Introduction
Estimates of population abundance and density are key metrics that can inform wild felid conservation. Initiatives such as IUCN Red List or WWF Living Planet assessments (Collen et al., Reference Collen, Loh, Whitmee, McRAE, Amin and Baillie2009; Santini et al., Reference Santini, Butchart, Rondinini, Benítez-López, Hilbers and Schipper2019; Ledger et al., Reference Ledger, Loh, Almond, Böhm, Clements and Currie2023), monitoring and management of wildlife populations (Callaghan et al., Reference Callaghan, Santini, Spake and Bowler2024), evaluation of conservation outcomes for protected areas (Geldmann et al., Reference Geldmann, Barnes, Coad, Craigie, Hockings and Burgess2013), as well as conflict mitigation actions promoting human–wildlife coexistence (Frank, Reference Frank2016), can all benefit from population estimates of the target species. Although some methodologies can estimate abundance based on counts of unmarked animals (e.g. Fiske & Chandler, Reference Fiske and Chandler2011; Dénes et al., Reference Dénes, Silveira and Beissinger2015; Moeller et al., Reference Moeller, Lukacs and Horne2018; Gilbert et al., Reference Gilbert, Clare, Stenglein and Zuckerberg2021) including unmarked spatial capture–recapture models (Chandler & Royle, Reference Chandler and Royle2013), they provide imprecise density estimates and are restricted to a Bayesian framework (Augustine et al., Reference Augustine, Royle, Murphy, Chandler, Cox and Kelly2019; Gilbert et al., Reference Gilbert, Clare, Stenglein and Zuckerberg2021; Twining et al., Reference Twining, McFarlane, O’Meara, O’Reilly, Reyne and Montgomery2022). In contrast, traditional spatial capture–recapture models, which rely on individual identification, are amongst the most reliable and readily available frameworks to determine population density, especially for species with distinct markings (e.g. Silver et al., Reference Silver, Ostro, Marsh, Maffei, Noss and Kelly2004; Sharma et al., Reference Sharma, Jhala, Qureshi, Vattakaven, Gopal and Nayak2010; Strampelli et al., Reference Strampelli, Andresen, Everatt, Somers and Rowcliffe2020). However, these models have been mainly used for conspicuous species such as large felids (Brooke et al., Reference Brooke, Bielby, Nambiar and Carbone2014), with small, subtly marked felid species receiving relatively little research attention (Brodie, Reference Brodie2009).
One such species is the güiña Leopardus guigna, the smallest wild cat of the Neotropics, with a body length of 59–64 cm and a typical weight of 1.3–2.5 kg (Dunstone et al., Reference Dunstone, Durbin, Wyllie, Freer, Jamett, Mazzolli and Rose2002; Napolitano et al., Reference Napolitano, Díaz, Sanderson, Johnson, Ritland, Ritland and Poulin2015; Peckham, Reference Peckham2023). Phenotypically the species shows two coat patterns: a brownish-yellow to grey-brown coat with black spotted markings, and melanistic, which also shows black spotted markings but at a low contrast as the coat’s base colour is also dark. The güiña occurs in central-southern Chile and south-western Argentina, and it is categorized as Least Concern on the IUCN Red List. However, three subpopulations are considered to be in critical condition (Gálvez et al., Reference Gálvez, Napolitano, Ibacache, Agostini and Pliscoff2025). Habitat loss, land-use intensification, forest fires and human–wildlife conflict linked to predation of poultry are the main threats to the species’ conservation (Napolitano et al., Reference Napolitano, Díaz, Sanderson, Johnson, Ritland, Ritland and Poulin2015; Gálvez et al., Reference Gálvez, Guillera-Arroita, St. John, Schüttler, Macdonald and Davies2018, Reference Gálvez, Infante, Fernandez, Díaz and Petracca2021a,Reference Gálvez, St John and Daviesb).
Camera traps are a powerful tool for identifying individuals with natural markings (e.g. spots, rosettes, stripes) in ecological and behavioural studies (Wearn & Glover-Kapfer, Reference Wearn and Glover-Kapfer2017). Researchers have used camera traps primarily on large carnivores to assess species richness and relative abundance, whereas population density has been studied less frequently (Burton et al., Reference Burton, Neilson, Moreira, Ladle, Steenweg and Fisher2015). Commonly, camera traps are set up facing horizontally, with one or more cameras per site (Smith & Coulson, Reference Smith and Coulson2012), to capture the flanks of passing animals. However, these markings may be insufficient to identify individuals in species that have more clearly distinguishable dorsal markings (Theimer et al., Reference Theimer, Ray and Bergman2017). Blair (Reference Blair2014) evaluated the overall coat pattern of the güiña and determined that the dorsal area shows considerably higher individual variation in the spotted patterns than the flanks. The potential of the dorsal coat pattern for individual identification, coupled with the absence of density metrics from camera-trap studies for güiñas despite considerable survey efforts during the last 20 years (Gálvez et al., Reference Gálvez, Infante-Varela, de Oliveira, Cepeda-Duque, Fox-Rosales and Moreira2023), underlines the need to explore ways to capture the dorsal markings of this felid with camera traps.
Here we describe a systematic procedure to identify güiña individuals based on their dorsal markings, using camera traps with zenith orientation (i.e. vertical placement; zenith camera traps henceforth). We provide recommendations for camera-trap installation and outline a systematic and transparent protocol for the identification of this small felid. We discuss the potential of this method for estimating population densities of the güiña and a range of other species with dorsal markings.
Study area
Our study area was in the Reñihué Valley, in the continental fjords of northern Chilean Patagonia (Fig. 1). The vegetation of this area is dominated by dense temperate rainforests containing evergreen species such as Nothofagus dombeyi and Eucryphia cordifolia in the upper arboreal layer, with a significant presence of Myrtaceae and Podocarpaceae. The forest has a dense understory dominated by Chusquea sp., tree seedlings and saplings, several fern species and bryophytes. The mean annual temperature is 10.6 °C and mean annual rainfall is 6,000 mm (Salazar, Reference Salazar2017). The area is situated within a 778 ha private property at the mouth of the Reñihué River, which is surrounded by Pumalín Douglas Tompkins National Park.
The Reñihué Valley study site, with the main rivers and the 40 zenith camera-trap locations.
Methods
Zenith camera-trap installation
Zenith camera traps (Bushnell Trophy Cam HD Aggressor, Bushnell, USA) were attached to tree branches or beams parallel to the ground and facing it (Fig. 2a), with the length of the camera’s field of view parallel to the expected direction of güiña movement underneath the camera (Fig. 2b). A camera-trap detection zone is composed of many detection windows (Fig. 2c), so when an animal moves from one window to the next, the passive infrared sensor registers a difference between windows in the radiation received, triggering the camera (Wearn & Glover-Kapfer, Reference Wearn and Glover-Kapfer2017). If the field of view’s length is perpendicular to the animal’s direction of movement, the camera may not be activated because of the small number of detection windows across the trail.
Zenith camera-trap installation. (a) Photograph of the camera-trap installation in situ. (b) Schematic representation of the installation, viewed in the direction of expected animal movement (e.g. along an existing trail). The length of the camera trap’s field of view is parallel to the expected direction of movement. (c) Diagram showing the camera’s field of view, with detection windows and target animal, whose body length should be c. one-third of the length of the field of view. (d) The length of the field of view can be adjusted via the height of the installation, with a higher installation resulting in a longer field of view (see Equation 1).
To maximize the size of the detection zone and field of view in zenith installations, we evaluated the optimal height of installation (Fig. 1d; Equation 1). For individual identification we aimed to obtain at least two consecutive full-body photographs. Based on an initial calibration study (see below), this occurred when the field of view’s length was approximately three times greater than the body length of the species, even if the animal is moving fast (as güiñas do). We used a basic trigonometric formula (Equation 1) to determine the installation height (h) from the body length of a species (L) and the angle covered by the camera’s field of view (α):
$$h = {{3L} \over {2 \times \tan {\alpha \over 2}}}$$
Given that güiñas have a body length of c. 0.6 m and the cameras’ field of view angle was 42°, the recommended installation height was calculated as 2.34 m. Additionally, because each camera model has a particular fixed focus distance not usually disclosed, we conducted a calibration of this height to refine the quality of photographs by comparing cameras placed at heights of 2.0, 2.5 and 3.0 m. We determined that the best focus was at a height of 2.5 m. This calibration is an important step in facilitating the subsequent individual identification.
During February 2019–November 2020, we placed 40 zenith camera traps in our study area; the mean distance between neighbouring cameras was 253 m, with a minimum of 51 and maximum of 486 m. The resulting camera-trap array had no spatial gaps based on home range estimates (Dunstone et al., Reference Dunstone, Durbin, Wyllie, Freer, Jamett, Mazzolli and Rose2002; i.e. all individuals could potentially be detected) and covered an area of 640 ha (calculated as the minimum complex polygon around all stations). Cameras were set to a triggering speed of 0.2 s and captured three images for each detection event, with a recovery time of 2 s between consecutive detection events. Additionally, we selected the highest available shutter speed to ensure sharp photographs of fast-moving güiñas. We defined independent events as detections at the same camera trap that were more than 30 min apart.
Individual identification of the dorsal pattern
The identification was carried out by one of the co-authors (TK; henceforth, identifier) who is a professional photographer. The process had four stages: filtering, editing, grouping and accuracy assessment (Fig. 3). Filtering involved scoring all photographs, from 1 (low) to 4 (high), based on two characteristics: visibility and sharpness (Fig. 3a). Visibility refers to the proportion of an individual’s body inside the photograph, including the head and tail. Sharpness refers to how clearly the details of natural marks were rendered in the photograph (i.e. spot and line borders in coat pattern). We discarded photographs when the sharpness and/or the visibility score were equal to 1 (i.e. the coat pattern was not discernible and/or less than 25% of an animal’s body was in frame) and kept the two highest scoring photos from a sequence.
Stages of the individual identification process using photographs of dorsal patterns of the güiña Leopardus guigna. (a) Filtering: suitable images are selected based on sufficient visibility and sharpness. (b) Editing: selected images are processed to obtain a high contrast edited image. Each edited image is assigned a unique ID associated with a particular event for accuracy assessment at stage (d). (c) Grouping: edited images are evaluated and at least two diagnostic marks are identified. Images with matching/similar diagnostic marks are grouped together (i.e. assumed to show the same individual). (d) Accuracy assessment: grouped edited images are checked to determine whether they belong to the same photo sequence (i.e. the same individual).
During the editing stage (Fig. 3b), we used a photo editing software (Adobe Systems, 2018) to digitally remove the background elements of each image, leaving only the target animal. We then converted photographs to grayscale, increased the contrast, printed the edited images at a resolution of 300 dpi and labelled each with a random code associated with the particular site and sequence in our database (see accuracy assessment). The grouping stage (Fig. 3c) was performed using a double-blind approach in which the identifier was unaware if two edited images belonged to the same independent event. Our identification protocol relied on recognizing at least two distinctive natural diagnostic marks to identify an individual, following a protocol for cheetah identification (Brassine & Parker, Reference Brassine and Parker2015). Most diagnostic marks become more evident as the time dedicated to their recognition increases (Dorning & Harris, Reference Dorning and Harris2019), hence several rounds of identifying and grouping images as belonging to the same individual were carried out by the identifier. Finally, to assess accuracy (Fig. 3d), we evaluated the identification process by checking if the pairs of edited images that belonged to the same independent event were grouped together as belonging to the same individual by the identifier. This assessment was only possible for events that had two edited images. The proportion of pairs that were successfully grouped together was used as a measure of the accuracy of the identification process.
Results
During a total of 12,784 trap-days we captured 1,386 photographs of güiñas in 586 independent events. Our trapping success was 4.6 events per 100 trap-days. More than half of the güiña photographs (n = 791; 57.1%) showed individuals with a melanistic or indeterminate morph, which were excluded from further analysis. Additionally, we excluded 101 photographs that scored 1 in sharpness and/or visibility. Sixty photos had low sharpness (59.4%), 22 had low visibility (21.8%), and 19 had both low sharpness and visibility (18.8%). Because we left only the top two scoring photos per sequence, we excluded an additional 144 photographs. In total, we included 350 photographs in the editing stage (Supplementary Fig. 1).
After editing (Fig 3b), the identifier searched for diagnostic marks and identified 320 of the 350 images at the individual level (91.4%). The remaining 30 were assigned as unknown individuals because they did not present two or more distinguishable marks to confirm their identity. In total, individual identification was accomplished in 210 of the 252 independent events of spotted güiñas obtained (83.3%). We identified 12 individuals of spotted güiñas (Fig. 4), with nine of them providing at least one spatial recapture. The identifier correctly assigned all of the 110 events with two identifiable edited images to the same individual in a particular sequence.
Dorsal pigmentation patterns of güiñas. (a) Library of the 12 identified individuals with the number of edited photographs assigned to each, and the number of different camera-trap stations where they were detected. (b) Examples of the pigmentation patterns of the cervical area. (c) Examples of pigmentation patterns of the sacral area.
Discussion
Individual identification can facilitate population density estimates of small wild felids, which in turn could provide valuable insights into the conservation status of many understudied species (Brodie, Reference Brodie2009). Our results show that zenith camera-trap placement can achieve individual identification of subtly marked species such as the güiña. Nonetheless, camera placement, image processing and identification accuracy need to be assessed systematically to obtain robust results. Our study confirms empirically that the dorsal pattern of the güiña can provide sufficient information to reliably identify individuals. The ability to distinguish individuals is the key first step towards implementing capture–recapture studies aiming to estimate abundance or density of güiña populations, data that do not currently exist for the species (Gálvez et al., Reference Gálvez, Infante-Varela, de Oliveira, Cepeda-Duque, Fox-Rosales and Moreira2023).
Future capture–recapture studies using the zenith camera placement and sampling design described here could have important implications for the conservation of the güiña. For example, the current IUCN Red List assessment (Gálvez, et al., Reference Gálvez, Napolitano, Ibacache, Agostini and Pliscoff2025) provides a coarse and conservative estimate of population size that may under- or overestimate the population, hence the extinction risk of the species is uncertain. Populations in some areas, particularly in the northern parts of the species’ distribution, are highly threatened by habitat loss as a result of deforestation and forest fires (Castillo et al., Reference Castillo, Plaza and Garfias2020; García et al., Reference García, Svensson, Bravo, Undurraga, Díaz-Forestier and Godoy2021; Beltrami et al., Reference Beltrami, Gálvez, Osorio, Kelly, Morales-Moraga and Bonacic2023). In Argentina the species is thought to be rare and population size estimates are needed urgently (Agostini et al., Reference Agostini, Mendizabal, Borla, Quiroga, Lambertucci and Cruz2024). Threat assessments for these populations would benefit significantly from density or abundance estimates. In the central and southern parts of its distribution, the species is also known to occupy landscapes with intensive agricultural production (Gálvez et al., Reference Gálvez, Guillera-Arroita, St. John, Schüttler, Macdonald and Davies2018, Reference Gálvez, Infante, Fernandez, Díaz and Petracca2021a) and exotic forest plantations (Pinus sp. and Eucalyptus sp.; Guzmán-Aguayo et al., Reference Guzmán-Aguayo, Magni-Pérez, González, Estades, Medel and Jaime Hernández2023). Density and/or abundance estimates in these human-modified landscapes may improve our understanding regarding the impact of human activities and facilitate evidence-based recommendations for landscape management and policy. Finally, there are records of the species in protected areas with camera traps but these are only presence data (CONAF, 2024). An account of the long-term population dynamics regarding density and/or abundance could provide evidence for the management of protected areas for the species under a variety of climate change scenarios (Cuyckens et al., Reference Cuyckens, Morales and Tognelli2015).
Some limitations with the zenith camera-trap placement may arise during sampling. For example, zenith camera traps may result in lower detection rates of species compared to lateral camera traps (Nichols et al., Reference Nichols, Glen, Garvey and Ross2017; Theimer et al., Reference Theimer, Ray and Bergman2017; Moore et al., Reference Moore, Champney, Dunlop, Valentine and Nimmo2020; Seidlitz et al., Reference Seidlitz, Bryant, Armstrong, Calver and Wayne2020). Our results showed a detection rate of 4.6 events per 100 trap-days, which is higher than the reported detection rate for güiñas in other areas using lateral cameras (3.1–4.3 events per 100 trap-days; Delibes-Mateos et al., Reference Delibes-Mateos, Díaz-Ruiz, Caro and Ferreras2014; Moreira-Arce et al., Reference Moreira-Arce, Vergara, Boutin, Carrasco, Briones, Soto and Jiménez2016; Gálvez et al., Reference Gálvez, Guillera-Arroita, St. John, Schüttler, Macdonald and Davies2018). However, this may be influenced by the sampling design or local population density. We recommend conducting a paired camera study, using one zenith camera and one lateral camera at each trapping location, to assess differences in detection rates (e.g. Smith & Coulson, Reference Smith and Coulson2012). In our study, we were able to identify individuals in 83.3% of photographic events, which is comparable to results from previous studies. For instance, Bashir et al. (Reference Bashir, Bhattacharya, Poudyal, Sathyakumar and Qureshi2013) identified 82.2% of leopard cat Prionailurus bengalensis photographs for density estimation. Although our identification rate is lower than that reported for species with more distinct markings, such as the 97.6% achieved for northern quolls Dasyurus hallucatus (Hohnen et al., Reference Hohnen, Ashby, Tuft and McGregor2013), it is higher than the 69.3% reported for more subtly marked species such as the giant panda Ailuropoda melanoleuca (Zheng et al., Reference Zheng, Owen, Nie, Hu, Swaisgood, Yan and Wei2016). In addition, testicles are not visible in zenith photographs, which makes sex determination difficult.
Zenith placement offers advantages such as always providing the same view of the animal (dorsal), regardless of its direction of movement. Additionally, by facing down, the camera lens is protected from mud and raindrops that reduce image quality (Apps & McNutt, Reference Apps and McNutt2018), which is of great benefit in ecosystems with high rainfall. Also, by having the cameras facing down, the size of the field of view area can be accurately estimated and hence utilized to fit a random encounter model (Rowcliffe et al., Reference Rowcliffe, Field, Turvey and Carbone2008), or to compare body measurements through photogrammetry (Cui et al., Reference Cui, Chen, Sun, Chu, Li and Jiang2020). We used photo editing software to create images that show clear and distinguishable dorsal patterns of each individual. An avenue for further research is the individual identification of such edited images using machine learning and deep learning methods. Finally, we anticipate that other species, particularly those inhabiting environments with well-defined trails (i.e. greater predictability of movement) and possessing distinct dorsal markings, could benefit from testing this survey method for individual identification. For instance, this may apply to species within the genus Neofelis (Christiansen & Kitchener, Reference Christiansen and Kitchener2011), most species of the genus Leopardus (Lescroart et al., Reference Lescroart, Bonilla-Sánchez, Napolitano, Buitrago-Torres, Ramírez-Chaves and Pulido-Santacruz2023), and those of the genus Prionailurus (Langle, Reference Langle2019). We encourage other researchers dealing with subtly marked species, particularly small wild cats in urgent need of population data, to assess individual dorsal patterns for identification using zenith camera placement.
Supplementary material
The supplementary material for this article is available at doi.org/10.1017/S003060532510238X
Author contributions
Study design, data analysis and writing: NG, TK, BG, EM, VA, GPM; fieldwork and funding: TK, VA, EM, BG.
Acknowledgements
This research project was funded by Reñihué Nature Conservancy Foundation and Charlie Clark. Our gratitude also goes to Consuelo Pivcevic for graphic design.
Conflicts of interest
None.
Ethical standards
This research abided by the Oryx guidelines on ethical standards. Animals were not handled or manipulated and thus the work did not require special permits by Chilean wildlife authorities. Our work was lead and carried out by members of Reñihue Foundation in their private reserve. Most authors are local, professional scientists. Because of the geographical isolation of this site, which is only accessible by air or water, we captured no photographs of people other than the camera-trapping team.
Data availability
The scoring and identification of sequences are available on request from the corresponding author.
Open access
