Movement and ranging patterns of the Common Chaffinch in heterogeneous forest landscapes

Study area and bird capture

Our study was conducted in south-western Germany in the Schwäbische Alb (approximate central coordinates: 48.41°N, 9.50°E, elevations of 500–800 m a. s. l.), a montane environment with a long history of anthropogenic forest management and small-scale agriculture. The land cover in the area was originally dominated by beech forest (Fagus sylvatica) but has been significantly altered by the logging of some of the forests and a variety of farming practices (Fischer et al., 2010). The forests are characterized by relatively small and discrete woodland stands with distinct forest management strategies (mean area of the forest stands 3.6 ± 5.8 ha; min/max: 0.004/84.5 ha; N = 1549) and interspersed open habitat. Forest stands range from old beech stands, neglected by management for more than 100 years, to mixed-deciduous/coniferous forests with limited intensive logging to intensively farmed monocultures of beech or Norway spruce (Picea abies). Each forest type is represented by a range of age classes from young plantations to stands with trees of up to 140 years old.

We captured and marked a total of 15 male adult chaffinches, Fringilla coelebs, in the field between May and July 2009 (5 individuals) and between April and June 2010 (10 individuals, one bird tracked in 2009 was recaptured and also tracked in 2010). During this time, most males had well-defined territories, which they defended against other males that usually stayed in close proximity to each other in the study area (K Kubiczek, pers. obs., 2010). Birds were captured in the vicinity of so-called experimental plots of the research platform ‘Biodiversity Exploratories’, of which each was characterized by homogeneous forest cover of at least 100 ×100 m (http: //www.biodiversity-exploratories.de; Fischer et al., 2010). A large range of abiotic and biotic information are collected at these plots, including temperature measures; however, we avoided trapping birds directly there in order to avoid disturbing other experiments and, also, as bird ranges cannot be directly linked to a plot-based monitoring scheme. We captured birds with nylon mist nets placed in front of playback units. We measured wing size as the length of the third primary with a pinned metal ruler to the nearest millimetre and body mass with a spring balance to the nearest 100 g/1 g division (Pesola, Baar, Switzerland).

Each captured bird was ringed with a uniquely numbered leg ring (Vogelwarte Radolfzell, Germany). For radio tracking, birds were equipped with a small radio transmitter (BD-2G, Holohil Systems Ltd., Ontario, Canada), which weighed 1.2 g and was therefore within the 3–4% body mass range assumed suitable for carrying without causing any harm to or having any impact on the behaviour of small passerine birds (Withworth et al., 2007). We constructed a backpack harness out of an elastic plastic band (High Power Gum, 0.6 mm, Lenzi) to which the transmitter was attached with a small drop of adhesive glue (Loctite, Henkel, Düsseldorf, Germany). The harness fixed the transmitter on the back of birds with loops around the legs; details of such bird harnesses have been described in (Naef-Daenzer, 2007). After checking for the accurate fit of the harness, we immediately released the bird at the point of capture.

Capturing and handling birds were performed in compliance with federal and state laws. All permits were granted by the “Regierungspräsidium Tübingen, Referat Artenschutz, Tierschutz” (RPT Tierversuch-Nr. -884, -907, -1056). All birds were handled to best practice following the guidelines of the bird banding laboratory “Vogelwarte Radolfzell” and we minimized suffering through swift release into the wild and professional harnessing techniques as recommended by the permit authorities. These guidelines on bird handling for scientific purpose implemented all steps requested by the animal welfare of the European Commission, which are implemented in the federal and state laws of Germany.

The mandatory training of the field workers was assessed during the permit procedure. Access to land was approved by all land owners.

Bird tracking

We used VHF radio-telemetry with a hand-held Yaesu VR-500 receiver (Yaesu Musen Co. Ltd., Tokyo, Japan) combined with a modified H-aerial antenna (Wagener Telemetrieanlagen, Köln, Germany) for tracking birds at regular time intervals with the so-called homing-in method. During tracking, we narrowed down the birds’ location by a single observer repeatedly checking the direction of the intensity of the radio signal from different angles. Birds usually perched in the canopy of larger trees, about 5–20 m above ground, and approaching these trees, we were able to estimate the geographic locations with a precision of ca. 10 m. With this method, we recorded bird locations in regular 10-min intervals. Only occasionally birds moved too large distances to follow them and accurately determine their location in 10-min intervals (about 20 times in >2,000 location records).

Each bird location was marked with a ribbon attached to the nearest branch or tree and an individual location number and we noted the time of observations. During tracking, we estimated coarse geographic position with a conventional handheld GPS with relatively large error of ca. 10–15 m (Garmin GPSmap 62CX, USA) for orientation only, whereas we determined the exact geographic coordinates of marked locations with a Trimble geoXTTM GPS and GeoBeaconTM receiver (Trimble, USA) with an accuracy of <1 m some days after radio tracking was finished.

We collected a first set of radio-telemetry data for each individual two days after the radio transmitter was attached to the bird, tracking birds from 10–30 min prior to sunrise for a minimum of 6 h and at a minimum of 30 accurate tracking locations at 10-min intervals. We radio-tracked individuals on three consecutive mornings. Another set of tracking was conducted after a break of 3–7 days, repeating the same protocol and gathering information for another three days. As we expected to cover the main daily activity period of birds with this tracking, we refer to ‘daily ranges’, being aware that this is only a relative measure that should not be confused with absolute home-range estimates.

Environmental characterization

At each bird location, we estimated local tree density and species composition by using a so-called dendrometer. For this, a metal bar of 15 mm width was held in a constant distance of 50 cm from the observer’s eye at the bird locations. All trees with diameters at breast heights ⩾15 cm and encountered as exceeding the width of the metal bar were counted. These trees were identified with regard to their species and diameter and their distance to the bird location were recorded. This method is used in forestry to estimate tree stock (Kramer & Akca, 1987) and proved to be an effective measure of local tree assemblages in the vicinity of bird locations.

For classifying area-wide forest cover and the forest stand types that birds used, we used a digital regional forest management map, which had information on tree species composition, stand age and management plan on the stand-level of discrete management units and which was obtained from the regional authorities (‘Forsteinrichtungsdaten’; ‘Fogis 2006’, Regierungspräsidium Tübingen). The digital maps accurately described at least 90% of forest stands, as confirmed by our own field inventories. We categorized forest stand composition (based on beech and spruce as the dominant overstory tree species, with a threshold level of 70% of dominant trees) into six categories: beech, beech-mixed, deciduous-mixed, spruce, spruce-mixed and coniferous-mixed (Wells et al., 2011). Stand age was defined as the oldest tree layer within stands. Forest management classes included age-class forest and unmanaged forests.

Local temperature data were obtained from meteorological measurement stations at nearby experimental plots (Rotronic HygroClip S3 Air Temperature sensor) in the vicinity of tracking locations and were measured at heights of 2 m at 60-min-intervals (we assigned the measured temperature value to all bird locations until the next measurement was available).

Modelling approach and data analysis

We performed data analysis at two different levels, namely at point-level movement distances (distance moved per 10-min time intervals) and the level of daily ranges (areal metric derived from subsets of points). Whereas these data sets are linked, in that ranges are estimated from subset of point locations, we point out that environmental attributes need different consideration at these two scales. In general, any point location used by a bird at a particular time can be described by local tree assemblages at this location nested within the forest stand in which this point is located (e.g., we can expect two point locations in a beech forest to differ in local tree assemblages if mixed with other tree species or if spatial clumping results in different tree densities; likewise, if two points with the same point-level attributes are located in a different forest stand, we can expect this to be a different environment, see Fig. 1). Each point location used at a particular time is further characterized by the local temperature and daytime. In contrast, at the range level, we can characterize the forest environment that has been used by a bird during a certain time as the average of point-levels attributes from utilized points. Independently of point estimates, we can characterize the overall forest landscape utilized during a certain time period from area-wide forest maps by describing the forest landscape underlying the resulting range. For a better understanding, we superscript environmental variables below with ‘^P’ for point-level, ‘^S’ for stand-level and ‘^R’ for range level attributes (see Table 1 for a detailed description of environmental variables).

We calculated movement distance as the bee-line distance between bird locations estimated at consecutive time steps of 10 min. For this, we considered all records with time lags between 8 and 12 min, as field conditions do not allow to sample in strictly regular time intervals. For analysing possible relationship between movement distances and environmental conditions, we matched covariates with the starting point from which movement distance had been calculated. We found point-level characterization of the forest at GPS locations more feasible than characterizing habitat along straight-line sections between points (see Fortin et al., 2005 for an example of step-selection functions) and also on the rationale that habitat along lines can be assumed to be less influential during flight. We constructed a hierarchical model, so that we were able to translate the hierarchical structure of points nested in different forest stand types into an analytical framework. Each observed movement distance λ(i, d, t) of bird individual i at observation day d and time t was characterized by the point-level number of beech trees B^P, the number of spruce trees S^P, the local density of all tree species X^P and the Shannon diversity index of all tree species H^P at respective point locations. T^P was the ambient temperature assigned to the respective location and time and we noted daytime Dt^P as the minute time difference of observations to sunrise time. Each point location was characterized by its surrounding forest stand type S.type^S and stand age S.age^S. We assumed that the effects of point-level conditions (variables with ‘^P’) were not independent of the surrounding forest stand type. Likewise, the effect of stand age could be assumed to vary with stand type.

We further assumed the movement distance to exhibit some autocorrelation in space and time because of similar environmental conditions or intrinsic factors such as the motivation of a bird to move, so that λ(i, d, t) might depend on λ(i, d, t−1), which is a 1st order autoregression term. Modelling movement distance at the log-scale with these assumptions gives the following model: ${\displaystyle log\left(\lambda \left(i,d,t\right)\right)\sim {\mu }_0\left(i\right)+\alpha \left(i\right)\lambda \left(i,d,t-1\right)+S.typ{e}^S+{\beta }_1\left(S.typ{e}^S\right)B^P+{\beta }_2\left(S.typ{e}^S\right)S^P+{\beta }_3\left(S.typ{e}^S\right)X^P+{\beta }_4\left(S.typ{e}^S\right)H^P+{\beta }_5\left(S.typ{e}^S\right)T^P+{\beta }_6\left(S.typ{e}^S\right)D{t}^P+{\beta }_7\left(S.typ{e}^S\right)S.ag{e}^S+\epsilon .}$ Here, μ₀(i) is the intercept that is allowed to vary among individuals. Nested/hierarchical structure of point-level attributes are implemented via coefficient estimates β, which are allowed to vary for different forest stand types; stand type is thus considered as both a group-level predictor and a grouping indicator. The zero-mean Gaussian error ε captures both variation due to observation error and residual variance of the process model, as our model framework did not explicitly account for measurement errors in a separate observation model. We did not consider variable selection in our modelling efforts, as we were more interested in inferring the strength and different magnitudes in coefficient estimates, rather than model parsimony.

We fitted the model with a Bayesian approach by using Gibbs sampler as implemented in the OpenBUGS software (Lunn et al., 2009) with the vague uniform priors of U(0, 100) assumed for all variance terms. All covariates were scaled to a mean of zero and one SD prior to analysis. There was no collinearity between covariates.

We ran two MCMC chains and found convergence and stationary posterior distribution after visual inspection of 100,000 iterations, which were discarded. Posteriors were estimated from 5,000 further iterations for each chain as the highest posterior density mode and 95% credible intervals (CI). Covariate relationships were assumed to be statistically significant if 95% CI did not overlap zero. We estimated finite sample variances for estimating the % variance explained in the data by different predictors in our model (see Appendix S1 for model code and further information).

For estimating the daily ranging patterns of birds, we calculated Convex hull polygons for each days’ range estimate (Getz & Wilmers, 2004), by using the first 30 tracking points per day. Daily ranges were then estimated with the fixed nearest neighbour method with k = 12 points and areal range sizes were estimated as 90% isopleths. Range analysis were conducted with the LoCoH.k() function of the adehabitatHR package in R (R Development Core Team, 2013).

For examining possible relationships between range size and environmental covariates, we used a linear mixed effects model fitted with a restricted maximum likelihood approach (using the lmer() function in the lme4 package in R) to account for the repeated measure of ranges from the same individual as a grouping structure, since our longitudinal set of ranges comprised variation in range sizes over time for any individuals and variation in range sizes among individuals under various environmental conditions. As a full model, we modelled daily range size at log-scale by using individual bird identity Bird^R, mean temperature Temp^R, julian day JulDay^R, stand numbers StandNo^R, proportion of beech stands BeechStand^R, proportion of spruce stands SpruceStand^R underlying the ranges and the average number of beech trees BeechTree^R, average number of spruce trees SpruceTree^R and total average number of tree species TotalTree^R per location within daily ranging as covariates (see Table 1). For the latter six covariates describing forest attributes within daily ranges, we introduced bird identity as random effects. We constructed various models with only single covariates and subsets and compared model outputs visually and based on Akaike’s information criterion and the size and variation in coefficient estimates (as none of the covariates proved to be of relevance, details not outlined).

Movement distances

Movement distances exhibited a highly skewed distribution towards small movements within 10-min intervals and only occasional large movements (Fig. 2). In 117 out of 2,316 records (5%), birds did not move but perched at the same location for at least 10 min. Median movement distance was 35 m during the 10-min intervals and the longest recorded movement was 730 m (note that a few longer movements were likely to be missed due to difficulties in following long distances during tracking).

Movement distances differed among forest stand types, smallest movements being recorded in beech forest stands and largest movement distances being recorded in non-spruce conifer and spruce-mixed forest stands, explaining 16% of the finite-sample variation in observed movement distances (Fig. 3, see Appendix S2 for all posterior values). Movement distances varied with stand age, notably with an evident effects of stand age only being found in beech forest stands: movement distances increased with stand age in beech stands but not in other stand types (Fig. 3). In contrast, the effects of point-level tree assemblages were all low, and we did not find any association with variation in movement distances. Movement distances increased further with daytime in spruce and beech-mixed forest stands, but this effect was less obvious in other stand types due to uncertainty in coefficient estimates (Fig. 3). Movement distance did not reveal any relationship to temperature. Variation in movement among bird individuals accounted for 5% of the finite-sample variation in movement distances but we found no relationship between individual average movement distances and wing size or body mass (results not shown in detail). Posterior modes for autoregression coefficients α were for most birds 0 < α < 0.5 and for one bird individual 0.5 < α < 1, suggesting that movement distances from previous time intervals were only of minor impact (Appendix S2).

Ranging patterns

Daily range sizes based on 90% isopleths of convex hulls ranged between 0.01 and 8.0 hectare (median of 0.86 ha), exhibiting a similar magnitude of intra-individual variation over time than variation among individuals foraging in the different forest types (Fig. 4). Indeed, 40% of the variation in daily range size was accounted for by intra-individual variation. None of the covariates characterizing underlying forest cover derived from point-level tree assemblages, regional forest maps, temperature or Julian day revealed any relationship with daily range sizes based on mixed effect models.

Notably, 70% (57 out of 81) of daily range estimates covered more than one forest stand type and also relatively small daily ranges were observed to cover more than three different stand types.