This study was conducted at the Mengla and Shangyong subreserves of the Xishuangbanna National Nature Reserve and Yiwu Prefectural Nature Reserve in Mengla County, Xishuangbanna Dai Autonomous Prefecture, Yunnan Province, southwestern China (21°08′~22°25′ N, 100°50′~101°06′ E). This region borders Laos and harbors rich biodiversity and provides important habitats for some key protected animal species in China, such as Indochinese tiger (Panthera tigris corbetti), Indochinese leopard (Panthera pardus delacouri), Asian elephant (Elephas maximus), clouded leopard (Neofelis nebulosa), Owston’s civet (Chrotogale owstoni), as well as Williamson’s mouse deer, yet anthropogenic disturbance is severe. Large amounts of tropical forests in this region have been displaced by cultivation (i.e., croplands and rubber plantations) and urbanization (e.g., cities, towns, villages, roads) since the 1950s (Zhang et al., 2019), which potentially affects the habitat use of Williamson’s mouse deer and other ungulate species in this region.

We used camera traps to monitor the ungulate species in the three protected areas (Mengla, Shangyong, and Yiwu) in Mengla County from January 2017 to January 2021 (Mengla: from January 2017 to June 2020; Shangyong: from November 2018 to January 2021; Yiwu: from July 2017 to October 2020). A total of 238 camera-trapping sites were set up along transects ( Figure 1 ). The transects primarily followed abandoned human or mammal trails, with the inclusion of sampling sites from the core to the edge of each of the protected areas. To ensure sampling independence, the minimum distance between any two adjacent camera trap sites that were concurrently operating was 500 m. Camera traps (Loreda L710, Yi’an Weishi Technology Co, Ltd, Shenzhen) were mounted at 0.5–2 m aboveground on the trunk of trees without baits or lures. The monitoring system of the camera traps was set to work 24 hours per day with high sensitivity, and three photographs were taken at each trigger without a delay. To reduce the disturbance of the observers, the batteries and SD cards for storing animal photographs were changed only every six months. After we collected the camera photographs, we recorded for each photograph the GPS location, species names (including livestock and humans), capture dates, and capture times within the day.

We used the relative abundance index (hereafter RAI) of the larger ungulates (RAI of northern red muntjac, RAI of wild boar, RAI of Chinese serow, RAI of sambar and RAI of all ungulates) as the major variables. Moreover, to eliminate the false signal of spatial interaction between Williamson’s mouse deer and the sympatric larger ungulates (Pollock et al., 2014; D’Amen et al., 2018), we added additional environmental covariates that can potentially affect the occupancy of Williamson’s mouse deer into the analyses. A total of 12 site-level covariates were selected (see more detailed description and data resources in Table S1 ), including two covariates that could reflect habitat quality (tree density and normalized difference vegetation index, NDVI), four covariates that could reflect topographical variation (elevation, slope, aspect and topographic position index, TPI), and six covariates that represent different types of anthropogenic disturbances (distance to nearest road, distance to nearest built-up area, distance to nearest cropland, distance to nearest plantation land, RAI of local people presence, and RAI of livestock presence. We also added site-level annual average temperature to explain the variation in detection probability because temperature has a potential effect on the detection capability of camera traps, and is often considered to influence detectability (Wang et al., 2018; Wang et al., 2021). All covariates were standardized with a mean of 0 and a standard deviation of 1.

Consecutive photographs of a given species at the same camera site within 30 minutes were not considered independent detections to avoid overestimating true occurrences of animals (O’Brien et al., 2003). The RAI of each species in each camera site was denoted by the number of independent detections recorded within 100 trapping days (Jenks et al., 2011), as it is an efficient index to represent the abundance of forest ungulates (Rovero and Marshall, 2009). Matrixes of detection history with 30 days as a sample segment for analysis were created for each species using the camtrapR package (Niedballa et al., 2016) for occupancy modeling. All analyses were conducted with R software 4.1.0 (R Core Team, 2022).

To examine how the relative abundance of sympatric larger ungulates affects the occupancy of Williamson’s mouse deer, single-species occupancy models were implemented by using the occu function in the unmarked package (Fiske and Chandler, 2011). Five full models were established with annual average temperature as the detection variable, and then differing occupancy variables: 12 covariates, 12 covariates + RAI of northern red muntjac, 12 covariates + RAI of wild boar, 12 covariates + RAI of Chinese serow and 12 covariates + RAI of sambar. All variance inflation factor (VIF) values of the occupancy variables were under 5 in the full models, as measured by the vif function, which indicates little collinearity (Akinwande et al., 2015). After the collinearity check of each full model, the candidate models were established with all possible combinations of the occupancy variables and then compared based on the Akaike information criterion (AIC) (Burnham and Anderson, 2002). Models with a ΔAIC < 2 from the top model (the model with the lowest AIC) were selected for weighted model average calculation by using the model.avg function in the AICcmodavg package (Mazerolle, 2020). The full averaged coefficients and 95% confidence intervals were finally reported. The variables were considered significant if their 95% confidence intervals did not cross 0 and the full averaged P value was < 0.05. We also modeled the four ungulate species separately with the same model selection procedure (only with the 12 covariates) in order to provide the precondition of establishment for the following two-species occupancy models (see below). For model validation, we ran goodness-of-fit tests, that is, calculating the Sum of Squared Errors, Freeman-Tukey Chi-squared test, and Pearson’s Chi-squared test with 1,000 bootstraps for the top model, using the parboot function (Fiske and Chandler, 2011). All P-values of goodness-of-fit tests were greater than 0.05, which supports the goodness of fit for our occupancy modeling ( Table S2 ).

We further used conditional two-species occupancy models to determine the spatial interaction between Williamson’s mouse deer and the larger ungulate species (Richmond et al., 2010). This model categorizes the species pair as a dominant competitor A and a subordinate competitor B, and allows the estimation of three occupancy parameters, ψA, ψBA, and ψBa, and 5 detection parameters, pA, rA, pB, rBA, and rBa (see the details in Richmond et al., 2010). The species interaction factor (SIF) can be calculated by the three occupancy parameters, as follows:

SIF = 1 suggests that the species pair is spatially independent, SIF > 1 suggests that species B is more likely to co-occur with species A than expected under the hypothesis of independence (i.e., spatial aggregation), and SIF < 1 suggests that species B is less likely to co-occur with species A than expected under the hypothesis of independence (i.e., spatial avoidance). We modeled the species pairs composed of Williamson’s mouse deer and each of the sympatric ungulate species. We considered the larger ungulates as the dominant species (species A) and Williamson’s mouse deer as the subordinate species (species B), and established four candidate models for each species pair, corresponding to four hypotheses: 1) the occupancy and detection of Williamson’s mouse deer are affected by the presence of larger ungulates, while the larger ungulates are independent from Williamson’s mouse deer (ψA, ψBA, ψBa, pA = rA, pB, rBA = rBa); 2) only the occupancy of Williamson’s mouse deer is affected by the presence of larger ungulates, while the larger ungulates are independent from Williamson’s mouse deer (ψA, ψBA, ψBa, pA = rA, pB = rBA = rBa); 3) only the detection of Williamson’s mouse deer is affected by the presence of larger ungulates, while the larger ungulates are independent from Williamson’s mouse deer (ψA, ψBA = ψBa, pA = rA, pB, rBA = rBa); and 4) Williamson’s mouse deer and the larger ungulates are independent from each other (ψA, ψBA = ψBa, pA = rA, pB = rBA = rBa). We did not hypothesize the effect of larger ungulate detection on Williamson’s mouse deer’s detection since we used a long sample segment (30 days). The detection variable, annual average temperature, and the significant occupancy variables from single-species occupancy models were added to each candidate two-species occupancy model. The candidate models with ΔAIC < 2 from the top model were selected for weighted model average calculation. The mean predicted value ± conditional 95% confidence intervals of the parameter’s coefficients were finally reported. Williamson’s mouse deer was considered significantly aggregated (or avoided) with the presence of larger ungulates if the conditional 95% confidence intervals were higher (or lower) than 1. All the two-species occupancy models were conducted by using the program PRESENCE (https://www.mbr-pwrc.usgs.gov/software/presence.html).

Nonparametric kernel density estimations were used to examine the overlap of the daily activity patterns of Williamson’s mouse deer and the sympatric larger ungulates. The captured clock time of day of independent events was transformed to circular solar time by using the solartime function in the activity package (Rowcliffe, 2022). The overlap of the activity pattern was measured by calculating the coefficient of overlap D-hat (Δ) with the overlapEst function in the overlap package (Ridout and Linkie, 2009), in which a range from 0 (no overlap) to 1 (complete overlap) was generated. Δ4 was used since the number of detections of all species was > 50. The overlap was considered high when Δ > 0.75, moderate when 0.5 < Δ < 0.75, and low when Δ < 0.5 (Monterroso et al., 2014). We also performed randomization tests by using the compareCkern function to test the significant difference in the daily activity patterns of Williamson’s mouse deer and the larger ungulates. We repeated the analysis for Williamson’s mouse deer with the different species of larger ungulates.

We performed time interval analyses to understand whether Williamson’s mouse deer displays direct temporal avoidance of the sympatric larger ungulates in their co-occupied sites. Two types of time intervals were calculated at each camera trap site: 1) the time interval from the detection of a given ungulate species to the next detection of Williamson’s mouse deer, and 2) the time interval between two successive detections of Williamson’s mouse deer ( Figure 2 ). Because of nonnormality, one-tailed Wilcoxon rank sum tests were performed to determine whether the time intervals of a given ungulate species to the next detection of Williamson’s mouse deer were significantly longer than those of two successive detections of Williamson’s mouse deer. A significant test result suggests that interspecific temporal avoidance exceeded temporal avoidance between conspecifics (Harmsen et al., 2009; de Oliveira et al., 2020), which would imply that Williamson’s mouse deer does show direct temporal avoidance of the given ungulate species. We repeated these analysis procedures for Williamson’s mouse deer with the different larger ungulate species. We also calculated the avoidance ratios by using the time interval from the detection of a given ungulate species to the next detection of Williamson’s mouse deer divided by that between two successive detections of Williamson’s mouse deer to determine the difference in the avoidance strength of Williamson’s mouse deer to different-sized larger ungulate species.

The results of single-species occupancy models showed that the occupancy probability of Williamson’s mouse deer was only significantly associated with elevation, in which it was higher in the lowlands (coefficient [95% CI] = −1.26 [−1.97, −0.56], P< 0.001), when interactions with other ungulates were not included ( Table S3 ). The relative abundance of all larger ungulate species had no significant effects on the occupancy probability of Williamson’s mouse deer (northern ren muntjac: coefficient [95% CI] = 0.4 [−0.026, 0.83], P = 0.066; wild boar: coefficient [95% CI] = 0.026 [−0.19, 0.25], P = 0.81; Chinese serow: coefficient [95% CI] = −0.5 [−1.83, 0.83], P = 0.46; sambar: coefficient [95% CI] = 0.0046 [−0.11, 0.12], P = 0.94) ( Figure 3 ).

We did not deploy two-species occupancy models to analyze the avoidance of Williamson’s mouse deer to the presence of northern red muntjac and wild boar. This is because its recorded sites were fully covered by those of the two larger ungulates, which suggested no avoidance of them. The averaged estimates of the two-species models showed that Williamson’s mouse deer had significant aggregation with Chinese serow (SIF [95% CI] = 1.041 [1.019, 1.064] > 1), and it was independent from the presence of sambar (SIF = 1) ( Table 2 ).

The daily activity pattern of Williamson’s mouse deer was significantly different from that of the sympatric larger ungulates (all P< 0.001), with moderate overlap with northern red muntjac (Δ4 = 0.63), wild boar (Δ4 = 0.66) and Chinese serow (Δ4 = 0.54) and low overlap with sambar (Δ4 = 0.44) ( Figure 4 ). The daily activity pattern of Williamson’s mouse deer was typically crepuscular with 2 peaks at 9:00:00 and 18:00:00 and a trough at 13:00:00. The other four species of larger ungulates had varied degrees of nocturnal activity, with inapparent peaks in the morning and twilight. In particular, sambar was mainly active at night without any obvious peaks and only had an obvious trough in the afternoon.

The time intervals of the detection of any larger ungulate species to the next Williamson’s mouse deer were significantly longer than those of the successive detections of Williamson’s mouse deer (all P< 0.05) ( Figure 5 ), which means that the interspecific temporal avoidance exceeded temporal avoidance between conspecifics, suggesting that Williamson’s mouse deer do display direct temporal avoidance to these species. Moreover, the strength of avoidance to Chinese serow (avoidance ratio = 3.36) > sambar (avoidance ratio = 3.026) > northern red muntjac (avoidance ratio = 1.96) > wild boar (avoidance ratio = 1.53).

It is usually considered that spatial avoidance is more frequent and important for alleviating interspecific competition than the other two major niche axes, i.e., resource and time (Schoener, 1974). Spatial avoidance would seem to indicate such intense competition that the species are unable to be present together at the same location, whereas in other forms of competition the species are able partition the resources, or temporal access to them. However, our results showed the occupancy probability of Williamson’s mouse deer was not negatively associated with the presence and relative abundance of the larger ungulates, which indicates that Williamson’s mouse deer did not avoid the larger ungulate species spatially. This suggested that spatial avoidance may not play a major role in the coexistence of Williamson’s mouse deer with the sympatric larger ungulates. This lack of spatial avoidance could largely be explained by temporal avoidance (i.e., differing in daily activity patterns and direct temporal avoidance; see below) (Kronfeld-Schor and Dayan, 2003). In addition, the differentiation in dietary niches may be another important reason that facilitates the steady coexistence of sympatric competitors (du Preez et al., 2017).

The diet of Williamson’s mouse deer is unclear (Meijaard, 2011), but it should be much different from the larger ungulates due to the extreme difference in body size (Cromsig and Olff, 2006). From interviews of local villagers in Mengla, a previous paper described the food selection of Williamson’s mouse deer to mainly include fallen fruits of wild figs (Ficus spp.), Phyllanthus emblica and Phyllanthi fructus, and the tender leaves of Alpinia kwangsiensis and Phrynium capitatum (Cao et al., 2010). If it is similar to other species of mouse deer, such as Javan mouse deer (Tragulus javanicus) and lesser mouse deer (T. kanchil), Williamson’s mouse deer’s diet may generally consist of small fallen fruits from Ficus, shoots, tender leaves from understory plants and small mushrooms (Matsubayashi et al., 2003; Meijaard, 2011). In contrast, the diets of larger ungulates are wider and more opportunistic (Ahrestani et al., 2016). The four larger ungulate species can also consume fallen fruits and understory plants, but they are not restricted to items that are of low height and small size, as are the small deer, and can consume more species of plants or other food items. For example, sambar can browse tender leaves, twigs and large fruits of larger plants (Ahrestani et al., 2016); wild boar even frequently consume small animals, such as earthworms, snakes, frogs, small rodents, and ground-dwelling birds (Ballari and Barrios-García, 2014). A previous study in Singapore showed that the presence of wild boar and sambar had no effect on the occupancy of lesser mouse deer, which may be partially explained by different food selection (Khoo et al., 2021).

The occupied sites of Williamson’s mouse deer were all covered by northern red muntjac and wild boar, and it was statistically aggregated with the presence of Chinese serow, which suggests that Williamson’s mouse deer is positively associated with these species in its space use. It is possible that Williamson’s mouse deer could reduce predation risk by co-occurring with the important prey of larger carnivores according to the “dilution effect” hypothesis (Dehn, 1990). In particular, the northern red muntjac is much more abundant than other species in our study area, and the number of independent detection events is almost three times (2.95 times) that of the second most abundant species, the wild boar ( Table 1 ), and muntjac may be the major prey of the Asian golden cat (Catopuma temminckii) and dhole (Cuon alpinus) (Kamler et al., 2012; Kamler et al., 2020). This may be why Williamson’s mouse deer had a near-significant positive association with the relative abundance of northern red muntjac. Additionally, although wild boar, Chinese serow and sambar may be important prey terms of the Asian golden cat, dhole and clouded leopard (Neofelis nebulosa) (Kamler et al., 2012; Kamler et al., 2020; Rasphone et al., 2022), the neutral associations with the relative abundance of wild boar, Chinese serow and sambar and the presence of sambar may be due to Williamson’s mouse deer having a trade-off between the benefits of reducing predation risk and habitat damage (reduced food availability and destruction of the refuges of Williamson’s mouse deer) by the abundant larger ungulates (Foster et al., 2014).

Although the possible dietary niche differentiation and dilution effects may be important reasons to explain the positive and neutral spatial associations between Williamson’s mouse deer and the sympatric larger ungulates, the temporal avoidance (i.e., differing in daily activity pattern and direct temporal avoidance) we assessed in this study provides an alternative way of avoiding competition from the perspective of niche theory.

Subordinate competitors commonly partition niches temporally with the dominant competitor (Schoener, 1974), and this is often found in ungulate guilds (de Pinho et al., 2022; You et al., 2022). In this study, Williamson’s mouse deer had significantly different daily activity patterns from the four sympatric larger ungulate species. The four larger ungulate species have more active times at night with inapparent peaks in the morning and twilight, especially sambar, which has primary nocturnal activities without any apparent peaks and only has a trough in the afternoon. Williamson’s mouse deer, in this region, is a typically crepuscular species, a finding which was different from the nocturnal activity pattern mentioned in a previous study in the same area that did not use camera traps (Cao et al., 2010). Similarly, another camera-trap study also reported that lesser mouse deer have a bimodal and crepuscular activity pattern (Khoo et al., 2021), which was different from the nocturnal pattern reported by earlier studies. Here the mouse deer was more active at night, which was different from the sympatric wild boar and sambar, which showed more diurnal and unimodal activities. This may suggest that Williamson’s mouse deer shift their daily activity pattern to avoid the sympatric larger ungulates. The moderate and low overlaps (< 0.7) and different activity peaks imply that Williamson’s mouse deer can use the co-occupied sites with the larger ungulates at different time periods of the day, which efficaciously reduces the likelihood of encounter, thus facilitating the coexistence between them (Kronfeld-Schor and Dayan, 2003).

Although the differentiation of daily activity patterns is an effective way to alleviate interspecific competition, the overlapping of daily activity time periods also implies that the subordinate species still have a low probability of encountering the dominant species (Kronfeld-Schor and Dayan, 2003). Our time interval analysis showed that Williamson’s mouse deer had direct temporal avoidance of the larger ungulates in the co-occupied sites. This suggested that direct temporal avoidance also helps Williamson’s mouse deer relieve competitive pressure from larger ungulates. Prolonged interactions may enable Williamson’s mouse deer to identify the tracks of larger ungulates by signs, scent and vocalizations. This avoidance strategy has rarely been found among the interactions of wild ungulates (Kavčić et al., 2021; Staudenmaier et al., 2021), although it has been found in the response of wild ungulates to livestock (Feng et al., 2021).

According to our second specific hypothesis, the coexistence of congeneric species with small differences in body size would prompt fiercely competitive interactions (Leyequién et al., 2007), which means that the smaller northern red muntjac should have had the strongest spatio-temporal exclusion to Williamson’s mouse deer, followed by wild boar, Chinese serow, and sambar. However, Williamson’s mouse deer had stronger avoidance of the much larger species spatially and temporally, which contradicts this hypothesis.

Spatially, Williamson’s mouse deer had positive associations with the presence of the three smaller ungulate species but had a neutral association with sambar (the largest species). Temporally, Williamson’s mouse deer had lower overlaps of daily activity patterns (wild boar [0.66] > northern red muntjac [0.63] > Chinese serow [0.54] > sambar [0.44]) and stronger direct temporal avoidance (wild boar [avoidance ratio = 1.53]< northern red muntjac [avoidance ratio = 1.96]< sambar [avoidance ratio = 3.026]< Chinese serow [avoidance ratio = 3.36]) with/of the much larger ungulate species, which was contrary to the expectation that the more similar the body sizes, the stronger the competition (Roughgarden, 1983). One possible reason is that sufficiently different body sizes represent sufficient differentiation of niches between sympatric species (Wilson, 1975), resulting in an unclear pattern between differences in body sizes and the degree of spatio-temporal competitive interactions. Previous studies pointed out that competitors can steadily coexist spatially if the body size ratio exceeds thresholds that were considered near 1.3 for body length and 2.0 for body mass (Hutchinson, 1959; Bowers and Brown, 1982). In Mengla, the sympatric larger ungulate species is much larger than Williamson’s mouse deer; for example, the body size and body mass of northern red muntjac, the next smallest ungulate species, are 2.32 and 8.14 times those of Williamson’s mouse deer, which implies that it has extremely different niches, representing distinct morphology and physiology (Helaouët and Beaugrand, 2009). Moreover, the sambar may utilize more open sites compared to Williamson’s mouse deer due to its extremely large body size, with a standing height reaching two meters from the foot to the antler. This difference in microhabitat preference may explain why Williamson’s mouse deer was positively associated with the presence of the three smaller ungulate species but neutrally associated with sambar. Another possible reason is that the strength of interspecific competition between the species may be related to multiple ecological factors instead of purely the difference in body size. Evaluation of interspecific competition focusing only on body size could miss processes such as the larger species excluding the smaller species in multiple ways (such as through habitat damage), or competitors having imperceptible symbiotic relationships (such as predator dilution effects). Consequently, we highlight that interspecific competition in this community appears to be more complicated than we expected, and the difference in body size is limited in explaining the coexistence of tropical forest ungulates.

Our study also can provide the first evidence of the relationship between Williamson’s mouse deer and anthropogenic disturbance. Fortunately, none of the anthropogenic variables had negative effects on Williamson’s mouse deer ( Table S3 ). Williamson’s mouse deer may have good adaptability in disturbed habitats, similar to common ungulate species such as muntjac and wild boar. According to studies of other species of mouse deer, these kinds of deer can occupy disturbed habitats. For example, Javan mouse deer can use dense tea plantations (Farida et al., 2006). Similarly, lesser mouse deer prefer fragmented and logged forest patches over continuous primary forests (Hazwan et al., 2022), and they can occupy fragmented urban forests and even prefer small fragments (Khoo et al., 2021).

Our study was conducted in protected areas with low levels of human disturbance. However, the potential habitats of Williamson’s mouse deer may also exist outside of protected areas, which we will be assessing in the future. Consistent with previous findings, our results showed that Williamson’s mouse deer prefer to use lowlands, and they may prefer to use sites closer to water resources (Cao et al., 2010). We thus suggest that future investigations of the population of Williamson’s mouse deer in nonprotected areas should mainly focus on lowlands closer to water. In Mengla County, the northern red muntjac and wild boar are becoming problematic to local ecosystems (e.g., damaging the local habitats and excluding the local species) due to their overabundance caused by habitat fragmentation and the declining population of larger predators (Cao and Zhang, 2022), which is hard to control and manage. However, we emphasize that these two species have no spatial exclusion from Williamson’s mouse deer. In summary, our study provides important information for understanding the threats to the population of Williamson’s mouse deer, which can guide management and conservation.