Journal of Applied Ecology 2014, 51, 42–52
doi: 10.1111/1365-2664.12176
Living in risky landscapes: delineating management units in multithreat environments for effective species conservation s Pedro P. Olea* and Patricia Mateo-Toma Instituto de Investigacion en Recursos Cinegeticos (IREC), CSIC-UCLM-JCCM, Ronda de Toledo s/n, Ciudad Real, 13071, Spain
Summary 1. Managing threatened species to reduce their extinction risk is a widely used, yet challeng-
ing, means of halting biodiversity loss. Species show complex spatial patterns of extinction risk, due to spatial variation in both threats and vulnerability across their ranges. Conservation practitioners, however, rarely consider this spatial variation and routinely apply uniform conservation schemes, either throughout the species’ ranges, or following administrative borders that do not match ecological boundaries. Most of these schemes are experience-based (e.g. expert opinion) and thus difficult to replicate. 2. We accounted for spatial variation in species’ threats by using multivariate techniques [i.e. cluster analyses and multidimensional scaling (MDS)] to delineate management units for more effective conservation. We grouped breeding territories of the endangered Egyptian vulture, according to interterritory similarity in presence and intensity of their threats. 3. The first three MDS axes explained 62% of the data variation. The first axis separated territories in protected areas, with low human presence, but high risk of illegal poisoning from areas highly dominated by humans. The second axis classified territories regarding the density of sheep/goats and griffon vultures and the presence of wind farms. The third axis confronted territories in protected areas with those in unprotected areas with wind farms. 4. We obtained 18 statistically supported groups (i.e. management units) including 86% of the territories. Territories within the same group were geographically close, agreeing with the underlying spatial autocorrelation of threats. However, six groups (33%) were distributed over more than one administrative region, which will require inter-regional coordination for cost-effective conservation. 5. Synthesis and applications. Our results show wide spatial variation for species’ threats and suggest incorporation of this heterogeneity into conservation schemes. We demonstrate how multivariate statistics, coupled with uncertainty analysis, can be employed in a systematic and repeatable way to deal with the heterogeneous landscapes of risk that species face across their ranges. Our approach allows researchers and managers to delineate management units according to similarity in species’ threats for any targeted organization level (e.g. individuals, territories, populations). The results can be visualized in Euclidean and geographical spaces for better interpretation, allowing managers to design more effective conservation actions. Key-words: biodiversity conservation, Egyptian vulture, evidence-based conservation, habitat heterogeneity, multivariate analysis, Neophron percnopterus, threatened species, uncertainty, wildlife management Introduction One widespread way to face the loss of biodiversity is to manage declining populations of threatened species to *Correspondence author. E-mail:
[email protected]
reduce their extinction risk (Caughley & Gunn 1996; Norris 2004; Wilcove 2010). To develop effective conservation practices for these species, it is essential to know what threats they face (Wilcove 2010) and where those threats occur. The former question is widely recognized and many attempts have been addressed world-wide to
© 2013 The Authors. Journal of Applied Ecology © 2013 British Ecological Society
Managing threat heterogeneity for conservation ascertain what factors jeopardize species conservation (Caughley 1994; Wilcove 2010). However, the spatial location or distribution of species’ threats has received far less attention and its knowledge is still rudimentary (but see Garthe & H€ uppop 2004; Mateo-Tomas & Olea 2010a; Mateo-Tomas et al. 2012). Growing evidence shows that the risk of extinction is unevenly distributed within a species’ range (Yackulic, Sanderson & Uriarte 2011). This is because species face different types and intensities of threat across their ranges (Wilcove 2010; Yackulic, Sanderson & Uriarte 2011) and exhibit intraspecific variation in their ability to tolerate these threats (i.e. vulnerability; Isaac & Cowlishaw 2004; Cowlishaw, Pettifor & Isaac 2009; Yackulic, Sanderson & Uriarte 2011). This intraspecific vulnerability varies among threats (Isaac & Cowlishaw 2004) and varies within a single threat across the species’ range, probably due to underlying environmental variability (Cowlishaw, Pettifor & Isaac 2009; Yackulic, Sanderson & Uriarte 2011). As a result, threats and vulnerability combine to create a wide spatial variation in extinction risk (Yackulic, Sanderson & Uriarte 2011). Consistent with this are recent studies that indicate an intraspecific regional variation in wildlife–habitat relationships (Whittingham et al. 2007; McAlpine et al. 2008). Further evidence comes from study cases that show that, within a single species, the threats highlighted in a site differ from threats reported
43
elsewhere (e.g. Egyptian vulture Neophron percnopterus L., Mateo-Tom as & Olea 2009, 2010b). Any species faces, therefore, a complex landscape of risk formed by the combination of different types and intensities of threats, coupled with a variation in vulnerability across its range. Besides the ecological relevance of knowing patterns and processes of this issue, the spatial variation in the response of individual species to threats has profound implications in conservation planning (McAlpine et al. 2008). For example, it precludes policy-makers and conservation managers from adopting a uniform conservation programme for individual species, particularly for those with large geographical ranges (Whittingham et al. 2007; McAlpine et al. 2008). Conservation managers and scientists, however, rarely recognize, or incorporate, this heterogeneity into species conservation plans and routinely apply uniform management schemes over national or international areas based on evidence raised from studies performed in small geographical areas (Whittingham et al. 2007; Mateo-Tom as & Olea 2010b). On many occasions, the management applied is based on artificial discontinuities, such as administrative borders (Kark et al. 2009; Fig. 1), where each region independently manages a fraction of the species’ population within their boundaries. While this boundary-driven management is widespread for practical purposes, it could be highly inefficient because administrative borders rarely coincide with
Fig. 1. Distribution of territories (points) of the endangered Egyptian vulture in northwest Spain. This vulture population distributes over three administrative regions (Castilla y Le on, Asturias and Cantabria; their borders are represented by thick black lines), which independently manage the species territories within their borders. © 2013 The Authors. Journal of Applied Ecology © 2013 British Ecological Society, Journal of Applied Ecology, 51, 42–52
44 P. P. Olea & P. Mateo-Tom as natural ecological boundaries (L opez-Hoffman et al. 2010) or conservation needs. Conservation programmes that do not match ecological reality reduce the efficacy and efficiency of conservation. Moreover, coordinated conservation actions across sociopolitical boundaries are expected to be economically more efficient than uncoordinated plans (Kark et al. 2009). An added complication is that much of the species’ management is experience-based (e.g. existing management plans, expert opinion) and thus difficult to replicate (Pullin & Knight 2005). One step towards improving conservation planning of threatened species is to identify the spatial variation in attributes (e.g. threat type and intensity) that the different species’ organization levels (i.e. populations, colonies, breeding pairs or individuals) face throughout their geographical ranges. This knowledge would enable conservation practitioners to better match the management actions required by the target species at each location (Ferrier & Wintle 2009). The best organization level to effectively undertake this task will depend on factors such as the target species, the extent of its range, its management needs and the available resources. Nonetheless, there will be elements at any species’ organization level sharing similar attributes (e.g. various territories sharing the same threats) that can be grouped within the same management unit. This type of clustering process should maximize the homogeneity in attributes within management units while maximizing the differences among them. In this study, we apply multivariate classification and ordination techniques to delineate management units in terms of the similarity of their attributes (i.e. threat type and intensity), at the territory level, for the globally endangered Egyptian vulture. The approach we develop is statistically based and undertaken in a systematic manner, which enhances the transparency and repeatability of conservation practise for threatened species. It also incorporates uncertainty into the classification process allowing one to assess the robustness of the created units. The resulting management units are projected in Euclidean and geographical space to improve visualization and thus facilitate their use in effective conservation planning.
Materials and methods STUDY AREA
The study area covers 24 639 km2 in north-west Spain, corresponding mainly to the Cantabrian Mountains (Fig. 1). This region is recognized for its high biodiversity and contains several protected areas (10 biosphere reserves, 18 sites within the Natura 2000 network of protected areas of the European Union, one national park and eight regional/natural parks). The landscape consists of a wide variety of habitats, including oak Quercus spp. and beech Fagus sylvatica woodlands, scrublands and pastures. The area supports several species of conservation concern – according to IUCN criteria – such as the critically endangered Cantabrian brown bear Ursus arctos (Kaczensky et al. 2013) and
the globally endangered Egyptian vulture (BirdLife International 2012a). Human population density is low (mean: 67 inhabitants km 2), especially in the south (27 inhabitants km 2; INE 2012). Extensive livestock rearing (mainly of cows) is the main activity in most of the area, including the protected areas. Livestock stay in the field for several months per year, depending on snow presence. A decreasing number of sheep flocks that are moved seasonally between pastures (i.e. transhumant) are still using some (sub)alpine pastures during the summer (Olea & Mateo-Tomas 2009).
STUDY SPECIES
The Egyptian vulture is a medium-sized, territorial scavenger distributed from the Mediterranean countries to India and South Africa. The species breeds on cliffs and usually uses the same nest, year after year, actively defending the nesting territory (Mateo & Olea 2007). The Egyptian vulture is listed as Endangered by the IUCN (BirdLife International 2012a). As much as 40% of the European breeding population is found in Spain (Del Moral 2009). The population of the study area is estimated to be 175 breeding pairs (i.e. 13–14% of the Spanish population); concern for conservation of vultures in this area has been previously highlighted (Mateo-Tomas, Olea & Fombellida 2010). The management of this vulture population depends on three autonomous regions (Asturias, Cantabria and Castilla y Le on), which independently manage the vultures present within their borders (Fig. 1).
DATA ACQUISITION
Our data set consisted of eight variables (Table 1) recorded in 169 Egyptian vulture territories (i.e. 966% of the total estimated regional population) (Fig. 1). We defined a territory as a site potentially occupied by an Egyptian vulture breeding pair, that is, a site once known to be occupied or to have shown some evidence of possible occupancy by the species. Information on territories was obtained from the review of previous censuses (Perea, Morales & Velasco 1990; Jubete 1997; Del Moral & Martı 2002; Picos de Europa National Park, F. Jubete [Fundaci on Global Nature] and J. Placer [2nd Egyptian Vulture National Census, SEO/BirdLife], pers. com.) and field surveys conducted during the 2005–2008 breeding seasons (Mateo-Tomas & Olea 2009; Olea & Mateo-Tomas 2011). All the variables were measured at each territory within a 25 km radius around the nest (known or suspected) according to half the average nearest-neighbour distance (NND/2) and habitat selection models for the species in the study area (Mateo-Tomas & Olea 2009, 2010b). Environmental variables measured around territories described potential threats, known or suspected, affecting Egyptian vultures (Table 1). We measured the density of sheep and/ or goats and cows, since livestock – especially sheep and goats – are a key food source for the Egyptian vulture (Cabrera-Garcıa, Mateo-Tomas & Olea 2012) and livestock decline is considered to be one of the main threats to European populations of Egyptian vultures (I~ nigo et al. 2008; Mateo-Tomas & Olea 2010b). We also considered other anthropogenic threats (e.g. poison, wind farms, human presence; Carrete et al. 2009; Hernandez & Margalida 2009; Mateo-Tomas et al. 2012), together with biological threats (e.g. interspecific competition with griffon vulture Gyps fulvus L.; Meretsky & Mannan 1999; P. Mateo-Tomas & P.P. Olea unpublished data).
© 2013 The Authors. Journal of Applied Ecology © 2013 British Ecological Society, Journal of Applied Ecology, 51, 42–52
Managing threat heterogeneity for conservation
45
Table 1. Main potential threats for the Egyptian vulture in the study area. All the variables but wind farms were measured at the territory level (i.e. 25 km radius around the nest) Variable Name Sheep (SHP) Cow (COW) Griffon (GFF) Parks (PRK)
People (PPL) Roads (RDS) Wind farms (WDF) Poison (POI)
Description
Hypotheses
Density of sheep and goats (animals km 2)*,§ Median=121, range: 09–601 Density of cows (animals km 2)*,§ Median = 172, range: 20–809
Livestock is an important food source for the Egyptian vulture
Density of griffon vulture breeding pairs (pairs km 2; Mateo-Tomas & Olea 2010a, 2011) Median = 002, range: 000–029 Degree of environmental protection of the area containing the territory. (0: non-protected, 1: locally protected, 2: regionally protected, 3: nationally protected, 4: more than one protected area overlapping)* Density of people (inhabitants km 2)†,§ Median = 86, range: 22– 6893
Griffon vultures mostly out-compete Egyptian vultures for food and breeding places Protected areas provide additional protection to wildlife through higher surveillance and implementation of specific conservation actions Human presence can increase disturbance to breeding territories but can also provide predictable sources of food
Density of roads (km km 2)*,‡ Median = 06, range: 00–31 Density of wind turbines within 25 km radius (Mateo-Tomas & Olea 2010a; Mateo-Tomas, Olea & Fombellida 2010)
Collision with wind turbines is an important cause of death for many bird species, especially large raptors
Maximum risk of illegal poison use based on a predictive map of 1 km2 pixel resolution (Mateo-Tomas et al. 2012) Median = 06, range: 01–08
Vultures (including Egyptian vultures) are frequently poisoned by feeding on illegal baits targeted at terrestrial predators of livestock and game
*Gobierno de Cantabria, Junta de Castilla y Le on and Principado de Asturias. † National Institute of Statistics (INE). ‡ ~ PNOA ©INSTITUTO GEOGRAFICO NACIONAL DE ESPANA. § Data obtained at municipality level.
STATISTICAL ANALYSIS
We created a territory-by-variable table, comprised of 169 rows (territories) and eight columns (variables), describing the territories. This table was converted into a two-dimensional matrix of dissimilarity, or distances, that were calculated for each pair of rows (i.e. territories). We used the Euclidean distance as a measure of association or similarity. This measure is appropriate for our data set, as it comprises environmental variables in which the presence of double zeros, when calculating the pairwise distances, contributes to similarity (Zuur, Ieno & Smith 2007). Euclidean distances were computed on standardized variables (z-scores, i.e. transformed to mean 0 and variance 1). We used classical multidimensional scaling (cMDS; also called principal coordinates analysis, PCoA) to graphically represent the interterritory distances as a plot formed by their best two or three dimensions (or principal axes). Variables were then overlaid on this plot using regression coefficients obtained from general linear models of the variables, onto the dimensions of the MDS (Greenacre 2010). The overlaid variables aid interpretation of the objects (territories) on the plot. Running a cMDS with Euclidean distance is equivalent to running a principal components analysis (PCA). However, cMDS is more flexible than PCA because the former can use more measures of association and thereby enables inclusion of different types of variables (e.g. continuous, ordinal, categorical, binomial; Legendre & Legendre 1998). Non-metric multidimensional scale (NMDS) also can be used as an alternative to cMDS or PCA because both multidimensional methods give similar results (Zuur, Ieno & Smith 2007), so either would be a good choice to represent interpoint distances. Nonetheless, unlike NMDS, cMDS allows one to obtain a more accurate
quantification of the interpoint distances (i.e. degree of similarity) in the plot (Zuur, Ieno & Smith 2007). To describe the spatial structure of the threats potentially affecting Egyptian vulture territories, we computed multivariate spatial correlation by performing a Mantel correlogram (i.e. a plot of spatial correlation values against geographical distance classes; Legendre & Legendre 1998). This correlogram measured correlation between the matrix of similarity, or distance among territories, and the matrix of geographical distances. Spatial correlation coefficients were tested for significance using 1000 permutations for each distance class after the data were detrended; P-values were corrected for multiple testing by applying Holm’s procedure (Oksanen et al. 2012). To delimit groups of territories, we performed cluster analysis on the dissimilarity matrix by applying hierarchical classification methods using a set of different algorithms: single and complete linkage agglomerative, unweighted and weighted average agglomerative (UPGMA and WPGMA, respectively), unweighted and weighted centroid agglomerative (UPGMC and WPGMC, respectively), and Ward’s minimum variance (see Legendre & Legendre 1998). We compared the different topologies resulting from these classification methods. To do this, we computed the cophenetic correlation (Legendre & Legendre 1998) as a measure of how well the produced dendrogram preserves the pairwise distances, that is, it measures the strength of the relationship between the pairwise distances of territories in the dendrogram and those of the dissimilarity matrix. The cophenetic correlation coefficient takes values between 1 and +1 (Legendre & Legendre 1998). The higher the absolute value of the coefficient, the better the clustering model produced. We assessed, also, the uncertainty in the cluster analysis by computing P-values for each cluster, via
© 2013 The Authors. Journal of Applied Ecology © 2013 British Ecological Society, Journal of Applied Ecology, 51, 42–52
46 P. P. Olea & P. Mateo-Tom as bootstrap resampling, using the ‘pvclust’ R package (Suzuki & Shimodaira 2011). P-values indicate how strongly a given cluster is supported by the data. We used approximately unbiased (AU) P-values which are computed by multiscale bootstrap resampling, since it is more accurate than an unbiased P-value bootstrap probability (BP) – which is computed by using normal bootstrap resampling (Suzuki & Shimodaira 2011). We considered clusters stable and strongly supported by data if they had AU P-values greater than or equal to 99% (Suzuki & Shimodaira 2011). Groups of territories generated in the cluster analysis were then projected in both Euclidean and geographical spaces. We also categorized the intensity of each variable at each one of the groups resulting from the cluster analysis. The values taken by each considered variable at the territories were classified according to the 10th, 25th, 50th, 75th and 90th percentiles. If the values of a variable at all territories within a group were ≤10th percentile, the intensity of that variable, at that group, was classified as ‘Very Low’. Values >10th and ≤25th were classified as ‘Low’, >25th and ≤75th as ‘Medium’, >75th and ≤90th as ‘High’, and >90th as ‘Very High’ (Table 2). All multivariate analyses were conducted in R version 2.15.2 (R Development Core Team 2012), using the packages ‘vegan’ (Oksanen et al. 2012), ‘MASS’ (Venables & Ripley 2002) ‘cluster’ (Maechler et al. 2012), ‘rgl’ (Adler & Murdoch 2012) and ‘pvclust’ (Suzuki & Shimodaira 2011).
Results The first two axes of the cMDS (Fig. 2) accounted for nearly 50% of the variation in the data (first axis: 298%;
second axis: 195%); the addition of a third axis (123%) increased it to 62%. Figure 2a displays a gradient on the first axis running from territories located in protected areas with high risk of illegal poison use and low human disturbance, in terms of density of inhabitants and roads (e.g. territories 1, 19, 40, 61), to territories located in human-dominated, unprotected areas with high densities of cows and under very low risk of illegal poisoning (e.g. 108, 121, 126, 151). Figure 2 also shows a positive correlation between density of roads, people and cows. These variables were negatively correlated with the presence of protected areas and risk of illegal poison use, which in turn, covary with each other. A second gradient was created, on the second axis, from territories characterized by high densities of sheep and/or goats and griffon vultures, to territories with low presence of these species but a high presence of wind farms. The third axis separated territories in protected areas from those located in unprotected areas with wind farms (Fig 2b,c). The three dimensions (i.e. axes) of the cMDS explained between 34% (cows) and 84% (sheep/goats) of the variation in each variable. While many territories in the three regions appeared mixed in Euclidean space, the first axis seems to separate the territories in Asturias from those in Castilla y Le on, suggesting some kind of spatial correlation (Fig. 2). Indeed, the threat-based environmental similarity of Egyptian vulture territories exhibited positive spatial correlation up to c. 20 km (i.e. statistical significance in the first
Table 2. Example of how the outputs of our approach can be effectively communicated to managers
Intensity of the considered threats at all the territories within each group according to the percentiles 10th, 25th, 75th and 90th. Very Low: ≤10th; Low: >10th and ≤25th, Medium: >25th and ≤75th, High: >75th and ≤90th, Very High: >90th. Low–Medium and Medium– High indicate that the value of that variable ranges within the two categories. ‘>‘ or ‘