1. Breeding territory choice constitutes a crucial antipredator behaviour for animals that determines reproductive success and survival during the breeding season. On arrival to breeding grounds migrant prey face a multitude of ‘waiting’ predators already settled within the landscape.
2. We studied territory selection and reproductive investment of migrant pied flycatcher (Ficedula hypoleuca) relative to breeding pygmy owls (POs) (Glaucidium passerinum) and Tengmalm’s owls (TOs) (Aegolius funereus). Diurnal POs present a greater predation threat to adult flycatchers (up to 80% songbirds in diet) compared with nocturnal TOs (up to 36%).
3. During territory selection, pied flycatchers strongly avoided POs (occupation: 42% in presence vs. 92% in absence of owl nest) but not TOs (80% vs. 75%). This suggests that flycatchers are able to distinguish between two potential predators, avoiding dangerous POs but not obviously responding to the less risky TOs.
4. Flycatchers responded to presence of PO nests with c. 4-day delay in the start of egg-laying. A significantly prolonged nest building period contributed to this potentially costly breeding delay. Flycatchers further significantly reduced initial reproductive investment in presence of POs by laying 8·2% smaller clutch sizes, even if laying date was controlled. No breeding delay and clutch size reduction was found relative to TO presence.
5. Our results highlight flexibility in breeding territory selection and reproductive strategies as antipredator responses to perceived risk in a multi-predator environment. This supports the idea that for prey, not all predators are equal.
Predators actively reduce prey numbers but may also significantly affect behaviour, sociality and reproductive success of prey. Prey animals may change, for example, their habitat use, activity periods and reproductive investment under the risk imposed by predators (reviewed by Caro 2005). From the prey perspective, antipredator behaviours are under the influence of many predator species. In a multi-predator environment, behaviours evolved to avoid one predator could alter the risk represented by other predators in the same habitat (Lima 1992; Sih, Englund & Wooster 1998). Therefore, understanding the impact of multiple predators on prey behaviour and decision making is vital for a comprehensive understanding of predator–prey interactions.
Predation risk has an important role in determining habitat quality. Recent evidence suggests that animals gather information about the quality of the habitat prior to making breeding territory decisions (Danchin et al. 2004). For example, settling birds are able to gather accurate information on the location of avian predator nests (Norrdahl & Korpimäki 1998; Thomson et al. 2006a; reviewed in Lima 2009). From this perspective, migrant birds represent an interesting system for studying breeding habitat selection because on arrival to breeding sites they face a multitude of ‘waiting’ predators already settled within the so-called predation risk landscape (Thomson et al. 2006a; Mönkkönen et al. 2007). Importantly, nesting territory choice can be seen as a crucial antipredator behaviour for animals and their offspring, a decision that has impacts for an entire breeding season.
We studied settlement of pied flycatchers (Ficedula hypoleuca, Pallas) in forest patches with predation risk induced by breeding pygmy owls (Glaucidium passerinum, L.; hereafter POs) and Tengmalm’s owls (Aegolius funereus, L.; hereafter TOs). Our study design included forest patches with breeding owls and patches where owls bred the year before but were not currently breeding. This design allows us to partly control for the influence of owl breeding habitat choice and permits to focus on passerine habitat selection decisions. The two owl species belong to the same functional group of predators and are visually quite similar. Both owls represent a risk to and elicit warning calls from small songbirds, which are their alternative prey (Kellomäki 1977; Korpimäki 1988; Cockrem & Silverin 2002; Hakkarainen et al. 2002). In reality although, the two owl species represent different actual predation risk to adult passerines. POs are diurnal, have flexible hunting strategies and show a high proportion of small birds in their diet in poor vole years (25–80% of diet, of which 1·8% are pied flycatchers; Kellomäki 1977; Kullberg 1995). TOs are nocturnal, display a single hunting strategy and have a lower proportion of small birds in their diet, also in poor vole years (20–36%, of which 0·1% are pied flycatchers; Korpimäki 1988; Bye, Jacobsen & Sonerud 1992).
In this multi-predator design at the forest patch scale, we expected lower nest box occupation, delayed start of egg-laying and reduced clutch size of pied flycatchers in presence of breeding owls. Because diurnal POs likely represent a larger predation threat to pied flycatchers than nocturnal TOs, we predicted accentuated responses to the presence of PO risk than to the presence of TO risk.
Materials and methods
This study was conducted in the Kauhava region, western Finland (63°N, 23°E), from May to July 2006 and 2007. The landscape consists of forest patches, mainly characterized by pine and spruce with some birch-dominated patches, interspersed with a high proportion of agricultural land and clear cut areas (Hakkarainen et al. 2003). The study area covers approximately 1300 km2 which contains approximately 500 nest-boxes of TO (Korpimäki & Hakkarainen 1991), and 200 forest patches with two PO nest boxes 100 m apart available to nesting owls.
Vole populations in our study area exhibit distinct high-amplitude (50–200-fold) 3-year cycles with sequential low, increasing and decreasing densities (Korpimäki et al. 2005). During the increasing phase of the 3-year cycle, vole densities are intermediate in the early stages of the breeding season of owls, and increase in the course of the summer. During the decreasing phase, vole density is intermediate when owls choose nest sites (February–March), but crashes in the course of the breeding season (April–July). In the low phase, vole densities remain low throughout both spring and summer (Korpimäki et al. 2005). Vole population cycles largely determine the predation risk on the alternative prey of the two owl species, with increasing impact on small songbirds during the low and decreasing years of vole abundance (Kellomäki 1977; Korpimäki 1988). This study took place during years of decreasing (2006) and low (2007) vole abundances when both owl species considerably increase the proportion of small songbirds in their diet. The pied flycatcher is a migrant passerine that arrives at its breeding quarters only in early May, 2–6 weeks after the start of egg-laying of owls.
Our study was conducted in forest patches containing either actively nesting TO or PO pairs or patches containing unoccupied PO or TO nest boxes that were occupied by a breeding TO or PO pair previously (mostly the year before). In all forest patches, we placed four pied flycatcher nest boxes on average at 80 m from the owl nest box (occupied or empty). In total, we set up pied flycatcher boxes in 62 forest patches. Of these, 26 were TO patches (14 with TO nest and 12 without, only in 2006), while 36 were PO patches (16 with PO nests and 20 controls, pooled data from 2006 and 2007). Pied flycatcher responses were then compared between forest patches with an active owl nest present and forest patches currently without an owl nest. The presence of TOs and POs in their nests were checked weekly throughout the duration of the study to ensure activity at the owl nests during flycatcher settlement and egg-laying; no forest patch was occupied by both TOs and POs. On average, PO patches were 12 km apart (range: 1·9–24·5 km in 2006 and 2·4–14·9 km in 2007), whereas TO patches were on average 11·72 km apart (range: 0·8–24·6 km). The linear distance between active nests of TOs and POs averaged 12·2 km in 2006 (range: 0·9–24·7 km). TO sites were only used in 2006, whereas PO sites both in 2006 and 2007. Forest patches near all known goshawk (Accipiter gentilis, L.) nests were avoided during the study design and no sparrowhawk (Accipiter nisus, L.) nest has been found in the study site since 2004.
Throughout the season pied flycatcher nest boxes were checked every 2 days. Tit species (Parus spp.) were prevented from breeding in our nest boxes by placing boxes in patches just prior to pied flycatcher arrival and by removing any tit nest material appearing in boxes. During pied flycatcher settlement both POs and TOs were incubating and male owls call during this time, both for defending nest-hole and for calling the female during incubation feeding.
We used four response variables. First, the occupation of all nest boxes were determined; a box was defined as occupied when at least one pied flycatcher egg was laid in it. Occupation is important in our study because it reflects the actual territory choice by flycatcher pairs. As pied flycatchers males may occasionally attract second females which would impact our occupation variable (Lundberg & Alatalo 1992), we trapped all adult flycatchers to control for this problem. In addition, natural cavities are uncommon in the managed forests of our study area and are not assumed to impact our occupation variable. Secondly, we recorded nest building duration, defined as the number of days from the date of nest initiation to the day before the first egg was laid. Thirdly, we recorded the laying date of the first egg and fourthly, final clutch size was checked. We were unable to check the response in brood size or quality because the fourth laid egg was collected from each nest (and replaced with a dummy egg) for a study on maternal effects. However, we found no differences in proportion of nest failure (i.e. nests where no chicks fledged) between PO and TO patches (21% vs. 24% nest failure).
We analysed our data both by (i) comparing responses of flycatchers to PO and TO presence using the data from 2006 only and (ii) by using the pooled PO data from 2006 to 2007. This was done to check for bias created by the unbalanced design; this was especially important since pied flycatcher perception of risk could have been different between years due to vole cycle phase (2006: decreasing vole year and 2007: low vole year). However, there were no biologically meaningful differences in the results obtained; all variables that statistically differ with pooled data, differ also considering 2006 alone. Therefore, we present analyses of the pooled data only.
We used Generalized Linear Mixed Models (GLMM; PROC GLIMMIX, SAS statistical software, version 9·2) to compare pied flycatcher territory selection in the presence and absence of PO and TO nests. To test this flycatcher response, we used occupation rate as a dependent variable (binomial distribution, logit link) in the models, while the owl nest presence (i.e. forest patch with or without owls), owl species currently or previously breeding in the patch and the interaction between the owl nest presence and owl species, were used as fixed factors. Forest patch was added as random effect to control for the fact that four flycatcher boxes were available in each forest patch.
We used linear mixed models (LMM; PROC MIXED, SAS statistical software, version 9·2) to analyse nest building duration and laying date. Nest building duration was log-transformed and laying date square-root transformed to achieve the normal distribution required by the model. In both models, the random and fixed factors were the same as for the occupation model but with an additional class variable, neighbour. The ‘neighbour’ variable represents the order of arrival of pied flycatchers within a particular patch and was added as a covariate in all models to control for the number of conspecific pairs already breeding in the patch (range 0–3). The model for nest building duration also included nest initiation as covariate as late arrival is generally characterized by faster nest building (Lundberg & Alatalo 1992). Nest initiation was defined as the day when a female was seen near the nest box or when nest material was first found in the nest box.
Clutch size did not fit either normal or Poisson distribution even after transformation. For clutch size analysis we used the forest patch as the sampling unit and calculated mean clutch size per forest patch. We used Generalized Linear Models (proc GLM, SAS statistical software 9·2) with owl nest presence (class), owl species (class) and mean laying date of the patch (continuous) entered in the models.
If models showed a statistically significant ‘owl nest presence × owl species’ interaction or differences between owl nest presence, we performed analyses separately for the two owl species to find the direction of these significant interactions. In this case, we have used the same model (GLMM, LMM or GLM) but without considering owl species as a fixed factor. In all final models, we did not consider the effect of year. Data from TOs originated solely from 2006, while we pooled all data from POs because there were no statistical differences in flycatcher responses to PO nest presence between 2006 and 2007. Year did not explain pied flycatcher occupation (F1,101 = 1·43, P = 0·235), nest building (F1,175 = 1·68, P = 0·197), laying date (F1,124 = 0·30, P = 0·58) and clutch size (F = 2·31, P = 0·135).
Pied flycatcher nest box occupation showed a statistically significant interaction between the presence vs. absence of the owl nest (owl nest presence) and owl species, indicating that pied flycatchers responded to presence of the two owl species differently (Tables 1 and 2). Flycatchers occupied boxes in the presence of POs significantly less frequently than in the presence of TOs (Fig. 1). The presence of a PO nest significantly explained pied flycatcher nest box occupation (F1,68·9 = 30·43, P < 0·001). In patches with a PO nest, 42% of boxes were occupied (12 of 36 in 2006 and 14 of 28 in 2007); whereas in sites without a PO nest, 92% of boxes were occupied (48 of 52 in 2006 and 25 of 27 in 2007). In contrast, the presence of a TO nest did not influence flycatcher nest box occupation (F1,25·5 = 0·33, P = 0·57). Pied flycatchers occupied 80% (45 of 56) of boxes in patches with a TO nest and 75% (36 of 48) in patches without TOs.
Table 1. GLMM analyses of pied flycatcher occupation and LMM for nest building duration and laying date. Owl nest presence (patches with or without owl nest) and owl species (pygmy or Tengmalm’s owl) have been used as class in the model and forest patch as random effect. The variable ‘neighbours’ represents the number of conspecifics already breeding in the patch and is included as a categorical variable in nest building and laying date models; the variable ‘nest initiation’ has been add to control for female arrival date on nest building duration model
a‘d.f.’, degrees of freedom of the numerator for the F-test and ‘Den d.f.’, degrees of freedom of the denominator for the F-test.
Owl nest presence
Owl nest presence × owl species
Nest building duration (log transformed)
Owl nest presence
Owl nest presence × owl species
Laying date (Square root transformed)
Owl nest presence
Owl nest presence × owl species
Table 2. Estimates of mean (least square) and 95% confidence limits (lower– upper CL respectively) of the interaction ‘owl nest presence × owl species’ (patches with/without owl × pygmy or Tengmalm’s owl) from all the models used to analyse pied flycatcher responses: GLMM (occupation), LMM (nest building duration and laying date), GLM (clutch size). Estimates and confidence limits for nest building duration and laying date have been back transformed
Nest building duration (days)
Laying date (1 = 1st May)
Nest building duration
Pied flycatchers breeding in PO patches significantly prolonged nest building duration compared to flycatchers living in TO patches (Tables 1 and 2; Fig. 2a). This longer nest building period was explained by the presence of PO nests even when controlling for the fact that flycatchers delay their nest initiation by 2 days in patches with PO nests compared to patches without POs (F1,29 = 14·04, P = <0·001). In patches with a PO nest, pied flycatchers in 2006 built nests in 6·8 ± 0·9 days vs. 7·0 ± 0·3 in patches without a PO nest. In 2007 nest building took 8·8 ± 0·8 days in patches with a PO nest vs. 7·1 ± 0·4 days in patches without a PO nest. There were no obvious differences in the nest building duration in patches with or without a TO nest (F1,23·8 = 0·20, P = 0·66) (with TO nest 6·8 ± 0·3 days vs. 6·7 ± 0·4 in patches without TO nest). Interestingly, if all data were pooled, regardless of the presence of an owl nest or of the owl species, we found that later settling pied flycatchers built nests significantly faster than earlier settling flycatchers (linear regression: r = −0·273, P < 0·001). However, flycatchers settled on average 2 days later in sites with a PO nest but still on average took the same time or longer to build their nests.
Laying date showed a statistically significant interaction between owl species and the presence of an owl nest (Tables 1 and 2). In the presence of a PO nest pied flycatchers showed a 4-day delay in laying date compared to the absence of POs (F1,30·5 = 14·48, P = <0·001; Fig. 2b). In 2006, pied flycatchers in patches with a PO nest delayed egg-laying by 3 days relative to flycatchers in patches without a PO nest (30·5 ± 1·4 vs. 27·5 ± 0·39), whereas in 2007 the delay was almost 4 days (30·9 ± 1·04 vs. 27·4 ± 0·47). In contrast, there was no obvious difference in laying dates between patches with or without a TO nest (F1,19·4 = 1·32 P = 0·26; Fig. 2b) (27·9 ± 0·36 vs. 28·8 ± 0·68).
We found a significant impact of owl nest presence on pied flycatcher clutch size (F = 6·53, P = 0·014), but the interaction of owl nest presence × owl species was not significant (Table 2; Fig. 3). However, the data were split by owl species as done for other variables. Flycatchers laid significantly smaller clutches in the presence of PO nests (F = 6·57, P = 0·015) independently of the influence of laying date (F = 8·55, P = 0·006). This translates into smaller clutches of almost one less egg when breeding in the presence of PO nests compared to sites in absence of PO nests, a reduction in initial reproductive investment of 8·2% (in 2006: with PO 6·1 ± 0·23 vs. 6·60 ± 0·1 without PO eggs; in 2007 respectively: 6 ± 0·22 vs. 6·88 ± 0·13). There was no clutch size differences between the presence (6·6 ± 0·15 eggs) and absence (6·7 ± 0·1 eggs) of a TO nest (F = 2·37, P = 0·14; Fig. 3).
We found marked differences in the settlement and duration of nest building of pied flycatchers between sites occupied by the two owl species. As expected, pied flycatchers showed avoidance of forest patches occupied by breeding PO. Moreover, pied flycatcher nest building was significantly prolonged in forest patches with PO, occurring independently of nest initiation date. Therefore, in the presence of a PO nest, pied flycatchers delayed their egg-laying by 4 days. Unexpectedly, there were no obvious differences in settlement, duration of nest building and initiation of egg-laying in patches with or without TO nests. Pied flycatcher females breeding in the vicinity of PO showed a decrease of the initial reproductive investment by laying 8·2% smaller clutches. There was no clutch size response relative to TO.
Strong pied flycatcher avoidance of PO nest sites, but lack of response to TO nest sites, suggests an evolved ability to discriminate between these predators. Less than half of boxes were occupied in patches with PO nests, whereas all other patches had occupation rates in excess of 75%. Despite these predators being superficially similar, the difference in habitat selection response is noteworthy. Both owl species hunt small birds as an important alternative prey and therefore both represent a risk to flycatchers (Kellomäki 1977; Korpimäki 1988), but it appears that the greater threat represented by POs is perceived by pied flycatchers.
Pied flycatchers showed significantly prolonged nest building in the presence of PO nests. This longer nest building period resulted in a c. 4-day delay in the start of egg-laying independently of the date of nest initiation. Delayed breeding in seasonal environments is known to reduce reproductive success (Sergio et al. 2007) and late arriving birds generally build their nests quicker (Lundberg & Alatalo 1992; this study). However, under PO predation risk, both early and late arriving flycatcher females showed prolonged nest building duration. Nest building appears to entail a physiological cost even to cavity nesting species like the pied flycatcher (Moreno et al. 2008). Therefore, as in other costly parental activities, the nest building period presents a trade-off situation that until now has been neglected from the literature. We expect that the costs associated with nest building would increase under predation risk. This appears to be the first evidence of increased predation risk prolonging nest building duration and adding to costly breeding delays. Nevertheless, this period of the breeding cycle is flexible (Cresswell & McCleery 2003) and an adaptive prolonging of nest building to avoid some aspect of predation risk cannot be excluded.
Smaller clutches were laid in the presence of PO nests while no response was observed in TO sites. Pied flycatchers may modify their reproductive investment depending on predator species breeding in the area. It has been shown previously that birds breeding in sites with increased predation risk lay smaller clutches (Doligez & Clobert 2003; Eggers et al. 2006), however, these studies focussed on the impact of only one predator. Smaller clutches serve to reduce both the probability of nest detection by predators (Lima 1987; Fontaine & Martin 2006b) and parental predation risk. It is difficult to separate the effect of individual quality on our clutch size response, but this result emerged in analyses that controlled for flycatcher arrival time which incorporates an individual quality component. Therefore, we suggest that in addition to delaying the initiation of breeding, the perceived predation risk further lowered prey reproductive investment.
Our study also controlled for effects of indirect habitat selection strategies. Pied flycatchers and other passerines prefer breeding in the proximity of tits Parus spp. (Forsman, Seppänen & Mönkkönen 2002; Thomson, Forsman & Mönkkönen 2003), but we prevented tits from nesting in our nest boxes and very few natural cavities exist due to the managed nature of the forests in our study site. Willow tits (P. montanus, von Baldenstein) excavate their own cavities; but are unlikely to bias our results as they tend to nest randomly in relation to breeding avian predators (Thomson et al. 2006b). Consequently, it is highly unlikely that background tit densities explain flycatcher habitat selection response relative to the two avian predators.
In our study, we were unable to completely eliminate the influence of habitat characteristics associated with year-specific owl territory choice. Therefore, we cannot completely exclude the possibility that flycatcher territory selection depends on factors other than the presence of PO nests. However, all forest patches without breeding owls, used in this study, contained owl nests in previous years and the habitat did not change visibly between years. We thus consider our ‘control’ sites adequate to suggest that the response of habitat generalist pied flycatchers is due to the presence of breeding PO.
An alternative, but unlikely, explanation for lack of response in TO patches is that flycatchers were not aware of the presence of the owl nest. The two owl species in this study may indeed differ in visibility to flycatchers. POs would be more easily detectable for passerines due to largely diurnal habits (Kellomäki 1977) whereas TOs are mostly nocturnal (Hakkarainen, Koivunen & Korpimäki 1997). However, short northern summer nights (between 3 and 6 h) necessarily mean an overlap in the activity of flycatchers and owls, and TOs elicit warning calls from diurnal passerines (Cockrem & Silverin 2002; pers. observation). Irrespective of these reasons, the probability of a passerine being unaware of a predator nest c. 80 m away seems unlikely. Passerines appear to have detailed knowledge of the sites of breeding avian predators (e.g. Norrdahl & Korpimäki 1998; Thomson et al. 2006a; Duncan & Bednekoff 2008).
Perhaps the most likely explanation for the lack of pied flycatcher avoidance of TO patches is that there may be benefits of associating with TOs in the form of protection from other smaller predators. Protective nesting associations are common in avian communities and owls in northern latitudes feature prominently as protector species (Quinn & Ueta 2008). In our system, POs appear to avoid patches used by TOs (Suhonen et al. 2007) and remains of POs have been found in TO nests (E. Korpimäki unpubl. data) suggesting intra-guild predation (Sergio & Hiraldo 2008). Although we found no evidence of pied flycatcher attraction to patches with breeding TOs (but see Hakkarainen et al. 1997), the high flycatcher occupation rates in patches without TOs makes the detection of an active association with TO nests unlikely. Studies on the interactions between POs and TOs are needed to understand passerine habitat selection strategies relative to these predators.
Our results highlight that breeding habitat selection and reproductive investment are plastic antipredator behavioural responses that differ even between functionally similar avian predators. Flycatcher breeding territory choice differed substantially between the two predator species: POs were avoided but not TOs. Moreover, pied flycatcher females showed prolonged duration of nest building, with subsequent delay in egg-laying and reduction in clutch sizes. This lends support to the idea that for prey, predators are not equal (Van Buskirk 2001; Quinn et al. 2003; Botham et al. 2008). This is one of the few attempts to understand prey habitat selection in a multi-predator landscape, which simplifies the decision relative to two avian predators while leaving the multitude of other predators for future studies. However, this model system of two avian predators likely has its benefits to identify the sensitivity of the information used by passerines in making habitat selection decisions and represent an important step to show plasticity in antipredator behaviour in a natural environment.
We are thankful to Axel Strauss, Eric Le Tortorec and Rauno Varjonen for extensive help with the field work. We are thankful to Annette Heisswolf, Katerine Hoset and Toni Laaksonen for help with statistical analyses. Moreover, we would like to thank Fabrizio Sergio, John Quinn, an anonymous referee and the entire PhD student seminar at the Section of Ecology, University of Turku, for helpful comments. This project was supported by the Center for International Mobility (CIMO, grant no. TM-07-4819 to EK), the Fondazione Ing. Aldo Gini, the Finnish Cultural Foundation, South Ostrobothnia Regional fund, The Turku University Foundation (personal grant to CM) and the Kone Foundation (personal grant to RLT).