Animal reactivity to camera traps and its effects on abundance estimate using distance sampling in the Taï National Park, Côte d’Ivoire

Study area

We conducted the survey in the Taï National Park (TNP), covering an area of 5,360 km², including the N’Zo Partial Faunal Reserve (PAG-PNT, 2020-2029). The TNP is located in the southwest of Côte d’Ivoire (Fig. 1A). The TNP was listed in the world network of Biosphere Reserves in 1978 and recognized as a World Heritage of the UNESCO since 1982. In addition, the TNP is one of the largest remaining and undisturbed lowland rainforests in West Africa (Lauginie, 2007). The average annual rainfall with the park is of 1,800 mm with an average annual temperature which swings between 24 °C and 30 °C (Anderson et al., 2005).

The large biodiversity observed in the TNP includes many endemic and threatened species. The park contains at least 1,366 plant species among which 25% are endemics (Scouppe, 2011). The northern part is mainly covered by Eremospatha macrocarpa and Diospyros mannittiera trees while the southwest includes water-demanding species such as Diospyros spp. and Mapania spp (Kolongo et al., 2006). About 145 mammal species were identified, among which the African forest elephant, Loxodonta cyclotis (Matschie, 1900), the Jentink’s duiker, Cephalophus jentinki (Thomas, 1892), the Zebra duiker, Cephalophus zebra (Gray, 1838), the Diana monkey, Cercopithecus diana (Linnaeus, 1758) and the pygmy hippopotamus, Choeropsis liberiensis (Morton, 1849) (Hoppe-Dominik, 1995; Chatelain et al., 2001). Moreover, 234 bird species are known in the TNP, including the White-breasted Guineafowl, Agelastes meleagrides (Bonaparte, 1850), a species of high conservation interest (Gartshore, Taylor & Francis, 1995; Fishpool, 2001). In addition, the TNP also holds 56 species of amphibians, 42 species of reptiles and 60 species of fish divided into 20 families and 37 genera inventoried in 2012 (Rödel & Ernst, 2004; Grell, Thiessen & Kouamelan, 2013).

Data collection

Data were collected from March 2019 to March 2020 in the TNP with the approval from the Office Ivoirien des Parcs et Réserves (OIPR). The infrared CTs (Bushnell Trophy Cam HD Aggressor, Model 119877) used, were installed on trees (with a Diameter Breast Height (DBH) of about 10 cm) within a 30 m radius of the design-specified locations at a height of 50 cm above-ground, heading toward the geographical north (0° ± 20°) to avoid backlight (Howe et al., 2017; Apps & McNutt, 2018; Cappelle et al., 2021). CTs have a field of view θ = 38° and the trigger delay was set to minimum (0.6 s).

Given the size of the TNP, the study area was divided into sectors (Fig. 1B), according to the administration limits defined by the OIPR, the Ivoirian parks and reserves authority: ADK-V6: 1,020 km²; Djapadji: 1,020 km²; Djouroutou: 980 km²; Soubré: 1,040 km² and Taï: 1,300 km². In each sector, two teams (10 teams in total) installed and collected 40 CTs Bushnell Trophy Cam HD Aggressor, totalling 200 CTs used in this study. We surveyed 291 locations systematically placed 4.3 km apart for the shortest distance and 6 km for the farthest (Fig. 1C). One CT was set up on each point transect and stayed for a minimum of three months (average = 102 days, SD = 8.31) on the same location, before being relocated to a new one until covering all points in the five sectors. Out of the 291 points, 24 did not provide usable data: six were not surveyed due to their proximity to rivers or to logistical and field constraints, 15 CTs were destroyed or stolen by poachers and three CTs stopped working after two days because they recorded leaves facing the CTs. Consequently, the data were collected on only 267 point transect instead of 291. We programmed the CTs to stay active 24 h/day and to record 60s-long videos when triggered. When the CT was set, reference videos of researchers holding signs indicating distances from the CT (from 1 to 15 m at 1-m intervals, in the centre, and at each side of the CT’s field of view) were recorded at each location so that we could subsequently measure distances between the animals and the CT (see Howe et al., 2017 for details). Time of installation, habitat type, presence of rivers and fruits, Global Positioning System (GPS) location, Secure Digital (SD) card number and CT name were recorded as well for each device.

Measurements and video analysis

After processing all videos (i.e., to record the species and count the number of individuals detected, along with location, date and time of recording), we selected four medium-to-large-bodied species that displayed a range of behaviour toward the CTs (from none to many and with different way of displaying) for our analysis: the Maxwell’s duiker which is least concerned (LC), the Jentink’s duiker and the pygmy hippopotamus which are Endangered (EN), and the Western chimpanzee which is critically endangered (CR) (IUCN, 2021).

Preliminary viewing of the Jentink’s duikers and the pygmy hippopotamus videos revealed that they obviously reacted to the CTs as well as spending long time very close to them, without obvious reactions. Thanks to meticulous analyses, we pointed out other reactions to the CTs such as animals sniffing the CT. Thereafter, we identified all the reactions to the CTs device for the four species and established precise ethograms to distinguish species-specific reactions to the CTs (Table 1).

We measured distances between CTs and the individuals of four species cited above at predetermined snapshot moments two seconds apart (at 0, 2, 4, . . . , 58 s after the even minute) as previously done (Howe et al., 2017; Bessone et al., 2020; Cappelle et al., 2019; Cappelle et al., 2021). To do so, the animal and the reference videos were compared to measure the distance between the animal and the CTs. As longer distances were more difficult to measure precisely than closer distances to the CTs, videos with animals more than 8 m from the CTs were assigned to one of the following categories: 8–10 m, 10–12 m, 12–15 m. At the same time, we also recorded the behaviour of all the animals, based on our ethogram.

Quantifying the bias in abundance estimation

To evaluate the bias resulting from fully or partially ignoring animal response to the CTs, we proceeded in three successive steps, (first) we estimate population size using all video clips of an animal species; then (second) we excluded the videos presenting obvious responses to the CTs (like was done by Bessone et al., 2020), and finally we excluded all videos presenting species-specific reactions to the CT following our ethogram (Table 2).

All densities were estimated by applying the following formula (Howe et al., 2017): ${\displaystyle \hat{D}=\frac{\sum _{k=1}^K{n}_k}{\pi w^2\sum _{k=1}^K{e}_k{\hat{P}}_k}\ast \frac{1}{A}}$ where “n_k” is the number of observations of distance between animals and CT “k”, “w” is the maximum distance between objects and CT or truncation distance, “${\hat{P}}_k$” is the estimated probability of an image of an animal that is within “θ” and “w” front of the CT at a snapshot moment and “$e_k=\frac{\theta T_k}{2\pi t}$” is the sampling effort at CT location “k”. Here, “T_k” is the sampling effort at CT “k” in seconds (all operational days, excluding days of CT installation and retrieval, were considered when calculating survey effort), “t” is the time interval of predetermined snapshots (set to 2 s), “θ” is the horizontal viewing angle of the CT in radians (i.e., 0.663), so “$\frac{\theta }{2\pi }$” represents the proportion of a circle covered by the CT (i.e., 0.106), and “$\frac{1}{A}$” is the availability correction factor defined by the proportion of time when individuals of a given species are available for detection.

A precisely-determined availability correction factor is necessary to avoid important biases in abundance estimates (Cappelle et al., 2019). The Maxwell’s duiker is active mostly during the day, and at the same time is very abundant in TNP (22,101 videos). Therefore, we decided to include only data from the peak of activity between 7:00 and 7:59 (Howe et al., 2017; Cappelle et al., 2021) when the availability correction factor is 1. The Jentink’s duiker which is nocturnal species, was considered to be fully available for the CTs between 18:00 to 5:59 (correction factor equal 1) and only the data from this period were analysed (>90% videos). The survey duration was defined as 12 h 59 min 59 s.

The pygmy hippopotamus is a rather nocturnal species with 12% of videos during the day. It was not available for detection at all times because of resting in the shallows and rivers. Similarly, the chimpanzees, diurnal and semiarboreal species, were also not available for detection at all times at the day. Therefore, we defined “T_k” as 24 h and estimated the proportion of time when pygmy hippopotamus and chimpanzees were available for detection using method described in Rowcliffe et al. (2014). We calculated the animal’s availability (A) using the r package “activity” (R Core Team, 2019) to determine the availability correction factor of these species.

We estimated the density using candidate models include the half-normal detection function with 0 and 1-hermite polynomial adjustment terms, the uniform detection function with 1 and 2-cosine adjustment terms and the hazard-rate detection function with 0, 1 and 2-cosine adjustment terms.

The best model was selected following a two-step procedure proposed by Howe et al. (2019). We selected the preferable models by comparing the adjusted model selection criteria (QAIC) scores within each key function. Then, we compared the goodness of fit test (GOF) statistic of the best models with different key function. We estimated variances using a nonparametric bootstrap, resampling points with replacement (Howe et al., 2017).

Following the exploratory analyses, we left-truncated the data when fewer detection near the CTs were recorded than expected, and right-truncated to eliminate the 10%–15% (approximately) of the farthest observations (Buckland et al., 2001). Therefore, longer distances were difficult to measure accurately, and we have very little data between 12–15 m (less than 1%) for Maxwell’s duiker, Jentink’s duiker and pygmy hippopotamus, either because of small size (Maxwell’s duiker) or a nocturnal behaviour (Jentink’s duiker and pygmy hippopotamus). Therefore, for these species we right-truncated distance observations > 12 m (Fig. 2). For pygmy hippopotamus, given its size, both body length and width were considered when measuring radial distances, so we combined distance observations into 2-m intervals. Most models of the detection function did not fit well for chimpanzees. In such cases, it is reasonable to left-truncate data for achieving accurate estimates (Buckland et al., 2001). We achieved a reasonable fit only when we left-truncated at 1-m and we combined distance observations of 1 to 5-m and 5 to 7-m (Fig. 2).

We finally estimated the population size by multiplying the estimated density by the study area size (5,360 km²), mentioned above.

Factors influencing animal’s reaction to camera trap

We attempted to determine which variables influenced the animals’ reaction to CTs. We hypothesised that CTs could be detected by animals for the following reasons: (1) olfactory signals (metal, plastic and human odours on the device), (2) visual signals (neophobia towards foreign objects and infrared LED illumination by night) and (3) auditory signals (emission of sounds from the electronic and mechanical components of the device) (Séquin et al., 2003; Larrucea et al., 2007; Meek et al., 2014; Caravaggi et al., 2020).

To measure whether the animal reacts to the CTs due to olfactory signals, we used the variable “CT duration on site” (i.e., the time elapsed between the date the CT was installed and the date the animal was recorded). The odour of the device itself or the odour left by the team on the device and on the nearby vegetation can remain for days. So, if animals react because of the odour of the device, then the effect should decrease the longer the CT stays in the field.

To determine whether the animal visually detected the CT, we used the distance to the CT (i.e., the distance at which the animal was located from the CT) and the time of the day (i.e., the time at which the reaction to the CT was observed). The time of the day was included for the species that were active at night (i.e., the Jentink’s duiker and the pygmy hippopotamus). If animals reacted to the CT because they detected it visually, we supposed that the probability to react should decrease the further away the animal was from the CT. If animals reacted to the LEDs of the CTs, which were only visible at night, then the reaction level was higher by night. Unfortunately, we have not witnessed behaviour that were representative of auditory reaction (e.g., freezing), so we did not analyse this factor.

We tested those variables using generalized linear models (GLMs) with binomial error structure and logit link function (McCullagh & Nelder, 1989) for each of the species. The three predictors were those described hereabove and the response was the reaction to CT (with reaction (1) or without reaction (0)). Since the time of the day is a circular variable, we transformed the time of the day into radian and included its sine and cosine into the model (Stolwijk, Straatman & Zielhuis, 1999).

Variance inflation factor (VIF; Field, 2005) were derived using the function VIF of the r-package “car” (Fox & Weisberg, 2019) applied to a standard linear model excluding the random effects. Thus, a variable is only retained in the model when its VIF is below the threshold value of 2.5 (Allison, 1999).

To establish the significance of the full model (Forstmeier & Schielzeth, 2011), we used a likelihood ratio test (Dobson, 2002) comparing its deviance with that of the null model comprising only the intercept. We considered a fixed effect being significant when the P-value is inferior to 0.05.

Reaction to the camera trap

Maxwell’s duikers only showed attraction to CTs in 14 observations representing 0.11% of the sample (Table 3). On the contrary, Jentink’s duikers strongly reacted to the CTs. Obvious reactions of avoidance and attraction represented 19.25% of the observations, while the cases of inspection behaviour represented 13.60% of the observations i.e., when the animal sniffed in all directions mainly towards the CT or sniffed branches cut or touched by field teams during CT installation (Table 1). Pygmy hippopotamus also strongly reacted to the CT but less visibly as 36.64% of the reactions were inspection behaviours whereas the obvious reactions constituted 16.31% of the data (Table 3). Finally, Western chimpanzees showed a similar rate of reaction to the pygmy hippopotamus regarding attraction and avoidance behaviour (15.23%) but a lower rate of inspection behaviour (1.17%) (Table 3).

After discarding all reactions to the CTs, the data including normal behaviour represented 99.89% of the total observations for the Maxwell’s duiker, 67.16% for the Jentink’s duiker, 47.04% for the pygmy hippopotamus and 83.60% for the Western chimpanzee (Table 3).

Quantifying the bias in abundance estimate

Density estimates of Maxwell’s duikers indicated 21.19 ind./km² keeping all reaction behaviours to CTs and 21.17 ind./km² when excluding reaction behaviours. Hence there was a slight overestimation of the density of 0.02 ind./km² (107 individuals).

Concerning the Jentink’s duikers, the analysis of all data including reactions provided an estimate of 0.75 ind./km². The density decreased to 0.68 ind./km² when excluding only attraction and avoidance behaviour and dropped to 0.26 ind./km² when excluding all reactions. Therefore, if we had kept the implicit reactions, we would have overestimated the density of 0.46 ind./km². Similarly, regarding the pygmy hippopotamus, the determined availability was 49% over a 24-hour period. The density estimate was 0.81 ind./km² including all reactions, 0.59 ind./km² when excluding the attraction and avoidance of the CT and 0.23 ind./km² when excluding all reactions.

Finally, the chimpanzees were available for detection 33% of the time every day. Using the availability correction factor, we found a density of 0.38 ind./km² when considering all reactions, 0.26 ind./km² without obvious reactions and 0.23 ind./km² excluding all reactions. We observed an overestimation of density of 0.15 ind./km² between analyses that included all reactions to CTs and those that excluded all reaction behaviours to CTs.

The abundance estimates remained precise for Maxwell’s duikers with a coefficient of variation (CV) of 10% for the three analyses while an increasing of CV was observed for the other three species, particularly for Western chimpanzees for which the CV increased from 30% to 43% (Table 2).

Factors influencing animal’s reaction to camera trap

The VIF of each of the different explanatory variables was less than 2.5, i.e., the threshold value. All the full-null model comparison of the four species were significant (Maxwell’s duiker: χ² = 47.02, p < 0.001; Jentink’s duikers: χ² = 1032.53, p < 0.001; pygmy hippopotamus: χ² = 1110.42, p < 0.001; Western chimpanzee: χ² = 509.67, p < 0.001).

All species reacted less to the CTs at longer distances than closer ones (Table 4; Fig. 3). In fact, Maxwell’s duikers only reacted when located in the first meter from the CT and with a probability of reaction of less than 10%. On the contrary, the probability to react close to the CT was higher for the Jentink’s duiker, pygmy hippopotamus and Western chimpanzees with 90%, 80% and 65% respectively. Western chimpanzees reacted mainly within 5 m from the CT, while the pygmy hippopotamus and the Jentink’s duiker showed reaction even farther away (beyond 5 m), with probability of reaction slowly decreasing with distance from the CTs (Fig. 3).

Both Jentink’s duikers and pygmy hippopotamus reacted significantly more at night than during the day. All the full-null model comparison of Jentink’s duikers and pygmy hippopotamus were significant (Jentink’s duikers: χ² = 83.7, p < 0.001; pygmy hippopotamus: χ² = 29.7, p < 0.001) (Table 4; Fig. 4).

The effect of the time CTs were on the field on the animal reaction was significant for all the species except for the Maxwell’s duiker (p = 0.96; Table 4; Fig. 5) and differed depending on the species. During the first days the CT was set up, the reaction probability for the Jentink’s duiker remained high, 70% of observations but decreased slightly over time. The reaction probability observed for the pygmy hippopotamus was 65% during the first days the CT was set up and slowly decreased until after 100 days from CT installation. The probability of reaction of Western chimpanzees remained low (<20%) despite a significant decrease with the time (Fig. 5).