Dog training

The dog team comprised one dog and the handler (first author), the handler was previously trained as an FWN observer (see below). The dog (female Belgian shepherd selected by a professional dog-breeder) had been living with its handler from the age of two months and was trained to recognize wolf scats. The training protocol, based on positive reinforcement, was adapted from previous conservation detection work (Smith et al. 2003, Wasser et al. 2004, Sentilles et al. 2016; Appendix S1) and remains fundamentally the same. The dog progressed to an interim training stage similar to that described in Statham et al. (2020), graduating from controlled settings with artificial scat provision into known, occupied wolf territories in order to encounter naturally occurring wolf scats. The handler trained the dog with samples from 60 genetically confirmed wolf scats originating in the French Alps. The selection of scats for training included genetically confirmed samples from target species (scats from male and female wolf haplotype group w22 (sensu Pilot et al. 2010), i.e. characteristic haplotype from grey wolves originating in the Apennine Peninsula and French Alps). Scats known to be from non-target species (red fox Vulpes vulpes and domestic dogs fed with wild ungulate meat) were also provided during training (as per DeMatteo et al. 2019). The training protocol consisted of searching off leash, with the dog indicating having found wolf scat by freezing, lying down, and barking. For each find the dog was rewarded by playing with a ball. Once the dog showed consistent recognition and detection of naturally occurring wolf scat, the handler deemed it ready to deploy on surveys.

French Wolf Network observer training

Wolf range expansion was monitored at the country level without standardized experimental design being implemented by the French Office of Biodiversity. Instead, opportunistic sampling was conducted by a network of field observers, the FWN. These field observers come from governmental (French Office of Biodiversity) or non-governmental organizations (e.g. game and wildlife services, hunting associations, forestry offices, naturalists, farmers). All observers completed a 3-day training course, covering species identification and surveying methods (Duchamp et al. 2012). After completing the training, the observers were asked to search for naturally occurring wolf scat samples on road trails and crossroads (detailed below).

Newly colonized areas and choice of sampled unit

The dog-handler team was deployed in 12 newly colonized areas declared as Permanent Presence Zones (PPZ), i.e. areas where wolf signs have been reported for two consecutive years and the identity of the species has been confirmed by DNA analysis or camera trapping. These sample units were surveyed by both the dog team and FWN observers. Each sample unit (i) were defined as 10 × 10 km cells (European Commission 2006). We used the same grid used by Louvrier et al. (2018b), available online at:  http://carmen.carmencarto.fr/38/Loup.map. We assumed that one grid cell of 100 km² corresponded to the minimal home range size for grey wolves at these latitudes (Okarma et al. 1998, Mech & Boitani 2010, Mancinelli et al. 2018). One sample unit was in the Maritime Alps (a new peri-urban colonized area), the others were outside the French Alps. In the case of the Maritime Alps, FWN observers suspected a new wolf pack, but only one genetic sample had been obtained. In one sample unit (in Camargue, an agricultural lowland, where one rogue wolf was suspected in sheep depredation) no genetic samples were available despite intense search effort by local FWN observers. For all these PPZ, we lacked genetic samples to implement the CMR model (i.e. no or few genetic samples collected each year). A designated wolf referee coordinated the network activity, controlling the technical reliability of each presence sign that was reported. We decided to deploy the dog team in each sample unit where FWN observers reported difficulties in finding wolf signs. Based on FWN coordinator expertise, we identified seven PPZ as held by a wolf pack (WP), and five zones with no wolf pack (no-WP) but where at least one solitary wolf had been reported for two consecutive years. We examined photographs and considered the quantity of prey remnants showing characteristic wolf attack marks, and the MNW on each PPZ (if available). During 2018-2019 FWN observers were asked to continue trying to collect wolf scats in the areas surveyed by the dog-handler team. All samples collected by FWN observers were genetically analysed. Dog team field surveys were conducted from October 2018 to May 2019. Surveys occurred mainly in mountainous areas (altitudes ranging from 600 to 1,460 m), but one sample unit occurred on agricultural lowland (in Camargue).

Dog-handler team sampling

The team surveyed predetermined transects that were preferably circular for logistical reasons (return to the vehicle), in some cases a member of the FWN transported the dog and handler, thus allowing a line-transect along forest roads. In each sample unit two to four transects were chosen, the aim being to cover more or less the same distance in each sample unit. The dog team surveyed trails previously covered by local FWN observers (including those where no scat had been detected but local knowledge considered wolf presence to be possible) or where camera trapping captured at least one wolf. The dog handler had no previous knowledge of the survey areas. We equipped both the dog and the handler with GPS devices (Dogtra pathfinder) to record survey tracks and distance covered, and to enable mapping of all scats indicated by the dog. All collected scats were genetically analysed, as described below. The dog team spent nine non-consecutive days surveying in 2018 and 10 non-consecutive days in 2019 (see Table 1 for details).

French Wolf Network opportunistic sampling

Both dog-team and FWN observers surveyed the same group of twelve sample units. The number of FWN observers per sample unit ranged from ten to > 55; the ratio of professionals/non-professionals ranged from 0.2 to 0.5 (Appendix S3b). In new areas of wolf colonisation, state rangers from the French Office of Biodiversity carried out baseline monitoring in all districts where wolf signs were still considered as unreliable. The rangers belong to the FWN and were already deployed across France for environmental policy and wildlife monitoring; they might opportunistically find scats during this time. Non-professional FWN observers usually go into the field specifically to search for wolf signs including scats. However, we did not require the FWN observers to record details of their time spent in the field.

Table 1.

Time and spatial parameters of the wolf detection dog surveys.

All scats collected by the FWN observers were genetically analysed (see below). According to Barja et al. (2004), wolves tend to travel along forest roads and trails and to mark them with tracks and scat, especially at forest crossroads. As such, we asked FWN observers to pay attention to forest crossroads and to collect all scat found there. The proportion (number of professionals over the total number of FWN observers) was 0.2 to 0.5. This proportion also depended on the unit (see also Louvrier et al. (2018b) for detailed analysis of sampling effort by FWN observers based on the number of observers per grid cell). However, no information on total time spent surveying opportunistically by FWN observers was available.

DNA extraction and genetic analysis

All scats were stored individually, without desiccant, in a sealed plastic bag. Samples were then frozen. Scat samples were analysed using a mitochondrial DNA (mtDNA) species identification test, as well as a Polymerase Chain Reaction (PCR) multitube approach based on 20 microsatellite markers and one sex-specific marker for individual and gender identification. Extraction methods, PCR protocols, protocols for individual identification, molecular sexing and for screening wolf samples from other animal species samples are described in detail in Valière et al. (2003) and Duchamp et al. (2012). We used these genetic analyses to test the reliability of the scat-detection dog's ability to discriminate wolf scats from other (nontarget) species scats. They were also essential to evaluate the efficiency of the scat collected by the dog-handler team vs. humans-only teams, to provide genotypes or detect different individuals. Genetic analyses were performed at the ANTAGENE company ( https://www.antagene.com/en).

Data modelling and analysis

To model the probability of detection by the dog team,weusedasingleseasonsite-occupancymodel to account of for what we termed “imperfect” detection (i.e. missing a scat when present) and to estimate the power of the dog survey method (MacKenzie 2006, MacKenzie et al. 2009) The survey was carried out over 12 separated grid cells (Fig. 1), and it lasted less than three days in each sample unit “i”. Each sample unit was in one PPZ. Within each sample unit “i” the protocol ensured that the dog team passed through a point where wolf presence signs were previously detected.

Fig. 1.

a) Wolf distribution in France (right) and localization of sample units (left), b) example of surveys in one sample unit. Permanent Presence Zone refers to areas where wolf signs have been reported for two consecutive years (PPZ), occasional presence refers to areas where wolf signs have been reported in less than two consecutive years.

The model (adapted from Hines et al. 2010; Appendix S2) was implemented in WinBugs (Kéry 2010) with 30,000 iterations and three chains. The distance covered was subdivided into segments “j” of equal length (i.e. 1 km) that were treated as spatial replicates in the occupancy analysis. The data was subdivided into spatial replicates (1 km transect segments) for the survey in each sample unit “i”. The number of 1-km replicate segments surveyed per sample unit ranged from six to 32. We checked convergence visually by inspecting the chains and by checking that the Rhat statistic was below 1.1 (Kéry 2010, Brooks & Gelman 2012; Appendixes S2, S3a).

We used a t-test to compare mean number of scats per kilometre of survey in areas where only dispersers were identified vs. wolf pack territories. All statistical analyses (other than occupancy with WinBugs) were performed using R-software.

Fig. 2.

Total number of scats collected by French Wolf Network (FWN) observers and the dog-handler team in twelve sample units. Grey bars – wolf haplotypes. White bars – false positive (i.e. nontargets). Black bars – scats not suitable for analysis (failed to amplify, see also Table 2).

To model the probability of detection by FWN observers, we used a single season site occupancy model. One approach to modelling detection/non detection from multiple seasons is to effectively apply a single season model to the data collected in each of the seasons (MacKenzie 2006). Under this approach, occupancy in one season is considered to be a random process in the sense that the occupancy status of a unit in the previous season has no effect on the probability of occupancy in the same unit in the current season. Regardless of the underlying processes of change in occupancy, only the resulting pattern of level of occupancy each season is modelled.

Table 2.

Species identification of wolf scat samples located by dog-handler team and FWN observers in France. ^aThe percentage of the total number of DNA-amplified samples confirmed as wolf.

Fig. 3.

Number of scats collected annually between 2016 and 2019 in twelve sample units. White bars – scats collected by trained observers from the FWN. Grey bars – scats collected via the dog-handler team. The year 2018 corresponds to the first year of using the dog team.

The model was implemented in WinBugs (Kéry 2010) with 30,000 iterations, three chains, a burn-in of 4,000 and a thinning of five. The time covered was subdivided into time surveys “j” of equal duration (i.e. one month) that were treated as temporal replicates in the occupancy analysis. The data were subdivided into temporal replicates (one month) for the survey in each sample unit “i”.

Fig. 4.

Comparison of the mean number of scats collected in each sample unit between 2018-2019 by FWN observers and by the dog-handler team. Opportunistic scat collection by FWN observers spanned one year, whereas the dog team spent two days of targeted monitoring to collect scats in each sample unit.

The number of one-month replicates surveyed per sample unit ranged from 12 to 48. The resulting data were treated as single detection histories for each of the sample units “i” (large cells). Each detection history corresponding to one season (i.e. 12 months) in the same sample unit “i” was grafted to the precedent to construct one unique detection history per sample unit “i” (Appendix S3c). Each detection history contained a “0” (no detected scat) or a “1” (detected scat) per month “j”.

Fig. 5.

Record of wolf haplotypes determined from analysis of scats found by FWN observers and by the dog-handler team. Dark grey circles show individual genotypes found as a result of the dog team surveys only, light grey indicates individual genotypes found by both the dog team and by FWN observers. A subset of the genotypes determined from scats (n = 7 of 29) found by the dog team were the same as those determined from scats recovered by FWN observers.

Fig. 6.

Cumulative probabilities of scat detection in territories held by a wolf pack (WP) or dispersers (no-WP). Lighter shades represent the estimated 95% confidence interval. a) Targeted monitoring using a trained detection-dog, b) Opportunistic monitoring by FWN observers.

Detection heterogeneity

Comparison of dog detection results in WP and no-WP areas suggest that the probability of detection seems to be correlated with the number of scats present (the higher the number of scats, the higher the probability of finding them). In areas where we suspected only dispersers, we found that using a dog team led to a detection probability that was half of that in wolf pack territories. Dispersers do not mark their territories as resident paired mates do (Mech & Boitani 2010). It is therefore logical that wolf faeces identified by the dog team were four times more frequent in WP territories than in no-WP. Despite that, the observed difference could also result more from the smaller potential quantity of faeces found in no-WP areas, rather than the difference in marking behaviour.

Complex biological phenomena may influence detection probability, and one should remain cautious when trying to interpret differences in detection probability. Some factors affecting detection heterogeneity by FWN observers include seasonal variation, as higher detection probabilities occur in winter (Marescot et al. 2011, Louvrier et al. 2018b). However, it is important to note that these different detection rates arise because of the reliance on visual factors influencing wolf sign detection by human observers. This may be especially true in the French Alps where the sampling effort is very high, and areas of historical wolf packs are well known (see map in Appendix S3b). Indeed, wolf sign probability detection by FWN observers is relatively high during winter in the French Alps core, where the sampling effort (defined as the number of observers per site and per year) is highest. In our study, detection probabilities by FWN observers did not differ between WP and no-WP territories. We suspect that since snow events were rare in these areas, sampling effort and experience of FWN trained observers was the main factor determining wolf scat detection in newly colonized areas. This is in accordance with results found in previous studies (Louvrier et al. 2018a, b).

Detection dog team performance and scat collection

The detection dog found 82.1% of the total wolf scats detected by all survey methods. Trained dogs have been used to find scat samples from numerous species (Engeman et al. 2002, Smith et al. 2003, Wasser et al. 2004, Cablk & Heaton 2006, Sentilles et al. 2016, Richards et al. 2018, Statham et al. 2020). The presence of nontargets within the surveyed environment may introduce complexity. For example, when looking for otter (Lutra sp.) faecal matter, handlers must be particularly vigilant if a communal latrine is encountered, wherein numerous other species may have defecated upon and/or overmarked target scats, to avoid reinforcing the dog to co-occurring nontarget scat (Richards et al. 2018). If the handler unintentionally rewards the dog for a nontarget, e.g. because it is visually indistinguishable from the target or has been overmarked, the dog may then seek it out in addition to their target, which can in turn decrease accuracy (Smith et al. 2003, MacKay et al. 2008).

The proportion of false positive responses (i.e. detection of nontargets) using scat detection dogs has previously been reported to be highly variable, ranging from 3.7 to 44.6% (DeMatteo et al. 2018). False positives were low (4.1%) in the current study and therefore did not pose a problem. The non-target scat found in our study were visually characteristic of wolf scats, we suspect these were target scats that had been urine-marked by other species, mainly red fox. According to Wikenros et al. (2017), meso-predators such as red foxes tend to urinate or defecate on the scat of apex predators. Systematic DNA analysis of recovered samples is key for effective monitoring (Hollerbach et al. 2018) and genetic confirmation to species is typically also recommended by detection professionals (see Statham et al. 2020). We note that many factors, including the age of the sample, exposure to elements and diet, influence the success of DNA amplification (Murphy et al. 2003, Piggott 2005, Panasci et al. 2011), which can also reflect on the performance of dog-handler teams and human-only teams. Collection and storage conditions of the samples also influences their viability for genetic analyses (Waits & Paetkau 2005, Statham et al. 2020). In the future, we could improve sample storage using cotton swabs in the field to isolate the mucous layer on the surface of fresh wolf scats (Rutledge et al. 2009).

In this work, the use of a detection dog and handler team helped improve the genetic monitoring of wolves in newly colonized areas of France. The team surveys enabled an improvement in quantitative (i.e. more genetic samples) and qualitative (individual genotyping) data collection. We showed that the use of the dog team allows considerable time savings (up to 99.6%) in comparison to opportunistic monitoring by human-only teams. As information sharing is an important aspect for mitigating conflict over wolves (Treves et al. 2009), detection dog utilization (and the time gain it represents) may be particularly useful in rapidly disseminating information to stakeholders. Moreover, detection dogs may play an ambassadorial and stakeholder engagement role, as livestock farmers often accompanied researchers in the field and were interested by our work.