Long‐term changes in habitat selection and prey spectrum in a reintroduced Eurasian lynx (Lynx lynx) population in Switzerland

Abstract When wild‐caught Eurasian lynx (Lynx lynx) from the Slovak Carpathian Mountains were reintroduced to Central Switzerland in the early 1970s and spread through the north‐western Swiss Alps (NWA), they faced a largely unfamiliar landscape with strongly fragmented forests, high elevations, and intense human land use. For more than 30 years, radio‐collared lynx have been monitored during three different project periods (in the 1980s, 1990s, and 2010s). Our study explored, how lynx over generations have learned to adjust to the alpine environment. We predicted that (1) lynx nowadays select more strongly for open habitats, higher elevations, and steep slopes compared to the early stages of recolonization and that (2) consequently, there were significant changes in the Eurasian lynx’ prey spectrum. To test our predictions, we analyzed telemetry data (VHF, GPS) of 13 adult resident lynx in the NWA over 35 years, using Resource Selection Functions. Furthermore, we compared kills recorded from different individuals inhabiting the same region during three project periods. In general, lynx preferred forested areas, but over the years, they avoided open habitat less. Compared to the early stage of the recolonization, lynx in the most recent project period selected for higher elevations and the proportion of chamois in their prey spectrum surmounted that of roe deer. Potential driving factors for the observed changes could be increasing tolerance to human presence, intraspecific competition, or fitness benefits through exploitation of new resources. Long‐term studies like ours provide important insight into how animals can respond to sudden environmental changes, e.g., in the course of translocations into new areas or anthropogenic alterations of their habitats.


| INTRODUC TI ON
Large carnivores were almost entirely eradicated from Western and Central Europe by the end of the 19th century. The main reasons for their demise were human persecution, habitat destruction, and loss of their prey base (Breitenmoser, 1998). In recent decades, carnivore species, such as the grey wolf (Canis lupus) or the Eurasian lynx (Lynx lynx), have successfully re-established stable or growing populations in many parts of Europe (Chapron et al., 2014). This conservation success has been aided by protective legislation, growing populations of wild ungulates, forest regeneration and-in the case of the Eurasian lynx-by several reintroduction programs (Breitenmoser et al., 2021).
However, with most natural areas in Europe largely destroyed and fragmented by humans due to land use practices, including forestry, agriculture, tourism, and power supply industry, wildlife species nowadays live in human-modified habitats. In these altered landscapes, the management and conservation of large carnivores are especially challenging due to their large spatial requirements and great potential for conflicts with human activities (Chapron et al., 2014;Schadt et al., 2002). Current threats to large carnivore populations include habitat fragmentation, illegal killing, and loss of genetic diversity in small and isolated populations (LCIE, 2021). For lynx conservation in such human-modified landscapes, it is essential to understand their space use and habitat choice, in order to predict how anthropogenic changes to landscape are likely to affect lynx habitat suitability and ultimately population viability (Bouyer et al., 2015;Grilo et al., 2019;Penteriani et al., 2018).
In Eastern, Central, and Western Europe, continuous forest cover is generally thought to be a prerequisite for the establishment of viable Eurasian lynx populations (Haller, 1992;Müller et al., 2014;Niedziałkowska et al., 2006;Rozylowicz et al., 2010;Schadt et al., 2002;. Indeed, the Eurasian lynx is often used as a flagship species for the protection of intact and unfragmented forest ecosystems (Niedziałkowska et al., 2006;Noss et al., 1996). However, recent studies have shown that Eurasian lynx can successfully deal with the trade-off between avoidance of human activities and the preference of areas with high prey densities, which often positively correlate with areas of intensive human land use (Bouyer et al., 2015;Filla et al., 2017;Gehr et al., 2017). Some studies show that Eurasian lynx increased their use of open habitats and areas with high human disturbance at night, while they preferred dense habitat in undisturbed areas for resting during the day (Bouyer et al., 2015;Filla et al., 2017;Gehr et al., 2017), especially where lynx are persecuted (Magg et al., 2016).
For most reintroductions, wild-caught Carpathian lynx (Lynx lynx carpathicus)-representing the geographically closest remnant population-were used (Breitenmoser et al., 2021). Between 1971 and 1980, lynx captured from the autochthonous population in the Slovak Carpathian Mountains were first released in Switzerland in the Alps and the Jura Mountains (Breitenmoser & Breitenmoser-Würsten, 2008). Not all of these releases led to a successful establishment of a population nucleus, but nevertheless growing lynx populations have since established in the Swiss and French part of the Jura mountains as well as in the north-western Swiss Alps (NWA, Breitenmoser & Breitenmoser-Würsten, 2008). In those parts of the Slovak Carpathians, where lynx occur, forest cover ranges between 60% and 90% (Kubala et al., 2019). Therefore, when lynx were translocated from the Carpathian Mountains to the NWA, they faced a largely unfamiliar environment. Elevations range up to 4273 m a.s.l, the valley bottoms are occupied by human settlements and instead of densely forested slopes, the NWA comprise a high amount of man-made pastureland in the montane and subalpine zone. Forests cover only about 30% of the area and are strongly fragmented, with timber line lowered by several hundred meters (Vogt et al., 2016).
In a long-term study area situated in the NWA, research on Eurasian lynx started in the 1980s and since then has been conducted throughout three different project periods: NWA I (1983NWA I ( -1985, NWA II (1997-2001), and NWA III (2011 (Breitenmoser & Haller, 1993;Breitenmoser-Würsten et al., 2001;Molinari-Jobin et al., 2007;Vogt et al., 2018). In the early 1980s (NWA I), lynx had only colonized the northern ranges of the Alps with a higher share of forest cover. Haller and Breitenmoser (1986) concluded that the more elevated and less forested landscapes further south did not provide enough suitable habitat for lynx to settle there permanently. In the 1990s (NWA II), however, lynx had also colonized these areas initially considered less suitable and had started using open habitats more and avoiding agricultural land less than expected (Breitenmoser-Würsten et al., 2001). This raised the question, whether lynx in Switzerland were adjusting their spatial behavior to the new habitat in the cultivated landscape of the NWA and whether the behavioral plasticity of this species, and possibly its adaptive potential, could be higher than originally anticipated.
The semi-open landscapes of the NWA, with their montane and subalpine altitudinal levels transferred into alpine pastures, imply a higher human presence but also offer valuable additional resources.
For example, open habitats above the timber line, as well as steep forested slopes are the preferred habitat of chamois (Schnidrig-Petrig & Salm, 2009). While the Tatra chamois (R. r. tatrica) is a rare mountain ungulate in the Carpathian Mountains (Anderwald et al., 2020;Corlatti et al., 2022), the Alpine chamois (R. r. rupicapra) is the most abundant ungulate in the NWA (LANAT, 2019).
The aim of our study was to investigate whether Eurasian lynx have adjusted their habitat selection subsequent to their reintroduction. We predicted that (1) lynx nowadays select more strongly for open habitats, higher elevations, and steep slopes compared to the early stages of recolonization (NWA I, NWA II) and that (2) consequently, there were significant changes in the Eurasian lynx' prey spectrum reflecting the altered habitat use, i.e., a higher predation of Alpine chamois. While behavioral plasticity of habitat selection in mammals has been studied along environmental gradients (Fortin et al., 2008), there are only few studies addressing long-term changes in habitat use (Ciach & Pęksa, 2019;Fierro-Calderón & Martin, 2020). Our unique dataset comprising telemetry data over a period of 35 years allows us to investigate changes in spatial behavior over time. Especially for long-living species like large carnivores, such studies remain rare and can provide important information for future conservation and reintroduction programs (Smith et al., 2017).

| Study area
The study site is mainly situated in the Bernese Oberland, a moun-  Vogt et al., 2019). Red deer occur at low numbers but have not yet been found in lynx prey . Yearly censuses of ungulate species are carried out by the cantonal game wardens using direct observations from vantage points for Alpine chamois and spotlight counts along transects for roe deer. Censuses were carried out in late autumn and early spring and census data represents population numbers in spring before reproduction (more detail in Vogt et al., 2019). While there is a potential bias to these census methods (e.g., Meriggi et al., 2008), censuses were carried out by the same observers throughout all three project periods (N. Blatter, pers. Comm.).

| Lynx captures and collaring
Lynx were captured between 1983 and 2016 by means of three capture systems. Since the beginning, lynx have been captured either with foot snares set at fresh kills or in large double-door live traps set on narrow paths (Breitenmoser & Haller, 1993;Vogt et al., 2016). Since 2005, a remotely controlled teleinjection system (Ryser et al., 2005) was additionally used. All lynx were captured following established standard protocols (described in (Ryser et al., 2005;Ryser-Degiorgis et al., 2002;Vogt et al., 2016) with all permits required by Swiss legislation for capturing, immobilizing, and radio-tagging lynx.

| Lynx location data
Data were gathered as part of three research projects (NWA I, 1984(NWA I, -1988NWA II, 1997andNWA III, 2011-2017;  Ecology and Wildlife Management, www.kora.ch). VHF telemetry was used during the NWA I and NWA II projects, while lynx observed during the NWA III project were fitted with GPS/GSM collars. This study is based on telemetry data obtained from 13 resident adult lynx (five males, eight females), which were observed for at least 9 months (Table 1). The average duration lynx were monitored was 23 months for VHF, and 17 months for GPS/GSM collar data.
The number of locations per individual used for this analysis ranged between 80 and 555 (mean = 274, SD = 155).

| Triangulation of VHF Signal and "homing in"
When lynx were tracked by means of VHF telemetry, five levels of accuracy were distinguished (Breitenmoser-Würsten et al., 2001).
Triangulation was used to determine the lynx' approximate position and the exact position was affirmed whenever possible by F I G U R E 1 Location of the study site within Switzerland (inset). The black polygon indicates the 95% Minimum Convex Polygon for all telemetry locations of observed lynx (N = 13). Green areas represent forest "homing in" on the signal until a level 3 or level 4 accuracy could be reached.
• Level 0: Lynx searched, but not found on that day; • Level 1: VHF signal heard from one direction, but not localized more precisely; • Level 2: position located with an accuracy of ±500 m (at least 3 bearings); • Level 3: position located with an accuracy of ±50 m; • Level 4: exact position affirmed through direct observation, a kill or tracks.
Lynx individuals were monitored ranging from once per week to several times per day. Lynx usually stay at one place during daytime (Breitenmoser & Haller, 1987) and return to their prey after dusk and during the night (Breitenmoser & Haller, 1993).
When a lynx was located at the same spot for a longer time during the evening, it was assumed to have a kill and the surrounding Note: Location data were obtained from 13 adult and resident lynx from three different study periods in the north-western Swiss Alps (NWA). The sample includes five males (in bold; 1324 locations) and eight females (2241 locations). VHF telemetry was used during the NWA I and NWA II study periods. GPS telemetry was used during the NWA III study period. The VHF dataset includes locations of accuracy levels 2 to 4. The number in brackets indicates the share of locations of accuracy 2. The GPS telemetry dataset was reduced to one location per day.

TA B L E 1 Sample size overview
area was searched for prey remains the following day, using a trained dog whenever possible.

| GPS location cluster control
With the use of GPS/GSM collars, searching for kill sites became a lot less time consuming. GPS collars were programmed to obtain seven GPS fixes per day, with a higher resolution during twilight and night hours: 01:00, 04:00, 13:00, 17:00, 19:00, 20:00, and 21:00 CET. To locate kill sites, ground-truthing of GPS location clusters (GLC) was used as described in Vogt et al. (2018).

| Statistical analyses
All statistical analyses were performed in R software (version 3.3.3, R Development Core Team, 2015).

| Data structure
Since the VHF location dataset was strongly biased toward daylight hours and lynx were often located only once per week to once per day, we restricted the GPS dataset to similar observation times to make it comparable to the field effort invested during the previous VHF telemetry studies. The reduction of the GPS dataset to one fix during daytime between 12:00 and 02:00 CET for all GPS-collared lynx resulted in 1956 GPS locations (Table 1). The VHF dataset contained 11.2% of localizations of accuracy levels 0 and 1, which were excluded from the analyses, as they lack coordinates. Localizations of accuracy level 2 made up 32.4% and were also excluded, as they do not allow for an accurate assignment to a specific habitat type (accuracy ±500 m). For the habitat analyses, we used localizations of levels 3 and 4 only (1024 locations; 56.4%). In total, 2980 lynx locations (GPS, VHF) were used for the purpose of this study (Table 1).

| Calculating home ranges
We estimated home range sizes by generating a 95% fixed kernel with a smoothing factor set to 1000, using the R package adehabitat (Calenge, 2006). The smoothing factor was evaluated visually to select the most biologically sensible estimate (Peters et al., 2015). Data points outside the estimated 95% Kernel were considered as outliers in individual range use and excluded from further analyses (Filla et al., 2017).

| Modelling habitat use (Resource Selection Function)
Based on proposed scales of habitat selection (Johnson, 1980), we used resource selection functions (RSFs) to assess lynx' habitat selection within their home ranges (third-order selection) under a useavailability design (Manly et al., 2002). RSFs analyze spatial patterns of animal locations obtained from telemetry studies, and have become a widespread method to identify habitat types that are used disproportionately in relation to their availability (Moorcroft & Barnett, 2008). To define resource availability, we generated a set of random 'available' locations within home ranges equal to the number of 'used' locations obtained from each individual lynx (Peters et al., 2015).
We To account for changes in habitat use among the three project periods, we included the variable 'project' as a factor. The latest project period (NWA III) served as reference because the main focus was on differences in behavior between now and then. The model does not include all possible two-way interactions, only the ones relevant for hypotheses testing. Individual identity was added as a random effect to our model in order to account for individual preferences and for differences in sample size (Gillies et al., 2006).   Figure 3). There was no significant difference in selection of slope between project periods ( Figure 4). Lynx strongly avoided unsuitable habitat during all three project periods.

| Model fit
Our

| Changes in prey spectrum
The ratio of abundance between Alpine chamois and roe deer in our study area increased from period NWA I (Ø 1.63) to NWA III (Ø 2.51) by 54% and declined from NWA II to NWA III (Ø 1.87) by 25% ( Figure 5). The lynx inhabiting the same region over different time periods showed a significant difference in prey spectrum (Fisher's exact test, p = 0.002). The proportion of Alpine chamois increased from 33% in NWA I to 53% of the total prey spectrum in NWA III ( Figure 6). The proportion of roe deer decreased over the years from 68% to 47%.

| Habitat selection
In parts of their distribution range, Eurasian lynx are inhabiting grasslands, Mediterranean shrublands, or alpine tundra (Linnell et al., 2021;Mahdavi et al., 2020;Mengüllüoğlu et al., 2018;Rauset et al., 2013). In Eastern, Central, and Western Europe, however, the Eurasian lynx is often used as a flagship species for the conservation of forest habitats (Niedziałkowska et al., 2006;Noss et al., 1996). Also, in our study, lynx preferred forest in all three project periods, which is in line with previous investigations (Basille et al., 2009;Breitenmoser-Würsten et al., 2001;Filla et al., 2017;Haller, 1992;Rozylowicz et al., 2010;Zimmermann, 2004;. However, we observed a weaker avoidance of open areas in more recent project periods (NWA II, NWA III) compared to the earliest project period few years after recolonization (NWA I). This suggests that lynx started using the open habitats of the montane and subalpine zone more frequently already in the 1990s (Breitenmoser-Würsten et al., 2001) compared to the 1980s, where observed lynx were located primarily in closed forest (Breitenmoser & Haller, 1993;Haller & Breitenmoser, 1986). The lynx observed by Breitenmoser and Haller (1993) (Lorenzini et al., 2022, in press) and Eurasian lynx face a trade-off between avoiding human disturbance and selecting areas with high prey abundance (Basille et al., 2009;Bouyer et al., 2015;Filla et al., 2017;Gehr et al., 2017). Two recent studies on Eurasian lynx habitat selection have found that lynx solved this trade-off by avoiding open habitats less during twilight and even by selecting meadows at night when human activity was low (Filla et al., 2017;Gehr et al., 2017). With most telemetry data in our study obtained during daylight hours, we can conclude that lynx also became less avoidant of open habitats during daytime, when they were mainly resting and not hunting.
All observed lynx selected for steeper terrain than available on average. However, contrary to our predictions, there was no significant difference in selection for slope over time. In the most recent project period (NWA III), lynx selected for higher elevations more strongly than in the earlier period of reintroduction. In humandominated landscapes, steep slopes and high elevation often correlate with lower levels of human disturbance (Basille et al., 2009;. Large carnivores incur higher energetic costs for movement in steeper terrain and avoid steep slopes in undisturbed areas. In the presence of humans, however, they are known to choose energetically suboptimal movement strategies and relax their avoidance of slope (Nickel et al., 2021;Nisi et al., 2021). Avoidance of humans was a likely driver of the selection for steep slopes across all project periods in our study area. However, it does not fully explain the stronger selection of high elevations in the most recent project period.
Increasing population densities can drive competitively inferior animals into habitats they would not normally prefer and which may or may not prove suboptimal for their reproductive success (López-Bao et al., 2011;O'Neil et al., 2020;Svanbäck & Bolnick, 2007). When the lynx population in our study area increased and suitable habitat became saturated (Breitenmoser-Würsten et al., 2001;Chapron et al., 2014;Kunz et al., 2017;Zimmermann et al., 2012Zimmermann et al., , 2016, lynx may have been forced to better utilize their home ranges vertically. Another driving factor for habitat selection in large carnivores is prey availability (Bouyer et al., 2015;Cristescu et al., 2019;Davidson et al., 2012;Oakleaf et al., 2006;Roder et al., 2020;Soyumert et al., 2019). Roe deer density has been shown to have a positive influence on lynx occurrence (Bouyer et al., 2015;Müller et al., 2014).
Accordingly, Basille et al. (2009) linked the avoidance of alpine areas and higher elevations shown by Eurasian lynx in Southern Norway to the lack of a suitable main prey species in these habitats. In our study area, higher elevations comprise Alpine pastures inhabited by suitable prey species like Alpine chamois, Alpine marmot Marmota marmota, and mountain hare Lepus timidus (Jobin et al., 2000;Vogt et al., 2018. Hence, the observed vertical home-range expansion by lynx could have resulted in better exploitation of additional resources (see Figure 6 on prey spectrum), which may in turn have implied fitness benefits through higher food intake (Holekamp & Dloniak, 2010).

| Methodological bias
The analysis of long-term datasets is often characterized by changes in animal tracking technology (Land et al., 2008). This is also the case in our study, where VHF telemetry was superseded by GPS telemetry in the most recent project period. With GPS telemetry, the physical presence of an observer in the field is no longer required to obtain a localization, alleviating potential biases of terrain accessibility. Indeed, in our VHF dataset, locations of lower accuracy were on average in steeper and higher terrain than locations of high accuracy, corroborating the fact that VHF telemetry in steep and inaccessible terrain may result in less accurate data. Potentially, the inability to get close enough to a lynx in inaccessible terrain could also result in failed localization attempts. However, localizations of level 0 (signal not found) or 1 (VHF signal heard only from one direction) only made up 11.2% of the VHF dataset we worked with and their inclusion in the models probably would not have changed the results substantially. Moreover, localizations of level 1 typically did not represent the failure of achieving a more accurate localization for a specific target lynx due to terrain inaccessibility, but were rather taken as complementary information, when the signal of a non-target lynx was heard during localization of a target lynx.
When accounting for a potential methodological bias by including locations of accuracy level 2 in our model, the observed effects were slightly less pronounced but still significant and the direction of the observed effects did not change (Appendix A). We thereby conclude that while lynx in the past could have used open habitat at high elevations more frequently than assumed, the observed changes in habitat selection represent a biological effect and not merely a methodological bias.

| Prey spectrum
The range of the Northern chamois in Slovakia is restricted to the Tatra mountains (Anderwald et al., 2020;Corlatti et al., 2022, in press show individual behavioral plasticity in their prey choice (Holekamp & Dloniak, 2010;Oriol-Cotterill et al., 2015). Subsequently, lynx have increased the use of Alpine chamois and it became the main prey species during the most recent project period (NWA III). Thus, space use patterns of lynx in our study area may be influenced less by roe deer availability and more by chamois availability than previously assumed (Gehr et al., 2017).
Alpine chamois are more common than roe deer in our study area (LANAT, 2021;Vogt et al., 2019). The relative abundance of chamois compared to roe deer increased in the study area during the 1990s and dropped toward NWA III (cf. Figure 5: ratio chamois/roe deer:  (Linnell et al., 2021;Mengüllüoğlu et al., 2018), the replacement of roe deer as main prey species (even in areas with low roe deer densities) has rarely been reported in previous studies from Western, Central, and Eastern Europe (Haller, 1992;Moa et al., 2006;Molinari-Jobin et al., 2007;Nilsen et al., 2009;Odden et al., 2006;Podgórski et al., 2008).

| CON CLUS ION
Our study shows that reintroduced Carpathian lynx were able to adapt their habitat selection and diet to the new environment of the NWA. Compared to earlier periods after their reintroduction, lynx today increased their selection of higher elevations and open areas and changed their main prey species from roe deer to Alpine chamois. Potential drivers for the observed changes could be increased tolerance toward human presence (Basille et al., 2009;Bouyer et al., 2015;Filla et al., 2017;Gehr et al., 2017), intraspecific competition (Boyce et al., 2002;O'Neil et al., 2020), or fitness benefits from ex-

ACK N OWLED G M ENTS
We thank the Federal Office of Environment and the hunting administration of the Canton of Bern for the permits to capture and tag lynx in our study area and for their professional support. We especially thank the game wardens of the Canton of Bern for their essential help with capturing and monitoring of lynx and for monitoring of prey species in our study area. We further thank the following wildlife veterinarians of the Institute for Fish and Wildlife Health (FIWI), Bern for their participation in lynx captures: Marie-Pierre Ryser-Degiorgis, Mirjam Pewsner, Roman Meier, Iris Marti, and Simone Pisano. We also thank all involved KORA personnel, interns, civil servants, and students for their help with field work. We are also grateful for the valuable feedback from two reviewers, which greatly improved the quality of this manuscript. This study could be realized thanks to a grant from a charity foundation from Liechtenstein.

CO N FLI C T O F I NTE R E S T
None declared.

A PPEN D I X B
Model coefficients of the Null models for lynx habitat selection with and without VHF localizations of low accuracy (level 2). Factor levels of the variable 'project' were randomly assigned to the dataset. 'Forest' was the reference category for habitat types. Study period NWA III was the reference category for variable 'project'. Lynx identity was included as random effect (without level 2 locations: estimated variance com-

A PPEN D I X C
Cross-validated Spearman-rank correlations (r s ) between RSF bin ranks and area-adjusted frequencies for individual and average model sets (as in Boyce et al., 2002)