Wolf predation on cattle in Portugal: Assessing the ...

Biological Conservation 207 (2017) 17–26

Contents lists available at ScienceDirect

Biological Conservation journal homepage: www.elsevier.com/locate/bioc

Wolf predation on cattle in Portugal: Assessing the effects of husbandry systems Virgínia Pimenta a,b,⁎, Inês Barroso c, Luigi Boitani d, Pedro Beja a,b a

CIBIO/InBio, Centro de Investigação em Biodiversidade e Recursos Genéticos, Universidade do Porto, Campus Agrário de Vairão, Rua Padre Armando Quintas, 4485–601 Vairão, Portugal CEABN/InBio, Centro de Ecologia Aplicada “Professor Baeta Neves”, Instituto Superior de Agronomia, Universidade de Lisboa, Tapada da Ajuda, 1349-017 Lisboa, Portugal ICNF, Instituto da Conservação da Natureza e das Florestas, Avenida da República, 16, 1050-191 Lisboa, Portugal d Dipartimento di Biologia e Biotecnologie, Sapienza Università di Roma, Viale dell'Università, 32, 00185 Roma, Italy b c

a r t i c l e

i n f o

Article history: Received 9 June 2016 Received in revised form 15 December 2016 Accepted 12 January 2017 Available online xxxx Keywords: Livestock predation Farming systems Damage compensation Large carnivores Cattle Portugal

a b s t r a c t Mitigating conflicts associated with predation on livestock is essential for conserving large carnivores in human dominated landscapes. This is generally addressed by targeting at individual management practices affecting predation risk, often disregarding that different livestock husbandry systems (i.e., groups of farms sharing similar resource bases, production patterns and management practices) with different vulnerabilities to predation may coexist within predator ranges, each of which requiring tailored prescriptions to reduce predation. Here we evaluated the importance of considering both husbandry systems and individual management practices to mitigate conflicts due to cattle predation by wolves in Portugal, where attacks on cattle increased N3 times in 1999–2013. Government records from 2012 to 2013 indicated that only b 2% of cattle farms suffered wolf attacks, of which b 4% had N10 attacks per year. We found that attacks were concentrated in the free-ranging husbandry system, which was characterized by multi-owner herds, largely grazing communal land far from shelter, and seldom confined. Protecting these herds at night in winter was the most important factor reducing wolf attacks, which could be achieved by changing practices of ≈25% of farmers in this system. Attacks were much lower in the semi-confined system, probably because herds grazed pastures closer to shelter, and they were often confined with fences or in barns. Farms bringing calves b 3 months old to pastures were associated with about 90% of attacks, but changing this practice would involve ≈50% of farmers in this system. Our results underline the importance of identifying livestock husbandry systems and to adjust mitigation strategies to each system. © 2017 Elsevier Ltd. All rights reserved.

1. Introduction Livestock predation by large carnivores is one of the main causes of human-wildlife conflicts worldwide (Treves and Karanth, 2003; Treves and Bruskotter, 2014). Therefore, the effective management of conflicts is key to the conservation of large carnivores, since people perceiving economic risks from wildlife can severely hinder conservation efforts (Treves and Karanth, 2003). Although there is growing evidence that coexistence between large carnivores and humans is possible, there is still considerable uncertainty on the most effective policies and management strategies to mitigate conflicts and thus to promote such coexistence (Linnell et al., 2001; Chapron et al., 2014). The wolf (Canis lupus) is often involved in major human-wildlife conflicts due to predation on livestock (e.g. Treves et al., 2004; Gazzola et al., 2008; Iliopoulos et al., 2009; Li et al., 2013). As a ⁎ Corresponding author at: CIBIO/InBio, Centro de Investigação em Biodiversidade e Recursos Genéticos, Universidade do Porto, Campus Agrário de Vairão, Rua Padre Armando Quintas, 4485–601 Vairão, Portugal. E-mail address: [email protected] (V. Pimenta).

http://dx.doi.org/10.1016/j.biocon.2017.01.008 0006-3207/© 2017 Elsevier Ltd. All rights reserved.

consequence, there is considerable controversy over wolf conservation, particularly in landscapes where extensive livestock production is an important economic activity, and thus wolves are often legally controlled or illegally killed (Treves et al., 2004; Woodroffe and Redpath, 2015). The problem has exacerbated in recent years, in part because successful wolf conservation over the last decades has allowed its geographic expansion and thus increased the contact between wolves and livestock (Breck and Meier, 2004; Chapron et al., 2014). In this context, predation on cattle is of particular concern, given its high socio-economic value (Iliopoulos et al., 2009). Furthermore, there is widespread extensive cattle rearing, virtually without vigilance and protection measures, in areas were the wolf has been absent for a long time and is recently recolonizing, such as pastureland in the European Alps and dehesas in western Spain (Blanco and Cortés, 2009; Marucco and McIntire, 2010; Kaczensky et al., 2013). Clearly, finding solutions to mitigate wolf predation on cattle would be useful to facilitate the sharing of landscapes by wolves and humans, particularly in regions holding important wolf populations within human dominated landscapes . Compensation for damages is one of the potential tools to mitigate conflicts with large carnivores including wolves (Boitani et al., 2010;

18

V. Pimenta et al. / Biological Conservation 207 (2017) 17–26

Dickman et al., 2011). In general, governmental agencies or conservation organizations pay for animals killed by wolves, which is expected to increase tolerance towards the species (Boitani et al., 2010; Treves and Bruskotter, 2014). Despite its potential value, this system is costly and may have limited impact to improve human attitudes towards predators (Naughton-Treves et al., 2003; Schwwerdtner and Gruber, 2007; Zabel and Holm-Müller, 2008). As a consequence, other solutions have been sought, often in combination with compensation schemes, by promoting management practices that reduce predation risk (Boitani, 2000; Boitani et al., 2010; Gazzola et al., 2008). For instance, livestockguarding dogs or fencing at night are often suggested as useful methods to reduce wolf predation, and thus may help reducing the cost of compensation schemes (Linnell et al., 2012; Gehring et al., 2010; Rigg et al., 2011). To be effective, however, this strategy requires detailed identification of management practices increasing the risk of wolf attacks, and the design of alternatives that can help reducing such risk. Furthermore, they require information on how to foster the uptake of favourable practices by livestock herders, as this often involves logistic difficulties and costs of implementation (Linnell et al., 2012). The farming system approach may be valuable to understand the interactions between livestock management and wolves. The concept of farming system was developed in agricultural economics, and it is based on the idea that there are groups of farms sharing similar resource bases, production patterns and management strategies, which are likely to impact on the landscape in similar ways, and to show similar responses to biophysical conditions, as well as policy and market drivers (Dixon et al., 2001; Köbrich et al., 2003; Ribeiro et al., 2014). A key aspect of this concept is that each farming system is associated with a particular set of practices, which are selected by farmers in response to economic, biophysical and logistic constraints (Ribeiro et al., 2016). Therefore, conservationists wanting farmers to adopt more environmentally friendly practices may need to understand the farming system as a whole, rather than focusing on specific practices on an individual basis (Ribeiro et al., 2016). This is essential because some practices may be impossible to change without changing the farming system, while other practices may be more flexible and thus easier to change. In the case of livestock-wolf conflicts, therefore, it should be essential to identify livestock husbandry systems and their vulnerability to wolf

predation, and how management practices within each system affect such vulnerability. In this study we evaluate the importance of considering both husbandry systems and individual management practices to address human-wildlife conflicts involving livestock predation by large carnivores. We focused on cattle predation in Portugal, where wolves are strictly protected, feed heavily on domestic livestock and predation on cattle is among the highest documented worldwide (Álvares, 2011). To reduce conflicts, an ex post compensation scheme managed by a governmental agency has been in place since 1990, and several conservation initiatives have tried to increase livestock protection (e.g., guarding-dogs, fencing) (IEA, 2008, 2014). However, the costs of compensation have escalated in recent years, particularly due to damages on cattle, though the wolf population remained stable (Álvares et al., 2015). There is thus a need to revise the strategy adopted so far, which requires a better understanding of the factors affecting cattle vulnerability to wolves. In this study, we (i) characterize the spatial and temporal patterns of cattle predation by wolves using official records of damage compensation payments; (ii) identify cattle husbandry systems and the practices associated to each system, based on enquiries to cattle breeders; and (iii) quantify wolf predation in relation to cattle husbandry systems and individual management practices. Results were then used to identify potential solutions for reducing conflicts between cattle breeders and wolf and, more generally, to discuss the value of the farming system approach to address conflicts due to predation on livestock. 2. Material and methods 2.1. Study area The study was conducted within the wolf distribution range in Portugal, corresponding to about 20,000 km2 (40° 11′–42° 9′ N, 41° 34′–41° 50′ E; Fig. 1). The area is characterized by low to medium altitude mountains, with 85% of the territory at N400 m (average 544 m) above sea level. Land cover is mainly agricultural land (48%), forests (33%) and shrub land (17%) (IGP, 2009). Human density is relatively low, with most of the area (82%) with b50 inhabitants/km2 and 64% with b25 inhabitants/km2 (INE, 2011a). Livestock production is an

Fig. 1. Map of the study area in northern Portugal, showing the four wolf population nuclei: A – Gerês; B – Alvão; C – Bragança; D – Sul Douro. The map also shows the average annual number of cattle killed by wolf reported per parish for 2012–2013, and the location of cattle farms where enquiries to livestock breeders were conducted.


important economic activity in rural areas, with average densities of 23.4 sheep, 6.0 goats, 7.8 cows and 1.3 equids per km2 (INE, 2011b). The total number of cattle heads declined from 1999 to 2005, and it remained largely stable thereafter (Supplementary material, Fig. A1; INE, 2016). However, there was a strong increase in the number of animals per farm, corresponding to an increase in the number of farms with N 100 cattle heads, and a sharp decline in the number of farms with smaller herds (Supplementary material, Fig. A1; INE, 2016). Sheep, goats and donkeys numbers have declined from 1999 to 2013, without major changes in the number of animals per farm. The number of horses has remained fairly stable during this period (INE, 2016). The last wolf census (2003) estimated about 60 packs and 300 wolves, distributed in four main areas: Gerês, Alvão, Bragança and Sul Douro (0.5 to 3 wolves/100 km2) (Pimenta et al., 2005). This information was recently updated through the review of unpublished reports from environmental impact assessment studies, and local and regional monitoring programs, which retrieved data for about 70% of the packs recorded in 2003 and suggest that the wolf population in Portugal has remained largely stable in 2004–2013 (Álvares et al., 2015). Wild boar (Sus scrofa) is the only abundant and widespread wild prey of wolves in the region, while roe deer (Capreolus capreolus) and red deer (Cervus elaphus) occur at lower densities (b 4/km2) and have a more restricted range (Torres et al., 2015a, 2015b). Predation on livestock is widespread, and domestic species make up most of wolf diet (Álvares, 2011; Torres et al., 2015c). This justified the creation in 1990 of an ex post compensation scheme managed by the Portuguese agency responsible for biodiversity conservation (ICNF - Instituto da Conservação da Natureza e Florestas), through which wolf damages on livestock are fully compensated. Livestock breeders and their associations within the wolf range are well aware of this program, and they are entitled to make claims to ICNF when there is a putative wolf damage to livestock. All claims are verified in the field by experienced technicians, evaluating whether damages can be attributed to wolf predation, following standardized procedures (e.g. Linnell et al., 2012) described in a field handbook. Analysis of DNA extracted from saliva left in bite wounds is undertaken in more uncertain situations to confirm predator's identity (IEA, 2014). To assure uniform and correct interpretation of field evidence, the technicians have to fill in an exhaustive form for each claim, which is then processed and databased at the regional level. Once a damage is attributed to wolf predation, the compensation is paid to farmers by ICNF. 2.2. Wolf predation on cattle Data on cattle attacks by wolves were obtained from records collected through the wolf damage compensation scheme. As there is no centralised database of wolf damages, we compiled and organised information from partial databases maintained by the regional departments of ICNF. For each attack attributed to wolves, we reviewed the record to extract the data available on the date and place of the attack, the number of animals of each livestock species killed or wounded, and the amount of the compensation paid. Data was compiled for 1999, 2003 and 2009–2013 to describe general temporal trends in wolf attacks. More detailed information on the monthly distribution of attacks and the age classes of livestock predated were only compiled for 2012– 2013, when there were more comprehensive records available. 2.3. Livestock husbandry systems We characterized livestock husbandry systems from enquiries conducted to the persons in charge of cattle farms located across the wolf range, between June 2013 and November 2014. Cattle farms were defined as any exploitation associated with a herd managed as a single unit, though sometimes including animals from different owners. From a total of ≈19,000 cattle farms within the wolf range (excluding dairy cattle), 260 (≈ 1.3%) suffered wolfs attacks in 2012–2013 and were included in the scheme of compensation for damages. From

19

these, we tried to visit all farms chronically affected (i.e. N10 attacks per year), as well as farms in the same or neighbouring parishes with different attack levels (0 to b10 attacks per year). This sampling strategy was designed to minimise spatial confounding effects, because farms within the same area were exposed to similar ecological conditions, including wolf and wild prey densities, and so variation in attack levels were more likely to be due to differences in husbandry systems. Overall, we conducted 68 enquiries, which were spatially distributed in proportion to the number of cattle farms existing in the four wolf population nuclei (Supplementary material, Table A1). The interviews were made personally by two of the authors (VP, IB), following a semi-structured questionnaire. We collected information on general farm characteristics (e.g. ownership of herds and pastures, production goal, beginning of activity), as well as management practices that were expected to affect the risk of wolf attacks (e.g. herd size, main breed, and protection and vigilance practices) (Supplementary material,Table A2). Each interview lasted about 30–60 min, and we used quality control strategies to test reliability of answers, like asking the same questions in different ways and cross-checking questions with potential contradictory answers. Each farm was also characterized in terms of the mean number of wolf attacks per year in the period 2012–2013, using the records of the damage compensation scheme. We focused only on these two years to achieve a better matching between the timing of the enquiries and the period of the attacks, as farm characteristics (e.g., herd size) and management practices may vary over time. 2.4. Data analysis The livestock husbandry systems were determined by non-hierarchical clustering based on the variables collected during the enquiries (Supplementary material, Table A2), using a Gower distance matrix (Pavoine et al., 2009) and the partition around medoids (PAM) clustering algorithm (Kaufman and Rousseeuw, 1990). Variables considered unreliable or that showed little variation among farms and would thus contribute little for differentiating livestock systems were excluded from analysis (Supplementary material, Table A2). Gower distances were used because the dataset included different types of variables, and involved the computation for each variable type of a particular distance metric that works well for that type, scaling it to fall between 0 and 1, and then calculating a weighted linear combination to create the final distance matrix. We used the range-normalized Manhattan distance for quantitative variables, the Manhattan distance adjusted for ties for ordinal variables, the Jaccard coefficient for asymmetrical binary variables, and the simple matching coefficient for symmetrical binary variables, and we combined the metrics with simple averaging (Kaufman and Rousseeuw, 1990, Pavoine et al., 2009). PAM clustering was used because it is more robust to outliers than other methods and it can deal with general dissimilarity coefficients such as Gower distances (Kaufman and Rousseeuw, 1990). In PAM, representative elements of each cluster (medoids) correspond to real observations, rather than centroids or averages, and so a medoid is a representative cattle farm in a cluster, whose average dissimilarity to all the other farms in the same cluster is minimal. Silhouette plots of the k-medoid partitions with k = 2 to 10 were computed, and we selected the k with the largest overall average silhouette width (Kaufman and Rousseeuw, 1990). The ability of the original variables to discriminate between clusters was evaluated using logistic regression, which can assume the binomial or multinomial form depending on the number of clusters (Legendre and Legendre, 1998). Dummy and orthogonal polynomial coding were used for binary and ordinal variables, respectively. We used linear regression to explore the relationship of husbandry system, farm characteristics and management practices with the average number (log10-transformed) of wolf attacks per farm and per year (Legendre and Legendre, 1998). Binary and ordinal variables were coded as for logistic regression. First, we used univariate

20


regression to estimate the effect of each variable on wolf attacks. Second, we controlled for the effect of husbandry system and estimated the additional effect of each farm characteristic and management variable. Third, within each farming system we built multivariate models relating wolf attacks to farm and management variables, based on the information theoretic approach and multi-model inference (Burnham and Anderson, 2002). To avoid biases due to multicollinearity among predictor variables, we adopted the approach suggested by Cade (2015), using estimates for predictors standardized by their partial standard deviations, and estimating the relative importance of each variable within each model as the ratio of its partial standardized regression coefficient to the largest partial standardized regression coefficient (absolute values in both cases) in the model. As candidate models we used all possible subsets of predictor variables. Model ranking and Aikaike weights (wi) were based on the Aikaike Information Criteria corrected for small sample sizes (AICc). Finally, we carried out a similar analysis, but including only those variables most amenable to management manipulation. This later analysis was used to identify the practices that should be targeted by managers to achieve a reduction in wolf attack rates. To check for eventual spatial autocorrelation problems that might bias model coefficient estimates (Legendre and Legendre, 1998), we used spline correlogram plots with 95% pointwise confidence intervals calculated with 1000 bootstrap resamples (Bjørnstad and Falck, 2001). We inspected correlograms for both the raw data and model residuals, to assess whether autocorrelation was effectively removed in the modelling process. We assumed that variable selection and parameter estimation was unbiased when there was no significant autocorrelation in model residuals (Legendre and Legendre, 1998). All analysis were carried out in R x64 3.2.5. (R Development Core Team, 2016), using packages ‘cluster’ for Gower distances, PAM and silhouette plots (Maechler et al., 2016), ‘MuMIn’ for model averaging (Barton, 2016), and ‘ncf’ for correlogram analysis (Bjørnstad and Falck, 2001). 3. Results 3.1. Wolf predation on cattle The number of wolf attacks on livestock recorded and the amount of compensation paid for damages in Portugal increased since the beginning of the compensation program in 1990 until about 2001, and then remained fairly stable in 2001–2013 (Supplementary material, Fig. A2). Considering the years for which we have more detailed data, attacks on cattle increased between 1999 and 2013 in terms of numbers of animals killed, predation events, and to a much lesser extent farms affected (Fig. 2). The average number of attacks per farm increased from 1000 900

Nr of killed animals

800

Nr of predation events

700

Nr of affected farms

600 500 400 300

1.44 (1999) to 2.95 (2013) per year. The proportion of predation events in which cattle was killed also increased over time, from 12% in 1999 to 28% in 2013. In 2012–2013, wolves were responsible for ≈ 2725 attacks to all livestock species per year, and the annual compensations paid for these damages amounted to ≈820 thousand euros (ICNF, unpublished data). During this period, cattle was involved in ≈28% (761) of the annual number of attacks, and represented ≈ 17% (919) of the animals killed and ≈ 45% (≈ 380,000€) of the compensations paid. From 260 cattle farms affected annually, 83% suffered between 1 and 3 attacks per year, while only 3.8% (10) suffered ten or more attacks per year. The ten most affected farms were responsible for 35% of wolf attacks, and 43% of the compensations paid for damages on cattle. The population nucleus most affected by wolf attacks on cattle in 2012–2013 was Gerês (78.9%), followed by Alvão (12.7%), Sul Douro (8.1%), and Bragança (0,2%). Wolf attacks peaked (2012 − 2013) in April and May, while they were fairly stable at about 60 attacks per month in the rest of the year (Supplementary material, Fig. A3). From 751 cattle predation events registered in 2013, 77% involved animals b2 years old, and 33% were calves b 3 months old (Supplementary material, Fig. A4). 3.2. Livestock husbandry systems A number of variables showed little variation among the 68 farms enquired and could not be used to differentiate livestock husbandry systems (Supplementary material, Table A2). In general, most farms reared beef cattle (96%) and used extensive pastures throughout the year (90%). Most herds were not attended by shepherds, either in summer (93%) or in winter (88%), and they had no livestock-guarding dogs (88%). Three additional variables were excluded from further analysis because the answers were considered unreliable. The best partitioning of data as assessed from the silhouette width plot (Supplementary material, Fig. A5) indicated the presence of two well-defined clusters describing the main livestock husbandry systems within the wolf range in northern Portugal. One of the clusters corresponded to a free-ranging system (n = 31 farms) and the other to a semi-confined system (n = 37), which differed significantly in a number of general characteristics and management practices (Table 1). In the free-ranging system, most farms began the activity in 1980–2000, the herd was most often owned by a family or it was communal, it mostly grazed communal pastures, and the main breeds were autochthonous. The animals were very rarely confined with fences or in barns, except at night during the winter, and they often used pastures far from shelter in summer (N5 km). In the semi-confined system, most farms also began the activity in 1980–2000, but they tended to be owned by individuals, they seldom grazed on communal pastures, and they used mainly autochthonous breeds, although to a somewhat lesser extent than the free-ranging farms. The animals were most often confined with fences or in barns all year around, both day and night, and the pastures were generally closer to shelter (b 5 km), particularly in winter (b1 km). There was significant variation in the distribution of the two farming systems across the four wolf population areas (χ2 = 46.21, df = 3, p b 0.001), with the free-ranging system prevailing in Gerês (90% of visited farms), while the semi-confined system was dominant in Bragança (100%), Alvão (91.7%) and Sul Douro (89%). 3.3. Wolf predation risk

200 100 0 1997

1999

2001

2003

2005

2007

2009

2011

2013

2015

Fig. 2. Temporal variation in wolf predation on cattle in Portugal (1999–2013), in terms of numbers of farms affected, predation events and animals killed.

In univariate analysis, there was a significant tendency for more wolf attacks per farm and per year in the free-ranging (9.0 ± 16.6SD, 0–89) than in the semi-confined husbandry system (2.4 ± 7.2SD, 0–43) (Table 2). Also, there were more attacks on herds with multiple owners (family/communal herds), on farms that started activity in 1980–2000, on large herds, on herds that were not confined in fences or barns, on


21

Table 1 Variation in characteristics of the two main livestock husbandry systems identified in 68 cattle farms located within the wolf range in northern Portugal. For each variable we indicate the values for the representative farm (medoid; M) of each system. Variables discriminating significantly (P b 0.05) between clusters as assessed from logistic regression are highlighted in light grey.

Free-ranging

Semi-confined

(n = 31)

(n = 37)

Individual

22.6%

78.4% (M)

Multiple

77.4% (M)

21.6%

2000

35.5%

21.6%

Never

3.2%

Partly Always

Z

P

-4.297

5 km (n= 17)

Distance of usual winter pastures to main shelter

Ordinal

1 – < 1 km (n= 32) 2 – 1-5 km (n= 31) 3 - > 5 km (n= 5)

Number of months/year with extensive grazing a

Binary asymmetric

Presence of shepherd in Summer a

Ordinal

1 – 12 (n=61) 2 – < 12 (n=7) 1 – no (n=63) 2 – temporary (n=4)

Extensive grazing year around was expected to increase exposure to wolf attacks. Herds guarded by shepherds were expected to be less vulnerable to wolf predation

3 – permanent (n=1) Presence of shepherd in Winter a

Ordinal

1 – no (n=60) 2 – temporary (n=3) 3 – permanent (n=5)

Presence of livestock guarding dog a

Birth periods

b

Binary asymmetric

0 – no (n=60) 1 – yes (n=8)

Categorical

A - All year around (n=50) B - (All year around but) more frequent in Spring/Summer (n=14) C - (All year around but) more frequent in Autumn/Winter (n=4)

Carcasses collected by governmental sanitary agency b

Binary asymmetric

0 – not always (n=14) 1 – always (n=54)

Herds guarded by livestock guarding dog were expected to be less vulnerable to wolf predation Herds with birth periods not controlled were expected to be more vulnerable to wolf predation given the higher vulnerability of new born calves.

Leaving carcasses in the field was expected to increase predation risk as these can serve as an attractant for wolves

a

The variable was excluded from statistical analysis because of little variation.

b

The variable was excluded because answers to these questions were perceived to be unreliable.

3

a) 450000

16,00

400000

14,00 12,00

350000 10,00 300000 8,00 Nr of animals

250000

6,00 Nr of animals per farm

Animals per farm (average)

Total nr of animals

Nr of animals

200000 4,00 1997 1999 2001 2003 2005 2007 2009 2011 2013 2015

b) 60000

900

55000

800

50000

700

45000

600

40000

500

35000

400

30000

300

25000

100 animals

100

Nr of farms > 100 animals

Nr of farms < 100 animals

Nr of cattle farms

15000 0 1997 1999 2001 2003 2005 2007 2009 2011 2013 2015

Figure A1. Temporal variation (1997-2015) of cattle numbers (except dairy cattle) in the Portuguese wolf range: (a) total number of animals and average number of animals per farm; and (b) number of cattle farms with more and less than 100 animals. Source: www.ine.pt (accessed May, 2016).

4

N

Number of Wolf Attacks

Compensation Value

3000

€ 900000 800000

2500 700000 2000

600000 500000

1500 400000 1000

300000 200000

500 100000 0

0 90 91 92 93 94 95 96 97 98 99 00 01 02 03 04 05 06 07 08 09 10 11 12 13

Figure A2. Temporal variation in the number of wolf attacks on livestock (all species) reported to the national conservation authority, and the amount of compensations paid for damages, since the outset of the compensation scheme (1990) until 2013. Source: ICNF, unpublished data.

5

100

Mena nr of attacks

80 60 40 20 0

1

2

3

4

5

6

7

8

9

10

11

12

Month

Figure A3. Monthly variation in the mean number of wolf attacks on cattle (2012-2013) in Portugal.

6

Percentage of animals killed 35 30 25 20 %

15 10 5 0 ≤3m

3 m - ≤ 8 m 8 m - 2 yrs 2 - 14 yrs Age class

> 14 yrs

Figure A4. Frequency distribution of cattle killed or wounded by wolves (2013) per age class in Portugal (n=751).

7

0.32 0.30 0.28 0.26 0.24 0.20

0.22

Silhouette Width

2

4

6

8

10

Number of clusters

Figure A5. Silhouette width plot of the k-medoid partitions with k = 2 to 10 used to estimate the best number of clusters to describe livestock husbandry systems within the wolf range in northern Portugal (see the main text for details).

8

-0.5

0.0

Correlation

0.0 -1.0

-1.0

-0.5

Correlation

0.5

0.5

1.0

1.0

a)

0

0

50000

100000

150000

100000

150000

200000

Distance

b)

Distance

d)

0.0 -1.0

-1.0

-0.5

-0.5

0.0

Correlation

0.5

0.5

1.0

1.0

c)

Correlation

50000

200000

0

50000

100000

150000

200000

Distance

0

50000

100000

150000

200000

Distance

Figure A6. Spatial correlogram plots with 95% pointwise confidence intervals calculated with 1000 bootstrap resamples, depicting spatial dependencies in (a) wolf attacks per farms (as log10), and the residuals of models relating wolf attacks (as log10) to (b) livestock husbandry systems, (c) livestock husbandry systems and herd size, and (d) livestock husbandry systems and use of communal pastures. Models are provided in Table 2 of the main text.

9

a)

b)

c)

d)

e)

f)

Figure A7. Spatial correlogram plots with 95% pointwise confidence intervals calculated with 1000 bootstrap resamples, depicting spatial dependencies in wolf attacks per farms (as log10) in the freeranging (a,c,e) and semi-confined (b,d,f) systems, based on the raw data (a,b), the residuals of average models with all variables (c,d), and the residuals of average models with management variables (e,f). Models are provided in Table 3 of the main text. 10