Movement is a fundamental characteristic of animals with each species adjusting the movement patterns differently with time and season to meet various ecological needs (Nathan et al. 2008) such as foraging (Grotta-Neto et al. 2019; Hanya et al. 2020); mating and reproduction (Mizumoto et al. 2020); thermoregulation (Spitz et al. 2018; Zollner et al. 2020); and avoidance of predators (Pokallus and Pauli 2016; Picardi et al. 2019), parasites (Kärvemo et al. 2020), and human disturbance (Tucker et al. 2018; Hertel et al. 2021). Researchers have used different methods to study animal movement. Among these, telemetry is an increasingly popular option that involves the attachment of Global Positioning System (GPS) devices, which allows collecting positional data of focal animals with high temporal and spatial resolution over several years (Hussey et al. 2015; Kays et al. 2015). This technology, however, requires capture and immobilization of the animal with additional energy expenditure while carrying the device, which raises ethical issues (McIntyre 2015) and confounds movement behavior (Dechen Quinn et al. 2012).

In cases where telemetry devices are not preferable or feasible, including studies on protected populations or elusive species, researchers have to employ other methods to collect the movement data. Direct tracking of individual animals or groups of animals is a traditional method to estimate the movement rate (distance traveled per unit time) and day range (daily travel distance) of large mammals (Matsumoto-Oda 2002; Kolowski et al. 2007). It allows detailed movement and behavioral recording without the need for capture (Peterson and Weckerly 2018; Saito and Idani 2020). This method, however, demands a high amount of time and labor, especially for species that require habituation to observe their natural movements. Consequently, records are often fragmented and collected from only a small number of focal individuals or groups, making it difficult to generalize findings to the population level. Moreover, tracking is highly challenging at night. These weaknesses are particularly evident when studying animals living in closed habitats with long day ranges.

Alternatively, motion-triggered trail cameras, often called camera traps, have the potential to indirectly elicit information on the temporal patterns of animal movement from image records. Camera trapping is now a standard method in studying animal behavior and ecology (O'Connell et al. 2011; Burton et al. 2015; Caravaggi et al. 2017). In addition to minimal impact on animal behavior, this technique allows 24-hour monitoring of terrestrial animals, albeit over a small area, and enables population-level inference by random or systematic camera placement.

Rowcliffe et al. (2016) proposed an approach to estimate the average movement rate and day range of terrestrial animals using camera traps. This method requires fine-scale measurements of recorded movement, which represents considerable labor in the field (Palencia et al. 2019). Instead, the relative variation in movement rates over time can be inferred based on the detection rate (the number of camera-trap records per unit time) of focal animal individuals or groups. Rowcliffe et al. (2014) formulated four determinants of detection rate at a given time: animal density, camera detection area size, movement rate, and active proportion (number of individuals or groups in the population on the move). It is reasonable to assume a constant animal density over the daily cycle when cameras are placed randomly or systematically. In addition, the size difference in the detection area between day and night should not be problematic for noncathemeral species. Therefore, if the constancy of active proportion can be confirmed by another method, time variation in detection rates directly translates into the temporal pattern of movement rates.

Each method used in animal movement research has its strengths and weaknesses. Therefore, it is essential to apply different approaches to the same question for reliable inferences. Comparison and integration of results from multiple methods, termed triangulation by Munafò and Davey Smith (2018), is crucial for verifying and generalizing the scientific findings (Nuñez et al. 2019). Studies on animal movement, however, have rarely applied multiple independent field methods to the same population (but see Fedigan et al. 1988; Thompson et al. 2018).

The present study examines seasonality in the temporal patterns of group movement rates of mandrills (Mandrillus sphinx) by integrating data from direct tracking and camera trapping. This endangered primate, classified as Vulnerable on the IUCN Red List (Abernethy and Maisels 2019), inhabits rainforests in central Africa and has an omnivorous diet with a preference for fruits (Abernethy and White 2013). Mandrills typically live in female-biased groups of 300 to >800 individuals (Abernethy et al. 2002; Hongo 2014) with a vast home range of ∼50 km² (White et al. 2010). Large groups travel mainly on the ground during the day and sleep in trees at night (Hoshino et al. 1984). Given that several studies have reported seasonal changes in their diet (Hoshino 1985; Nsi Akoue et al. 2017), habitat use (Hongo et al. 2018), and reproduction (Setchell et al. 2002; Setchell and Wickings 2004; Hongo et al. 2016), we expected that the movement rate patterns would also vary seasonally. Nevertheless, our knowledge of mandrill movement patterns is quite limited because the dense rainforest vegetation, large group size, and large home range size prevent continuous tracking. To date, only a tracking (Hoshino 1985) and telemetry (Brockmeyer et al. 2015) study have estimated day ranges of small groups (95–120 individuals) and reported that the groups traveled more during the fruiting season.

We hypothesized that large mandrill groups would change their temporal movement patterns in response to the seasonal environments of African rainforests. We conducted three lines of data collection and analysis. First, we employed a direct tracking method on several unidentified groups for as long as possible to record their movement patterns. The constancy of the active proportion and the seasonal difference in the group movement rate patterns were then examined. We predicted that mandrill groups seasonally change their temporal movement patterns according to fruit availability. Second, we used camera traps systematically installed in the study area to record the same mandrill groups. The seasonal difference in temporal patterns of the detection rate was examined employing the method of Rowcliffe et al. (2014). We predicted that daily patterns of the detection rates also vary with the seasonal change in fruit availability. Lastly, we compared the results from the two methods by examining the similarity in the shape of temporal patterns between the direct-tracking movement rate and the camera-trap detection rate. We predicted that if the two methods accurately capture the seasonality in daily movement behaviors, the resulting patterns from the two data sets will be similar to each other.

Materials and Methods

This study complied with American Society of Mammalogist guidelines (Sikes et al. 2016) and the Gabonese Republic laws. We conducted fieldwork with approval from the Centre National de la Recherche Scientifique et Technologique (CENAREST, N° AR0031/11/MENESRSIC/CENAREST/CG/CST/CSAR) and the Agence Nationale des Parcs Nationaux (ANPN, N° 000017/PR/ANPN/SE/CS/AEPN, N° 000022/PR/ANPN/SE/CS/AEPN).

Camera trapping

YN, SH, EFA-O, and FLM-N conducted a camera-trap survey continuously from January 2012 to February 2014. This survey was initially designed to estimate the population density of forest duikers (Nakashima et al. 2020). We used the Bushnell Trophy Cam (model numbers 119435 and 119436, Overland Park, Kansas) with a maximum detection distance of approximately 10 m and a 50° field of view. Cameras recorded a 30-s video file in response to an animal passage with a 30-s minimum interval before a new recording could be triggered.

We installed cameras at 157 stations with three to 10 stations set at regular intervals of 200 or 400 m along 20 line transects of 1–2 km (Fig. 1b). We did not intentionally select animal trails or particular vegetation types for camera stations, nor did we use bait or lures. We strapped each camera to a tree 10 m away from the transect and adjusted it to be parallel to the ground at the height of 30 cm. We then cleared the undergrowth in front of cameras to ensure detection of animal passages. We checked the cameras on a monthly basis.

Fig. 1.

Map of the study area in Moukalaba-Doudou National Park, Gabon, depicting tracking paths of mandrill groups (2009–2013) and camera-trap stations (2012–2014).

Results

Tracking data.—In the 46 tracking sessions, we tracked mandrill groups for 2.6 ± 2.0 (x̄ ± SD) km during a mean of 3.8 ±2.3 (x̄ ± SD) hours (Table 1). We never observed mandrill groups traveling through the savannah (Fig. 1).

Group active proportion was high throughout the daytime with tracking points considered to be a group pause (i.e., movement rate of <100 m/h) accounting for only 3.1% in the fruiting season and 7.5% in the nonfruiting season (Fig. 2). We observed group members sleeping >20 m high in the canopy during four sessions, of which all were after 17:44. In the early morning of the following days, we found the groups in almost the same places.

As a result of likelihood ratio tests, both the fixed interaction effect (ΔD₁ = 8.70, P = 0.013) and the random smooth effect (ΔD₁ = 9.74, P = 0.0077) of the movement rate GAMM were statistically significant. This suggests that temporal patterns of the movement rate variations randomly differed between sessions and there was a general difference between the two seasons. According to the optimal model (Table 2), the movement rate variation showed a bimodal shape in the fruiting season, with peaks at 10:10 and 15:10 and a slight trough at 12:20 (Fig. 2a). The nonfruiting season trend, in contrast, was a monotonous increase with time (Fig. 2b). The curve in the nonfruiting season did not turn downward until sunset, probably because of the small sample size near dusk. Mean day ranges predicted by the optimal model were longer in the fruiting season (6.98 km) than in the nonfruiting season (6.06 km).

Table 1.

Descriptive statistics on direct tracking (2009–2013) and camera trapping (2012–2014) of mandrill groups in Moukalaba-Doudou National Park, Gabon.

Camera-trap data.—Our terrestrial camera traps detected mandrill groups 309 times between 06:39 and 18:17 (Table 1). The shapes of temporal variation in the detection rate were statistically different between the fruiting and nonfruiting seasons (randomization test: observed overlap index = 0.829, mean null overlap index = 0.904, SD of the null distribution = 0.030, P = 0.015). In the fruiting season, the temporal variation had a bimodal shape with peaks at 08:50 and 16:20 and a trough at 14:10 (Fig. 3a). In the nonfruiting season, the shape was uni-modal with a peak at 14:50 (Fig. 3b).

Comparison between direct tracking and camera trapping.—After correcting for the movement rate effect, temporal variation patterns of the camera-trap detections changed their shapes. The shape for the fruiting season model altered only slightly: The two peaks slightly shifted towards twilight with the morning peak becoming more pronounced (Fig. 4a). For the nonfruiting season model, the afternoon peak became lower after the correction, leading to a flatter pattern throughout the daytime (Fig. 4b).

Discussion

Animal movement is a fascinating, but difficult behavior to measure noninvasively and different methods have strengths and weaknesses. We examined the seasonality in movement rate patterns of mandrill groups using two methods—direct tracking and camera trapping. Direct tracking revealed that the groups continuously moved throughout the daytime, without long pauses, and avoided the savannah. Group movement rate varied with time and season. In the fruiting season, the group day range was longer, and the temporal movement rate formed a bimodal pattern. In the nonfruiting season, the day range was shorter, and the movement rate increased monotonously with time during the day. Terrestrial camera traps detected the groups only in the daytime. The detection rate generally replicated the movement rate patterns. Correcting for the movement rate effect, the pattern of daytime detection rate flattened in the nonfruiting season, but a flattening did not occur in the fruiting season.

Fig. 2.

Temporal variations in the movement rate of mandrill groups during (a) the fruiting season and (b) the nonfruiting season in Moukalaba-Doudou National Park, Gabon. White circles show data points at which groups were moving, and filled circles represent those for temporal pauses. Bold lines indicate the predicted general trend based on the optimal model, and shaded areas represent the 95% CI. Dotted curves show the random smooth effect—predictions for each tracking session.

Table 2.

Parameter estimates of the optimal model for the movement rate of mandrill groups in Moukalaba-Doudou National Park, Gabon.

By integrating the two methods, we reliably revealed the seasonality in the movement rate patterns, as predicted. Our direct tracking showed a constantly high active proportion of the groups through the daytime (Fig. 2), suggesting that the camera-trap detection rate was a function of the movement rate alone (Rowcliffe et al. 2014). The estimated patterns of the detection rates confirmed this suggestion. Similarly for both methods, the temporal variation formed a bimodal pattern in the fruiting season and a monotonous increase in the nonfruiting season (Figs. 2 and 3). Our tracking data had a small sample size and most tracking sessions did not last throughout the daytime (Table 1). Patterns derived from this fragmented tracking data, nevertheless, have been replicated by our systematic camera trapping. This allows us to generalize our findings to the study population.

Why do mandrills change the group movement patterns between seasons? We hypothesize that they respond to seasonal changes in food resources and temperature. Slower general movement rates and a shorter day range in the nonfruiting season may be due to dietary shifts. In the fruiting season, mandrills mainly eat fresh fruits, which are clumped in trees. In the nonfruiting season, they feed on dispersed foods such as buried seeds and woody tissue in the forest litter (Hoshino 1985; Hongo et al. 2018). Large mandrill groups in our study area may need to move slower to forage for the dispersed foods during the season when the preferred foods are scarce, as observed in smaller groups in other sites (Hoshino 1985; Brockmeyer et al. 2015) and other primates (Hemingway and Bynum 2005; Tsuji 2010; Reyna-Hurtado et al. 2018; Green et al. 2020).

Fig. 3.

Temporal variations in the camera-trap detection rate of mandrill groups during (a) the fruiting season and (b) the nonfruiting season in Moukalaba-Doudou National Park, Gabon. Histograms represent the data frequency. Bold and dotted curves show the estimated kernel densities and their 95% CI, respectively.

Fig. 4.

Corrected temporal variations in camera-trap detection rates of mandrill groups during (a) the fruiting season and (b) the nonfruiting season in Moukalaba-Doudou National Park, Gabon. The weighted patterns corrected for the movement rate effect (bold lines) are compared with the original patterns (fine lines, same as in Fig. 3). Solid and dotted curves show the estimated functions and the 95% CI, respectively.

The bimodal movement pattern observed in the fruiting season may be the result of thermoregulation. The ambient temperature increases during the fruiting season, often exceeding 35°C around noon (Takenoshita et al. 2008). The observed decrease in the movement rate during this season suggests that many individuals in the group relaxed their foraging activity in midday to avoid excessive body temperature increases. Behavioral thermoregulation is crucial for mammals to maintain homeostasis (Weiss and Laties 1961; Terrien et al. 2011). Reduced movement and activity during the hottest hours are reported in many diurnal mammals, including Cabrera voles (Microtus cabrerae) (Grácio et al. 2017), Verreaux's sifakas (Propithecus verreauxi) (Erkert and Kappeler 2004), and moose (Alces alces) (Alston et al. 2020), as well as birds (Silva et al. 2015) and lizards (Foa and Bertolucci 2001).

Although the two methods exhibited similar temporal patterns in each season, our analytical comparison showed that the fruiting-season patterns were not exactly consistent between the methods (Fig. 4a). Part of this inconsistency may be because two study periods did not overlap exactly. Direct tracking was conducted between 2009 and 2013 and the camera-trap survey was carried out from 2012 to 2014. Group movement patterns were different from day to day (Figs. 1 and 2), so the noncoincident study periods may have led to the observed difference in the results. Also, the analytical methods were different (GAMM vs. kernel density function). These methodological differences may have generated the mismatch in patterns, as also reported in other studies applying multiple methods (Kamgaing et al. 2018; Steinbeiser et al. 2019; Wei et al. 2020). Notwithstanding these differences, we suggest there is a bimodal movement rate pattern in the fruiting season, but further investigation is needed to determine the intensity and time of the peaks.

An interesting secondary finding was that mandrill groups moved exclusively during the daytime. On four occasions, we observed mandrills beginning to sleep high in trees before 18:00, and the groups hardly moved until the next morning. Camera traps strongly supported these observations because they did not detect the groups between 18:30 and 06:30 (Table 1), regardless of the season. Several primate species living in much smaller groups sometimes travel on the ground at night (Hanya et al. 2018; Tagg et al. 2018), so it is surprising that the large mandrill groups totally avoided terrestrial movement at night, although they may be active in the trees (Mochida and Nishikawa 2014). Nocturnal ground movement is also reported in wild ring-tailed lemurs (Lemur catta)—a smaller-sized primate living in smaller groups than mandrills (LaFleur et al. 2014). Although mandrills are known prey of leopards (Panthera pardus) and central African pythons (Python sebae) (Henschel et al. 2011; Abernethy and White 2013), it is questionable whether this avoidance of nocturnal movement can be solely explained by predator avoidance. Future research is required to more closely study their nocturnal behavior to better understand predator–prey interactions and their influence on movements.

In conclusion, our multiple-method approach reliably revealed the seasonality in temporal movement patterns and the avoidance of nocturnal movement in mandrill groups. Direct tracking demonstrated that the groups constantly moved throughout the daytime. This enabled us to interpret the temporal variations in camera-trap detections as the movement rate variations. The absence of nighttime recordings during the extensive camera-trap survey reinforced the anecdotal observations of sedentary groups at night. The observed differences in the movement rate patterns during the two seasons suggested different responses to seasonal food resources and temperature. The strict avoidance of nocturnal movement implied predator avoidance, but further investigation is required to confirm this behavior. Because each wildlife research method has specific weaknesses, and researchers may have confirmation bias (Marsh and Hanlon 2007), applying multiple techniques to the same questions is essential for robust conclusions, particularly when studying elusive and endangered animals.