What determines prey selection in owls? Roles of prey traits, prey class, environmental variables, and taxonomic specialization

Abstract Ecological theory suggests that prey size should increase with predator size, but this trend may be masked by other factors affecting prey selection, such as environmental constraints or specific prey preferences of predator species. Owls are an ideal case study for exploring how predator body size affects prey selection in the presence of other factors due to the ease of analyzing their diets from owl pellets and their widespread distributions, allowing interspecific comparisons between variable habitats. Here, we analyze various dimensions of prey resource selection among owls, including prey size, taxonomy (i.e., whether or not particular taxa are favored regardless of their size), and prey traits (movement type, social structure, activity pattern, and diet). We collected pellets of five sympatric owl species (Athene noctua, Tyto alba, Asio otus, Strix aluco, and Bubo bubo) from 78 sites across the Mediterranean Levant. Prey intake was compared between sites, with various environmental variables and owl species as predictors of abundance. Despite significant environmental impacts on prey intake, some key patterns emerge among owl species studied. Owls select prey by predator body size: Larger owls tend to feed on wider ranges of prey sizes, leading to higher means. In addition, guild members show both specialization and generalism in terms of prey taxa, sometimes in contrast with the expectations of the predator–prey body size hypothesis. Our results suggest that while predator body size is an important factor in prey selection, taxon specialization by predator species also has considerable impact.

ideal group for addressing this problem is owls (order Strigiformes).
Owls form a guild-defined as a group of species exploiting the same class of resources in a similar way (sensu Root, 1967; see also Simberloff & Dayan, 1991)-that offer two important advantages for studying the role of predator body size in prey selection: The relative ease of collecting pellets and identifying prey remains and the subsequent plethora of literature on this subject (e.g., Dor, 1947;Glue, 1967;Gotta & Pigozzi, 1997;Hardy, 1977;Hayward & Garton, 1988;Herrera & Hiraldo, 1976;Obuch, 2011Obuch, , 2014Romanowski, 1988;Zhao, Song, Liu, & Shao, 2011), allowing comparisons across species, time, and space.
Many studies have addressed prey selection in owl guilds, with conflicting results. While some studies found support for size-based prey selection among owls, others found contradicting patterns. Yalden (1985) studied Tyto alba, Asio otus (Figure 1), Asio flammeus, and Strix aluco in Britain. He found that the largest owl studied (Strix aluco) preyed more than the others on relatively large mammals, but also considerably more on invertebrates. In Greece, prey body size differed greatly between the largest (Bubo bubo) and the smallest (Athene noctua) owls, but not between Tyto alba and the larger Asio otus (Alivizatos, Goutner, & Zogaris, 2005). That study also found significant intraspecific differences in prey size. A comparison of sympatric Athene noctua and Tyto alba in Italy found that the larger owl took larger prey items (Gotta & Pigozzi, 1997). Hardy (1977) concluded that while Tyto alba hunted smaller prey than the larger Asio otus, Asio flammeus, and Strix aluco, little difference in prey size was found between the latter three species, despite their interspecific size differences. A study in Idaho comparing five species of owls found that the smallest (Glaucidium gnoma) and the largest (Bubo virginianus) owls differed in prey size from the three intermediate sized species (Aegolius acadicus, Aegolius funereus, and Otus kennicotii), which in turn differed according to habitat but not prey size (Hayward & Garton, 1988). Another study conducted in Idaho found that Tyto alba preyed on considerably larger prey than the sympatric and similarly sized Asio otus, although the difference they found (42 vs. 31 g) was much greater than in comparable studies elsewhere in North America (Marks & Marti, 1984 and references therein). Thus, while some evidence of the effect of predator size on prey size selection can be discerned, clearly, other selecting factors such as differences in prey taxonomic identity and environmentally driven variation in prey availability must be considered.
Consequently, any study of prey selection by owls must be conducted over spatial scales that incorporate such environmental variation.
To address this, here we studied sympatric owl prey diets using prey traits and environmental factors as predictors. We focused on Israel, the West Bank and the Golan Heights, but also examined the generality of our results in light of the ample owl diet literature from around the world. We analyzed the prey of five owl species, spanning an order of magnitude in body mass (from 110 to 1,550 g). Each owl's prey niche is described here in terms of various taxonomic levels (from phylum to species) as well as prey mass and life history traits. We also explored to what extent prey selection is related to environmental variability within our study area and described the emerging patterns of intraguild prey selection independent of the environment (i.e., via a model of prey selection whose predictors are not only owl species, but also environmental variables). Finally, we discuss the implications of our results for understanding the factors driving prey selection in general.
In this study, we test the hypothesis that prey size (and its amplitude) will increase with predator size within the owl guild, having accounted for various environmental factors affecting prey abundance in the field, prey traits, preferences for particular prey taxa, and other possible sources of bias. To test this idea, we analyze the prey composition of owls across a wide geographical range and F I G U R E 1 A long-eared owl (Asio otus) from Tel Aviv, Israel. This individual belongs to the country's migrant population of long-eared owls. Photo credit: Oded Comay incorporate both owl species and the environment as predictors of prey taxonomic composition. A significant effect of the owl species in a model incorporating environmental effects and prey traits on owl prey composition would indicate that owl species prey on different animals not only as the outcome of their environment (and thus, the available prey) but as an inherent trait. Once the existence of different owl prey niches is established, we continue to describe each owl's diet in terms of taxonomic composition and life history strategies of the prey (including mass, but also other traits such as temporal activity patterns, social structure, movement type) in an attempt to rule out the possibility that larger owls favor larger prey for reasons unrelated to their size. For instance, if larger prey and owl species are also more nocturnal than smaller ones, then any apparent predator-prey size pattern could be an artifact of not accounting for temporal activity patterns. In addition to life history traits, we also account for potential taxonomic bias in owl prey selection that may lead to the appearance of predator-prey body size relationships. For example, shrews (Soricidae) are the smallest mammals in our study system, and if smaller owls favor shrews (for reasons other than their size), an artificial predator-prey size relationship could emerge.

| Study area and sampling design
A total of 3,165 owl pellets and bone assemblages were collected from 78 nesting and roosting sites throughout the Mediterranean zone of Israel, the West Bank and Golan Heights ( Figure 2; Table 1). The study sites were chosen by their proximity to less disturbed areas, where F I G U R E 2 Study area and study sites. Additionally, the contents of 475 owl pellets and prey assemblages from the Steinhardt Museum of Natural History at Tel Aviv University were added to the database. These pellets (collected in the 1980s-2000s) were significant to the analysis, as they greatly expanded the sample size of the largest owl studied, the Eurasian eagle owl Bubo bubo.

| Prey mass
We used mass (g) as our measure of prey body size. Adult body mass of all 38 mammalian and reptilian prey species and genera identified as prey items was retrieved from the Israeli literature (Arbel, 1984;Mendelssohn & Yom-Tov, 1988;Shalmon et al., 1993), as prey size may vary between regions and we needed local measurements. As Strix aluco preyed on more birds than other owls (Figure 4), and many of those were not identified to the genus level or lower, relatively few Strix aluco prey specimens were assigned mass. When the prey taxon could only be identified to the genus level, we used the mean of all the species belonging to this genus that occur in the study area. This averaging was conducted for specimens of Pipisterllus (two specimens, five local species), Crocidura (all 483 specimens, four local species), Rhinolophus (one specimen, six local species), Mus (all 1,814 specimens, two local species), Gerbillus (16 specimens, six local species), Apodemus (76 specimens, two local species), Meriones (11 specimens, three local species), and Rattus (126 specimens, two local species), listed here from the lightest genus to the heaviest. A total of 316 specimens were attributed to taxonomic groups higher than the genus level and were excluded from the prey mass analysis.

| Invertebrate prey
Invertebrate prey individuals were identified to their class (mainly arthropods, but also three gastropods), but were not identified to lower taxonomic levels, nor were they counted, because the preservation state of their remains rendered the identification and counting process extremely tentative and difficult. Nevertheless, as invertebrates were the smallest prey taken, we wanted to test whether smaller owls prey more on invertebrates than larger ones, and hence their quantification was required. Thus, instead of using the minimum number of individuals (MNI; calculated as the greatest number of paired skeletal elements of a given species divided by two), we simply calculated the proportion of pellets containing arthropod remains. For the purpose of this specific analysis, only whole pellets were examined and bone assemblages were excluded.

| Vertebrate prey traits
In order to assess the contribution of prey size in isolation from unrelated traits affecting prey selection, we included prey traits in the owl prey model. For instance, if some owls specialize on volant prey (bats, birds) and these taxa have smaller mass for a given size, then failing to account for volancy could lead to spurious association between predator and prey size. We used the following prey species traits as predictors of abundance in owl diets in interaction with the other predictors stated above: mass (g), diet (granivore, insectivore, grazer), movement type (fly, jump, climb, and burrow), social structure (solitary, social), and diel activity pattern (diurnal, nocturnal). All traits were inferred from the species descriptions in Israel (Mendelssohn & Yom-Tov, 1988;Paz, 1986;Shalmon et al., 1993). Table 2 details the traits assigned to each of the main prey taxa of owls in this study.
T A B L E 1 Mammalian and reptilian prey mass (g) per owl species, along with owl body mass data (Paz, 1986). Sample size (n) is cranial MNI of all mammalian and reptilian prey individuals that were assigned a mass value (see text). Owl species are ordered by ascending mean prey mass

| Environmental analysis
To account for the potential effects of the environment on prey availability, we quantified the environmental characteristics of each sampling site using the following protocol.

| Vicinity
We extracted environmental data (see below) from a four km radius area around each nesting and roosting site (referred to here as its "vicinity"), using ArcGIS 10.4 (ESRI 2015). This radius was chosen as a rule of thumb, as the hunting ranges of owls in Israel were studied both with radio telemetry and GPS transmitters, with the latter suggesting larger hunting ranges than the former (Charter, 2016;Motro, 2011). While telemetric studies suggested a hunting radius of ~500 m for Tyto alba (Motro, 2011), recent data from GPS-tags indicate ranges of 6-8 km and up to 14 km for the same species. Given these results, the range of four km was used a compromise, not only for Tyto alba but for all owl species (for which no equivalent data were available). Spatial data were transformed to raster grids with cell size of 100 × 100 m, as detailed below. Numerical values were attributed to cells, and each owl site was attributed a Vegetation Index (see below) calculated based on all cells in its vicinity, regardless of their relative distance.

| Geographical analysis
The following layers were downloaded from Open Spaces Portal habitats, including park-forest/maquis and low-density shrubland, high-density forest and maquis, high-density maquis, medium-density maquis/medium-high-density shrubland, grassland/sparse shrubland and semi-arid grassland/sparse vegetation/exposed soil. These separate polygonal layers were converted to raster layers with 100-m edge length cell size and combined to create a layer of land cover, from which we calculated the Vegetation Index (see below).

| Vegetation Index
For the purpose of this analysis, we devised a Vegetation Index to represent the key stages of botanical succession in the Mediterranean region of Israel: dwarf-shrub steppe, garrigue, maquis, or forest, from the sparsest to the thickest, in successional order (Waisel, 1991). Each undeveloped land grid cell in the study area was assigned a value of 1-3, based on these succession

| Climate
Climatic variables were downloaded from worldclime.org (Hijmans, Cameron, Parra, Jones, & Jarvis, 2005 attributed to each site: mean annual precipitation (mm), mean annual temperature (°C; referred hereafter as "temperature"), mean temperature of the coldest quarter (°C; referred to hereafter as "winter temperature"), and mean temperature of the warmest quarter (°C; referred to hereafter as "summer temperature").

| Statistical analysis
We used PAST version 3.12 (Hammer, Harper, & Ryan, 2001) and the R language (R Core Team 2016) for conducting statistical tests.
The procedure detailed below follows the recommendations for model-based thinking in community ecology (Warton, Foster, De'ath, Stoklosa, & Dunstan, 2015) and the mvabund package documentation. We used the mvabund (Wang, Naumann, Wright, & Warton, 2012, 2017 package for R to create a model of prey abundance in owl diets by owl species and the environmental variables detailed above. MNI counts of prey taxa occurring more than four times were used as the response variables, assuming a negative binomial distribution for model errors. This assumption was verified via visual inspection of the Dunn-Smyth residuals vs. the linear predictor value plot. The unequal sampling effort (i.e., unequal prey MNI) between sites was taken into account by adding the total MNI of all prey species at each site as an offset term to the model (as recommended by Warton, Foster et al., 2015). This resulted in modeling the effects on relative rather than absolute abundances. We tested the statistical significance of the predictors and removed nonsignificant predictors from further analysis, starting with the full model and using a stepwise backward deletion of predictors, until only significant (α = 0.05) predictors remained. Next, we used the LASSO algorithm (Osborne, Presnell, & Turlach, 2000) to penalize the coefficients of the significant predictors found in the stepwise backward deletion, as recommended by the mvabund package documentation. In sum, the LASSO algorithm adjusts predictors' coefficients in accordance with their correlation to the response variable (in our case, each species' abundance in owl diets). Predictors with little correlation to the response could be reduced to zero, thus removing them from the model altogether.

| Prey traits as predictors of relative abundance in owl diets
"Fourth corner" models in community ecology are models that examine the effects of species traits on their abundance (Warton, Foster et al., 2015;Warton, Blanchet et al., 2015). While standard community ecology models study predictor-by-species effects (e.g., how are micromammal species affected by vegetation density), "fourth corner" models focus on predictor-by-trait effects (e.g., how does species size affect its abundance along a temperature gradient).
Categorical traits were converted to binary (true/false) variables for the purpose of this analysis, which allowed attributing several (for instance) movement types to each taxon, when appropriate. We used the mvabund R package (Wang et al., 2012(Wang et al., , 2017 for analysis. The fourth corner coefficients were plotted using the lattice R package (Deepayan, 2015).
Taxonomy-based prey selection was analyzed by chi-squared tests using adjusted residuals for post hoc analysis (Sharpe, 2015).

| RESULTS
A full list of the vertebrate prey taxa (by MNI) found in each site is available at Dryad.org, https://doi.org/10.5061/dryad.9m84np6.

| Predictors of owl prey
Owl species, temperature, and the Vegetation Index significantly contributed to prey selection of owls, indicating both an environmental impact and an inherent prey selection pattern among owls, independent of external factors (Table 3). The LASSO-penalized coefficients of each significant predictor on prey abundance are depicted in Figure 3 (by prey taxon) and in Figure 4 (by prey traits and taxonomy-a "fourth corner" model). Prey mass differed considerably between owl species but not with environmental gradients (Figure 4). Some prey taxa (Apodemus spp. and Acomys dimidiatus) varied strongly in their abundance between owls of different habitats, but only marginally between owl species; most prey species, however, demonstrated the opposite pattern (Figure 3).

| Vertebrate prey mass
Prey mass differed between all owl species except between Strix aluco and Bubo bubo (Table 4). Prey mass mean and standard deviation increased with owl mass on the log 10 scale (for prey mass mean: Pearson r: R 2 = .92619, p-value = .023805; for prey mass standard deviation: Pearson r: R 2 = .89772, p-value = .03866). Figure 5 depicts the proportion of pellets containing arthropod remains of the total analyzed pellets of each owl species. All owl species preyed on arthropods, to significantly varying degrees (χ 2 test, p < 10 −4 ). Cells whose adjusted residuals were greater than 3 in absolute value were considered as significantly different from the null hypothesis of equal arthropod consumption (Sharpe, 2015).

| DISCUSSION
We found a strong relationship between predator body size and prey size and amplitude. These results suggest a significant pattern of prey selection between sympatric owl guild members, despite considerable dietary overlap and environmental effects on prey availability. Prey mass had the greatest effect on predation by owls, more than any other prey trait or environmental factor (Figure 4): prey size mean and variance increased with predator size. Additionally, owls specialized F I G U R E 3 LASSO-penalized coefficients of prey taxa vs. owl species and the significant environmental predictors (fourth corner model). Colors indicate the effect size from negative (red) to positive (blue) coefficients. Absolute values of coefficients are not comparable between predictors due to differences in units. See text F I G U R E 4 LASSO-penalized coefficients of prey traits and taxonomy vs. owl species and the significant environmental predictors (fourth corner model). All traits but mass (g) are binary. Colors indicate the effect size from negative (red) to positive (blue) coefficients. Absolute values of coefficients are not comparable between predictors due to differences in units. See text on specific taxa to a varying degree. Some taxon specializations contradicted expectations of the predator-prey size hypothesis: Tyto alba (rather than the smaller Athene noctua) specialized on the smallest vertebrate prey (shrews, Soricidae; Figure 4) and arthropod consumption did not correlate well with owl size ( Figure 5).
Owl body size plays a major role in prey selection, as prey mass had large coefficients in our prey-by-trait fourth corner model (Figure 4).
Both the mean and the variance of prey mass increased with predator size (Table 1), as expected by theory (Cohen et al., 1993). Notably, all owl species, regardless of size, capture small prey items (arthropods and vertebrates up to 28 g; Table 1, Figure 5). That is, large owls do not simply give up small prey items in favor of large ones, but their size allows them to take large prey items which are apparently unobtainable for smaller owls, in addition to the smaller ones (thus the difference in prey mass variance).
These results accord with those of previous works. For instance, Hardy (1977) found that Strix aluco (the largest owl he studied) preyed more often on large animals than Asio otus and Tyto alba. Capizzi and Luiselli (1998) found that among sympatric Tyto alba, Asio otus, and However, other works also found contradicting results. Leader (2003) found that Tyto alba in the Negev desert feed on larger prey than the sympatric (larger) Asio otus. Another interesting point is the two smallest vertebrate prey categories (0-8 g), composed of shrews and bats: Tyto alba preys on these taxa more than the smaller owl, Athene noctua ( Figure 4). Thus, we suggest that other factors besides prey size effect prey selection, as discussed below.
While predator body size is significant for understanding prey selection among owls, it cannot account for taxon-specific preferences.
While Athene noctua's preference for arthropods ( Figure 5 Past works showed similar patterns to those reported here; that is predator size is important in owl prey selection yet this trend is limited by taxon specializations. Several studies report Strix aluco preying on more invertebrates than smaller sympatric owls (Hardy, 1977;Herrera & Hiraldo, 1976;Obuch, 2011;Siracusa, Sarà, La Mantia, & Cairone, 1996;Yalden, 1985). The relatively large intake of birds by Asio otus and Strix aluco was also reported elsewhere (Hardy, 1977 and references therein;Yalden, 1985;Bertolino, Chiberti, & Perrone, 2001;Kiat et al., 2008;Birrer, 2009; but see Davorin, 2009 andObuch, 2011). Tyto alba's preference of shrew prey was also found in previous studies comparing sympatric owls (Capizzi & Luiselli, 1998;Georgiev, 2005;Kitowski, 2013). Therefore, our results-the overbearing importance of body size and the taxonomic specializations that contrast its general trend-likely represent a general pattern in owls (and potentially in other guilds as well) and should not be regarded as unique to our study area.
While our conclusion that whole genera of predators' may specialize on prey taxa in contrast of the general predator-prey size trend relies on data from owls' diets, it might apply in other guilds as well. Despite the long scientific interest in predator-prey size relations (e.g., Brose et al., 2006;Cohen et al., 1993;Hespenheide, 1973;Klompmaker et al., 2017;Naisbit et al., 2011;Nakazawa, 2017;Schoener et al., 1986), to the best our knowledge, this issue was not explicitly pointed out before. The relative ease of studying owl diets (i.e., by collecting pellets with mostly intact skeletal remains) allowed us not only to plan a geographically extensive study of multiple sympatric guild members, but also to corroborate our results using the rich owl diet literature. Given a similar research or a meta-analysis focusing on different predator guilds, other deviations from the predator-prey size ratios trend may emerge.
Paleontological studies of intraguild predator-prey size ratios could track the evolutionary histories and mechanisms of these specializations (e.g., when and in what environmental context did the lineage of modern-day Tyto first started specializing on small mammals more than Strigid owls?).
In sum, prey selection by the owl species studied here, based on both predator size and taxonomic specialization, overrides variation due to species life history or environmental differences across space.
Disentangling the ecological or evolutionary causes of prey selection beyond body size thus warrants further research, which could shed more light on the fascinating subject of prey selection.

ACKNOWLEDGMENTS
We wish to thank the naturalists who aided in locating the owls' roosts and nests that we studied, who are too numerous to individually name herein. We are also grateful to the anonymous reviewers and the associate editor, whose comments helped us to improve the manuscript.

CONFLICT OF INTEREST
None declared.

AUTHOR CONTRIBUTIONS
OC and TD conceived the ideas and designed the methodology. OC collected and analyzed the data. OC and TD led the writing of the manuscript.
All authors contributed critically to the drafts and gave final approval for publication.