Density estimation is a key issue in wildlife management but is particularly challenging and labour-intensive for elusive species. Recently developed approaches based on remotely collected data and capture-recapture models, though representing a valid alternative to more traditional methods, have found little application to species with limited morphological variation. We implemented a camera trap capture-recapture study to survey wolf packs in a 560-km2 area of Central Italy. Individual recognition of focal animals (alpha) in the packs was possible by relying on morphological and behavioural traits and was validated by non-invasive genotyping and inter-observer agreement tests. Two types (Bayesian and likelihood-based) of spatially explicit capture-recapture (SCR) models were fitted on wolf pack capture histories, thus obtaining an estimation of pack density in the area.
Despite the wolf is one of the most studied large predators worldwide, no study has so far reported abundance and/or density estimation of wolves by camera traps. In a pilot investigation, Galaverni and colleagues  combined CT with NGS to test their effectiveness in monitoring a wolf population. The authors stated that, although identifying individual wolves from photographic material during the study was rarely possible, CT data allowed an estimation of the minimum packs size that was similar to that obtained through NGS. Actually, the main issue related to the use of a camera-trap capture-recapture (CTCR) approach with wolves primarily concerns the presence of sufficient phenotypic variation for individual recognition. Although wolves lack evident natural markings, external idiosyncrasies (e.g. permanent injuries, blind eyes) often occur in a population, allowing for individual recognition. Moreover, wolf-dog hybridization events, documented in many areas of Europe , can introduce phenotypic variation in traits like body size, pelage colour, length, shape and carriage of tail and ears. This source of morphological variation, possibly combined with individual difference in behavioural traits associated to social status (e.g. scent-marking display), can allow individual recognition in wolf populations.
A second issue is the applicability of CR approaches to group-living species, given that CR models assume uncorrelated activity centres of individuals (i.e., independence of capture events . As wolves live in packs, capture events are often correlated, violating this assumption. For these reasons, the adoption of a CTCR method in wolves is challenging and needs further validation. Moreover, accurate estimates of wolf density are very infrequent in Europe and limited to a few radio-tracked wolf populations [25,26,27]. In Italy, estimates of wolf population density are scarce [28, 29], whereas a large amount of grey literature reporting on local abundance was used to extrapolate large-scale density values [30, 31]. In this study, we tested for the first time the applicability of CT to obtain robust estimates of wolf density using CR methods. We chose a pack-based approach, inferring total wolf density from pack density and average pack size as a conversion factor . We studied an Italian wolf population, where individual recognition was facilitated by the introgression of canine genes  and validated by the support of NGS data coming from a long-term research project. Our aims were (i) to obtain a robust density estimate for the wolf population of our research area in the province of Arezzo, Central Italy, comparing a Bayesian approach with a likelihood based one; (ii) to test for repeatability of our method by evaluating the effect of inter-observer disagreement in wolf identification on density estimates; and (iii) to test the effect of different survey periods and CT sampling design on wolf density estimate. Finally, we discuss our results in the light of the up-to-date knowledge on wolf density at local and continental scale, and evaluate pros and cons of the application of a CR approach on camera trapping data in this species.
During 303 sampling occasions (5197 trap days), from the 1st of April 2014 to the 11th of June 2015, we achieved a total of 909 wolf videos corresponding to 657 independent capture events (CE) (1.38 videos/CE) and 1240 individuals captured (1.89 individuals/CE). In 130 CE, wolf recognition was not possible because of low video quality (e.g. fuzzy videos of running individuals, recording only part of the silhouette, excessive distance of the individual from the camera, or bad weather conditions). Of the remaining 527 usable CE, 427 (81.0%) were assigned to a specific pack and the remaining 100 were classified as undetermined. One or both members of the breeding pair were captured in 356 out of these 427 CE (54.2% of the total) and, after having selected one single focal individual/pack, its relative capture histories were created (in total 295 CE). The data of the pilot study and the two sessions are summarized in Table 1. During the pilot study and the first session (2014), we identified 10 packs constituted of 20 alpha (α), 14 beta (β) and 9 pups for a total of 43 wolves (see Additional file 1). In the subsequent session (2015), 9 out of 10 packs were confirmed and 3 new packs were identified, two of which at trap sites that were not used in 2014 (Fig. 1), for a total of 12 different packs. In the 2015 session we ascertained the presence of 50 wolves, 24 of which were α and 26 were β (the latter including individuals born in 2014). In total, 14 different focal individuals were considered: 10 individuals belonged to the packs observed in the 2014 survey, three individuals to three new packs discovered in 2015 and the last one was the new α male M22 that in 2015 replaced the α male detected in one of the 2014 packs (Table 2). The distinctive features used for their recognition are described in Additional file 1. A selection of the video captures of each focal animal used for density calculations is available in Additional file 2.
Camera trapping study area in Arezzo province (Italy), where wolf density was estimated in 2014 and 2015. Approximate locations of the 13 detected packs were reconstructed by the video-captures of focal animals at distinct trap sites during the study sessions. The large dotted area is the habitat suitable to wolves and is formed by dots representing potential wolf pack activity centres (spaced 666 m and buffering the trap array by a 15-km radius). Unsuitable habitats for wolf were excluded from calculations and are shown in white
Comparison of wolf pack density estimates for the study area (Arezzo province, Italy) during 2014 and 2015 camera trap capture-recapture sessions, obtained by four models (NE-_NULL, NE_TP, HN_NULL and HN_TP, see text and Table 4 for details) using SPACECAP and secr estimators. Bars represent the 95% confidence intervals
When these prerequisites are satisfied, the classic CTCR method needs some adjustments to be successfully applied to wolf. First of all, the focus must be on packs instead of individuals. Three years of integrated camera trapping and NGS suggested that individual recognition is feasible for α individuals but not for all resident wolves. We were able to recognize α pairs of different packs by means of morphological and behavioural traits inasmuch as NGS validated all fifteen video assignments of ten α individuals belonging to four different packs (Additional file 1). Although we correctly assigned some β individuals to some given packs, we failed in tracking some between-packs movements of certain individuals and some changes in social status within packs that occurred during the study. However, our CR approach assumes no misidentification of α individuals only within a single session, i.e. few weeks or months, and not between years, so these limitations in individual recognition had no effect on capture history and density estimation. For a territorial group-living species like the wolf, referring to packs in population monitoring is more feasible and anyway well informative on total population size when adequate conversion factors - from packs to total individuals - can be applied .
The last issue concerns study duration. In our approach, the actual CR session was a time window within a longer monitoring period, preceded by a pilot study necessary to collect preliminary video material for focal animal identification, to verify the presence of an adequate variability and to test for effectiveness of camera locations. Since focal animal identification is assisted by observation of wolves during scent-marking, checking putative camera station points as effective scent-marking sites is crucial to have a uniform detection probability. Moreover, an adjustment of camera position and settings is necessary to maximize captures and individual recognition. During our pilot study, we identified nine out of ten focal animals, that were captured by videos during the subsequent session 2014, and we also obtained a preliminary estimation of the animal movement parameter that led us to improve trap spacing in our sampling design. Similarly, our second CR session was followed by a control period to verify focal animal persistence and to collect additional data on pack size.
According to Foster and Harmsen , when monitoring species without evident natural markings, one has to clearly indicate the level of inter-observer agreement in individual recognition. Video analysis of all capture events and compilation of capture histories for model analysis were performed by a single operator (LM). However, to test the repeatability of our protocol, we selected two subsamples of 20 and 40 videos among the capture events collected during the 2014 survey, representative of most trapping sites. The first sample of 20 videos simulated the pilot survey and was intended as a training for inexperienced operators. The second group of 40 videos was used as the actual test. We tested the level of agreement between LM and three other operators, two of which were field assistants (EB, EF) and the third one (AC) was inexperienced both in field work and video analysis. The test was blind, operators having information only about location and date of video captures. Inter-observer agreement/disagreement was expressed as the percentage of matches in pack assignment and as the difference in the number of different packs identified by the three observers. We also evaluated the effect of inter-observer disagreement on density estimation by performing a model analysis of the 40 videos capture histories compiled by the different operators and by comparing differences in model outputs. 2b1af7f3a8