Estimation of White‐Handed Gibbon Density and Population Size in Huai Kha Khaeng Wildlife Sanctuary, Western Thailand

ABSTRACT

All 20 gibbon species (Hylobatidae) are listed as Threatened or Endangered on the IUCN Red List. The four species of gibbons occurring in Thailand survive only in well‐protected forests and need to be effectively monitored for conservation management. Because of their habit of duetting and living in small‐group territories, gibbons are usually surveyed using acoustic methods employing point counts. We surveyed the white‐handed gibbon (Hylobates lar) population in Huai Kha Khaeng Wildlife Sanctuary in western Thailand to determine the long‐term viability of the population. We combined the listening‐point method using arrays of listening posts with the point transect Distance method with prior random placement of sample points. We placed 39 sample arrays in dry (seasonal) evergreen forest (DEF), the optimal forest type for gibbons in the sanctuary, and 34 arrays in selected areas of mixed deciduous forest (MDF). During the 20 months of survey (2017–2021), we estimated an average density of 3.74 groups km⁻² for DEF and 1.10 groups km⁻² for MDF. The total number of gibbon groups in the habitats was estimated at 2568 (95% CI: 2156–3063) in DEF and 1482 (95% CI: 1172–1887) in MDF. A multiple covariate distance sampling analysis revealed that forest type had a major effect on gibbon density, while altitude and distance to poaching signs had little or no effect. The most serious threat to gibbon populations in the WEFCOM complex in the future is likely to be increased dryness and degradation of forests caused by the global anthropomorphic increases in temperature.

Summary

• A census of the white‐handed gibbon (Hylobates lar) was carried out in Huai Kha Khaeng Wildlife Sanctuary, western Thailand.

• Gibbons were most abundant in dry evergreen forest (an average of 3.74 groups per square kilometer), but occurred at lower density (about 1.1 groups per square kilometer) in the drier mixed deciduous forest.

• Average group size was 3.24 individuals, and the total population in the sanctuary is estimated to be at least 4000 individuals, higher than expected.

A census of the white‐handed gibbon (Hylobates lar) was carried out in Huai Kha Khaeng Wildlife Sanctuary, western Thailand. Gibbons were most abundant in dry evergreen forest (an average of 3.74 groups per square kilometer), but occurred at lower density (about 1.1 groups per square kilometer) in the drier mixed deciduous forest. Average group size was 3.24 individuals, and the total population in the sanctuary is estimated to be at least 4000 individuals, higher than expected.

1. Introduction

Gibbons (Hylobatidae) have been inhabiting the historically contiguous and pristine forests of tropical Asia over 1 MY (Fleagle 2013). They are highly arboreal, spending most of their lives in the forest canopy (Carpenter 1940; Fleagle 2013), where they require a continuous canopy and a diverse range of plant foods (for Hylobates spp.: Bartlett 2009; Ellefson 1974; Kappeler 1984; Marshall 2010; Marshall et al. 2009; McConkey et al. 2002; Raemaekers 1977; Srikosamatara 1084; Suwanvecho and Brockelman 2017; Whitten 1982). Gibbons are recognized as crucial seed dispersers and, therefore, contribute to habitat maintenance (Brockelman et al. 2022; Corlett 2017; Fan et al. 2008; Hai et al. 2018; McConkey and Chivers 2007; McConkey 2009; McConkey 2018). Due primarily to habitat loss from expansion of agriculture, and the pet trade, gibbons are among the most threatened of primates (Brockelman 1975; Tunhikorn et al. 1994). The IUCN Red List of Threatened Species recognizes all 20 species of gibbons as being threatened: one species as Vulnerable, 14 species as Endangered and 5 species as Critically Endangered (IUCN 2022). The four species of gibbons in Thailand (white‐handed gibbon Hylobates lar, pileated gibbon Hylobates pileatus, black‐handed or agile gibbon Hylobates agilis, and the siamang Symphalangus syndactylus) are all listed as Endangered (IUCN 2022), To ensure sufficient protection and appropriate conservation actions for gibbon species, it is important to have accurate estimates of their populations and their trends, as well as threats to their habitats.

In Thailand, gibbons occur only in protected areas, primarily wildlife sanctuaries and national parks. Gibbon surveys have been conducted in a few such areas, including Khao Soi Dao Wildlife Sanctuary, Khao Yai National Park, and Khao Ang Rue Nai Wildlife Sanctuary for H. pileatus (Brockelman and Srikosamatara 1993; Phoonjampa and Brockelman 2008; Phoonjampa et al. 2011) and Hala Bala Wildlife Sanctuary for H. agillis and S. syndactylus (Nongkaew et al. 2018). Populations of H. lar and H. pileatus were very roughly estimated during the population and habitat viability analysis (PHVA) meeting conducted in 1994 (Tunhikorn et al. 1994). At that meeting, the average density of gibbons in the extensive remote areas of forest of Thailand was assumed to be 2 groups (about 8 individuals) km⁻², and a total population of H. lar of approximately 110,000 individuals in 31 separate populations. There have been no population estimates of H. lar, the most widespread species in Thailand, since that meeting. H. lar reaches a high density of about 4 groups km⁻² in some well‐protected areas around the headquarters and research areas of Khao Yai National Park and Huai Kha Khaeng Wildlife Sanctuary (Brockelman and Saralamba, personal observations).

Gibbons live in stationary territories from which mated pairs regularly give duetted songs which can be heard from a distance of more than 1 km under favorable terrain and atmospheric conditions (Brockelman and Ali 1987; Bartlett 2009, 2011; Fan et al. 2009). Gibbon researchers have, therefore, developed acoustic survey methods that involve listening for duetted gibbon songs from prominent listening posts and mapping their locations.

Huai Kha Khaeng Wildlife Sanctuary is considered to be the core wildlife area of the Western Protected Area Complex of Thailand, which contains 17 contiguous national parks and wildlife sanctuaries (Figure 1). The complex has extensive populations of gibbons (Wildlife Conservation Society Thailand 2019), but we do not have a population estimate for any protected area within it. The area contains a variety of both evergreen and deciduous forest types, but gibbons have been observed inhabiting mainly seasonal evergreen forest (also referred to as dry evergreen forest [DEF]). Huai Kha Khaeng has been invaded by wildlife poachers over the years from the numerous villages near its borders, and it is unclear how such poaching has affected the survival of gibbons. Hence, our objectives were to (1) provide estimates of gibbon density and population size in Huai Kha Khaeng Wildlife Sanctuary, (2) Study the effects of forest type, altitude and poaching pressure on gibbon density, and (3) Evaluate the importance of the Western Forest Complex, and Huai Kha Khaeng Wildlife Sanctuary in particular, to the long‐term conservation of the species in Thailand.

Figure 1.

Map showing the 17 protected areas, including 12 national parks (NP) and 5 wildlife sanctuaries (WS), included in WEFCOM, western Thailand. Their names and map abbreviations are as follows (starting with the northernmost): Khao Sanam Preang NP (KSP), Klong Wang Chao NP (KW), Klong Lan NP (KL), Umpang WS (UMP), Mae Wong NP (MW), Huai Kha Khaeng WS (HKK), Thung Yai East WS (TYE), Thung Yai West WS (TYW), Tong Pha Phum NP (TPP), Khao Laem NP (KLM), Phu Toei NP (PT), Khuean Srinagarindra NP (KS), Lum Klong Ngu NP (LKN), Chalerm Rattanakosin NP (CR), Salakpra WS (SLP), Erawan NP (ERW), Sai Yok NP (SY).

2. Methods

2.1. Study Site

Huai Kha Khaeng Wildlife Sanctuary (15°00′–15°47′ N, 99°00′–99°27′ E; Figure 1), established in 1972, encompasses an area of about 2780 km² in the center of the Western Protected Area Complex, which contains about 19,000 km² of forest habitat (Trisurat et al. 2010). The sanctuary covers an altitudinal range of 200–1600 m a.s.l, but most of the sanctuary lies between 600 and 1000 m a.s.l. The sanctuary contains four major vegetation types: DEF (24.7% of area) mixed deciduous forest (MDF) (48.3%), deciduous dipterocarp forest (6.9%), and hill evergreen forest (13.4%) (Trisurat et al. 2010). Hill evergreen forest in Thailand is usually defined as forest above 1000 m a.s.l. and ecologically is approximately equivalent to lower montane forest (Ashton 2014). DEF, also known as seasonal evergreen forest, has a 4–6‐month dry season with little or no rainfall (Bunyavejchewin et al. 2001). Its drier stands grade into what may be considered semi‐evergreen forest (Ashton 2014), which is not a forest type recognized on maps of the Department of National Parks, Wildlife and Plant Conservation. Preliminary observations indicated that DEF is the main habitat of gibbons in Thailand, although there were reports of gibbons inhabiting MDF. The deciduous dipterocarp forest is a drier forest type with more open canopy and grassland understory, and has few or no gibbons. Hence, we focused our survey on dry evergreen and MDF, but our most reliable data come from the evergreen forest; our survey of the MDF encountered difficulties of accessing remote sample points due to limited road access and flooding in the wet season as well as lack of water in the dry season.

2.2. Survey Methods

We conducted the gibbon survey during five non‐consecutive periods: May–September 2017, May–September 2018, May–June 2019, July–September 2019, and July–November 2021, for a total of 185 days. Survey timing was based on the availability of research assistants, field staff, field volunteers, environmental factors such as rain and the COVID‐19 situation which caused delays. We conducted 2‐week‐long training workshop on site for field personnel to practice all techniques needed for field data collection and processing.

Traditionally, gibbons have been surveyed using listening‐point sampling and recording acoustic information to map gibbon groups (Brockelman and Ali 1987; Brockelman and Srikosamatara 1993; Brockelman et al. 2009; Brockelman et al. 2020; Gilhooly et al. 2015; Haimoff et al. 1986; Nongkaew et al. 2018; Pang et al. 2022; Phoonjampa and Brockelman 2008; Phoonjampa et al. 2011; Ray et al. 2015; Vu et al. 2018; Yanuar et al. 2020). In this study, we combined listening point sampling with point‐transect Distance sampling (Buckland et al. 2001; Thomas et al. 2010) to estimate group density from the mapped groups.

The Distance program models the decline in probability of detection with increasing distance from a sample point or transect line, by fitting a “key function” to the frequency of detections at different distances from the point or line. The value of the fitted curve where it crosses the origin (zero distance) provides the best estimate of density. In the case of gibbon sampling, groups do not duet every day, so that a correction factor may have to be applied to the density estimate.

We describe our methods in a series of procedures to help readers follow the logic: (1) selection of areas to be sampled and sample points to be used in the point‐transect analysis; (2) selection listening post array around each sample point; (3) collecting acoustic data from listening posts; (4) processing of data on maps to allow determination of the number of groups heard and their locations; (5) measuring the distance of each group's mean location to the sample point, the basic radial distance used in the point‐transect analysis; (6) estimation of singing probability, (7) estimation of average group size, (8) use of the data in a multiple‐covariate distance analysis; (9) stratified conventional distance analysis to determine densities for each of the two main forest types.

2.2.1. Selection of Sample Areas and Points

The main forest types in Huai Kha Khaeng have been mapped by the Royal Forest Department at 1:50,000 scale from Landsat satellite images (shown color‐coded in Figure 2 at reduced scale). We first surveyed gibbons in the DEF, which mostly covers the foothills and mountainsides up to about 1000 m a.s.l. We randomly selected 40 sample points (with a minimum distance apart of 2 km) from a 1:50,000 scale topographic map (Royal Thai Survey Department) in DEF. At one sample point the survey was rained out and so we ended up with 39 points for analysis (Figure 2). After this survey, we decided to sample the more extensive mixed‐deciduous forest in lowland areas, and placed 40 sample points in this forest type. We were unable to survey all the randomly placed points due to the weather and terrain problems mentioned above. Most of the points actually sampled, shown in Figure 2, were along rivers where road access was possible. Due to budget and time restraints, we managed to complete data collection at only 34 points.

Figure 2.

Map showing all sample points (with 2‐km spacing) in the gibbon survey. Circles (●) represent samples in dry evergreen forest and triangles (▲) samples in mixed deciduous forest. Hollow dotted points represent selected samples that were not surveyed.

2.2.2. Selection of Listening Post Arrays

The objective is to estimate the density of gibbon groups around each sample point. The Distance method of analysis using point transects (Buckland et al. 2001, 2015) assumes that the probability of detecting an animal or group of animals (in our case, a family group of gibbons) over the sample point is 100%, and that the probability of detection declines with increasing distance away. The locations of all animals detected, therefore, must be estimated and mapped. In the analysis, the distribution of distances of all animals detected is fitted with a mathematical function which allows prediction of the density where the function meets the sample point (distance zero). This is the “key function,” which describes how detectability declines with increasing radial distance from the sample point.

In most point transect analysis the observer or listener will stand on the sample point for a specified period of time and record the estimated distances of all animals detected (Buckland et al. 2015). However, gibbons cannot be visually detected in evergreen forest unless the observer is very near them. We, therefore, use their duetted songs to detect them. Gibbon songs can be heard up to about 2 km away under favorable conditions, but estimating their distance away cannot be done accurately, as audibility of songs depends greatly on the intervening terrain and atmospheric conditions. Therefore, we use a method of mapping their singing locations called “triangulation,” in which listeners from two locations spaced a known distance apart measure the directions of songs heard with hand compasses. Group locations can then be determined from a map where the compass direction lines from the listening posts intersect (Brockelman and Ali 1987).

In this survey we used an array of four listening posts around each sample point to collect data on singing groups (Figure 3). The listening posts were placed on hills or ridges a few hundred meters from the point. Initial selection was made on a 1:50,000 topographic map, and verified on foot. The listening posts were established, and access trails cut to them, at least 1 day before data collection. Our experience in surveying gibbons enabled us to select an array of four listening posts such that no gibbon group that sang within a distance of 400 or 500 m from the sample point would be missed. The location of the sample point and each listening post was determined with a hand‐held GPS (GARMIN GPS map 62s, Garmin International Inc., Kansas, USA). The average distance from the listening posts to the sample point was 444 m (SE = ±20 m) for DEF samples and 493 (SE = ±17 m) for MDF samples. The average altitude a.s.l. of sample points in the DEF was 680 m (range: 388–960 m) and in the MDF, 342 m (range: 180–742 m) (Table 1).

Figure 3.

Method of determining the number of groups from mapped singing locations. After singing locations are mapped for each day, they are color‐coded and shown on a single map, as in this figure, to determine the actual number of groups heard. For this, we rely on information on the timing of duets heard on the same day, and on the distinctiveness of duets from different locations. Duets that overlapped in time, or were distinguishable acoustically, are connected with line segments indicating that they were given by different groups. This enables us to draw lines around locations that were likely to represent different groups. As a rule of thumb, locations farther than 500 m apart (the approximate diameter of a gibbon territory) are considered to be in different groups. In the example illustrated, eight groups were estimated to be present. The procedure is a conservative one that provides a minimal estimate of the number of groups. After measurement of radial distances from estimated group average locations to the sample point, average group density is estimated from the Distance software (which estimates the number of groups missed at greater distances from the sample point).

Table 1.

Summary of details of datasets used in conventional distance sampling analysis.

Dataset	Right truncation (m)	N within RT	Excluded N	Mean altitude, m (range)	Density groups km−2 (95% CI)	Mean group size (range, N)
DEF	1600	446	13	680 (388–960)	3.74 (3.14–4.46)	3.39 (2–7, 68)
MDF	1600	203	18	342 (180–742)	1.10 (0.87–1.40)	3.43 (2–5, 30)
MDF	2000	218	3	342 (180–742)	1.12 (0.88–1.42)	3.43 (2–5, 30)

Abbreviations: DEF, dry evergreen forest; MDF, mixed deciduous forest.

As most sample points were remote from human facilities, the survey team set up a base camp at a suitable site near a source of water, but outside the array of listening posts to minimize disturbance of the gibbons present.

2.2.3. Collection of Acoustic Data

On each day of survey two persons quietly walked to each listening post at dawn with synchronized watches and hand compasses. The listening period began before 0700 h, and all listeners returned to the base camp afternoon. A single mated pairs of adults constitutes the core of each territorial group, which is most easily identified by their duets. A duet may be sung at any time during the morning, and consists of a series of “great‐calls” given by the female, each lasting 12–20 s, given at 1–2 min intervals. Each great‐call is followed by a male “coda” consisting of a rapid sequence of several hoots rising in pitch. Duets may last from a few minutes to a half hour or more. Gibbons seldom duet in the afternoon (although intergroup conflicts may occur). From no later than 0700 h until noon, the listeners noted down the exact time and direction of each great‐call sequence of each gibbon duet heard, as well as solo calls and intergroup conflict calls. They also noted special characteristics of the duets that may help in distinguishing groups, such as female great‐call length or pitch, male call timing, and the calls of subadult females that sometimes sing in synchrony with their mothers.

The “sampling effort” was 4 days at each site. Rainy or windy days were avoided or excluded. Collection of four good‐weather days of data allows for satisfactory separation and counting of groups (see below), and for estimation of the probability of singing per day.

2.2.4. Processing of Data on Maps

The analysis of singing location data on maps involves two processes: the mapping of singing locations, and the determination of the number of groups heard. Groups do not duet on every day, and groups may duet more than once on some days, so that the number of groups does not equal the number of duetted bouts heard. Determining the number of groups from the number and locations of bouts heard is the key difficulty in using this method of censusing gibbons.

In a relatively dense gibbon population such as the one sampled here in evergreen forest in Huai Kha Khaeng, the mapped singing locations are scattered more or less at random and do not cluster in a way that would allow identification of group territories. Groups give duets from all parts of their territories, and territories are virtually all contiguous. Therefore, in a dense population, it is not possible to identify groups from their singing locations without acoustic or other types of information. For this reason, a procedure, outlined below, has been developed over the years for identifying groups in gibbon surveys (Brockelman et al. 2009; Brockelman et al. 2020; Brockelman and Srikosamatara 1993; Phoonjampa and Brockelman 2008). This procedure is needed for generation of data required for mark‐recapture analyses, in which groups calling on successive days are individually recognized (“marked”}, as are mice or other mammals caught in traps on successive days. Mark‐recapture analysis helps us to determine the number of groups detected in the sample area, and their singing frequency (described below).

For each recording day, the data from the four listening posts in each array were mapped at a scale of 1:20,000 (5 cm = 1 km) (or similar, using computer mapping software). Lines from each listening post were drawn indicating sound trajectories to the gibbon groups heard. If the times of the calls in the bouts heard from two or three listening posts were highly correlated (starting within a few seconds of one another), the intersection of the lines was taken as the location of a group. At the conclusion of the 4 days, all triangulated group positions were color coded and transferred to a single map (Figure 3), and all singing locations judged to be from the same group were circled. This judgment was based on three criteria: (1) If the bouts of nearby singing locations overlapped in time on any given day, they were regarded as belonging to different groups; (2) If the acoustic or timing characteristics of different bouts were distinctly different from those of other bout locations, they were regarded as coming from a different group; (3) If the bout locations were greater than 500 m apart, they were judged to be from different groups. Therefore, an outlying point more than 500 m from any other song location is considered to represent a separate group, as this is considered to be a reasonable minimum estimate of the width of a typical gibbon home range. Criterion (1) was the most important one and 4 days of accumulated data are sufficient to separate nearly all groups in a listening area. Bouts not so classified are regarded as belonging to the nearest identified group. The assignment of an unidentified bout to one or another cluster does not affect the determination of the number of groups, although it will affect the frequency distribution of song days per group. The procedures for the recognition of a separate group likely causes some underestimation of groups, as a singing location is assumed to be part of the nearest identified group unless the decision rules prove otherwise.

2.2.5. Determining Radial Distances

The locations of bouts assumed to be from the same group were averaged on x and y axes and converted to a single point. These locations included bouts given on the same day or on different days. We measured the distance of the averaged group locations to the central sample point of the array to the nearest 10 m. These are the radial distances used in the Distance analysis. We tabulated 459 distances from the 39 sample points in deciduous evergreen forest, and 221 distances from the 34 sample points in the MDF (Table 1).

2.2.6. Determining Singing Probability

A preliminary estimate of singing probability may be obtained by dividing the number of days each group sang, summed over all groups, by the total number of listening days summed over all groups. This will be an overestimate because the total number of groups within hearing distance is not known, as it is missing the groups present that did not sing on any of the survey days; however, the estimate will improve by increasing survey effort (number of days of listening per listening post). Non‐singing groups must be estimated and included to the total number of listening days. We may do this by modeling the distribution of singing days per group with the binomial distribution, in which the probability of singing per day, p(1), is a random Bernoulli trial. We estimate p(1) using data on the numbers of groups actually heard singing on different numbers of days. In general, the probability of singing at least once in n days is p(n) = 1 − [1 − p(1)] ⁿ , which is one minus the probability that the group did not sing on any of the n days. The distribution will be truncated because it is missing the term (1 − p)⁴ (where n = 4), the probability of not singing for 4 days (Brockelman et al. 2020). We present a program in R code for estimating p in the truncated binomial distribution using maximum likelihood, provided courtesy of David L. Borchers (Supporting Information S1).

Estimation of singing probability should be made only for groups close enough to the listening posts so that we may assume that they will be heard with near 100% certainty. An empirical measure of the decline in detection probability can be obtained by plotting the density of groups within circles increasing in radius from the sample point (Brockelman et al. 2020) (Figure 4). The number of gibbon groups within 500 m have been pooled because the relatively low numbers within very small radii produce highly variable estimates of density. Figure 4 shows that cumulative density does decline beyond about 600 m in deciduous evergreen forest and 800 m in MDF, so we decided use groups mapped within 600 m in dry evergreen and 700 m in MDF for determination of singing frequency.

Figure 4.

Densities of groups in circular listening areas in relation to radius for dry evergreen forest (DEF) (A) and mixed deciduous forest (MDF) (B).

2.2.7. Estimating Average Group Size

Group size was determined from groups observed opportunistically. If a gibbon group called near a listening post (up to 100 or 200 m away), one listener attempted to observe them for at least 1 h and obtain a count of group members, including infants being carried by the female.

2.3. Analysis

2.3.1. Selection of Key Detection Function

Using the radial distances in each array of listening posts, we first carried out an analysis on the aggregate data (combining forest types) using the point transect analysis in DISTANCE version 7.5 (Buckland et al. 2015; Thomas et al. 2010). The datasets were first right‐truncated at 1.6 km in DEF and 1.6 km and at 2.0 km (in separate analyses) in MDF to remove approximately 5% of the observations most distant from the sample points, as detections at these distances have little effect on the shape of the probability density function near the origin (at distance = zero), and very large distances near the limit of audibility are likely to be less accurately estimated. We ran the analysis using two truncation distances in the deciduous forest to determine if the truncation distance affected the results.

Four key detection probability functions are offered in the Distance program: the negative exponential, half‐normal, hazard rate and uniform functions. The negative exponential can be excluded on a priori grounds, as observations of wildlife do not decline steeply close to the origin, as this distribution implies (Buckland et al. 2015). The uniform distribution could also be excluded as we know from past experience that detections will decline with distance. The half normal function has been used in some gibbon surveys (e.g., Lwin et al. 2022), but we excluded it because in our method of sampling, we assume that the probability of detecting groups is level at 100% to some distance beyond the listening posts before it declines. The hazard rate key function, which has been used previously (Brockelman et al. 2020), has this property, and hence was our function of choice.

2.3.2. Multiple‐Covariate Analysis (Multiple Covariate Distance Sampling [MCDS])

We employed the MCDS analysis (Thomas et al. 2010) to test the effects of forest type, altitude (digital evaluation model, or DEM), and distance to poaching incidence as covariates, and compared the results with conventional distance sampling (CDS) which analyzes the data without covariates. Akaike's information criterion (AIC) was used to test the effects of covariates and their combinations. The model that has the lowest AIC is the one that is best supported, and is given the value 0; other models are given the value ΔAIC, or their AIC minus the lowest AIC (Burnham and Anderson 2002). The Distance program also provides three conventional goodness‐of‐fit‐tests on the data (see also Buckland et al. 2015). The Chi‐square test tests the fit of the binned frequencies against frequencies predicted by the model. The Kolmogorof–Smirnov test presented is considered more powerful because it is applied to the continuous data; it tests the fit of the observed cumulative continuous data to the data predicted by the function. The Cramer–von Mises test is similar to the Kolmogorof–Smirnov test except that it uses squared differences between the predicted cumulative distance function and the observed distance data.

The AIC penalizes models with more parameters and correction factors, and does not always reflect the data's goodness‐of‐fit. Other things being equal, models with fewer parameters and required correction factors are considered more desirable.

2.3.3. Stratified Distance Analysis

After finding that forest type was the major variable affecting group density (Table 2, upper rows), we decided to stratify the distance analysis. Separate estimates of density for the two forest types are desirable for several reasons: (1) the detection function might not be identical for the two forest types, which are structurally different; (2) the results for the two forest types differ in reliability; (3) application of the data to estimate gibbon populations in other protected areas with different frequencies of forest types requires separate density estimates for each; (4) it is hoped that separate estimates will facilitate research on the effects of forest type on gibbon ecology and behavior; and (5) having separate estimates of density for the two forest types allows us to better evaluate the ability of the wildlife sanctuary as a whole to conserve the population in the face of future climate change.

Table 2.

Details of models fitted to the gibbon group data, using the hazard‐rate key function with cosine adjustments.

Dataset	Analysis	Covariates	Pars	∆AIC	EDR	χ2	CvM	KS
DEF + MDF	MCDS	forest + altitude	4	0.00	1023	0.044	> 0.50	0.664
DEF + MDF	MCDS	forest	3	0.33	1029	0.081	> 0.50	0.571
DEF + MDF	MCDS	altitude	3	38.00	1125	0.095	> 0.15	0.108
DEF + MDF	CDS	—	2	33.59	1079	0.190	> 0.90	0.807
DEF	MCDS	altitude + poach dist	4	8.54	1052	0.118	> 0.025	0.049
DEF	MCDS	altitude	3	6.58	1052	0.147	> 0.025	0.049
DEF	MCDS	poach dist	3	6.54	1049	0.145	> 0.025	0.053
DEF	CDS	—	2	0.00	982	0.426	> 0.800	0.748
MDF	MCDS	altitude + poach dist	4	0.13	1262	0.005	> 0.20	0.114
MDF	MCDS	altitude	3	3.49	1376	0.004	> 0.15	0.172
MDF	MCDS	poach_dist	3	3.57	1374	0.004	> 0.20	0.143
MDF	CDS	—	2	0.00	1315	0.013	> 0.40	0.358

Note: Forest = DEF or MDF forest, altitude distance from the sample point to nearest poaching sign(s); Pars = number of parameters in model; ∆AIC values; EDR = effective detection radius (m); and p‐values for χ², Cramer‐von Mises (CvM), and Kolmogorov‐Smirnov (KS) goodness of fit tests.

Abbreviations: CDS = conventional distance sampling, DEF = deciduous evergreen forest, MCDS = multiple‐covariate distance sampling, MDF = mixed deciduous forest.

2.4. Ethical Statement

Our noninvasive research was permitted by the Department of National Park, Wildlife and Plant Conservation (DNP), Thailand under the MOU between the DNP and the Wildlife Conservation Society (WCS), Thailand program. This study was conducted in compliance with the American Society of Primatologists Principles for Ethical Treatment of Nonhuman Primates, and the Code for Best Practices in field Primatology. This study complied with the Animals for Scientific Purposes Act (A.D. 2015) of Thailand under the Committee for Supervision and Promotion of Procedures on Animals for Scientific Purposes.

3. Results

3.1. Multiple Covariate Distance Sampling

The initial MCDS analysis was carried out to determine what major variables affected gibbon group density the most. Data for the two forest types were pooled and forest type was used as a covariate, along with altitude (Table 2). The model with forest type + altitude, and that with forest type alone, performed about equally well (with the lowest ΔAIC values), but all models (including the CDS model with no covariates) without forest type did very poorly. It is clear that forest type had the major effect on density, and altitude by itself had little or no effect. Altitude varies considerably within forest type (Table 1), and is not a determining factor for forest type.

We decided to stratify the analysis after finding the overriding effect of forest type as a covariate. Analyses for the separate forest types included altitude and distance to the nearest poaching added as covariates. For the DEF, altitude and poaching did not improve prediction, and the inclusion of both covariates caused the ΔAIC value to increase more than 6.5 above that for CDS (Table 2). For MDF, inclusion of altitude + poaching had a relatively small advantage over the omission of either variable from the model alone, or the CDS model with no covariates. Most poaching signs encountered were in the MDF where most tracks and trails are located. The larger values of ΔAIC in the MCDS analyses appear to be caused mainly by an increase in the number of parameters, but the goodness‐of‐fit tests also show better fits to the CDS model, especially in the DEF.

3.2. Analysis of Density Using CDS

The hazard rate function assumes that all gibbon groups will be heard up to some distance from the central sample point. In DEF, with a truncation distance of 1.6 km selected, 446 points were available for analysis, and in the MDF, 218 points were available within 2.0 km and 203 within 1.6 km. The hazard rate model with cosine extension was selected as the best model and gave a density estimate of 3.74 (95% CI: 3.14–4.46) groups km⁻² for DEF, and 1.10 (CI: 0.87–1.40) in MDF with truncation distance of 1.6 km and 1.12 (CI: 0.88–1.42) in MDF with truncation distance of 2.0 km (Table 1, Figure 5). Truncation distance has a negligible effect on estimated density. In the MDF model, binning all observations below 600 m from the sample point was done because of the small number of groups within closer distances. According to the χ ² values, the fit of the hazard rate function was better in the case of DEF than of MDF. This may be due to the higher group densities in DEF, which are limited by competition for space and hence, have lower variation, and by the relatively lower numbers of groups detected at smaller distances in the MDF.

Figure 5.

Hazard rate detection function models fitted to frequencies of gibbon groups heard in point transect distance analysis. A, data from dry evergreen forest (DEF). B1 and B2, data from mixed deciduous forest (MDF), with two different truncation distances (2000 and 1600 m). The two truncation distances have little effect on the density estimate.

The graph showing the cumulatve density of groups with increasing distance (Figure 4) shows that, in the MDF, there is a shortage of groups within about 800 m of the sample point, which is also evident in the detection function graphs in Figure 5B. The only explanation we have for this is that the listening personnel disturbed the groups while moving to their LPs, and the gibbons close to the sample points either moved outward or remained silent. This effect is also common in bird surveys, but movement of animals away from the sample point does not greatly affect the estimate of density (Buckland et al. 2015).

3.3. Estimation of Singing Probability

The value of p(1), the probability of a group singing in a single day, is estimated from the truncated binomial distribution at 0.661 (CI: 0.602–0.716, N = 575 groups) and p(4), the probability of singing at least once in 4 days, is 0.986 (CI: 0.975–0.993). There was no significant difference in p(1) or p(4) between dry evergreen and MDF, so we have combined the data for the two habitats. This indicates that we have missed less than 1% of groups overall due to lack of singing. The two forest types have very different group densities, but singing frequency was similar in both.

The estimate of singing probability from summing the frequencies of days of singing divided by the sum of survey days across groups was 0.672. This lies well within the confidence limits of the value obtained from solving the truncated binomial, and indicates that with a relatively high probability of detection per day of 0.66, 4 days of survey is sufficient to detect virtually all groups and estimate density without the need for a correction factor.

3.4. Gibbon Group Size

Throughout the survey, 68 gibbon groups were seen in deciduous evergreen forest and 30 in MDF. We found that evergreen forest gibbons had an average group size of 3.39 individuals (range: 2–7) and MDF groups averaged 3.43 individuals (range: 2–4). Clearly, habitat did not significantly affect average group size. Average group size for the combined habitats is 3.40 individuals (Figure 6). The most common group size was four individuals.

Figure 6.

Gibbon group size frequency distribution, including both forest types.

3.5. Gibbon Population in Huai Kha Khaeng WS

We estimate the number of gibbon groups within each habitat using the average densities of groups and the areas of each forest type. We estimated a total of 2568 groups in DEF (CI: 2156–3063 groups) and 1482 groups in MDF (CI: 1172–3063), for a combined population of 4050 groups. These habitats combined comprise 73% of the area of the sanctuary. We have more confidence in the estimate for the evergreen forest based on random sampling, but we believe that the samples in MDF are sufficiently representative to be worthy of reporting. Correction of the estimates for singing probability [p(4)] increases these estimates by only about 1%, a negligible amount.

4. Discussion

4.1. Gibbon Density

The density of gibbons found in Huai Kha Khaeng turned out to be higher than initially expected. The wildlife sanctuary contains a spectrum of habitat types ranging from low stature, dry (deciduous) dipterocarp forest, MDF containing bamboo, dense bamboo forest, and seasonal (or “dry”) evergreen forest grading into lower montane forest at higher elevations. The preferred habitat of gibbons is seasonal evergreen forest which is expected to have the highest diversity of tree species, and the tallest and most complex structure. This forest type is dominant in the hilly areas of continental Southeast Asia where it has not been logged and converted to agriculture, and has the highest biodiversity (Ashton 2014). However, semi‐evergreen forest also contains a relatively high density of gibbons. The Huai Kha Khaeng ForestGEO plot, located in the northern part of the sanctuary at an altitude of 549–638 m a.s.l., is covered with semi‐evergreen forest and is surrounded by numerous gibbon groups, four of which overlap the 50‐ha plot (WYB, unpublished observations).

Mixed deciduous, or tall deciduous, forest is utilized by gibbons as well, although the density of groups is less than one‐third that in evergreen forest. In addition, Light et al. (2021) found that gibbon groups frequently crossed the ecotone between evergreen and deciduous forest to reach food sources and spent considerable time in deciduous forest when food trees were available. Gibbons are flexible in their feeding behavior and can switch to flowers, young leaves, and other non‐fruit foods when needed (Light et al. 2021).

We may have underestimated the habitat area utilized by gibbons because they likely use some forest above 1000 m altitude. We lack estimates of gibbon density in this forest type. Such areas are steep and more difficult to access, and few surveys of wildlife include this forest type, which has lower stature and reduced diversity of food trees.

4.2. Forest Protection and Poaching

The evidence that protected areas have been effective in preventing declines in wildlife populations world‐wide has been challenged by Geldmann et al. (2013) as being inconclusive. Hameed et al. (2024) have shown that primate populations have declined in most protected areas of India. In a study based on actual sampling data, Tun et al. (2023) documented a 46% decline in density of hoolock gibbons (Hoolock leuconedys) over a 16‐year period in Mahamyaing Wildlife Sanctuary, Myanmar. Poaching by non‐timber forest product collectors in Khao Soi Dao Wildlife Sanctuary in Southeast Thailand severely reduced the pileated gibbon population there (Kolasartsaree and Srikosamatara 2014). Results of primate surveys rarely provide “good news” for conservation; ours is a welcome exception.

In 2004, the Department of National Parks, Wildlife and Plant Conservation devoted increased resources to the protection of WEFCOM, the largest and most important protected area complex in Thailand (and perhaps in continental Southeast Asia), and with the support of the Wildlife Conservation Society and other international conservation organizations, initiated measures to improve personnel training, patrolling and enforcement. Nevertheless, some wildlife poaching persists from the villages adjacent to the sanctuary.

Given the impact of poaching on wildlife populations (Duangchantrasiri et al. 2016), our study's initial prediction was that poaching distance would likely be a factor affecting the abundance of gibbon groups. In Vietnam, Tran and Vu (2020) found that the distance‐to‐village variable influenced the northern yellow‐cheeked gibbon (Nomascus annamensis). In Huai Kha Khaeng, however, poaching distance had little or no effect as a covariate; hence our results suggest that it is not a major threat to gibbons in the sanctuary. The absence of using gibbons as a food source in Thailand, the increased enforcement of the ban on the capture of gibbons for the pet trade, and the ban on export of primates, have also likely helped to reduce the poaching pressure on gibbon populations in Thailand.

4.3. Huai Kha Khaeng as a Gibbon Conservation Area

In addition to having relatively high densities of gibbons, Huai Kha Khaeng Wildlife Sanctuary is configured in a way that is favorable to their long‐term survival. Firstly, the existence of two large upland areas of evergreen forest, on the eastern and western mountain ranges (Figure 1), provides better assurance of their survival than would a single large area, or perhaps fragmentation of the evergreen forest into numerous small areas with gibbon populations below the minimum viable population size (Tunhikorn et al. 1994). Secondly, the lowland area between the two mountain ranges contains mostly MDF which connects the evergreen forest blocks. This lowland forest contains a low density of gibbons so that all habitats in Huai Kha Khaeng containing gibbons are connected.

Global warming could affect the gibbon population in Huai Kha Khaeng (and throughout the WEFCOM complex) by increasing dryness of the deciduous forest and increasing the incidence of fire (Grimm et al. 2013; Seidl et al. 2017). Warmer and drier conditions would degrade the MDF and result in its conversion to scrub vegetation or to deciduous dipterocarp forest which is less suitable for gibbons. Global climate models predict an increase in precipitation throughout most of the tropics (IPCC 2023), but this increase will be accompanied by an increase in variability and unpredictability (Zhang et al. 2024) which may be detrimental to forests. Tangang et al. (2019) have analysed the effects of a selected ensemble of global climate models on future precipitation in Thailand and found that the best models are not consistent in their predictions. Regardless of the changes in precipitation caused by global warming, the overriding effect of temperature increase on water availability will likely be increasing dryness caused by increase in evapotransporation. Climate change will likely cause species to expand their ranges into higher altitude forest, and forests above 1000 m are available in Huai Kha Khaeng to make this possible.

Huai Kha Khaeng Wildlife Sanctuary is an important conservation area for Hylobates lar and has a wide variety of forest types, many of which offer habitat suitable for gibbon survival. Two research priorities remain, however, which are (1) to study further the ability of gibbons to thrive in the drier, deciduous forest types which will likely become more prevalent in the future, and (2) survey evergreen forests above 1000 m in altitude to study their suitability for gibbons and determine the current upper limit of gibbon distribution, which at present is poorly known. An additional conservation priority is to develop more rapid procedures for surveying gibbons that require less manpower and expense than those used in this study.

Author Contributions

Study design: All authors. Funding acquisition: Anak Pattanavibool. Project administration: Anak Pattanavibool. Data collection: Somphot Duangchantrasiri, Mayuree Sornsa, and the WCS Thailand team. Data curation, data analysis, and writing the article: All authors.

Conflicts of Interest

The authors declare no conflicts of interest.

Supporting information

S1‐est.of p.

Data Availability Statement

Data on the singing frequencies of gibbons at all sample points, the locations of sample points, and the models used in the distance functions and their outputs, are available from the authors.