Large carnivore habitat suitability modelling for Romania and associated predictions for protected areas

Introduction

Long-term persistence of many large carnivore species relies on the existence of vast natural areas of core protected habitat that act as sources for the surrounding landscape (Noss et al., 1996; Soulé & Terborgh, 1999). Identifying patterns of carnivore occurrence in relation to the distribution of natural habitats and human land use can be used to inform protected area designation (Carroll, Noss & Paquet, 2001) or management (Reed & Merenlender, 2008). In Europe, carnivore habitat suitability has been quantified in Scandinavia (e.g.,  May et al., 2008) and central European countries (e.g.,  Kobler & Adamic, 2000; Huck et al., 2010). However, uncertainties over habitat suitability for species of European Community interest (e.g., large carnivores) remain prevalent in countries that have only recently joined the European Union (EU).

Many large carnivores have been extirpated from their historic habitats in Europe (Enserink & Vogel, 2006; Dalerum et al., 2009). However, carnivore decline has not been uniform and while some populations have been eradicated, others have survived and even expanded or increased (Linnell, Swenson & Andersen, 2001; Chapron et al., 2014). Currently population statuses of large carnivores vary widely among EU member states. While the brown bear (Ursus arctos Linnaeus), gray wolf (Canis lupus Linnaeus) and Eurasian lynx (Lynx lynx Linnaeus) are now making a come-back in the EU, Romania (which joined the EU in 2007) has historically housed large and stable populations of these species (Boitani, 2000; Breitenmoser et al., 2000; Swenson et al., 2000; Van Maanen et al., 2006).

Brown bears, wolves and lynx are protected in Romania by national and EU legislation, as well as international Conventions. Carnivore habitat in Romania is protected in EU-legislated Natura 2000 sites, national parks (IUCN category II) and nature parks (IUCN category V). Because Natura 2000 sites were designated based on expert opinion, accurate spatial information (e.g., GIS) and, to a lesser degree, by incorporating previous ecological modeling outputs, Natura 2000 sites might have higher habitat value for carnivores than national-level protected areas (national parks and nature parks). Historical decisions to designate national parks and nature parks mostly revolved around areas with little anticipated economic potential (such as rugged mountainous regions), and/or were spearheaded by scientists with interests in specific habitats or ecological communities. This latter approach to protected area designation has been common practice worldwide (Scott et al., 2001; Joppa & Pfaff, 2009). Investigating differences in habitat suitability across protected area types could help strategize the channeling of limited conservation resources to protected areas that host the best carnivore habitat. In addition, comparing carnivore occurrence from confirmed distribution records and habitat suitability analyses could help identify management needs for protected areas of certain types, thereby improving protection effectiveness.

In Romania carnivore research at the country-wide extent has focused on assessing distribution patterns from raw footprint tracking data (with outputs such as Figs. S1–S3) with little consideration of underlying habitat characteristics. Yet, carnivore habitat might exist outside the current extent of carnivore distribution, and habitat delineation could assist recovery efforts and range expansion. Only two quantitative studies have investigated potential habitat for brown bear, gray wolf and Eurasian lynx in Romania. Salvatori (2004) applied a Mahalanobis distance to identify environmental suitability for carnivores in the Carpathian Mountains using carnivore observation records, environmental variables and expert opinion. Van Maanen et al. (2006) used Marxan software to identify a network of potential protected areas for Romania, under the assumption that forest, grassland and shrub represent prime habitat for carnivores. Subsequently to these studies, a reassessment of the occurrence of EU-listed species in Romanian protected areas was recommended (Iojă et al., 2010).

Using an alternative design that incorporates additional information in the modelling procedure, while also contrasting candidate models in an information theoretic approach, we (1) identify habitat characteristics associated with the occurrence of the three large carnivore species present in Romania and (2) assess whether the Natura 2000 network supersedes the national (pre-Natura 2000) protected area network in relation to carnivore habitat suitability, by inspecting the predicted values of carnivore occurrence obtained at (1). Our objectives are therefore to quantitatively select a set of habitat suitability predictors that are meaningful for the carnivore occurrence dataset; and evaluate which protected area types in the current protected area system have the highest probability of carnivore presence.

Materials & Methods

Results

Brown bear habitat was predicted to occur primarily in the Central, Central-Western and Northern parts of Romania (Fig. 1). These regions coincide roughly with the Carpathian Mountains chain and their foothills. Gray wolf habitat was also predicted in these areas, but was more widely distributed throughout the country (Fig. 2). Predicted habitat for Eurasian lynx was located in the same broad geographic areas as bear habitat (Fig. 3).

Eurasian Lynx

Habitat suitability models

The top model for lynx occurrence explained >72% of the deviance (Table 2) and had excellent predictive power (mean cross-validation estimate of accuracy 0.951). For 1 unit increase in proportion of mixed forest cover, the odds of lynx occurrence increased by 177% (Table 3). As proportion of crops and orchards increased by 1 unit, the odds of lynx presence decreased by 84%. Areas with pastures had a positive non-linear association with lynx occurrence. The probability of lynx occurrence increased at a steeper slope of the fit curve when proportion pasture was higher than 0.20, than when pasture was at lower values (Fig. 4). Density of communal roads was also positively non-linearly associated with lynx occurrence, with low occurrence probability at roads densities of 0–0.25 km/km², sharply increasing odds at densities of 0.25–0.50 km/km², and maximum odds at densities greater than 0.50 km/km².

Habitat in protected areas

Similar to the other carnivores, lynx raw presence data suggested that lynx were proportionately mostly present in national parks (87%), followed by nature parks (71%) and Natura 2000 sites (42%). Mean probabilities of occurrence were also high for national parks, lower for nature parks and lowest for Natura 2000 areas (Fig. 5). As was the case for bear and wolf, habitat suitability for lynx varied between and within protected area types, with some Natura 2000 sites having better lynx habitat than the mean suitability of national parks.

Discussion

Statistical habitat models of carnivore occurrence (presence/absence) enabled us to evaluate and predict country-wide landscape responses of carnivores in Romania and habitat suitability of protected areas. The occurrences of brown bear and Eurasian lynx were best predicted by models with identical structure (Model set 4, H16) that included variables for Natural (biotic) and Human-generated habitat characteristics. The occurrence of gray wolf was best predicted by a model with variables denoting Natural (abiotic & biotic) and Human-generated features (Model set 5, H20). Our interpretations below regarding the associations between predictor variables and the response variable are based on top models, which explained most of the variation in the variable of interest (presence or absence of carnivore). This is important to emphasize because for a given predictor variable, the observed strength of its relationship with the response variable is modulated by the other predictors in the model. Basing our interpretations on outcomes of best models as identified via ranking a candidate model set, instead of on a singular global model, provided a strong foundation for inferences on carnivore habitat suitability.

Terrain ruggedness occurred in the best wolf habitat model, but not in the top bear and lynx models. Wolves showed strong selection for rugged areas, similar to wolves in Scandinavia (May et al., 2008). We highlight the fact that the scale of the analysis was carnivore home range level, and therefore our interpretations are home range level inferences. For example, while rugged terrain was an important predictor of wolf occurrence, this does not imply that wolves move through extremely rugged terrain, but that ruggedness is an important home range component for their persistence, possibly because it provides refuge from persecution. Throughout much of their range, wolves are one of the most highly persecuted carnivore, including in areas with legal protection (Liberg et al., 2012). Although evidence is limited, in Romania wolf persecution is probably widespread and has been one of the drivers for implementing applied research and conservation programs with wolf-human conflict mitigation components (Carpathian Large Carnivore Project, WOLFLIFE “Implementing best practices for the in-situ conservation of the species Canis lupus in the Eastern Carpathians”).

Irrespective of forest type, higher proportions of forest cover were associated with increased probability of occurrence for all 3 carnivores. However, not all forest types were included in top models. Conifer forest was the only forest type that occurred in top models for all 3 carnivores, but confidence intervals overlapped zero for lynx. Brown bears can exploit vegetative foods such as berries in conifer forests even the spring following autumn berry ripening (B Cristescu, 2011, unpublished data), and they also select conifer forest for bedding (Cristescu, Stenhouse & Boyce, 2013). Red deer (Cervus elaphus) have a high proportion of conifers in their winter diets (Gebert & Verheyden-Tixier, 2008) and this might explain the association of wolves with conifer forests at this time of the year, given that red deer are a major prey for wolves in European ecosystems (Okarma, 1995). The use of mixed forests by bear and lynx has been previously documented (Große, Kaczensky & Knauer, 2003; Boutros et al., 2007) and is likely associated with food distribution, such as winter-killed ungulate carcasses (Green, Mattson & Peek, 1997) or berries from previous autumn for bears; and roe deer (Capreolus capreolus) for lynx, given that roe deer are Eurasian lynx’s main prey (Jedrzejewski et al., 1993). The wolf’s association with broadleaf forest might indicate that wolves are able to exploit ungulates which are known to use broadleaf forest in spring (Bongi et al., 2008).

Occurrence of lynx in areas with low proportion of cultivated land could be because human land use could displace lynx outside the cultivation season. It is unknown whether lynx in Romania show seasonal variability in occurrence on agricultural lands, or avoid these altogether. Brown bears sometimes use these areas during the timing of agricultural production (C Domokos, 2017, unpublished data), which would explain the slight association of bears with cultivated areas we found, although modelled probabilities of bear presence in these areas were low.

Wolves and lynx occurred in areas with high proportion of pastures, a pattern possibly associated with ungulate use of pastures for feeding (Putman, 1986; Godvik et al., 2009). The year-round use of pastures by wolves has been previously reported (Meriggi et al., 1991). Habitat edges at the forest-pasture interface were not captured in our modelling but probably present good opportunities for carnivores to hunt ungulates (Podgórski et al., 2008). In Romania, bears also use pastures for foraging but at different times of the year than the timing of footprint surveys in the snow (Roellig et al., 2014).

As expected, probabilities of occurrence in relation to artificial (built) areas were low for all carnivores especially bears, for which confidence intervals for the artificial parameter estimate did not overlap zero. Built areas generally provide little natural forage, although in some areas they may provide human-originated foods that can attract carnivore and result in human-wildlife conflict situations (Cristescu et al., 2016). Regions with high proportion of built-up areas could also have low carnivore occurrence because of human-mediated carnivore removal, or because carnivores avoid these unsafe areas.

Avoidance of high national road density by brown bears is possibly indicative of an adverse response to vehicular traffic and comparable to grizzly bear avoidance of areas with major roads and traffic in Canada (Gibeau et al., 2002). In North America it is generally accepted that road development is unfavorable to grizzly bear conservation (Nielsen, Stenhouse & Boyce, 2006). Confidence interval overlap with zero for national road density in wolf and lynx occurrence models suggests a weaker response of these species to traffic at the current road density. This is in contrast with the avoidance of high road density by wolves in Poland (Jedrzejewski et al., 2005) and with extensive road networks impeding lynx movement in Germany (Kramer-Schadt et al., 2004). We caution that the ongoing and projected increase in the density and quality of transport infrastructure in Romania might surpass within-home range road tolerance thresholds for these species. Occurrence of brown bear and lynx in areas with medium or high communal road density, and lack of influence of communal roads on wolf occurrence, suggest that the first two species might be less responsive to human traffic on these small roads than wolf. Alternatively, lack of avoidance of areas with communal roads by all carnivores may be related to lower overall use of smaller roads by humans in spring, when many communal roads are still covered by snow, or are inaccessible because of heavy mud.

Based on both modelling predictions and raw footprint (presence-absence) data, the national park system in Romania has higher habitat suitability for bear, wolf and lynx than EU-level protected areas. In general, national parks have the strictest protective regulations and could therefore, in theory, perpetuate natural habitat types free of major human intervention, or even areas without human presence (e.g., core areas). Because national parks are strongholds for carnivores in Romania, managers of these areas should continue to enforce strict protective measures and also strive to maintain connectivity to other suitable carnivore habitat, that is ideally also formally protected.

For all carnivore species, nature parks had lower habitat suitability value than national parks, but higher mean suitability than Natura 2000 sites. However, there was substantial variability in habitat suitability both between and within protected area types, which would have remained undetected were we to explore carnivore occurrence from raw presence-absence data only. For example, identifying specific Natura 2000 sites with better habitat than some nature or national parks can assist conservation prioritization and site-level management for carnivore conservation. These differential outputs between raw and modelled carnivore occurrence illustrate the importance of incorporating quantitative techniques in assessments of protected area suitability for large carnivores. Our results reveal that such differences are present outside protected areas also (Figs. 1–3) and are possibly indicative of carnivore range restrictions from portions of suitable habitat due to human threats. Areas with suitable carnivore habitat but where carnivores were not tracked should be targeted by management to assist in carnivore recovery and recolonization, which are possible provided favorable management (Chapron et al., 2014).

Nature parks have fewer restrictions on human activity than national parks, with high levels of tourism. Managers of nature parks could improve protected area suitability for carnivores and minimize the risk of carnivore-human interactions by restricting tourism activities in the best carnivore habitat. Both national and nature parks have zoning, whereas zoning is not required in the case of Natura 2000 sites. An additional key difference between the three protected area categories is that national parks and nature parks fall under Romanian legislation, whereas Natura 2000 sites are under EU legislation. While we showed that habitat suitability is greatest for national parks, the observed variability in habitat suitability for large carnivores in Romania is likely to translate differentially into true conservation effects, with EU-level legislation operating in Natura 2000 sites potentially acting as a protective framework against local or national interests that might otherwise undermine protected area effectiveness. For example, a number of nationally protected areas experience threats such as illegal logging, transport infrastructure development, mining and the construction of hydroelectric power plants. In the case of Natura 2000 sites, any potential conflictual cases between site authorities and other stakeholders, such as developers or recreationists, can be solved at the European Commission level if possibilities offered by relevant national legislation have been exhausted.

Because the Natura 2000 network has incorporated many of the pre-Natura 2000 protected areas, for a total land base that is more extensive than the national and nature parks combined (Iojă et al., 2010), the relative differences in habitat suitability between Natura 2000 sites and national protected areas are likely even greater than we documented. Nonetheless, the large spatial extent of the Natura 2000 network and its EU-governed legislative framework suggest that Natura 2000 sites have potential to play a significant role in conservation of large carnivore populations in Romania and likely other European countries also. Law enforcement, environmental education of local communities and viable human-wildlife conflict mitigation solutions are necessary if Natura 2000 sites and carnivore habitat in Romania in general are to achieve higher conservation value for large carnivores and species under their umbrella.

Conclusions

We identified a variety of habitat characteristics that are associated with carnivore occurrence in Romania. Based on habitat suitability modelling, we showed that national as well as EU-legislated protected areas in Romania contain suitable habitat for large carnivores, but that the habitat values differ by carnivore species and according to protected area type. Quantitative predictions from this work could be used for border amendments of existing sites, to delineate additional protected areas, and to establish corridors for ecological connectivity (e.g., Figs. 1–3B). Even if current resources might not enable protected area expansion, the spatial outputs from habitat suitability modelling can be used to focus management for safeguarding carnivores in areas with high habitat suitability and confirmed carnivore presence (from footprint tracking data). The results can also be used to guide the spatial prioritization of sites for implementing human-carnivore conflict mitigation programs in highly suitable carnivore habitat but where carnivore presence was not confirmed with footprint tracking; as well in areas with medium suitability for carnivores but that can facilitate connectivity between carnivore strongholds. Law enforcement and working with local communities to empower them through education and by providing tools for human-wildlife conflict mitigation can pave the way to maintaining carnivore populations, carnivore re-colonization of suitable habitat, and co-existence with humans.

Supplemental Information