Introduction Effective conservation of a species necessitates basic information about densities, population sizes, and trends. Relative densities, or indices, can be used to indicate trends in numbers and compare areas with respect to conservation priority, but absolute estimates of numbers, and hence densities, are required for estimates of population size and vulnerability to extinction (Cowlishaw and Dunbar 2000) . Obtaining estimates of absolute numbers can, however, be very difficult. Orangutans (Pongo spp.), for example, are semi-solitary, non-territorial, arboreal and cryptic forest animals that often live at low densities. It is therefore difficult to count them directly and, if animals hide from observers, as in areas where they are hunted, line transects will systematically underestimate densities. However, great apes regularly build sleeping platforms, or nests, that can be counted with greater ease. Some two decades ago, chimpanzee researchers began using estimates of nest densities obtained through line transects, which can be converted into animal densities when the values of various parameters are known (Ghiglieri 1984; Hashimoto 1995) . Since its application to orangutans in the early 1990s (van Schaik et al. 1995) , this technique has been improved incrementally (Buij et al. 2003; Johnson et al. 2005) , and is now widely used to produce estimates of orangutan densities (Russon et al. 2001; Buij et al. 2002; Morrogh-Bernard et al. 2003; Ancrenaz et al. 2005) . In this paper, we first briefly review the technique then describe an experiment to test its adequacy, and finally present the results of a simple alternative plot-count method that is more accurate, as assessed by comparison with a direct estimate of density.

d ¼

L

N w

2 ; where d is the density of nests in numbers/km2, N the number of nests observed along the transect, L the length of the transect line in km, and w the estimated strip width in km. This nest density can be converted into orangutan density in number/km2 (Ghiglieri 1984; van Schaik et al. 1995) , using: D ¼ p d r

t ; where p is the proportion of the population making nests, r the rate of nest production (number of nests per capita per day), and t is the time to disappearance of the average nest in days. The line transect technique must overcome (1) the problems of estimating the additional parameters, and (2) the difficulties inherent in the use of line transects.

The problem with the additional parameters is that they need to be estimated, and are only known with some error, compounding the error in the eventual orangutan density estimate. The parameters p and r must be estimated from long-term field studies of habituated animals. The results show that p is fairly constant across sites, but that r can vary appreciably, with Sumatran populations (van Schaik et al. 1995; Singleton 2000) building nests at far higher rates than Bornean ones (Gunung Palung: Johnson et al. 2005; Kinabatangan: Ancrenaz et al. 2005; Tuanan: see Results) . It turns out that it is most difficult to estimate t, which appears to be quite variable, but without straightforward environmental correlates (Buij et al. 2003; Johnson et al. 2005) . In long-term studies, direct monitoring of the survival of nests made at known dates produces reliable estimates. Where repeat surveys are available, t can also be estimated by recording the transitions between decay stages of the nests and by considering the decay process a Markov chain with an absorbing end state (van Schaik et al. 1995) , which has been shown to produce good estimates (Buij et al. 2003; Johnson et al. 2005) . Thus, while the additional parameters needed to convert nest density into orangutan density cannot be estimated without error, these estimates have become fairly accurate.

The second difficulty stems from the line transect technique itself. Among the basic assumptions of this technique (Buckland et al. 1993) , the most problematic one is that all objects on the line (in forests: above the line) are encountered. Although it is not known what proportion of nests directly above or near the transect line are missed, there will always be some that escape detection. The existence of this problem, illustrated in Fig. 1, is demonstrated by two observations. First, more experienced observers generally produce higher density estimates than less experienced observers, while, if this assumption is met, they should merely detect objects in a broader strip and thus produce similar densities, albeit with narrower confidence limits. Second, repeat surveys of the same line in the opposite direction generally not only add nests, but actually yield higher density estimates when the samples are pooled (van Schaik et al. 1995; Buij et al. 2003; Johnson et al. 2005) .

Workers have therefore attempted to produce a correction for this deficit. First, the repeat survey can be used to produce a corrected estimate (Johnson et al. 2005) . However, although the new estimate is closer to the true density, it remains unknown how close (see Fig. 1b). Second, one can develop an empirical correction factor by comparing calculated densities with known densities (Buij et al. 2003) . However, this procedure also appears to be unsatisfactory because there is no general prediction as to the size of this factor, nor its relationship with forest structure or other obvious ecological factors. In conclusion, even if all parameters have been estimated correctly, the method will generally still produce an underestimate of the true density by an unknown margin.

In this paper, we make two independent attempts to develop more accurate density estimates. First, we examine the results of an experiment in which the results of two experienced pairs of observers who surveyed the same transect lines in both directions were compared to the combined sample of these two and an additional six pairs. Second, we examined a simple alternative method, based on complete counts of small plots. The results of Fig. 1 The effect of observer experience on estimated densities of objects surveyed through line transects. In (a) experienced (dotted line) and naı¨ ve observers (solid line) behave as expected by the method, i.e. both observe all objects on the line, but differ in how well they observe objects away from the line. In practice, their difference resembles that in (b), where experienced observers observe more objects, both on the transect line and away from it.

It is unknown, however, whether even experienced observers detect all objects on the line. The effect of including a second pass of the trail in opposite direction can also be represented in this way: new nests are detected mainly away from the line (a) or everywhere (b) these two attempts are compared with an estimate based confidence limits in orangutan density (i.e. assuming no on direct observations of orangutans, in particular the error in the other parameters). size and overlap of female home ranges. We established 11 plots of 0.2 ha each, by taking a 50-m stretch of the transect line and adding a 20-m strip perpendicular to the line on either side. Plots along the Methods same transect line were 200 m apart in order to avoid sampling the same clusters of nests if nests tend to be The study was conducted in August 2003 at the Tuanan distributed in a clumped way around major food trees study area in the Mawas Reserve, Central Kalimantan, and, thus, ensures independence of plot counts and Indonesia. This site (2 09¢06.1¢S, 114 26¢26.3¢E) consists similar habitat coverage as the line transects. An addiof peat swamp on shallow peat, of varying thickness, up tional ten similarly spaced plots were established along to about 2 m. It is disturbed, having been subject to two trails elsewhere in the study area. In each plot, a selective commercial logging in the early 1990s, followed team of three experienced observers attempted to locate by almost a decade of informal logging by local people. all nests. They were allowed to move both inside and Line-transect surveys of nests of Bornean orangutans outside the plot and for as long as they deemed neces(Pongo pygmaeus wurmbii) were conducted along a sary. The location of each nest was carefully measured recently established 2.5-km-long narrow boardwalk, to establish whether it was located within the plot. which follows trails that had been cut in a regular grid. Estimated densities using these indirect methods were

In a 2-day period, eight different pairs of observers also compared with our best estimate of the true density walked the same transect (1.25 km/day). On a given day, based on long-term follows of identified orangutans. a pair walked the transect line in both directions Adult female density was estimated as the number of (henceforth referred to as the first and second pass, home ranges among six identified individuals stacked at respectively). To facilitate comparisons among teams, a single point divided by mean female home range size; for each nest encountered, each team recorded location this density was subsequently extrapolated to the whole (trail coordinate), perpendicular distance to the board- population based on estimated population composition walk, decay class, and height in 5-m classes. Perpendic- (cf. Singleton and van Schaik 2002) . We superimposed a ular distances were measured to the nearest 0.5 m. grid on the 200-ha portion of the study area that was Location was estimated by reference to 50-m trail most intensively sampled. In this grid, we counted for markers, using the length of boards (4 m each) for each of 42 regularly spaced points the number of home additional reference. range of adult females known to include this point.

The teams varied in experience: two of them were Mean home range size for four intensively studied adult composed of people who had much experience (teams 1, females (>750 h each) was 245 ha (Wich and van 2) in nest surveys, whereas the other six teams (students Schaik, unpublished). Extrapolation to the whole popand staff of Universitas Nasional, Jakarta) had no ulation was done using the following estimated compoexperience, although several had followed orangutans sition: 0.9 infants; 0.4 juveniles and adolescents; 0.8–1 and all were at least familiar with orangutan nests. unflanged and flanged sexually mature males (the first

At the completion of the eight team nest surveys, the two of these numbers are based on direct observations, two most experienced teams rechecked all nests to pro- the latter is an estimate based on male-biased sex ratios duce a combined sample in which all nests recorded by at birth combined with male-biased adult mortality; cf. at least one team were included. This step was necessary Singleton and van Schaik 2002). Hence, 3.1–3.3· adult to avoid the double counting of nests that differed female density is an estimate of overall orangutan denslightly in recorded location, perpendicular distance, or sity. height class, and to ensure use of the same definition of nests among observers (which is especially important for the final stages of decay, as defined in van Schaik et al. Results 1995) .

Strip width (w) was estimated using DISTANCE 4.0 Line transect counts (Thomas et al. 2001) . Perpendicular distances were truncated at 30 m (six cases, or approximately 5%). We The main results are summarized in Table 1. Experifollowed the same procedure as Buij et al. (2003) , based enced teams detected significantly more nests after the on the five models recommended by Buckland et al. first pass (76, 87) and after both passes (85, 98) along the (1993), to which the observed distribution of perpen- 2.5-km line transect than any of the six inexperienced dicular distances were fitted. Calculations were made for teams (means: 42.5 and 49 nests, respectively; Mann– each team, one for the first pass and one for the first and Whitney U-tests: U=0, P<0.05, in both cases). They second pass combined. Only a single calculation could did not differ much from the inexperienced teams in the be done for the combined sample, representing both percentage of nests added due to the second pass in the passes. A 95% confidence limits were calculated for the opposite direction (12 and 13% for the two experienced estimated strip width based on the analytical variance teams vs mean of 15.7% for the six inexperienced option in DISTANCE, and were used to estimate the teams). Experienced teams tended to produce the highest aReporting only density estimates for first and second pass combined, using p=0.88 and r=1.15 nest densities, at 10.0 and 10.9 nests/ha, but one of the inexperienced teams also reached a similar density (10.4) (Mann–Whitney U-test, U=1; 0.05<P<0.10). The increase in nest density due to the second pass was marginally lower for the experienced pairs (4.4 and 9.5%) than for the other teams (mean: 14.4). Overall, then, experienced observers detect more nests and tend to produce higher density estimates.

The number of confirmed nests seen by at least one team, which provides the best estimate of the number of nests that could be seen by human observers walking the transect line, was 129. This number corresponds to 14.33 nests/ha, using the w estimate for all these nests (w = 18.01, choosing the perpendicular distance randomly from among the teams that had recorded the nest). The comparison indicates that the experienced teams still missed a considerable number of nests that could potentially be seen. Thus, they achieved only 73 and 64% of the best estimate after the first pass, and 76 and 70%, respectively, of this best estimate after passing the transect line in both directions. This finding suggests that single-pass line transect estimates of even experienced nest counters can underestimate the best possible density based on line-transects by at least 30%, whereas double-pass estimates still underestimate it by an average of at least 27%.

The 11 plots along the transect lines yielded a mean number of 3.09 nests (SD 1.87), corresponding to a nest density of 15.5 nests/ha. The estimated nest density based on the combined sample for the same part of the transect line as covered by the plots was 15.0 (based on 27 nests, and using the w estimate for the whole sample of 129 nests). The value of 15.0 nests/ha, however, is still 44% higher than the mean value of the two experienced line-transect teams (who had 10.0 and 10.9 nests/ha).

An attempt was made for the plot counts to be complete. Plot counts were more complete than the line transect counts of the same transect segments. Given a w value of 18.0 m in the combined sample, the expected number of nests in the plot based on the transect count would have been 27·(20/18)=30. Thus, with 34 observed nests, plots provide a 13% better coverage than transect lines, although plots were still not perfect: three nests that had been recorded along the transect line were subsequently missed during the plot counts. Correcting again for the w value of 18.0 m in the combined sample, this would translate to a number of missed nests of 3·(20/18), or 3.33 nests of a total of 34 recorded nests. Hence, the overall best estimate of the true density of nests would be 37.33 nests in 2.2 ha, or 16.97 nests/ha.

Because the number of independent and hence, widely spaced plots along the line transect was limited, we also placed an additional ten plots elsewhere in the study area in similar habitat. Their mean number of 3.00 nests per plot hardly changes the mean estimated nest density: 15.24 nests/ha. Thus, even the small sample of plots considered here can be considered adequate to characterize the nest density. The mean and 95% confidence limits for the density based on 11 plots are therefore 15.45±6.28, and for the one based on all 21 plots are 15.24±5.31. Obviously, inclusion of more plots narrows the confidence limits.

ranges of known adult females. Female density is

2 therefore 1.37 adult females/km , and total orangutan

2 density at Tuanan is therefore 4.25–4.5 individuals/km .

Thus, densities based on indirect methods can be compared to this best estimate of the true density.

Comparisons with the nest-based estimates are made possible by extensive direct observations at Tuanan, which produced local estimates of p, and r. Because 34 of 37 focal animals built nests, we estimate p=0.88. and 23.7%, respectively, compared to the mean increase Based on a sample of 391 complete focal follow days, observed here of 15% (and 12 and 13% for the two and using an average of the means per age–sex class experienced teams). Nonetheless, it is unlikely that weighted for representation in the population, we esti- multiple passes produced a ‘true’ density in their studies. mate r=1.15. To estimate t, we follow Morrogh-Ber- Both Buij et al. (2003) and the present study used nard et al.’s (2003) most conservative estimate for independent data to estimate the true density, and found Sebangau, the nearest site with similar habitat, at that double-pass line transect estimates, were still far 350 days. below the estimated true densities. Increasing the num

In Table 2, we summarize the results of the various ber of passes is unlikely to improve this situation: techniques. Whereas the line transect technique under- Johnson et al. (2002) did a third pass in some cases, but estimates densities (although the cumbersome combined found very few new nests. sample including all observers does so to the smallest Densities estimated through the line transect method extent), the plot counts provide estimates close to the were considerably lower than those obtained by plot estimated true density of 4.5 individuals/km2 . Confi- counts (Table 2), although the combined line transect dence limits are relatively higher for the plots than for sample yielded only marginally lower estimates than the line transects based on individual pairs, as expected those based on plots or female home ranges. Plot counts because of the larger number of nests sampled in the line still missed a small proportion of nests, but came very transects. Thus, a relatively larger number of plots close to the estimated true density. Thus, the plot would be needed to reach similar reliability as in line method is superior to the line transect method in this transects, but a small number of plots already gives a case. First, even the best line transect estimates remained more accurate estimate. well below the true density, by a margin that normally cannot be estimated without resorting to other methods (cf. Fig. 1b). Second, the plot method does not take Discussion much more time to produce these better estimates. A double pass of the two line transects would take a pair of The simple experiment in nest surveying produced three observers 2 days (progress is slow and the high number interesting findings about line transects of nests. First, of measurements in the dense forest takes much time). In experience matters: inexperienced teams generally pro- the same amount of time, this team can cover 10–12 duced lower estimates than experienced teams, contrary plots. Although estimates based on the combined line to the naı¨ ve expectation that lack of experience will be transect sample are also reasonably accurate, this procompensated for by narrower strip widths. Second, line cedure requires the presence of multiple teams and transects underestimate nest densities: even highly tedious and time-consuming data sorting. experienced teams tended to underestimate nest density The plot method also has a weak point, namely its by an average of 27%. This estimate is conservative lower reliability. A relatively larger number of plots is because even the combined sample missed some nests required to reach similar confidence limits to those right above or near the trail. Third, the second pass, reached by line transect methods (assuming homogewhile increasing the estimated nest density, still yielded neous habitats). However, it is obviously better to have density estimates well below the realistic minimum esti- an unbiased estimate that can be made reliable with mate. R. Buij (personal communication) and Johnson some additional effort (the plot method) than a method et al. (2002) also noted an increase in the number of that is more reliable but produces an unknown downnests on the second pass. Their increases were higher, 20 ward bias (line transects). By adding more well spaced plots, one can reduce the confidence limits to the lowest values admitted by habitat heterogeneity. Stratified Table 2 Summary comparison of the results of different methods random sampling of clearly identified habitat types is of estimating orangutan densities in Tuanan (assuming p=0.88, also easy, because plots can still be classified well after r=1.15, and t=350) sampling.

In areas where plots can be laid out and traversed Method (Dde/knmsit2y) l9i5m%itsconfidence with relative ease (as in this peat swamp forest), the results of plot counts should be superior to those of line Line transect: experienced pairs A: 2.95 A: 2.40–3.62 transects. Indeed, a replication by Simon Husson (A and B), single pass B: 2.58 B: 2.30–2.89 (unpublished) in another Bornean peat swamp conLine transect: experienced pairs A: 3.08 A: 2.52–3.76 firmed our conclusion. However, it is not necessarily L(iAneatnradnBse)c,td: ocuobmlebipnaesds sample, 4B.:024.82 B3.:126.–458.–137.21 true that plot counts are equally superior in structurally double pass more complex forests (most dry-land forests). Even Plots (complete sample), 4.30 2.80–5.80 though trees can be examined form all angles and all uncorrected distances, more nests may be missed in plots in such Plots (complete sample), 4.79 - forests. Hence, more comparisons are needed before we corrected for missed nests Female home range estimation 4.25–4.5 - can recommend the plot method as universally superior to line transect counts. Nonetheless, we expect that plots are more likely to overcome the inherent underestimation of densities due to the nests missed above or near the transect line, assuming that observers are experienced.

The size and shape of plots should depend on forest structure. Plots can be sited around trail intersections to facilitate access and measurement. Great care should not only be taken to identify all the nests in the plots, but also to measure their location and exclude those outside the plot, especially in narrow plots with a relatively long edge. Obviously, using plots does not obviate the need to estimate t, but these estimates can be obtained through repeat visits to plots, and are becoming increasingly available for different habitats.

The experiment reported here implies that orangutan densities in the literature are often too low. Most are based on line transects walked by small teams, often involving a double pass. In this study, such estimates (including only experienced teams) were 28–37% below the best estimate of the true density based on the home range methods, and 28–34% below the density based on nest counts in plots. Nonetheless, even if actual numbers are significantly higher than many reported to date, the trends in numbers noted by all studies that compared different points in time remain unambiguous. Orangutan habitats are declining seriously in both quality and extent through forest conversion, logging, fires, and poaching; and orangutan numbers almost certainly show the same trend (e.g. Rijksen and Meijaard 1999; van Schaik et al. 2001; Wich et al. 2003) .

Acknowledgements We thank the BOS-Foundation for permission to work at Mawas, and its staff for the great support of all activities; the staff and students from Universitas Nasional, Jakarta (Tatang Mitra Setia, Imran Said L. Tobing, Didik Prasetyo, Dwi Mulyawati, Tirza Yohana, Lili Aries Sadikin, Adi Hadinata, Ari Meididit, Fikty Apirlinayati, Fitriah Basalamah, and Ika Mian Karlina) for participating in the line transect experiment; the Indonesian Institute of Sciences (LIPI) for permission to work in Indonesia; the L.S.B. Leakey Foundation for financial support of the Tuanan Orangutan Project; the Netherlands Organisation for Scientific Research for financially supporting S.A. Wich, and Meredith Bastian and Maria van Noordwijk for comments on the MS.