likelihood using LRT between models. In an LRT, −2 times the difference in the likelihood of a model with and without an effect was tested as a chi-squared value with the effect’s degrees of freedom (Pinheiro & Bates 2000). The significance of the random effect was also based on an LRT. Significance of fixed effects in LMMs was based on F-tests.


A total of 351 clutches were monitored in the study population between 1981 and 1995. Of these clutches, 300 produced hatchlings and 274 succeeded to fledge offspring. In 294 breeding attempts, the female parents were caught and identified and, in 278 cases, also the male parents were caught and identified.

Pair formation

Pairs were formed assortatively with respect to breeding experience as most (68%, 125 of 184) pair formations were between either two inexperienced breeders or two experienced breeders (Table 1). After first pair formation mate change was rare: in total 32 of 122 (26%) males changed mate during the study, of which 13 changed more than once [in total 51 mate changes of which 43 changes (84%) was due to death of the partner]. Even fewer females (29/129, 16%) changed mate, with only six females changing more than once (in total 37 mate changes of which 28 changes (76%) was due to death of the partner).

Table 1.   Pair formation in the tawny owl population
First breeder (%)Experienced breeder (%)
  1. There were in total 184 new pair formations of which most were between two first breeders.

First breeder96 (52)26 (14)
Experienced breeder33 (18)29 (16)

Territory occupancy and breeding activity in relation to the vole cycle

Territories were considered occupied when an owl responded to play-back calls prior to the breeding season. Once established on a territory, most owls remained faithful to their territory. Only 23/122 (19%) males and 29/136 (21%) females changed territory during the study. Territory change occurred after the first breeding event in 15/23 cases (65%) for males and in 19/29 (66%) of the cases for females.

There was variation in the number of occupied territories between vole cycle phases (Fig. 1a): in low phases 28 ± 4 territories were occupied and the number increased to 34 ± 3 territories in increase phases and further to 39 ± 2 territories in decrease phases [anova: phase, F2,12 = 3.76, P = 0.05; decrease phase vs. low phase, t12 = 2.74, P = 0.02; year (phase) dropped]. This difference in territory occupancy between years was less clear when prey abundance in the preceding autumn was entered as an explanatory variable (Fig. 1a; linear regression: R2 = 0.14, B = 2.70 ± 1.96, F1,12 = 1.89, P = 0.19).

Figure 1.

 Number of occupied territories in which tawny owls responded to play-back calls (a) and proportion of occupied territories where a clutch was produced (b) as a response of prey abundance. Prey abundance stands for vole numbers caught per 100 trap nights. Vole cycle phases are denoted by different symbols: filled diamonds represent low phase, open circles represent increase phase and filled triangles represent decrease phase.

The proportion of occupied territories where the residing owl pair attempted to breed increased as vole abundance increased (Fig. 1b; logistic regression, B = 0.27 ± 0.10, χ2 = 7.39, d.f. = 1, P = 0.007). Also when categorizing vole abundance into vole cycle phases, a substantially higher proportion of territorial pairs bred in the increase and decrease phase years than in the low phase years [Fig. 1b; GLM: phase, χ2 = 9.64, d.f. = 2, P < 0.0001; year(phase) dropped].

Some nests failed at an early stage before parents could be caught and identified (16.2%, 57/351). There was no difference in the proportion of failed nests between vole cycle phases (χ2 = 1.09, d.f. = 2, P = 0.58). Among the failed nests, 48% (31/65) were in newly established territories and this proportion of failed newly established territories did not vary between vole cycle phases (χ2 = 1.80, d.f. = 2, P = 0.41). Therefore, we found no bias in the data with respect to capture of individuals from newly established territories (indicating that they are first breeders) and formerly established territories (indicating that they are experienced breeders) in different vole cycle phases.

The numerical response of breeding male and female owls to the yearly prey density was affected by the owls’ breeding experience (Fig. 2): the yearly vole density tended to explain the number of first breeding males (linear regression, R2 = 0.27, B = 2.20 ± 1.03 SE, F1,13 = 4.54, P = 0.055) and females (R2 = 0.26, B = 2.06 ± 0.99 SE, F1,13 = 4.31, P = 0.060). In contrast, there was no evidence for a relationship between the number of experienced owls and prey density (males, R2 = 0.05, B = 0.89 ± 1.08 SE, F1,13 = 0.67, P = 0.43; females, R2 = 0.13, B = 1.35 ± 1.00 SE, F1,13 = 1.81, P = 0.20).

Figure 2.

 Relationship of owl and vole numbers separately for first-breeding and experienced (a) male and (b) female tawny owls. Prey abundance stands for vole numbers caught per 100 trap nights. Owl numbers are actual numbers. Symbols are as in Fig. 1. Statistics are given in the text.

Age structure in the population

The vole cycle had differential effects on the age structure of first and experienced breeders (Fig. 3). Despite the age variation between years [Table 2, year(phase)], the vole cycle caused variation in the age structure of the population (Fig. 3), where especially the first breeders were younger in the increase and decrease phase than in the low phase (Fig. 3; Table 2, phase). First breeders and experienced breeders showed opposite patterns in age composition in different vole phases (Fig. 3). Both male and female individuals that started to breed in a decrease phase were younger than those that started in a low phase, whereas experienced breeders were the oldest in the decrease phase (Fig. 3; Table 2, phase × experience).

Figure 3.

 Mean age (±SE) of first-breeding and experienced females (open circles) and males (filled diamonds) in different phases of the vole cycle. Included are individuals ringed as nestlings, which were born either within or outside the study area, but which bred within the study area are included. See Table 2 for statistics.

Table 2.   Minimal adequate model of factors that explain the variation in age in the tawny owl population
Fixed effects
 Phase28.242, 122<0.0001
 Experience3059.321, 122<0.0001
 Year (phase)856.903, 122<0.0001
 Phase × experience21.732, 122<0.0001
 Variance (95% CI)%χ2P
  1. The results are from a linear mixed effects model with normal errors and individual ID as the random effect. The significance of the random effect is tested with a likelihood ratio test (χ2).

Random effect
 Individual0.98 (0.84–1.13)99.4268.29<0.0001

Reproduction, patterns of recruitment and immigration

Recruits were produced in all three phases of the vole cycle (Fig. 4). Recruitment probability depended on the vole phase, as eggs produced in the increase phase had the highest recruitment probability (phase: females, χ2 = 19.07, d.f. = 2, P < 0.0001; males, χ2 = 16.36, d.f. = 2, P = 0.0003). Furthermore, experienced individuals produced proportionally more recruits than first breeding individuals (experience: females, χ2 = 7.61, d.f. = 1, P = 0.006; males, χ2 = 4.36, d.f. = 1, P = 0.04). Nearly all (39/40) of the locally produced recruits were recruited to the breeding population as 1- or 2-year olds. More specifically, a larger proportion of these local recruits were able to start breeding as 1-year olds if they were born in an increase phase compared with those that were born in a low or decrease phase (Table 3).

Figure 4.

 Proportion of offspring that recruited to the breeding population in low, increase and decrease vole phases produced by first breeders and experienced breeders. Error bars (mean ± SE, n = 5 years for each phase) are clustered by the sex of parent (open circles: females; filled diamonds: males). Total number of eggs produced is given above bars.

Table 3.   Age at recruitment of offspring born in different phases of the vole cycle
 Phase at birth (%)
  1. Included are local recruits that were born in the study area between 1981 and 1995 (n = 40, 21 males and 19 females). Data from both sexes are pooled as there was no difference in age at recruitment between sexes (Fisher’s exact test, P = 0.59). The age distribution of recruits differ between phases of the vole cycle (Fisher’s exact test, P = 0.002).

Age at recruitment
11/3 (33)24/28 (85.7)3/9 (33)
22/3 (67)3/28 (10.7)6/9 (67)
31/28 (3.6)

The proportion of locally recruited and immigrant first breeders in the population varied between vole cycle phases. These proportions were similar for both sexes and we therefore pooled the sexes. The proportion of immigrant first breeders was phase-dependent (χ2 = 6.17, d.f. = 2, P = 0.046). In low vole phases, 82% (32/39) of first breeders were immigrants, whereas in increase vole phases the proportion of immigrants was higher (102/117, 87%) and in decrease vole phases lower (91/122, 77%).


A CMR analysis of the data revealed clear differences in survival probabilities between first and experienced breeders (Table 4). Survival was fairly stable across different phases of the vole cycle as the most parsimonious model had constant survival over years in both first and experienced breeders. However, some indications of a vole cycle effect on survival could be detected as models including phase-dependent survival in either first (phasefirst/cexp) or experienced breeders (cfirst/phaseexp) or both (phasefirst/phaseexp) had almost as low QAIC values as the most parsimonious model (Table 4). Because none of the models could be considered a single best model, we employed model averaging to calculate the survival and recapture values for first and experienced breeders. Estimates of annual variation in survival did vary to some extent according to phases with higher survival of experienced breeders (Table 4). There were no differences in survival between sexes as all sex-specific models received lower support. Also recapture probabilities were equal for both sexes, but they varied differently for first and experienced breeders in the models that received the highest support (Table 4). Model averaging revealed that recapture probability of first breeders was dependent on phase, and was intermediate after low phases (average 0.60 ± 0.13), highest after increase phases (average 0.78 ± 0.08) and lowest after decrease phases (average 0.25 ± 0.06). Among experienced breeders, recapture probability varied annually but independently of the vole cycle (average 0.67 ± 0.09).

Table 4.   Survival and recapture probabilities of tawny owl males and females between 1981 and 1995
ModelQAICQAIC weightN parLikelihood
 First breeder ΦExperienced breeder Φ
  1. The models separate between survival (Φ) and recapture (p) and have two age categories (reported as ‘first breeder/experienced breeder’) that can be either constant (c), phase dependent (phase) or time dependent (time). We tested all possible model permutations for both experience categories, and for both sexes including their interactions. Shown here are the four best models as judged by their QAIC values, which together provide 89.4% support for the data (models including sex-specific effects received little support). Statistics given are quasi-likelihood Akaike information criterion (QAIC; the AIC corrected for overdispersion, ĉ = 1.15), proportional support for the model (QAIC weight), number of parameters (N par) and model likelihood. The lower part of the table shows the estimates of survival (Φ) ±SE (95% CI) achieved by averaging the estimates over all candidate models.

Model averaging
 Low0.56 ± 0.07 (0.427–0.688)0.73 ± 0.05 (0.620–0.819)
 Increase0.55 ± 0.06 (0.437–0.635)0.78 ± 0.07 (0.598–0.896)
 Decrease0.60 ± 0.08 (0.440–0.739)0.71 ± 0.06 (0.578–0.823)

A population dynamical model

Model structure and simulation

We found heterogeneity in the contribution different stages (first and experienced breeders) make to the population in terms of their reproduction and survival. Nevertheless, in a cyclic environment where survival and fecundity vary between cycle phases, it is not immediately obvious how a change in these values would affect the overall population dynamics. We therefore constructed a stage-specific matrix model, where we considered first and experienced breeders as separate stages. Because both sexes showed very similar dynamics, we considered females only.

A standard matrix population model is based on annual values of reproduction and survival, forecasting year-to-year dynamics. In a cyclic environment, annual values for reproduction and survival are not the same from year to year. However, cycles will repeat themselves with a fixed period. Hence, matrices describing the yearly population contributions (reproduction and survival) over one cycle can be combined in a block matrix to describe the cycle-to-cycle population dynamics (Brommer, Kokko & Pietiäinen 2000).

We denoted recruitment (average number of offspring per breeding pair that joined the breeding population in a future year) for phase ‘ph’ and a bird of experience ‘exp’ as Fph,exp. By definition, all first breeders breed, but not all experienced females with a territory breed in a given phase (see above), and we therefore included the fraction breeding bph. Because this population is not closed, we included a phase-specific immigration rate mph (see above), which was assumed to act as a scaling parameter for all reproductive output. Some recruits join the population in the next year, whereas others join the breeding population in later years (in practice this was never later than 3 years, Table 3). We therefore denote the fraction of recruits born in phase ‘ph’ and recruiting ‘y’ (1, 2 or 3) years later as rph,y.

Survival for phase ‘ph’ to the next phase for a bird of experience ‘exp’ was denoted by Pph,exp. The cycle-to-cycle population dynamics of a three-phase cycle are then given by the block matrix


We parameterized the model with the observed data. The fractions (r) of offspring that were born in a low, increase or decrease phase and recruited 1, 2 or 3 years later are given in Table 3. The values for the reproductive parameters m, b and F and their rationale are outlined in Table 5. Stage- and phase-specific survival values are reported in Table 4.

Table 5.   Data summary per vole cycle phase of parameters used in the population dynamical model
  1. The data includes observations of all clutches produced in the population during five vole cycles. Parents of those clutches that failed (and where parents could therefore not be caught) are categorized according to whether a clutch was produced in the territory in the previous year (experienced breeder) or not (first breeder). Nfirst and Nexperienced are the total number of observed first and experienced (female) breeders in the population in a given phase. Parameter bph is the fraction of experienced breeders that produced a clutch over the total number of experienced breeders (reproducing and territorials) in the population. Ffirst and Fexperienced are fecundity estimates of first and experienced breeders (number of recruits/total number of pairs), and mph is the immigration coefficient (mph = Nimmigrants/Nrecruits) when both sexes are pooled (data reported in text).

Low264141/110 (0.37)0/26 (0)5/41 (0.12)4.57
Increase705757/97 (0.59)20/70 (0.29)25/57 (0.43)6.80
Decrease778080/117 (0.68)4/77 (0.05)17/80 (0.21)2.94

The dominant eigenvalue λ1 of B1(B)] denotes the change in population size (i.e. the number of females with a territory) across cycles. Elasticity analysis is a prospective analysis in which one can explore how much λ would change in response to changes in reproduction and survival (Caswell 2000). Elasticity can be calculated for each stage/phase combination by using standard methods (Caswell 2001) to calculate the elasticities of B. Each elasticity value gives a proportional contribution to λ1(B) as the sum of elasticities always equals 1. In a stable cyclic population λ1(B) = 1. The right eigenvector of B that corresponds with the dominant eigenvalue [w1(B)] gives the stable stage distribution, i.e. the proportion of first and experienced breeders and non-breeders in each phase during a complete cycle. Again, this vector can be calculated using standard methods (Caswell 2001).

The elasticities of matrix B indicate the proportional contribution of each matrix element to population dynamics. The sum of all elasticities equals 1 by definition, and the elasticity values therefore provide a ranking of importance. This procedure models both demographic stochasticity and uncertainty in the estimates of fecundity and survival. To gain insight into the variability of our findings with respect to stochasticity in fecundity and survival, we calculated λ1(B) and the elasticities for matrix B based on 10 000 simulations varying Fph,exp and Pph,exp. We treated all other parameters as constants as we have no prior expectation of their uncertainty. Hence, the variability addressed by our simulations is conservative.

Simulated reproductive output was the average of a randomly drawn Poisson distribution over the number of phase- and stage-specific territories with mean Fph,exp (Table 5 for values). We used the estimated error of the CMR survival values (Table 4) to vary the survival values. The logit of the simulated survival value was inline image, where inline image is the estimated mean survival (Table 5), and inline image is a random draw of a normal distribution with zero mean and variance inline image (i.e. the square of the standard error in Table 5). For each of the simulations, we calculated the dominant eigenvalue and the elasticities.

We used the simulated dominant eigenvalues to calculate the 95% confidence interval (CI) around λ1(B), and tested whether the elasticities showed a significant difference using the distribution of the simulated elasticities. Pairwise comparisons were made between all simulated fecundity elasticities and between all simulated survival elasticities. We first calculated the proportion pof the differences between the 10 000 simulated values that was either smaller or larger than zero (whichever was smallest). This gives a pairwise test value (e.g. Manly 1997) for whether two elasticities are equal after taking into account demographic uncertainty in the estimates of fecundity and survival. Overall significance of the multiple comparison of all pairwise P-values was based on the Holm–Bonferroni method (Holm 1979) retaining the overall significance level α = 0.05. The elasticity values can then be grouped into groups that differ from each other significantly.

Population dynamics: model results

The dominant eigenvalue of the population dynamical model was 0.981 (95% CI: 0.867–1.100). We considered this a reasonable reflection of the observed dynamics, because the CI clearly overlaps with 1 (stable cycle-to-cycle population dynamics). The expected stage/phase distribution (given by the right eigenvector) produced a satisfactory fit (χ2 = 9.60, d.f. = 5, P = 0.09) to the observed distribution (Fig. 5b), although the model predicted a relatively large proportion of first breeders in the increase phase. In the field data, the number of experienced breeders stayed fairly constant across phases (Figs 5b,c,f and 2). The matrix model showed that the survival of experienced breeders was more important for population growth than that of inexperienced breeders and that the main difference stemmed from the survival elasticity of an experienced breeder in the low phase (Fig. 5a). The elasticity values of reproduction showed only small differences, with none of the non-zero elasticities differing significantly (Fig. 5a).

Figure 5.

 The graph at the top shows elasticity of survival (filled dots) and fecundity (open dots) for both stages in the different phases of the vole cycle (x-axis is the same for both panels). Grouping of elasticities in significantly different groups is indicated with letter coding and is based on Bonferroni–Holm-corrected multiple comparison of 10 000 simulated elasticity values that take into account demographic stochasticity in fecundity and survival values. Elasticities that share a common letter do not differ significantly. Dotted lines are for visual presentation only. The graph at the bottom shows the expected number of reproductively active territories (black bars) derived from the population matrix model compared with the observed number of territories (white bars) where a clutch was produced. In those territories of the actual data where parents were not identified, the parents were assumed to be first breeders if a clutch was not produced in the territory in the previous year, and experienced breeders if a clutch was produced in the territory in the previous year. There is no significant difference in the number of territories expected by the model and observed in nature.


Numerical response of tawny owls

We have analysed 15 years (five 3-year vole cycles) of data on both sexes of tawny owls and their small mammal prey in a population in southern Finland. Based on responses to playback, the number of active territories has been censused in this population, allowing us to investigate the response of the total population size to fluctuations in vole abundance. We find that territorial activity (potential breeders) in the tawny owl population in southern Finland varies between vole cycle phases in response to fluctuations in the abundance of their main prey. Thus, our result differs from studies of tawny owls in temperate regions without multi-annual fluctuations in the abundance of their main prey. In particular, territory numbers in temperate regions are stable between years (Southern 1970; Hirons 1985; Jedrzejewski et al. 1996; Sunde & Bølstad 2004; Desfor, Boomsma & Sunde 2007). We find that the variation in territory number between vole cycle phases is not driven by extensive mortality after the decrease phase as in the closely related Ural owl living in the same environment (Brommer et al. 1998, 2002), because tawny owl survival is fairly stable between vole phases. Instead, recruitment of new breeders to the population (including immigration) drives the variation in territory numbers between vole phases and constitutes the main part of the observed numerical response of the species. Because first breeders survive less well than experienced breeders, they are less important for population growth, and a population dynamical model emphasizes that the survival of experienced breeders is most important for population growth above other components.

Stage-dependent reproduction and survival

A numerical response of the number of breeding pairs to the abundance of their main prey has been documented in several owl studies (Tengmalm’s owl Aegolius funereus and voles: Korpimäki & Norrdahl 1991; barn owl Tyto alba and voles: Taylor 1994; tawny owl and voles: Jedrzejewski et al. 1996; Ural owl and voles: Brommer et al. 2002; Northern spotted owl Strix occidentalis caurina and deer mice: Rosenberg, Swindle & Anthony 2003). These studies have, however, not accounted for any sex- or stage-related differences. We find that fluctuating prey abundance has distinctive consequences for the age composition of first breeding and experienced individuals. In particular, the majority of tawny owl offspring hatched in an increase phase starts to breed already as 1-year olds in the following decrease phase when conditions (at least at the onset of reproduction) are favourable (see also Brommer et al. 1998). This flexible and potentially young age at first breeding allows for a rapid tracking of the vole cycle. As a further consequence, one stage in the population (first breeders) becomes younger as voles increase in abundance, whereas already established territorial individuals get older when voles become more abundant. Hence, over the course of a cycle, the structure of the population changes, in terms of both the proportion of first vs. experienced breeders and their age distributions.

Bird populations consist of two sexes, but most studies on birds of prey consider one sex only. In birds of prey, males and females are sexually dimorphic in size and have markedly different parental roles. Males and females may therefore respond differently to variations in prey availability. The smaller males are the main hunters during courtship, prior and during incubation, and they do most of the hunting during the nestling stage. The larger female incubates and concentrates on defending the brood (e.g. Lundberg 1986; Wallin 1987; Newton 1989; Meijer, Daan & Hall 1990; Sunde, Bølstad & Möller 2003). However, in spite of these different sex roles, both males and females largely respond similarly to variations in prey availability in terms of their numerical response, reproduction and survival. Our results thus correspond with those of Krüger (2002), who found that the lifetime reproductive success of male and female common buzzard Buteo buteo were affected by similar factors.

We find strong contrasts between first and experienced breeders also in terms of their reproduction. First-breeding tawny owls have a much lower production of recruits in all phases of the vole cycle. The future prospects of tawny owl offspring thus depend to a large extent on whether their parents have reproduced before or not. Again, this effect may be due to the parents gaining experience (individually or together, as most pairs breeding the first time consist of two inexperienced parents and parents rarely divorce). Alternatively, poor parents are selected out of the population after their first breeding attempt. In addition to parental experience, the vole cycle itself creates different future prospects for the offspring. Over 80% of the local recruits produced in the increase phase are able to start breeding as 1-year olds, whereas the majority of recruits born in low or decrease phases start breeding as 2-year olds. This aspect of tawny owl reproduction is similar to Tengmalm’s owls (Korpimäki 1988, 1992) and Ural owls (Brommer et al. 1998). Intriguingly, experienced tawny owls are rather successful in producing recruits even in a low phase as recruitment probability of offspring hatched in a low phase is almost equal to that of offspring hatched in the decrease phase. This latter finding is in contrast to Tengmalm’s owls, which rarely breed during low phases (Korpimäki & Norrdahl 1989; Laaksonen, Korpimäki & Hakkarainen 2002), and is also in contrast to Ural owls, which do breed in low numbers, but rarely produce recruits in low phases (Brommer et al. 1998). This difference is likely caused by the fact that tawny owl reproduction is less dependent on small voles than the reproduction of the other two owl species. In particular, water voles Arvicola terrestris, which do not follow the 3-year vole cycle, are an important prey item affecting tawny owl reproduction in Finland (Kekkonen et al. 2008).

We find only marginal effects of the vole cycle on tawny owl survival. This is in contrast to findings in Finnish Ural owls, where CMR analysis revealed a reduced survival probability after a decrease phase (Brommer et al. 2002). Similarly, in a recapture and recovery analysis of long-term nationwide data on Finnish tawny owls, Francis & Saurola (2004) found that most (more than 50%) of the variation in adult survival is explained by winter temperature, with only 9% of variation explained by the vole cycle. In addition to winter temperature, which increases energy expenditure in owls (Mosher & Henny 1976), also the depth of the snow cover may affect mortality in tawny owls. A deep snow cover makes vole prey less accessible during winter. For example, in Swiss barn owls Tyto alba, the number of days with snow cover during winter explained substantial variation in annual survival (Altwegg et al. 2003, 2006). We find that first-breeding tawny owls of both sexes have lower survival than experienced individuals. It is therefore possible that, in Northern tawny owl populations such as the one studied here, winter severity has prominent effects on survival, especially in first breeders. We do not have sufficient data to distinguish whether the higher survival of experienced breeders is due to their older age or because they have gained more experience than first breeders. In addition, experienced birds represent, by definition, a subset of owls that have already undergone survival selection as a first breeder and may therefore, because of intrinsic or extrinsic factors, enjoy a higher survival than the unselected group of first breeders. Whatever the reason for the observed differences in survival, our primary interest here is to understand its consequences for population dynamics.

Our CMR analysis also revealed that recapture probability varied between vole cycle phases for first breeders whereas it varied annually for experienced breeders. Previously recapture rate has been found to vary in tawny owls due to individual characteristics (Roulin et al. 2003). We find here that recapture rate of first breeders is the lowest in a low vole phase. This is unlikely to be due to movement out of the population as we find that both first and experienced breeders are highly site tenacious. Instead, we find it most likely that these individuals refrain from breeding under such poor food conditions (see also Southern 1970; Brommer et al. 2002) and not that they move out of the population, because we find significantly lower breeding activity in the population in low phases compared with other vole phases. Therefore, we suggest that refraining from breeding is more common after the first breeding event than at later stages and explains the vole cycle effect on recapture rate of first breeders.

The vole cycle and tawny owl population dynamics

Individuals of different stages of a population are often affected differently in an environment where food availability fluctuates annually or in a cyclic manner (Pietiäinen 1988; Ratcliffe, Furness & Hamer 1998; Cam & Monnat 2000; Krüger & Lindström 2001; Laaksonen et al. 2002). To draw conclusions on the persistence of a population, it is therefore crucial to evaluate how these environmental fluctuations affect reproduction and survival of different stages, and how these differences translate into population dynamics (Caswell 2001). We have here taken a pragmatic view, and rather than considering age-dependent performance (on which we have incomplete information), we distinguish stages according to their breeding experience (first time vs. experienced breeders). Our cycle-to-cycle population dynamical matrix model, which was parameterized with stage- and phase-specific estimates of reproduction, survival and immigration, produced a reasonable estimate of population growth and the numerical distribution of the two stages across the vole cycle’s phases. First breeders have a strong numerical response as prey abundance increases, whereas the number of territories of experienced breeders remain fairly stable between vole cycle phases. Hence, our population dynamical model captures the essentials of the tawny owl population dynamics.

The matrix model shows that experienced breeders are most important for population growth mainly due to their higher survival. Similarly, barn owl population growth rate is highly sensitive to changes in the survival of adults but markedly less to changes in the survival of yearlings (Altwegg et al. 2007). To some extent, these differences are inflated by structuring the population into two stages (cf. Reid et al. 2004). The first-breeder stage necessarily equals 1year, whereas the experienced-breeder stage lumps all experienced breeders into a single category independently of how long they have been experienced breeders [breeding life span in the population is c. 2.5 years (Brommer, Ahola & Karstinen 2005)]. However, clear differences between phase-specific elasticity in tawny owl survival between the two stages remain. Intriguingly, the population-dynamical impact of survival of the experienced breeders after a low vole phase is higher compared with increase and decrease phases, whereas this pattern is absent in the first breeders. This difference is because experienced breeders that survive a low phase contribute strongly to the population during increase phases when reproduction is favourable. The elasticity of survival of first breeders mainly tends to increase from low to increase and decrease phases as they become more abundant in the population.

A second striking finding of our modelling exercise was that there was no strong population-dynamical impact of the almost threefold difference in offspring recruitment probability between increase and decrease phases that occurred in both first and experienced breeders. In fact, the proportionally much higher recruitment of offspring hatched in the increase phase than in any other phase is a general characteristic of an avian predator preying on cyclically fluctuating vole populations (Korpimäki 1992; Brommer et al. 1998). Increase vole phases have therefore been considered to be of major importance for population persistence and viability. For example, offspring hatched in the increase phase starts to breed at an earlier age (this study; Brommer et al. 1998; Laaksonen et al. 2002). However, no study has – to our knowledge – investigated the actual population dynamical impact of this difference in recruitment probability by integrating fecundity and survival differences in a proper model. We find that the elasticity of reproduction in the increase phase was not any higher than the value for the decrease phase in both first and experienced breeders. Hence, this finding shows that explicit modelling of the dynamics can provide non-intuitive insights.


Experience-dependent reproductive performance and survival are key aspects to be taken into account when modelling population dynamics in variable environments. This is because changes in food supply will have a clear impact on the proportion of first breeders in the population. Whenever first breeders have a markedly lower fecundity and survival than experienced breeders, such changes in the composition of the population will have population dynamical consequences, especially when the population dynamics are primarily driven by the recruitment of new breeders in years of abundant food supply. Our findings further caution for interpreting possibly large differences in fecundity across years of varying quality as indicative of having population dynamical impact, and we emphasize the need to integrate fecundity and survival in a model for making robust conclusions on the population dynamics.


This is report number 6 of Kimpari Bird Projects. We thank the other members of KBP – Juhani Ahola, Pentti Ahola, Bo Ekstam, Arto Laesvuori and Martti Virolainen – for the many hours spent conducting field work. We thank Hannu Pietiäinen, Peter Sunde and two anonymous referees for insightful comments. Jari Valkama and Seppo Niiranen from the Finnish Ringing Bureau kindly provided data on tawny owl recruits. Author contributions: all data were collected by KA and TK; initial analyses were carried out by AZ; final analyses and writing by PK and JEB. AZ was supported by CIMO, PK by the Academy of Finland (project 1118484 to JEB) and JEB was employed as an Academy Researcher.