|Habitat-mediated predation risk and decision making of small birds at forest edges
Alejandro Rodr´ıguez, Henrik Andre´n and Gunnar Jansson
We studied the effect of four components of predation risk, namely predation pressure, spatial variation in woody cover, visibility, and flock size, on the behaviour of four species of the Parus guild (crested tit, P. cristatus, willow tit, P. montanus, coal tit, P. ater, and goldcrest, Regulus regulus ) at the edge of patches of mature boreal forest. Birds used the exposed side of the edge (matrix) mainly during periods with low levels of predation pressure by pygmy owls (Glaucidium passerinum ). Some species avoided edges under low light conditions. Birds in large groups were more prone to cross the edge, whereas group cohesion increased in risky situations, especially in the most vulnerable species (coal tit and goldcrest). The effects of these three components of predation risk were not general in that only some species responded to them, and in that intra-specific responses were not always consistent. In contrast, all behaviours examined in all four bird species (occurrence, matrix cross- ing, edge crossing, and group cohesion) appeared to be strongly affected by habitat- mediated predation risk. Mature boreal forests appeared to be qualitatively superior to any type of matrix in terms of protection against predators. Birds generally avoided open matrix, and seemed to move towards and across more developed stages in forest regeneration according to the ratio between food intake and predation risk attributable to a given matrix type. Open matrix (farmland and clearcuts) and very young plantations strongly restricted the rate of bird movement between old forest patches. Our results contrast with the widespread thought that birds have a great potential to use fragmented landscapes in a fine-grained manner. These limited movements across the landscape during winter might have important repercussions on the occupation of some forest patches by tits, their subsequent reproduction, and in turn their population dynamics.
A. Rodr´ıguez, H. Andre´n and G. Jansson, Grimso¨ Wildlife Research Station, Dept of Conservation Biology, Swedish Univ. of Agricultural Sciences, SE -730 91 Riddarhyt - tan, Sweden (present address of AR: Dept of Applied Biology, Estacio´ n Biolo´ gica de Don˜ ana, CSIC, Avda. Mar´ıa Luisa s /n, E -41013 Sevilla, Spain [firstname.lastname@example.org]).
In fragmented landscapes, ecological processes that take place at the individual level (e.g. foraging or dispersal) often depend on the organisms’ ability to move between adjacent patches of different quality (Stamps et al. 1987, Dunning et al. 1992, Johnson et al.
1992). Animal movements are the outcome of activity patterns and space use which in turn are largely deter- mined by the perceived risk of predation (Lima 1998). Predation risk changes greatly with place and time due to variation in predator habits and predation pressure
(Korpima¨ ki et al. 1991, 1994, Bouskila 1995, Go¨ tmark and Post 1996), habitat structure and cover availability (Lima 1990, Hughes and Ward 1993, Moreno et al.
1996), predator detectability (Lima and Dill 1990), and behavioural responses of prey as grouping or vigilance (Lendrem 1983, Elgar 1989, Richardson 1994). Since all these components of predation risk can be viewed as patch-specific, the effect of predation risk on animal movement should be best detected at marked disconti- nuities in the landscape, that is, at patch boundaries.
In this paper we study the effect of predation risk on the behaviour of bird species of the Parus guild (cf. Suhonen et al. 1992, 1993), as a mechanism determining bird movement towards and across the edge of mature forest fragments. We restrict our observations to win- ter, when predation is a major component of small passerine mortality in the Scandinavian boreal forest (Ekman et al. 1981, Jansson et al. 1981). As survival during the non-breeding season is a good predictor of individual fitness (Caraco 1979), it is expected that birds behave in a way that minimizes predation risk in winter. The four components of predation risk listed above are hypothesized to affect the behaviour of birds as follows.
In the boreal forests of Europe, sparrowhawks (Ac - cipiter nisus ) and, especially, pygmy owls (Glaucidium passerinum ) are the main predators of the Parus guild in winter (Solheim 1984, Ekman 1986, Suhonen 1993a, Kullberg 1995, Go¨ tmark and Post 1996, Krams 1998). Suhonen (1993a, b) found that predation pressure by pygmy owls on tits increases remarkably in winters with very low vole (Clethrionomys glareolus and Microtus agrestis ) density, i.e. low phase of the population cycle. Thus, owl prey switch might make vole-crash years more dangerous to small birds, thereby increasing the weight of predation risk in their decision making at forest edges. In our study area, during the winter of
1990 – 1991 vole populations were at the highest popula- tion peak in the nine-year period studied by Helldin (1999), whereas in the winter of 1991 – 1992 vole abun- dance was one of the lowest records during the same period. Assuming that pygmy owls had looked more actively for birds in the winter of low vole abundance and that owl density did not decrease during that year, predation risk should have been higher in the winter of
1992. We predict that birds will approach forest edges less often during the winter with depressed vole populations.
Habitat-mediated predation risk often operates through the amount of available cover provided by vegetation which can be used as shelter against preda- tor attacks (Lima and Dill 1990). Parids and parid-like species usually exhibit escape tactics based on flying into woody cover (Ekman 1987, Lima 1993). Cover provided by different habitat (patches in the landscape) and microhabitat types (trees within a patch and place within a tree) differs in density, distribution, and struc- ture, all of which reflects exposure to predators (Opdam
1978, Ekman 1986, Suhonen 1993a, Go¨ tmark and Post
1996). Due to their relatively closed branch structure and foliage density, spruce provides more shelter than pine (Ekman 1987), and conifers provide more shelter than the open-branched and deprived of leaves birch trees (Suhonen 1993a). To minimize predation, birds at the boundary between two patches should opt for the habitat having fewer open spaces, or the one containing more tree species with a relatively compact structure.
Therefore, we predict that the frequency at which birds within a patch of mature forest approach and cross its edges will increase with the amount of woody cover in the adjoining patch. We also expect birds to travel less across patches poor in woody cover. Besides, the differ- ence in tree height across old-forest edges is a direct measure of predation risk. Perching or foraging height relates inversely to predation risk (Ekman 1986, Kull- berg 1995, Go¨ tmark and Post 1996). Pygmy owls are attracted by prey contact calls and attack small birds from above, diving from a nearby tree (Ekman 1986, Kullberg 1995). Prey can escape predation by perching or foraging higher up in the tree (Ekman 1986, Hogstad
1988a, Suhonen 1993a, Krams 1998). This behavioural option disappears as the contrast in tree height in- creases between edge sides, making birds perched in small trees more vulnerable to owl attacks from higher trees in adjacent old forest. Therefore, we predict that bird occurrence and edge crossing rates will be inversely related to the difference in tree height between edge sides.
Regarding predator detectability, Lahti et al. (1997) suggest that at dawn and dusk predation risk may be greater than during the central hours of the daylight period, since the vision of diurnal birds is not well adapted to dimness. Similarly, light levels are rather low in winter time at high latitudes, especially with cloudy or misty weather, and under these conditions predators might be difficult to spot. Hence, one might expect birds to be less prone to use risky boundaries or the more risky side of a boundary during low light periods. Pygmy owls, however, have symmetrical ears and cannot localize prey by sound alone, so that they have to combine vision and hearing to localize prey, or use only vision (Sonerud 1986). Thus, hunting success of pygmy owls may decrease with dimness too. In spite of this, small birds might still perceive a high predation risk under low light conditions because they are uncer- tain that a predator has not detected them. Bouskila and Blumstein (1992) suggest that prey would increase survival chances by overestimating predation risk, i.e. by using decision rules leading to behave as if a preda- tor was actually present when there is just perception of vulnerability. Under this hypothesis, we predict that birds will tend to avoid old-forest edges near twilights and during cloudy or misty days.
Finally, we analyse the relationship between bird aggregation and predation risk. Tits and other small forest birds spend the winter in single- or mixed-species groups (Ekman 1979, Hogstad 1988a, Suhonen 1993a), the larger the group, the more likely to detect predators and avoid predation (Elgar 1989). However, large groups also imply increased competition between group members for scarce, non-renewable winter food and other benefits of holding territories (Ekman 1979). Since optimal group size is achieved through the addi- tion of individual decisions (Ekman 1989), and spatio-
temporal changes in individual trade-offs make group size dynamic (Ekman 1979, Hogstad 1988b, Lens and Dhondt 1992), we expect that flock size at risky boundaries between adjacent patches will be larger and more cohesive than at less contrasted boundaries in terms of cover.
The study was conducted during the winters (January to March) of 1991 and 1992 around Skinnskatteberg and in the contiguous Grimso¨ Wildlife Research Area, Bergslagen region, south-central Sweden (59°45’N,
15°30’E), totalling 840 km2. In this area, boreal forest
dominates the landscape (roughly > 70%), which also contains bogs, lakes, and open farmland. The propor- tion of farmland increases towards the south of the study area. The most common tree species in the forest are Norway spruce (Picea abies ) and Scots pine (Pinus sylvestris ). Broad-leaved tree species (chiefly birch, Be - tula pendula and B. pubescens, and aspen, Populus tremula ) are scarce. Forests are subjected to modern forestry, based on clearcutting, planting and thinning, which generates a mosaic of even-aged monoculture stands of different tree sizes and densities (Esseen et al.
1997). Stand size ranged between 1 and 25 ha (x¯ = 10.2 ha, n = 462 stands within the Grimso¨ Wildlife Research Area). Therefore, maximum diameters of forest stands, clearcuts, and also many agricultural fields, are usually well below 1 km. The percentage of the landscape covered by forest stands older than 80 yr is 15.3%. Clearcutting and subsequent planting produce distinct edges between adjacent stands.
Habitat patches were assigned to one of six habitat categories, which correspond to stages in forest man- agement and regeneration: Old forest (coniferous trees older than 80 yr), Mid-age forest (trees between 50 and
80-yr old), Young2 (trees younger than 50 yr and higher than 1.5 m), Young1 (small seedlings, with heights in the range 0.5 – 1.5 m, form a shrub-like continuous cover), Clearcut (recently cut areas with or without seedlings lower than 0.5 m, sometimes with a few scattered seed trees), and Farmland (trees and shrubs absent, grassland or agricultural crops).
The behaviour of the four bird species that compose the Parus guild in the study area (crested tit, P. cristatus, willow tit, P. montanus, coal tit, P. ater, and goldcrest, Regulus regulus ; see Alatalo et al. 1987) was observed in well-contrasted edges between habitat and matrix. We tentatively identified habitat with Old forest
and defined matrix as any other habitat type. Although this is an extreme simplification of the gradual transi- tion between best habitat and completely unsuitable habitat, titmice and goldcrests prefer old trees which offer an increased volume of foraging substrate and high food availability, whereas these birds rarely visit fields, clearcuts, shrublands or young plantations (Hilde´n 1965, Ekman 1979, Lens and Dhondt 1992, Hansson 1994). Further, all species are bound to conif- erous forest, with the noticeable exception of willow tits which can be found in pure deciduous forest (Nour et al. 1997). Since the preferred Old-forest stands were in short supply, we assumed that all contained resident birds of every species.
Bird songs were played back in a randomly chosen point or station along a randomly chosen edge from the landscape. Each station was operated just once during the study. All stations visited by birds were at least 1.0 km apart. The mean minimum distance ( ± SD) to the nearest station was 2.4 ( ± 4.5) km during the first winter, and 3.0 ( ± 3.2) km during the second winter. Given the separation between stations and the average area used by winter flocks (14.7 ha in the study area, Alatalo et al. 1987; a circular territory of this size has a diameter of 0.4 km), we assume that each bird was exposed just to one playback session. Playbacks were performed for 20 min during the daylight period, be- tween 08:00 and 16:00. We prepared a tape containing
15 repeats of a series of four sequential song bouts, each 20 s long, one for each species. This tape was played back at nearly the maximum volume on a 6-W Hitachi Cassette Recorder TRK-510E which was set on the ground with a single loudspeaker oriented upwards. Although acoustic signals rapidly degrade with dis- tance, they can be correctly identified by birds within
200 m even when degraded sound is played (Naguib
1996). Hence, we assume that all birds present in winter territories, whose average radius in the study area is close to 0.2 km, identified our non-degraded playbacks. These were not operated in rain, snow or strong winds. We observed from the playback point.
Within 30 m around each station we categorized the structure of Old forest as one-layer (only large trees) or multi-layer (large trees plus saplings or shrubs). The recorded number of layers agreed with data for the whole stand contained in forest inventories. In Old forest and Mid-age forest habitat types, we measured timber volume (m3/ha) of each tree species (spruce, pine, birch, and aspen) with a relascope following a standard forestry procedure (Skogsstyrelsen 1988). We converted timber volume figures to proportions which were used as estimates of the tree species composition. In Young1 and Young2 habitat types we estimated the proportion of each tree species within a semicircle of 3 m radius around the station. The average height of even-aged plantations was also estimated in all matrix habitats. We recorded time of day and weather (tem-
perature, sky cover, and the occurrence of wind, haze or mist).
We were able to detect birds within 50 m of the playback point. While approaching or inspecting the station many individuals called or sang which helped us to detect them. We are confident that most, if not all, birds that responded to playbacks were recorded, since all birds first seen even beyond 50 m checked the loudspeaker from a few metres at least once. For every bird observed at the station we recorded species, the side of the edge it came from, and the side of the edge in which it disappeared.
The behaviour of birds was described by five re- sponse variables. Firstly, we noted presence or absence at the station. Secondly, for stations with birds we considered the original habitat from which they were attracted to the station. Given that only in a few stations birds came from both sides of the edge, we made this variable dichotomic: birds coming only from Old forest, birds coming only from the matrix or from both sides. Thirdly, only for stations with birds we computed the number of birds attracted by play-backs. Fourthly, for each individual we recorded whether it crossed the edge or not. We defined a crossing as the loss of visual or auditory contact with a bird (i.e. beyond 50 m from the station) in the edge side opposite to the one it came from. We considered that birds did not cross the edge when they disappeared into the original habitat. Finally, to describe flock cohesion, we calculated the proportion of recorded birds that made the majority decision, either crossing the edge or not, for stations that were visited by at least two birds. Thus, this cohesion index varied in the range 0.5 to 1. In a few cases there were two computable flocks, one from each side of the edge; the cohesion index was then calculated on the total number of non-solitary birds. Note that the sampling unit in the analysis of crossing was the individual instead of the playback station, and if all group members consistently made the same deci- sions our approach would lead to pseudoreplication. However, we could not assume that group membership implied correlated crossing behaviour since bird groups are often loose entities and individuals move indepen- dently at short time scales (Ekman 1979, Hogstad
1988b, Matthysen 1990). We rather treated the indepen- dence of individual decisions regarding edge crossing as an hypothesis which is explicitly tested in the analysis of cohesion.
According to their hypothesised biological meaning, we divided explanatory variables into three groups. The first one included wind and temperature as factors expressing the potential effect of environmental stress on bird energetics and activity levels. The second group consisted of variables describing habitat features in Old forest, including the number of layers and the propor- tions of spruce, pine, birch, and aspen. The third group included variables expressing four components of pre- dation risk:
1) The annual variation in predators’ active search for small birds, represented by the factor YEAR (2 levels: 1991 and 1992).
2) The differential antipredatory shelter provided by habitat elements at the matrix side of the edge. Descrip- tors of this component were the factor TYPE (5 levels, ordered by decreasing availability of woody cover: Mid-age, Young2, Young1, Clearcut, and Farmland), HEIG (average tree height), and the proportions of spruce (SPRM), pine (PINM), birch (BIRM), and conifers (CONM = SPRM + PINM). Since aspen was seldom found, the proportion of deciduous trees in the matrix was expressed by BIRM.
3) Visibility, which may affect bird efficiency in spot- ting potential predators, includes the factor TIME (2 levels: central hours, between 10:00 and 14:00, and crepuscular hours, i.e. 08:00 – 10:00 or 14:00 – 16:00) and the factor SKY (2 levels: clear, and cloudy or misty).
4) Bird aggregation was expressed by four variables: SIZE1, the total number of observed individuals of the same species; SIZE2, the number of conspecifics that came from the same side of the edge, a quantity which conceivably represented an estimate of the minimum size of social groups (cf. Ekman 1979); ALL1, the total number of birds observed at the station; and ALL2, the number of accompanying birds that came from the same side of the edge, which could be assimilated to minimum total flock size.
We used Generalized Linear Models (GLM) to examine the effect of predation risk on the response of birds to play-backs and their subsequent behaviour at old-forest edges. Since our focus was on predation risk, the effect of variables expressing bird activity and old forest quality was controlled for but will not be shown in this paper. We made analyses for each species separately and for the four species together, as they form mixed winter flocks in the study area (Suhonen et al. 1992). The cohesion index and the habitat descriptors ex- pressed as proportions were arcsine transformed. We used binomial errors for dichotomic response variables, Poisson errors for the analyses of number of birds and normal errors for the analyses of cohesion. All explana- tory variables were introduced in a forward stepwise manner, selecting the most powerful model at each step. Missing values in some explanatory variables led us to repeat every analysis with 15 datasets of different length and, consequently, minimum adequate models (sensu Crawley 1993) sometimes consisted of different variables.
We chose the best final models on the basis of three criteria: 1) maximizing the number of variables analysed together: we examined first the model result- ing from the analysis of the shortest dataset, i.e. with all
explanatory variables, and if the effect of the variable responsible for a reduced sample size was not signifi- cant, we proceeded with the next analysis (without that variable and with a larger number of cases); we iterated this rule until the condition was fulfilled; 2) maximizing explanatory power: we also chose the model explaining the highest proportion of deviance, if different from the selected model under criterion 1; and 3) we also consid- ered minimum adequate models containing variables that had a significant effect in univariate analyses after sequential Bonferroni adjustments but were not con- tained in models chosen by either criterion 1 or 2. Bonferroni adjustments were independently calculated for three families of univariate tests that addressed a common null hypothesis (Chandler 1995), i.e. bird ac- tivity, habitat selection inside Old forest, and predation risk.
In order to isolate the effect of predation risk on bird behaviour, we analysed groups of explanatory variables in a sequential way. In the analysis of bird presence/ab- sence we controlled first for the potential effect of bird activity, we found the best model, and on its residuals we started a new stepwise analysis with the variables indicating habitat features in the old forest. Again we found a new best model before analysing the effect of variables expressing predation risk.
The same procedure was followed for the other re- sponse variables, but the rationale for choosing differ- ent groups of explanatory variables varied. In the analysis of bird original habitat we were interested in the frequencies with which birds came from different suboptimal matrix habitats. In this case, we only con- trolled for the effect of weather that could affect the decision of staying in exposed open matrix or crossing into the forest because of energetic reasons.
The third analysis concerned crossing behaviour. We considered weather conditions, habitat quality in Old forest, and indicators of predation risk in the matrix but, because it is difficult to hypothesize the relative importance of each factor in bird decision making, there was no a priori logical order in which these groups had to be examined. Therefore, in this case groups of variables with different biological meaning were not introduced in the model in a sequential way. We analysed the effect of social factors as a component of predation risk only regarding crossing behaviour.
In our fourth analysis, we focused on how flocks modify their size in places where birds perceive different levels of predation risk. Because birds came in most cases from Old forest, and resource abundance in it might influence group size, we controlled for habitat features in that side of the edge before analysing preda- tion risk variables.
In the fifth analysis, only the effect of variables indicating predation risk on the cohesion index was examined.