There are no robust estimates of snow leopard population size range wide or at national level (Jackson et al. 2010) but such estimates are urgently needed in order to detect population trends and inform conservation action.
Throughout the 1970s and 1980s, indirect evidence in the form of field sign (pugmarks, feces, scrapes and other scent markings) served as the primary means for confirming snow leopard presence and mapping its distribution. During the 1990s, systematic recording of field sign along transects under the SLIMS protocol (Jackson and Hunter 1997) was also used in crudely assessing snow leopard populations. SLIMS is a valuable tool for “presence-absence” surveys, and may have some use in assessing relative abundance if (1) restricted to scrapes or well-defined pugmarks, (2) aging criteria can be consistently applied and (3) feces are verified through genetic analysis. However, this technique if widely prone to error and/or bias related to observer-based differences in judging the age of signs, effects of rainfall and snowfall that influence sign detectability by damaging, obliterating or covering sign, and disturbance from livestock or humans (similar results) as well as observer skills at selecting and conducting surveys.
Compared to other large cats like the tiger (Panthera tigris), jaguar (P. onca), leopard (P. pardus) and puma (Puma concolor), very few snow leopards have ever been radio-collared, due in large part to the difficult terrain and their land home ranges which together present significant difficulties in obtaining enough fixes. While radio-telemetry may offer precise estimates of home range, such studies are generally too costly or time consuming to apply except in a few cases.
In the last few years, far more robust, sophisticated and reliable techniques have become available in the form of remote camera trapping, non-invasive genetics based on laboratory analysis of feces, and GPS/satellite collars that allow individual animals to be tracked across time and space. These technological innovations have been accompanied by advances in statistical computing that greatly increase the power of analyses.
A major limitation of all monitoring initiatives has been the relatively small study areas (a few hundred km²), making extrapolations to larger landscapes difficult or problematic unless based on reliable maps of habitat suitability. Much of this is believed to be a direct consequence of logistic challenges of working in snow leopard habitats, and the limited resources at disposal (Janečka et al. 2011).
Modern camera-trapping of wild felids, using individual photo identification coupled with use of the Capture-Recapture (CR) algorithm, began on tigers in the 1990s, Since then, this technique has been applied to many species, including snow leopards. The advent of relatively cheap, reliable, and much more sophisticated digital cameras has led to camera trap surveys being applied widely across all countries of snow leopard range.
Important elements to consider in using camera trapping to generate population density estimates include individual identification, confidence in avoiding false IDs, ensuring adequate sample size and capture probability, proper camera location and spacing, size of study area and ad hoc density estimation from the calculation of an effective trapping area (Foster and Harmsen 2011). Jackson et al. (2006) recommended achieving capture probabilities of about 0.30. A common constraint, imposed by a combination of access and logistics involved in periodically moving traps to new sites, and the availability of cameras, is the size of area that can be surveyed within a time frame that minimizes the likelihood of violating key underlying assumptions of CMR and yet that permits sufficient captures and recaptures upon which to base abundance estimates.
Typically, sampling should be done over relatively large areas sufficient to support a population of at least 20 adults or more, although this may not prove feasible for snow leopards for several reasons: large home range, low densities and rugged terrain with limited access significantly limit the size of area which can be surveyed without danger of violating key assumptions related to population closure (immigration and emigration) and births/deaths (ca. up to 60-80 days). Large samples with adequate numbers of recaptures are necessary to ensure reasonably narrow confidence limits around the population estimates in order to detect changes over time robustly and accurately: thus, this goal may only be possible under special circumstances. Also, continued long-term abundance monitoring aimed at generating valuable additional data on life history parameters like birth rates, mortality, and migration are needed, but are typically costly since such studies hinge upon satellite telemetry and are labor intensive with respect to monitoring trail cameras, changing batteries and retrieving and processing images. The huge volume of photos generated, often including numerous false images, in nearly all surveys presents its own challenge, and represents a research topic for examining the utility of computer-assisted image recognition algorithms.
Another important issue in abundance estimation involves translating abundance values into density estimates. Conventional buffer-based approaches such as Mean Minimum Distance Moved (MMDM), ½ MMDM, and radius of home range (if available) have been used to estimate the effective trapping area and to calculate density (e.g. Karanth & Chundawat, 2004). These approaches have been criticized for underestimating the area and inflating density estimates (Efford 2011; Obbard et al 2010; Sharma et al. 2010; Soisalo & Cavalcanti 2006), with researches recommending combining telemetry and camera-trap studies.
A more feasible approach involves use of Spatially Explicit Capture Recapture (SECR) methods (Efford 2004; Royle & Young 2008; Linkie et al. 2010). These generate more realistic, geographically aligned density estimates, and are also compatible with information from other sources (e.g. DNA, occupancy surveys) as well as allowing for covariate analysis aimed at identifying key environmental correlates of high capture rates.
SECR models do not require estimation of the effective survey area, but instead estimate density directly using maximum likelihood (Borchers and Efford 2008) or hierarchical Bayesian (Royle and Young 2008) approaches. Such models use spatial information (i.e. capture locations) in conjunction with the capture probability of different individuals to estimate the number of sampled individuals likely having activity centers within the sampling area. Additionally, these models are robust to individual capture heterogeneity, and do not require strict geographic closure, a problematic assumption with many other CR models. SECR models have been used in camera-trapping studies (e.g. Royle et al. 2009; Sollmann et al. 2013) as well as in non-invasive genetic studies using hair snares (Obbard et al. 2010; Kery et al. 2011) as described below.
Photo capture–recapture density estimation may be of little value when population sizes are extremely low and individuals are elusive and highly dispersed; here simple photo capture rates may provide more reliable results as an index of relative abundance than capture–recapture density estimation. McCarthy et al (2008) considered photo-rates to provide a legitimate index of snow leopard abundance in their study area based on similarity with genetic individual identification. They noted that this relative index may be suitable when true densities are not needed, although this metric is also affected by multiple factors, from camera placement to spacing densities with regard to prevailing travel patterns of the resident or dispersing snow leopard cohort.
14.3. Fecal genetics
A more promising method of monitoring of snow leopard populations may involve fecal DNA analysis, especially given the generally lower costs and ability to cover large areas of habitat within a relatively short time-frame. However, important issues relating to aging and the relative longevity of genetic material in high elevation arid climates need to be resolved and standard procedures for collecting and analyzing feces put in place.
Schwartz et al. (2006) highlighted the use of genetic monitoring as a tool for conservation and management. Non-invasive fecal genetics is especially applicable to monitoring rare and elusive carnivores like the snow leopard, and may serve as a viable and more cost-effective alternative to intensive techniques such as fitting radio/GPS collars to free-roaming animals. The tendency of snow leopards to use well-defined travel routes (Ahlborn and Jackson 1986; Jackson and Ahlborn 1989) facilitates the collection of feces, especially in mountains with well-defined ridges and drainages. The dry, cold climate helps retard bacterial action and breakdown of nucleic acids that otherwise degrades genetic material, so that amplification success rates tend to be higher than for felids inhabiting tropical climes.
Among the many applications aiding in the conservation of elusive cats are species identification to establish species distribution; habitat requirements and diet, determination of the sex of individuals within a population; and identification of individuals within a population, allowing for estimates of population abundance and breeding rates (Rodgers and Janečka 2012). Fecal DNA enables researchers to investigate evolutionary, population, landscape or conservation genetic hypotheses like the rate of gene flow, metapopulation dynamics, habitat connectivity, genetic diversity and phylogeography. Additionally, species ID from scat can be used to identify those individuals most responsible for livestock related depredation conflicts with humans and to provide data for modeling occupancy across different habitat and study sites (Mackenzie et al. 2002).
Monitoring snow leopard population trends over time using non-invasive genetic methods should enable managers to better identify population declines and threats of local extinction, as well as assessing the effectiveness of conservation actions or the outcome of re-introduction programs. Individual identification can also be used to examine ecological and behavioral parameters such as home range size, spatial overlap between individuals and population turnover, as well as to identify ‘problem animals’ responsible for attacks on livestock or conflict with humans. Finally, individuals must be identified among the scat sampled before population, landscape and conservation genetics analyses can be conducted (Rodgers and Janečka 2012).
Polymorphic microsatellite markers within each population provide the basis for identifying individuals. For any two individuals, the probability that both will share the same allele at a given microsatellite locus is dependent upon the frequency of that allele in the population. As more loci are analyzed, the probability that two individuals will possess the same alleles at all loci decreases multiplicatively. The number of microsatellite loci that should be used to identify individuals is a balance between achieving sufficiently low probability of individual identity, whilst minimizing costs by using the least number of loci necessary. Where different populations are sampled and/ or where large numbers of individuals are being sampled (>100), additional loci may be required since some loci may not be variable across all populations.
Once individuals have been identified, the simplest approach to population estimation is to determine the minimum number of individuals present within the area surveyed. Capture–Recapture approaches comparable to camera trapping can be employed to generate more rigorous estimates of abundance based on sampling over multiple, independent sampling occasions; however, the important issue of aging of feces remains to be resolved.
The other important study design element – still under study –involves how best to delineate and estimate the actual area being sampled, since this suffers from similar biases associated with camera trap studies. The relationship between extraction success and scat age needs to be more thoroughly investigated in order to establish robust temporal limits for genetic surveys. Several population estimation methods are available if samples cannot be collected over sequential and multiple sampling occasions (Rodgers and Janečka 2012). Janečka et al. (2011) recommend that to minimize such bias, non-invasive scat survey transects should be uniformly distributed and oriented to maximize the area surveyed (see Box 2). These investigators also suggest that it may be preferable to sample a greater number of shorter transects, as opposed to a few long transects.
However, fecal genetics has a number of downsides: it is time consuming, and sequencing costs (roughly 5–10 USD/sample for species identification and 10-20 USD/sample for individual and gender identification) may be prohibitive for large-scale studies or for those individuals and agencies that do not have access to the latest, but costly sequencing equipment. Rodgers and Janečka (2012), Janečka et al. (2011) and Lampa et al. (2013) offer guidelines for sample collection, storage and DNA extraction. Rigorous error-detection procedures must be followed, including highly recommended exchange of blind samples between different laboratories. As noted by numerous investigators, the techniques employed must be carefully matched and customized to the task and questions being posed; these are undergoing constant change and improvement as more sophisticated sequencing equipment becomes available and new or more efficient approaches are identified. In addition, because allele sizes can vary, care must be taken when combining samples across different areas, time periods, or studies unless a subset of the same samples is analyzed together in order to calibrate the allele designations and arrive at reliable estimates of the number of individuals present.
Lampa et al. (2013) reviewed potential sources of error associated with non-invasive genetic capture-mark-recapture analysis, including low amplification success rates and genotyping errors that can substantially bias subsequent analysis. In order to attain reliable results and minimize time and costs required for non-invasive microsatellite genotyping, one must carefully choose a species-specific sampling design, methods that maximize the amount of template DNA, and methods that could overcome genotyping errors, especially when using low-quality samples. A key goal involves generating consensus genotypes to minimize errors that lead to overestimated population sizes. The literature includes many other methodological reviews that are beyond the scope of the SLSS document.
Research has indicated that even experienced field researchers mistakenly allocate feces to snow leopards which were deposited by foxes (Janečka et al. 2008). Scat misidentification leading to presumed species presence in an area where it does not occur may waste limited conservation resources. Therefore, genetically-based verification of feces is an essential component of any diet study, since it is impossible to definitively identify all snow leopard scats from that of other carnivores in the field.
In comparing these two methods, camera-trapping and fecal genetics, for surveying and monitoring snow leopards, Janečka et al. (2011) cautioned that estimates from the two approaches may be difficult to compare because of differences in the distribution of observations and because sub-adults are typically excluded from population estimates derived from camera-trapping, whilst differentiating between adults and sub-adults is not possible using non-invasive genetic techniques. However, these investigations concluded that the costs of estimating abundance of snow leopards using non-invasive genetic sampling is lower than for camera-trapping primarily because a much larger area can be covered within a shorter period of field time. Therefore, scat sampling has the potential to enable large- scale (i.e. regional) distribution and abundance surveys of snow leopards at significantly lower costs than camera trap surveys. This assumes equivalent laboratory capabilities between range states and/or that samples can be shared between designated laboratories located in different countries. Important advantages of such information sharing include the ability of different facilities to identify the same individuals (important in transboundary locations), use of standardized methods with increased rigor and robustness of the resulting data and cross-validation of data sets.
In genetic sampling, it is difficult to separate dependents (cubs) from adults, and as noted above the sampling period is also difficult to ascertain, because some collected scats might well be older than the desired sampling period.
However, a major advantage of camera trap and fecal genetics techniques is that all of the above can be done without having to capture or directly observe snow leopards, thus greatly increasing the number of individuals sampled (compared to costly telemetry) and the amount of data that can be collected (compared to camera-trapping). In addition, snow leopard density estimates from various sites are in turn expected to help understand relationships between these estimates and sign-based occupancy estimates, at various scales.
14.4. Occupancy modelling
One issue influencing estimates of population density of snow leopards and other secretive species is that of ‘detectability’ – the probability of detecting a snow leopard during a survey even if it is present at the site. A new type of statistical modelling technique (occupancy models) was developed to deal with problems created by imperfect detectability (Mackenzie et al. 2006). These models use information from repeated observations and the proportion of sites occupied. The records of whether the species was detected or not detected at each site during each survey are then converted to mathematical statements. Occupancy modelling offers potential in increasing the robustness of density estimates, but is dependent on meeting a set of assumptions to avoid bias. Few if any applications to snow leopards have so far been reported.
14.5. Radio and satellite telemetry
Radio-collaring of snow leopards was pioneered in western Nepal in the 1980s (Jackson and Ahlborn 1989 and generated valuable new information on home ranges, movement patterns and density. Subsequently, similar operations were undertaken in NW India (Chundawat 1990), western Mongolia (McCarthy 2000), Nepal (Oli 1997) and Pakistan (McCarthy et al. 2007). The feasibility of remote telemetry has been dramatically improved with the availability of miniaturized radio collars and GPS satellite technology which allows multiple fixes to be obtained daily and downloaded remotely, thus providing detailed data on movement patterns, habitat use, home range size and dispersal. This technique has been deployed in an ongoing, long-term study in the South Gobi, Mongolia during which 19 snow leopards have been collared. The first results of this study have now been published (Sharma et al. 2014). In other areas, in 2008 one animal was equipped with a satellite collar in Baga Bogd, Mongolia (B.Munkhtsog, pers comm.) and five snow leopards were fitted with these devices in Wakhan, Afghanistan in 2013 (P. Zahler, pers. comm. 2014). Researchers recently collared a male snow leopard in eastern Nepal (WWF-Nepal, pers comm.). Further, a device to automate and enable constant monitoring of trap transmitters while capturing snow leopards has been invented (Johansson et al. 2011).
14.6. Monitoring prey populations
Developing robust methods of estimating prey populations that allow for statistical comparison over time and space are crucial. Singh and Milner-Gulland (2011) explored the pros and cons of various methods available for monitoring snow leopard prey populations. Suryawanshi et al. (2012) propose the use of double observer methods to monitor snow leopard prey. The methods have been implemented at a few sites including India and Mongolia, and the results have been found comparable to predicted prey populations. Although some minor modifications may need to be made in the method to cater to specific issues such as behavioral response of animals in areas where hunting is prevalent, or have highly variable prey group sizes (Tomotsukh (2013), the method can efficiently be used to estimate and monitor prey populations in specific landscapes. For larger landscapes that include an entire range or a province, information on local extinctions and colonisations can be generated using site occupancy framework. This method may not provide abundance estimates, but provides useful probabilistic estimates of local colonization and extinctions taking place for the species of interest. Further testing and development of prey abundance monitoring techniques is warranted.
Attaining robust estimates of the size of mountain ungulate populations is problematic due to a range of factors including broken terrain hindering visibility and clumped distributions and consequent issues with detectability. It is notoriously difficult to meet the assumptions for Distance Sampling and a combination of block and point-counts is still widely used, e.g. as developed for censusing argali and ibex in Kyrgyzstan by the IUCN Caprinae Specialist Group and others. Such raw or uncorrected counts are often the best available, but biases cannot be quantified and so the results should be interpreted with care (see e.g. Wingard et al. 2011).
Harris et al. (2010) concluded that fecal DNA analysis of prey species like the argali is too expensive, although this study demonstrated the importance of sex identification and separate sex-based abundance estimates, especially where movement ecology differs by sex.
14.7. Aerial photography
Recent technical advances in aerial photography have greatly improved the counting of ungulates in open terrain and trials should be conducted in mountain habitats to assess their potential, especially in conjunction with lightweight unmanned aerial vehicles (UAVs), commonly termed drones, which are being increasingly deployed as a tool in biodiversity surveys and environmental monitoring.
For larger landscapes that include an entire range or a province, information on local extinctions and colonisations can probably be generated using site occupancy framework, assuming ungulates can be detected using technologies like thermal sensing; feasibility studies in the high mountains have yet to be conducted. Presumably winter, fall or spring represent the best time, when ambient temperatures are low and the animal’s warm-bodied thermal signatures most different from that of the background. Even where this method cannot provide abundance estimates, it might enable useful probabilistic estimates of local colonization and extinctions for the species of interest. Further testing and development of prey abundance monitoring techniques is warranted in order to establish optimal grid dimension and sampling replicates.
We also postulate that the visual or thermal detection of snow leopard by remotely controlled UAVs is probably not possible, except in unusual circumstances, given this species’ thick pelage and excellent thermal conservation as well as its preference for seeking protected daytime beds.
We also recommend investigating the option of involving local students and herders as “citizen naturalists or scientists” in helping to document ungulate sightings over time under a standardized occupancy framework managed by a trained biologist. While these individuals would probably not have access to GPS units, their records could be based on the pre-designated watershed and/or livestock pasture polygons or blocks along with use of simple data forms for gathering and tabulating sightings and/or sign.
14.8. Monitoring snow leopards: the need for regional and global monitoring alliances
Monitoring serves as an alarm system alerting managers to possible population declines and for enabling the creation of population baselines against which conservation interventions and targets can be designed, measured and compared between areas or over time. However, it requires a well-implemented monitoring program in order to evaluate snow leopard status and performance of conservation initiatives. Any monitoring program should be structured in order to facilitate adaptive management.
There are two simple but real constraints in monitoring the abundance of snow leopards. First, estimating abundance over the entire habitat, or even in the area of interest, is usually difficult, and in the specific cases of snow leopard, it may not be feasible. Second, while it is usually simpler to detect the presence of the species within a given sampling unit, it is far more challenging to ascertain or confirm its absence (the problem of false absence). Appropriately designed sampling schemes, together with analytical frameworks such as occupancy modeling or capture-recapture analyses can help overcome these challenges.
The key elements in monitoring snow leopards involve tracking population size and trend. At the very least, the presence/absence of snow leopards (based on sign, sightings, reports of local informants etc) within carefully designated sampling units could be monitored, yielding distribution maps and changes in distribution over time. At the level of landscapes, changes in the relative abundance of snow leopards could be monitored, estimated through various indices including snow leopard sign, camera trap capture rates and individual IDs and fecal DNA sampling. Finally, the most desirable — albeit also the most difficult indicator to monitor — is the actual population size of snow leopards, and how each population or subpopulation in the general region changes over time, along with the level of functional connectivity between separate entities.
How frequently should the monitoring be carried out? That is no simple answer but snow leopard abundance in the key conservation landscapes using robust mark-recapture or SCER techniques could ideally be estimated annually or once every in three years. Large-scale (country-wide) snow leopard occurrence could be monitored once in 5 years based on occupancy frameworks, and supplemented by more localized camera trapping or fecal DNA surveys.
The importance of coordinating snow leopard landscape identification and population monitoring cannot be over-estimated. SLN therefore recommends the GSLEP sponsors organize a workshop bringing together experts from range state governments, and national and international research institutions and NGOs to address the questions posed above with the objective of generating a Range-Wide Plan for monitoring progress.
Ahlborn, G. and Jackson, R. (1988). Marking in free-ranging snow leopards in west Nepal: a preliminary assessment. Pages 25-49 In: H. Freeman (Ed). Proceedings 5th International Snow Leopard Symposium International Snow Leopard Trust and Wildlife Institute of India, Seattle, Washington. 269 pages.
Chundawat, R. S., Panwar, H.S and Rawat, G.S. (1990). The ecological studies of snow leopard and its associated prey species in Hemis High Altitude National Park, Ladakh. Technical Report No. RR-l, Wildlife Institute of India, Dehra Dun.
Borchers, D.L .and Efford, M.G. (2008). Spatially explicit maximum likelihood methods for capture-recapture studies. Biometrics, 64, 377-385.
Efford, M. (2004). Density estimation in live-trapping studies. Oikos, 106, 598–610.
Efford, M. G. (2011). Estimation of population density by spatially explicit capture-recapture analysis of data from area searches. Ecology, 92, 2202–7.
Efford, M.G., Dawson, D.K and Robbins, C.S. (2004). DENSITY: software for analyzing capture-recapture data from passive detector arrays. Animal Biodiversity and Conservation, 27(1), 217-228.
Foster, R.J. and Harmsen, B.J. (2011). A critique of density estimation from camera-trap data. Journal Wildlife Management, 76(2), 224–236.
Gopalaswamy, J.M., Royle, J.A., Hines, J.E., Singh, P., Jathanna, D., Kumar, N.S. and Karanth, K.U. (2012). Program SPACECAP: software for estimating animal density using spatially-explicit capture-recapture models. Methods in Ecology and Evolution, 3(6), 1067-1072.
Harris, R.B., Winnie, J. Jr., Amish, S.J., Beja-Pereira, A., Godinho, R., Nia Costa, V. and Luikart, G. (2010). Argali abundance in the Afghan Pamir using capture–recapture modeling from fecal DNA. Journal of Wildlife Management, 74(4), 668–677.
Jackson, R. and Ahlborn, G. (1989). Snow leopards (Panthera uncia) in Nepal: home range and movements. National Geographic Research, 5, 161-175.
Jackson, R. and Hunter, D.O. (1996). Snow leopard survey and conservation handbook. International Snow Leopard Trust, Seattle, and U.S. Geological Survey, Biological Resources Division, 154 pages + appendices.
Jackson, R., Roe, J.D., Wangchuk, R. and Hunter, D.O. (2006). Estimating snow leopard population abundance using photography and capture–recapture techniques. Wildlife Society Bulletin, 34, 772–781.
Jackson, R., Munkhtsog, B., Mallon, D.P., Naranbaatar, G. and Khurlee Gerelmaa, K. (2009). Camera-trapping snow leopards in the Tost Uul region of Mongolia. Cat News, 51, 18-21.
Janečka, J.E., Jackson, R., Yuguang, Z., Diqiang, L., Munkhtsog, B., Buckley-Beason, V. and Murphy, W.J. (2008). Population monitoring of snow leopards using noninvasive collection of scat samples: a pilot study. Animal Conservation, 11, 401-411.
Janečka, J.E, Munkhtsog, B., Jackson, R., Naranbaatar, G., Mallon D.P., and Murphy, W.J. (2011). Comparison of noninvasive genetic and camera-trapping techniques for surveying snow leopards. Journal of Mammalogy, 92, 771–783.
Karanth, K., & Chundawat, R. (2004). Estimation of tiger densities in the tropical dry forests of Panna, Central India, using photographic capture–recapture sampling. Animal Conservation, 7, 285–290.
Kéry, M., Gardner, B., Stoeckle, T., Weber, D. and Royle, J.A. (2011). Use of spatial capture-recapture modeling and DNA data to estimate densities of elusive animals. Conservation Biology, 25, 356–364.
Lampa, S., Henle, K., Klenke, R., Hoehn, M. and Gruber, B. (2013). How to overcome genotyping errors in non-invasive genetic mark-recapture population size estimation—a review of available methods illustrated by a case study. Journal of Wildlife Management, 77(8), 1490–1511.
Li, J., Wang, D., Yin, H., Zhaxi, D., Jiagong, Z., Schaller, G. B., and Lu, Z. (2014). Role of Tibetan Buddhist monasteries in snow leopard conservation. Conservation Biology, 28(1), 87-94.
Linkie, M., Guillera-Arroita,G., Smith, J. and Rayan, D.M. (2010). Monitoring tigers with confidence. Integrative Zoology, 5, 342-350.
Mackenzie, D. I., Nichols, J.D., Lachman, G.B., Droege, S., Royle, J.A. and Langtimm, C.A. (2002). Estimating site occupancy when detection probabilities are less than one. Ecology, 83, 2248–2255.
Mackenzie, D. I., Nichols, J.D., Royle, J.A., Pollock, K.H., Bailey, L.L and Hines, J.E. (2006). Occupancy estimation and modeling: inferring patterns and dynamics of species occurrence. Elsevier, Amsterdam.
McCarthy T.M. (2000). Ecology and conservation of snow leopards, Gobi brown bears and wild Bactrian camels in Mongolia. University of Massachusetts, Amherst, MA, USA.
McCarthy T.M., Khan J., Ud-Din J. and McCarthy K. (2007). First study of snow leopards using GPS-satellite collars underway in Pakistan. Cat News, 46, 22-23.
McCarthy, K. P., Fuller, T. K., Ming, M., McCarthy, T. M., Waits, L., and Jumabaev, K. (2008). Assessing estimators of snow leopard abundance. Journal of Wildlife Management, 72, 1826–1833.
Oli M.K. (1997). Winter home range of snow leopards in Nepal. Mammalia, 62, 355-360.
Obbard, M. E., Howe, E. J., and Kyle, C. J. (2010). Empirical comparison of density estimators for large carnivores. Journal of Applied Ecology, 47(1), 76–84.
Rodgers, T.W. and Janečka, J.E. (2013). Applications and techniques for non-invasive fecal genetics research in felid conservation. European Journal of Wildlife Research, 59, 1-16.
Royle, J.A, and Young, K. (2008). A hierarchical model for spatial capture-recapture data. Ecology, 92, 526–528.
Royle, J.A, Karanth, K.U., Gopalaswamy,A.M. and Kumar, N.S.S (2009). Bayesian inference in camera trapping studies for a class of spatial capture–recapture models. Ecology, 90, 3233–3244
Sharma, R.K., Jhala, Y., Qureshi, Q., Vattakaven, J., Gopal, R., and Nayak, K. (2010). Evaluating capture–recapture population and density estimation of tigers in a population with known parameters, Animal Conservation, 13(1), 94-103.
Singh, N.J. and Milner-Gulland, E.J. (2011). Monitoring ungulates in Central Asia: current constraints and future potential. Oryx, 45, 38–49.
Soisalo, M.K. and. Cavalcanti, S.M.C. (2006). Estimating the density of a jaguar population in the Brazilian Pantanal using camera-traps and capture-recapture sampling in combination with GPS radio-telemetry. Biological Conservation, 129, 487-496.
Sollmann, R.,Tôrres, N.M., Furtado, M.M, Tereza de Almeida, A., Jácomo, F., Palomares, S., Roques, S. and Silveira, L.( 2013). Combining camera-trapping and noninvasive genetic data in a spatial capture–recapture framework improves density estimates for the jaguar. Biological Conservation, 167, 242-247.
Stanley, T. R. and Burnham, K.P. (1999). A closure test for time-specific capture–recapture data. Environmental and Ecological Statistics, 6, 197–209.
Suryawanshi, K.R., Bhatnagar, Y.V. & Mishra, C. (2012) Standardizing the double-observer survey method for estimating mountain ungulate prey of the endangered snow leopard. Oecologia, 169, 581–90.
Swartz, M.K., Luikart, G. and Waples, R.S. (2006). Genetic monitoring as a promising tool for conservation and management. Trends in Ecology and Evolution, 22(1), 25-33.
Tobler, M.W. and Powell, G.V.N. (2013). Estimating jaguar densities with camera traps: Problems with current designs and recommendations for future studies. Biological Conservation, 159, 109–118.
Wingard, G.J., Harris, R.B., Amgalanbaatar, S. and Reading, R.P. (2011). Estimating abundance of mountain ungulates incorporating imperfect detection: argali (Ovis ammon) in the Gobi Desert, Mongolia. Wildlife Biology, 17, 93-101.
World Bank (2013). Global Snow Leopard & Ecosystem Protection Program: Conference Document for Endorsement (71 pages) and Annex Document of Complete National Programs (234 pages). Washington DC.
Box 1: Specific Recommendations for Camera-trap Surveys of Snow Leopards
Box 2: Specific Recommendations for Non-invasive Fecal Genetic Surveys
BOX 1: Recommended Guidelines for Remote Trail Camera Surveys of Snow Leopard
Camera Trapping, has become a preferred tool for investigating snow leopard abundance and distribution patterns. Nonetheless, we believe more attention should be devoted toward standardizing sampling procedures and minimizing the underlying sources of bias that result from this felid’s low numbers, sparse distribution and relatively large home ranges (Jackson et al, 2006, McCarthy et al 2008). The snow leopard’s rugged and largely roadless habitat hampers on-the-ground logistics, making the deployment and servicing of cameras both time-consuming and relatively expensive compared with other large cats like jaguar and tiger.
We urge all researchers to consult the literature prior to undertaking their camera survey. Unless carefully planned and then rigorously executed, the resulting survey may end up reporting little more than the “number of photographs taken.” Many surveys fail in properly documenting photo-capture rates, or to even list the minimum number of individuals detected — thus limiting their usefulness for conservation.
Camera reports often only consist of a series of photos disseminated through social media. Unless robust and defensible abundance and/or density estimates are undertaken, snow leopard conservation will suffer and valuable resources may be needlessly squandered.
The key elements and associated questions that need to be addressed during any remote camera survey should include the following:
- Pre-survey planning, including launching a pilot or feasibility study to address the question, “what are the prospects of the survey generating sufficient information on population size and/or occupied range in Area X and Y?”
- Critical parameters for population estimation under the Capture-Recapture (C-R) framework, including a working understanding of the assumptions underlying each model
- Size and configuration of the survey area (how big should an area should be surveyed and how many trail cameras do I need to complete the survey?)
- Spacing of trail cameras (where should I place cameras and how far apart should they be?)
- Duration of the survey, including identifying an optimal sampling period and session interval
- Data analysis and interpretation (Which population estimation model or models are most appropriate for my dataset? How much information will I need to collect to complete the computations?)
- Lessons learned and recommendations (how can future surveys be improved?)
- Dissemination of survey results (Who should I share the results with? What information is most important to share with others?)
Survey Pre-Planning: We urge all investigators to familiarize themselves thoroughly with the available literature, and to consult experts for advice. Further, it is highly recommended that researchers carry out simulation studies in order to test their study design prior to deploying camera traps in the field. These simulations are readily implemented using the SECR package in R (Efford, 2011b) or the software DENSITY (Efford, 2011a). Both packages allow users to use a real camera trap layout and to define realistic parameters as the basis for simulations. By varying parameters like trap density, home range size or encounter rates, one can evaluate the range of conditions under which the proposed study design would be expected to give the best results.
Paying attention to even small details, like minimizing false-triggering, maintaining camera battery life and freeing memory card space, pays major dividends in the long-run through more time-efficient and site-sensitive data collection. In order to estimate population densities robustly, researchers must carefully plan and adopt sampling protocols that are based on a biologically meaningful understanding of snow leopard habitat use and movement patterns in the survey area. This, in turn, will help ensure cameras are well located, thus more likely to generate high capture and recapture rates which are required by most computer-enabled capture-recapture computations.
Demographic and Geographic Closure: Abundance and density estimates depend upon meeting the basic assumptions underlying the C-R framework, namely demographic and geographic closure of the sampled population over the entire sampling period. In addition, one should be familiar with known or potential
sources of bias that lead to unequal capture rates between individuals or variations over time – all of which need to be addressed through proper survey design, camera placement etc.
Study Area Size: Camera trap surveys are only useful if they encompass most or all of the home ranges of multiple individuals, preferably for a sample size in excess of 15-20 individuals. While home range sizes varies widely, from around 50-100 km2 in the Himalaya to as much as 800 km2 in Mongolia, individual snow leopard home ranges tend to overlap to a large extent, especially between gender. Dominant males occupy home ranges 2 – 4 times greater than those of females, along with exhibiting greater spatial exclusivity. Females with cubs gradually expand their area of use as their litter ages and become more mobile. Therefore, on the basis of home range size and assuming a population in excess of 5-10 individuals, the survey area should total at least 300 km2 and 1,000 km2 or more in size. The study area boundaries and configuration should follow the natural local mountain terrain, and may include patches of less suitable habitat like an intervening alluvial plain or high-elevation snowfield.
Camera Placement: Trail cameras should be placed at sites in ways that maximize the detection of a passing snow leopard (i.e. at communal sign sites and along well used travel routes, including well-defined ridgelines, the base of cliffs or along narrow drainages). We recommend striving for capture probabilities of ≥ 0.30. Such capture rates serve to improve the precision of survey estimates and resulting upper and lower population bounds (95% confidence intervals), of course assuming a sufficiently large survey area and optimized camera spacing. Note that camera arrays placed randomly within the landscape have low or very low capture probabilities, resulting in poor if any usable data.
Size of Camera Polygon and Camera Spacing: The camera polygon or grid size for a density study should be at least the size of one home range (usually female). However, researchers are still debating whether to use the average size of an adult female or male range, a metric that should not be confused with total extent of habitat covered during the course of an entire camera survey. Extending the survey area helps reduce potential bias, increase sample size, and include greater habitat heterogeneity, thereby making it more representative of the general area and thus better at supporting the extrapolation of density values to a larger regional scale. Tobler and Powell (2013) recommend that when the area that can be covered by a camera polygon is limited, using a more rectangular grid should improve density estimates by SECR models. In that case, the survey area design should attempt to have the long side of the polygon be at least the length of one home range diameter.
Camera Spacing: Jackson et al. (2006) used a camera density of two cameras per 16-32 km2 for his survey in Ladakh and which relied upon analog trail cameras that require more regular servicing because of the maximum of 36 images per roll of film. Photo capacity is no longer a limitation with the advent of digital trail cameras that are capable of storing many thousands of pictures — meaning trail cameras can be left unattended longer and dispersed over a wider area. Periodic visit to ensure the units are functioning properly are essential, however.
Tober and Powell (2013) concluded that the maximum distance between cameras depends on the female home range which is generally much smaller than those of males. According to simulations by these investigators, surveys with fewer stations will likely result in biased or at least imprecise results unless capture probabilities are very high. If the number of cameras available is smaller than the total number needed, a blocked design can be used where cameras are periodically moved during the survey to ensure there are no gaps large enough to enable an individual snow leopard to remain undetected during the course of the survey. Procedures for enabling this are well documented in the literature.
Based on this and other surveys, we recommend spacing cameras at least 2-3km apart and preferably at distances of 4-5 km, with a total of at least 20-25 trap stations deployed and concurrently operational in the same area, over a minimum of 30-40 consecutive days.
Survey Duration and Sampling Intervals: Camera trap surveys require a minimum of 40-50 days, although 60-80 days are preferred as long as these do not violate assumptions of population closure. This is best assessed using the program Closure (Stanley and Burnham 1999). Sampling intervals can vary from 1 day to 3 or 5 day intervals with aggregated data providing higher capture probabilities. Based on pre-survey trials, investigators need to estimate mean times to first captures, then using this interval as a preferred duration before shifting the camera array to new sites.
Data Analysis and Available Models: C-R models have become far more sophisticated since the original work by White et al. (1982); they are implemented through software programs like Capture (Otis et al 1878) and Mark (White and Burnham 1999) and more recently R program modules such as RMark and RCapture. The program SpaceCap automates implementation of the Bayesian SECR abundance estimation method (Gopalaswamy et al. 2012). For details consult the original papers as well as the camera trap handbook available for download from the Snow Leopard Conservancy’s website: (http://www.snowleopardconservancy.org/pdf/screen111705.pdf)
We recommend using SECR models since these accommodate differences in capture probability between sites, and offer more robust estimation of the area sampled than prior methods that relied on buffering trap arrays (for example, mean maximum distance moved, MMDM or ½ MMDM). Under Mark or RMark, one can examine the influence of predictive site and individual covariates (e.g., habitat type, sex/age cohort etc). Besides altering camera polygon size, another means for increasing sample size entails combining data from multiple surveys where home ranges and habitat are similar. When applying the SECR model, it is straight forward to include the exact number of days each camera was active, thus accounting for camera failure or blocked designs where not all cameras were active at the same time. Not accounting for camera failure can lead to biased density estimates resulting from the underestimation of capture probabilities. Another approach receiving attention recently is the hierarchical Bayesian multi-state mark–recapture model (Calvert et al. 2009) that permits partitioning of complex parameter variation across space or time, as well as simultaneous analysis of multiple data sets. Depending upon data availability, subsampling of environmental covariates greatly improves the utility of population estimates. The assumption of equal detectability rarely holds, so estimating capture probabilities based on sex and/or age should receive high priority.
Like other felids, male and female snow leopards appear to have different home range sizes with varying encounter rates and not accounting for this may lead to biased density estimate. Therefore, try to include sex covariates both for the k0 and the r parameter. While the maximum likelihood implementation of SECR models requires the sex of each individual to be known, the Bayesian implementation allows for such missing data (Sollmann et al., 2011). With the inclusion of covariates, however, the data are divided up into smaller groups and thus larger sample sizes will be needed. SECR models with sex covariates have been run with 10 individuals in jaguars (Sollmann et al., 2011), but a sample size of 30 or more individuals is required for more precise estimates with smaller confidence intervals. Clearly, such a target is difficult or very difficult to achieve in snow leopards, perhaps another reason why genetic means for estimating population merit more attention.
Use of Baits and Attractants: As noted above, generating unbiased sample of populations is not simple: due to behavioral differences between species and individuals, the abundance of photo-captures is being constantly influenced by how each target animal reacts to the remote camera, human activity, the presence (or absence) of roads or trails, and differences in habitat. Therefore, we do not recommend using baits or artificial lures on formalized abundance surveys, since these may result in varying responses (e.g., camera-happy individuals may be attracted while camera-shy individuals deterred by such scent).
Detecting Population Change over Time: The ability to detect population change over time with any degree of confidence represents an unresolved challenge because of typically low detection probabilities, limited sample sizes, and site specific variation and changes in snow leopard social status and structure over the short-term and upon which optimal trap placement may depend to some degree. Linkie et al. (2010) in referring to low density tiger populations (e.g. <1 adult tigers/100 km2) noted obtaining sufficient precision for state variable estimates from camera trapping represented a major challenge because of insufficient detection probabilities and/or sample sizes. These investigators suggested that occupancy surveys might overcome this problem by redefining the sampling unit (e.g. grid cells and not individual tigers).
BOX 2: Suggested Guidelines for Fecal (Scat) Genetic Surveys
SLSS recommends a collaborative effort to test, evaluate and formulate standard protocols for undertaking non-invasive fecal genetic surveys aimed at quantifying snow leopard presence/absence (occupancy) at regional scales, along with population size estimation based on the number of unique individuals detected over representative time frames. These protocols are needed in support of the Global Snow Leopard Environmental Protection Plan (World Bank 2013) goal of “20 landscape-level populations of snow leopards secured by the year 2020.”
We suggest that 2-3 contiguous blocks of habitat which could support >100 breeding snow leopards be selected in each country. Within these blocks a noninvasive genetic survey should be carried out in 3 sites; the core area of the block, representative peripheral area at the edges, and an area of intermediate habitat quality. Field staff would collect recent (fresh) scats along transects in 10 predefined 25 km2 sampling grids located in each of the 3 areas for a total of 30 grids sampled. Late winter, after the snow cover has melted but before daytime ambient temperatures rise much above about 50 °F represents a prime time for sample collection. Late fall, before the onset of snowfall offers another suitable window. Approximately 30 scats would be collected per grid, for a total of up to 900 scats per survey block.
All samples will be collected using a standardized method, stored and then analyzed at an approved laboratory according to a mutually agreed methodology that includes rigorous quality controls implemented at the country, regional and international levels. Since snow leopard habitat and populations of range countries extend across their national boundary to neighboring states, it is essential that all samples be aligned along a known base pair scale, so that common (i.e. the same) individuals can be identified in locations where (a) snow leopards are known to move between separate populations including international borders, and (b) where young animals (especially males) disperse further from their natal area, and may thus range into a neighboring country or adjacent snow leopard landscape unit.
Detailed information on methodologies for sequencing samples successfully while maintaining high quality control standards is beyond the scope of the SLSS document, but will need to be addressed by both researchers and supporting government agencies.