Wolf (Canis lupus) populations have recovered and expanded across many parts of the world thanks to conservation efforts, including improved legal status and restoration of their prey. Concurrently, public concerns regarding the risk of wolf attacks on humans and livestock are increasing as wolves occupy human-dominated landscapes. We examined a unique case in Europe allegedly involving wolves in the death of a female British tourist, aged 64, in northern Greece in September 2017. This incident received extensive international media attention and yet many fundamental details of the case area are lacking, including whether local livestock guarding dogs played a role. To assist in resolving the case, we conducted an extensive literature review which documented 13 criteria linked to the risk of either a wolf and/or a dog attacking a human. We also conducted a camera trap survey (October to December 2017) soon after the fatal attack to calculate the activity overlap among humans, dogs and wolves. Sufficient data were available for assessing 11 of the 13 criteria. For the remaining two, the required data were either not analysed (i.e. canid DNA collected from the attack site), not appropriately collected (i.e. DNA from the mouths of suspected dogs) or were collected, but misinterpreted (i.e. the post-consumption patterns of the victim’s corpse). Via this combination of evidence, we conclude that this case involved a fatal dog attack. This assertion is supported by evidence such as the: a) high dog-human activity overlap at the attack site which peaked during the attack time as opposed to near zero wolf-human activity overlap at the same time, b) presence of a large pack of unsupervised dogs, c) high ratio of male dogs in the dog pack, d) close vicinity of the attack site to dog owner’s property and e) previous documented aggression of these dogs towards humans. The consumption patterns, time scale and location of the victim’s remains indicate a posthumous consumption of the corpse possibly by the same dogs and/or by wild scavengers including wolves. A multidisciplinary approach, such as this one, in the assessment of putative wildlife attacks on humans can reduce misidentifications of the responsible species by forensic authorities and, therefore, prevent unfounded decrease in public tolerance for large carnivores.

animal attacks on humans, forensic analysis, large carnivores, livestock guarding dogs, multidisciplinary approach, wolf

The wolf (Canis lupus L.) has expanded across many parts of the world over the last three decades, thanks to conservation efforts related to its legal protection and restoration of its habitats (Chapron et al. 2014; Cimatti et al. 2021). This recovery of wolf populations, although often heralded as important for restoring ecological processes (Ripple and Beschta 2012; Boyce 2018), is not without risk, including the potential of increasing human-wildlife conflict due to wolf depredation of livestock and game animals (Janeiro-Otero et al. 2020). Moreover, concerns about the risk that wolves pose to human safety are gaining public attention (Linnell et al. 2003), as wolves increasingly occupy human-dominated landscapes (Kuijper et al. 2019).

Despite the fact that there are more than 17,000 wolves in Europe (Boitani 2018) and 75,000 in north America, with presence also in landscapes which are home to millions of people, reports of wolves attacking humans since the early 20^th century are rare (Penteriani et al. 2016; Linnell et al. 2021). Wolf attacks resulting in human injuries or fatalities have been reported in published literature, unpublished/historical reports and the media (Linnell et al. 2002; Linnell et al. 2021). In general, wolf attacks on humans can be categorised as: 1) attacks by unhealthy/injured wolves, 2) provoked or defensive attacks and 3) predatory attacks (Linnell et al. 2021). Attacks by unhealthy wolves have been reported mainly in areas with rabies prevalence in wildlife, like India (Isloor et al. 2014), China (Wang et al. 2014), Iran (Gholami et al. 2014), Turkey (Turkmen et al. 2012; Ambarli 2019), Russia (Sidorov et al. 2010), as well as other countries from the Middle East, Eurasia and Asia (Linnell et al. 2021). Provoked wolf attacks on humans are rarer and have involved cases where wolves injured humans in defence of their life, prey or offspring (Linnell et al. 2021). Predatory attacks were mainly reported from areas with low natural wolf prey availability, wolf habituation with anthropogenic food sources, such as livestock, offal remains and garbage, high human density in rural settings and absence of firearms (McNay 2002; Löe and Röskaft 2004; Lescureux and Linnell 2014). Such conditions are mostly found in the Middle East and Asia, including India (Jhala and Sharma 1997; Rajpurohit 1999), Iran (Behdarvand et al. 2014; Behdarvand and Kaboli 2015) and Israel (Linnell et al. 2021). Predatory attacks on humans by healthy wolves have been also reported in Europe and North America in the 20^th and 21^st century (Linnell et al. 2002; McNay 2002; McNay and Mooney 2005; McNay 2007; Butler et al. 2011; Penteriani et al. 2017; Linnell et al. 2021; Nowak et al. 2021). However, in the last 40 years, since scientific studies on wolves have been carried out, only two people have been killed by wolves in North America, while in western Europe (excluding Russia and some neighbouring countries where rabies is still prevalent), no wolf predatory attack on humans has been verified (Linnell et al. 2021).

As reports of wolf attacks on humans and increased livestock depredation may impede species conservation efforts (Linnell et al. 2021), great care is needed in assessing attacks putatively attributed to the species. Reports of wild predator attacks on humans and livestock – especially when receiving wide media coverage – have the potential to disproportionately decrease public tolerance towards those species, with broader repercussions for wildlife conservation (Kansky and Knight 2014; Hoffmann et al. 2017; Penteriani et al. 2017; Bombieri et al. 2018). Unsubstantiated incidents of human-wolf conflict, in particular involving human injuries or fatalities, can increase public fear of wolves. Fear has been used by individuals or interest groups to promote public dislike for wolves and to reduce the legislative protection afforded to the species (Linnell and Alleau 2015).

We examine here the 2017 case study of an alleged wolf involvement in the fatal attack of a female British tourist, aged 64, in northern Greece, because of the potential ramifications it could have for wolf public perception across Europe, especially given the extensive media attention it received both nationally and internationally (Arbieu et al. 2021). Specifically, we tested the hypothesis that wolves were indeed responsible for the fatality by examining factors associated with: a) wolf and dog attacks on humans reported in literature, b) relying on case-related evidence from official reports and c) our own field investigations, initiated soon after the attack, regarding wolf and dog occurrence and activity patterns at the attack site and adjacent areas. The criteria and assessment protocols proposed and used may help reduce the chance of predator misidentification in future human attack cases.

According to the official police report to the District Attorney, on 21 September 2017, the victim called her family in Britain at about 17:00 h Greek local time (GMT+2), which was still daylight, to report that she was being “attacked by fierce dogs” along the Petrota-Maroneia section of the E6 European long distance tourist trail in Rhodope Prefecture, northern Greece. Local witnesses, including two Austrian tourists whose written report to WWF Greece was shared to us with their consent, confirmed having seen her one hour earlier (4 pm) leaving Petrota Beach (~ 3.7 km from the attack site) heading on foot towards Maroneia Village along the car-accessible dirt/forest road which is part of the E6 European long-distance hiking trail (Figs 1–2). In response, the fire brigade initiated, that same evening, a missing person search in the area, but returned without success. The next morning (22 September 2017), the search resumed with additional personnel from the local police department, a special search and rescue unit of the fire brigade and local volunteers, continuing through the night and beyond. Eventually, using tracking dogs, some personal belongings (incl. passport, clothing) and bodily remains of the victim were discovered on the morning of 23 September 2017, approximately 36 hours after the initial distress call. The collection of additional remains concluded 72 h following the attack. The remains were dispersed along an approximately 15,000 m² area adjacent to the E6 trail/road, 3.7 km from Petrota Beach – or approximately 1 hour of leisure walk (Figs 1–2). According to the official police report, the attack site (60 m a.s.l.) was determined to be approximately 40 m from the forest road, 200 m from an actively used seasonal night-time corral for a 750-strong goat herd and 90 m from the herd’s watering troughs (Figs 2–4). The herd was protected by at least ten free-roaming livestock guarding dogs (Figs 5–6).

Figure 1. 

Area delimited by solid line indicates the approximate boundaries of the study area, the dashed line indicates the approximate location of the local wolf pack rendezvous site. Star symbols indicate camera traps deployed at the “Attack area”, while closed square symbols indicate location of camera traps deployed in the “Broader area”.

Figure 2. 

Location of the attack site in relation to a nearby goat herd corral and watering spot.

According to the Coroner’s report, the tourist’s body was almost completely consumed with only ten bone fragments (parts of skull, lower jaw, femur and tibia) and soft tissue (both lungs, part of the heart and small parts of skin) retrieved by the police. All remains had signs of animal bites and/or consumption. Amongst the retrieved belongings were torn clothing (incl. jacket, trousers, shirt, hat) with traces of dry blood.

Initial national and international media reported stray dogs or livestock guarding dogs as being responsible for the death of the victim (e.g. “The Guardian” 2017, “Express” 2017). However, a 26 September 2017 article in The Times (London) (de Bruxelles and Carassava 2017) cited the Coroner’s belief, prior to the completion of the laboratory analysis of remains, that the victim “may have been attacked by other wild animals, like rabid wolves” judging by the state of the victim’s remains. This statement, which was widely reproduced by national and international media, contributed to an already ongoing debate on the future of the protection status of wolves in European countries, including a call by the German Federal Ministry of Agriculture, just four days after The Times’ article, to regulate their national wolf population by hunting (Heine 2017). Recognising the potential long-term impact of this case on public perceptions of human-wolf interactions in Europe, we initiated this study on 2 October 2017.

Our study area extended over 21 km² in the Prefecture of Rodope in northern Greece (Fig. 1). It was defined as the area encompassing: (a) the attack site, (b) the grazing area of the goat herd kept at the corral adjacent to the attack site and (c) potential nearby wolf pack home sites (i.e. rendezvous sites) as defined by habitat modelling (Iliopoulos et al. 2014) because wolf presence in the area was unknown at the onset of the study. The study area ranges from 0–612 m a.s.l and is characterised by dense evergreen Mediterranean shrubland dominated by Quercus coccifera. Cultivated and abandoned olive groves extend across parts of the lower elevation and coastal areas. In addition to the goat herd that are grazed in this area, there is one more free-ranging livestock herd within the study area and five around the periphery (Fig. 1). The dirt forest road that forms the E6 European long-distance hiking trail traverses the study area parallel to the coast (Figs 2–3).

Figure 3. 

The attack site over the E6 trail (forest road passable for vehicles).

Figure 4. 

Watering spot of the goat herd corral adjacent to the attack site at 90 m, occupied throughout the study duration and until early November 2017 where remains of the victim were also retrieved.

Our study was complementary to the on-going police investigation at the time and tasked under a permit issued by the Hellenic Ministry of Environment and Energy with: (a) assessing wolf and dog presence and activity patterns in the study area and (b) providing an expert wildlife opinion on the case. All material (e.g. torn clothes, human bones, animal hair and faeces) collected at the attack site, which could be used to extract genetic material for DNA analysis of the involved predator, were held by the police department and were not accessible for external analysis. Nevertheless, in March 2019, the District Attorney handling the case granted access to the police and coronary-related reports and, hence, we included their findings in our analysis.

Reviewing both canid ecology literature and forensic cases involving large carnivore and dog attacks on humans, we identified a list of criteria and factors either linked to the risk of a human being attacked or useful for distinguishing the responsible predator (Table 1). We then used these criteria to assess our case. This approach was deemed necessary because the full consumption of the victim’s corpse and the time it took to locate the victim’s remains severely compromised the collection of evidence related to the attack for predator identification, such as size and shape of bite marks, location of injuries to the soft tissue corpse (i.e. Fonseca and Palacios 2012; Fonseca et al. 2015) or analysis of saliva DNA from fresh body wounds collected soon after the attack and prior to any post-mortem consumption (i.e. Caniglia et al. 2013; Caniglia et al. 2016). Furthermore, multi-criteria assessment was also needed due to an additional complication in our case; the inability to determine if the biological samples (hair, faeces) collected at the attack site were from the predator or subsequent scavengers, because it took 36–72 h from the initial distress phone call to locate all the victim’s remains. Even though these samples were not ultimately analysed for canid DNA, the interpretation of such analysis would have been problematic since scavenging could have taken place prior to collection. Other authors have also urged to not depend solely on forensic pathologist reports in cases of animal-related human fatalities, but to include ecological and environmental characteristics of the attack as well (i.e. Fegan-Earl 2005; Shields et al. 2009; Fonseca and Palacios 2012).

In total, our camera trapping effort was 293 trap days (mean 49 ± 15 SD, n = 6); 120 for the “Attack area” and 173 for the “Broader area” (Table 2). Wolf presence was confirmed via records at all six cameras, consisting of individuals of one reproductive pack (breeding pair and four pups). In addition, we confirmed the presence of golden jackals (Canis aureus), which is a potential scavenger for the case study (Table 2). The maximum jackal family group observed was two animals, with most detections recorded in the “Attack area”. All jackal detections occurred after dusk and before dawn. The encounter rate of natural wolf prey species was very low, with roe deer (Capreolus capreolus) not detected at all and wild boar (Sus scrofa) detected only once in the “Broader area” (Table 2). Specifically, considering both the camera trap data and the field observations, we were able to obtain the following information regarding the presence, activity patterns and overlap of wolves, livestock guarding dogs and humans at the two sites.

Vicinity of attack site – “Attack area”

We recorded 11 wolf detection events. All occurred during night-time with the activity pattern of wolves at the “Attack area” peaking at midnight and prior to sunset (Fig. 7, chart a). Until 11 November, wolf detections (n = 4) involved solely the two adults of the pack. By the time the pups, born in spring 2017, appeared in the cameras, the corral adjacent to the attack site was not systematically used by the goat herd.

We recorded 56 detection events of humans, of which 34 were of shepherds. The recreationists (hunters n = 5, hikers/cyclists n = 17) ranged in group size from 1–3 individuals. Most detections were around noon and peaked during afternoon (Fig. 7, chart c).

We recorded 69 LGD detection events and identified in total 11 LGD individuals belonging to the goat herd stationed at the corral adjacent to the attack spot, of which at least eight were males. All but two of the dog pack were of large-bodied breed in good physical condition (Figs 5–6). Activity at the corral (vehicles, shepherds, herd, LGDs) was recorded daily until approximately 12 November, when the herd moved out of the coral. After this date, the presence of the LGDs and its herd was periodic. As of early November, a cattle herd started appearing in the cameras accompanied by two additional LGDs, accounting for six LGD detections only. Hunting dog detections were rare in this site (n = 3). No feral dogs were detected at the “Attack area”.

Figures 5, 6. 

Unsupervised livestock guarding dog images from the goat flock occupying the corral adjacent to the attack site.

From those 69 LGD detection events, 45 were of “unsupervised dogs” (i.e. without livestock or humans/vehicles recorded at the given camera within 1 h). Supervised LGDs accompanying the goat herd were mostly detected during morning hours (9 am – 12 noon), (Fig. 7, chart e). In contrast, unsupervised LGD activity was multimodal, with most detections occurring either between 9 am -10 am or 4 pm - 7 pm (Fig. 7, chart g). The evening activity of the goat herd LGDs was also corroborated by the field team during the camera trap deployment at the “Attack area”. Specifically, at 7 pm (night-time) on 3 October 2017, while the livestock were corralled, the authors (YI, EC) observed and heard the LGDs barking spread out over a radius of several hundred meters from the corral.

Figure 7. 

Density plots showing the activity pattern of wolves at (a) “Attack area” (n = 47) and (b) “Broader area” (n = 43), humans (excluding researchers and shepherds) at (c) “Attack area” (n = 49) and (d) “Broader area” (n = 32), supervised large guarding dogs with the herd at (e) “Attack area” (n = 83) and (f) “Broader area” (n = 33) and unsupervised large guarding dogs at (g) “Attack area” (n = 86) and (h) “Broader area” (n = 29). The time of the original observations at the camera traps are displayed as black ticks below the x axes. Sample sizes refer to the sum of all detections (Table 2).

Livestock presence was pervasive throughout the study period. We recorded 92 detection events of the goat herd from the “Attack area” and one cattle herd. In addition, we observed several livestock carcasses in various stages of consumption along the E6 trail. Their cause of death could not be determined in all cases (e.g. disease, disposal, predation).

Activity overlap

The activity overlap between wolves and humans (hikers, cyclists, hunters) at the “Attack area”, where the fatal attack took place, was significantly lower than the overlap between unsupervised LGDs and humans (0.04 vs. 0.55; Table 3). At 5 pm, which is the time of the attack, the probability of wolves encountering humans during our study period was near zero (Fig. 8, chart a). On the contrary, the overlap of unsupervised LGDs and humans was at its peak at that same time (Fig. 8, chart b).

Table 3.

Download as

CSV

XLSX

Activity overlaps (Δ) between humans, wolves and LGD’s at the “Attack area” and “Broader area”.

Activity overlap pairs	Overlap coefficient Δ (95% CI)
	“Attack area”	“Broader area”
LGDs (unsupervised) & Humans	0.55 (0.42–0.68)	0.72 (0.56–0.89)
Wolves & Humans	0.04 (0–0.12)	0.48 (0.32–0.59)
LGDs (unsupervised) & Wolves	0.22 (0.13–0.30)	0.42 (0.27–0.58)

Figure 8. 

Activity overlap at the “Attack area” of humans (“recreationists” – i.e. hikers, bikers, hunters, excluding researchers and shepherds) and (a) wolves (coefficient of overlap Δ = 0.04) and (b) unsupervised livestock guarding dogs (Δ = 0.55). Sunrise and sunset are marked at their average time for the study period (GMT+2).

The very low activity overlap between wolves and humans was observed only at the “Attack area” and not in the “Broader area”; the latter of which represents the activity at the broader landscape beyond the immediate vicinity of the attack site (Table 3). Moreover, activity overlap of LGDs by wolves at the “Attack area” was half of that in the “Broader area” during the study period.

Based on the assessment of our findings, there is substantial circumstantial evidence that the fatal attack on the British tourist in Greece on 21 September 2017 could have been caused by the livestock guarding dogs (LGDs) of the goat herd corralled adjacent to the attack site. Our evidence includes high activity overlap between recreationists and unsupervised LGDs (with it peaking at 5 pm, which is the same time as that of the attack), lack of LGD supervision, large LGD pack size, high ratio of males among LGDs, close vicinity of the attack site to the LGDs and the livestock corral, and a record of LGD aggression towards humans as recently as a few days before the attack. Specifically, regarding LGD aggression, apart from the author’s own field observations, two Austrian tourists reported seeing a helmet-mounted video of two Belgian bikers who were attacked by seven aggressive dogs on the same section of the trail a couple days before the attack. This was reported as a written statement to WWF-Greece and submitted to the police. It was also shared with us with their consent.

However, there are certain inherent limitations due to the nature of our study which dictate that the LGDs’ involvement in the fatal attack is highly probable, but not certain. This includes that our data collection unavoidably occurred after the attack and, therefore, cannot with certainty reflect the exact conditions on the day of the attack. Additionally, the behaviour of the wolves could have changed through time, due to the brief (72 h), but intense human activity near the attack site by search and forensic teams. Therefore, in principle, a wolf attack cannot be excluded despite the very low probability supported by our findings. In terms of posthumous body consumption, our findings suggest that the same dogs and/or wild scavengers, including wolves, could have contributed.

Although our study confirmed the presence of golden jackals in the study area and the coroner’s media statements and report suggested the species could be involved in the fatal attack, we did not consider the species as a potential predator in our assessment. We decided this because there is no literature reference of jackals fatally attacking humans anywhere in the world and injury reports are limited to countries where rabies is endemic and human and jackal densities are very high (i.e. India, Akhtar and Chauhan 2009), whilst rabies in Greece is no longer endemic. Moreover, jackals were exclusively nocturnal at the study area and exhibited no overlap with human activity. Finally, the resident jackal family group consisted of just two adults (low density).

Our findings stem from a set of criteria developed to assist, along with standard forensic protocols, the predator identification in cases such as ours (i.e. nearly complete body consumption and delayed location of remains). On the contrary, the coroner’s conclusion was based primarily on the post-mortem consumption pattern observed and the lack of human DNA detection in the mouths of the dogs. Importantly, there was no differentiation in the report between the two distinct phases of the incident – the attack and the post-mortem consumption of the body. Recognising these two phases is important, as the species responsible for the fatal attack may not consume (or may not be the only one to consume) the body. Therefore, in that particular case, the post-mortem consumption pattern cannot be used as a criterion for excluding dog responsibility for the fatal attack. Moreover, the reported form and extent of post-mortem consumption can result from dog consumption according to forensic literature (e.g. Borchelt et al. 1983; Haglund et al. 1989; Avis 1999; Steadman and Worne 2007; Buschmann et al. 2011; Fonseca and Palacios 2012; Fonseca et al. 2015).

Regarding the search for human DNA in the dog cheek swabs, the results could not be informative in this specific case. The sampling occurred > 72 h after the attack and, critically, not from a location that would identify the attacker, especially if only bites had been inflicted on the victim. Several forensic publications recommend to sample for predator DNA from the saliva left on the wounds and clothes of the victim, soon after the attack (e.g. Sundqvist et al. 2007; Caniglia et al 2013; Harms et al. 2015; Caniglia et al. 2016; López-Bao et al. 2017; Fabbri et al. 2018; Plumer et al. 2018). Regardless, even if genetic material is collected using swabs from the mouth of suspected animals, a recent study by Iarussi et al. (2020) showed that no traces of hexogen (i.e. the victim’s) DNA could be traced in the swab samples after 4 h. Therefore, the absence of human DNA in the dog cheek swabs analysed in our case study is not informative about the involvement or not of the dogs in the attack.

Results from this case study are in line with the main conclusion of Linnell et al. (2021) that human fatalities from healthy wolves in Europe are extremely rare (e.g. non-existent for the period 2015–2018), while rabies remains the most prevalent cause for verified human injuries or deaths from wolves. Rabies is no longer endemic in Greece, at least since 2015 (EFSA 2021), while not a single case of a rabid wolf has been recorded in the country since rabies was first detected in 2012. On the contrary, in 2016 alone, at least 45 human fatalities caused by dogs were recorded in Europe (Sarenbo and Svensson 2020).

We recognise that the rarity of wild carnivore fatal attacks on humans in Europe, especially with such extensive posthumous consumption, complicated efforts to resolve this case. Forensic science literature of similar cases recommends a more interdisciplinary approach to the evaluation of the evidence, which includes also ecological and environmental characteristics of the attack (e.g. Fonseca and Palacios 2012). The value of such a broader approach to the case was recognised by the Police Department, Forest Service and Ministry of the Environment, which expeditiously provided the necessary permits and support for our study. Moreover, the Public Prosecutor’s office took into consideration an earlier version of this report, amongst all other information available, prior to bringing the owner of the goat herd to court with the accusation of involuntary manslaughter. After many adjournments of the court, the case verdict was issued on September 23, 2022. The judges concluded that the death was due to attack by the livestock guarding dogs of the goat herd being kept at the time at the temporary coral near the attack site, and found the owner guilty of involuntary manslaughter.

Resolving the case is important for carnivore conservation in Europe, as it received considerable national and international media coverage (Arbieu et al. 2021). The alleged involvement of wolves, despite the lack of concrete evidence to support it, has left a lingering impression to the public regarding the threat that wolves pose to humans, as deduced by the frequent enquiries addressed to the authors regarding the case. Though rare, even alleged wildlife involvement in human attacks have the potential to decrease public tolerance towards these species, which can have broader repercussions on wildlife conservation (Bombieri et al. 2018). Therefore, we consider the detailed description of what transpired in the death of the British tourist to be of value to the broader public, forensic authorities and scientific/wildlife management community.

Our assessment regarding the responsibility of LGDs in the fatal attack is also not without potential negative impacts on large carnivore conservation. LGDs are recognised as an effective and socially acceptable tool for wildlife-human co-existence (Rigg et al. 2011). The case could amplify existing concerns about the broader risk that LGDs may pose to recreationists (Linnell and Lescureux 2015). Therefore, ensuring both adequate livestock protection from large carnivores and public safety is important in areas where nature tourism and extensive livestock raising co-occur, since both activities are important for rural economies. In Greece, a large LGD group size, like the one observed in this case, is common in areas with large carnivores (wolf and brown bear). Moreover, LGDs often need to work independently from the shepherd’s supervision to locate and deter predators (Landry et al. 2020).

The location of the seasonal livestock corral next to a tourist trail was a decisive factor in the case, as it greatly increased the chance of the LGDs encountering the victim. In addition, the victim’s response to approaching dogs, though unknown, could have contributed to the initiation and escalation of the attack, as documented in similar studies (e.g. Borchelt et al. 1983; Rezac et al. 2015; Reese and Vertalka 2020).

Based on the assessment of our findings, we conclude that there is no concrete evidence that wolves were responsible for the death of the British tourist in the Maroneia-Petrota region, northern Greece. On the contrary, evidence provided by the assessable risk criteria, based on literature review and field observations, suggest an increased probability of a dog attack as a probable cause of death of the British tourist. The court’s decision issued around the same time as the acceptance of this article was also in line with our assessment.

There are lessons to be learned from this tragic case study. Risks related to livestock guarding dogs need to be carefully evaluated in areas with frequent tourist activity to avoid human injuries and fatalities. We propose three practical measures to reduce the probability of future LGD attacks on humans. First, all permanent and seasonal livestock corrals in the vicinity of tourist areas (e.g. hiking trails, open-air archaeological sites, mountain lodges) should be mapped, their potential risk to users evaluated including assessment of dog behaviour and, if necessary, relocated. In all cases, warning signs can be added both on site and in hiking maps or applications. Secondly, together with the guidelines provided for encounters with potentially dangerous wildlife (e.g. brown bears; Bombieri et al. 2019), the public should also be informed about best ways of avoiding and handling dangerous LGD encounters. Additionally, in countries where deterrence tools, such as pepper sprays, are not allowed, legal exceptions to their use for animal deterrence in rural areas could be considered.

Finally, a multidisciplinary approach in the assessment of putative wildlife attacks on humans can reduce misidentifications of the responsible species and, therefore, prevent unfounded decrease in public tolerance for large carnivores, which is difficult to achieve and even harder to maintain. To reduce future controversies, police routines should be established that automatically integrate wildlife expertise into such investigations, as well as establishing best practices for the collection of appropriate forensic data.

The study was implemented under a research licence from the Hellenic Ministry of Environment (no. 154593/2487). The authors would like to thank the Head of the Rhodope Forest Directorate, M. Gotzaridou for assistance and support during field work, Sapes Police Department for facilitating ongoing research, P. Maragou from WWF Greece for providing access to testimonials from the study area, R.A. Montgomery and three other anonymous reviewers for providing valuable feedback on an earlier version of this manuscript and Callisto Wildlife Society for providing all necessary equipment needed for the research.