Interactive effects of concealment, parental behaviour. and predators on the survival of open passerine nests KAREL WEIDINGER

Transcription

1 Ecology , Interactive effects of concealment, parental behaviour Blackwell Science Ltd and predators on the survival of open passerine nests KAREL WEIDINGER Laboratory of Ornithology, Palacky University, tr. Svobody 26, Olomouc, Czech Republic Summary 1. A simple model of relative effects of parental behaviour (parents present vs. absent) and nest concealment on probability of nest predation was evaluated by measuring survival of paired natural/artificial nests of four open nesting passerines over 3 years. 2. The ratio of rodent to corvid predation (i) decreased from yellowhammer (small eggs, ground/near ground nests) through blackcap (small, shrub) to song thrush and blackbird (medium sized, shrub/subcanopy); (ii) was highest in years when rodent abundance peaked this effect was clear in yellowhammer, detectable in blackcap, but not found in thrushes. An inverse relationship was found between mean annual nest survival and abundance of the major nest predator for each species. Predators differed between poorly (corvids) and well-concealed (rodents) nests in blackcap. 3. The effects on nest survival differed among species, including: positive effect of parental behaviour combined with neutral effect of concealment (thrushes); independent positive effects of behaviour and concealment (yellowhammer); neutral effect of behaviour combined with positive effect of concealment (blackcap). These patterns are consistent with hypothesis that relatively larger species with conspicuous nests (thrushes) should either engage in more vigorous nest defence or their defence is more efficient, compared with smaller species with less conspicuous nests (blackcap). 4. The positive relationship between nest concealment and survival resulted either from an effect of nest site per se (yellowhammer) or from an effect of parental behaviour (blackcap) that changed from negative (nest disclosure) to positive (nest defence) between poorly and well-concealed nests, respectively. The view that lack of a withinspecies relationship between nest concealment and survival (thrushes) results from parental behaviour compensating for predation risk associated with poor nest concealment was not supported. 5. This study demonstrated (i) multiple interactions among factors influencing the rate of nest predation, both between and within species; (ii) potential bias associated with drawing general conclusions from small-scale experiments. Key-words: anti-predator strategy, nest defence, nest predation, nest site, parental activity. Ecology (2002) 71, Ecological Society Introduction Nest predation is one of the major selective agents shaping avian life histories. Anti-predator strategies evolved by birds include direct effects of parental behaviour (nest defence, sensu Montgomerie & Weatherhead 1988) as well as indirect ones, such as decision where (nest site selection) and when (timing) to breed. Nest defence may be active, including direct Correspondence: Karel Weidinger, Laboratory of Ornithology, Palacky University, tr. Svobody 26, Olomouc, Czech Republic. Fax: , weiding@prfnw.upol.cz attacks, mobbing, distraction display, nest guarding and vocalization, or passive through sitting on the nest (nest attentiveness), thereby hiding the eggs from view and serving to camouflage the nest and to discourage some of the potential predators (Montgomerie & Weatherhead 1988; Martin 1992a). The effectiveness of different kinds of nest defence depends on the bird s capacity to perceive risk associated with the particular predator (McLean & Rhodes 1991; Burhans 2000; Veen et al. 2000) and capacity to deter or distract it (Blancher & Robertson 1982; Komdeur & Kats 1999; Schmidt et al. 2001a). Apart from cost through allocation of time between nest defence and competing activities (Martin 1992a; Komdeur & Kats 1999),

2 425 Nest predation in passerines parental activity associated with nesting can be costly because it can attract predator s attention to the nest (Martin et al. 2000a; Martin, Scott & Menge 2000b). The relative strength of the positive ( nest defence ) and negative ( nest disclosure ) effects of parental activity on nest survival in particular bird species is likely to depend on the type of nest predators involved and their foraging behaviour. Selection of nest sites by birds is supposed to be nonrandom and adaptive with regard to the risk of predation (Martin 1998; Clark & Shutler 1999; but see Misenhelter & Rotenberry 2000 and Willson & Gende 2000). By selecting safe nest sites, birds can either decrease nest encounter rate for incidental nest predators (Vickery, Hunter & Wells 1992; Schmidt, Goheen & Naumann 2001b) or decrease nest detectability for actively searching predators. Among the various nest site characteristics, the concealment by vegetation is probably the most important factor for investigators attention. Yet, previous studies yielded conflicting evidence for the expected positive relationship between nest concealment and survival (Martin 1992b; Götmark et al. 1995; Howlett & Stutchbury 1996; Burhans & Thompson 1998). The lack of consistent patterns in nest predation studies has traditionally been attributed to spatial/ temporal variation in abundance of nest predators, or to interactions among the effects of nest site, parental and predator behaviour (Götmark et al. 1995; Hogstad 1995; Murphy, Cummings & Palmer 1997; Söderstrom, Pärt & Rydén 1998; Rangen, Clark & Hobson 1999; Schmidt 1999; Dion, Hobson & Larivière 2000; Martin et al. 2000b; Schmidt et al. 2001b). For example, Cresswell (1997) ascribed the lack of uniform (across species) effect of nest concealment to a compensating effect of parental nest defence operating on the within-species level. This hypothesis predicts that species capable of efficient nest defence will show a weaker relationship between nest concealment and survival than species that cannot deter predators and must rely on passive protection through nest concealment. Assumption implicit in this explanation is that on the within-species level, parent birds can respond to nest site associated risk of predation by altering their behaviour (McLean, Smith & Stewart 1986; Hobson, Bouchart & Sealy 1988; Burhans & Thompson 2001). Ricklefs (1977) found some support for an interspecific relationship between mean nest concealment and intensity of nest defence. Yet, sound interpretation of the observed patterns of predation is often limited by unknown identity of principal nest predators in the particular study system (Söderstrom et al. 1998; Pietz & Granfors 2000; Willson & Gende 2000). Relative effects of parental behaviour and nest site have often been inferred from comparisons between artificial (without parents) and natural (with parents) nests (Cresswell 1997; Wilson, Brittingham & Goodrich 1998; King et al. 1999; Komdeur & Kats 1999; Davison & Bollinger 2000; Martin et al. 2000b). Considering the diversity of patterns documented in such studies, there has been little effort to interpret them within a conceptual framework permitting formulation of testable hypotheses about the underlying mechanisms (Schmidt 1999; Schmidt et al. 2001a,b). Next I present a simple model of the possible mechanisms behind the patterns in survival of natural and artificial nests (Fig. 1). It is assumed that survival of experimental nests reflects effect of nest site (e.g. concealment) regardless of parental behaviour, whereas survival of natural nests reflects combined effects of nest site and behaviour. The positive (nest defence) or negative (nest disclosure) effects of parental behaviour are broadly defined as any effects due to presence of parents at the nest that have positive or negative effect on nest survival. These positive and negative effects may combine, thus yielding the net effect of behaviour that can be neutral (positive and negative effects offset each other), positive or negative. This net effect can be inferred from comparison of the mean survival rates between natural and experimental nests: no difference (neutral effect; Fig. 1A); higher survival of natural nests (positive effect; Fig. 1B); lower survival of natural nests (negative effect; Fig. 1C). Effect of nest concealment unaffected by parental Fig. 1. Model relating nest survival to possible effects of parental behaviour and nest concealment, as reflected in survival of natural (with parents; solid lines and filled points) and experimental (without parents; dashed lines and open points) nests. Points indicate the mean survival rate across all levels of nest concealment. The net effect of parental behaviour can be: (A) neutral either no effect or antagonistic effects that offset each other; (B) positive nest defence ; (C) negative nest disclosure. Effect of nest concealment unconfounded by parental behaviour can be: (a, d) neutral; (b, c) positive. Effects of parental behaviour and nest concealment can be: (a, c) independent; (b, d) interacting. Each subplot is identified by column (A C) and row (a d) coordinates.

3 426 K. Weidinger behaviour can be inferred from the relationship shown by experimental nests: neutral effect (Fig. 1a,d); positive effect (Fig. 1b,c). The null model is that neither behaviour nor concealment effect is present (Fig. 1Aa). Any of the behaviour effects may combine with any of the concealment effects, both effects being either independent (Fig. 1a,c) or interacting (Fig. 1b,d). Interaction means that the effect of behaviour is not constant across different levels of nest concealment and can be of two opposite directions. First, parental behaviour can compensate for the effect of nest concealment, yielding no relationship between concealment and survival in natural nests (Fig. 1b). Alternatively, parental behaviour can generate relationship between concealment and survival of natural nests in cases, where effect of nest concealment per se is neutral (Fig. 1d). The model shows that each of the two basic patterns exhibited by survival of natural nests (Fig. 1a,b vs. c,d) can be explained by several alternative mechanisms. First, lack of relationship between concealment and survival of natural nests (Fig. 1a,b) can result either from no effect of concealment combined with independent effect of parental behaviour (Fig. 1a), or from positive effect of concealment masked by an interacting effect of behaviour (Fig. 1b). Second, the positive relationship between concealment and survival of natural nests (Fig. 1c,d) can result either from positive effect of concealment combined with independent effect of parental behaviour (Fig. 1c), or may be due to effect of behaviour interacting with no effect of concealment (Fig. 1d). The above alternatives generate different predictions for survival of artificial nests relative to that of natural nests, and thus can be evaluated experimentally. This study was designed to test these alternative hypotheses using data on four open nesting passerines characterized by different body size (presumed correlate of defence potential) and different nest size and placement (correlates of overall nest conspicuousness, accessibility to predators and risk of disclosure through parental activity). The survival of natural and experimental nests of each species was concurrently monitored to separate the relative effects of parental behaviour and nest site. Cresswell s (1997) hypothesis generates the following predictions for the species under study. Song thrush (Turdus philomelos, Brehm) and blackbird (T. merula, Linnaeus) relatively high nest defence potential and conspicuous nests: natural nests should survive on average better than experimental nests; there should be no relationship between concealment and survival for natural nests but a positive one for experimental nests (Fig. 1Bb). Blackcap (Sylvia atricapilla, Linnaeus) and yellowhammer (Emberiza citrinella, Linnaeus) relatively low nest defence potential and concealed nests: there should be no difference in mean survival rates between natural and experimental nests, survival should increase with concealment in both natural and experimental nests (Fig. 1Ac). Methods STUDY AREA The study was conducted in a farmland area of c 83 km 2 north of the village Luze (49 54 N, E, m above sea level) in the Eastern Bohemia, Czech Republic, in The local landscape is a mosaic of arable land with settlements and remnants ( 25 ha) of deciduous woodland. The tree vegetation is dominated by Fraxinus excelsior, Populus nigra, Alnus glutinosa, Quercus petraea and Salix spp., and the dense shrub layer is formed by Sambucus nigra, Padus avium and Prunus spinosa. The study took place on plots (2 10 ha) representative of local woodland habitats and located in distinct habitat patches surrounded by arable land or grassland. The median distance to the nearest neighbouring plot was 750 m, minimum was 300 m. MODEL SPECIES Three (1998, 1999) or four (2000; see Data analysis) open-cup nesting passerine species that differ in nest placement (Fig. 2), nest size and egg size and crypsis were studied. Yellowhammer: medium sized nest placed on or close to ground and usually well hidden in herbaceous vegetation; cryptic eggs (21 6 mm 16 2 Fig. 2. Relationship between concealment category and height above ground for active ( ) and experimental ( ) nests. Median and interquartile range are shown. Relative frequency of concealment categories for active (left bars) and experimental (right bars) nests shown by histogram. Note that there is a partial overlap between the samples of active and experimental nests (see Methods for samples definition); the total number of physical nests was 201 in yellowhammer, 885 in blackcap and 532 in thrushes.

4 427 Nest predation in passerines mm); only female incubates. Blackcap: small nest placed typically on thin branches of shrubs, m above ground, moderately concealed; cryptic eggs (19 6 mm 14 7 mm); male and female take about equal share of incubation. Song thrush: conspicuous bulky nest, placed in shrubs or trees, m above ground, often little concealed; blue-green eggs (27 3 mm 20 6 mm) contrast with yellowish and unlined nest cup; only female incubates. Blackbird: similar nest as in song thrush but with cup lining; cryptic eggs (29 4 mm 21 6 mm); only female incubates. NATURAL NESTS Nests were systematically searched for in the shrub and herbaceous vegetation with approximately constant effort from about middle of April until the end of July (Weidinger 2001a). Nest content was inspected every 5th day to determine fate. All successive visits to each individual nest were approximately (±2 h) at the same time of day, thus keeping the exposure time comparable among nests. All nests that failed from reasons other than predation (inclement weather, disturbance by forestry or agriculture) were excluded from the analysis. The following nest site characteristics were recorded on the last visit to a nest: nest supporting plant, height above ground (m), concealment category (subjective score ranked from 0 = low to 4 = high). The latter was obtained as a sum of horizontal and overhead concealment (each scored as: 0 = nest well visible from most directions; 1 = intermediate; 2 = nest not visible in any direction from a distance of c. 2 m). A total of 146 (yellowhammer), 654 (blackcap), 225 (song thrush) and 78 (blackbird; only in 2000) nests were analysed. EXPERIMENT 1A Natural nests with known fate (successful or predated) were left in their original position and baited with artificial clutches about 15 days after the natural nesting attempt had terminated. The artificial clutches consisted of four plastic eggs attached to the nest and mimicking the size and coloration of real eggs (for details see Weidinger 2001b). The levels of disturbance to natural and experimental nests were maintained at a similar level. Because natural nests were inspected from close proximity and eggs were handled to determine the stage of incubation, human scent was present on both experimental and natural nests. Nest markers were not used. Experimental nests were exposed for a 15-day period, which was close to the mean duration of the egg stage (laying and incubation) of the natural nests (16 days in yellowhammer and blackcap, 17 days in thrushes), and were checked in 5-day intervals in the same way as the natural nests. A nest was considered depredated when at least one egg was damaged, pulled out of the nests, missing or buried in nest lining. Because of development of vegetation throughout the breeding season, I recorded nest concealment also after termination of each experimental nest; the other nest characteristics remained the same as for the natural nest. Data were collected in 1998 and 1999 from a total of 105 (yellowhammer), 418 (blackcap) and 183 (song thrush) experimental nests. EXPERIMENT 1B All procedures were the same as in Experiment 1a, except that the artificial clutches consisted of one Japanese quail (Coturnix japonica) egg not attached to the nest. Data were collected in 2000 from 52 (yellowhammer), 325 (blackcap), 125 (song thrush) and 129 (blackbird) experimental nests. EXPERIMENT 2 Active nests of blackcap found from the stage of laying ( 2 eggs laid) to the 2nd day after clutch completion (i.e. 10 days before hatching) were paired with an experimental nest undamaged blackcap nests collected on the study plots in the previous breeding season. These nests were baited with a clutch of four tree sparrow (Passer montanus) eggs, the size (mean: 19 1 mm 14 2 mm) and coloration (degree of crypsis) of which resembled those of blackcap. Experimental nests were placed within m of the paired active nest. Nest sites were kept similar within each nest pair: the same species of nest supporting plant or at least similar type of vegetation, height above ground, concealment and microhabitat (e.g. isolated shrub vs. closed shrub layer). Before placing an experimental nest, I tried to make sure that no other nest was active within a radius c. 10 m except the focal blackcap nest. Given the difficulties with satisfying all the above conditions, not all blackcap nests found could be included in this experiment. Paired nests were checked simultaneously in 5- day intervals to determine fate (active nest) or survival over the 15 day exposure period (experimental nests). When either of the paired nests was depredated, I continued visiting the remaining nest. Nest site characteristics were recorded in the same way for both active and experimental nests. In all, 46 (1999) and 74 (2000) nest pairs provided useful data. PREDATORS The occurrence of potential avian predators (corvids) was recorded visually throughout the breeding season. A frequency of species occurrence in the total number of plot-visits (528 per year; each lasted about 2 hours) was used as a rough measure of relative abundance. Small mammals were snap-trapped on the study (plus two additional) plots in the second half of August (24 plots 2 transects 30 traps 1 night exposure = 1440 trap-nights per year). One transect was placed along the habitat edge and the other inside the patch within each study plot. No attempt was made to determine predators of natural nests from egg remains or nest

5 428 K. Weidinger appearance (Pietz & Granfors 2000). Predators (classified as bird vs. mammal) of experimental nests could be inferred from tooth or beak marks left in the soft layer of paint on the plastic eggs (Experiment 1a) or from remains of quail eggs (Experiment 1b). In the latter case it was only possible to distinguish eggs attacked by small mammals (unbroken eggs or egg shells with tooth scratches; Marini & Melo 1998) from those destroyed by a large predator. Many quail eggs disappeared from the nests, which prevented any inferences about predators involved. A study aimed at identification of predators of artificial nests was initiated in 2000 on the same plots where the experiments were running. Automatic cameras (Olympus AF-10 mini) equipped with a photocell trigger were placed at a distance of c. 1 5 m from inactive nests of the model species, baited with plastic and quail eggs in the same way as the experimental nests. The first year of the camera study provided only qualitative results. The frequently documented nest predators included jay (Garrulus glandarius), pine marten (Martes martes), wood/yellow-necked mouse (Apodemus sylvaticus/a. flavicolis) and bank vole (Clethrionomys glareolus). Occasional predators included red squirrel (Sciurus vulgaris), weasel/ stoat (Mustela nivalis/m. erminea) and great-spotted woodpecker (Dendrocopos major). DATA ANALYSIS The unit of analysis in Experiment 1 was a nest, on which two repeated measurements of survival were made: one for natural nesting attempt (i.e. active nest; A) and the second for the experimental nest (E). I defined two data sets of different extent according to whether only paired nests (nests for which both repeated measurements were available) or all nests (including those nests with either A or E measurement missing) were included. Paired nests (total n = 882) represent a subset of all nests (total n = 1618). The unit of analysis in Experiment 2 was a pair of simultaneously exposed nests (A and E; total n = 120). Two response variables were used to measure nest survival. (i) Daily survival rate ( DSR = number of successful days divided by the number of exposure days; Aebischer 1999); applicable to all experimental nests and to those natural nests that have been observed active during the incubation stage (from laying of the penultimate egg to hatch day). Number of exposure days was estimated by the Early termination variant of the Mayfield method (Manolis, Andersen & Cuthbert 2000) with an exception that the exposure period of successful (i.e. hatched) nests was truncated by the hatching day and not by the last visit before the potential hatching interval. For the purposes of graphical presentation, I extrapolated DSR to a 15-day period (Hensler 1985). (ii) Survival over 15-day exposure period ( FATE = binary response coded 0/ 1); applicable to all experimental nests and to those natural nests found active more than 15 days before the earliest possible fledging date (defined as the day when nestlings were 8 days old). Nests analysed using the FATE response represent a subset of nests analysed using the DSR response. Simultaneous effects of several predictors on either type of response variable were analysed by fitting generalized linear models, with logit link and binomial error distribution (PROC GENMOD; SAS Institute Inc. 2000), under general guidelines by Crawley (1993) and Aebischer (1999). The original models containing all two- and three-way interactions were reduced by sequentially deleting nonsignificant (P > 0 05) interaction terms. Significance was assessed by the Type III tests based on the generalized score statistic, which is asymptotically χ 2 distributed. When a significant interaction of species/year effect with a categorical predictor was detected, the analysis was partitioned for species/years. When a significant interaction with a covariate (date, nest height) was detected, a model with separate slopes was fitted. Significance of terms included in the models derived from the all nests sample was further evaluated by fitting these models to a restricted sample of paired nests. Data collected in 2000 did not reveal significant differences between song thrush and blackbird in nest site characteristics, mean nest survival and the examined patterns (direction and magnitude of differences or trends). Hence, data on these two species were merged in one category referred to as thrushes. Conclusions did not change in any way when blackbird was excluded from the analyses, but it was retained to increase sample size and power of the tests. The year was included as a predictor to account for an effect of the type of artificial eggs used (Experiment 1) and effects of uncontrolled factors that may have changed among years (see Discussion). The study plot was considered as a blocking variable to account for the site-specific effects (e.g. individual predators). It follows from spacing and size of the study plots that each plot in this study probably represented only one pair of corvids, but many individual small mammals, home range of which is much less than plot area. The effect of study plot was not included in the analyses of sparse data sets (yellowhammer and Experiment 2) because of many empty cells in the data table. To simplify analysis, and because an overall concealment differed markedly among species (Fig. 2), I dichotomized nest concealment within each species as relatively good (score 4 in yellowhammer, scores 3 4 in blackcap, scores 2 4 in thrushes) or poor (the remaining scores). Height of nests above ground was partially redundant with factor species (Fig. 2), it was therefore included only in the separate-species models. The effect size observed in this study was evaluated by estimating the 95% confidence interval on differences between the estimated DSR for A and E nests. To get some impression about statistical power of the tests performed, power analysis (Steiger 1999) was conducted

6 429 Nest predation in passerines for the simplest case encountered in this study, the McNemar s test for paired binary responses. Given the number of nests available per species (years pooled), type I error rate = 0 05, power = 0 80, and η (proportion of nests with different outcomes on the two measurement occasions) = 0 4, the minimum detectable difference between A and E nests was estimated at 19% (yellowhammer), 9% (blackcap) and 13% (thrushes) in Experiment 1 and 16% in Experiment 2. Results NEST PREDATORS Jay, probably the only important avian nest predator of the studied species, increased numbers steadily, while abundance of rodents showed more than two-fold fluctuations (Table 1). The relative contribution of rodents to overall predation on experimental nests (Experiment 1) decreased, on average, from yellowhammer (81%) through blackcap (51%) to thrushes (44%). A logistic model relating the type of predator (mammal vs. bird) to species, year and nest concealment (Fig. 3), revealed significant species year (χ 2 4 = 10 1, P = 0 040) and species concealment (χ2 2 = 9 8, P = 0 007) interactions. Because of relatively sparse data sets, the partitioned analyses were based on exact tests of independence in contingency tables (shown are exact P-values). Mammalian share of total nest predation differed: (i) among species in years of high rodent abundance (1998: P < 0 001; 2000: P = 0 012), but not in the year of low rodent abundance (1999: P = 0 708); (ii) among years in yellowhammer (P = 0 003) and blackcap (P = 0 010), but not in thrushes (P = 0 222); (iii) between nest concealment categories in blackcap (P < 0 001), but not in yellowhammer (P = 0 294) and thrushes (P = 0 326). EXPERIMENT 1 A simple comparisons of mean survival rates (not adjusted for effects of nest site and date) between active and experimental nests showed differences of both directions and generally less than c. 15%. A marked exception was yellowhammer in 1998, where active Table 1. Relative abundance of potential nest predators on the study plots. Corvids: frequency of occurrence per 100 plot-visits (528 plot-visits per year); mammals: numbers captured per 100 trap-nights (1440 trap-nights per year) Jay Hooded crow Magpie Total corvids Wood/yellow-necked mouse Bank vole Total small mammals Fig. 3. Proportion of predation events on experimental nests (Experiment 1) ascribed to small mammals, according to year and nest concealment (poor = left bars, good = right bars). Annual means (concealment categories combined) shown by points connected by a dashed line. Sample size refer to the number of predation events (= 100%) that allowed predator discrimination. Relative abundance of rodents indicated as high/low (see Table 1). nests have survived by about 30% better than experimental nests (Fig. 4). Significance tests on these raw data were precluded because of the time lag between active and experimental nests and seasonal trends in nest survival rate. Nest survival measured by DSR and FATE responses was similar within all species-year samples in both active and experimental nests (Fig. 4). The most complex multispecies model (including effects of nest activity, concealment, year, species, date and study plot) was examined for presence of interactions and for sensitivity to the type of response variable and the extent of data included. Using DSR response and all nests sample (n = 1618) revealed significant three way interactions: activity species year (χ 2 4 = 10 1, P = 0 038) and activity concealment year (χ 2 2 = 7 8, P = 0 021). Using DSR response and paired nests sample (n = 822) revealed similar patterns, but only two-way interactions among the above effects were significant: activity species (χ 2 2 = 8 1, P = 0 018), species year (χ 2 4 = 10 9, P = 0 027) and concealment year (χ 2 2 = 15 7, P = 0 001). Using FATE response led to essentially the same conclusions as using DSR, hence the results are not reported. Because of the significant interactions with species effect, I next performed separate species analyses. The model for thrushes (Table 2; Figs 5 7) showed that DSR differed among years and that all other effects were consistent across years. The only significant interaction was detected between effects of activity and

7 430 K. Weidinger Fig. 4. Observed (unadjusted) survival rates (means with 95% CI) of active (hatched bars) and experimental (open bars) nest. Comparison based on: DSR daily survival rates extrapolated to 15-day period; FATE binary proportion of nests that have survived over 15 days since discovery (active nests) or initial exposure date (experimental nests). The nests used for FATE comparison represent a subsample of those used for DSR comparison (see Methods for sample definition). Only paired nests are included number of nests shown above bars applies to both active and experimental nests. Table 2. Experiment 1 thrushes. Generalized linear models (binomial error, logit link) relating the daily survival rate to nest activity (active vs. experimental: repeated factor within individual nests), concealment (poor vs. good), year ( ), date and height above ground (continuous covariates), and effect of study plot (blocking variable). Given for each effect or interaction of effects is P-value associated with the generalized score statistic based on Type III test. A model including only main effects (M) and a model with interactions (I) is shown. The latter was obtained by sequential deleting nonsignificant (P > 0 05) interaction terms from an original model containing all two- and three-way interactions. Significance of terms included in the models derived from the all nests sample was further evaluated by fitting these models to a restricted sample of paired nests (see Methods for sample definition) Effect d.f. All nests (n = 532) Paired nests (n = 208) M I M I Activity (within nest) Concealment Year Date Height Study plot Activity Date Main effects involved in the interaction terms retained in the model. P < 0 05 shown in bold. date: DSR increased with date, which effect was stronger in active ( 2 37 ± 0 49; regression slope ± SE, 0 01; logit scale) than in experimental nests (0 45 ± 0 27). Nevertheless, for any particular date of the breeding season, DSR was on average higher for active than for experimental nests. The positive effect of concealment on DSR, though strong when considered separately, was only marginal and not significant after controlling for date. The model for blackcap revealed significant threeway interactions: activity height year (χ 2 2 = 9 4, P = 0 009) and activity concealment year (χ 2 2 = 7 5, P = 0 024). Next separate year analyses were performed (Table 3; Figs 5 7). DSR tended to decrease with date in 1998 ( 0 78 ± 0 45) but increased in 1999 (1 85 ± 0 44) and 2000 (0 78 ± 0 31). In two of three years a marginally significant interaction was found between effects of nest activity and height. The only significant effect of height revealed by the separate slopes models was that for active nests in 1998 (106 3 ± 43 3). Accordingly, the main effect models did not reveal an overall effect of nest height in any year. A significant interaction between effects of nest activity and concealment was found in two of three years (Table 3; Fig. 7). DSR increased with concealment in active nests (1998: χ 2 1 = 14 8; 1999: χ 2 1 = 22 6, both P < 0 001) but not in experimental nests (P > 0 1 in both years). As a result, active nests showed significantly higher survival than experimental nests within the good-concealment

8 431 Nest predation in passerines Table 3. Experiment 1 blackcap. Generalized linear models (binomial error, logit link) relating the daily survival rate to nest activity (active vs. experimental: repeated factor within individual nests), concealment (poor vs. good), date and height above ground (continuous covariates), and effect of study plot (blocking variable). The separate year models were obtained by partitioning the total blackcap model, that revealed significant (P < 0 05) activity concealment year and activity height year interaction. For other details see Table 2 All nests (n = 885) Paired nests (n = 512) 1998 (n = 236) 1999 (n = 289) 2000 (n = 360) 1998 (n = 112) 1999 (n = 168) 2000 (n = 232) Effect d.f. M I M I M = I M I M I M = I Activity (within nest) Concealment Date Height Study plot Activity Concealment Activity Height Main effects involved in the interaction terms retained in the model. P < 0 05 shown in bold. Fig. 5. Nest survival rates ( least square means with 95% CI) estimated for the main effects of year, concealment and activity (A = active, E = experimental). Based on the separatespecies models containing all two- and three-way interactions, fitted to the total nest sample (Experiment 1) or paired nest sample (Experiment 2). The annual number of nests is shown for each species. Fig. 6. Nest survival rates ( least square means with 95% CI) estimated for two-way interactions: year concealment (poor = hatched, good = solid); year activity (active = hatched, experimental = open); concealment activity (active = hatched, experimental = open). Based on the separate-species models containing all two- and three-way interactions, fitted to the total nest sample. category (1998: χ 2 1 = 12 8, P < 0 001; 1999: χ 2 1 = 5 0, P = 0 026), while nonsignificant (P > 0 3 in both years) effects of opposite direction were found in the poor-concealment category (Figs 6 and 7). Only marginally significant decrease of DSR with concealment, but neither significant interactions, nor overall effect of nest activity was found in one of three years (2000). The model for yellowhammer (Table 4, Figs 5 7) showed that DSR was higher for active than for experimental nests and increased with nest concealment, which effects were consistent across years. The only

9 432 K. Weidinger Table 5. Experiment 2 blackcap. Generalized linear models (binomial error, logit link) relating the daily survival rate (DSR) or nest survival over 15-day period (FATE) to nest activity (active vs. experimental: repeated factor within nestpairs), concealment (poor vs. good), year (1999 vs. 2000), date and height above ground (continuous covariates). For other details see Table 2 DSR (n = 120) FATE (n = 116) Effect M I M I Activity (within nest-pair) Concealment Year Date Height Concealment Date Μain effects involved in interaction terms retained in the model. P < 0 05 shown in bold. d.f. = 1 for all effects. Fig. 7. Nest survival rates ( least square means with 95% CI) estimated for three-way interaction year concealment activity (active = hatched, experimental = open). Based on the separate-species models containing all two- and three-way interactions, fitted to the total nest sample. Table 4. Experiment 1 yellowhammer. Generalized linear models (binomial error, logit link) relating the daily survival rate to nest activity (active vs. experimental: repeated factor within individual nests), concealment (poor vs. good), year ( ), date and height above ground (continuous covariates). For other details see Table 2 Effect significant interaction was detected between effects of date and year: DSR did not change with date in 1998 ( 0 48 ± 0 66) and 1999 (1 71 ± 1 07) but increased in 2000 (2 12 ± 0 82). The main effect model suggests marked annual variation in mean DSR. EXPERIMENT 2 d.f. All nests (n = 201) Paired nests (n = 102) M I M I Activity (within nest) Concealment Year Date Height Year Date Main effects involved in the interaction terms retained in the model. P < 0 05 shown in bold. The fates (success/ predation) over 15-day exposure period of paired active and experimental blackcap nests tended to be positively associated, but this effect was not significant (exact test of independence in a 2 2 table) in either year (1999: n = 46, P = 0 376; 2000: n = 70, P = 0 357). In contrast to Experiment 1, a simple paired comparison of survival rates between active and experimental nests (exact McNemar s test for paired binary responses) was relevant here because of simultaneous exposure of paired nests. No significant effect was found in either year (1999: P > 0 9; 2000: P = 0 110). A generalized linear model (Table 5; Fig. 5) showed consistent results for the two response variables and verified the findings of simple paired comparisons. All main effects were consistent across years and generally matched those found for blackcap in Experiment 1. DSR decreased from 1999 to 2000 but did not differ significantly between active and experimental nests. DSR increased with date in poor-concealment nests (4 75 ± 1 51) but decreased in good-concealment nests ( 3 29 ± 1 26). Discussion METHODOLOGICAL CONSIDERATIONS Artificial nest experiments suffer from many drawbacks (Major & Kendal 1996). While it was hardly possible to check for all of them simultaneously, an attempt was made to eliminate different biases in turn in the separate experiments. The experimental nests were designed to resemble as closely as possible the real models in all aspects but the target variable presence of parents. Experiment 2 came close to this goal, but at the expense of limited sample size and coverage of only one species. In spite of this effort, the natural and experimental nests inevitably differed in many uncontrolled or only statistically controlled traits. Interpretation of the experimental results thus relies on the assumption that the difference between survival of

10 433 Nest predation in passerines natural and experimental nests is mostly, but not necessarily completely, attributable to parental effects. Review of past work on artificial nest methodology made this assumption plausible. It has long been realized that survival of artificial nests does not measure an absolute survival of natural nests, but as relative measure can be compared among experimental treatments (Major & Kendal 1996). The interactions between effect of nest type (i.e. presence of parents) and other factors (nest site, date, year, species) cast some doubt on general validity of this assumption, supporting instead the recent critical view (Butler & Rotella 1998; Wilson et al. 1998; Buler & Hamilton 2000; Davison & Bollinger 2000; Weidinger 2001b). The two types of artificial eggs used in Experiment 1 were associated with different biases. The plastic eggs were of natural size and colour, whereas the quail (= natural) eggs were of unnatural size (all species) and colour (thrushes). Only the latter provided reward (food) to the predator but their shell was probably difficult to break for mice (Marini & Melo 1998; Maier & DeGraaf 2000; Rangen, Clark & Hobson 2000). The effect on nest survival of fluctuating rodent abundance would probably be easier to detect by using natural eggs of appropriate size, as suggested by the larger year effect revealed in Experiment 2 (sparrow eggs). Coloration of artificial eggs may not represent serious limitation of this study, as it was shown to have little effect on the risk of predation in these species (Weidinger 2001c). The effect of year was partially confounded with that of egg type in Experiment 1, but the differences among years can not be simply explained by the different type of eggs used in the last year of the study. Marked year-to-year variation occurred throughout the study and was not parallel across all species (Figs 5 8). Moreover, the annual variation in nest survival was not the focus of this study. Instead, annual variation (of whatever cause) was statistically controlled for when evaluating the effects of primary interest (behaviour, concealment). Perhaps the main impression from this study is the high variability in results. None of the patterns observed was consistent across all species and/ or years, but all results were consistent for the two response variables. Contrary to main effects, conclusions about significance of interactions were sensitive to the extent of data included. The paired nests sample (about half size) was not sufficient to detect significant interactions revealed by the analysis of the all nests sample (Tables 2 4), although all effects detected were of the same direction and of comparable size. The number of nests observed and their spatial distribution seem sufficient to exclude spurious effects typical of small data sets, but not all species year samples were sufficient to detect effects of clear biological significance (e.g. 10% difference in nest survival over the incubation period) with an acceptable power (Fig. 8). Also representativeness of the analysed nest samples differed among species. The wellconcealed ground (vs. less concealed off-ground) nests Fig. 8. Differences (with 95% CI) in daily survival rates between active and experimental nests estimated by the separate-species models (see Figs 5 and 6). Positive values indicate better survival of active nests. The two pooled values shown for yellowhammer refer to combined years 1999 and 2000 (point) or all three years (bar), respectively. in yellowhammer and canopy (vs. shrub) nests in thrushes were under-represented (Weidinger 2001a), whereas blackcap nests, being located in a searchable vegetation, represented the least biased sample. NEST PREDATORS A review of past work (Söderstrom et al. 1998) showed that birds prevailed over nonavian predators for shrub compared to ground nests, and a similar trend has also been predicted from foraging theory (Schmidt 1999). Mice are known to be important predators of passerine nests (Ketterson et al. 1996; Bures 1997; Honza et al. 1998; Maier & DeGraaf 2000; Pietz & Granfors 2000; Schmidt et al. 2001a). This study supported these findings; the overall ratio of mammalian to avian predation decreased from yellowhammer (small eggs, ground/ near ground nests) through blackcap (small, shrub) to thrushes (medium sized, shrub/ subcanopy). Several pieces of evidence suggest a link between annual changes of nest predation and the predators involved. The ratio of mammalian to avian predation was higher in years when rodent abundance peaked; an effect that was clear in yellowhammer, detectable in blackcap, but not found in thrushes. Finally, for each species, the mean annual nest survival was inversely related to abundance of the major nest predator and its share of total predation.

11 434 K. Weidinger The dominant nest predators differed not only among species or broadly defined nest sites (ground/ shrub), but also among nest sites within species (see also Rangen et al. 1999; Dion et al. 2000; Hansson, Bensch & Hasselquist 2000). Visually hunting corvids took mainly the poorly concealed nests, whereas rodents preyed chiefly upon the nests well hidden in vegetation. This could be because the most conspicuous nests were quickly lost to diurnal corvids before nocturnal predation by rodents could take place, while the opposite may happen to well-concealed nests. Alternatively, rodents may avoid foraging in open places to reduce risk of avian predation on themselves (Korpimäki, Koivunen & Hakkarainen 1996; Manson & Stiles 1998; but see Schmidt et al. 2001b), whereas the most concealed nests may not be detectable for corvids. Such a clear pattern was consistently found in blackcap, whose nests showed a high variation in overall concealment and were heavily preyed upon by both rodents and corvids. A similar (though not significant owing to low number of nests) effect was found also in yellowhammer, but only in the year of relatively low rodent abundance, when corvids became important nest predators. In this case the increased predation by corvids did not compensate for reduced predation by rodents and disproportionately affected the poorly concealed nests. No such effects were found in thrushes with their conspicuous nests, where concealment had little effect on nest survival and where apparently all nests were detectable for corvids. NEST CONCEALMENT This study revealed generally positive relationship between concealment and survival of active nests (see also Martin 1992b), which effect disappeared in thrushes after controlling for date. Previous studies reported similar relationships in blackcap (positive: Hoi-Leitner, Nechtelberger & Hoi 1995), song thrush (neutral: Götmark et al. 1995) and blackbird (neutral: Cresswell 1997; positive: Hatchwell, Chamberlain & Perrins 1996). The relationships between nest concealment and survival were parallel for active and experimental nests in thrushes and yellowhammer. In blackcap, however, survival of experimental nests was independent of concealment, indicating that the positive relationship found in active nests was largely due to parental behaviour. Hence, this study concurs with others that the effect of concealment may differ between active and artificial nests, but the patterns reported are conflicting: either the effect was found in active nests but not in artificial ones (this study; Dion et al. 2000), or vice versa (Götmark et al. 1995; Cresswell 1997). PARENTAL BEHAVIOUR The net effect of parental behaviour on nest survival differed among species, being either positive (thrushes, yellowhammer) or neutral (blackcap). The finding that positive effects (nest defence) prevailed over the negative ones (nest disclosure) (Fig. 8) is further supported by the lack of increase in overall predation rates between the egg (low activity) and nestling (high activity) stages in all four species (K. Weidinger, unpublished data; see Martin et al. 2000b for an opposite result). Two other similarly designed experiments on comparable species also have shown higher survival for active than for artificial nests (Cresswell 1997; Wilson et al. 1998). On the within-species level, the effect of behaviour was either independent of nest concealment (thrushes, yellowhammer) or interacted with it (2 of 3 years in blackcap). In the latter case, effect of behaviour was positive for well-concealed nests but negative for poorly concealed nests. The interaction was thus of opposite direction than that reported for thrushes (Götmark et al. 1995; Cresswell 1997), and not consistent with expectations derived from optimality theory of nest defence. According to the latter, parents with exposed nests should exhibit more vigorous nest defence than those with concealed nests (McLean et al. 1986; Hobson et al. 1988; Montgomerie & Weatherhead 1988). The above prediction was not supported for thrushes, but the test was inconclusive for blackcap, where the effect of parental behaviour on survival of differently concealed nests (but not necessarily the intensity of nest defence) was influenced by concealment predator interaction. Finally, an interaction between effects of behaviour and date was found in thrushes; the positive effect of parental behaviour on nest survival tended to increase as the breeding season progressed. This is consistent with hypothesis predicting seasonal increase in intensity of parental nest defence as re-nesting potential declines (Montgomerie & Weatherhead 1988). All species under study have similar re-nesting capacity, but presumably differ in nest defence potential. Accordingly, the above seasonal effect was detectable only in thrushes. This effect cannot be accounted for by changes in nest defence between nesting stages (Montgomerie & Weatherhead 1988; McLean & Rhodes 1991) because the analysis was restricted to the incubation stage, neither by seasonal changes in nest concealment (Hobson et al. 1988), which effect was controlled statistically. ANTIPREDATOR STRATEGIES Thrushes are capable of active nest defence against birds, including direct attacks (own observations; Cresswell 1997; Grim & Honza 2001), and their nests are unlikely to be threatened by mice when attended by parents. On the other hand, the large nests are conspicuous on their own, so the risk of nest disclosure is unlikely to be greatly increased by parental activity. Resulting net effect of parental behaviour on nest survival is thus positive (Table 6) and independent of nest concealment, which itself has little effect (Fig. 1Ba).