The African golden cat Caracal aurata is an elusive, medium-sized felid categorized as Vulnerable on the IUCN Red List, regarded as the least-known wildcat in Africa and one of the least-studied felines (Bahaa-El-Din et al., Reference Bahaa-El-Din, Mills, Hunter and Henschel2015b). As the only truly forest-dwelling feline in Africa, it is threatened by deforestation, habitat destruction and snaring (Bahaa-El-Din et al., Reference Bahaa-El-Din, Sollmann, Hunter, Slotow, Macdonald and Henschel2016). Although the population is believed to be decreasing (Bahaa-El-Din et al., Reference Bahaa-El-Din, Henschel, Butynski, Macdonald, Mills, Slotow and Hunter2015a), robust knowledge of the species is scant. The golden cat occurs in West and Central Africa, with an easternmost population recently confirmed in Kenya from a roadkill in the Aberdare Mountains (Hatfield et al., Reference Hatfield, Mwaura, Musila and O'Meara2019). Western Tanzania has been regarded as potentially suitable for the species (Butynski et al., Reference Butynski, Douglas-Dufresne and de Jong2012), but no records have previously been reported. Here, we present the first record and results of habitat association modelling for a population of golden cat found in Tanzania in the course of a biodiversity survey in Minziro Nature Forest Reserve. We used a camera-trap survey, and occupancy analysis, to determine the distribution of the golden cat and its association with habitat edges and human disturbance.

Camera trapping was conducted in Minziro Nature Forest Reserve, north-west Tanzania, during 3 October–28 December 2018. Established in 2016 and formerly a reserve with lower protection status, it comprises 257 km² of flat, moist forest at a mean altitude of 1,150 m (Rovero et al., Reference Rovero, Shinyambala, Perkin and Greco2019). The northern boundary is the country border, merging with the Sango Bay Forest Reserve in Uganda. All other boundaries adjoin human-modified areas with settlements, plantations and patches of drier woodland and grassland (Fig. 1), with Kagera river and seasonal flooding to the east, and a paved road and several villages to the west. The Reserve is the easternmost extension in Tanzania of the Congo–Guinea forest, with fauna typical of West and Central Africa. The Reserve is heavily disturbed by illegal logging, burning, livestock grazing and fishing occurring in the area (Rovero et al., Reference Rovero, Shinyambala, Perkin and Greco2019). We found evidence of bushmeat hunting: we encountered hunters with dogs on one occasion and found a number of snares, presumably set for ungulates but potentially a threat to the golden cat and other species.

Fig. 1 Minziro Nature Forest Reserve, Tanzania, with camera-trap sites and records of the African golden cat Caracal aurata.

We surveyed 70 sites across the Reserve using a regular grid of 1–2 km² cell size, firstly surveying 40 camera sites and then 30 (Fig. 1), for a minimum of 30 days (mean 32 days) each. We used Browning (Morgan, USA), Cuddleback (Green Bay, USA) and UOVision (Shenzhen, China) cameras, with infrared flash and motion-activated sensor. The trigger speed was 0.15 s for Browning, and 1–2 s for Cuddleback and UOVision. Cameras were unbaited and attached to trees at c. 50 cm from the ground, facing a presumed animal trail.

The survey yielded 2,219 camera-days from 68 camera traps (two were stolen). We annotated images using Wild.ID and ran analyses using R 3.6 (R Core Team, 2019). We calculated the number of golden cat detections per day, and used them to derive a relative abundance index. We plotted the species’ daily activity pattern using hourly events. We then used single-season occupancy modelling (MacKenzie et al., Reference MacKenzie, Nichols, Lachman, Droege, Royle and Langtimm2002) to estimate golden cat occupancy (ψ) and detection probability (p) and their association with presumed covariates. We arranged golden cat detections/non-detections into a site-by-sampling occasion matrix; we chose 5 days as the resolution of the occasions as with poorly detected species this provides the best estimation of detectability, and hence occupancy (Rovero & Spitale, Reference Rovero, Spitale, Rovero and Zimmermann2016). We built parsimonious models using the distance from each site to the closest Reserve border (i.e. forest edge) as a covariate. We derived these distances using QGIS (QGIS Development Team, 2019) and a 1:100,000 scale map. Based on our survey of human disturbance, we considered greater distance to the border as a proxy for increasing habitat quality and decreasing human disturbance. We assumed that both ψ and p increase with distance to border. For detection probability we also considered that camera-trap model could potentially influence detection. Using unmarked (Fiske & Chandler, Reference Fiske and Chandler2011), we built all model combinations and ranked them according to the Akaike information criterion (AIC). We considered statistically supported models with ΔAIC < 2.00 and used AICcmodavg (Mazerolle, Reference Mazerolle2019) to average parameter estimates.

After discarding 3,297 blank and set-up images, camera trapping yielded 4,244 photographs of 23 species of medium to large wild mammals (Supplementary Table 1), including 169 images of golden cats. These corresponded to 33 independent detection events of the species within 24-hour intervals at 17 sites (naïve occupancy = 0.26; Fig. 1). Based on individual pelage marks, we identified a minimum of 10 individuals. Pelage colouration varied from dark brown to sandy and golden (Plate 1). Animals were mainly active at night, with activity peaking at 19.00, 22.00 and 3.00, but were also detected in daylight (Supplementary Fig. 1). Based on occupancy model ranking (Table 1), we averaged the two most supported models, and the resultant mean estimated occupancy was 0.41 ± SE 0.12. Occupancy increased significantly with distance to the Reserve border (1.30 ± SE 0.54; Table 2, Fig. 2). Mean detection probability was 0.13 ± SE 0.05 and did not vary significantly with distance to the Reserve border but did vary between camera models, with Browning cameras having significantly higher values (Supplementary Table 2).

Fig. 2 Estimated occupancy (dotted lines show SE) of the golden cat in Minziro Nature Forest Reserve in relation to distance from the closest forest edge.

Plate 1 African golden cat Caracal aurata photo-trapped in Minziro Nature Forest Reserve, Tanzania. The survey revealed golden/brown-reddish (a, c) and dark/light grey (b, d) morphotypes.

Table 1 Model ranking for occupancy (ψ) and detection probability (p) of the African golden cat Caracal aurata detected by camera trapping in Minziro Nature Forest Reserve (Fig. 1), Tanzania.

¹ Models were ranked according to the Akaike information criterion (AIC), with ΔAIC < 2 considered supported.

Table 2 Parameter estimates from the averaging of the two best supported models testing the effect of covariates on detection probability (p) and occupancy (ψ) of the African golden cat in Minziro Nature Forest Reserve (Fig. 1).

Our findings suggest that habitat intactness and human disturbance affect golden cat occurrence, with increased occupancy with increasing distance from the Reserve border and human settlements. This matches the feline's known sensitivity to anthropogenic disturbance and its vulnerability in human-disturbed areas (Martinez Marti, Reference Martinez Marti2011; Bahaa-El-Din et al., Reference Bahaa-El-Din, Sollmann, Hunter, Slotow, Macdonald and Henschel2016). We documented a high incidence of human activities around and within the Reserve, potentially explaining the species’ relatively higher occurrence in areas with dense, closed-canopy, continuous forests. A similar pattern has also been reported in Uganda (Mugerwa et al., Reference Mugerwa, Sheil, Ssekiranda, van Heist and Ezuma2012). Detection probability was greater with faster cameras, highlighting the importance of standardizing camera models in wildlife surveys: the best performing devices increase the probability of detecting elusive species. Although not exclusively nocturnal, the golden cat activity pattern showed peaks at night, mirroring the cathemeral behaviour reported by Bahaa-El-Din et al. (Reference Bahaa-El-Din, Henschel, Butynski, Macdonald, Mills, Slotow and Hunter2015a). Peaks of golden cat activity may overlap with periods of low human presence and higher prey activity. The minimum number of 10 individual cats detected gives a naïve density of 4 per 100 km², suggesting that this population may not be as abundant as in other areas. Densities of 16 individuals per 100 km² have been reported in pristine habitats in Gabon, and densities similar to our estimates have been reported in highly disturbed and hunted areas (Bahaa-El-Din et al., Reference Bahaa-El-Din, Sollmann, Hunter, Slotow, Macdonald and Henschel2016). It is possible that the presence of seasonally flooded forest and swamps, mainly on the eastern side of Minziro Nature Forest Reserve, make this habitat type suboptimal for the golden cat.

This is the first record of the little-known golden cat in Tanzania, c. 200 km east-south-east from the closest known populations in Uganda and Rwanda, potentially confirming anecdotal reports of the species in the contiguous Sango Bay Forest in Uganda in the mid 1990s (T. Davenport, pers. comm., 2019). Pelage pattern suggests the population in Tanzania may be the subspecies Caracal aurata aurata, with spots only on the abdomen and limbs (Bahaa-El-Din et al., Reference Bahaa-El-Din, Henschel, Butynski, Macdonald, Mills, Slotow and Hunter2015a). Given its isolation and the heavily modified habitat, we assume this population is genetically isolated, and thus Minziro Nature Forest Reserve may be an important site for the species’ conservation. We also recorded the tree pangolin Phataginus tricuspis and giant pangolin Smutsia gigantea, both Endangered, and the first record in Tanzania of the fire-footed rope squirrel Funisciurus pyrropus. Considering the human encroachment we detected, and ineffective law enforcement, these findings indicate the need for appropriate protection of this forest.