OUP user menu

Estimating Bobcat Population Sizes and Densities in a Fragmented Urban Landscape Using Noninvasive Capture–Recapture Sampling

Emily W. Ruell, Seth P. D. Riley, Marlis R. Douglas, John P. Pollinger, Kevin R. Crooks
DOI: http://dx.doi.org/10.1644/07-MAMM-A-249.1 129-135 First published online: 18 February 2009


Bobcats (Lynx rufus) are valuable indicators of connectivity in the highly fragmented landscape of coastal southern California, yet their population sizes and densities are largely unknown. Using noninvasive scat sampling in a capture–recapture framework, we estimated population sizes for 2 similar areas of natural habitat with differing levels of isolation by human development in Santa Monica Mountains National Recreation Area, California. We used scat transects with geographic information system land-use layers and home-range sizes of bobcats to estimate effective sampling area and population densities. Estimates of population size in the study area connected to a much larger habitat area (26–31 individuals) were similar to estimates for the area that was completely surrounded by development (25–28 individuals). Bobcat densities for the 2 study areas also were similar (ranging from 0.25 to 0.42 bobcat/km2) and likely represent recent population declines because of notoedric mange likely interacting with toxicants. These methods proved effective despite particularly low densities of bobcats and may be especially useful when study areas are geographically isolated, reducing the uncertainty in size of the sampling area.

Key words
  • bobcat
  • capture–recapture
  • effective sampling area
  • Lynx rufus
  • population density
  • population size
  • scat

Mammalian carnivores can be particularly vulnerable to extinction in fragmented habitat because of their low population densities, relatively large ranges, and direct persecution by humans (Crooks 2002; Noss et al. 1996; Soulé and Terborgh 1999). Thus, mammalian carnivores are focal species for large-scale conservation planning (Crooks 2002). Estimates of population size and density are important for monitoring population trends, predicting the long-term persistence of small populations, and identifying populations with dangerously low numbers (Creel et al. 2003; Prugh et al. 2005). However, obtaining reliable population estimates for mammalian carnivores is often difficult and expensive using conventional trapping techniques, especially for small populations (Creel et al. 2003; Prugh et al. 2005). Here we show, in a capture–recapture sampling framework, that noninvasive genetic sampling allows for valid estimation of population size and density even for small populations (Creel et al. 2003; Miller et al. 2005; Soisalo and Cavalcanti 2006).

Because of the high level of species endemism in coastal southern California, massive habitat fragmentation in the region has created a global hotspot of endangerment and extinction (Dobson et al. 1997). Bobcats (Lynx rufus) are sensitive to habitat fragmentation and are valuable indicators of connectivity in this highly fragmented landscape (Crooks 2002; Riley et al. 2003). However, as is often the case with mammalian carnivores, population sizes and densities of bobcats are largely unknown and are likely affected both by habitat loss and fragmentation and by anthropogenic sources such as vehicle collisions and toxicants in this region (Riley et al. 2003, 2007). Accurate population sizes and densities are particularly difficult to determine when populations decline and sampling requires greater effort. In Santa Monica Mountains National Recreation Area in coastal southern California, annual survival rates of bobcats dropped from 0.77 (5-year average from 1997 to 2001) to 0.28 in 2003, likely due to interactions between anticoagulant exposure and a notoedric mange epizootic that began in 2002. The number of bobcat scats collected in these areas along established scat transects also showed a significant decrease in bobcat presence beginning in 2002 through spring 2004 (Riley et al. 2007).

Our objectives were 2-fold: to construct and evaluate population and density estimation methods for mammalian carnivores in fragmented landscapes and in multicarnivore systems, and to determine if these methods were effective for small bobcat populations of conservation concern in urban southern California. We used noninvasive scat surveys and a DNA-based capture–recapture sampling framework to estimate population sizes of urban bobcats in sites that had large numbers of scats of nontarget species. We used and compared estimates from 2 closed-population heterogeneity estimators in program CAPTURE (Otis et al. 1978) and the capwire estimator, which was created specifically for non-invasive genetic sampling (Miller et al. 2005). We then used the scat transects with geographic information systems land-use layers and bobcat home-range sizes to estimate effective sampling areas and hence population densities. We compared bobcat population sizes and densities in 2 similar-sized areas of natural habitat with differing levels of isolation in the Santa Monica Mountains National Recreation Area; 1 habitat area (Simi Hills) was fully surrounded by human development and the other (Topanga) was partially connected to larger natural areas. We expected greater bobcat densities with increased isolation, because roadways and urban development may limit dispersal from isolated habitat fragments and bobcat home ranges have greater overlap and are smaller in size along urban and roadway boundaries (Riley et al. 2003, 2006). Population-size and density estimation using noninvasive sampling proved effective even when bobcat densities were likely much lower than normal due to recent population declines and despite the occurrence of large numbers of scats of nontarget species.

Materials and Methods

Study areas.—Scat samples were collected in summer 2004 from 2 study areas located within Santa Monica Mountains National Recreation Area, a national park, north of Los Angeles, California (Fig. 1). The study area partially surrounded by human development, called Topanga, was located within the eastern portion of the Santa Monica Mountains. Topanga was separated from the rest of the Santa Monica Mountains by Topanga Canyon Boulevard to the west, was bordered by the Pacific Coast Highway (State Route 1), Sunset Boulevard, and residential development to the south, United States Highway 101 and residential development to the north, and United States Interstate 405 and residential development to the east. The study area fully surrounded by development, called Simi Hills, was located south of Simi Valley, separated from habitat to the north by State Route 118, separated from the Santa Monica Mountains to the south by United States Highway 101, and surrounded by residential development and secondary roads. Vegetation for both study areas consisted primarily of coastal sage scrub habitat interspersed with patches of chaparral, oak woodlands, nonnative grasslands, and riparian woodlands. This mosaic of habitat in coastal southern California supports bobcats, pumas (Puma concolor), coyotes (Canis latrans), gray foxes (Urocyon einereoargenteus), and several other native and nonnative mesopredator species (Crooks 2002).

Fig. 1

The Topanga and Simi Hills study areas in Santa Monica Mountains National Recreation Area, California, in summer 2004. Scat transects also are shown to demonstrate scat transect coverage of study areas.

Field sampling.—We systematically conducted surveys for scats along roads, trails, and dry creek beds that thoroughly covered both study areas. Nonrandom sampling transects were necessary to obtain adequately high capture probabilities (Mowat and Strobeck 2000; Woods et al. 1999); bobcats and other carnivores frequently defecate along roads and trails, which are regular paths of movement (Kohn et al. 1999; MacDonald 1980), and random scat transects placed off roads or trails would have had low probabilities of encountering scats (E. W. Ruell, pers. obs.). Unbiased population estimation requires that every individual had a reasonable chance of being sampled (White et al. 1982). Because home ranges of female bobcats (55 km2 ± 1.44 SD, n = 19, in nearby fragments) are smaller than those of males and often overlap (Riley et al. 2003), we divided study areas into 1-km2 cells to avoid missing bobcat territories and made a concerted effort to search for scats along carnivore movement routes within each cell (Mowat and Strobeck 2000; Soisalo and Cavalcanti 2006). Each study area contained approximately 64 km of total transect (Figs. 1 and 2).

Fig. 2

The Simi Hills effective sampling area size and 95% CI, estimated by buffering scat transects with the radius and 95% CI of the average home range of a male. Only areas of natural habitat are included in the estimated effective sampling area.

Before sampling, we cleared all scats from sampling routes so that scats collected in subsequent sampling occasions were of known age. Scat transects were then sampled once every 4 days for 4 consecutive sampling occasions over a total sampling period of 16 days. This sampling regime represented a balance of sampling intensity with the risk of violating the geographic closure assumption, which can result in serious bias in population estimates, over longer sampling periods. Collecting scat samples every 4 days also helped to prevent DNA degradation and ensure genotyping success, while allowing adequate opportunity for individuals to deposit new scat samples. We traveled the identical route on scat transects during each sampling occasion and collected all large-carnivore scats encountered because of the overlap in scat morphology among carnivore species within the study areas. Scat sampling of mammalian carnivores was noninvasive, was conducted following guidelines of the American Society of Mammalogists (Gannon et al. 2007), and was approved by the Colorado State University Animal Care and Use Committee (03-187A-03).

Scat samples were collected using inverted resealable freezer bags, which were then reversed and sealed. Samples were sorted into potential-felid and nonfelid categories based on scat dimensions, composition, and tracks found nearby. We were conservative in our identification and included ambiguous scats in the felid category to avoid missing bobcat scats. This culling method was effective in limiting the number of nonbobcat samples processed while minimizing the chance of missing bobcat samples (Ruell and Crooks 2007). We added 1.905-cm silica gel beads (Sigma Aldrich, Inc., St. Louis, Missouri) to each bag to desiccate samples in an approximate 5:1 (silica: scat) weight ratio (Wasser et al. 1997).

Genotyping.—Genomic DNA was extracted from all potential felid scat samples using the QIAamp DNA Stool Mini Kit (Qiagen, Inc., Valencia, California) following the manufacturer's instructions. To maximize DNA yield, extractions were performed within 12 months of their collection (Roon et al. 2003). Samples were genetically identified to species using the 16S mitochondrial DNA (mtDNA) method (Mills et al. 2000b). Those samples that failed to generate mtDNA restriction profiles were considered to be of poor quality and culled from the data set. Bobcat samples were identified to individual using genotypes from 4 microsatellite loci (FCA026, FCA045, FCA077, and FCA132—Menotti-Raymond et al. 1999; Ruell and Crooks 2007). Individual genotypes were used as unique “tags” identifying each sampled individual in capture-recapture sampling. An advantage of this approach is that there can be no mark loss. However, genotyping errors can result in assigning incorrect genotypes to samples and consequently adding false individuals to the list of marked individuals. False individuals cause an erroneously large list of marked individuals and an underestimated probability of an animal being encountered at least once, which leads to an overestimation of population size by a capture–recapture estimator (Lukacs and Burnham 2005). Therefore, for capture–recapture studies, care must be taken to reduce genotyping error to a negligible rate (Taberlet et al. 1996).

The overall expected genotyping error rate for these 4 microsatellite loci was 0.0004% and estimated from known samples using a multiple tubes approach (Ruell and Crooks 2007). These 4 loci could differentiate individuals with confidence (P(ID)sib = 0.02 and P(ID)obs = 0—Ruell and Crooks 2007; Waits et al. 2001). We were less concerned about failing to discriminate between individuals (termed the “shadow effect” by Mills et al. [2000a]) than about adding ghost individuals due to genotyping error, because the shadow effect causes little bias in population estimation compared to even low levels of genotyping error (Mills et al. 2000a; Waits and Leberg 2000). Using a multiple tubes approach with 3–6 replicates per locus, only samples that provided genotypes at all 4 loci were analyzed further. Scat sample genotypes were matched using the Excel Microsatellite Toolkit (Park 2001). Closely matched genotypes with only 1 or 2 allele differences were rechecked for scoring errors, particularly when allelic dropout at 1 locus could have caused the difference. When in doubt, these samples were lumped as the same individual to avoid inflating population estimates with false individuals (Prugh et al. 2005; Waits and Leberg 2000). Closely matching samples of poorer quality (based on band intensity and ease of scoring) were culled from the data set, particularly if they were unique genotypes that did not match any other samples. We then identified the sex of individuals using the zinc-finger region test (Pilgrim et al. 2005).

Population-size estimation.—Because there was substantial heterogeneity in individual capture probabilities, and we cannot precisely determine the sampling intensity (proportion of population sampled), population sizes (N) were estimated using 3 closed-population heterogeneity estimators: Mh-jackknife (Burnham and Overton 1979) and Mh-Chao (Chao 1988) in program CAPTURE, and capwire, using the TIRM model (Miller et al. 2005). For small population sizes, Mh-jackknife is an efficient estimator that performs well with substantial heterogeneity and is robust to some variation in capture probability over time with high sampling intensity (Manning et al. 1995; Otis et al. 1978; White et al. 1982). However, when sampling intensity is lower, Mh-Chao and capwire display less (although slightly positive) bias and have greater coverage than Mh-jackknife, which displays negative bias and has poor coverage unless sampling intensity is high (Miller et al. 2005). Mh-Chao displays the least bias with low sampling intensity, but capwire provides much narrower confidence intervals (CIs) with good coverage (Miller et al. 2005).

Capture histories for Mh-jackknife and Mh-Chao were constructed in 2 ways. For the 1st, multiple captures during a sampling occasion were pooled and recorded as a single capture for that occasion for each of the 4 sampling occasions. Miller et al. (2005) found that the Chao estimator performed slightly better with the maximum number of captures than with 4 discrete sampling occasions (i.e., less bias, smaller CI width, and better coverage). Given that individuals are effectively sampled with replacement during each sampling occasion (any individual can deposit multiple independent scat samples during a 4-day sampling occasion), alternatively the number of sampling occasions was the maximum number of captures per individual over the total 16-day sampling period (Miller et al. 2005). Similarly, capwire used total captures per individual over the total 16-day sampling period (Miller et al. 2005). We assumed population closure, both demographic (i.e., no births, deaths, permanent immigration, or emigration) and geographic (i.e., no movement on and off a study area between capture occasions), during the 16-day sampling period. We combined both sexes in population-size estimates, because sample sizes for each sex would have been too low to obtain reasonable estimates.

Density estimation.—Bobcat densities (D) were estimated using population estimates from Mh-jackknife (maximum number of captures) and capwire, because they provided the smallest CI widths and represent estimates that may be subject to negative (Mh-jackknife) and positive bias (capwire), given that the sampling intensity was unknown (Miller et al. 2005). We estimated D in each study area by dividing TV by effective sampling area size, which we calculated in 2 ways. Effective sampling area sizes were estimated using geographic information systems by buffering scat transects with the radius (Arad: 1.009 km) and then the diameter (Adiam: 2.018 km) of the average home range of male bobcats (3.21 km2Riley et al. 2003) and measuring the area of natural habitat included in these buffers (Fig. 2). Average home-range size of males was used because ranges of males are larger than ranges of females, and we chose to potentially underestimate rather than overestimate bobcat densities because these populations are of conservation concern. We only included natural habitat because bobcats in this area primarily use and rely on native vegetative cover and avoid urban areas (Riley et al. 2003). Ninety-five percent CIs for Arad and Adiam were estimated conservatively from the standard error of the average home range of males and were in turn used with CI of N to calculate CI of density estimates.


Field sampling and genotyping.—In Topanga we collected 421 mammalian carnivore scat samples during the 4 sampling occasions; of these, 27.8% (n = 121) of samples were labeled “maybe felid.” In Simi Hills we collected 498 scat samples during the 4 sampling occasions, from which 28.3% (n = 141) of samples were categorized as “maybe felid.” Using mtDNA restriction profiles, we were able to identify to species 77% of the “maybe felid” samples in Topanga and 89% in Simi Hills. Of the samples that identified to species, 45% (n = 42) in Topanga and 41% (n = 51) in Simi Hills were bobcat scats. From the identified bobcat samples, 88% (n = 37) in Topanga and 90% (n = 46) in Simi Hills could be genotyped at all 4 microsatellite loci using the multiple tubes approach replication criteria. In 6 cases (1 Topanga and 5 Simi Hills) we recorded a 1-allele genotype for a particular locus with the other allele scored as missing data.

Population-size estimation.—Among the bobcat scat samples that were genotyped, we identified 19 unique genotypes in Topanga and 19 in Simi Hills. There was considerable variation between individuals in capture histories. We sampled multiple scats from 53% (n = 10) of the individuals sampled in Topanga (range 2–5 samples per individual) and 53% (n = 10) of the individuals sampled in Simi Hills (2–11 samples per individual). The remaining sampled individuals in Topanga and Simi Hills were only captured once. We sampled 7 males and 12 females in Topanga and 8 males, 10 females, and 1 unknown in Simi Hills.

Population estimates were generally similar for all estimation methods (Table 1). The performance of Mh-jackknife and Mh-Chao when using the maximum number of captures instead of 4 sampling occasions was inconsistent across study areas. In Topanga, using the maximum number of captures provided identical or slightly larger estimates of N and larger CI widths than using 4 sampling occasions. In Simi Hills, using the maximum number of captures provided slightly smaller estimates of TV and similar CI widths relative to using 4 sampling occasions. Estimates of N from capwire were slightly higher than Mh-jackknife and Mh-Chao in Topanga and very similar in Simi Hills. For all estimates, Mh-jackknife and capwire had considerably smaller CIs than Mh-Chao. Given the range of N, we sampled scat from 61–73% of the population in Topanga and 66–76% of the population in Simi Hills.

View this table:
Table 1

Population and density estimates for bobcats (Lynx rufus) for the Topanga and Simi Hills study areas in Santa Monica Mountains National Recreation Area, California, in summer 2004.

Study areas
Topangaa (n= 19)Simi Hillsa (n= 19)
Population size estimatorsN95% CICI widthN95% CICI width
Mh-jackknife (4 sampling occasions)2622–39172823–4320
Mh-jackknife (maximum no. captures)2622–43212521–4120
Mh-Chao (4 sampling occasions)2621–48272922–5836
Mh-Chao (maximum no. captures)2922–64422721–5433
Densities using Mh-jackknife (maximum no. occasions)D95% CICI widthD95% CICI width
Density/km2 from Arad0.350.26–0.740.480.330.24–0.720.48
Density/km from Adiam0.270.20–0.540.340.250.18–0.500.32
Densities using capwireD95% CICI widthD95% CICI width
Density/km2 from Arad0.420.22–0.760.540.370.22–0.680.46
Density/km2 from i4diam0.320.18–0.550.370.280.16–0.480.32
  • a Estimated study area sizes ranged from 75 to 96 km2 for Topanga and 76 to 100 km2 for Simi Hills.

Density estimation.—Effective sampling area sizes for Topanga were Arad = 74.6 km2 (95% CI = 57.9–85.4) and Adiam = 96.3 km2 (95% CI = 80.3–108.3) and for Simi Hills they were Arad = 75.8 km2 (95% CI = 57.3–87.6) and Adiam = 100.3 km2 (95% CI = 81.9–116.6). Bobcat density estimates were almost identical for Topanga and Simi Hills when using N from either Mh-jackknife (maximum number of captures) or capwire, although slightly larger in Topanga when using capwire (Table 1). Density estimates were greater when using Arad than Adiam, but not significantly so, because the CIs overlap (Table 1).


Scat surveys are a relatively easy and efficient method to noninvasively sample bobcats even when their population numbers and densities are likely low. By only testing scats that could be felid based on size and composition, we greatly reduced the number of nontarget samples processed. Although only approximately 10% of the mammalian carnivore scats collected in both study areas proved to be bobcat, we only had to genetically test approximately 29% of the scat samples encountered to find them.

The validity of estimates from closed-population estimators relies on demographic and geographic closure during sampling. The assumption of demographic closure was likely not violated in our study because of the short sampling duration of 16 days. The assumption of geographic closure also was reasonable given that human development surrounded both study areas. Bobcats largely avoid human development and crossing major roadways, and roadways often become home-range boundaries for bobcats that live adjacent to them in southern California (Riley et al. 2003, 2006). In the event of geographic-closure violations, overestimation of population sizes is minimized when sampling duration is short and home-range size is small relative to study area size, as in this study (Mowat and Strobeck 2000; White et al. 1982).

Heterogeneity in capture probability is a long-standing issue for population estimators and can bias population estimates low (Otis et al. 1978; White et al. 1982). Our capture histories indicated that there was substantial individual heterogeneity in capture probability with scat sampling. Heterogeneity could have been caused by differences in defecation rates, road and trail use, and territory marking between individuals, sexes, or age groups (Bellemain et al. 2005; Eggert et al. 2003). Some individuals may have consistently deposited poorer-quality samples than others because individual animals may leave varying amounts of DNA in scats, resulting in nonrandom culling of poor-quality samples (Alessandrini et al. 2003; Lukacs and Burnham 2005). With regard to sex bias, we would expect higher densities of females because they have smaller home ranges and philopatry with close relatives (Janecka et al. 2006; Riley et al. 2003). In accordance with this expectation, we did sample a greater number of females than males in both study areas. However, a larger proportion of males sampled had multiple “captures” than females sampled, likely due to larger male ranges containing more scat transects within them; this may have caused additional heterogeneity in capture probability among individuals.

Although heterogeneity was evident, for small population sizes choosing the best heterogeneity estimator is not straightforward when the sampling intensity is unknown (Miller et al. 2005). The Mh-jackknife was likely an appropriate estimator given that it performs better than other estimators during the 1st few sampling occasions for small populations (Frantz et al. 2003; Manning et al. 1995; Otis et al. 1978; White et al. 1982). Further, managers may prefer to underestimate population sizes using Mh-jackknife than to overestimate them using capwire when bobcat populations are of conservation concern.

Density estimation is difficult when using capture–recapture methods and the true size of the sampling area is unknown. Accurate density estimates are dependent on the level of sampling biases and on the accuracy of the estimate of effective sampling area size. We estimated effective sampling areas by buffering scat transects with average home-range size based on telemetry data, because home-range size is a better estimate of individual movement than the commonly used mean maximum distance moved methods and therefore is less likely to overestimate bobcat densities (Parmenter 2003; Soisalo and Cavalcanti 2006). Density estimates when using Arad and Adiam were similar because scat transects thoroughly covered each study area, only natural habitat was included in area estimation, and both study areas were largely surrounded by development.

Contrary to our expectations, we found little evidence that habitat isolation affects bobcat densities in our study areas, given that density estimates and associated CIs were almost identical in Topanga and Simi Hills despite different levels of surrounding human development. Potentially, the greater degree of development surrounding Simi Hills had not detectably inflated bobcat density, either because a substantial proportion of bobcat home ranges were in the interior region and did not abut development (Riley et al. 2006) or because current levels of habitat isolation might not have sufficiently limited dispersal out of Simi Hills. Alternately, high mortality caused by the notoedric mange epizootic that began in 2002 may have reduced bobcat densities within both Topanga and Simi Hills and thus masked the impact of habitat isolation on density by summer 2004.

Our density estimates in Topanga and Simi Hills (0.25–0.42 individuals/km2) were considerably lower than previous estimates of bobcat densities in coastal southern California. Densities of 1.27–1.53 individuals/km2 were estimated from a small (6.7–km2) study area within the Cleveland National Forest, south of Los Angeles, although these densities were not derived from population estimates but rather calculated from the fraction of time that radiocollared bobcats occupied the study area, which may not be representative of all bobcat habitat in coastal southern California (Lembeck 1986). However, a radiotelemetry study conducted before the mange epizootic within smaller fragments (3.15 and 4.45 km2) in Santa Monica Mountains National Recreation Area near our study areas found that at least 3 females were using each fragment and male bobcats were using multiple fragments (Riley et al. 2003), suggesting higher densities (≥0.6 individuals/km2) than we estimated in Topanga or Simi Hills in 2004. These smaller fragments could have contained unusually high bobcat densities because of increased overlap and reduced size in home ranges and core areas along the boundaries of surrounding urban development and roadways (Riley et al. 2003, 2006).

More likely, however, the low population densities we recorded in both Topanga and Simi Hills were indicative of the recent population declines resulting from the notoedric mange epizootic, likely interacting with anticoagulant exposure, that began in 2002 (Riley et al. 2007). After the epizootic, densities within the 2 smaller fragments in Santa Monica Mountains National Recreation Area also had declined (from >0.6 to ∼0.2 individuals/km2—S. P. Riley, pers. obs.). Even though detection of bobcat scat samples was unusually low from 2002 to 2004 (Riley et al. 2007), our methods worked effectively and efficiently to estimate population sizes and densities with minimal disturbance. This suggests that noninvasive scat sampling in a capture–recapture framework is a promising tool for population estimation even when populations are smaller than normal and more difficult to sample using traditional methods. This is particularly important when monitoring populations of conservation concern, such as mammalian carnivores that may serve as focal species for conservation planning, as they become increasingly threatened and sensitive to disturbance.


Thanks to C. E. Lee at the University of Wisconsin, Madison, R. K. Wayne at the University of California, Los Angeles, and M. E. Douglas and M. F. Antolin at Colorado State University for laboratory facilities, laboratory expertise, and valuable guidance on this project and manuscript. E. York and M. Cegelski provided field assistance. C. Talbert and M. Ackerman assisted with laboratory analyses. Thanks to P. M. Lukacs for assistance with population estimation analyses. J. A. Tracey and L. Lee assisted with geographic information system study area size estimation. Thanks to C. Handelsman for feedback and support. The Oscar and Isabel Anderson Graduate Scholarship, Douglas L. Gilbert Memorial Scholarship, Colorado State University College of Natural Resources Need-Based Scholarship, Theodore Roosevelt Memorial Fund Grant, and the Rocky Mountain Goats Foundation's Bill Burtness Fellowship supported this work.


  • Associate Editor was John A. Yunger.

Literature Cited

View Abstract