Herd demography, sexual segregation and the effects of forest management on Bornean banteng Bos javanicus lowi in Sabah, Malaysian Borneo

Between 1973 and 2010, 39.5% of Sabah’s (Malaysian Borneo) natural forest cover was lost to deforestation and conversion to agriculture, therefore the remaining population of endangered Bornean banteng Bos javanicus lowi is being driven towards extinction. The Bornean banteng’s herd demography, sexual segregation and the effects of forest management were investigated at 393 camera locations in 6 forest reserves using generalised estimating equations (GEE) fitted via a generalised linear model (GLM). A total of 43344 camera trap nights and 832 independent banteng events were captured at 93 locations. The identification of 183 bantengs included 22 herds (>1 individual) and 12 solitary bulls, with a herd size range of 2 to 21. Significantly larger herds were observed in forest with <8 yr of post-logging regeneration (PLR), whereas herds were smaller in forest with <3, 4 and 16 yr of PLR. Within these forests, herds were significantly larger along logging roads than in open sites and on forest trails. Herds were significantly larger in upland compared to lowland dipterocarp forest, but significantly smaller when closer to the forest border. Bachelor herds were observed as frequently as mixed-sex herds, and there was a significantly higher capture frequency of female herds in the dry season, supporting the theory of sexual segregation. Frequency of calf births was highest in March and September, and calf captures peaked in June and July. This study contributes to a better understanding of banteng ecology, and will assist in the production of effective management strategies aimed at providing suitable habitat for re-population and enabling banteng population persistence.


INTRODUCTION
Tropical forests are the richest terrestrial eco system on Earth (Gentry 1992) and contain many of the world's 'biodiversity hotspots' (Myers et al. 2000), yet are experiencing the greatest forest loss of all forest do-mains, with loss increasing by 210 100 ha yr −1 (Hansen et al. 2013).Between 1990 and 2010, tropical forest cover was reduced from 1635 mil lion ha to 1514 million ha, with 32.9 million ha lost in Southeast Asia (Achard et al. 2014).Overexploitation and agricultural activities are the primary threats to species worldwide OPEN PEN ACCESS CCESS (Maxwell et al. 2016).Crop, livestock and tree plantations are the major direct causes of tropical deforestation.Logging, mining and petroleum development also contribute directly to tropical de forestation; however, they also indirectly promote deforestation by increasing the accessibility to otherwise remote areas and facilitating poaching (Butler & Laurance 2008).Sparsely populated areas are being cleared rapidly; for example, the Amazon for large-scale cattle ranching and industrial soy farming, and Southeast Asia, including Borneo, Sumatra, and New Guinea, for oil palm and rubber plantations (Sodhi & Ehrlich 2010).Of the 3 major tropical regions, Southeast Asian forests are experiencing the highest rates of forest loss (Sodhi et al. 2004), therefore requiring urgent conservation attention.
Forest cover in Borneo is being lost at approximately twice the rate of other tropical forests in the world (Gaveau et al. 2014).Between 1973 and2010, 39.5% of forest in Sabah, Malaysian Borneo, was lost to selective logging, fire and conversion to oil palm and timber plantations (Gaveau et al. 2014).Of this forest loss, 97% occurred in habitat suitable for the Bornean banteng Bos javanicus lowi (Gaveau et al. 2014, Gardner et al. 2016), likely the rarest mammal in Sabah now that the Sumatran rhinoceros Dicerorhinus sumatrensis has been declared extinct in the wild in Malaysia (Havmøller et al. 2016).
Banteng Bos javanicus is a sexually dimorphic wild cattle species and is categorised as 'Endangered' by the IUCN Red List of Threatened Species (Gardner et al. 2016).The most recent global population estimate was 8000 individuals (Gardner et al. 2016).Three subspecies are recognised: Java banteng B. j. javanicus found in Java and Bali, Burma banteng B. j. birmanicus existing on the Asian mainland, and Bor nean banteng B. j. lowi present in Borneo (Hassanin & Ropiquet 2007), in Sabah, Kalimantan and possibly Sarawak (Gardner et al. 2014).The first survey of the Bornean banteng carried out in Sabah in the early 1980s estimated the population at 300 to 550 individuals (Davies & Payne 1982); however, this probably declined to < 300 in the late 1990s (Boonratana 1997).Precise present-day Bor nean banteng population and subpopulation sizes are unknown, although there is possibly 1 subpopulation of > 50 individuals (the agreed minimum viable population size) present in Sabah (Gardner et al. 2016).The reduction and fragmentation of habitat and its conversion to agriculture, poaching and increased risk of disease transmission from domesticated cattle are severely threatening the Bornean banteng, with many of the remaining subpopulations confined to protected areas (Gardner et al. 2016).
Banteng form cohesive social groupings (Srikosamatara 1993), which are important aspects of their social behaviour and environment.Herd sizes of large herbivores are primarily functions of foraging strategy and anti-predator behaviour (Kie 1999), and explain, for example, increases in herd sizes in open habitat with reduced canopy cover (Gerard & Loisel 1995, Kie 1999).Additionally, herd size and dynamics, including herd formation and division, can determine the habitat selection of large herbivores, as observed for bison Bison bison (Fortin et al. 2009).This is supported by larger herds of banteng being observed in open forest (Gray 2012) and grasslands (Pudyatmoko & Djuwantoko 2006) in comparison to smaller herds observed in dense forest with continuous canopy cover, in Cambodia and Java, respectively.Herd sizes may vary in response to frequent human disturbance causing weak bonds in herds, thus instigating less permanent herd sizes (Pudyatmoko & Djuwantoko 2006).Herd sizes may also differ be tween seasons (Pudyatmoko & Djuwantoko 2006), as banteng herds often aggregate around water holes in the dry season when rainfall is limited (Nguyen 2009).
Herd sex (i.e.gender composition) provides information on ecological factors including the expression of sexual segregation, which is commonly exhibited by sexually dimorphic ungulates outside the mating season (Ruckstuhl 2007).Banteng bachelor herds of mixed ages, and banteng cow and calf groups are known to occur frequently, with mixed temporary assemblages occurring during the mating season or in large open areas (Gardner et al. 2016).Sexual segregation is more likely to occur during the birth period because the behavioural differences between males and females become more pronounced (Bon & Campan 1996, Ruckstuhl 2007); females become more asocial, timid (Copland 1974) and more dependent on water and rich food sources as a result of the additional demands of gestation and lactation (Bon & Campan 1996).No rut or calving season has been observed for B. j. lowi (Gardner et al. 2014), although exploring the expression of sexual segregation may allow rut or calving seasons to be observed.Research into understanding the sexual segregation of ungulates has been conducted; the causes are still poorly understood (Ruckstuhl 2007), but could include ecological, physiological, social or foraging factors (Main et al. 1996).
Species with wider ecological niches, particularly herbivores, are more tolerant towards logging and may even benefit from post-logging conditions (Meijaard & Sheil 2008).B. j. javanicus occupy secondary forest formations resulting from logging and fires; however, this has not been ob served in B. j. birmanicus (Gardner et al. 2016).B. j. lowi thrive on the temporary abundance of pioneer species present in the early stages of post-logging regeneration when not hunted, but evidence of bark stripping by B. j. lowi suggests a lack of grassland forage, hunting pressure or heavy disturbance (Gardner 2015).Timber harvesting that creates open spaces may be beneficial in providing sufficient space for larger banteng herds; however, energy-demanding behaviour is reduced in open areas during hot hours (Gardner 2015).Increased anthropogenic disturbance may cause herds to separate and also in crease stress levels, which is likely to impact banteng breeding activity and behaviour (Gardner et al. 2014).Furthermore, B. j. lowi have reduced body conditions in conventionally logged forests compared to reduced-impact logging (RIL) forests (Prosser et al. 2016).
Collaborations between scientists, managers and conservationists to produce science-based wildlife management strategies is increasing and has been identified as a requirement for the effective management and conservation of ungulates (Apollonio et al. 2017).Understanding the mechanisms that in fluence the demography of increasingly small and isolated ungulate populations is a conservation priority, and is essential to prevent future extinctions (Tatin et al. 2009).Bornean banteng are important ecosystem engineers, but are severely threatened and are being driven towards extinction (Gardner et al. 2016).
The aims of this study were to investigate the herd demography of the Bornean banteng, specifically the herd size, sex and composition, and to explore the expression of sexual segregation in regenerating forest in Sabah.Understanding banteng herd demography will provide baseline data on their behaviour and ecology -including their vigilance (Roberts 1996), predation risk and population density -and on habitat structure (Marino & Baldi 2014).This enhanced scientific understanding of their demography will enable the desired management and conservation of the Bornean banteng and the complex system they inhabit.Informative baseline data will indicate changes in the population and environment, thus will facilitate future management.Since timber harvesting creates open spaces and facilitates increased growth of pioneer species that provide temporary resources, we hypothesised that banteng herds would be larger in forests with less than 8 yr post-logging regeneration, and that herds would be larger in open sites than on forest trails and logging roads.We also hypothesised that banteng express sexual segregation and that more banteng calves would be born in the dry season than in the wet season.

Study sites
Six forest reserves in Sabah (Malaysian Borneo) were surveyed using remote infrared camera traps: Tabin Wildlife Reserve (TWR), Malua

Camera trapping
Behavioural data on the Bornean banteng was collected using non-invasive, remote, passive infrared camera traps: Reconyx HC500, Reconyx PC800 and Reconyx PC850 (Reconyx).Camera trap data originated from 2 different studies using (1) a grid layout and cameras positioned on an ad hoc basis where banteng signs (tracks and dung) were located in TWR and MFR (Gardner 2015) and (2) a state-wide survey for banteng across Sabah in which camera traps were deployed where signs of banteng were located (P. C. Gardner & B. Goossens unpubl. data).Table 1 shows the camera trapping method used in each forest reserve location.A camera trap station consisted of 2 Reconyx Professional Hyperfire cameras fixed to opposing trees, approximately 1 to 1.5 m above the ground to maximise the chances of capturing bantengs and identifying individuals.A minimum distance of 0.5 km was maintained between camera trap stations to maximise the chance of photographing banteng in the area.Camera traps detected heat and movement, triggering 3 consecutive photographic captures at 1 s intervals, with no time delay between activations.All camera trap photo graphs were digitally stamped with the event date, time and temperature.Camera traps operated for a minimum survey period of 90 d, and were checked every 28 d to ensure functionality.Camera trapping effort, which refers to the survey duration (sum of all 24 h operational camera trap nights), was calculated from the date the camera trap was set to the date it was retrieved for all forests.If the camera was no longer functioning, the date of the last event was used.The habitat vegetation (lowland dipterocarp, upland dipterocarp, seasonal freshwater swamp, scrub, lower montane forest or industrial tree plantation) and elevation were recorded at each camera trap station.Percentage leaf cover was extracted from photographs of the canopy, taken directly above each station using a Samsung WP10 waterproof allweather 12.2 MP × 5.0 digital zoom compact camera on minimum optical zoom.Percentage leaf cover was estimated from monochrome photographs using the software Leaf Cover Calculator v. Table 1.Location of each Bornean banteng survey, camera trap model used, sampling method (grids or ad hoc), survey period and study (in m) between each camera trap station, nearest village and forest border was extracted post hoc using ArcGIS v.10.1 (ESRI) (H.Y. Lim et al. unpubl. data).Incidences of poaching, which included armed and unarmed people, shotgun cartridges, snares, carcasses, lone dogs, gaharu/sandalwood harvesters and poachers' camps, or camera trap stations that were stolen (2 camera traps station −1 ) was recorded (P.C. Gardner et al. unpubl. data).

Herd demography
Bantengs captured on camera trap within each forest reserve were defined as subpopulations, including forests that were adjacent (MFR and KMTFR); therefore, 6 subpopulations and their respective herd sizes were studied.Individuals were identified using a series of natural marks, including scars on the body, ear tears, horn shape and size, and natural coat colourations (P.C. Gardner & B. Goossens unpubl. data).Recognition and recaptures of solitary individuals and herds (>1 individual) by morphology and scars were recorded in each forest reserve.Herd size was estimated from the number of bantengs photographed.Male and female morphological characteristics were used to calculate adult sex percentages for each event (Gardner et al. 2014).Bantengs were categorised into 3 broad age classes based on clear differences in body size: adult (male or female), juvenile (up to 50% smaller than adult cow) and calf (more than 50% smaller than adult cow).This classification has been applied to a banteng population in Baluran National Park, Indonesia, using direct ob servations (Pudyatmoko & Djuwantoko 2006).Calf births were estimated from the first date of appearance on camera and from their approximate body size.Herd composition was categorised as: (1) male herd, (2) female herd, (3) mixed-sex herd, (4) mixed-sex herd including calf (calves), (5) female(s) and calf (calves), (6) solitary male, (7) solitary female and (8) unknown, due to low light levels or photographs obscured by vegetation.Juveniles could not be reliably sexed so were disregarded when categorising the herd composition.All captures were classified according to 1 of 2 seasons: (1) the wet season, which was defined as between October and March, and (2) the dry season, between April and September, for all forest reserves, due to the reduced impacts of drought and El Niño-Southern Oscillation events (Walsh 1996).Daily rainfall data (mm) collected in Danum Valley, east Sabah, provided by the South East Asia Rainforest Research Partnership (SEARRP), was applied to all forest re serves.

Data preparation
The data set comprised discrete explanatory variables (number of bulls, cows, juveniles and calves, herd size and poaching and stolen camera trap station incidences), categorical explanatory variables (study design, herd composition, years of PLR, camera trap site, salt lick presence, season and habitat vegetation) and continuous explanatory variables (percentage encounter rates of herd compositions, temperature, rainfall, elevation, canopy cover and the distances of each camera trap station to the nearest village and forest border).Yasuda (2004) studied medium-to large-sized mammals using camera traps and defined successive photographs as independent when separated by at least 30 min, and Phan & Gray (2010) suggested 20 minutes for Bos javanicus birmanicus.Preliminary observations showed that an intermission length of 90 min between camera trap captures accommodated a range of herd behaviours: travelling, foraging and resting.Therefore, longitudinal independence was defined by discounting any banteng individual or herd captured within 90 min of the previous event, unless the individual or herd was identified as different.It was assumed that the banteng's choice to join others was not constrained by availability of other banteng.

Statistical analysis
All statistical analyses were conducted using the statistical software package R v.2.15.2 (R Development Core Team 2012).As a result of the clustered, longitudinal and repeated measures data and the individual observations not being statistically independent, generalised estimating equations (GEE) were fitted using a generalised linear model (GLM) following the protocol of Vaughan et al. (2007), using the error distribution 'Poisson' and library 'geepack'.This allowed the analysis of hierarchical and correlated data, and accounted for spatial autocorrelation (Højsgaard et al. 2006).GEEGLM models were used to test for differences in herd sizes between study design (grid or ad hoc), forest reserve regeneration age, season, camera trap site, salt lick presence, herd sex, temperature, rainfall, habitat vegetation, elevation, canopy cover, distance to the nearest village and forest border, and poaching and stolen camera trap station incidences (Table 2).Backwards stepwise deletion was used to produce the final, most robust GEEGLM model including significant (and 1 marginally non-significant) explanatory terms.Pearson residuals were used to validate model output.GEEGLM results were transformed into odds ratios (OR), a measure of association between an environment and an outcome (Szumilas 2010), by taking the exponential.OR compared the relative odds of an outcome of interest occurring in a particular environment as: OR = 1: environment does not affect odds of outcome; OR > 1: environment related with higher odds of outcome; and OR < 1: environment related with lower odds of outcome (Szumilas 2010).
One-way analysis of variance (ANOVA) and Tukey post hoc tests were used to investigate the expression of sexual segregation by testing for differences in capture frequencies and the effect of season upon each herd sex in all forest reserves collectively.A Poisson GLM was used to test for significant differences in the number of calves born between the seasons due to the approximately equal variance to the mean and the acceptable degree of over-dispersion.
A Poisson generalised additive model (GAM) with a log link function was fitted to analyse the association between the number of calves captured by camera traps and the month of the year.Month was fitted with a cyclic cubic smoothing spline to allow for cyclical non-linear trends (Table 2).

Survey effort
During the study period (April 2011 to April 2015), a total of 832 independent events of banteng were captured from 93 camera traps over 43 344 camera trap nights in 6 forest re serves (Table 3).Over 2400 camera trap nights were discounted because of electronic failure or camera trap theft.A total of 30 banteng events were discounted because they were captured within 90 min of the previous event and violated our assumption of independence.
Inter-forest comparisons of herd sizes and forest sites revealed that herds on logging roads were significantly larger (OR = 1.30±1.10, p < 0.01, Model 1; Table 6) than herds in open sites, whilst herd sizes on forest trails (OR = 1.17 ± 1.10, p = 0.1055) were not significantly different (Model 1; Table 6).Herd sizes at sites with salt licks present were significantly larger (OR = 1.73 ± 1.19, p < 0.01) than herds at sites with no salt licks present (Model 1; Table 6).
Habitat vegetation had a significant effect on herd size, whereby herd sizes were larger in upland dip terocarp (OR = 2.56 ± 1.34, p < 0.01) habitats when compared to lowland dipterocarp (OR = 2.01 ± 1.19), the intercept (Model 1; Table 6).Distance to the forest border had a significantly negative effect on banteng herd size, with herd sizes significantly smaller closer to the forest border (OR = 1.00 ± 1.00, p < 0.001; Fig. 4).

Sexual segregation
We observed a significant effect of banteng sex on capture frequencies when all captures from all forests were pooled (1-way ANOVA: F 2,17 = 6.20, p < 0.05, Model 2; Table 7).We found that female herds occurred significantly less frequently than male herds (Tukey post hoc test: p < 0.05) and mixed-sex herds (Tukey post hoc test: p < 0.05).There was no significant difference between capture frequencies of male herds and mixed-sex herds (Tukey post hoc test: p = 0.983).Male and mixed-sex herds contributed to 40.9 and 42.3% of the camera trap events respectively, whilst females contributed only 16.8% of the events.Survey durations were longer in TWR, MFR and MBCABZ and encompassed both wet and dry seasons, therefore only data from these forests was used to explore the effect of season on sex.Season had a significant effect on the capture frequency of female herds (1-way ANOVA: F 1, 5 = 19.89,p < 0.05, Model 3; Table 7), with more captures obtained in the dry season (72%) compared to the wet season (28%); however, the seasonal effect was only marginal for male and mixed-sex herds (Models 4 and 5; Table 7).

Calf births
A Poisson GLM revealed that the number of calves born during the survey period did not significantly differ between the dry and wet season (Poisson GLM: LRT = 2.28, p = 0.1308, Model 6).However, a Poisson GAM with a cyclic cubic smoothing spline revealed a significant non-linear trend of number of calves captured across months (Poisson GAM: edf = 4.23, χ 2 = 15.84,p < 0.01, Model 7).According to the month of first capture of each identified calf, births were most frequent in March and September (Fig. 5), and calf captures peaked in June and July (Fig. 6).

DISCUSSION
A total of 183 bantengs, including 25 calves, were identified in 6 forest reserves, in herd sizes ranging up to 21 individuals.Forest regeneration age, type of site, presence of salt licks, sex, habitat vegetation and distance to the nearest forest border all had significant effects on banteng herd size.A significant effect of banteng sex was found on capture frequencies.The frequency of calf births was highest in March and September, and calf captures peaked in June and July.

Herd demography
The fact that only 183 bantengs were identified in our 6 forest reserves is strong evidence of the Bornean banteng's current vulnerability.The banteng encounter range of 1 to 21 individuals differs from the 5 to 40 banteng estimated from villagers' perceptions in Sabah in 1982 (Davies & Payne 1982) Table 7. One-way ANOVA determining any significant differences in the capture frequency of each herd sex: male, female and mixed in all forest reserves, together with the effect of season.Significance denoted as: (.) p < 0.1; *p < 0.05  number of identified banteng had a larger herd size range and average, therefore, a declining population likely causes smaller herd sizes, as observed in a population of Bos javanicus birmanicus in Vietnam (Nguyen 2009).Low population densities and restrictions to home range are causing reductions, and subsequent ly extinctions, in other banteng populations (Pedrono et al. 2009).This is likely occurring with the Bornean banteng.Movements identified between MFR and KMTFR, which supported the largest herds, suggest that substantial and continuous forest patches are important and required for larger herds to form.Conse quently, habitat reduction and fragmentation threatens the banteng population and reduces herd sizes.SPTFR had a small number of bulls identified (10 individuals) but the highest number of bulls consistently travelling alone (6 individuals).KMTFR had the highest number of bulls identified (21 individuals), and was the only forest reserve to have a bachelor herd and to contain no solitary bulls.Here, the bachelor herd was dynamic in composition, with individuals regularly leaving and new bulls joining.The reduced number of males in SPTFR possibly affects bull behaviour and results in solitary lifestyles.Moreover, the number of herds repeatedly observed in SPTFR was high, therefore the choice or chance of interacting with a herd was higher than in many of the other forest reserves.Despite this, bulls in SPTFR remained solitary.KMTFR had fewer herds but they were larger, and this may have made locating and acceptance into the herd more likely.
The bachelor herd in KMTFR was observed in 26 different combinations involving 15 mature bulls, which suggest they are very tolerant and highly social; however, the maximum herd size encountered was comprised of only 7 individuals.Additionally, due to the evidence of illegal activity en countered when surveying KMTFR, hunting and human disturbance may have caused males to form bachelor herds in order to increase vigilance and therefore survival, a behaviour that has been observed in other threatened ungulates (Averbeck et al. 2010).This dynamic bachelor herd may be an easier target for hunters; therefore, considering the evidence of illegal activity encountered during this survey, this is a major concern.

Banteng herd sizes
This study revealed that banteng herd size was significantly affected by the years of PLR.Forest in the onset stages of regeneration had the smallest herd size range, suggesting that regular use of heavy machinery and human disturbance may have weakened the bonds of association between individuals within herds.This may have resulted in smaller herds that are less permanent and prone to splitting more frequently (Pudyatmoko & Dju wantoko 2006).Logging increases the abundance of pioneer species (Imai et al. 2012), in cluding grasses, vines and shrubs favourable to banteng (S.Ridge unpubl.data), and regenerating vegetation benefits banteng body condition and breeding (Gardner et al. 2014, Prosser et al. 2016), which is likely to have influenced the larger herds observed in forest with < 8 yr of PLR.Reduced disturbance from heavy machinery in the years following logging activity may have helped herds re-form and allowed herds to aggregate more frequent ly.MFR and KMTFR are adjacent, and banteng movement between these forest reserves was identified, showing that banteng individuals have had access to 2 forests and therefore a larger range of PLR.This would have provided in creased optimal conditions to allow larger herds to form.Forest with 22 yr of PLR would have increased closed areas that reduce ambient temperatures and thermal stress (P.C. Gardner et al. unpubl. data), and reduce conflict between dominant individuals and their vulnerability to poaching.This likely contributed to the larger banteng herds observed in forest with 22 yr of PLR.However, the overall effects of deforestation, heavy machinery and human presence may have raised banteng stress levels enough to negatively affect breeding, and possibly result in increased mortality (Gardner et al. 2014).
Herd sizes not significantly differing between open sites and dense forest contrasts with B. j. javanicus in Baluran National Park (Pudyatmoko & Djuwantoko 2006).Gerard & Loisel (1995) (Gardner et al. 2014); however, disturbance is likely altering this behaviour and reducing herd size.Larger banteng herd sizes in upland compared to lowland dipterocarp forest suggests banteng are being forced to higher habitats for space and forage, likely a result of habitat destruction, fragmentation and human disturbance.
Larger herds on logging roads than in open sites suggest that the former provide sufficient space for larger ag gregations and the opportunity to travel as a larger unit.Internal abandoned logging roads also provide banteng forage due to the increased regeneration of pioneer species over time (Gardner 2015).Moreover, logging roads provide easy access to the previously cultivated areas that provide secondary growth which banteng benefit from (Pedrono et al. 2009).Thus, more individuals will be attracted to the available forage, causing larger herds to be vulnerable to human conflict on logging roads.
Larger herds were observed in sites with salt licks than when no salt licks were present.This suggests that banteng may well be deficient in sodium and need this additional source; this is supported by a small-scaled diet-supplementation survey in Sabah that captured banteng at sites with mineral blocks and loose salt (G.H. Phillips unpubl.data), together with Davies & Payne (1982) and Matsubayashi et al. (2007) who state that banteng require and frequently visit salt licks.It may be advantageous to implement salt licks for monitoring herds; however, doing so may make them more vulnerable to poaching.
Smaller herd sizes occurring closer to the forest border suggests they are influenced by human disturbance in the vicinity and that forest reserves must be substantial enough for banteng to express their natural demography.In contrast to the findings of Nguyen (2009), season did not have a significant influence on banteng herd size.This may imply that sufficient water sources were available in the study sites for the banteng to not aggregate, or that the conditions in the wet and dry seasons were not distinct enough to have an effect.
Although the incidences of poaching and stolen camera traps did not have a significant influence on banteng herd size, illegal activity occurred in all forest reserves (P.C. Gardner et al. unpubl. data), which likely affects their stress levels and therefore possibly breeding success, in addition to increasing their vulnerability.

Sexual segregation
Bachelor herds were observed as frequently as mixed-sex herds, therefore bulls segregated from cows, and thus this study provides evidence to support the theory of sexual segregation and is a required behaviour of bentang.Although one or several of the proposed factors (ecological, physiological, social or foraging) may cause males to segregate, they may not affect females as strongly (or at all), since female herds occurred significantly less often than bachelor herds.In addition, it is possible that females forced the bulls to leave mixed-sex herds, either due to differences in the previously stated factors, or because of birthing.Female herds being significantly larger than male herds implies that females remain in herds, and that males are possibly forced out.This is possible as Bornean cows and calves have been observed to assert authority, and younger bulls have been ob served to force older bulls from herds in Java and Burma (Gardner et al. 2014).
Female herds were observed significantly more often during the dry than the wet seasons, indicating that a possible calving period is more likely to occur during the dry season.Female ungulates are more likely to segregate from males during the birth period be cause of behavioural and nutritional differences, and more specifically to locate suitable birthing places and to give birth (Bon & Campan 1996Ruckstuhl 2007).Evidence of a female banteng actively segregating itself in preparation for the birthing period was observed (see Fig. A1 in the Appendix).In contrast, no significant difference in the observation of male herds between seasons suggests that males do not experience the same pressures as females for niche habitat or nutritional requirements, and that they do not have a need to segregate from females during gestation.

Calf births
No evidence was found to suggest that births were elevated in either the wet or the dry season, which indicates that females experience gestation through both seasons.It is possible that both seasons provided optimal environmental conditions for breeding and therefore season was not a constraining factor.More calves born in March and September and peaks in calf captures in June and July suggests a possible calving season.More calf captures in the months following March was due to calves taking approximately 2.5 to 3 months to be categorised as juvenile.This increase in calf capture would be expected after September; however, 2 calves born in September in TWR were born at the end of camera trapping in this forest.Increased calf births in March and captures through to June is supported by the calving season of B. j. javanicus between April and June in Baluran National Park (Pudyatmoko & Djuwantoko 2006).The very presence of calves indicates the potential for a population expansion; however, this area of Bornean banteng ecology requires further research.

Methodology limitations
Camera traps are a useful tool for wildlife behavioural studies, and are increasingly being used to improve species conservation (Caravaggi et al. 2017).It is, however, important to acknowledge their limitations.Dark photographs and vegetation obstruction made banteng identification impossible for some encounters.Camera trap placement has major influences on group size estimates.This study's camera trap height and location were specifically chosen to capture banteng.It was acknowledged that trails and logging roads had dense vegetation surrounding camera trap stations, making banteng more likely to travel between the 2 cameras in order to remain within the herd.It is, however, important to state that banteng are not inhibited by thick thorny vegetation, and can penetrate dense vegetation either side of trails and logging roads.When undisturbed, banteng show tendencies to follow their own trails, which was factored into the camera trap sensor area when positioning cameras.Open sites had far less vegetation to obstruct view, and when temperature decreased in open sites the range of the camera sensor had a tendency to extend.Despite the limitations of camera trapping, this survey method has been identified as superior when compared to the use of signs for detecting banteng (Gardner 2015).Camera trapping has provided conservation-relevant behavioural data of the Bornean banteng, and will act as baseline for future ecological studies aiming to assess the Bornean banteng demography.

CONCLUSION
This study has supplemented the limited knowledge and understanding of Bornean banteng ecology.The identification of banteng subpopulations and their respective herd sizes in Sabah has provided evidence of their vulnerability.This study presents baseline data of the Bornean banteng which can assist in producing appropriate management procedures that will work towards their conservation.Future land management must acknowledge the extent of habitat reduction and fragmentation, and thus ensure substantial and continuous forest patches to allow large banteng herds to form, particularly away from forest borders, and to reduce their disturbance and stress levels.Forest management should consider the stages of PLR and habitat vegetation in each forest reserve, ensuring the availability of < 8 and 22+ yr of PLR, and suitable lowland and upland dipterocarp forest respectively, to maximise herd sizes.Banteng have adapted to habitat modifications by utilising logging roads; therefore, forest management should include restricted human access and constant monitoring.Bachelor herds being ob served as frequently as mixed-sex herds, and a significantly higher capture frequency of female herds in the dry season, showed a strong indication of sexual segregation.Future research to understand banteng sexual segregation should be conducted to enhance understanding of banteng ecology and behaviour.The number of calves born did not significantly differ between seasons; however, frequency was highest in March and September, and calf captures peaked in June and July; thus, forest disturbance should be minimal during these months.The current banteng demography should be compared to the results of future research and monitoring in these 6 forest reserves, to enhance understanding and to stimulate any necessary protection measures.Incidences of illegal activity, in cluding poaching, occurred in all forest reserves, therefore anti-poaching patrols should be conducted.This enhanced scientific understanding of Bornean banteng demography can facilitate science-based wildlife and forest management strategies to allow banteng re-population and ensure their long-term existence.

Fig. 4 .
Fig. 4. Bornean banteng herd sizes at varying distances to the nearest forest border (in m).Dashed lines: prediction intervals

Fig. 6 .
Fig. 6.Average number of Bornean banteng calves captured over the survey period in each forest reserve.Dashed lines: SE

Fig
Fig. A1.Female banteng actively segregating herself in preparation for the birthing period (top image), followed by the same female banteng and her newborn calf (bottom image) on the same day (06/04/2014) in Maliau Basin Conservation Area Buffer Zones.This identified female banteng was observed regularly in a mixed-sex herd of 6 individuals before and after the birth Malaysian Borneo (inset), and location of the 6 study sites: Kuamut Forest Reserve and Malua Forest Reserve in central Sabah, Maliau Basin Conservation Area Buffer Zones and Sapulut Forest Reserve in south central, Sipitang in the west and Tabin Wildlife Reserve in the east

Table 2 .
The questions investigated (response term ~ explanatory term), and their respective statistical test and model number.Explanatory terms: study design (grid or ad hoc), forest regeneration age (years of post-logging regeneration), season, site (camera trap location), salt lick (presence), sex (herd sex), temperature, rainfall, habitat vegetation, elevation, canopy cover, distance_village (distance to the nearest village), distance_forest (distance to the nearest forest border), poaching, stolen camera stations and month.GEEGLM: generalised estimating equations fitted using a generalised linear model; OR: odds ratio;GAM: generalised additive model

Table 3 .
Total number of camera trap (CT) stations, number of CT stations that captured banteng, total number of CT nights (operational for 24 h), number of independent events (banteng individual or herd not captured within 90 min of the previous event, unless the individual or herd was identified as different), herds (>1 individual), solitary bulls and identified banteng for each forest reserve

Table 6
. Final, most robust generalised estimating equations fitted using a generalised linear model (GEEGLM) (Model 1) including the different factors of the terms regeneration age (years of post-logging regeneration [PLR]), site (camera trap location), salt lick presence, sex (herd sex), habitat vegetation, distance_forest (distance to the nearest forest border) and season.GEEGLM model estimates were converted to odds ratio, which is a measure of association between an environment and an outcome.Intercept included < 8 yr PLR (Kuamut Forest Reserve), open site, absent salt lick, female herd sex, lowland dipterocarp forest and dry season.SPTFR: Sipitang Forest Reserve; MFR: Malua Forest Reserve; MBCABZ: Maliau Basin Conservation Area Buffer Zones; TWR: Tabin Wildlife Reserve.Significance denoted by: (.) p < 0.1; *p < 0.05; **p < 0.01; ***p < 0.001 . At the time of the 1982 survey, the estimation of 40 bantengs was thought to be an underestimate (Da vies & Payne 1982), suggesting a decline in herd size over time.The forest re serves containing the highest Frequency of Bornean banteng calves born over the survey period in each forest reserve according to the month of the first capture of each identified calf and cow and calf approximate body size.TWR: Tabin Wildlife Reserve; MFR: Malua Forest Reserve; MBCABZ: Maliau Basin Conservation Area Buffer Zones; SPTFR: Sipitang Forest Reserve; SPLFR: Sapulut Forest Reserve and KMTFR: Kuamut Forest Reserve stated that large herbivores, including roe deer Capreolus capreolus and Alaskan moose Alces alces, generally have larger herds in open habitats with less canopy cover, when there is minimal disturbance.Our results, therefore, suggest that Bornean banteng are not analogous to the majority of large herbivores or even to other banteng subspecies.Bornean banteng have been observed to forage in open grasslands and socialise in open spaces