Biodiversity is highly threatened due to increasing human influence on the planet. We are currently witnessing the so-called sixth mass extinction, primarily due to rapid habitat loss and fragmentation, which is further exacerbated by climate change (Ceballos et al., 2017). Habitat alteration restricts species to small populations in isolated patches with a high rate of ecosystem decay (Chase et al., 2020). This reduces movement and gene flow between populations (Crooks et al., 2017), and ultimately elevates the risk of species extinction (Gilpin and Soulé, 1986). Halting biodiversity loss therefore requires urgent, concerted and sustained efforts to protect high-quality sites capable of sustaining viable populations, while simultaneously supporting connectivity among populations and movement of individuals across the landscape (Boyd et al., 2008).

Landscape connectivity buffers the effects of local extinction processes by facilitating the movement and effective dispersal of individuals, thus maintaining gene flow between populations, and supporting species range-shifts in response to changing climate and land-use regimes (Robillard et al., 2015). By increasing the potential for species dispersal into climatically suitable areas (Belote et al., 2017), ecological connectivity fosters resilient ecosystems that are effective in delivering ecosystem services (Mitchell et al., 2013). Consequently, connectivity has been recognized as an integral component of biodiversity conservation in international agreements, e.g. Aichi Target 11 of the United Nations Convention on Biological Diversity (CBD, 2010), the current draft of the post-2020 CBD framework (CBD, 2020), and in the European Union’s Habitats Directive and Biodiversity Strategy for 2030 (European Commission, 2020).

Surrogate approaches are frequently used in conservation planning to compensate for incomplete ecological knowledge on all species in any ecosystem (Wiens et al., 2008). Two types of surrogates used in connectivity planning are species-based (or fine-filter) approaches where one or few species are used as surrogates for several co-occurring species (Caro and O’Doherty, 1999), and habitat-based (or coarse-filter) approaches wherein landscape naturalness (Theobald et al., 2012) or land-facets (Brost and Beier, 2012) are used as surrogates for the species that inhabit them. Evaluations of coarse vs fine-filter approaches suggest significant trade-offs (Krosby et al., 2015), but overall the fine-filter approach has been found to be more effective (Meurant et al., 2018).

The species-based surrogate approach should aim to identify a suite of surrogate species, which we refer to as a “connectivity umbrella” in this paper, that encompass areas most likely to be used by several co-occurring species. Connectivity umbrellas are often chosen based on a few traits such as body mass or home range size, their charisma, conservation status (Beier et al., 2009; Wang et al., 2018), or because they have previously been identified for other purposes such as flagships for fund-raising, indicators of ecosystem health, or umbrellas for habitat management (Caro and Girling, 2010). Pre-selected connectivity umbrellas based on few criteria such as charismatic large species, have been shown to be ineffective at representing other species in multiple studies (Beier et al., 2009; Cushman and Landguth, 2012; Wang et al., 2018).

In addition to species-based surrogates being better performing than habitat based surrogates, research suggests that multiple species are better than any single surrogate species (Meurant et al., 2018; Wang et al., 2018), surrogate species are effective within their own taxonomic group (Brodie et al., 2015), and a diverse set of species reflects the needs of other co-occurring species (Cushman and Landguth, 2012). Despite the emphasis on multi-species connectivity models (Wood et al., 2022), there are only a few methods that use data-driven approaches to identify which and how many species could represent the connectivity needs of all species in any given region.

In this study, we address this research gap in connectivity science. Our goal is to develop an approach that increases the objectivity in selecting connectivity umbrella species from a pool of candidate species. We demonstrate the approach at multiple scales and evaluate the interaction of landscape characteristics and connectivity umbrella selection using the terrestrial large-mammal community in continental Europe as a case study.

For the large mammal community in Europe, we modelled potential connectivity between protected areas (PAs) and created a connectivity target representing high-probability connectivity areas for multiple species present in the region. We then ranked each species in the region using a cumulative measure that indicates how well a species represents the target and connectivity network of co-occurring species. Starting with the highest-ranked species, we sequentially added the connectivity network of all species, and plotted the contribution, i.e., percent increase of the connectivity target for each addition (Figure 1). The final connectivity umbrella comprises the combination of species that are required to reach a certain coverage of the target (set at 95% in this study), as adding more species contributed only marginally towards reaching the connectivity target. All steps of the analyses are presented in greater detail in Figure S1. All quantitative analysis was performed using R statistical software (https://www.r-project.org/) and all maps were created in ArcGIS 10.3.

The four main steps involved in this approach are: (1) Modelling connectivity for each species, (2) Generating connectivity targets per region, (3) Ranking species in each region, and finally (4) Determining the suite of umbrella species.

Using the approach presented in Figure 1, we demonstrate that a systematic and objective assessment of which and how many species are required to meet connectivity needs of multiple co-occurring species is indeed possible. In this process, we discovered some important patterns.

Multiple species are required to meet connectivity goals at both scales considered. Ninety-two percent of the biomes (Figure 2) and 83% of the ecoregions (Figure 3) had more than one connectivity umbrella species. Both the number and identity of species are scale dependent. More species are selected at the biome level (average 4.15, range 1–9 species; Figure 2) than at the ecoregion level (average 3.16, range 1–9 species; Figures 3A, S4). The identity of the species selected also varies across scales. For example, six species are selected in the temperate broadleaf mixed forest biome (Figure 3A), whereas several of the species selected in the eleven nested ecoregions (ranging from 1 to 6) are exclusive to the ecoregions, i.e., not selected in the biome-level analysis (Figure 3B).

Connectivity coverage achieved using single species is highly variable (Figure S5). For example, if only the highest-ranking species were to be selected, an average of 75% of the regional targets would be covered, ranging from ~14% by the Iberian lynx (Lynx pardinus) in Southwest Iberian Mediterranean sclerophyllous and mixed forests to ~97% by the wolf (Canis lupus) in South Apennine mixed montane forests. If only the highest-ranking species of conservation concern is selected, an average of 25% of the individual regional targets would be covered, ranging from ca. 1% by the bison in the Carpathian Mountains to 95% by the wolverine in the Scandinavian Tundra. Some ecoregions in the Scandinavian and Apennine peninsulas are exceptions to this general pattern.

Our results indicate that even when species of conservation concern are present in a region, they may not necessarily be the most suitable connectivity umbrellas. A total of seven conservation-concern species are present in 17 ecoregions (47% of ecoregions, average 1.3 conservation-concern species/region), but are selected as connectivity umbrellas in only 9 ecoregions (~53% where they are present). The number or proportion of conservation-concern species present in any region is not significantly correlated to the number or proportion of conservation-concern species in the suite of connectivity umbrella species (Pearson’s r = 0.04 and 0.26, respectively; p > 0.05 in both cases). We used Red List status at the European level because it was consistent with the European extent of our study, but threat assessments at regional scales may also be relevant. For example, none of the species selected as connectivity umbrellas in the temperate broadleaf mixed forest biome are of conservation concern at the European scale (Figure 3B), but the brown bear (Ursus arctos) and Eurasian lynx (Lynx lynx), selected at the ecoregion scale are categorized as vulnerable and endangered in the Rhodope and Dinaric Mountain ecoregions (Temple and Terry, 2007).

Of the 24 species evaluated, not even one species is an effective connectivity umbrella across its entire range (Figure 4). Even the species selected most often, the wolverine (Gulo gulo) and moose (Alces alces), were only selected as connectivity umbrellas in 60% of the regions in which they occur. While there is a significantly higher chance of being selected as an umbrella species for wider ranging species, i.e., species that occur in many regions (Pearson’s r = 0.73, p<.05), this effect is lost when comparing the occurrence of species across multiple ecoregions with the proportion of the times that it is selected as a connectivity umbrella (Pearson’s r = 0.13, p = 0.66).

We identified five unique clusters of ecoregions that ranged from very low to very high human influence (Figure S6). Mountain ranges and regions at higher latitudes are generally less fragmented and impacted by humans than regions in the plains or at lower latitudes. The maximum dispersal distance (Kruskal-Wallis chi-squared = 10.942, df = 4, p-value = 0.027), home range (chi-squared = 12.401, df = 4, p-value = 0.014), proportion of range in Europe (chi-squared = 21.014, df = 4, p-value = 0.0003), and conservation status (chi-squared = 9.9392, df = 4, p-value = 0.041) of connectivity umbrella species are significantly different between the clusters. On average, species selected as connectivity umbrellas in regions with low human influence have higher mean body mass, larger home ranges, longer dispersal distances, smaller geographic ranges, occur at lower population densities, and are of higher conservation concern than connectivity umbrellas in more human-influenced regions (Figures 5A–F). Proportionally more species are required to meet connectivity targets in regions with high human influence (Figure 5G).

Our approach increases objectivity when selecting umbrella species for connectivity conservation. It can be applied to multiple-taxonomic groups and to varying spatial scales. Unlike static target-setting in conservation prioritization exercises, our approach is flexible and can be adapted to test which species would suit different conservation goals and scenarios.

Two methods have been developed to specifically select species-based surrogates for connectivity planning. Lechner et al. (2017) use a cluster analysis on habitat requirements, connectivity elements, and gap-crossing distance thresholds to group twelve candidate mammal species into five distinct dispersal guilds. Connectivity for each guild is then modelled separately to identify which guild of species are more sensitive to habitat fragmentation. Meurant et al. (2018) present an approach to select multi-taxa surrogates from a pool of candidate species that represent a diversity of habitat needs and movement abilities. From a set of fourteen candidate species, the authors create a reference spatial prioritization map and run multiple permutations and combinations of the fourteen candidate species to identify the species subset that best retains different thresholds of the reference spatial priority map. Our approach is different from both methods. First, we include the entire terrestrial large mammal community in Europe without going through a pre-selection process as in Meurant et al. (2018), where they selected fourteen candidate species from a pool of 48 mammals, 216 birds, and 32 amphibians and reptiles based on a previous study (Albert et al., 2017). Second, we do not group species into dispersal guilds, diversity of habitat, or movement requirements. We treat each species independently. Further, instead of using a conservation concern filter in selecting the species to include in the candidate pool, we use the results to guide and test how frequently species of conservation concern are suitable connectivity umbrellas. Finally, in addition to addressing which and how many species are required for connectivity planning, we also evaluate the species-surrogates approach across two different geographic scales and assess the characteristics of connectivity umbrellas in landscapes with varying levels of human influence.

Our results re-confirm previous findings on the fine-filter surrogate approach that multiple species are better than any single surrogate species (Meurant et al., 2018; Wang et al., 2018) and a priori selection of certain groups, e.g., carnivores, may not be suitable connectivity umbrellas (Beier et al., 2009). Most suites of species selected in our analyses consist of a combination of herbivores and carnivores. Such mixed suites of species have diverse ecological attributes and dispersal abilities and are therefore more likely to accommodate the ecological requirements of diverse terrestrial mammal communities and their differing sensitivities to environmental conditions. In addition to corroborating previous findings, we show that both the number and the identity of species differ by the scale of analysis, and that species selected at finer scales are not subsets of those selected at broader scales (Figure 3).

Our approach is flexible in two ways. First, the result is a suite of species that are needed to cover 95% of the connectivity target. Depending on the regional context, it is possible to sub-select from within this suite of species, especially when species provide very little percent improvement towards the overall target. Second, the evaluation of sensitivity to different connectivity targets (Figure S3) provides a way to select species that are suitable for a diversity of conservation goals. For example, moose and wolverine in the northern regions, red deer (Cervus elaphus) and Iberian Ibex (Capra pyrenaica) in the Mediterranean forests, woodland and scrub of the Iberian Peninsula and wolf in the Apennine peninsula, red deer and red fox (Vulpes vulpes) in temperate broadleaf and mixed forest in central Europe, and the Alpine ibex in the Alps, are species selected in all three connectivity targets which were created with very different conservation goals. In general, target EQ required more species than target Q or T. This is intuitive because target EQ represents all species, target Q represents only high-quality corridors irrespective of species, and target T only represents species of conservation concern. The identity of species selected across the targets changed in their rank, but in general the top-ranking species were consistent across the different targets, and the pool of species that was most highly ranked remained similar. This redundancy in the selection of species across disparate conservation goals can be a valuable tool to incorporate more certainty about including a particular species to be an umbrella for connectivity conservation.

An important finding is the interaction between the attributes of landscape and species selected as connectivity umbrellas. The interaction between species and landscape indicates that species selected to be surrogates for connectivity conservation need to be tailored to the level of human influence in the landscape under consideration. To be effective in highly humanized landscapes, numerically more species with a certain set of attributes (widely distributed but smaller-bodied species that occur at higher population densities and disperse over shorter distances) may be required. These attributes are generally not associated with umbrella species but may be the key to developing region-specific effective connectivity plans.

The aim of our study is not to prescribe which species should be used for connectivity conservation in Europe. Rather, we use continental Europe as an example to demonstrate an approach that is quite flexible and can accommodate many different conservation and modelling choices. Due to the continental scale of analyses, we use several simplifications in our modelling to optimize computational time. We used coarse-scale (1 km) existing habitat suitability maps to generate resistance surfaces and model connectivity between PAs, but it is noteworthy that source populations may also be present in habitat patches outside of the PA network, and individual species are likely to be impacted by more and different sets of variables. Therefore, we caution the use of the species presented here to be effective and “correct” connectivity umbrellas. The species presented here can be starting points, but we recommend users to create resistance surfaces using variables and resolutions that are relevant to their region of interest. It will most likely be computationally plausible to run regional models using non-protected habitat patches as source populations at a finer resolution and include more variables to develop more realistic resistance surfaces.

Our approach can easily be extended to other systems not presented here. For example, we analyze only terrestrial mammals, but other taxonomic groups of interest can easily be used to conduct similar analyses. We define three different connectivity targets based on different conservation goals, but connectivity targets may be defined in many other ways. Finally, we use a threshold of 95% of the target as our goal (guided by our observation that an asymptote is reached around 95%), but users may decide to use a different threshold. Our approach should be applicable in any focal landscape where it is possible to generate a connectivity network for multiple species.

Overall, we demonstrate that an objective selection of surrogate species for connectivity planning is possible. Considering multiple species, not just in relation to species attributes but in relation to landscape scale and characteristics may be the key towards identifying connectivity umbrellas that better represent the needs of entire species communities. In practice, decision-making in conservation is a complex process, involving many stakeholders, and is highly contextual. When species are tied in rank and contribution to the connectivity target, managers or conservationists could select species with a higher social acceptance or other issues that are relevant locally or regionally. Rather than advocating for one species over another, we present an approach to remove the guesswork over which species should be selected as connectivity umbrellas and increase objectivity in the selection of multispecies conservation strategies.

Conservation targets, whether modest at 30% of terrestrial land (CBD, 2023), or ambitious at 50% (Dinerstein et al., 2017), are unachievable without integrating connectivity conservation, especially in rapidly changing ecosystems in the Anthropocene. The key to selecting umbrella species for connectivity conservation is to move beyond arbitrary selection of species, for the selection process to be objective, and in relation to the scale and attributes of the landscape. Only then can the overall goal of connectivity conservation be realized and be effective in restoring ecosystems and decreasing biodiversity loss.

The original contributions presented in the study are publicly available. This data can be found here: https://figshare.com/s/23831fdc52a9dcc60a36.

TD, NB, and MDB contributed to conception and design of the study. JS contributed towards the analyses and SF’s contributions helped improve the figures. TD wrote the first draft of the manuscript. All authors contributed to the article and approved the submitted version.