Population parameters.

From 2002 to 2012 we studied an eagle owl population located in the Sierra Norte of Seville (37°30′N, 06°03′W, SW Spain; details in [35]). We located 56 nest sites, where a total of 132 breeding attempts were monitored. Laying dates ranged from December 24 to April 8, and the mean (± SD) number of fledglings was 2.18±1.03 per brood (range: 1–4 chicks). Mortality rates were calculated on the basis of 130 radio-tagged individuals (date of first animal tagged: 01/03/03; date of last animal tagged: 22/04/09): mortality of breeders (35.29%; 8 males and 4 females) and dispersers (36.45%; 18 males, 11 females and 4 individuals of sex unknown) were similar. Given the scope of this study, we described the population by means of: (i) two measures of productivity, i.e. the mean and the coefficient of variation (CV) of young fledged per breeding pair [34]; (ii) an estimate of the quality of breeding sites via census methods of the main eagle owl prey species in the study area, the rabbit Oryctolagus cuniculus (mean number of latrines per km of transect ± SE = 20.6±12.4 km⁻¹; range: 7.7–46.0 km⁻¹); (iii) an analysis of the diet through the collection of a minimum of 100 pellets (and as much of prey remains as possible) for each nesting site (mean biomass percentage of rabbit in the diet ± SD = 62.0±19.1%, range = 16–94%; for more details see [32]); and (iv) landscape characteristics by intersecting a digital layer representing the boundaries of the owls’ home ranges with a map of landcover elements (scale 1∶25,000). Following the studies of Aebischer et al. [36], and with the aim of selecting only those habitat types that were most relevant for eagle owls [14], [32], we (a) first classified the landscape into 10 landcover types: urban areas, water bodies, forests, dense scrublands with trees, sparse scrub with trees, herbaceous vegetation with trees, scrublands, low vegetation, woody crops and herbaceous plants. Additionally, we used edge density (i.e., the total length of the patch edge per unit area within each landscape; [37]) as a proxy for the effect of habitat heterogeneity [38]–[40], which has been shown to be important in determining breeders’ movements and rhythms of activity [41]. Then (b) we performed a compositional analysis to test owl habitat selection (for more details, see [32]). We used ArcView 3.2 (Geographic Information System, GIS) and its extension Patch Analyst [37] for the analyses of landscape characteristics.

Finally, we analyzed the genetic structure of the population by using a set of loci developed for eagle owls, the spotted owl (Strix occidentalis lucida) and the lanyu scops owl (Otus elegans botelensis). We extracted DNA, following a Hotshot protocol [42], from blood samples (2 mL, taken from the brachial vein by V.P., who was initially accompanied and trained by an expert veterinary; date of first animal sampled for DNA: 01/03/03; date last animal sampled for DNA: 22/04/09) of 22 adult individuals in our study population. Blood samples were collected under the Junta de Andalucía–Consejería de Medio Ambiente permit nos. SCFFSAFR/ GGG RS-260 / 02 and SCFFS-AFR/CMM RS-1904 / 02. Based on polymorphism of the loci, we finally selected the following 10 loci: Oe3-7, Oe045, Oe054, Oe128, Oe2-57 (GenBank accession no. AY312418, AY312422, AY312425, AY312427, AY312420, respectively) [43]; Bb42, Bb126, Bb131 (GenBank accession no. AF32093, AF32097, AF32098, respectively) [44]; 15A6 and 13D8 [45]. Fluorescently-labeled PCR products were amplified in a reaction with a final volume of 20 µl, which included 50–80 ng of DNA, 67 mM of Tris-HCl, 16 mM of (NH₄)₂ SO4, 2 mM of MgCl₂, 0.2 mM of dNTPs, 0.1 ng/µl of bovine serum albumin (BSA, Biomol), 0.5 µM of reverse and fluorescently-labeled universal M13 primer, 0.041 µM of forward primer and 0.5 units of Taq polymerase (BIOTAQ, Biomol). Reaction conditions were as follows: an initial denaturation step of 2 min at 94°C, 30 s at 55°C annealing temperature decreasing 1°C/cycle for 15 cycles, and 30 s at 72°C, followed by 27 additional cycles with an annealing temperature of 40°C and a final step of 5 min at 72°C. Products were analyzed on an ABI PRISM® 3100 DNA Genetic Analyser (Applied Biosystems) and alleles were scored with GeneMapper 4.0 (Applied Biosystems, Inc.).

Individual parameters.

We trapped and radio-tagged 34 breeding individuals (24 males and 10 females) from 24 nests, as well as 96 juveniles (54 males and 42 females) from 21 different nest sites (for more details about the radio-tracking procedure see [14], [32]). Each individual was fitted with a 30 g radio-transmitter using a Teflon ribbon backpack harness (Biotrack Ltd, Wareham, BH20 5AJ, Dorset, UK; www.biotrack.co.uk). The mass of the backpack was less than 3% of the mass of the smallest adult male (1550 g; mean ± SE = 1667±105 g) in our population. This telemetry study allowed us to collect detailed information at the individual level concerning both the dispersal process [14], [31] and the breeders’ home ranging behavior [32]. Radiotracking data were analyzed under the framework of animal movement analyses (see [14], [31]–[32] for more details). We found dispersal distances to be very short in most cases, ranging from 1.5 to 34.3 km (mean ± SD = 6.0±4.2 km). In fact, 35% of the individuals which dispersed established a stable range close to their natal population. In general, breeders showed high site fidelity; their home range behavior being simultaneously affected by different internal and external factors acting at different spatio-temporal scales. However, we also recorded nine cases of breeding dispersal (5 males and 4 females), as well as ten cases of replacement of a breeder (5 males and 5 females).

The individual monitoring of this nocturnal species is extremely demanding, especially considering the intrinsic difficulties and relatively low success rates of breeder trapping. Given that two previous studies [21], [30] showed that eagle owl vocalizations are individually distinctive, we were expecting to be able to recognize, over the course of a year, each territory owner within our population by the characteristics of its call sonograms. This procedure would have also favored the use of a technique less intrusive than breeder trapping, i.e. the individual discrimination by territorial and sexual call recording of breeders. Thus, from 2002 to 2006 we recorded 15 males and 10 females at 15 breeding sites using a Sony digital audiotape recorder (TCD-D100) and a Sennheiser directional microphone (condenser microphone ME 67+ powering module K6). Some individuals were recorded over different years, namely six males from the 15 breeding sites that were also captured and radio-tagged. The characteristics of the territorial call of eagle owls are well described in [21].

We strictly followed a rigorous recording protocol. (1) Recordings were always made at sunset for birds positioned on known call posts in close proximity to their nests [46] during calm days (without wind or rain), and the observer was never too far from the birds (less than 100 m). Recordings were made during the pre-breeding period (i.e. September–December in our study area), when males and females are in general more vocally active [28], [47]. (2) Recordings were performed by the same two observers (V.P. and M.D.). (3) We were helped by an expert (P.L.; see acknowledgement) who has a great knowledge of bird recordings and sound analysis. Therefore, we are confident that we carried out a well-designed recording of the breeders in our population, where recorded information was combined with data from radio-tagged individuals, when possible.

We extracted the acoustic features of the 478 calls that were recorded on audiotape by performing a spectrographic analysis. For this analyses, we used Avisoft SASLab Pro software (Version 3.91; [48]), performing a Fast Fourier Transform (sampling frequency 11,025 Hz, FFT length 512, time resolution 8.9 ms, bandwidth of frequency resolution 43 Hz, Window Function: Bartlett). For both male and female calls, four temporal variables were measured (Fig. 1A): total duration of the bout (Dtot), duration of the portion of increasing (D1), stable (D2) and decreasing (D3) frequency. Four frequency variables were also measured (Fig. 1A): minimum and maximum frequency (Fmin and Fmax), dominant frequency (DOM; i.e. the frequency with the highest energy) and the range of frequencies (range = Fmax-Fmin) in a bout.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Figure 1. Graphical representation of the spectrograms of the hooting of the eagle owl (A) of male (below) and female (above) calls of the eagle owl.

Parameters measured to characterize the call are: (a) parameters in the time domain: D1, D2 and D3; (b) parameters in the frequency domain: Fmax, Fmin and FDOM (see text for explanations). (B) Four spectrograms of the territorial calls uttered by different eagle owl males in south-western Spain. The high similarity among eagle owl calls is apparent even by visual inspection. Owing to the considerable overlap observed, individuals could not be discriminated on the basis of the information concerning their vocalization.

https://doi.org/10.1371/journal.pone.0077557.g001

Population structure analyses.

Following Penteriani et al. [34], we analyzed the spatial structure of the population (i.e. population heterogeneity) using several procedures. First, to test the effect of breeding site quality on overall population fecundity, we eliminated the year effect on productivity. Owing to the existing annual variations, we controlled for the year effect by subtracting annual mean from the row data. For the number of fledglings, negative values indicate a poorer breeding performance than average, whereas positive values indicate a better one. Relative productivity was analyzed by a general mixed model, with the breeding site as a random factor to correct for pseudoreplication. Second, we tested a variable designated % of contributing pairs, which allowed us to detect intrinsic variability of the population through the evaluation of the distribution of fecundity among nesting territories. Our assumption was that a heterogeneous population structure, characterized by differences in quality among breeding sites, should lead to low variance in production of young during good years and high variance during poor years (i.e. a few pairs will produce the majority of the fledglings). To accomplish this, we considered the percentage of breeding pairs producing at least 50% of the annual fledged young. We calculated this parameter by summing the number of fledged young (starting from the pairs with highest productivity) necessary to attain 50% of the annual young production. Finally, to detect whether landscape structure, diet (rabbit biomass) and resource abundance explain differences in mean reproductive output and its annual variance within the population, we ran two multiple regression models using (a) mean number of fledglings and (b) CV as dependent variables. We used the open-source software R, version 2.10.1 [49] to build the linear models. We always explored the residuals for: (i) normality, (ii) homogeneity of variance, and (iii) spatial independence. For the latter, we used the package Gstat [50] to verify the independence of the data by plotting the residuals versus their spatial coordinates; the resulting bubble plot did not show any spatial pattern. All tests are two-tailed, statistical significance was set at α <0.05, and ± deviations for means are either SD or SE, depending on whether the factor of interest was variability or precision, respectively.

Population genetic analyses.

Genetic diversity, i.e. observed (H_O) and expected heterozygosity (H_E), mean number of alleles per locus (N_a), and the population inbreeding coefficient (F_IS), was estimated for each locus for the population using FSTAT. Significance of F_IS was determined by bootstrapping over loci to obtain a 95% confidence interval based on 10,000 replications. The same program was used to perform tests for Hardy-Weinberg equilibrium (HWE) using 10,000 permutations of alleles among individuals. Sequential Bonferroni corrections were applied to correct for multiple simultaneous comparisons. To analyze the genetic structure of the population we used two approaches. First, we used Structure v.2.2 software [51] to assess the number of different genetic clusters (k) in the population. Simulations were run with a burn-in period of 20,000 followed by an additional 2×10⁶ MCMC steps. The number of populations was varied from1 to 5, and for each k 20 replicates were run under an admixture model with correlated gene frequencies. We assessed the support for k populations based on visual inspection of the plot of the algorithm of the posterior density (lnP (D)) as a function of k, and Δk, following [52]. Convergence was assessed by checking that the posterior density and the log-likelihood levels reached a plateau before the end of the MCMC runs. Second, by using GenAlEx and following the method proposed by [53], we investigated the genetic spatial autocorrelation at the individual level within this population (SA). This analysis allowed us to determine whether related individuals were clustered in space, which might suggest that dispersal is limited by distance, even within the same population. We used a pairwise geographical distance between individuals calculated as the linear distance separating them based on their breeding location, and a pairwise genotypic distance. We estimated the average genetic similarity between pairs of individuals in specific distance classes (thresholds at 500, 1000, 1500 and 2000 m) through the autocorrelation coefficient (r) obtained from 9,999 permutations.

Individual acoustic analyses.

To identify the presence of sound information concerning individuality [30], [20], [21], we first performed a nonparametric analyses of variance (Kruskal-Wallis ANOVA) to identify the characteristics of calls for which inter-individual variation was higher than intra-individual variation. We assumed that a much greater inter-individual value indicates a factor which is better able to describe individual variation. As a measure of call individuality [30] we also estimated for each variable the ratio between the inter-individual coefficient of variation (CVb) and the individual coefficient of variation (CVi). Once these variables were identified, we performed a discriminant function analysis (DFA) on standardized data [24], [25] to test for the discriminant power of the acoustic features. For this analysis, we used only the calls of the six recorded and radio-tagged males whose identity was known (as in [20]). For classification purposes, we finally applied similarity techniques to define threshold values of similarity within individuals, i.e. calculating the Euclidean distances between the acoustic features of pairs of birds. Following previous studies [20], [25], when a new recorded bird fell outside the intra-individual threshold for all marked birds, it was classified as a new individual. As those birds whose identity was known were all male, we performed classification analyses for this sex only. Discriminant function analysis (DFA) and Euclidean distance estimations were performed with SPSS (version 20). Finally, we used regression analysis to explore whether acoustic similarity was higher for closer neighbors. In fact, because neighboring birds can form local communication networks [54] and match their songs to those of their neighbors [16], a change of vocalization structures and acoustic matching over distance was expected to occur.