# Do line-transect surveys systematically underestimate primate densities in logged forests.

код для вставкиСкачатьLmerican Journal of Primatology 13:l-9 (1987) RESEARCH ARTICLES Do Line-Transect Surveys Systematically Underestimate Primate Densities in Logged Forests? JOSEPH P. SKORUPA Graduate Group in Ecology, Department of Anthropology, University of California, Davis, California The rate at which social groups of primates are encountered in disturbed rain forest may be biased relative to undisturbed rain forest. A recently reported case study revealed a 25%reduction in postlogging raw encounter rates even though the true density of primates remained at the prelogging level. If biased raw encounter rates are typical of disturbed forests, and if they translate into equally biased line-transect density estimates, results of many comparative surveys might prove misleading (ie, apparent declines of primates in disturbed forest may not be real). Here a set of line-transect density estimates from logged forest are tested for systematic bias by comparing them to range-mapping density estimates, and the response of a Fourier series detectability function to several hypothetical patterns of bias in raw encounter rates is illustrated. Tests of line-transect density estimates from logged forest provide no evidence of systematic bias. The Fourier series results suggest that biased raw encounter rates may often be ameliorated by line-transect density estimators. Available evidence suggests that linetransect density estimates or similarly transformed encounter rates usually provide reliable comparative results within the limits of a particular study’s resolution. In contrast, conclusions drawn directly from comparative raw encounter rates (without transforming them into density estimates) are more prone to error. Key words: census techniques, disturbed rain forest, bias, Hylobates lar, Presbytis melalophos, Presbytis obscum, Colobus guereza INTRODUCTION As human influence on tropical rain forests continues to expand at increasing rates (eg, Myers, 1984; Caufield, 19851, it has become apparent that assessments of the capacity of primate species to survive in disturbed rain forest should be an integral component of long-term conservation plans [Johns, 1985a; Skorupa, 1986; Johns & Skorupa, in press]. Forests that are managed for sustainable selective logging represent one category of disturbed forest that may hold particular promise for primate conservation [eg, Johns, 1983al. To accurately assess the conservation value of selectively logged forests, it is essential that strictly comparable estimates of primate densities in unlogged and logged forest be obtainable. Johns [1985b] recently reported that changes in primate behavior induced by logging disturbance resulted in reduced detectability of primates in logged forest. Received July 8,1986; revision accepted November 28,1986. Address reprint requests to Joseph P. Skorupa, Dept. of Anthropology, Univ. of California, Davis, CA 95616. 0 1987 Alan R. Liss, Inc. 2 I Skorupa Johns compared matched seasonal samples collected before and after logging at a West Malaysian study site and found that encounter rates with primate social groups declined by 25% after logging, even though primate densities were verified to have remained unchanged. More importantly, Johns [1985b] argued that although raw encounter rates after logging were being reduced by an increased propensity for primates to “freeze,” by more time spent resting, and by a decreased propensity to vocalize, line-transect estimates of effective census strip width should not be expected to change (eg, “detectability functions [of line-transect estimators] . . . cannot consider differences in certain behavioral parameters between habitats. . . . the cumulative effect is to reduce the proportion of groups (sighted) within the (same) effective sample strip width”). Consequently, Johns implied that the biased (reduced) raw encounter rates he observed could be expected to translate into equally biased (reduced) line-transect density estimates. This implication is not trivial, because the detectability functions that some line-transect density estimators are based on were formulated to be responsive to changes in detectability [ie, altered encounter rates; cf Burhnam et al, 19801. Although Johns [1985b] tenders no such claim, his paper does leave one wondering whether line-transect surveys in logged forest might not systematically underestimate primate densities-a possibility with significant implications for conservation planning (eg, are reported declines of primates in logged rain forest merely sampling artifacts?). In this study data from Johns’s West Malaysian study area were used t o test whether line-transect estimates of primate density in logged forest show a consistent negative bias (ie, underestimation). The results of these tests do not indicate a consistent negative bias. “herefore, census data from Kibale Forest, Uganda were used t o simulate a 25%reduction in encounter rates [as Johns, 1985b, reported]; it was found that under reasonable selection criteria (for dropping 25%of the original data points), a Fourier series line-transect density estimator yields closely comparable “before” and “after-logging” density estimates. Contrary to Johns’s inference, estimates of effective strip width can be altered by biased encounter rates. This result provides at least one set of conditions consistent with the failure to detect systematic negative bias in density estimates for logged forest. There is, however, at least one selection criterion that doesn’t ameliorate the effects of biased encounter rates, and that is when bias is uniformly distributed with respect to distance from the census line. MATERIALS AND METHODS Johns [1983c, 19861 spent two- to three-week observation periods surveying primate densities in each of three logged hill dipterocarp forest plots in the Sungai Tekam Forestry Concession, Pahang, West Malaysia. Primate densities were estimated by two methods: (1)mapping the approximate home ranges of primate social groups and (2) using a Fourier series line-transect estimator [Johns,1983bl. Range mapping is generally considered to provide the most accurate approximation of true density for rain forest primates [eg, Struhsaker 1975; National Research Council, 19811, and therefore John’s line-transect density estimates for logged forest can be tested for systematic bias by comparing them to the range-mapping estimates of true density. This test does not depend on the absolute accuracy of the rangemapping estimates, but it does assume that the range-mapping estimates consistently will be closer to the true density than will the line-transect estimates. This assumption is particularly likely to be valid when line-transect estimates are based on minimal sampling effort, as is the case here [cf Johns,1983b]. To test for systematic bias the frequencies of apparent overestimation (positive bias) and underestimation (negative bias) were tested against the null hypothesis of random error (ie, equal frequencies of positive and negative bias) using the binomial Problems of Censusing Primates / 3 test [Siegel, 19561. Second, differences in the average magnitude of apparent positive vs negative bias were assessed by using the Mann-Whitney U test [Siegel, 19561. To examine the response (or nonresponse) of a Fourier series detectability function to biased (reduced)encounter rates, a set of census data was used that had been generated from sightings of red colobus monkeys (Colobus badius) in unlogged portions of Kibale Forest, Uganda, during a recent two-year field study [Skorupa, 19861. First, a Fourier series density estimate was calculated for the baseline data set to serve as a hypothetical “before-logging” standard. Then, a 25% reduction in encounter rates was simulated by removing 25% of the baseline data points according to the selection criteria described below. Case I It is assumed here that all primates within 10 m of the census line will flush and therefore be detected regardless of whether the observer is in logged or unlogged forest. It is also assumed that primates sufficiently removed from the census line will not react (ie, fi-eeze,hide, etc) to the observer in either type of forest. Therefore, the 25% reduction in data points is achieved by assuming that any point between 10 and 60 m from the census line has an equal probability of being deleted. The specific probability is set by the number of points to be deleted in order to reduce the entire data set by 25%. In this case there are 74 points between 10 and 60 m and 25% of the total sample (n=112)equals 28; therefore the probability of deletion equals 281 74 (0.378). This approach explicitly solves for the theoretically expected mean outcome of the process whereby intermediate points are randomly deleted. Case I1 Here it is assumed that all primates within 10 m of the census line will flush but that behaviors such as resting or vocalizing will otherwise be unaltered at all other distances. Therefore all data points beyond 10 m from the census line are assumed to have equal probability of deletion. Case I11 Here the 25% reduction in data points is achieved by assuming that all points have an equal probability (0.25)of deletion. For each of the above hypothetical patterns of bias, Fourier series density estimates (ie, theoretically expected mean values) were calculated for comparison against the previously established nonbiased standard. RESULTS Tests for Systematic Bias Data for the three most common primate species at Sungai Tekam- Hylobates lac Presbytis melalophlos, and I! obscura-provide nine pairwise comparisons of range-mapping and line-transect density estimates for logged forest (Table I). Linetransect density estimates appear t o overestimate true density in four cases while underestimating true density in four cases (with one case showing no apparent bias); a 1:l ratio of overestimates to underestimates is exactly the ratio expected from random error: Therefore the null hypothesis cannot be rejected (binomial test, N =8, x =4, P> .60, one-tailed probability). The magnitudes of overestimates range from +8% to +50% while the magnitudes of underestimates range from -7% to -46%. There is no significant difference in the average magnitudes of apparent positive and negative bias (nl=4, n2=4, U=7, P> .40, one-tailed probability). Thus, both the frequency and the magnitude of negative bias exhibited by line-transed density estimates for logged forest seem 4 I Skorupa TABLE I. Comparison of Line-Transect Estimates of Primate Densities (Groupsh') to Range-Mapping Estimates in 3 Logged Study Areas* Species Hylobates lar Range-mapping estimates Line-transect estimates Relative bias Presbytis melalophos Range-mapping estimates Line-transect estimates Relative bias Presbytis obscura Range-mapping estimates Line-transect estimates Relative bias 1-2-year-old logged 3-4-year-old logged 5-6-year-old logged 2.75 1.76 - 36% 2.50 3.76 +50% 2.75 1.51 2.26 3.01 +33% 4.52 4.89 +8% 3.39 3.39 0% 1.90 1.76 - 7% 2.37 3.02 +27% 1.42 0.76 -46% -45% * *Line-transect estimates of primate densities were calculated by using the Fourier series estimator presented by Burnham et a1 [1980]. Data i n this table are from Johns [1983b: Table 4.19; see also Johns, 1986: 2091. consistent with the hypothesis that, just as expected in unlogged forest, error is random (ie, nonbiased). Fourier Series Detectability Function Responses The baseline data set (Table II) results in a Fourier estimate of effective stri width (ESW) that equals 100.0 m; this leads to a density estimate of 10.4 groupskm (Table 110. The absolute accuracy of the above density estimate is not an issue that is relevant to this paper: This paper is concerned with the accuracy of betweenhabitat comparisons. Under the assumptions of case I (see Materials and Methods), the theoretically expected Fourier estimate of ESW is 79 m, which largely compensates for the simulated 25% negative bias (reduction) in encounters, leading to a n expected density estimate of 9.8 groups/km2 (Tables 11,111).Thus, while the bias in encounter rates is 25%, the bias in the expected line-transect density estimate is only 5%. In case I1 the expected line-transect density estimate is again substantially less biased than the raw encounter rate that the density estimate is based on. The completely random selection criterion of case I11 results in an unaltered expected estimate of strip width (ESW) and therefore a 25% underestimate of true density (Tables 11, 111). !l DISCUSSION Failure to detect systematic negative bias in Johns' [1983b, 19861 line-transect density estimates for primates inhabiting logged forest is consistent with several possible explanations. It could indicate that the assumptions of the hypothetical cases I and I1 are generally valid in logged forest, and therefore that biased raw encounter rates are ameliorated by line-transect density estimators to a point that the bias is insignificant relative to the coarse resolution (ie, high random sampling error) associated with most primate surveys in rain forest. Even with intensive sampling, resolution only approaches +25% for common primate species and is more likely to be around +45% for species of typical rarity [Skorupa, 19861. Alternatively, Problems of Censusing Primates / 5 the results may simply indicate that the effect Johns observed is temporary, because Johns [1985b] recorded biased encounter rates shortly following logging, and the data tested here come from older logged forest in which encounter rates could already have returned to normal. As a third alternative, the biased encounter rates in Johns’s main study plot may have been an artifact of having habituated the study groups of primates prior to logging (with the groups regaining normal wariness after the trauma of logging). For surveys of unhabituated primates this plausible source of bias would not be a concern (A.D. Johns, personal communication). Finally, the results may only reflect the low power of tests based on small sample sizes. The “low-power” alternative seems unlikely since there is not the slightest hint of deviation from expectations based on random error. Furthermore, another data point that can be added from Kibale Forest also shows no obvious negative bias. Despite the fact that black-and-white colobus monkeys (Colobus guereza) are extremely fearful of humans in logged forest, the line-transect density estimate of 5.9 groups/km2 fits well with the range-mapping estimate of 5.0-6.2 groups/km2 (Skorupa, unpublished data). Unfortunately, none of the other Kibale primates concentrate their activities within small, distinct core areas to the extent that C. guereza does [Oates, 1974; cf. Struhsaker Leland, 19791. Therefore, obtaining range-mapping density estimates for unmarked social groups of the other species occupying logged forest is problematic. Regardless of which explanation best applies in this particular case, the hypothetical test cases indicate that biased raw encounter rates need not necessarily lead to an expectation of equally biased line-transect density estimates (cases I, II,Table 111).Most methods for transforming raw encounter rates into density estimates (ie, line-transect density estimators)assume that there is some region around the census line in which the detection probability equals 1.0 [eg, Burnham et al, 19801. Furthermore, most methods assume that a plot of detection frequencies (as a function of distance from the census line) should show a shoulder roughly corresponding to the 1.0 detection probability region [eg, Burnham et al, 19801. Ultimately, it is the height of the shoulder that determinesthe density estimate. Thus, for the entire class of what are referred to here as “shoulder density estimators,” biased raw encounter rates will only influence density estimates to the extent that they alter the estimated height of the frequency distribution’sshoulder. The estimated height of the shoulder is determined primarily by the frequency of encounters very close to the census line (ie, inside the 1.0 detection probability region). This explains why the patterns of bias in cases I and I1 had so little affect on density estimates. The hypothesized “flush response” of primates near the census line insured that bias within the shoulder region was lower than the general level of bias. Estimates of ESW, by contrast, are measuring how far the shoulder can be extended by compressing the tail of the detection frequency curve back toward the origin and to the height of the shoulder. Clearly, estimates of ESW should decline as points are preferentially deleted from the tail (ie, the region beyond the shoulder). In fact, if all deleted points came from the tail, the new ESW estimate should be expected to exactly compensate for the deletions (bias) and lead to unbiased density estimates. Depending on the specific density estimator, the width of the shoulder region, the width of the flush response region, and the magnitude of encounter rate bias, line-transect density estimates can range from showing no bias to being nearly as biased as the underlying encounter rates. However, as long as any flush region exists, density estimators will be less biased than raw encounter rates. Apparently, Johns [1985133 expected biased raw encounter rates to translate into equally biased line-transect density estimates because he was assuming that the 6 / Skorupa TABLE 11. Distributions of Perpendicular Sighting Distances Used to Calculate the Fourier Series Density Estimate Statistics Presented in Table III* No. sightings Perpendicular distance (m) 0 1 2 3 5 6 7 8 10 11 12 13 15 16 17 18 19 20 21 22 23 25 26 27 28 30 31 32 33 35 36 37 38 40 41 42 43 45 46 47 49 50 51 54 56 57 58 60 61 62 64 Baseline data set 3 4 4 2 2 1 1 3 3 2 2 3 3 2 2 4 1 4 3 2 3 3 3 2 2 2 2 1 2 2 1 1 1 2 1 1 3 2 2 1 1 1 1 1 2 1 1 1 1 1 1 Case I Case I1 3 4 4 2 2 1 1 3 3 1.243 1.243 1.865 1.865 1.243 1.243 2.486 0.622 2.486 1.865 1.243 1.865 1.865 1.865 1.243 1.243 1.243 1.243 0.622 1.243 1.243 0.622 0.622 0.622 1.243 0.622 0.622 1.865 1.243 1.243 0.622 0.622 0.622 0.622 0.622 1.243 0.622 0.622 1 1 1 1 3 4 4 2 2 1 1 3 3 1.371 1.371 2.056 2.056 1.371 1.371 2.741 0.685 2.741 2.056 1.371 2.056 2.056 2.056 1.371 1.371 1.371 1.371 0.685 1.371 1.371 0.685 0.685 0.685 1.371 0.685 0.685 2.056 1.371 1.371 0.685 0.685 0.685 0.685 0.685 1.371 0.685 0.685 0.685 0.685 0.685 0.685 Case I11 2.25 3 3 1.50 1.50 0.75 0.75 2.25 2.25 1.50 1.50 2.25 2.25 1.50 1.50 3 0.75 3 2.25 1.50 2.25 2.25 2.25 1.50 1.50 1.50 1.50 0.75 1.50 1.50 0.75 0.75 0.75 1.50 0.75 0.75 2.25 1.50 1.50 0.75 0.75 0.75 0.75 0.75 1.50 0.75 0.75 0.75 0.75 0.75 0.75 Problems of Censusing Primates / 7 TABLE 11. Distributions of Perpendicular Sighting Distances Used to Calculate the Fourier Series Densitv Estimate Statistics Presented in Table III* (continued) No. sightings Perpendicular distance (m) 66 68 69 71 75 83 84 89 90 Samde size Baseline data set 2 1 1 1 1 1 2 1 1 112 Case I Case I1 Case III 2 1 1 1 1 1 2 1 1 84 1.371 0.685 0.685 0.685 0.685 0.685 1.371 0.685 0.685 84 1.50 0.75 0.75 0.75 0.75 0.75 1.50 0.75 0.75 84 *The baseline data are generated from actual observations of red colobus monkeys in unlogged portions of Kibale Forest. See Materials and Methods for a discussion of how the other distributions were derived. TABLE 111. Parameter Values for Fourier Series Line-Transect Density Estimates Correspondingto Several Hypothetical Patterns of Bias in Raw Encounter Rates+ Parameter* n W" (m) 1/W* 2/nW* 1/W*(2/n+1P2 a1 a2 a3 a4 lla ESW (m) L (km) Density (groups/km2) % bias Baseline 112 90 0.01111 0.00020 0.00148 0.00890 0.00115 - 0.02001 100.0 108.0 10.4 NIA Case I1 I 84 90 0.01111 0.00026 0.00170 0.00848 0.00348 0.00222 0.00078 0.02529 79.0 108.0 9.8 -5.2b 84 89.08 0.01122 0.00027 0.00172 0.00996 0.00285 0.00156 - 0.02403 83.2 108.0 9.3 -9.9b III 84 89.35 0.01119 0.00027 0.00172 0.00884 0.00115 - 0.02003 99.9" 108.0 7.8 -24.gb +See Burnham et a1 [1980]for definition of parameters and for the details of calculating Fourier series line-transect density estimates. Parameters calculated from data presented in Table II. Effective strip width (ESW) and density values are rounded to the nearest tenth. aThe difference between this value and the baseline value is due to rounding error. bBias calculated from precise density estimates, not from rounded estimates. *W*= maximum value of perpendicular distance; n = sample size; ak = kth term of Fourier series; a = l/Z ESW; L = census strip length. bias in detectability would be uniformly distributed with respect to distance from the census line [as in case III; Johns, personal communication]. The "flush" assumption of test cases I and I1 probably leads to more realistic and generally applicable expectations than does case III. Johns [1985b] reported that even freezing (hiding) primates flushed when approached closely, and there is no reason to doubt that resting and nonvocalizing primates would behave similarly. Accordingly, patterns of bias should tend to be concentrated away from the census line. 8 I Skorupa Johns [1985b] rightfully cautions that comparative census data from undisturbed and disturbed forest must be interpreted very carefully; however, any doubts about the validity of reported declines of primates in disturbed rain forest that may have been engendered by his paper should be tempered by the knowledge that biased raw encounter rates in disturbed forest do not automatically translate into equally biased line-transect density estimates. If Johns’ results don’t represent a temporary effect or an artifact of habituation, then the full weight of his cautionary advice does apply to those who attempt to draw conclusions directly from comparative raw encounter rates. In summary, between-habitat variation of primate densities is best assessed via intensive long-term studies that use a combination of censusing techniques as crosschecks on results [eg, Johns 1983b], but, when short-term surveys are all that can be afforded, available evidence (n= 10) suggests that line-transect density estimates or similarly transformed encounter rates provide reasonably reliable comparative results within typically coarse limits of resolution. CONCLUSIONS 1. The effect that biased raw encounter rates with primate social groups in disturbed rain forest will have on line-transect density estimates depends on the distribution of bias relative to distance from the census line and on the type of density estimator being employed. When bias is unequally distributed because of a “flush response” (ie, when normally detectable groups that go undetected due to biasing behaviors are concentrated away from the census line), and when a “shoulder density estimator” is being employed, line-transect density estimates ameliorate the biased raw encounter rates. 2. Although it is not clear that biased encounter rates are typical of censuses in disturbed rain forest, if they are, the sparse available data (n= 10) appear consistent with the expectations derived from “flush response” test cases and inconsistent with the assumption that bias is uniformly random relative to distance from the census line. 3. Between-habitat variation of primate densities is best assessed via intensive long-term studies that utilize a combination of censusing techniques as cross-checks, but, when short-term surveys are all that can be afforded, line-transect density estimates or similarly transformed encounter rates usually provide reasonably reliable comparative results within coarse limits of resolution. 4.Conclusions about habitat suitability that are drawn directly from comparative raw encounter rates are more prone to error than similar conclusions based on density estimates. ACKNOWLEDGMENTS The baseline data for the test cases were collected during a field study supported by a grant from World Wildlife Fund-US (project No. 1969) and by the New York Zoological Society (both through Dr. Thomas T. Struhsaker). The author is also grateful to the Ugandan National Research Council, Uganda Department of Forests, and the Department of Zoology, Makerere University, for local sponsorship. This paper benefited from comments on earlier drafts offered by L. Berenstain, T.M. Butynski, A.D. Johns, J.F. Oates, and B.O. Manullang. I am particularly indebted to Dr. Andrew D. Johns for freely sharing numerous insights relevant to this paper. REFERENCES Burnham, K.P.; Anderson, D.R.; Laake, J.L. ESTIMATION OF DENSITY FROM LINE TRANSECT SAMPLING OF BIOLOGI- CAL POPULATIONS. WILDLIFE MONOGRAPHS NO. 72, Washington, D.C., The Wildlife Society, 1980. Problems of Censusing Primates / 9 Caufield, C. IN THE RAINFOREST. New York, Alfred A. Knopf, Inc., 1985. Johns, A.D. Tropical forest primates and logging: Can they coexist? ORYX 17:114-118, 1983a. Johns, A.D. ECOLOGICAL EFFECTS OF SELECTIVE LOGGING IN A WEST MALAYSIAN RAIN FOREST. PhD Dissertation, University of Cambridge, England 1983b. Johns, A.D. Selective logging and wildlife conservation in tropical rain-forest: Problems and recommendations. BIOLOGICAL CONSERVATION 31:355-3 75,1985a. Johns, A.D. Differential detectability of primates between primary and selectively logged habitats and implications for population surveys. AMERICAN JOURNAL OF PRIMATOLOGY 8:31-36,1985b. Johns, A.D. The effects of commercial logging on a West Malaysian primate community, (pp 206-211 in PROCEEDINGS OF THE IXth IPS CONGRESS. F.A. King; D. Taub, eds. New York, Van Nostrand Reinhold, 1986. Johns, A.D.; Skorupa, J.P. Responses of rainforest primates to habitat disturbance: A review. INTERNATIONAL JOURNAL OF PRIMATOLOGY, in press. Myers, N. THE PRIMARY SOURCE. New York, W.W. Norton and Co., 1984. National Research Council. TECHNIQUES FOR THE STUDY OF PRIMATE POPULATION ECOLOGY. Washington, D.C., National Academy Press. 1981. Oates, J.F. THE ECOLOGY AND BEHAVIOUR OF THE BLACK AND WHITE COLOBUS MONKEY (COLOBUS GUEREZA RUPPELL) IN EAST AFRICA. PhD Dissertation, University of London, England, 1974. Siegel, S. NONPARAMETRIC STATISTICS. New York, McGraw-Hill Book Co., 1956. Skorupa, J.P. Responses of rainforest primates to selective logging in Kibale Forest, Uganda: A summary report, pp 57-70 in PRIMATES: THE ROAD TO SELF-SUSTAINING POPULATIONS. K. Benirschke, ed. New York, Springer-Verlag, 1986. Struhsaker, T.T. THE RED COLOBUS MONKEY. Chicago, Illinois, University of Chicago Press, 1975. Struhsaker, T.T.; Leland, L. Socioecology of five sympatric monkey species in the Kibale Forest, Uganda. ADVANCES IN THE STUDY OF BEHAVIOR 9:159-228,1979.

1/--страниц