Using Search Time and Regression to Estimate Abundance of Territorial Spotted OwlsAuthor(s): James P. Ward, Jr., Alan B. Franklin, R. J. GutierrezReviewed work(s):Source: Ecological Applications, Vol. 1, No. 2 (May, 1991), pp. 207-214Published by: Ecological Society of AmericaStable URL: http://www.jstor.org/stable/1941813 .Accessed: 15/03/2012 13:30

Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at .http://www.jstor.org/page/info/about/policies/terms.jsp

JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range ofcontent in a trusted digital archive. We use information technology and tools to increase productivity and facilitate new formsof scholarship. For more information about JSTOR, please contact support@jstor.org.

Ecological Society of America is collaborating with JSTOR to digitize, preserve and extend access toEcological Applications.

http://www.jstor.org

Ecological Applications, 1(2), 1991, pp. 207-214 ? 1991 by the Ecological Society of America

USING SEARCH TIME AND REGRESSION TO ESTIMATE ABUNDANCE OF TERRITORIAL SPOTTED OWLS'

JAMES P. WARD, JR., ALAN B. FRANKLIN, AND R. J. GutiuRREz Department of Wildlife, Humboldt State University, Arcata, California 95521 USA

Abstract. Using conventional sampling methods, unbiased, precise estimates of the number of individuals in a population can be difficult to obtain for rare, secretive species. We used a Leslie regression model (LRM) to estimate the number (No) and the variance of the number (V[No]) of adult and subadult Northern Spotted Owls in a territorial pop- ulation from direct counts within a 292 km2 study area. Estimates of No from day counts were more accurate and precise and more robust to fluctuations in survey effort than estimates from night counts. LRM estimates from day counts were not significantly different from two different maximum-likelihood estimates and required 30-64% less effort. These findings suggest a less costly method for statistically comparing Spotted Owl abundance between spatial or temporal units.

Key words: abundance; call survey; Leslie regression model; maximum likelihood estimator; North- ern Spotted Owl; population size; Strix occidentalis caurina; territorial.

INTRODUCrION

The number of individuals alive in a population at a given time (N,) is a fundamental parameter of many ecological studies. Methods of estimating size of ani- mal populations are well documented and thorough reviews can be found in Caughley (1977), Seber (1982, 1986), and Krebs (1989). However, unbiased and pre- cise estimates of (N,) using conventional sampling methods may be difficult to obtain for some species (e.g., Burnham et al. 1980, Reynolds et al. 1980, White et al. 1982) because sampling assumptions are difficult to meet and estimators are not robust to violations of assumptions. Specifically, species that use large areas, occur in low densities, or are difficult to observe (e.g., large predators, nocturnal or secretive species) cannot be sampled adequately without tremendous field effort.

Ironically, it is the large or rare predators that may be promoted as "umbrella" (Peterson 1988), "key- stone" (Norton 1988), or "indicator" (Salwasser et al. 1982, Salwasser 1988) species in an effort to preserve biodiversity. For example, in the Pacific Northwest, maintenance of Northern Spotted Owl (Strix occiden- talis caurina) populations is promoted as a way to con- serve old-growth coniferous forests (Thomas et al. 1988). Proposed plans to maintain viable owl popu- lations consist of a network of suitable habitat areas (United States Forest Service 1988). The size and struc- ture of those areas needed to maintain owl occupancy are becoming well documented (Solis 1983, Forsman et al. 1984, Ganey and Balda 1988, 1989, LaHaye 1988, Laymon 1988, Bias 1989, Carey et al. 1990, Sisco 1990). However, the arrangement and number of hab- itat areas needed to maintain viable populations of the

owl are poorly understood (Dawson et al. 1987). A logical starting point for understanding viability of Spotted Owl populations would be to model the ex- isting population, which would require an ability to estimate and monitor N. Any reasonable comparison of N, over time or between habitats would also require a measure of the sampling variance associated with Nt.

Spotted Owls are territorial, nocturnal predators that primarily live in old-growth conifer forests of the Pa- cific Northwest (Solis 1983, Forsman et al. 1984). Be- cause these owls are territorial and use large (400-1800 ha) tracts of habitat, populations attain relatively low densities (Forsman et al. 1984, Franklin et al. 1990). The owl's nocturnal habit and low density, coupled with steep, rugged terrain prohibit use of line transect (Burmham et al. 1980), or variable plot (Reynolds et al. 1980) methods for estimating their abundance. However, during March through August both members of an owl pair actively defend a breeding territory using vocal advertisements (Forsman et al. 1984). Young depend on their parents until dispersal and do not de- fend a territory before dispersing in the fall (Gutierrez et al. 1985). After their juvenile molt, subadult owls (1-2 yr old) can be recognized by plumage characters, and these birds may obtain and defend a territory. Thus, territorial, adult, and subadult Spotted Owls can be located consistently by imitating their calls and elic- iting a vocal response during the breeding period (Fors- man 1983).

We infer from banding studies that Spotted Owls are long lived and have high (0.96) annual adult survi- vorship (A. B. Franklin et al., unpublished data). Mor- tality in territorial adults occurs primarily in the non- breeding period (A. B. Franklin et al., personal observation). We occasionally encounter unbanded birds during nighttime calling surveys that we cannot relo-

1 Manuscript received 22 January 1990; revised and ac- cepted 23 July 1990.

208 JAMES P. WARD, JR. ET AL. Ecological Applications Vol. 1. No. 2

cate during the day. Since these owls are usually within the territory of a marked pair, we believe that these birds are nonterritorial members ofthe population. We know little about the demography and ecology of non- territorial birds. We also do not know about the night- time response rate of nonterritorial birds. However, the number of banded birds detected while spot-light- ing owls at night or observing owls at day roosts in- dicates that nonterritorial birds (both adult and sub- adults) do not call frequently, often have unstable home ranges, and are not consistently observable throughout a short time period. Thus, our observations of banded birds indicate that we are locating territorial birds al- most exclusively and that there is little immigration or emigration, and few births (e.g., juvenile owls are not located by calls) or deaths in the study population dur- ing the census period. These conditions allow us to meet some ofthe basic biological assumptions of catch- effort methods (Seber 1982:296, Krebs 1989:162).

In this paper, we use a Leslie regression model (LRM; Leslie and Davis 1939) to estimate abundance and the variance of abundance of Spotted Owls from direct counts. We also compare LRM estimates based on nighttime counts of owls with estimates based on day- time counts, and evaluate the utility of this method by comparing LRM estimates with two maximum like- lihood estimates (Zippin 1956, Jolly 1965, Seber 1965, White et al. 1982). This method may be applicable for estimating abundance of other species that behave sim- ilarly to the Spotted Owl.

METHODS

Study area Adult and subadult Northern Spotted Owls were

counted in a 292.4 km2 study area near Willow Creek, California, USA. Vegetation within this study area was classified as either mixed evergreen forest or Klamath montane forest (Kiichler 1977), and has been described in detail elsewhere (Solis 1983, Franklin et al. 1990). Approximately 43% of the study area was considered suitable Spotted Owl habitat (Franklin et al. 1990).

Terrain was steep and rugged; elevations and slopes ranged from 150 to 1700 m and from 18? to 550, re- spectively. Study area boundaries were drawn along prominent topographical features to minimize overlap of owl territories between the study area and adjacent habitat (Franklin et al. 1990).

Owl surveys Owls were located and counted using vocal imita-

tions of the species' calls during surveys at night (1/2 h prior to dusk to 1/2 h prior to dawn) and during the day, April through August, 1985 to 1987 (Forsman 1983, Franklin et al. 1990). Night surveys were conducted at call stations established along roads and, generally, were used to locate and count owls. Day surveys consisted of more extensive searches and, specifically, were used to locate and count owls by visual observation at roosts.

For each year, the time (in minutes) devoted by each observer to calling for, searching for, and locating owls was recorded separately for nighttime and daytime sur- veys and, hereafter, will be referred to as survey effort. Survey effort did not include travel time between call stations or survey routes, nor any additional field effort.

Surveys were conducted within geographic subunits of the study area. Subunits were delineated by observ- ers using topographic and habitat features, and rep- resented potential owl territories. These territories were used to assign survey effort to the number of owls counted. Once a pair of owls was located in a territory, additional owl responses and additional survey effort within the territory were ignored. If only a single bird was detected in a territory, additional effort was de- voted to locating a potential mate until a mate was counted or the annual census ended. If an area con- tained more than one pair, the number of territories and survey effort within each area were adjusted ac- cordingly. Thus, we used the association of an owl with a given territory to infer owl identity and the amount of time necessary to locate an individual without knowledge of banding records. Concurrently, we at- tempted to capture and mark all individuals with col- ored leg bands to confirm identity and evaluate this assumption (Franklin et al. 1990). Effort required for marking was also recorded.

Leslie regression model (LRAM) development

Call surveys provided a direct count without a mea- sure of precision. An LRM (Leslie and Davis 1939) was used to estimate the number of adults and sub- adults alive at the beginning of each breeding season (NO), and a 95% confidence interval for No.

The LRM is a closed-population model and works on the tenet that the number of individuals removed by sampling from a given area will decline over time (Leslie and Davis 1939, Seber 1982:296, Krebs 1989: 162). Thus, a regression line of the form

Y, = a + bX, (1)

can be fit to the number of individuals captured per unit of effort (Y1), and the cumulative number of in- dividuals captured prior to each survey interval i (Xi), using a least squares analysis (Zar 1984:263-265). Fol- lowing the notation of Seber (1982:296), and using the slope of the regression line (b) as an estimate of the coefficient of catchability (K), No can be calculated as

No =X+ (Y/K). (2)

The intercept of the x axis by the regression line also provides a point estimate of No (Seber 1982:298).

The variance of No, or V[NO], can be estimated as follows:

= 2i 1 (N- X)2]

May 1991 ESTIMATING OWL ABUNDANCE 209

where 12 is the mean square error from the least squares analysis and s is the regression sample size (Seber 1982: 298). A 95% confidence interval can be estimated from V[NO] using the delta method when s is large (> 10),

No ? 1.96(/N01])/2, (4)

or by using the procedure described by Seber (1982: 299) when s is small.

To use the LRM, we considered our vocal imitations as "bait" to entice owls to respond. Owls that were detected, either by sight or sound, were considered "captured." We calculated catch per unit effort, Yi, as the number of owls detected (n,) divided by the asso- ciated number of hours of survey effort (f), in each of s sampling periods. The latter were created by system- atically grouping survey results from a given year with an arbitrary window of time, m hours long (Fig. 1). Grouping of data was necessary because the amount of effort devoted and the number of owl territories sampled during any given survey were small relative to total survey effort (Fi) and the total number of owl territories sampled each year, respectively. Without grouping the data over time, too few owl territories would have been sampled and the numbers of owls detected during each sampling period would not have declined through time, a requisite for using the LRM.

For example, grouping the cumulative number of owls detected for every m = 23 h of effort devoted during daytime surveys in 1985 gives Y1 = 0.708 owls detected per hour and X1 would equal 0 owls detected previous to sampling period I (Appendix). Because the arbitrary selection of m might have influenced LRM results, X and Y values for all surveys were calculated by varying m from 2 to (F1/1 1) h, incremented every 1 h. The latter terminus was used to ensure that con- fidence intervals were calculated using s > 10 (Seber 1982:12, Krebs 1989:164). The data set that best adhered to model assumptions

was selected for estimating No, for each survey type, for each year. There are three assumptions of the Leslie method: the population is geographically and demo- graphically closed; the probability of an individual be- ing captured (i.e., detected; Pi) is constant throughout the experiment; and detection probabilities are equal among individuals in the population (Krebs 1989). De- tection probabilities can be considered constant and equal if sampling is a Poisson process with respect to effort (Seber 1982:296). The generalized removal method (White et al. 1982:109) also can be used to evaluate constancy and equality of pi. There are no simple tests for population closure (White et al. 1982: 162). However, if all assumptions are valid the data will fit a straight line and the variance about the re- gression line will be constant (Krebs 1989:162-164). Thus, data that (1) provided ni most likely distributed as a Poisson, (2) resulted in a regression slope least likely to equal 0, and (3) provided the lowest variance of observed values about the regression were used as

input into the LRM. The above criteria were quantified and evaluated using a chi-square goodness-of-fit test, a Student's t test, and the standard error of the regres- sion divided by Y, respectively. Statistical tests were deemed significant at the 5% level (a = .05). A BASIC computer program was developed to keep this proce- dure standardized among data sets and users.

LRM estimates of No were compared to (1) the min- imum number of individuals known to be alive (Nm) based on marking results (Franklin et al. 1990), (2) maximum likelihood estimates calculated using the generalized removal model (Mbh) of program CAP- TURE (Zippin 1956, Otis et al. 1978, White et al. 1982), and (3) Jolly-Seber, model D estimates from program JOLLY (Jolly 1965, Seber 1965,1982, Frank- lin et al. 1990).

The n, tallied from daytime surveys represented owls found at a roost. Because daytime roosts of individual owls were spatially distinct from roosts in adjacent owl territories, roosting owls could be counted and effec- tively "removed" from the population. Hence, the n, from the daytime data that best fit model assumptions were treated as removal data and input into the CAP- TURE Mbh model. Jolly-Seber estimates were inde- pendently calculated from mark-recapture results (Franklin et al. 1990).

Coefficients of variation for No were calculated as:

cv(NO) = {[V(No)]'2/NN} x 100. (5)

Expected estimates of No that deviate from true pa- rameter values are biased. Because we did not know the true population size we could not estimate statis- tical bias. However, we could evaluate each estimator by comparing estimates relative to the minimum num- ber of owls known to be alive, Nm, using the calculation:

o DAY-1985 70 -~f213"

ni tm h 11 n.

70-

C o) 501

Z 30- w I

10

0

0 50 100 150 200 250

CUMULATIVE HOURS SURVEYED

FIG. 1. Procedure for grouping Spotted Owl call survey data for input into a Leslie regression model, where m is the grouping (time) interval, ni is the number of owls detected in time group i, and f is the time in hours required to locate and count n,

210 JAMES P. WARD, JR. ET AL. Ecological Applications Vol. 1, No. 2

TABLE 1. Survey effort devoted to locating, marking, and recapturing Northern Spotted Owls, April through August, 1985 to 1987, near Willow Creek, California.

Hours of effort

Night survey Day survey Mark-recapture Total al No. Year h %* h %* h %* survey recapture owlst 1985 208.5 31.0 266.3 39.5 198.8 29.5 474.8 673.6 65 1986 95.5 12.8 199.4 26.6 453.3 60.6 294.9 748.2 65 1987 122.1 10.7 286.6 25.0 736.3 64.3 736.3 1145.0 67

* Percent of total hours devoted to survey and mark-recapture effort. t Number of adult and subadult individuals known to be alive in a 292.4 km2 area based on banding data.

[(No - Nm)/Nm] x 100. Negative results would indicate that the LRM provided an underestimate of No. Costs to produce LRM and Jolly-Seber model D estimates were considered a function of survey effort, and the sum of survey effort and mark-recapture effort, re- spectively.

RESULTS

Annual direct counts of adult and subadult Spotted Owls varied only slightly during 1985 to 1987, even though night and day survey effort fluctuated through time (Table 1). On average 91% of the owls counted within the study area each year were banded after de- tection.

Tests of assumptions

Probabilities for the Poisson goodness-of-fit tests from the three best fitting data sets generated by night sur- veys were, P < .25, .001, and .001 during 1985 through 1987, respectively, and P < .001, .10, and .001 for day surveys. All regression line slopes were significantly >0

(1987 night LRM, P < .05; all other LRMs, P < .00 1; Fig. 2). Standard errors of all regressions ranged from 29 to 68% of the mean Yvalue from each survey type. Thus, the assumption of constant detection probability was violated for 4 of the 6 best fitting regressions. All other assumptions appeared valid for selected regres- sion models. Additionally, the general removal method of program CAPTURE tested for and selected an es- timator of No assuming equal detection probabilities for day survey data collected in 1985 and 1986. Nine different detection probabilities were assumed in es- timating No from diurnal counts tallied in 1987.

Population estimates Estimates of the number of territorial adult and sub-

adult Spotted Owls in the study area varied with the method of estimation (Fig. 3). Confidence limits cal- culated using the procedure for small sample sizes (i.e., a quadratic formula; Seber 1982:299) were asymmet- rical, and upper limits were consistently greater than confidence limits calculated using the delta method (e.g., confidence intervals for 1985 night and day sur- veys were 71-692 and 51-71, respectively). The delta method was used to evaluate differences among esti- mators because (1) the upper limit calculated for 1985 night-based estimates using the quadratic formula ap- peared biologically unrealistic, (2) all samples were > 10 (Seber 1982:12, Krebs 1989:164), and (3) No appeared to be normally distributed (Seber 1982:12).

LRM estimates of No based on night surveys differed significantly (P < .05) from LRM estimates based on day surveys in 1986, and from maximum likelihood estimates in 1986 and 1987 (Fig. 3). During the 1st yr of sampling, night surveys resulted in an overestimate of Nm by as much as 71%, whereas night surveys in subsequent years underestimated Nm (Table 2). Annual LRM estimates based on night surveys appeared to fluctuate proportionally with annual survey effort.

In contrast, LRM estimates based on day surveys were generally consistent, with the exception of the 1 st yr when Nm was underestimated by 11% (Table 2). LRM estimates of No based on day surveys were equal to or less than maximum-likelihood estimates on 3 of 6 occasions. All but the 1985 night-based estimates were precise (cv[NO] < 10%; Table 2). LRM estimates

TABLE 2. Percent difference between direct counts and es- timates of the number of territorial Spotted Owls (No) in a 292.4 kM2 stuidy area near Willow Creek, Califomia, and precision of N,

Percent cv[No] Year Model* nt difference (%) 1985 Night LRM11 19 +70.7 34.5

Day LRM23 12 -10.8 6.9 Day CAP.Mbh 12 -6.2 7.5 Jol-Seb D 0 4.7

1986 Night LRM7 14 -20.0 9.8 Day LRM12 17 +7.7 7.8 Day CAP.Mbh 17 +13.8 9.8 Jol-Seb D +7.7 2.6

1987 Night LRM1 12 -9.0 9.2 Day LRM24 12 +6.0 7.0 Day CAP.Mbh 12 +1.5 4.9 Jol-Seb D +13.4 2.6

* LRM = Leslie regression model, CAP.Mbh = Program CAPTURE removal model, Jol-Seb = Jolly-Seber model D; subscripts represent the interval, m hours wide, used to group the data prior to model input.

t LRM n = number of x, y coordinates; CAP.Mbh = number of sampling sessions.

May 1991 ESTIMATING OWL ABUNDANCE 211

NIGHT SURVEYS 1985 DAY SURVEYS 1985 3 - Y 0.741 - 0.007X 1.5 Y- 0.612 - 0.OO I X

21

0 1 * .5 - Li..

t 0 20 40 60 80 0 20 40 60 80 0

NIGHT SURV/EYS 1986 DAY SURVEYS 1986 tk 3 Y:I.1823 -0.036 X 1.5 *Y:~0.735 -.0.010OX

CL w

a:

oH 2 1- - \

35 U . 03

U 0*0 U

I-. 0 20 40 60 80 0 20 40 60 80 0

X NIGHT SURVEYS 1987 DAY SURVEYS 1987 3 Y- 1.132-0.019X 1.5 Y 0.551 -0.008X

z 2 1

O '_ U .5

0 * g 55033

0 20 40 60 80 0 20 40 60 80

OWLS PREVIOUSLY DETECTED

FIG. 2. Regression of the number of individual Spotted Owls detected per hour of survey effort against the cumulative number of individual owls previously detected during nighttime and daytime surveys within a 292.4 km2 area near Willow Creek, Califomia, 1985 to 1987.

based on day surveys were less precise than the Jolly- Seber estimates but more precise than the LRM esti- mates based on night surveys during all years, and more precise than the general removal model of program CAPTURE during 1985 and 1986.

LRM estimates cost 30-64% less than mark-recap- ture estimates, considering the additional time re- quired to mark and capture the birds (Table 1). More time was devoted to mark-recapture activities in 1986 and 1987 to mark trap-shy owls with color bands. Re- sighting owls, rather than physically recapturing owls, would require less effort in subsequent years but at least

30% more effort than only counting owls, because the effort and methods- used to lure owls close enough to identify color bands are similar to those used for cap- turing previously unmarked birds, as was the case in 1985 (Table 1). Most important, however, is the fact that LRM estimates can be generated with 1 yr of survey effort rather than a minimum of 3 yr of data required for Jolly-Seber estimates of population size.

DIsCUssIoN

Despite meeting many of the tested assumptions, the Leslie regression model was a poor estimator of No

212 JAMES P. WARD, JR. ET AL. Ecological Applications Vol. 1, No. 2

1985 210

(11,19) 129

150-

112

90- (23,12) 70 71

63 61 165 ~65 I58 5 54 52

C/) 30 95

O > 1986 o o 100 L1J4

o ~~~~~~~~~~~~89 02. (12,17) { C/)c 80- 176 HQ4) v ' 74 73

70 7

v 60 (7,14) 67 65

cD 60 58 65 59

U)06 { 52

D O 40

LL 3 1987 O ^ 100- 0

z_

(24,12) 80 - 76 79

(11,12) 75 76 67 71 67

60 - 61 66 61

55

40 LRM LRM CAP JOL COUNT

NIGHT DAY MBH SEB

FiG. 3. Annual estimates of the number of territorial Northern Spotted Owls in a 292.4 km2 area near Willow Creek, California. Model acronyms, beneath horizontal axis, are: LRM = Leslie regression (numbers in parentheses are the interval in hours used to group census data and sample size of regression, respectively); CAP-MBH = general removal method of program CAPTURE; JOL-SEB = Jolly-Seber re- duced parameter (D); COUNT = direct count of owls.

when night surveys were used. The LRM will work only if the number of individuals captured declines over time (Krebs 1989). A plot of cumulative number of individual owls detected at night as a function of cumulative hours expended during night surveys was linear for each year, indicating that the "catch" did not decline. The same plot for day surveys was curvilinear, showing fewer owl detections over time.

Capture rates will not decline if population closure is violated or a large enough portion of the studied population is not sampled (Zippin 1956, White et al. 1982). With few exceptions, Spotted Owls were site faithful during the census period (A. B. Franklin et al.,

personal observation). Also, home range size for indi- viduals during this period ranged from 0.6 to 2.3% of the study area (Solis 1983), indicating that few home ranges would be intersected by study area boundaries (e.g., see White et al. 1982:122). Furthermore, study area boundaries were established to minimize move- ments by owls across them (Franklin et al. 1990). Thus, immigration and emigration were considered negligi- ble. By definition, No was not affected by births, and mortality during the census period was assumed to be close to zero based on radio telemetry (Solis 1983) and banding data (A. B. Franklin et al., unpublished data). However, immigration could be "mimicked," and hence closure violated, when individuals are counted more than once. This source of error was more likely at night when owls were mobile (i.e., hunting or moving toward an observer who was calling), making it difficult for observers to infer owl identity.

Moreover, the number of owls counted during all day surveys was >80% of No. Also, considering home range size for paired owls (X = 676.8 ha), the propor- tion of home range overlap between neighboring owls (X = 0.39) and the total amount of habitat potentially used by Spotted Owls within the study area (12 626 ha; Solis 1983, Franklin et al. 1990), we would expect roughly ({12626/[(1 - 0.39) x 676.8]} x 2) or 62 adult and subadult owls in the study population. There- fore, the portion of the population sampled appeared adequate. This led us to conclude that the overestimate produced by night surveys during 1985, and subse- quent fluctuations of No in 1986 and 1987 were likely caused by overcounting.

Regression estimators are notorious for underesti- mating No (Van Etten et al. 1965, Lewis and Farrar 1968, White et al. 1982, Schnute 1983) and, under the assumption of constant capture probability, are infe- rior to maximum likelihood estimators (White et al. 1982:115). In addition, estimation of V[NO] using Eq. 3 may not be valid because regression techniques as- sume independence and equality of error terms (White et al. 1982:116). However, our LRM estimates are comparable to maximum likelihood estimators and, therefore, appear robust when day surveys are used to estimate No and V[NO]. Counts of Spotted Owls during the day provide adequate estimates of No when input into the LRM because, all other considerations equal, it is unlikely that individuals are miscounted when visually observed at daytime roosts.

Relative differences were calculated using a total count (Nm) as a reference value. Because direct counts invariably result in an underestimate of the true pop- ulation size (Caughley 1977), our estimates of differ- ence likely represented minimum differences between No and No. However, evaluation of the relative per- formance among models was valid because the same value was referenced in all calculations. Total counts probably represented 90-95% of the territorial popu- lation. This judgment was based on the amount of

May 1991 ESTIMATING OWL ABUNDANCE 213

suitable habitat present and the number of repeated visits to territories where owls were not detected after the first visit each year (maximum = 21 visits).

Observer characteristics could affect counts of birds (Verner 1985). Surveys in 1985 and 1986 were con- ducted by the same two observers, with 2 and 4 yr of experience in calling and locating owls at the beginning of the 1985 census. All observations were verified by both observers. Despite this level of observer experi- ence, the LRM overestimated Nm using night counts and underestimated Nm using day counts, indicating that at least two census periods may be necessary for reliable estimates. In 1987, one of the original observ- ers was replaced with less experienced observers. How- ever, prior training of new observers coupled with re- sidual expertise of the remaining original observer, prevented any noticeable loss of information.

Our results suggest that the LRM can be used to estimate the number of territorial adult and subadult Spotted Owls present at the beginning of their breeding season, and to estimate a measure of precision. Given the same study design, No can be converted to crude or ecological density (Franklin et al. 1990). Although nocturnal counts should not be used for estimating No with the LRM, night surveys are useful for assessing the general vicinity of owl roost sites. Once owls are visually observed at roosts, they can be counted with little duplication; nests can be located and reproductive parameters sampled with little extra effort (Forsman 1983, Franklin et al. 1990). Thus, day surveys, in com- bination with the LRM, would provide more infor- mation for land management decisions than night sur- veys alone.

We recognize that the use of the LRM with direct count data is not a substitute for mark-recapture ex- periments, as the latter provide information on sur- vivorship, recruitment, and dispersal in addition to estimates of N,. However, the LRM will provide a less expensive, simpler alternative for estimating and mon- itoring abundance of territorial Spotted Owls and other species that behave similarly. Additional testing of this method is recommended, particularly to assess effects of observer experience, geographic variation, and spe- cies variation on model estimates.

ACKNOWLEDGMENTS

We are indebted to T. J. Evans, C. A. Moen, and K E. Young for their devoted assistance in collecting and entering data, to G. M. Allen, C. H. Biles, and B. R. Noon for dis- cussions on mathematical procedure, and to E. D. Forsman and W. S. LaHaye for discussions of field techniques. The computer programming skills of D. Call and R. Waters are also greatly appreciated. D. R. Anderson, J. A. Blakesley, R. J. Cooper, J. L. Ganey, D. G. Hankin, J. Vemer, T. A. Weist, and G. C. White kindly reviewed drafts of this manuscript. This research was funded by California Department of Fish and Game (Federal Aid in Wildlife Restoration Projects W-65-R-3 and 4), U.S. Forest Service, Pacific Southwest Range and Experiment Station (Cooperative Agreement Number

PSW-87-001 ICA), and the McIntire-Stennis Program (HSU Project Number 85).

LITERATURE CITED

Bias, M. 1989. Habitat use by California Spotted Owls in the central Sierra Nevada. Thesis. Humboldt State Uni- versity, Arcata, California, USA.

Burnham, K. P., D. R. Anderson, and J. L. Laake. 1980. Estimation of density from line transect sampling of bio- logical populations. Wildlife Monographs 72.

Carey, A. B., J. A. Reid, and S. P. Horton. 1990. Spotted Owl home range and habitat use in southern Oregon Coast Ranges. Journal of Wildlife Management 54:11-17.

Caughley, G. 1977. Analysis of vertebrate populations. Wi- ley, London, England.

Dawson, W. R., J. D. Ligon, J. R. Murphy, J. P. Myers, D. Simberloff, and J. Verner. 1987. Report of the advisory panel on the Spotted Owl. Condor 89:205-229.

Forsman, E. D. 1983. Methods and materials for locating and studying Spotted Owls. United States Forest Service General Technical Report PNW-162.

Forsman, E. D., E. C. Meslow, and H. M. Wight. 1984. Distribution and biology of the Spotted Owl in Oregon. Wildlife Monographs 87.

Franklin, A. B., J. P. Ward, R. J. Guti6rrez, and G. I. Gould. 1990. Density of Northern Spotted Owls in northwest Cal- ifornia. Journal of Wildlife Management 54:1-10.

Ganey, J. L., and R. P. Balda. 1988. Distribution and habitat use of Mexican Spotted Owls in Arizona. Condor 91:355- 361.

Ganey, J. L., and R. P. Balda. 1989. Home range charac- teristics of Spotted Owls in northern Arizona. Journal of Wildlife Management 53:1159-1165.

Gutierrez, R. J., A. B. Franklin, W. S. LaHaye, V. J. Meretsky, and J. P. Ward. 1985. Juvenile Spotted Owl dispersal in northwestern California: preliminary results. Pages 60-65 in R. J. Guti6rrez and A. B. Carey, editors. Ecology and management of Spotted Owls in the Pacific Northwest. United States Forest Service General Technical Report PNW-185.

Jolly, G.M. 1965. Explicit estimates from capture-recapture data with both death and immigration-stochastic model. Biometrika 52:225-247.

Krebs, C. J. 1989. Ecological methodology. Harper & Row, New York, New York, USA.

Kulchler, A. W. 1977. The map of the natural vegetation of California. Pages 909-938 in M. Barbor and J. Majors, editors. Terrestrial vegetation of California. John Wiley & Sons, New York, New York, USA.

LaHaye, W. S. 1988. Nest site selection and nesting habitat of the Northern Spotted Owl (Strix occidentalis caurina) in northwestem California. Thesis. Humboldt State Univer- sity, Arcata, California, USA.

Laymon, S. 1988. Ecology of the Spotted Owl in the central Sierra Nevada, California. Dissertation. University of Cal- ifornia, Berkeley, California, USA.

Leslie, P. H., and D. H. S. Davis. 1939. An attempt to determine the absolute number of rats on a given area. Journal of Animal Ecology 8:94-113.

Lewis, J. C., and J. W. Farrar. 1968. An attempt to use the Leslie census method on deer. Journal of Wildlife Man- agement 32:760-764.

Norton, B. 1988. Commodity, amenity and morality, the limits of quantification in valuing biodiversity. Pages 200- 205 in E. 0. Wilson and F. M. Peter, editors. Biodiversity. National Academy Press, Washington, D.C., USA.

Otis, D. L., K. P. Burnham, G. C. White, and D. R. Anderson. 1978. Statistical inference from capture data on closed animal populations. Wildlife Monographs 62.

Peterson, R. 0. 1988. The pit or the pendulum: issues in

214 JAMES P. WARD, JR. ET AL. Ecological Applications VnV I Nn- 2

large carnivore management in natural ecosystems. Pages 105-117 in J. K. Agee and D. R. Johnson, editors. Eco- system management for parks and wilderness. University of Washington Institute of Forest Resources Contribution Number 65. University of Washington Press, Seattle, Washington, USA.

Reynolds, R. T., J. M. Scott, and R. A. Nussbaum. 1980. A variable circular plot method for estimating bird num- bers. Condor 82:309-313.

Salwasser, H. 1988. Managing ecosystems for viable pop- ulations of vertebrates: a focus for biodiversity. Pages 87- 104 in J. K. Agee and D. R. Johnson, editors. Ecosystem mangement for parks and wilderness. University of Wash- ington Institute of Forest Resources Contribution Number 65. University of Washington Press, Seattle, Washington, USA.

Salwasser, H., I. D. Luman, and D. Duff. 1982. Integrating wildlife and fish into public land forest management. Pro- ceedings of the Western Association of Fish and Wildlife Agencies 62:293-299.

Schnute, J. 1983. A new approach to estimating populations by the removal method. Canadian Journal of Fisheries and Aquatic Sciences 40:2153-2169.

Seber, G. A. F. 1965. A note on the multiple recapture census. Biometrika 52:249-259.

1982. The estimation of animal abundance and re- lated parameters. Second edition. Griffin, London, En- gland.

1986. A review of estimating animal abundance. Biometrics 42:276-292.

Sisco, C. L. 1990. Seasonal home range and habitat ecology of Spotted Owls in northwestern California. Thesis. Hum- boldt State University, Arcata, California, USA.

Solis, D. M. 1983. Summer habitat ecology of Spotted Owls in northwestern California. Thesis. Humboldt State Uni- versity, Arcata, California, USA.

Thomas, J. W., L. F. Ruggiero, R. W. Mannan, J. W. Schoen, and R. A. Lancia. 1988. Management and conservation of old-growth forests in the United States. Wildlife Society Bulletin 16:252-262.

United States Forest Service. 1988. Final supplement to the environmental impact statement for an amendment to the Pacific Northwest regional guide. United States Forest Ser- vice, Portland, Oregon, USA.

Van Etten, R. C., D. F. Switzenberg, and L. Eberhardt. 1965. Controlled deer hunting in a square-mile enclosure. Journal of Wildlife Management 29:59-73.

Verner, J. 1985. Assessment of counting techniques. Pages 247-302 in R. F. Johnston, editor. Current ornithology. Volume 2. Plenum, New York, New York, USA.

White, G. C., D. R. Anderson, K. P. Burnham, and D. L. Otis. 1982. Capture-recapture and removal methods for sampling closed populations. Los Alamos National Labo- ratory Publication LA-8787-NERP.

Zar, J. H. 1984. Biostatisticalanalysis. Second edition. Pren- tice-Hall, Englewood Cliffs, New Jersey, USA.

Zippin, C. 1956. An evaluation of the removal method of estimating animal populations. Biometrics 12:163-169.

APPENDIX Example calculations of variables input into the Leslie re-

gression model, given a 23-h grouping of the cumulative num- ber of Spotted Owls detected and the cumulative number of hours of effort devoted during 1985 daytime surveys in a 292.4 km2 area near Willow Creek, California.

Per sampling Cumulative data interval Regression data

Owls Owls Survey Cum. detect- prev. inter- no. No. No. ed per de- val No. survey owls hours hour tected (1) owls hours (n) () (Y) (X) 1 2 0.83

4 2.57 6 5.17 8 7.23

10 8.40 12 10.03 13 16.70 15 21.20 15 21.20 0.708 0

2 16 24.77 18 27.50 20 29.03 21 35.03 22 35.83 24 35.93 25 38.80 26 45.93 11 24.73 0.445 15

3 27 48.10 29 50.90 31 55.77 32 65.37 6 19.44 0.309 26

4 33 70.73 34 74.27 35 78.03 3 12.66 0.237 32

5 36 101.77 38 113.70 3 35.67 0.084 35

6 39 121.90 41 125.67 3 11.97 0.251 38

7 43 142.80 2 17.13 0.117 41 8 45 180.83 2 38.03 0.052 43 9 47 188.43

48 200.63 49 201.23 4 20.40 0.196 45

10 50 207.27 51 210.73 52 216.90 3 15.67 0.191 49

11 54 239.50 55 250.10 56 250.27 4 33.37 0.120 52

12 57 266.27 1 16.00 0.063 56