Study area

The Bohemian Forest represents a forested mountain range along the border between Bavaria (Germany) and the southwestern part of the Czech Republic. The core area of our study was located in two contiguous protected areas, the Bavarian Forest National Park (240 km²; 49°3‘19″N, 13°12‘9″E) and the Šumava National Park (690 km²; 49°7‘0″N, 13°36‘0″E). Data were also collected from the adjacent Bavarian Forest Natural Park (3,007 km²) and the Bohemian Forest Protected Landscape Area (1,000 km²). Together, the four parks form the largest strictly protected continuous forest expanse in Central Europe, the so-called Bohemian Forest Ecosystem. Compared to elsewhere in Europe, the human population density is very low; < 2 inhabitants/km² in the core area and at the margins (outside the national parks) approximately 70 inhabitants/km² in Germany and < 30 inhabitants/km² in the Czech Republic. The elevation ranges from 600 to 1,453 m a.s.l., with an average annual temperature of 6.5°C at lower elevations and 3.0°C at higher elevations. The mean annual precipitation is between 830 and 2,230 mm. Snow cover persists for 7-8 months at the higher elevations and for 5-6 months in the valleys.

Three major forest types dominate the vegetation. Above 1,200 m a.s.l., a montane spruce forest with Norway spruce Picea abies and mountain ash Sorbus aucuparia is found. Between 600 and 1,000 m a.s.l., mixed montane forests with Norway spruce, silver fir Abies alba, European beech Fagus sylvatica and sycamore maple Acer pseudoplatanus grow on slopes; and spruce forests with Norway spruce, silver birch Betula pendula, downy birch Betula pubescens and mountain ash grow in wet depressions with cold air pockets (Heurich & Neufanger 2005). Large parts of the forest, as well as some secondary treeless areas, are left to develop naturally without human intervention (Albrecht 2003).

The primary species of wild ungulates are red deer, roe deer and wild boar Sus scrofa. The Eurasian lynx is the only large carnivore species currently living in this area.

Data recording

In 2005, 2010 and 2011, we captured six free-living Eurasian lynx (four males and two females without kittens) in box traps (Table 1). After trapping, they were anaesthetised, tagged with GPS-collars (GPS Plus Mini; Vectronic Aerospace, Berlin, Germany) and later set free at the same place (Heurich 2011). The collars carried dual-axis acceleration sensors (x- and y-axes) that detected the movement of the lynx every eight second and summarised it in 5-minute intervals. Since 96% of the x- and y-activity data correlated, we only analysed the x-activity; as Löttker et al. (2009) did for red deer. We monitored the lynx during periods of different length. In summary, we acquired 403,467 data points of activity at 5-minute intervals on a total of 1,360 monitoring days (see Table 1).

We defined movement of the lynx body as activity, because it included all possible actions (e.g. climbing, feeding or patrolling the territory). The more the lynx body moved the higher were the values detected by the acceleration sensors. The value scale ranged from 0 (no activity) to 255 (very high activity). However, it was not possible to distinguish the different types of behaviour on the basis of the sensors, except the distinction between activeness and inactiveness, which included resting and sleeping. Therefore, we defined all values from 0 to 27 as inactive behaviour. This limit was chosen because the acceleration sensors were very sensitive to movements of the collar and also detected light head shakings of the lynx, which cannot be stated as activity. This threshold was established in a preliminary study with a captive lynx (P. Löttker, M. Traube, A. Rummel, A. Stache & M. Heurich, unpubl. data).

The mean air temperature per day was calculated from data of one meteorological station in Germany (Waldhäuser) and one in the Czech Republic (Zhuři).

On the basis of 133 clustered GPS positions, we identified 109 kills (82%) in the field during the whole monitoring period: 87 kills were roe deer, 17 were red deer and five were European hares Lepus europaeus. We identified a cluster visually using the program ArcGIS 9.2 (Esri 2009) and it was defined as GPS positions within a circle of about 200 m in diameter (Ersson 2011). We defined days ‘with kill’ (N = 357) as the days on which we knew a killed prey was available because the lynx visited the same place during at least two consecutive nights, and we could find the carcass in the field. These days were subdivided into four categories: day 1 (N = 109), i.e. the day with the first GPS position of the lynx at the kill site (assumed to be the day when the prey was killed); day 2 (N = 105), the day following the kill if the lynx still visited the kill; day 3 (N = 85), the day following day 2 if the lynx still visited the kill; and day > 3 (N = 58), representing the group of days following day 3 if the lynx still visited the kill. The period of time a lynx fed on a kill was determined relying on GPS positions. On average, the lynx consumed a killed roe deer within three days (SD = 1), a red deer within four days (SD = 1) and a hare within two days (SD = 1). The days when we assumed that the lynx had no kill, because the GPS positions did not show any cluster, were summarised into a fifth category ‘no kill’ (N = 316). The last category ‘unknown’ (N = 687) included all days when GPS positions were lacking (although activity data were recorded) or when we in the field could not confirm whether the lynx had a kill or not.

Data analysis

We analysed the daily activity patterns of four lynx: M1, M2, F1 and F2. The other two lynx (M3 and M4) were not included in this first descriptive part of the analysis as their collars only recorded activity data for a few months (see Table 1). We calculated the average activity of the single values (5-minute intervals) per hour for each lynx to obtain a general circadian activity pattern as well as seasonal variation in this general pattern. We defined seasons as spring (March-May), summer (June-August), autumn (September-November) and winter (December-February).

To speed up the following calculation, we used a more parsimonious 15-minute time interval grid instead of the original activity observations on a 5-minute time interval grid. We calculated the respective average values by pooling adjacent original continuous activity observations corresponding to one 15-minute interval, getting 67,048 data points of activity at 15-minute intervals. This data reduction was not accompanied by any substantial loss of information.

To determine the effects on activity as dependent variable, we pursued two different settings, based on generalised additive mixed models (GAMM; Bolker 2008). In a first model, we analysed the duration of activity as response variable, defined as hours per day when the lynx was active (Table 2). In a second model, we analysed the strength of activity as response variable, which is represented by the single continuous values (average values on 15-minute intervals) of the activity. Date, daytime (defined below), sex of the lynx, mean air temperature per day and prey usage were considered as covariates. Temporal dependence of activity measurements was considered by the smooth estimation of the daytime effect. The variable prey usage included the categories ‘with kill’ and ‘no kill’ and is therefore based on 673 days. We included the name of the lynx as random effect to consider the consecutiveness of data collection for each of the lynx (Fahrmeir et al. 2007).

We based analysis on all six lynx, relying on data of all four ‘with kill’ categories (‘day 1’ - ‘day > 3’) and the category ‘no kill’, in the first model and to the 15-minute intervals in the second model. As a further descriptive analysis, we calculated the variation of the circadian activity pattern among these five categories (‘with kill’ and ‘no kill’). Tests for differences in the activity patterns should be based on simultaneous bands for penalised spline estimators. However, these are difficult to achieve and subject to current research in generalised models and correlated data (Krivobokova et al. 2010).

Since we were interested in mean effect of daytime across the year in the second model, the daytime variable was included in the model, which was transformed to become independent of the seasonality of day length. We used daily geo-information of sunrise and sunset and fixed these to 6 am and 6 pm at the transformed daytime, thus twilight was always at the same time across the whole year. The different numbers of 15-minute intervals in between these two time points at each day were stretched adequately. For all analysis, the Central European Time (CET = UTC+1) was used.

All calculations were carried out with the statistical software R 2.11.1 and version 1.6-2 of R package mgcv (Hornik 2011, Wood 2011).

Results

Discussion

Although the activity of an animal would be best analysed by direct observations (Naguib 2006), other methods have to be applied in the case of wild, cryptic species with a low population density (Altmann 1974, Karanth et al. 2010). Camera trapping has been used to analyse activity patterns of ocelots Leopardus pardalis (Kolowski & Alonso 2010), jaguars and pumas (Harmsen et al. 2011), but in these studies, the activity is represented only as the percentage of all collected photos and the individuals cannot always be identified. Both radio-telemetry and GPS-telemetry focus on the movement of an animal and provide more detailed data on activity (Odden & Wegge 2005, Mattisson et al. 2010). However, both methods are invasive (requiring the physical capture of the animals), they can be financially and technically demanding and they imply dealing with ethical matters about animal welfare. Therefore, in general only a limited number of specimens can be monitored by telemetry techniques. Nevertheless, since this is currently the best way to monitor lynx, we used GPS-telemetry, but we acquired fine-scale information on the animals' activity by using GPS-collars with acceleration sensors (Löttker et al. 2009).

The free-living Eurasian lynx in our study area were nocturnal and predominantly crepuscular. In contrast, Reinhardt & Halle (1999) described different daily activity patterns of lynx in Slovenia, which had a higher level of activity during daytime than at night. However, they studied one female lynx with kittens, and other studies have also described different activity patterns for leading females (Schmidt 1999, Kolbe & Squires 2007). The female lynx in our study had no kittens, which could have accounted for the similarities in the activity patterns between the sexes. Bernhart (1990) and Schmidt (1999) found activity patterns of lynx similar to us, with the highest activity at twilight and a higher activity at night than during daytime. Roe deer, the main prey species of Eurasian lynx, and red deer are most active at twilight (Clutton-Brock et al. 1982, Chapman et al. 1993, Danilkin 1996), which may have an influence on the lynx activity patterns, as predators often adapt their behaviour to the activity of their main prey (Ferguson et al. 1988, Beier et al. 1995). Furthermore, in general, lynx and cats hunt predominantly by using their highly developed senses of sight, smell and hearing (Sunquist & Sunquist 2002). Therefore, they are optimally adapted to the conditions at twilight when they are not easily detected by the prey and still can see enough for hunting. Another potential explanation for this activity pattern could be that the lynx avoided humans, whose activity is concentrated during the day. The human population density in our study area is low, but touristic activity must be mentioned and also poaching is a problem (Wölfl et al. 2001). This avoidance behaviour has been shown for snow leopards (Wolf & Ale 2009) and wolves (Theuerkauf et al. 2003), but has not been confirmed yet in the case of the lynx (Sunde et al. 1998, Basille et al. 2009).

The daily activity patterns of the lynx varied among seasons, with activity in winter being concentrated at dusk. This is consistent with the observations of the Canadian lynx Lynx canadensis by Kolbe & Squires (2007). This may be at least indirectly related to the higher ambient temperature at dusk than at dawn. The roe deer, which is the main prey of the lynx, shifts its activity peak from dawn to later in the morning in winter when it is frosty (Clutton-Brock et al. 1982). Daylight may not ensure the best conditions for hunting activity by the lynx. In fact, in our study, we observed no increase in the activity of the lynx during daylight hours in winter. Nonetheless, although we are aware that lynx is not active only when hunting, the change in roe deer activity (related to changes in temperature) may have played a role in the observed decrease in lynx activity at dawn, making it less profitable for the lynx to hunt at dawn than at dusk. The one exception was lynx F2, which was as active at dawn as at dusk. It is unclear whether the subadult status and lower hunting experience of lynx F2 influenced this pattern.

As observed in our study, Beltran & Delibes (1994) and Schmidt (1999) also found that lynx avoid high ambient temperatures, especially in summer.

In general, it takes longer than one night for lynx to consume large ungulate prey (Breitenmoser & Breitenmoser-Würsten 2008). This allowed us to detect the presence of potential kills based on cluster analysis of the positions provided by the GPS-collars. Using this method, we could find and verify most of the killed ungulate prey in the field, while smaller prey animals were difficult to recognise, only five hares were identified as consumed prey. However, we performed a scat analysis and only found hair from hares in 5% of the scats. In about 15% of the scats, we found other small mammals, like mice that do not change the activity or movement behaviour. This indicates that we did not miss much.

In accordance with previous studies (Bernhart 1990, Reinhardt & Halle 1999, Schmidt 1999), we found that both the duration and the strength of lynx activity were significantly lower on days when a kill was available than on days with no kill. This could be explained by the lack of need to hunt for several days (Okarma et al. 1997), and by the lynx staying near the kill to prevent losses to scavengers. This would also explain why the level of activity on consecutive days with a kill remained constant. In contrast, Schmidt (1999) found a lower activity on the first day ‘with kill’ than on the next days spent consuming the kill. However, his analysis was limited to the movement of the lynx, which is just one aspect of all activities. Jedrzejewski et al. (2002) found that activity is highest on days without a kill and lowest on the first day after a kill.

In our study, the activity pattern of the days with a kill differed from the pattern of days without a kill only during a 2-hour period, from 9 pm to 11 pm. On days without a kill, lynx may use this time interval to patrol their territories and visit various hunting grounds (Jedrzejewski et al. 2002, Breitenmoser & Breitenmoser-Würsten 2008), whereas on days with a kill, they probably consume their prey during this time interval. The lynx activity pattern during the rest of the day was independent of the availability of a kill. Other environmental factors, such as the temperature, clearly had a greater effect on the activity pattern.

In conclusion, our study indicated that there are detectable differences in duration, strength and patterns of activity between the days with and without consumption of a prey. On this basis, it may be possible to develop a method to detect the presence of a kill by analysing the activity data (in combination with the information from the GPS positions). This would greatly reduce the time effort required for searching for kills in the field, and it would especially help when studying the lynx, whose weight and lifestyle impose limitations to battery size and therefore to the obtainable number of GPS positions. Increasing the availability of data on the hunting rates of this species, would improve our knowledge about an important aspect of this species' ecology.